first stab

Updates the requirements on [lru](https://github.com/jeromefroe/lru-rs) to permit the latest version.
- [Release notes](https://github.com/jeromefroe/lru-rs/releases)
- [Changelog](https://github.com/jeromefroe/lru-rs/blob/master/CHANGELOG.md)
- [Commits](https://github.com/jeromefroe/lru-rs/compare/0.6.5...0.7.0)

---
updated-dependencies:
- dependency-name: lru
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>

Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

.github/FUNDING.yml

@@ -0,0 +1,12 @@
+# These are supported funding model platforms
+
+github: fulmicoton
+patreon: # Replace with a single Patreon username
+open_collective: # Replace with a single Open Collective username
+ko_fi: # Replace with a single Ko-fi username
+tidelift: # Replace with a single Tidelift platform-name/package-name e.g., npm/babel
+community_bridge: # Replace with a single Community Bridge project-name e.g., cloud-foundry
+liberapay: # Replace with a single Liberapay username
+issuehunt: # Replace with a single IssueHunt username
+otechie: # Replace with a single Otechie username
+custom: # Replace with up to 4 custom sponsorship URLs e.g., ['link1', 'link2']

.github/ISSUE_TEMPLATE/actions.md

@@ -0,0 +1,13 @@
+---
+name: Actions
+about: Actions not directly related to producing code.
+
+---
+
+# Actions title
+
+Action description. 
+e.g. 
+- benchmark
+- investigate and report
+- etc.

.github/dependabot.yml

@@ -0,0 +1,8 @@
+version: 2
+updates:
+- package-ecosystem: cargo
+  directory: "/"
+  schedule:
+    interval: daily
+    time: "20:00"
+  open-pull-requests-limit: 10

.github/workflows/coverage.yml

@@ -0,0 +1,25 @@
+name: Coverage
+
+on:
+  push:
+    branches: [ main ]
+  pull_request:
+    branches: [ main ]
+
+jobs:
+  coverage:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+      - name: Install Rust
+        run: rustup toolchain install nightly --component llvm-tools-preview
+      - name: Install cargo-llvm-cov
+        run: curl -LsSf https://github.com/taiki-e/cargo-llvm-cov/releases/latest/download/cargo-llvm-cov-x86_64-unknown-linux-gnu.tar.gz | tar xzf - -C ~/.cargo/bin
+      - name: Generate code coverage
+        run: cargo llvm-cov --all-features --workspace --lcov --output-path lcov.info
+      - name: Upload coverage to Codecov
+        uses: codecov/codecov-action@v1
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }} # not required for public repos
+          files: lcov.info
+          fail_ci_if_error: true

.github/workflows/long_running.yml

@@ -0,0 +1,24 @@
+name: Rust
+
+on:
+  push:
+    branches: [ main ]
+
+env:
+  CARGO_TERM_COLOR: always
+  NUM_FUNCTIONAL_TEST_ITERATIONS: 20000
+
+jobs:
+  functional_test_unsorted:
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v2
+    - name: Run indexing_unsorted
+      run: cargo test indexing_unsorted -- --ignored
+  functional_test_sorted:
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v2
+    - name: Run indexing_sorted
+      run: cargo test indexing_sorted -- --ignored
+

.github/workflows/test.yml

@@ -0,0 +1,30 @@
+name: Rust
+
+on:
+  push:
+    branches: [ main ]
+  pull_request:
+    branches: [ main ]
+
+env:
+  CARGO_TERM_COLOR: always
+
+jobs:
+  test:
+
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v2
+    - name: Build
+      run: cargo build --verbose --workspace
+    - name: Install latest nightly to test also against unstable feature flag
+      uses: actions-rs/toolchain@v1
+      with:
+            toolchain: nightly
+            override: true
+            components: rustfmt
+    - name: Run tests
+      run: cargo test --all-features --verbose --workspace
+    - name: Check Formatting
+      run: cargo fmt --all -- --check

.gitignore

@@ -1,4 +1,5 @@
 tantivy.iml
+proptest-regressions
 *.swp
 target
 target/debug
@@ -11,3 +12,4 @@ cpp/simdcomp/bitpackingbenchmark
 *.bk
 .idea
 trace.dat
+cargo-timing*

.travis.yml

@@ -10,7 +10,7 @@ env:
   global:
     - CRATE_NAME=tantivy
     - TRAVIS_CARGO_NIGHTLY_FEATURE=""
-    - secure: eC8HjTi1wgRVCsMAeXEXt8Ckr0YBSGOEnQkkW4/Nde/OZ9jJjz2nmP1ELQlDE7+czHub2QvYtDMG0parcHZDx/Kus0yvyn08y3g2rhGIiE7y8OCvQm1Mybu2D/p7enm6shXquQ6Z5KRfRq+18mHy80wy9ABMA/ukEZdvnfQ76/Een8/Lb0eHaDoXDXn3PqLVtByvSfQQ7OhS60dEScu8PWZ6/l1057P5NpdWbMExBE7Ro4zYXNhkJeGZx0nP/Bd4Jjdt1XfPzMEybV6NZ5xsTILUBFTmOOt603IsqKGov089NExqxYu5bD3K+S4MzF1Nd6VhomNPJqLDCfhlymJCUj5n5Ku4yidlhQbM4Ej9nGrBalJnhcjBjPua5tmMF2WCxP9muKn/2tIOu1/+wc0vMf9Yd3wKIkf5+FtUxCgs2O+NslWvmOMAMI/yD25m7hb4t1IwE/4Bk+GVcWJRWXbo0/m6ZUHzRzdjUY2a1qvw7C9udzdhg7gcnXwsKrSWi2NjMiIVw86l+Zim0nLpKIN41sxZHLaFRG63Ki8zQ/481LGn32awJ6i3sizKS0WD+N1DfR2qYMrwYHaMN0uR0OFXYTJkFvTFttAeUY3EKmRKAuMhmO2YRdSr4/j/G5E9HMc1gSGJj6PxgpQU7EpvxRsmoVAEJr0mszmOj9icGHep/FM=
+    # - secure: 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
 
 addons:
   apt:
@@ -38,20 +38,21 @@ matrix:
     # Linux
     #- env: TARGET=aarch64-unknown-linux-gnu
     #- env: TARGET=i686-unknown-linux-gnu
-    - env: TARGET=x86_64-unknown-linux-gnu CODECOV=1
+    - env: TARGET=x86_64-unknown-linux-gnu CODECOV=1 #UPLOAD_DOCS=1
     # - env: TARGET=x86_64-unknown-linux-musl CODECOV=1
-
     # OSX
-    - env: TARGET=x86_64-apple-darwin
-      os: osx
+    #- env: TARGET=x86_64-apple-darwin
+    #  os: osx
 
 before_install:
   - set -e
   - rustup self update
+  - rustup component add rustfmt
 
 install:
   - sh ci/install.sh
   - source ~/.cargo/env || true
+  - env | grep "TRAVIS"
 
 before_script:
   - export PATH=$HOME/.cargo/bin:$PATH
@@ -60,14 +61,25 @@ before_script:
 
 script:
   - bash ci/script.sh
+  - cargo fmt --all -- --check
 
 before_deploy:
   - sh ci/before_deploy.sh
 
-cache: cargo
-before_cache:
-  # Travis can't cache files that are not readable by "others"
-  - chmod -R a+r $HOME/.cargo
+after_success:
+  # Needs GH_TOKEN env var to be set in travis settings
+  - if [[ -v GH_TOKEN ]]; then echo "GH TOKEN IS SET"; else echo "GH TOKEN NOT SET"; fi
+  - if [[ -v UPLOAD_DOCS ]]; then cargo doc; cargo doc-upload; else echo "doc upload disabled."; fi
+
+#cache: cargo
+#before_cache:
+#  # Travis can't cache files that are not readable by "others"
+#  - chmod -R a+r $HOME/.cargo
+#  - find ./target/debug -type f -maxdepth 1 -delete
+#  - rm -f  ./target/.rustc_info.json
+#  - rm -fr ./target/debug/{deps,.fingerprint}/tantivy*
+#  - rm -r target/debug/examples/
+#  - ls -1 examples/ | sed -e 's/\.rs$//' | xargs -I "{}" find target/* -name "*{}*" -type f -delete
 
 #branches:
 #  only:
@@ -77,4 +89,4 @@ before_cache:
 
 notifications:
   email:
-    on_success: never
+    on_success: never

ARCHITECTURE.md

@@ -0,0 +1,295 @@
+# Tantivy
+
+## What is tantivy?
+
+Tantivy is a library that is meant to build search engines. Although it is by no means a port of Lucene, its architecture is strongly inspired by it. If you are familiar with Lucene, you may be struck by the overlapping vocabulary.
+This is not fortuitous.
+
+Tantivy's bread and butter is to address the problem of full-text search :
+
+Given a large set of textual documents, and a text query, return the K-most relevant documents in a very efficient way. To execute these queries rapidly, the tantivy needs to build an index beforehand. The relevance score implemented in the tantivy is not configurable. Tantivy uses the same score as the default similarity used in Lucene / Elasticsearch, called [BM25](https://en.wikipedia.org/wiki/Okapi_BM25).
+
+But tantivy's scope does not stop there. Numerous features are required to power rich-search applications. For instance, one may want to:
+- compute the count of documents matching a query in the different section of an e-commerce website,
+- display an average price per meter square for a real estate search engine,
+- take into account historical user data to rank documents in a specific way,
+- or even use tantivy to power an OLAP database.
+
+A more abstract description of the problem space tantivy is trying to address is the following.
+
+Ingest a large set of documents, create an index that makes it possible to
+rapidly select all documents matching a given predicate (also known as a query) and
+collect some information about them ([See collector](#collector-define-what-to-do-with-matched-documents)).
+
+Roughly speaking the design is following these guiding principles:
+- Search should be O(1) in memory.
+- Indexing should be O(1) in memory. (In practice it is just sublinear)
+- Search should be as fast as possible
+
+This comes at the cost of the dynamicity of the index: while it is possible to add, and delete documents from our corpus, the tantivy is designed to handle these updates in large batches.
+
+## [core/](src/core): Index, segments, searchers.
+
+Core contains all of the high-level code to make it possible to create an index, add documents, delete documents and commit.
+
+This is both the most high-level part of tantivy, the least performance-sensitive one, the seemingly most mundane code... And paradoxically the most complicated part.
+
+### Index and Segments...
+
+A tantivy index is a collection of smaller independent immutable segments. 
+Each segment contains its own independent set of data structures.
+
+A segment is identified by a segment id that is in fact a UUID.
+The file of a segment has the format
+
+ ```segment-id . ext ```
+
+The extension signals which data structure (or [`SegmentComponent`](src/core/segment_component.rs)) is stored in the file.
+
+A small `meta.json` file is in charge of keeping track of the list of segments, as well as the schema.
+
+On commit, one segment per indexing thread is written to disk, and the `meta.json` is then updated atomically.
+
+For a better idea of how indexing works, you may read the [following blog post](https://fulmicoton.com/posts/behold-tantivy-part2/).
+
+
+### Deletes
+
+Deletes happen by deleting a "term". Tantivy does not offer any notion of primary id, so it is up to the user to use a field in their schema as if it was a primary id, and delete the associated term if they want to delete only one specific document.
+
+On commit, tantivy will find all of the segments with documents matching this existing term and create a [tombstone file](src/fastfield/delete.rs) that represents the bitset of the document that are deleted.
+Like all segment files, this file is immutable. Because it is possible to have more than one tombstone file at a given instant, the tombstone filename has the format ``` segment_id . commit_opstamp . del```.
+
+An opstamp is simply an incremental id that identifies any operation applied to the index. For instance, performing a commit or adding a document.
+
+
+### DocId
+
+Within a segment, all documents are identified by a DocId that ranges within `[0, max_doc)`.
+where `max_doc` is the number of documents in the segment, (deleted or not). Having such a compact `DocId` space is key to the compression of our data structures.
+
+The DocIds are simply allocated in the order documents are added to the index.
+
+### Merges
+
+In separate threads, tantivy's index writer search for opportunities to merge segments.
+The point of segment merge is to:
+- eventually get rid of tombstoned documents
+- reduce the otherwise ever-growing number of segments.
+
+Indeed, while having several segments instead of one does not hurt search too much, having hundreds can have a measurable impact on the search performance.
+
+### Searcher
+
+The user of the library usually does not need to know about the existence of Segments.
+Searching is done through an object called a [`Searcher`](src/core/searcher.rs), that captures a
+snapshot of the index at one point of time, by holding a list of [SegmentReader](src/core/segment_reader.rs).
+
+In other words, regardless of commits, file garbage collection, or segment merge that might happen, as long as the user holds and reuse the same [Searcher](src/core/searcher.rs), search will happen on an immutable snapshot of the index.
+
+## [directory/](src/directory): Where should the data be stored?
+
+Tantivy, like Lucene, abstracts the place where the data should be stored in a key-trait
+called [`Directory`](src/directory/directory.rs).
+Contrary to Lucene however, "files" are quite different from some kind of `io::Read` object.
+Check out [`src/directory/directory.rs`](src/directory/directory.rs) trait for more details.
+
+Tantivy ships two main directory implementation: the `MMapDirectory` and the `RAMDirectory`,
+but users can extend tantivy with their own implementation.
+
+## [schema/](src/schema): What are documents?
+
+Tantivy's document follows a very strict schema, decided before building any index.
+
+The schema defines all of the fields that the indexes [`Document`](src/schema/document.rs) may and should contain, their types (`text`, `i64`, `u64`, `Date`, ...) as well as how it should be indexed / represented in tantivy.
+
+Depending on the type of the field, you can decide to
+- put it in the docstore
+- store it as a fast field
+- index it
+
+Practically, tantivy will push values associated with this type to up to 3 respective
+data structures.
+
+*Limitations*
+
+As of today, tantivy's schema imposes a 1:1 relationship between a field that is being ingested and a field represented in the search index. In sophisticated search application, it is fairly common to want to index a field twice using different tokenizers, or to index the concatenation of several fields together into one field.
+
+This is not something tantivy supports, and it is up to the user to duplicate field / concatenate fields before feeding them to tantivy.
+
+## General information about these data structures.
+
+All data structures in tantivy, have:
+- a writer
+- a serializer
+- a reader
+
+The writer builds an in-memory representation of a batch of documents. This representation is not searchable. It is just meant as an intermediary mutable representation, to which we can sequentially add
+the document of a batch. At the end of the batch (or if a memory limit is reached), this representation
+is then converted into an on-disk immutable representation, that is extremely compact.
+This conversion is done by the serializer.
+
+Finally, the reader is in charge of offering an API to read on this on-disk read-only representation.
+In tantivy, readers are designed to require very little anonymous memory. The data is read straight from an mmapped file, and loading an index is as fast as mmapping its files.
+
+## [store/](src/store): Here is my DocId, Gimme my document!
+
+The docstore is a row-oriented storage that, for each document, stores a subset of the fields
+that are marked as stored in the schema. The docstore is compressed using a general-purpose algorithm
+like LZ4.
+
+**Useful for**
+
+In search engines, it is often used to display search results.
+Once the top 10 documents have been identified, we fetch them from the store, and display them or their snippet on the search result page (aka SERP).
+
+**Not useful for**
+
+Fetching a document from the store is typically a "slow" operation. It usually consists in
+- searching into a compact tree-like data structure to find the position of the right block.
+- decompressing a small block
+- returning the document from this block.
+
+It is NOT meant to be called for every document matching a query.
+
+As a rule of thumb, if you hit the docstore more than 100 times per search query, you are probably misusing tantivy.
+
+
+## [fastfield/](src/fastfield): Here is my DocId, Gimme my value!
+
+Fast fields are stored in a column-oriented storage that allows for random access.
+The only compression applied is bitpacking. The column comes with two meta data.
+The minimum value in the column and the number of bits per doc.
+
+Fetching a value for a `DocId` is then as simple as computing
+
+```
+min_value + fetch_bits(num_bits * doc_id..num_bits * (doc_id+1))
+```
+
+This operation just requires one memory fetch.
+Because, DocSets are scanned through in order (DocId are iterated in a sorted manner) which
+also help locality.
+
+In Lucene's jargon, fast fields are called DocValues.
+
+**Useful for**
+
+They are typically integer values that are useful to either rank or compute aggregate over
+all of the documents matching a query (aka [DocSet](src/docset.rs)).
+
+For instance, one could define a function to combine upvotes with tantivy's internal relevancy score.
+This can be done by fetching a fast field during scoring.
+One could also compute the mean price of the items matching a query in an e-commerce website.
+This can be done by fetching a fast field in a collector.
+Finally one could decide to post-filter a docset to remove docset with a price within a specific range.
+If the ratio of filtered out documents is not too low, an efficient way to do this is to fetch the price and apply the filter on the collector side.
+
+Aside from integer values, it is also possible to store an actual byte payload.
+For advanced search engine, it is possible to store all of the features required for learning-to-rank in a byte payload, access it during search, and apply the learning-to-rank model.
+
+Finally facets are a specific kind of fast field, and the associated source code is in [`fastfield/facet_reader.rs`](src/fastfield/facet_reader.rs).
+
+# The inverted search index.
+
+The inverted index is the core part of full-text search.
+When presented a new document with the text field "Hello, happy tax payer!", tantivy breaks it into a list of so-called tokens. In addition to just splitting these strings into tokens, it might also do different kinds of operations like dropping the punctuation, converting the character to lowercase, apply stemming, etc. Tantivy makes it possible to configure the operations to be applied in the schema (tokenizer/ is the place where these operations are implemented).
+
+For instance, the default tokenizer of tantivy would break our text into: `[hello, happy, tax, payer]`.
+The document will therefore be registered in the inverted index as containing the terms
+`[text:hello, text:happy, text:tax, text:payer]`.
+
+The role of the inverted index is, when given a term, gives us in return a very fast iterator over the sorted doc ids that match the term.
+
+Such an iterator is called a posting list. In addition to giving us `DocId`, they can also give us optionally the number of occurrence of the term for each document, also called term frequency or TF.
+
+These iterators being sorted by DocId, one can create an iterator over the document containing `text:tax AND text:payer`, `(text:tax AND text:payer) OR (text:contribuable)` or any boolean expression.
+
+In order to represent the function
+```Term ⟶ Posting```
+
+The inverted index actually consists of two data structures chained together.
+
+- [Term](src/schema/term.rs) ⟶ [TermInfo](src/postings/term_info.rs) is addressed by the term dictionary.
+- [TermInfo](src/postings/term_info.rs) ⟶ [Posting](src/postings/postings.rs) is addressed by the posting lists.
+
+Where [TermInfo](src/postings/term_info.rs) is an object containing some meta data about a term.
+
+
+## [termdict/](src/termdict): Here is a term, give me the [TermInfo](src/postings/term_info.rs)!
+
+Tantivy's term dictionary is mainly in charge of supplying the function
+
+[Term](src/schema/term.rs) ⟶ [TermInfo](src/postings/term_info.rs)
+
+It is itself broken into two parts.
+- [Term](src/schema/term.rs) ⟶ [TermOrdinal](src/termdict/mod.rs) is addressed by a finite state transducer, implemented by the fst crate.
+- [TermOrdinal](src/termdict/mod.rs) ⟶ [TermInfo](src/postings/term_info.rs) is addressed by the term info store.
+
+
+## [postings/](src/postings): Iterate over documents... very fast!
+
+A posting list makes it possible to store a sorted list of doc ids and for each doc store
+a term frequency as well.
+
+The posting lists are stored in a separate file. The [TermInfo](src/postings/term_info.rs) contains an offset into that file and a number of documents for the given posting list. Both are required and sufficient to read the posting list.
+
+The posting list is organized in block of 128 documents.
+One block of doc ids is followed by one block of term frequencies.
+
+The doc ids are delta encoded and bitpacked.
+The term frequencies are bitpacked.
+
+Because the number of docs is rarely a multiple of 128, the last block may contain an arbitrary number of docs between 1 and 127 documents. We then use variable int encoding instead of bitpacking.
+
+## [positions/](src/positions): Where are my terms within the documents?
+
+Phrase queries make it possible to search for documents containing a specific sequence of terms.
+For instance, when the phrase query "the art of war" does not match "the war of art".
+To make it possible, it is possible to specify in the schema that a field should store positions in addition to being indexed.
+
+The token positions of all of the terms are then stored in a separate file with the extension `.pos`.
+The [TermInfo](src/postings/term_info.rs) gives an offset (expressed in position this time) in this file. As we iterate throught the docset,
+we advance the position reader by the number of term frequencies of the current document.
+
+## [fieldnorms/](src/fieldnorms): Here is my doc, how many tokens in this field?
+
+The [BM25](https://en.wikipedia.org/wiki/Okapi_BM25) formula also requires to know the number of tokens stored in a specific field for a given document. We store this information on one byte per document in the fieldnorm.
+The fieldnorm is therefore compressed. Values up to 40 are encoded unchanged.
+
+
+## [tokenizer/](src/tokenizer): How should we process text?
+
+Text processing is key to a good search experience.
+Splits or normalize your text too much, and the search results will have a less precision and a higher recall.
+Do not normalize, or under split your text, you will end up with a higher precision and a lesser recall.
+
+Text processing can be configured by selecting an off-the-shelf [`Tokenizer`](./src/tokenizer/tokenizer.rs) or implementing your own to first split the text into tokens, and then chain different [`TokenFilter`](src/tokenizer/tokenizer.rs)'s to it.
+
+Tantivy's comes with few tokenizers, but external crates are offering advanced tokenizers, such as [Lindera](https://crates.io/crates/lindera) for Japanese.
+
+
+## [query/](src/query): Define and compose queries
+
+The [Query](src/query/query.rs) trait defines what a query is.
+Due to the necessity for some queries to compute some statistics over the entire index, and because the
+index is composed of several `SegmentReader`, the path from transforming a `Query` to an iterator over documents is slightly convoluted, but fundamentally, this is what a Query is.
+
+The iterator over a document comes with some scoring function. The resulting trait is called a
+[Scorer](src/query/scorer.rs) and is specific to a segment.
+
+Different queries can be combined using the [BooleanQuery](src/query/boolean_query/).
+Tantivy comes with different types of queries and can be extended by implementing
+the `Query`, `Weight`, and `Scorer` traits.
+
+## [collector](src/collector): Define what to do with matched documents
+
+Collectors define how to aggregate the documents matching a query, in the broadest sense possible.
+The search will push matched documents one by one, calling their
+`fn collect(doc: DocId, score: Score);` method.
+
+Users may implement their own collectors by implementing the [Collector](src/collector/mod.rs) trait.
+
+## [query-grammar](query-grammar): Defines the grammar of the query parser
+
+While the [QueryParser](src/query/query_parser/query_parser.rs) struct is located in the `query/` directory, the actual parser combinator used to convert user queries into an AST is in an external crate called `query-grammar`. This part was externalized to lighten the work of the compiler.

CHANGELOG.md

@@ -1,5 +1,291 @@
+Tantivy 0.16.1
+========================
+- Major Bugfix on multivalued fastfield.  #1151
+
+Tantivy 0.16.0
+=========================
+- Bugfix in the filesum check. (@evanxg852000) #1127
+- Bugfix in positions when the index is sorted by a field. (@appaquet) #1125
+
+Tantivy 0.15.3
+=========================
+- Major bugfix. Deleting documents was broken when the index was sorted by a field. (@appaquet, @fulmicoton) #1101
+
+
+Tantivy 0.15.2
+========================
+- Major bugfix. DocStore still panics when a deleted doc is at the beginning of a block. (@appaquet) #1088
+
+Tantivy 0.15.1
+=========================
+- Major bugfix. DocStore panics when first block is deleted. (@appaquet) #1077
+
+Tantivy 0.15.0
+=========================
+- API Changes. Using Range instead of (start, end) in the API and internals (`FileSlice`, `OwnedBytes`, `Snippets`, ...)
+  This change is breaking but migration is trivial.
+- Added an Histogram collector. (@fulmicoton) #994
+- Added support for Option<TCollector>.  (@fulmicoton)
+- DocAddress is now a struct (@scampi) #987
+- Bugfix consistent tie break handling in facet's topk (@hardikpnsp) #357
+- Date field support for range queries (@rihardsk) #516
+- Added lz4-flex as the default compression scheme in tantivy (@PSeitz) #1009
+- Renamed a lot of symbols to avoid all uppercasing on acronyms, as per new clippy recommendation. For instance, RAMDirectory -> RamDirectory. (@fulmicoton)
+- Simplified positions index format (@fulmicoton) #1022
+- Moved bitpacking to bitpacker subcrate and add BlockedBitpacker, which bitpacks blocks of 128 elements (@PSeitz) #1030
+- Added support for more-like-this query in tantivy (@evanxg852000) #1011
+- Added support for sorting an index, e.g presorting documents in an index by a timestamp field. This can heavily improve performance for certain scenarios, by utilizing the sorted data (Top-n optimizations)(@PSeitz). #1026
+- Add iterator over documents in doc store (@PSeitz). #1044
+- Fix log merge policy (@PSeitz). #1043
+- Add detection to avoid small doc store blocks on merge (@PSeitz). #1054
+- Make doc store compression dynamic (@PSeitz). #1060
+- Switch to json for footer version handling (@PSeitz). #1060
+- Updated TermMerger implementation to rely on the union feature of the FST (@scampi) #469
+- Add boolean marking whether position is required in the query_terms API call (@fulmicoton). #1070
+
+
+Tantivy 0.14.0
+=========================
+- Remove dependency to atomicwrites #833 .Implemented by @fulmicoton upon suggestion and research from @asafigan).
+- Migrated tantivy error from the now deprecated `failure` crate to `thiserror` #760. (@hirevo)
+- API Change. Accessing the typed value off a `Schema::Value` now returns an Option instead of panicking if the type does not match.
+- Large API Change in the Directory API. Tantivy used to assume that all files could be somehow memory mapped. After this change, Directory return a `FileSlice` that can be reduced and eventually read into an `OwnedBytes` object. Long and blocking io operation are still required by they do not span over the entire file.
+- Added support for Brotli compression in the DocStore. (@ppodolsky)
+- Added helper for building intersections and unions in BooleanQuery (@guilload)
+- Bugfix in `Query::explain`
+- Removed dependency on `notify` #924. Replaced with `FileWatcher` struct that polls meta file every 500ms in background thread. (@halvorboe @guilload)
+- Added `FilterCollector`, which wraps another collector and filters docs using a predicate over a fast field (@barrotsteindev)
+- Simplified the encoding of the skip reader struct. BlockWAND max tf is now encoded over a single byte. (@fulmicoton)
+- `FilterCollector` now supports all Fast Field value types (@barrotsteindev)
+- FastField are not all loaded when opening the segment reader. (@fulmicoton)
+- Added an API to merge segments, see `tantivy::merge_segments` #1005. (@evanxg852000)
+
+This version breaks compatibility and requires users to reindex everything.
+
+Tantivy 0.13.2
+===================
+Bugfix. Acquiring a facet reader on a segment that does not contain any
+doc with this facet returns `None`. (#896)
+
+Tantivy 0.13.1
+===================
+Made `Query` and `Collector` `Send + Sync`.
+Updated misc dependency versions.
+
+Tantivy 0.13.0
+======================
+Tantivy 0.13 introduce a change in the index format that will require
+you to reindex your index (BlockWAND information are added in the skiplist).
+The index size increase is minor as this information is only added for
+full blocks.
+If you have a massive index for which reindexing is not an option, please contact me
+so that we can discuss possible solutions.
+
+- Bugfix in `FuzzyTermQuery` not matching terms by prefix when it should (@Peachball)
+- Relaxed constraints on the custom/tweak score functions. At the segment level, they can be mut, and they are not required to be Sync + Send.
+- `MMapDirectory::open` does not return a `Result` anymore.
+- Change in the DocSet and Scorer API. (@fulmicoton).
+A freshly created DocSet point directly to their first doc. A sentinel value called TERMINATED marks the end of a DocSet.
+`.advance()` returns the new DocId. `Scorer::skip(target)` has been replaced by `Scorer::seek(target)` and returns the resulting DocId.
+As a result, iterating through DocSet now looks as follows
+```rust
+let mut doc = docset.doc();
+while doc != TERMINATED {
+   // ...
+   doc = docset.advance();
+}
+```
+The change made it possible to greatly simplify a lot of the docset's code.
+- Misc internal optimization and introduction of the `Scorer::for_each_pruning` function. (@fulmicoton)
+- Added an offset option to the Top(.*)Collectors. (@robyoung)
+- Added Block WAND. Performance on TOP-K on term-unions should be greatly increased. (@fulmicoton, and special thanks
+to the PISA team for answering all my questions!)
+
+Tantivy 0.12.0
+======================
+- Removing static dispatch in tokenizers for simplicity. (#762)
+- Added backward iteration for `TermDictionary` stream. (@halvorboe)
+- Fixed a performance issue when searching for the posting lists of a missing term (@audunhalland)
+- Added a configurable maximum number of docs (10M by default) for a segment to be considered for merge (@hntd187, landed by @halvorboe #713)
+- Important Bugfix #777, causing tantivy to retain memory mapping. (diagnosed by @poljar)
+- Added support for field boosting. (#547, @fulmicoton)
+
+## How to update?
+
+Crates relying on custom tokenizer, or registering tokenizer in the manager will require some
+minor changes. Check https://github.com/tantivy-search/tantivy/blob/main/examples/custom_tokenizer.rs
+to check for some code sample.
+
+Tantivy 0.11.3
+=======================
+- Fixed DateTime as a fast field (#735)
+
+Tantivy 0.11.2
+=======================
+- The future returned by `IndexWriter::merge` does not borrow `self` mutably anymore (#732)
+- Exposing a constructor for `WatchHandle` (#731)
+
+Tantivy 0.11.1
+=====================
+- Bug fix #729
+
+
+Tantivy 0.11.0
+=====================
+
+- Added f64 field. Internally reuse u64 code the same way i64 does (@fdb-hiroshima)
+- Various bugfixes in the query parser.
+    - Better handling of hyphens in query parser. (#609)
+    - Better handling of whitespaces.
+- Closes #498 - add support for Elastic-style unbounded range queries for alphanumeric types eg. "title:>hello", "weight:>=70.5", "height:<200" (@petr-tik)
+- API change around `Box<BoxableTokenizer>`. See detail in #629
+- Avoid rebuilding Regex automaton whenever a regex query is reused. #639 (@brainlock)
+- Add footer with some metadata to index files. #605 (@fdb-hiroshima)
+- Add a method to check the compatibility of the footer in the index with the running version of tantivy (@petr-tik)
+- TopDocs collector: ensure stable sorting on equal score. #671 (@brainlock)
+- Added handling of pre-tokenized text fields (#642), which will enable users to
+  load tokens created outside tantivy. See usage in examples/pre_tokenized_text. (@kkoziara)
+- Fix crash when committing multiple times with deleted documents. #681 (@brainlock)
+
+## How to update?
+
+- The index format is changed. You are required to reindex your data to use tantivy 0.11.
+- `Box<dyn BoxableTokenizer>` has been replaced by a `BoxedTokenizer` struct.
+- Regex are now compiled when the `RegexQuery` instance is built. As a result, it can now return
+an error and handling the `Result` is required.
+- `tantivy::version()` now returns a `Version` object. This object implements `ToString()`
+
+Tantivy 0.10.2
+=====================
+
+- Closes #656. Solving memory leak.
+
+Tantivy 0.10.1
+=====================
+
+- Closes #544.  A few users experienced problems with the directory watching system.
+Avoid watching the mmap directory until someone effectively creates a reader that uses
+this functionality.
+
+
+Tantivy 0.10.0
+=====================
+
+*Tantivy 0.10.0 index format is compatible with the index format in 0.9.0.*
+
+- Added an API to easily tweak or entirely replace the
+ default score. See `TopDocs::tweak_score`and `TopScore::custom_score` (@fulmicoton)
+- Added an ASCII folding filter (@drusellers)
+- Bugfix in `query.count` in presence of deletes (@fulmicoton)
+- Added `.explain(...)` in `Query` and `Weight` to (@fulmicoton)
+- Added an efficient way to `delete_all_documents` in `IndexWriter` (@petr-tik).
+  All segments are simply removed.
+
+Minor
+---------
+- Switched to Rust 2018 (@uvd)
+- Small simplification of the code.
+Calling .freq() or .doc() when .advance() has never been called
+on segment postings should panic from now on.
+- Tokens exceeding `u16::max_value() - 4` chars are discarded silently instead of panicking.
+- Fast fields are now preloaded when the `SegmentReader` is created.
+- `IndexMeta` is now public.  (@hntd187)
+- `IndexWriter` `add_document`, `delete_term`. `IndexWriter` is `Sync`, making it possible to use it with a `
+Arc<RwLock<IndexWriter>>`. `add_document` and `delete_term` can
+only require a read lock. (@fulmicoton)
+- Introducing `Opstamp` as an expressive type alias for `u64`. (@petr-tik)
+- Stamper now relies on `AtomicU64` on all platforms (@petr-tik)
+- Bugfix - Files get deleted slightly earlier
+- Compilation resources improved (@fdb-hiroshima)
+
+## How to update?
+
+Your program should be usable as is.
+
+### Fast fields
+
+Fast fields used to be accessed directly from the `SegmentReader`.
+The API changed, you are now required to acquire your fast field reader via the
+`segment_reader.fast_fields()`, and use one of the typed method:
+- `.u64()`, `.i64()` if your field is single-valued ;
+- `.u64s()`, `.i64s()` if your field is multi-valued ;
+- `.bytes()` if your field is bytes fast field.
+
+
+
+Tantivy 0.9.0
+=====================
+*0.9.0 index format is not compatible with the
+previous index format.*
+- MAJOR BUGFIX :
+  Some `Mmap` objects were being leaked, and would never get released. (@fulmicoton)
+- Removed most unsafe (@fulmicoton)
+- Indexer memory footprint improved. (VInt comp, inlining the first block. (@fulmicoton)
+- Stemming in other language possible (@pentlander)
+- Segments with no docs are deleted earlier (@barrotsteindev)
+- Added grouped add and delete operations.
+  They are guaranteed to happen together (i.e. they cannot be split by a commit).
+  In addition, adds are guaranteed to happen on the same segment. (@elbow-jason)
+- Removed `INT_STORED` and `INT_INDEXED`. It is now possible to use `STORED` and `INDEXED`
+  for int fields. (@fulmicoton)
+- Added DateTime field (@barrotsteindev)
+- Added IndexReader. By default, index is reloaded automatically upon new commits (@fulmicoton)
+- SIMD linear search within blocks (@fulmicoton)
+
+## How to update ?
+
+tantivy 0.9 brought some API breaking change.
+To update from tantivy 0.8, you will need to go through the following steps.
+
+- `schema::INT_INDEXED` and `schema::INT_STORED`  should be replaced by `schema::INDEXED` and `schema::INT_STORED`.
+- The index now does not hold the pool of searcher anymore. You are required to create an intermediary object called
+`IndexReader` for this.
+
+    ```rust
+    // create the reader. You typically need to create 1 reader for the entire
+    // lifetime of you program.
+    let reader = index.reader()?;
+
+    // Acquire a searcher (previously `index.searcher()`) is now written:
+    let searcher = reader.searcher();
+
+    // With the default setting of the reader, you are not required to
+    // call `index.load_searchers()` anymore.
+    //
+    // The IndexReader will pick up that change automatically, regardless
+    // of whether the update was done in a different process or not.
+    // If this behavior is not wanted, you can create your reader with
+    // the `ReloadPolicy::Manual`, and manually decide when to reload the index
+    // by calling `reader.reload()?`.
+
+    ```
+
+
+Tantivy 0.8.2
+=====================
+Fixing build for x86_64 platforms. (#496)
+No need to update from 0.8.1 if tantivy
+is building on your platform.
+
+
+Tantivy 0.8.1
+=====================
+Hotfix of #476.
+
+Merge was reflecting deletes before commit was passed.
+Thanks @barrotsteindev  for reporting the bug.
+
+
+Tantivy 0.8.0
+=====================
+*No change in the index format*
+- API Breaking change in the collector API. (@jwolfe, @fulmicoton)
+- Multithreaded search (@jwolfe, @fulmicoton)
+
+
 Tantivy 0.7.1
 =====================
+*No change in the index format*
 - Bugfix: NGramTokenizer panics on non ascii chars
 - Added a space usage API
 
@@ -22,7 +308,7 @@ Tantivy 0.6.1
         - Exclusive `field:{startExcl to endExcl}`
         - Mixed `field:[startIncl to endExcl}` and vice versa
         - Unbounded `field:[start to *]`, `field:[* to end]`
- 
+
 
 Tantivy 0.6
 ==========================
@@ -30,10 +316,10 @@ Tantivy 0.6
 Special thanks to @drusellers and @jason-wolfe for their contributions
 to this release!
 
-- Removed C code. Tantivy is now pure Rust. (@pmasurel)
-- BM25 (@pmasurel)
-- Approximate field norms encoded over 1 byte. (@pmasurel)
-- Compiles on stable rust (@pmasurel)
+- Removed C code. Tantivy is now pure Rust. (@fulmicoton)
+- BM25 (@fulmicoton)
+- Approximate field norms encoded over 1 byte. (@fulmicoton)
+- Compiles on stable rust (@fulmicoton)
 - Add &[u8] fastfield for associating arbitrary bytes to each document (@jason-wolfe) (#270)
     - Completely uncompressed
     - Internally: One u64 fast field for indexes, one fast field for the bytes themselves.
@@ -41,7 +327,7 @@ to this release!
 - Add Stopword Filter support (@drusellers)
 - Add a FuzzyTermQuery (@drusellers)
 - Add a RegexQuery (@drusellers)
-- Various performance improvements (@pmasurel)_
+- Various performance improvements (@fulmicoton)_
 
 
 Tantivy 0.5.2

Cargo.toml

@@ -1,61 +1,78 @@
 [package]
 name = "tantivy"
-version = "0.7.1"
+version = "0.16.1"
 authors = ["Paul Masurel <paul.masurel@gmail.com>"]
 license = "MIT"
 categories = ["database-implementations", "data-structures"]
 description = """Search engine library"""
-documentation = "https://tantivy-search.github.io/tantivy/tantivy/index.html"
+documentation = "https://docs.rs/tantivy/"
 homepage = "https://github.com/tantivy-search/tantivy"
 repository = "https://github.com/tantivy-search/tantivy"
 readme = "README.md"
 keywords = ["search", "information", "retrieval"]
+edition = "2018"
 
 [dependencies]
-base64 = "0.10.0"
-byteorder = "1.0"
-lazy_static = "1"
-regex = "1.0"
-fst = {version="0.3", default-features=false}
-fst-regex = { version="0.2" }
-lz4 = {version="1.20", optional=true}
-snap = {version="0.2"}
-atomicwrites = {version="0.2.2", optional=true}
-tempfile = "3.0"
-log = "0.4"
-combine = "3"
-tempdir = "0.3"
-serde = "1.0"
-serde_derive = "1.0"
-serde_json = "1.0"
-num_cpus = "1.2"
-itertools = "0.7"
-levenshtein_automata = {version="0.1", features=["fst_automaton"]}
-bit-set = "0.5"
-uuid = { version = "0.7", features = ["v4", "serde"] }
-crossbeam = "0.4"
-crossbeam-channel = "0.2"
-futures = "0.1"
-futures-cpupool = "0.1"
-owning_ref = "0.4"
-stable_deref_trait = "1.0.0"
-rust-stemmers = "1"
-downcast = { version="0.9" }
-matches = "0.1"
-bitpacking = "0.5"
-census = "0.1"
-fnv = "1.0.6"
-owned-read = "0.4"
-failure = "0.1"
+base64 = "0.13"
+byteorder = "1.4.3"
+crc32fast = "1.2.1"
+once_cell = "1.7.2"
+regex ={ version = "1.5.4", default-features = false, features = ["std"] }
+tantivy-fst = "0.3"
+memmap2 = {version = "0.5", optional=true}
+lz4_flex = { version = "0.9.0", default-features = false, features = ["checked-decode"], optional = true }
+brotli = { version = "3.3", optional = true }
+snap = { version = "1.0.5", optional = true }
+tempfile = { version = "3.2", optional = true }
+log = "0.4.14"
+serde = { version = "1.0.126", features = ["derive"] }
+serde_closure = "0.3"
+serde_json = "1.0.64"
+num_cpus = "1.13"
+fs2={ version = "0.4.3", optional = true }
+levenshtein_automata = "0.2"
+uuid = { version = "0.8.2", features = ["v4", "serde"] }
+crossbeam = "0.8.1"
+futures = { version = "0.3.15", features = ["thread-pool"] }
+tantivy-query-grammar = { version="0.15.0", path="./query-grammar" }
+tantivy-bitpacker = { version="0.1", path="./bitpacker" }
+common = { version = "0.1", path = "./common/", package = "tantivy-common" }
+fastfield_codecs = { version="0.1", path="./fastfield_codecs", default-features = false }
+ownedbytes = { version="0.1", path="./ownedbytes" }
+stable_deref_trait = "1.2"
+rust-stemmers = "1.2"
+downcast-rs = "1.2"
+bitpacking = { version = "0.8.4", default-features = false, features = ["bitpacker4x"] }
+census = "0.4"
+fnv = "1.0.7"
+thiserror = "1.0.24"
 htmlescape = "0.3.1"
-fail = "0.2"
+fail = "0.4"
+murmurhash32 = "0.2"
+chrono = "0.4.19"
+smallvec = "1.6.1"
+lru = "0.7.0"
+fastdivide = "0.3"
+itertools = "0.10.0"
+measure_time = "0.7.0"
+wasm-mt = "0.1"
+wasm-mt-pool = "0.1"
 
 [target.'cfg(windows)'.dependencies]
-winapi = "0.2"
+winapi = "0.3.9"
 
 [dev-dependencies]
-rand = "0.5"
-maplit = "1"
+rand = "0.8.3"
+maplit = "1.0.2"
+matches = "0.1.8"
+proptest = "1.0"
+criterion = "0.3.5"
+test-env-log = "0.2.7"
+env_logger = "0.9.0"
+
+[dev-dependencies.fail]
+version = "0.4"
+features = ["failpoints"]
 
 [profile.release]
 opt-level = 3
@@ -67,14 +84,35 @@ debug-assertions = true
 overflow-checks = true
 
 [features]
-# by default no-fail is disabled. We manually enable it when running test.
-default = ["mmap", "no_fail"]
-mmap = ["fst/mmap", "atomicwrites"]
-lz4-compression = ["lz4"]
-no_fail = ["fail/no_fail"]
+default = ["mmap", "lz4-compression" ]
+mmap = ["fs2", "tempfile", "memmap2"]
+
+brotli-compression = ["brotli"]
+lz4-compression = ["lz4_flex"]
+snappy-compression = ["snap"]
+
+failpoints = ["fail/failpoints"]
 unstable = [] # useful for benches.
+wasm-bindgen = ["uuid/wasm-bindgen"]
+
+[workspace]
+members = ["query-grammar", "bitpacker", "common", "fastfield_codecs", "ownedbytes"]
 
 [badges]
 travis-ci = { repository = "tantivy-search/tantivy" }
 
+# Following the "fail" crate best practises, we isolate
+# tests that define specific behavior in fail check points
+# in a different binary.
+#
+# We do that because, fail rely on a global definition of
+# failpoints behavior and hence, it is incompatible with
+# multithreading.
+[[test]]
+name = "failpoints"
+path = "tests/failpoints/mod.rs"
+required-features = ["fail/failpoints"]
 
+[[bench]]
+name = "analyzer"
+harness = false

Makefile

@@ -0,0 +1,3 @@
+test:
+	echo "Run test only... No examples."
+	cargo test --tests --lib

README.md

@@ -1,10 +1,10 @@
 
-[![Build Status](https://travis-ci.org/tantivy-search/tantivy.svg?branch=master)](https://travis-ci.org/tantivy-search/tantivy)
-[![codecov](https://codecov.io/gh/tantivy-search/tantivy/branch/master/graph/badge.svg)](https://codecov.io/gh/tantivy-search/tantivy)
+[![Docs](https://docs.rs/tantivy/badge.svg)](https://docs.rs/crate/tantivy/)
+[![Build Status](https://github.com/tantivy-search/tantivy/actions/workflows/test.yml/badge.svg)](https://github.com/tantivy-search/tantivy/actions/workflows/test.yml)
+[![codecov](https://codecov.io/gh/tantivy-search/tantivy/branch/main/graph/badge.svg)](https://codecov.io/gh/tantivy-search/tantivy)
 [![Join the chat at https://gitter.im/tantivy-search/tantivy](https://badges.gitter.im/tantivy-search/tantivy.svg)](https://gitter.im/tantivy-search/tantivy?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge&utm_content=badge)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
-[![Build status](https://ci.appveyor.com/api/projects/status/r7nb13kj23u8m9pj/branch/master?svg=true)](https://ci.appveyor.com/project/fulmicoton/tantivy/branch/master)
-[![Say Thanks!](https://img.shields.io/badge/Say%20Thanks-!-1EAEDB.svg)](https://saythanks.io/to/fulmicoton)
+[![Crates.io](https://img.shields.io/crates/v/tantivy.svg)](https://crates.io/crates/tantivy)
 
 ![Tantivy](https://tantivy-search.github.io/logo/tantivy-logo.png)
 
@@ -17,74 +17,118 @@
 [![](https://sourcerer.io/fame/fulmicoton/tantivy-search/tantivy/images/6)](https://sourcerer.io/fame/fulmicoton/tantivy-search/tantivy/links/6)
 [![](https://sourcerer.io/fame/fulmicoton/tantivy-search/tantivy/images/7)](https://sourcerer.io/fame/fulmicoton/tantivy-search/tantivy/links/7)
 
+[![Become a patron](https://c5.patreon.com/external/logo/become_a_patron_button.png)](https://www.patreon.com/fulmicoton)
 
 
-**Tantivy** is a **full text search engine library** written in rust.
+**Tantivy** is a **full text search engine library** written in Rust.
 
-It is closer to [Apache Lucene](https://lucene.apache.org/) than to [Elastic Search](https://www.elastic.co/products/elasticsearch) and [Apache Solr](https://lucene.apache.org/solr/) in the sense it is not
+It is closer to [Apache Lucene](https://lucene.apache.org/) than to [Elasticsearch](https://www.elastic.co/products/elasticsearch) or [Apache Solr](https://lucene.apache.org/solr/) in the sense it is not
 an off-the-shelf search engine server, but rather a crate that can be used
 to build such a search engine.
 
 Tantivy is, in fact, strongly inspired by Lucene's design.
 
+# Benchmark
+
+The following [benchmark](https://tantivy-search.github.io/bench/) break downs 
+performance for different type of queries / collection.
+
+Your mileage WILL vary depending on the nature of queries and their load.
+
 # Features
 
 - Full-text search
+- Configurable tokenizer (stemming available for 17 Latin languages with third party support for Chinese ([tantivy-jieba](https://crates.io/crates/tantivy-jieba) and [cang-jie](https://crates.io/crates/cang-jie)), Japanese ([lindera](https://github.com/lindera-morphology/lindera-tantivy) and [tantivy-tokenizer-tiny-segmenter](https://crates.io/crates/tantivy-tokenizer-tiny-segmenter)) and Korean ([lindera](https://github.com/lindera-morphology/lindera-tantivy) + [lindera-ko-dic-builder](https://github.com/lindera-morphology/lindera-ko-dic-builder))
 - Fast (check out the :racehorse: :sparkles: [benchmark](https://tantivy-search.github.io/bench/) :sparkles: :racehorse:)
 - Tiny startup time (<10ms), perfect for command line tools
-- BM25 scoring (the same as lucene)
-- Natural query language `(michael AND jackson) OR "king of pop"`
-- Phrase queries search (`"michael jackson"`)
+- BM25 scoring (the same as Lucene)
+- Natural query language (e.g. `(michael AND jackson) OR "king of pop"`)
+- Phrase queries search (e.g. `"michael jackson"`)
 - Incremental indexing
 - Multithreaded indexing (indexing English Wikipedia takes < 3 minutes on my desktop)
 - Mmap directory
-- SIMD integer compression when the platform/CPU includes the SSE2 instruction set.
-- Single valued and multivalued u64 and i64 fast fields (equivalent of doc values in Lucene)
+- SIMD integer compression when the platform/CPU includes the SSE2 instruction set
+- Single valued and multivalued u64, i64, and f64 fast fields (equivalent of doc values in Lucene)
 - `&[u8]` fast fields
+- Text, i64, u64, f64, dates, and hierarchical facet fields
 - LZ4 compressed document store
 - Range queries
 - Faceted search
 - Configurable indexing (optional term frequency and position indexing)
 - Cheesy logo with a horse
 
-# Non-features
+## Non-features
 
-- Distributed search is out of the scope of tantivy. That being said, tantivy is meant as a
+- Distributed search is out of the scope of Tantivy. That being said, Tantivy is a
 library upon which one could build a distributed search. Serializable/mergeable collector state for instance, 
-are within the scope of tantivy.
+are within the scope of Tantivy.
 
 
-# Supported OS and compiler
-
-Tantivy works on stable rust (>= 1.27) and supports Linux, MacOS and Windows.
-
 # Getting started
 
-- [tantivy's simple search example](http://fulmicoton.com/tantivy-examples/simple_search.html)
-- [tantivy-cli and its tutorial](https://github.com/tantivy-search/tantivy-cli).
-`tantivy-cli` is an actual command line interface that makes it easy for you to create a search engine,
-index documents and search via the CLI or a small server with a REST API.
-It will walk you through getting a wikipedia search engine up and running in a few minutes.
-- [reference doc]
-    - [For the last released version](https://docs.rs/tantivy/)
-    - [For the last master branch](https://tantivy-search.github.io/tantivy/tantivy/index.html)
+Tantivy works on stable Rust (>= 1.27) and supports Linux, MacOS, and Windows.
 
-# Compiling
+- [Tantivy's simple search example](https://tantivy-search.github.io/examples/basic_search.html)
+- [tantivy-cli and its tutorial](https://github.com/tantivy-search/tantivy-cli) - `tantivy-cli` is an actual command line interface that makes it easy for you to create a search engine,
+index documents, and search via the CLI or a small server with a REST API.
+It walks you through getting a wikipedia search engine up and running in a few minutes.
+- [Reference doc for the last released version](https://docs.rs/tantivy/)
 
-## Development
+# How can I support this project?
 
-Tantivy compiles on stable rust but requires `Rust >= 1.27`.
-To check out and run tests, you can simply run :
+There are many ways to support this project. 
 
-    git clone git@github.com:tantivy-search/tantivy.git
+- Use Tantivy and tell us about your experience on [Gitter](https://gitter.im/tantivy-search/tantivy) or by email (paul.masurel@gmail.com)
+- Report bugs
+- Write a blog post
+- Help with documentation by asking questions or submitting PRs
+- Contribute code (you can join [our Gitter](https://gitter.im/tantivy-search/tantivy))
+- Talk about Tantivy around you
+- [![Become a patron](https://c5.patreon.com/external/logo/become_a_patron_button.png)](https://www.patreon.com/fulmicoton)
+
+# Contributing code
+
+We use the GitHub Pull Request workflow: reference a GitHub ticket and/or include a comprehensive commit message when opening a PR.
+
+## Clone and build locally
+
+Tantivy compiles on stable Rust but requires `Rust >= 1.27`.
+To check out and run tests, you can simply run:
+
+```bash
+    git clone https://github.com/tantivy-search/tantivy.git
     cd tantivy
     cargo build
+```
 
-## Running tests
+## Run tests
 
 Some tests will not run with just `cargo test` because of `fail-rs`.
-To run the tests exhaustively, run `./run-tests.sh`. 
+To run the tests exhaustively, run `./run-tests.sh`.
 
-# Contribute
+## Debug
 
-Send me an email (paul.masurel at gmail.com) if you want to contribute to tantivy.
+You might find it useful to step through the programme with a debugger.
+
+### A failing test
+
+Make sure you haven't run `cargo clean` after the most recent `cargo test` or `cargo build` to guarantee that the `target/` directory exists. Use this bash script to find the name of the most recent debug build of Tantivy and run it under `rust-gdb`:
+
+```bash
+find target/debug/ -maxdepth 1 -executable -type f -name "tantivy*" -printf '%TY-%Tm-%Td %TT %p\n' | sort -r | cut -d " " -f 3 | xargs -I RECENT_DBG_TANTIVY rust-gdb RECENT_DBG_TANTIVY
+```
+
+Now that you are in `rust-gdb`, you can set breakpoints on lines and methods that match your source code and run the debug executable with flags that you normally pass to `cargo test` like this:
+
+```bash
+$gdb run --test-threads 1 --test $NAME_OF_TEST
+```
+
+### An example
+
+By default, `rustc` compiles everything in the `examples/` directory in debug mode. This makes it easy for you to make examples to reproduce bugs:
+
+```bash
+rust-gdb target/debug/examples/$EXAMPLE_NAME
+$ gdb run
+```

appveyor.yml

@@ -18,5 +18,6 @@ install:
 build: false
 
 test_script:
-  - REM SET RUST_LOG=tantivy,test & cargo test --verbose --no-default-features --features mmap -- --test-threads 1
+  - REM SET RUST_LOG=tantivy,test & cargo test --all --verbose --no-default-features --features lz4-compression --features mmap
+  - REM SET RUST_LOG=tantivy,test & cargo test test_store --verbose --no-default-features --features lz4-compression --features snappy-compression --features brotli-compression --features mmap
   - REM SET RUST_BACKTRACE=1 & cargo build --examples

benches/analyzer.rs

@@ -0,0 +1,22 @@
+use criterion::{criterion_group, criterion_main, Criterion};
+use tantivy::tokenizer::TokenizerManager;
+
+const ALICE_TXT: &str = include_str!("alice.txt");
+
+pub fn criterion_benchmark(c: &mut Criterion) {
+    let tokenizer_manager = TokenizerManager::default();
+    let tokenizer = tokenizer_manager.get("default").unwrap();
+    c.bench_function("default-tokenize-alice", |b| {
+        b.iter(|| {
+            let mut word_count = 0;
+            let mut token_stream = tokenizer.token_stream(ALICE_TXT);
+            while token_stream.advance() {
+                word_count += 1;
+            }
+            assert_eq!(word_count, 30_731);
+        })
+    });
+}
+
+criterion_group!(benches, criterion_benchmark);
+criterion_main!(benches);

bitpacker/Cargo.toml

@@ -0,0 +1,15 @@
+[package]
+name = "tantivy-bitpacker"
+version = "0.1.1"
+edition = "2018"
+authors = ["Paul Masurel <paul.masurel@gmail.com>"]
+license = "MIT"
+categories = []
+description = """Tantivy-sub crate: bitpacking"""
+repository = "https://github.com/tantivy-search/tantivy"
+keywords = []
+
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]

bitpacker/benches/bench.rs

@@ -0,0 +1,33 @@
+#![feature(test)]
+
+extern crate test;
+
+#[cfg(test)]
+mod tests {
+    use tantivy_bitpacker::BlockedBitpacker;
+    use test::Bencher;
+    #[bench]
+    fn bench_blockedbitp_read(b: &mut Bencher) {
+        let mut blocked_bitpacker = BlockedBitpacker::new();
+        for val in 0..=21500 {
+            blocked_bitpacker.add(val * val);
+        }
+        b.iter(|| {
+            let mut out = 0;
+            for val in 0..=21500 {
+                out = blocked_bitpacker.get(val);
+            }
+            out
+        });
+    }
+    #[bench]
+    fn bench_blockedbitp_create(b: &mut Bencher) {
+        b.iter(|| {
+            let mut blocked_bitpacker = BlockedBitpacker::new();
+            for val in 0..=21500 {
+                blocked_bitpacker.add(val * val);
+            }
+            blocked_bitpacker
+        });
+    }
+}

bitpacker/src/bitpacker.rs

@@ -0,0 +1,142 @@
+use std::{convert::TryInto, io};
+
+pub struct BitPacker {
+    mini_buffer: u64,
+    mini_buffer_written: usize,
+}
+impl Default for BitPacker {
+    fn default() -> Self {
+        BitPacker::new()
+    }
+}
+impl BitPacker {
+    pub fn new() -> BitPacker {
+        BitPacker {
+            mini_buffer: 0u64,
+            mini_buffer_written: 0,
+        }
+    }
+
+    #[inline]
+    pub fn write<TWrite: io::Write>(
+        &mut self,
+        val: u64,
+        num_bits: u8,
+        output: &mut TWrite,
+    ) -> io::Result<()> {
+        let val_u64 = val as u64;
+        let num_bits = num_bits as usize;
+        if self.mini_buffer_written + num_bits > 64 {
+            self.mini_buffer |= val_u64.wrapping_shl(self.mini_buffer_written as u32);
+            output.write_all(self.mini_buffer.to_le_bytes().as_ref())?;
+            self.mini_buffer = val_u64.wrapping_shr((64 - self.mini_buffer_written) as u32);
+            self.mini_buffer_written = self.mini_buffer_written + num_bits - 64;
+        } else {
+            self.mini_buffer |= val_u64 << self.mini_buffer_written;
+            self.mini_buffer_written += num_bits;
+            if self.mini_buffer_written == 64 {
+                output.write_all(self.mini_buffer.to_le_bytes().as_ref())?;
+                self.mini_buffer_written = 0;
+                self.mini_buffer = 0u64;
+            }
+        }
+        Ok(())
+    }
+
+    pub fn flush<TWrite: io::Write>(&mut self, output: &mut TWrite) -> io::Result<()> {
+        if self.mini_buffer_written > 0 {
+            let num_bytes = (self.mini_buffer_written + 7) / 8;
+            let bytes = self.mini_buffer.to_le_bytes();
+            output.write_all(&bytes[..num_bytes])?;
+            self.mini_buffer_written = 0;
+            self.mini_buffer = 0;
+        }
+        Ok(())
+    }
+
+    pub fn close<TWrite: io::Write>(&mut self, output: &mut TWrite) -> io::Result<()> {
+        self.flush(output)?;
+        // Padding the write file to simplify reads.
+        output.write_all(&[0u8; 7])?;
+        Ok(())
+    }
+}
+
+#[derive(Clone, Debug, Default)]
+pub struct BitUnpacker {
+    num_bits: u64,
+    mask: u64,
+}
+
+impl BitUnpacker {
+    pub fn new(num_bits: u8) -> BitUnpacker {
+        let mask: u64 = if num_bits == 64 {
+            !0u64
+        } else {
+            (1u64 << num_bits) - 1u64
+        };
+        BitUnpacker {
+            num_bits: u64::from(num_bits),
+            mask,
+        }
+    }
+
+    #[inline]
+    pub fn get(&self, idx: u64, data: &[u8]) -> u64 {
+        if self.num_bits == 0 {
+            return 0u64;
+        }
+        let num_bits = self.num_bits;
+        let mask = self.mask;
+        let addr_in_bits = idx * num_bits;
+        let addr = addr_in_bits >> 3;
+        let bit_shift = addr_in_bits & 7;
+        debug_assert!(
+            addr + 8 <= data.len() as u64,
+            "The fast field field should have been padded with 7 bytes."
+        );
+        let bytes: [u8; 8] = (&data[(addr as usize)..(addr as usize) + 8])
+            .try_into()
+            .unwrap();
+        let val_unshifted_unmasked: u64 = u64::from_le_bytes(bytes);
+        let val_shifted = (val_unshifted_unmasked >> bit_shift) as u64;
+        val_shifted & mask
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::{BitPacker, BitUnpacker};
+
+    fn create_fastfield_bitpacker(len: usize, num_bits: u8) -> (BitUnpacker, Vec<u64>, Vec<u8>) {
+        let mut data = Vec::new();
+        let mut bitpacker = BitPacker::new();
+        let max_val: u64 = (1u64 << num_bits as u64) - 1u64;
+        let vals: Vec<u64> = (0u64..len as u64)
+            .map(|i| if max_val == 0 { 0 } else { i % max_val })
+            .collect();
+        for &val in &vals {
+            bitpacker.write(val, num_bits, &mut data).unwrap();
+        }
+        bitpacker.close(&mut data).unwrap();
+        assert_eq!(data.len(), ((num_bits as usize) * len + 7) / 8 + 7);
+        let bitunpacker = BitUnpacker::new(num_bits);
+        (bitunpacker, vals, data)
+    }
+
+    fn test_bitpacker_util(len: usize, num_bits: u8) {
+        let (bitunpacker, vals, data) = create_fastfield_bitpacker(len, num_bits);
+        for (i, val) in vals.iter().enumerate() {
+            assert_eq!(bitunpacker.get(i as u64, &data), *val);
+        }
+    }
+
+    #[test]
+    fn test_bitpacker() {
+        test_bitpacker_util(10, 3);
+        test_bitpacker_util(10, 0);
+        test_bitpacker_util(10, 1);
+        test_bitpacker_util(6, 14);
+        test_bitpacker_util(1000, 14);
+    }
+}

bitpacker/src/blocked_bitpacker.rs

@@ -0,0 +1,178 @@
+use crate::{minmax, BitUnpacker};
+
+use super::{bitpacker::BitPacker, compute_num_bits};
+
+const BLOCK_SIZE: usize = 128;
+
+/// `BlockedBitpacker` compresses data in blocks of
+/// 128 elements, while keeping an index on it
+///
+#[derive(Debug, Clone)]
+pub struct BlockedBitpacker {
+    // bitpacked blocks
+    compressed_blocks: Vec<u8>,
+    // uncompressed data, collected until BLOCK_SIZE
+    buffer: Vec<u64>,
+    offset_and_bits: Vec<BlockedBitpackerEntryMetaData>,
+}
+impl Default for BlockedBitpacker {
+    fn default() -> Self {
+        BlockedBitpacker::new()
+    }
+}
+
+/// `BlockedBitpackerEntryMetaData` encodes the
+/// offset and bit_width into a u64 bit field
+///
+/// This saves some space, since 7byte is more
+/// than enough and also keeps the access fast
+/// because of alignment
+#[derive(Debug, Clone, Default)]
+struct BlockedBitpackerEntryMetaData {
+    encoded: u64,
+    base_value: u64,
+}
+
+impl BlockedBitpackerEntryMetaData {
+    fn new(offset: u64, num_bits: u8, base_value: u64) -> Self {
+        let encoded = offset | (num_bits as u64) << (64 - 8);
+        Self {
+            encoded,
+            base_value,
+        }
+    }
+    fn offset(&self) -> u64 {
+        (self.encoded << 8) >> 8
+    }
+    fn num_bits(&self) -> u8 {
+        (self.encoded >> 56) as u8
+    }
+    fn base_value(&self) -> u64 {
+        self.base_value
+    }
+}
+
+#[test]
+fn metadata_test() {
+    let meta = BlockedBitpackerEntryMetaData::new(50000, 6, 40000);
+    assert_eq!(meta.offset(), 50000);
+    assert_eq!(meta.num_bits(), 6);
+}
+
+impl BlockedBitpacker {
+    pub fn new() -> Self {
+        let mut compressed_blocks = vec![];
+        compressed_blocks.resize(8, 0);
+        Self {
+            compressed_blocks,
+            buffer: vec![],
+            offset_and_bits: vec![],
+        }
+    }
+
+    /// The memory used (inclusive childs)
+    pub fn mem_usage(&self) -> usize {
+        std::mem::size_of::<BlockedBitpacker>()
+            + self.compressed_blocks.capacity()
+            + self.offset_and_bits.capacity()
+                * std::mem::size_of_val(&self.offset_and_bits.get(0).cloned().unwrap_or_default())
+            + self.buffer.capacity()
+                * std::mem::size_of_val(&self.buffer.get(0).cloned().unwrap_or_default())
+    }
+
+    #[inline]
+    pub fn add(&mut self, val: u64) {
+        self.buffer.push(val);
+        if self.buffer.len() == BLOCK_SIZE as usize {
+            self.flush();
+        }
+    }
+
+    pub fn flush(&mut self) {
+        if let Some((min_value, max_value)) = minmax(self.buffer.iter()) {
+            let mut bit_packer = BitPacker::new();
+            let num_bits_block = compute_num_bits(*max_value - min_value);
+            // todo performance: the padding handling could be done better, e.g. use a slice and
+            // return num_bytes written from bitpacker
+            self.compressed_blocks
+                .resize(self.compressed_blocks.len() - 8, 0); // remove padding for bitpacker
+            let offset = self.compressed_blocks.len() as u64;
+            // todo performance: for some bit_width we
+            // can encode multiple vals into the
+            // mini_buffer before checking to flush
+            // (to be done in BitPacker)
+            for val in self.buffer.iter() {
+                bit_packer
+                    .write(
+                        *val - min_value,
+                        num_bits_block,
+                        &mut self.compressed_blocks,
+                    )
+                    .expect("cannot write bitpacking to output"); // write to in memory can't fail
+            }
+            bit_packer.flush(&mut self.compressed_blocks).unwrap();
+            self.offset_and_bits
+                .push(BlockedBitpackerEntryMetaData::new(
+                    offset,
+                    num_bits_block,
+                    *min_value,
+                ));
+
+            self.buffer.clear();
+            self.compressed_blocks
+                .resize(self.compressed_blocks.len() + 8, 0); // add padding for bitpacker
+        }
+    }
+    #[inline]
+    pub fn get(&self, idx: usize) -> u64 {
+        let metadata_pos = idx / BLOCK_SIZE as usize;
+        let pos_in_block = idx % BLOCK_SIZE as usize;
+        if let Some(metadata) = self.offset_and_bits.get(metadata_pos) {
+            let unpacked = BitUnpacker::new(metadata.num_bits()).get(
+                pos_in_block as u64,
+                &self.compressed_blocks[metadata.offset() as usize..],
+            );
+            unpacked + metadata.base_value()
+        } else {
+            self.buffer[pos_in_block]
+        }
+    }
+
+    pub fn iter(&self) -> impl Iterator<Item = u64> + '_ {
+        // todo performance: we could decompress a whole block and cache it instead
+        let bitpacked_elems = self.offset_and_bits.len() * BLOCK_SIZE;
+        let iter = (0..bitpacked_elems)
+            .map(move |idx| self.get(idx))
+            .chain(self.buffer.iter().cloned());
+        iter
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    #[test]
+    fn blocked_bitpacker_empty() {
+        let blocked_bitpacker = BlockedBitpacker::new();
+        assert_eq!(blocked_bitpacker.iter().collect::<Vec<u64>>(), vec![]);
+    }
+    #[test]
+    fn blocked_bitpacker_one() {
+        let mut blocked_bitpacker = BlockedBitpacker::new();
+        blocked_bitpacker.add(50000);
+        assert_eq!(blocked_bitpacker.get(0), 50000);
+        assert_eq!(blocked_bitpacker.iter().collect::<Vec<u64>>(), vec![50000]);
+    }
+    #[test]
+    fn blocked_bitpacker_test() {
+        let mut blocked_bitpacker = BlockedBitpacker::new();
+        for val in 0..21500 {
+            blocked_bitpacker.add(val);
+        }
+        for val in 0..21500 {
+            assert_eq!(blocked_bitpacker.get(val as usize), val);
+        }
+        assert_eq!(blocked_bitpacker.iter().count(), 21500);
+        assert_eq!(blocked_bitpacker.iter().last().unwrap(), 21499);
+    }
+}

bitpacker/src/lib.rs

@@ -0,0 +1,81 @@
+mod bitpacker;
+mod blocked_bitpacker;
+
+pub use crate::bitpacker::BitPacker;
+pub use crate::bitpacker::BitUnpacker;
+pub use crate::blocked_bitpacker::BlockedBitpacker;
+
+/// Computes the number of bits that will be used for bitpacking.
+///
+/// In general the target is the minimum number of bits
+/// required to express the amplitude given in argument.
+///
+/// e.g. If the amplitude is 10, we can store all ints on simply 4bits.
+///
+/// The logic is slightly more convoluted here as for optimization
+/// reasons, we want to ensure that a value spawns over at most 8 bytes
+/// of aligned bytes.
+///
+/// Spanning over 9 bytes is possible for instance, if we do
+/// bitpacking with an amplitude of 63 bits.
+/// In this case, the second int will start on bit
+/// 63 (which belongs to byte 7) and ends at byte 15;
+/// Hence 9 bytes (from byte 7 to byte 15 included).
+///
+/// To avoid this, we force the number of bits to 64bits
+/// when the result is greater than `64-8 = 56 bits`.
+///
+/// Note that this only affects rare use cases spawning over
+/// a very large range of values. Even in this case, it results
+/// in an extra cost of at most 12% compared to the optimal
+/// number of bits.
+pub fn compute_num_bits(n: u64) -> u8 {
+    let amplitude = (64u32 - n.leading_zeros()) as u8;
+    if amplitude <= 64 - 8 {
+        amplitude
+    } else {
+        64
+    }
+}
+
+pub fn minmax<I, T>(mut vals: I) -> Option<(T, T)>
+where
+    I: Iterator<Item = T>,
+    T: Copy + Ord,
+{
+    if let Some(first_el) = vals.next() {
+        return Some(vals.fold((first_el, first_el), |(min_val, max_val), el| {
+            (min_val.min(el), max_val.max(el))
+        }));
+    }
+    None
+}
+
+#[test]
+fn test_compute_num_bits() {
+    assert_eq!(compute_num_bits(1), 1u8);
+    assert_eq!(compute_num_bits(0), 0u8);
+    assert_eq!(compute_num_bits(2), 2u8);
+    assert_eq!(compute_num_bits(3), 2u8);
+    assert_eq!(compute_num_bits(4), 3u8);
+    assert_eq!(compute_num_bits(255), 8u8);
+    assert_eq!(compute_num_bits(256), 9u8);
+    assert_eq!(compute_num_bits(5_000_000_000), 33u8);
+}
+
+#[test]
+fn test_minmax_empty() {
+    let vals: Vec<u32> = vec![];
+    assert_eq!(minmax(vals.into_iter()), None);
+}
+
+#[test]
+fn test_minmax_one() {
+    assert_eq!(minmax(vec![1].into_iter()), Some((1, 1)));
+}
+
+#[test]
+fn test_minmax_two() {
+    assert_eq!(minmax(vec![1, 2].into_iter()), Some((1, 2)));
+    assert_eq!(minmax(vec![2, 1].into_iter()), Some((1, 2)));
+}

ci/script.sh

@@ -7,7 +7,7 @@ set -ex
 main() {
     if [ ! -z $CODECOV ]; then
         echo "Codecov"
-        cargo build --verbose && cargo coverage --verbose && bash <(curl -s https://codecov.io/bash) -s target/kcov
+        cargo build --verbose && cargo coverage --verbose --all && bash <(curl -s https://codecov.io/bash) -s target/kcov
     else
         echo "Build"
         cross build --target $TARGET
@@ -15,7 +15,8 @@ main() {
             return
         fi
         echo "Test"
-        cross test --target $TARGET --no-default-features --features mmap -- --test-threads 1
+        cross test --target $TARGET --no-default-features --features mmap
+        cross test --target $TARGET --no-default-features --features mmap query-grammar
     fi
     for example in $(ls examples/*.rs)
     do

common/Cargo.toml

@@ -0,0 +1,17 @@
+[package]
+name = "tantivy-common"
+version = "0.1.0"
+authors = ["Paul Masurel <paul@quickwit.io>", "Pascal Seitz <pascal@quickwit.io>"]
+license = "MIT"
+edition = "2018"
+description = "common traits and utility functions used by multiple tantivy subcrates"
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]
+byteorder = "1.4.3"
+ownedbytes = { version="0.1", path="../ownedbytes" }
+
+[dev-dependencies]
+proptest = "1.0.0"
+rand = "0.8.4"

common/src/bitset.rs

@@ -1,11 +1,14 @@
-use std::fmt;
+use ownedbytes::OwnedBytes;
+use std::convert::TryInto;
+use std::io::Write;
 use std::u64;
+use std::{fmt, io};
 
 #[derive(Clone, Copy, Eq, PartialEq)]
-pub(crate) struct TinySet(u64);
+pub struct TinySet(u64);
 
 impl fmt::Debug for TinySet {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         self.into_iter().collect::<Vec<u32>>().fmt(f)
     }
 }
@@ -14,6 +17,7 @@ pub struct TinySetIterator(TinySet);
 impl Iterator for TinySetIterator {
     type Item = u32;
 
+    #[inline]
     fn next(&mut self) -> Option<Self::Item> {
         self.0.pop_lowest()
     }
@@ -28,21 +32,54 @@ impl IntoIterator for TinySet {
 }
 
 impl TinySet {
+    pub fn serialize<T: Write>(&self, writer: &mut T) -> io::Result<()> {
+        writer.write_all(self.0.to_le_bytes().as_ref())
+    }
+
+    #[inline]
+    pub fn deserialize(data: [u8; 8]) -> io::Result<Self> {
+        let val: u64 = u64::from_le_bytes(data);
+        Ok(TinySet(val))
+    }
+
     /// Returns an empty `TinySet`.
+    #[inline]
     pub fn empty() -> TinySet {
         TinySet(0u64)
     }
 
+    /// Returns a full `TinySet`.
+    #[inline]
+    pub fn full() -> TinySet {
+        TinySet::empty().complement()
+    }
+
+    pub fn clear(&mut self) {
+        self.0 = 0u64;
+    }
+
+    #[inline]
     /// Returns the complement of the set in `[0, 64[`.
+    ///
+    /// Careful on making this function public, as it will break the padding handling in the last
+    /// bucket.
     fn complement(self) -> TinySet {
         TinySet(!self.0)
     }
 
+    #[inline]
     /// Returns true iff the `TinySet` contains the element `el`.
     pub fn contains(self, el: u32) -> bool {
         !self.intersect(TinySet::singleton(el)).is_empty()
     }
 
+    #[inline]
+    /// Returns the number of elements in the TinySet.
+    pub fn len(self) -> u32 {
+        self.0.count_ones()
+    }
+
+    #[inline]
     /// Returns the intersection of `self` and `other`
     pub fn intersect(self, other: TinySet) -> TinySet {
         TinySet(self.0 & other.0)
@@ -50,40 +87,58 @@ impl TinySet {
 
     /// Creates a new `TinySet` containing only one element
     /// within `[0; 64[`
-    #[inline(always)]
+    #[inline]
     pub fn singleton(el: u32) -> TinySet {
         TinySet(1u64 << u64::from(el))
     }
 
-    /// Insert a new element within [0..64[
-    #[inline(always)]
+    /// Insert a new element within [0..64)
+    #[inline]
     pub fn insert(self, el: u32) -> TinySet {
         self.union(TinySet::singleton(el))
     }
 
-    /// Insert a new element within [0..64[
-    #[inline(always)]
+    /// Removes an element within [0..64)
+    #[inline]
+    pub fn remove(self, el: u32) -> TinySet {
+        self.intersect(TinySet::singleton(el).complement())
+    }
+
+    /// Insert a new element within [0..64)
+    ///
+    /// returns true if the set changed
+    #[inline]
     pub fn insert_mut(&mut self, el: u32) -> bool {
         let old = *self;
         *self = old.insert(el);
         old != *self
     }
 
+    /// Remove a element within [0..64)
+    ///
+    /// returns true if the set changed
+    #[inline]
+    pub fn remove_mut(&mut self, el: u32) -> bool {
+        let old = *self;
+        *self = old.remove(el);
+        old != *self
+    }
+
     /// Returns the union of two tinysets
-    #[inline(always)]
+    #[inline]
     pub fn union(self, other: TinySet) -> TinySet {
         TinySet(self.0 | other.0)
     }
 
     /// Returns true iff the `TinySet` is empty.
-    #[inline(always)]
+    #[inline]
     pub fn is_empty(self) -> bool {
         self.0 == 0u64
     }
 
     /// Returns the lowest element in the `TinySet`
     /// and removes it.
-    #[inline(always)]
+    #[inline]
     pub fn pop_lowest(&mut self) -> Option<u32> {
         if self.is_empty() {
             None
@@ -109,22 +164,12 @@ impl TinySet {
     pub fn range_greater_or_equal(from_included: u32) -> TinySet {
         TinySet::range_lower(from_included).complement()
     }
-
-    pub fn clear(&mut self) {
-        self.0 = 0u64;
-    }
-
-    pub fn len(self) -> u32 {
-        self.0.count_ones()
-    }
 }
 
 #[derive(Clone)]
 pub struct BitSet {
     tinysets: Box<[TinySet]>,
-    len: usize, //< Technically it should be u32, but we
-    // count multiple inserts.
-    // `usize` guards us from overflow.
+    len: u64,
     max_value: u32,
 }
 
@@ -133,8 +178,41 @@ fn num_buckets(max_val: u32) -> u32 {
 }
 
 impl BitSet {
+    /// serialize a `BitSet`.
+    ///
+    pub fn serialize<T: Write>(&self, writer: &mut T) -> io::Result<()> {
+        writer.write_all(self.max_value.to_le_bytes().as_ref())?;
+
+        for tinyset in self.tinysets.iter() {
+            tinyset.serialize(writer)?;
+        }
+        writer.flush()?;
+        Ok(())
+    }
+
+    /// Deserialize a `BitSet`.
+    ///
+    #[cfg(test)]
+    pub fn deserialize(mut data: &[u8]) -> io::Result<Self> {
+        let max_value: u32 = u32::from_le_bytes(data[..4].try_into().unwrap());
+        data = &data[4..];
+
+        let mut len: u64 = 0;
+        let mut tinysets = vec![];
+        for chunk in data.chunks_exact(8) {
+            let tinyset = TinySet::deserialize(chunk.try_into().unwrap())?;
+            len += tinyset.len() as u64;
+            tinysets.push(tinyset);
+        }
+        Ok(BitSet {
+            tinysets: tinysets.into_boxed_slice(),
+            len,
+            max_value,
+        })
+    }
+
     /// Create a new `BitSet` that may contain elements
-    /// within `[0, max_val[`.
+    /// within `[0, max_val)`.
     pub fn with_max_value(max_value: u32) -> BitSet {
         let num_buckets = num_buckets(max_value);
         let tinybisets = vec![TinySet::empty(); num_buckets as usize].into_boxed_slice();
@@ -145,6 +223,23 @@ impl BitSet {
         }
     }
 
+    /// Create a new `BitSet` that may contain elements. Initially all values will be set.
+    /// within `[0, max_val)`.
+    pub fn with_max_value_and_full(max_value: u32) -> BitSet {
+        let num_buckets = num_buckets(max_value);
+        let mut tinybisets = vec![TinySet::full(); num_buckets as usize].into_boxed_slice();
+
+        // Fix padding
+        let lower = max_value % 64u32;
+        tinybisets[tinybisets.len() - 1] = TinySet::range_lower(lower);
+
+        BitSet {
+            tinysets: tinybisets,
+            len: max_value as u64,
+            max_value,
+        }
+    }
+
     /// Removes all elements from the `BitSet`.
     pub fn clear(&mut self) {
         for tinyset in self.tinysets.iter_mut() {
@@ -154,10 +249,11 @@ impl BitSet {
 
     /// Returns the number of elements in the `BitSet`.
     pub fn len(&self) -> usize {
-        self.len
+        self.len as usize
     }
 
     /// Inserts an element in the `BitSet`
+    #[inline]
     pub fn insert(&mut self, el: u32) {
         // we do not check saturated els.
         let higher = el / 64u32;
@@ -169,7 +265,21 @@ impl BitSet {
         };
     }
 
+    /// Inserts an element in the `BitSet`
+    #[inline]
+    pub fn remove(&mut self, el: u32) {
+        // we do not check saturated els.
+        let higher = el / 64u32;
+        let lower = el % 64u32;
+        self.len -= if self.tinysets[higher as usize].remove_mut(lower) {
+            1
+        } else {
+            0
+        };
+    }
+
     /// Returns true iff the elements is in the `BitSet`.
+    #[inline]
     pub fn contains(&self, el: u32) -> bool {
         self.tinyset(el / 64u32).contains(el % 64)
     }
@@ -179,7 +289,7 @@ impl BitSet {
     ///
     /// Reminder: the tiny set with the bucket `bucket`, represents the
     /// elements from `bucket * 64` to `(bucket+1) * 64`.
-    pub(crate) fn first_non_empty_bucket(&self, bucket: u32) -> Option<u32> {
+    pub fn first_non_empty_bucket(&self, bucket: u32) -> Option<u32> {
         self.tinysets[bucket as usize..]
             .iter()
             .cloned()
@@ -194,23 +304,149 @@ impl BitSet {
     /// Returns the tiny bitset representing the
     /// the set restricted to the number range from
     /// `bucket * 64` to `(bucket + 1) * 64`.
-    pub(crate) fn tinyset(&self, bucket: u32) -> TinySet {
+    pub fn tinyset(&self, bucket: u32) -> TinySet {
         self.tinysets[bucket as usize]
     }
 }
 
+/// Serialized BitSet.
+#[derive(Clone)]
+pub struct ReadSerializedBitSet {
+    data: OwnedBytes,
+    max_value: u32,
+}
+
+impl ReadSerializedBitSet {
+    pub fn open(data: OwnedBytes) -> Self {
+        let (max_value_data, data) = data.split(4);
+        let max_value: u32 = u32::from_le_bytes(max_value_data.as_ref().try_into().unwrap());
+        ReadSerializedBitSet { data, max_value }
+    }
+
+    /// Number of elements in the bitset.
+    #[inline]
+    pub fn len(&self) -> usize {
+        self.iter_tinysets()
+            .map(|tinyset| tinyset.len() as usize)
+            .sum()
+    }
+
+    /// Iterate the tinyset on the fly from serialized data.
+    ///
+    #[inline]
+    fn iter_tinysets<'a>(&'a self) -> impl Iterator<Item = TinySet> + 'a {
+        assert!((self.data.len()) % 8 == 0);
+        self.data.chunks_exact(8).map(move |chunk| {
+            let tinyset: TinySet = TinySet::deserialize(chunk.try_into().unwrap()).unwrap();
+            tinyset
+        })
+    }
+
+    /// Iterate over the positions of the elements.
+    ///
+    #[inline]
+    pub fn iter<'a>(&'a self) -> impl Iterator<Item = u32> + 'a {
+        self.iter_tinysets()
+            .enumerate()
+            .flat_map(move |(chunk_num, tinyset)| {
+                let chunk_base_val = chunk_num as u32 * 64;
+                tinyset
+                    .into_iter()
+                    .map(move |val| val + chunk_base_val)
+                    .take_while(move |doc| *doc < self.max_value)
+            })
+    }
+
+    /// Returns true iff the elements is in the `BitSet`.
+    #[inline]
+    pub fn contains(&self, el: u32) -> bool {
+        let byte_offset = el / 8u32;
+        let b: u8 = self.data[byte_offset as usize];
+        let shift = (el % 8) as u8;
+        b & (1u8 << shift) != 0
+    }
+
+    /// Maximum value the bitset may contain.
+    /// (Note this is not the maximum value contained in the set.)
+    ///
+    /// A bitset has an intrinsic capacity.
+    /// It only stores elements within [0..max_value).
+    #[inline]
+    pub fn max_value(&self) -> u32 {
+        self.max_value
+    }
+}
+
 #[cfg(test)]
 mod tests {
 
     use super::BitSet;
+    use super::ReadSerializedBitSet;
     use super::TinySet;
-    use docset::DocSet;
-    use query::BitSetDocSet;
-    use std::collections::BTreeSet;
+    use ownedbytes::OwnedBytes;
+    use rand::distributions::Bernoulli;
+    use rand::rngs::StdRng;
+    use rand::{Rng, SeedableRng};
     use std::collections::HashSet;
-    use tests;
-    use tests::generate_nonunique_unsorted;
+    use std::convert::TryInto;
 
+    #[test]
+    fn test_read_serialized_bitset_full() {
+        let mut bitset = BitSet::with_max_value_and_full(5);
+        bitset.remove(3);
+        let mut out = vec![];
+        bitset.serialize(&mut out).unwrap();
+
+        let bitset = ReadSerializedBitSet::open(OwnedBytes::new(out));
+        assert_eq!(bitset.len(), 4);
+    }
+
+    #[test]
+    fn test_read_serialized_bitset_empty() {
+        let mut bitset = BitSet::with_max_value(5);
+        bitset.insert(3);
+        let mut out = vec![];
+        bitset.serialize(&mut out).unwrap();
+
+        let bitset = ReadSerializedBitSet::open(OwnedBytes::new(out));
+        assert_eq!(bitset.len(), 1);
+
+        {
+            let bitset = BitSet::with_max_value(5);
+            let mut out = vec![];
+            bitset.serialize(&mut out).unwrap();
+            let bitset = ReadSerializedBitSet::open(OwnedBytes::new(out));
+            assert_eq!(bitset.len(), 0);
+        }
+    }
+
+    #[test]
+    fn test_tiny_set_remove() {
+        {
+            let mut u = TinySet::empty().insert(63u32).insert(5).remove(63u32);
+            assert_eq!(u.pop_lowest(), Some(5u32));
+            assert!(u.pop_lowest().is_none());
+        }
+        {
+            let mut u = TinySet::empty()
+                .insert(63u32)
+                .insert(1)
+                .insert(5)
+                .remove(63u32);
+            assert_eq!(u.pop_lowest(), Some(1u32));
+            assert_eq!(u.pop_lowest(), Some(5u32));
+            assert!(u.pop_lowest().is_none());
+        }
+        {
+            let mut u = TinySet::empty().insert(1).remove(63u32);
+            assert_eq!(u.pop_lowest(), Some(1u32));
+            assert!(u.pop_lowest().is_none());
+        }
+        {
+            let mut u = TinySet::empty().insert(1).remove(1u32);
+            assert!(u.pop_lowest().is_none());
+        }
+    }
     #[test]
     fn test_tiny_set() {
         assert!(TinySet::empty().is_empty());
@@ -236,6 +472,21 @@ mod tests {
             assert_eq!(u.pop_lowest(), Some(63u32));
             assert!(u.pop_lowest().is_none());
         }
+        {
+            let mut u = TinySet::empty().insert(63u32).insert(5);
+            assert_eq!(u.pop_lowest(), Some(5u32));
+            assert_eq!(u.pop_lowest(), Some(63u32));
+            assert!(u.pop_lowest().is_none());
+        }
+        {
+            let u = TinySet::empty().insert(63u32).insert(5);
+            let mut data = vec![];
+            u.serialize(&mut data).unwrap();
+            let mut u = TinySet::deserialize(data[..8].try_into().unwrap()).unwrap();
+            assert_eq!(u.pop_lowest(), Some(5u32));
+            assert_eq!(u.pop_lowest(), Some(63u32));
+            assert!(u.pop_lowest().is_none());
+        }
     }
 
     #[test]
@@ -252,6 +503,16 @@ mod tests {
                 assert_eq!(hashset.contains(&el), bitset.contains(el));
             }
             assert_eq!(bitset.max_value(), max_value);
+
+            // test deser
+            let mut data = vec![];
+            bitset.serialize(&mut data).unwrap();
+            let bitset = BitSet::deserialize(&data).unwrap();
+            for el in 0..max_value {
+                assert_eq!(hashset.contains(&el), bitset.contains(el));
+            }
+            assert_eq!(bitset.max_value(), max_value);
+            assert_eq!(bitset.len(), els.len());
         };
 
         test_against_hashset(&[], 0);
@@ -264,27 +525,6 @@ mod tests {
         test_against_hashset(&[62u32, 63u32], 64);
     }
 
-    #[test]
-    fn test_bitset_large() {
-        let arr = generate_nonunique_unsorted(100_000, 5_000);
-        let mut btreeset: BTreeSet<u32> = BTreeSet::new();
-        let mut bitset = BitSet::with_max_value(100_000);
-        for el in arr {
-            btreeset.insert(el);
-            bitset.insert(el);
-        }
-        for i in 0..100_000 {
-            assert_eq!(btreeset.contains(&i), bitset.contains(i));
-        }
-        assert_eq!(btreeset.len(), bitset.len());
-        let mut bitset_docset = BitSetDocSet::from(bitset);
-        for el in btreeset.into_iter() {
-            bitset_docset.advance();
-            assert_eq!(bitset_docset.doc(), el);
-        }
-        assert!(!bitset_docset.advance());
-    }
-
     #[test]
     fn test_bitset_num_buckets() {
         use super::num_buckets;
@@ -337,12 +577,33 @@ mod tests {
         assert_eq!(bitset.len(), 2);
         bitset.insert(104u32);
         assert_eq!(bitset.len(), 3);
+        bitset.remove(105u32);
+        assert_eq!(bitset.len(), 3);
+        bitset.remove(104u32);
+        assert_eq!(bitset.len(), 2);
+        bitset.remove(3u32);
+        assert_eq!(bitset.len(), 1);
+        bitset.remove(103u32);
+        assert_eq!(bitset.len(), 0);
+    }
+
+    pub fn sample_with_seed(n: u32, ratio: f64, seed_val: u8) -> Vec<u32> {
+        StdRng::from_seed([seed_val; 32])
+            .sample_iter(&Bernoulli::new(ratio).unwrap())
+            .take(n as usize)
+            .enumerate()
+            .filter_map(|(val, keep)| if keep { Some(val as u32) } else { None })
+            .collect()
+    }
+
+    pub fn sample(n: u32, ratio: f64) -> Vec<u32> {
+        sample_with_seed(n, ratio, 4)
     }
 
     #[test]
     fn test_bitset_clear() {
         let mut bitset = BitSet::with_max_value(1_000);
-        let els = tests::sample(1_000, 0.01f64);
+        let els = sample(1_000, 0.01f64);
         for &el in &els {
             bitset.insert(el);
         }

common/src/lib.rs

@@ -0,0 +1,167 @@
+use std::ops::Deref;
+
+pub use byteorder::LittleEndian as Endianness;
+
+mod bitset;
+mod serialize;
+mod vint;
+mod writer;
+
+pub use bitset::*;
+pub use serialize::{BinarySerializable, DeserializeFrom, FixedSize};
+pub use vint::{read_u32_vint, read_u32_vint_no_advance, serialize_vint_u32, write_u32_vint, VInt};
+pub use writer::{AntiCallToken, CountingWriter, TerminatingWrite};
+
+/// Has length trait
+pub trait HasLen {
+    /// Return length
+    fn len(&self) -> usize;
+
+    /// Returns true iff empty.
+    fn is_empty(&self) -> bool {
+        self.len() == 0
+    }
+}
+
+impl<T: Deref<Target = [u8]>> HasLen for T {
+    fn len(&self) -> usize {
+        self.deref().len()
+    }
+}
+
+const HIGHEST_BIT: u64 = 1 << 63;
+
+/// Maps a `i64` to `u64`
+///
+/// For simplicity, tantivy internally handles `i64` as `u64`.
+/// The mapping is defined by this function.
+///
+/// Maps `i64` to `u64` so that
+/// `-2^63 .. 2^63-1` is mapped
+///     to
+/// `0 .. 2^64-1`
+/// in that order.
+///
+/// This is more suited than simply casting (`val as u64`)
+/// because of bitpacking.
+///
+/// Imagine a list of `i64` ranging from -10 to 10.
+/// When casting negative values, the negative values are projected
+/// to values over 2^63, and all values end up requiring 64 bits.
+///
+/// # See also
+/// The [reverse mapping is `u64_to_i64`](./fn.u64_to_i64.html).
+#[inline]
+pub fn i64_to_u64(val: i64) -> u64 {
+    (val as u64) ^ HIGHEST_BIT
+}
+
+/// Reverse the mapping given by [`i64_to_u64`](./fn.i64_to_u64.html).
+#[inline]
+pub fn u64_to_i64(val: u64) -> i64 {
+    (val ^ HIGHEST_BIT) as i64
+}
+
+/// Maps a `f64` to `u64`
+///
+/// For simplicity, tantivy internally handles `f64` as `u64`.
+/// The mapping is defined by this function.
+///
+/// Maps `f64` to `u64` in a monotonic manner, so that bytes lexical order is preserved.
+///
+/// This is more suited than simply casting (`val as u64`)
+/// which would truncate the result
+///
+/// # Reference
+///
+/// Daniel Lemire's [blog post](https://lemire.me/blog/2020/12/14/converting-floating-point-numbers-to-integers-while-preserving-order/)
+/// explains the mapping in a clear manner.
+///
+/// # See also
+/// The [reverse mapping is `u64_to_f64`](./fn.u64_to_f64.html).
+#[inline]
+pub fn f64_to_u64(val: f64) -> u64 {
+    let bits = val.to_bits();
+    if val.is_sign_positive() {
+        bits ^ HIGHEST_BIT
+    } else {
+        !bits
+    }
+}
+
+/// Reverse the mapping given by [`i64_to_u64`](./fn.i64_to_u64.html).
+#[inline]
+pub fn u64_to_f64(val: u64) -> f64 {
+    f64::from_bits(if val & HIGHEST_BIT != 0 {
+        val ^ HIGHEST_BIT
+    } else {
+        !val
+    })
+}
+
+#[cfg(test)]
+pub mod test {
+
+    use super::{f64_to_u64, i64_to_u64, u64_to_f64, u64_to_i64};
+    use super::{BinarySerializable, FixedSize};
+    use proptest::prelude::*;
+    use std::f64;
+
+    fn test_i64_converter_helper(val: i64) {
+        assert_eq!(u64_to_i64(i64_to_u64(val)), val);
+    }
+
+    fn test_f64_converter_helper(val: f64) {
+        assert_eq!(u64_to_f64(f64_to_u64(val)), val);
+    }
+
+    pub fn fixed_size_test<O: BinarySerializable + FixedSize + Default>() {
+        let mut buffer = Vec::new();
+        O::default().serialize(&mut buffer).unwrap();
+        assert_eq!(buffer.len(), O::SIZE_IN_BYTES);
+    }
+
+    proptest! {
+        #[test]
+        fn test_f64_converter_monotonicity_proptest((left, right) in (proptest::num::f64::NORMAL, proptest::num::f64::NORMAL)) {
+            let left_u64 = f64_to_u64(left);
+            let right_u64 = f64_to_u64(right);
+            assert_eq!(left_u64 < right_u64,  left < right);
+        }
+    }
+
+    #[test]
+    fn test_i64_converter() {
+        assert_eq!(i64_to_u64(i64::min_value()), u64::min_value());
+        assert_eq!(i64_to_u64(i64::max_value()), u64::max_value());
+        test_i64_converter_helper(0i64);
+        test_i64_converter_helper(i64::min_value());
+        test_i64_converter_helper(i64::max_value());
+        for i in -1000i64..1000i64 {
+            test_i64_converter_helper(i);
+        }
+    }
+
+    #[test]
+    fn test_f64_converter() {
+        test_f64_converter_helper(f64::INFINITY);
+        test_f64_converter_helper(f64::NEG_INFINITY);
+        test_f64_converter_helper(0.0);
+        test_f64_converter_helper(-0.0);
+        test_f64_converter_helper(1.0);
+        test_f64_converter_helper(-1.0);
+    }
+
+    #[test]
+    fn test_f64_order() {
+        assert!(!(f64_to_u64(f64::NEG_INFINITY)..f64_to_u64(f64::INFINITY))
+            .contains(&f64_to_u64(f64::NAN))); //nan is not a number
+        assert!(f64_to_u64(1.5) > f64_to_u64(1.0)); //same exponent, different mantissa
+        assert!(f64_to_u64(2.0) > f64_to_u64(1.0)); //same mantissa, different exponent
+        assert!(f64_to_u64(2.0) > f64_to_u64(1.5)); //different exponent and mantissa
+        assert!(f64_to_u64(1.0) > f64_to_u64(-1.0)); // pos > neg
+        assert!(f64_to_u64(-1.5) < f64_to_u64(-1.0));
+        assert!(f64_to_u64(-2.0) < f64_to_u64(1.0));
+        assert!(f64_to_u64(-2.0) < f64_to_u64(-1.5));
+    }
+}

common/src/serialize.rs

@@ -1,6 +1,6 @@
+use crate::Endianness;
+use crate::VInt;
 use byteorder::{ReadBytesExt, WriteBytesExt};
-use common::Endianness;
-use common::VInt;
 use std::fmt;
 use std::io;
 use std::io::Read;
@@ -14,6 +14,20 @@ pub trait BinarySerializable: fmt::Debug + Sized {
     fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self>;
 }
 
+pub trait DeserializeFrom<T: BinarySerializable> {
+    fn deserialize(&mut self) -> io::Result<T>;
+}
+
+/// Implement deserialize from &[u8] for all types which implement BinarySerializable.
+///
+/// TryFrom would actually be preferrable, but not possible because of the orphan
+/// rules (not completely sure if this could be resolved)
+impl<T: BinarySerializable> DeserializeFrom<T> for &[u8] {
+    fn deserialize(&mut self) -> io::Result<T> {
+        T::deserialize(self)
+    }
+}
+
 /// `FixedSize` marks a `BinarySerializable` as
 /// always serializing to the same size.
 pub trait FixedSize: BinarySerializable {
@@ -61,6 +75,11 @@ impl<Left: BinarySerializable, Right: BinarySerializable> BinarySerializable for
         Ok((Left::deserialize(reader)?, Right::deserialize(reader)?))
     }
 }
+impl<Left: BinarySerializable + FixedSize, Right: BinarySerializable + FixedSize> FixedSize
+    for (Left, Right)
+{
+    const SIZE_IN_BYTES: usize = Left::SIZE_IN_BYTES + Right::SIZE_IN_BYTES;
+}
 
 impl BinarySerializable for u32 {
     fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
@@ -89,6 +108,19 @@ impl FixedSize for u64 {
     const SIZE_IN_BYTES: usize = 8;
 }
 
+impl BinarySerializable for f32 {
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
+        writer.write_f32::<Endianness>(*self)
+    }
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
+        reader.read_f32::<Endianness>()
+    }
+}
+
+impl FixedSize for f32 {
+    const SIZE_IN_BYTES: usize = 4;
+}
+
 impl BinarySerializable for i64 {
     fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
         writer.write_i64::<Endianness>(*self)
@@ -102,6 +134,19 @@ impl FixedSize for i64 {
     const SIZE_IN_BYTES: usize = 8;
 }
 
+impl BinarySerializable for f64 {
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
+        writer.write_f64::<Endianness>(*self)
+    }
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
+        reader.read_f64::<Endianness>()
+    }
+}
+
+impl FixedSize for f64 {
+    const SIZE_IN_BYTES: usize = 8;
+}
+
 impl BinarySerializable for u8 {
     fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
         writer.write_u8(*self)
@@ -115,6 +160,28 @@ impl FixedSize for u8 {
     const SIZE_IN_BYTES: usize = 1;
 }
 
+impl BinarySerializable for bool {
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
+        let val = if *self { 1 } else { 0 };
+        writer.write_u8(val)
+    }
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<bool> {
+        let val = reader.read_u8()?;
+        match val {
+            0 => Ok(false),
+            1 => Ok(true),
+            _ => Err(io::Error::new(
+                io::ErrorKind::InvalidData,
+                "invalid bool value on deserialization, data corrupted",
+            )),
+        }
+    }
+}
+
+impl FixedSize for bool {
+    const SIZE_IN_BYTES: usize = 1;
+}
+
 impl BinarySerializable for String {
     fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
         let data: &[u8] = self.as_bytes();
@@ -135,9 +202,9 @@ impl BinarySerializable for String {
 #[cfg(test)]
 pub mod test {
 
+    use super::VInt;
     use super::*;
-    use common::VInt;
-
+    use crate::serialize::BinarySerializable;
     pub fn fixed_size_test<O: BinarySerializable + FixedSize + Default>() {
         let mut buffer = Vec::new();
         O::default().serialize(&mut buffer).unwrap();
@@ -172,6 +239,11 @@ pub mod test {
         fixed_size_test::<i64>();
     }
 
+    #[test]
+    fn test_serialize_f64() {
+        fixed_size_test::<f64>();
+    }
+
     #[test]
     fn test_serialize_u64() {
         fixed_size_test::<u64>();
@@ -181,10 +253,7 @@ pub mod test {
     fn test_serialize_string() {
         assert_eq!(serialize_test(String::from("")), 1);
         assert_eq!(serialize_test(String::from("ぽよぽよ")), 1 + 3 * 4);
-        assert_eq!(
-            serialize_test(String::from("富士さん見える。")),
-            1 + 3 * 8
-        );
+        assert_eq!(serialize_test(String::from("富士さん見える。")), 1 + 3 * 8);
     }
 
     #[test]

common/src/vint.rs

@@ -0,0 +1,234 @@
+use super::BinarySerializable;
+use byteorder::{ByteOrder, LittleEndian};
+use std::io;
+use std::io::Read;
+use std::io::Write;
+
+///   Wrapper over a `u64` that serializes as a variable int.
+#[derive(Clone, Copy, Debug, Eq, PartialEq)]
+pub struct VInt(pub u64);
+
+const STOP_BIT: u8 = 128;
+
+pub fn serialize_vint_u32(val: u32, buf: &mut [u8; 8]) -> &[u8] {
+    const START_2: u64 = 1 << 7;
+    const START_3: u64 = 1 << 14;
+    const START_4: u64 = 1 << 21;
+    const START_5: u64 = 1 << 28;
+
+    const STOP_1: u64 = START_2 - 1;
+    const STOP_2: u64 = START_3 - 1;
+    const STOP_3: u64 = START_4 - 1;
+    const STOP_4: u64 = START_5 - 1;
+
+    const MASK_1: u64 = 127;
+    const MASK_2: u64 = MASK_1 << 7;
+    const MASK_3: u64 = MASK_2 << 7;
+    const MASK_4: u64 = MASK_3 << 7;
+    const MASK_5: u64 = MASK_4 << 7;
+
+    let val = u64::from(val);
+    const STOP_BIT: u64 = 128u64;
+    let (res, num_bytes) = match val {
+        0..=STOP_1 => (val | STOP_BIT, 1),
+        START_2..=STOP_2 => (
+            (val & MASK_1) | ((val & MASK_2) << 1) | (STOP_BIT << (8)),
+            2,
+        ),
+        START_3..=STOP_3 => (
+            (val & MASK_1) | ((val & MASK_2) << 1) | ((val & MASK_3) << 2) | (STOP_BIT << (8 * 2)),
+            3,
+        ),
+        START_4..=STOP_4 => (
+            (val & MASK_1)
+                | ((val & MASK_2) << 1)
+                | ((val & MASK_3) << 2)
+                | ((val & MASK_4) << 3)
+                | (STOP_BIT << (8 * 3)),
+            4,
+        ),
+        _ => (
+            (val & MASK_1)
+                | ((val & MASK_2) << 1)
+                | ((val & MASK_3) << 2)
+                | ((val & MASK_4) << 3)
+                | ((val & MASK_5) << 4)
+                | (STOP_BIT << (8 * 4)),
+            5,
+        ),
+    };
+    LittleEndian::write_u64(&mut buf[..], res);
+    &buf[0..num_bytes]
+}
+
+/// Returns the number of bytes covered by a
+/// serialized vint `u32`.
+///
+/// Expects a buffer data that starts
+/// by the serialized `vint`, scans at most 5 bytes ahead until
+/// it finds the vint final byte.
+///
+/// # May Panic
+/// If the payload does not start by a valid `vint`
+fn vint_len(data: &[u8]) -> usize {
+    for (i, &val) in data.iter().enumerate().take(5) {
+        if val >= STOP_BIT {
+            return i + 1;
+        }
+    }
+    panic!("Corrupted data. Invalid VInt 32");
+}
+
+/// Reads a vint `u32` from a buffer, and
+/// consumes its payload data.
+///
+/// # Panics
+///
+/// If the buffer does not start by a valid
+/// vint payload
+pub fn read_u32_vint(data: &mut &[u8]) -> u32 {
+    let (result, vlen) = read_u32_vint_no_advance(*data);
+    *data = &data[vlen..];
+    result
+}
+
+pub fn read_u32_vint_no_advance(data: &[u8]) -> (u32, usize) {
+    let vlen = vint_len(data);
+    let mut result = 0u32;
+    let mut shift = 0u64;
+    for &b in &data[..vlen] {
+        result |= u32::from(b & 127u8) << shift;
+        shift += 7;
+    }
+    (result, vlen)
+}
+/// Write a `u32` as a vint payload.
+pub fn write_u32_vint<W: io::Write>(val: u32, writer: &mut W) -> io::Result<()> {
+    let mut buf = [0u8; 8];
+    let data = serialize_vint_u32(val, &mut buf);
+    writer.write_all(data)
+}
+
+impl VInt {
+    pub fn val(&self) -> u64 {
+        self.0
+    }
+
+    pub fn deserialize_u64<R: Read>(reader: &mut R) -> io::Result<u64> {
+        VInt::deserialize(reader).map(|vint| vint.0)
+    }
+
+    pub fn serialize_into_vec(&self, output: &mut Vec<u8>) {
+        let mut buffer = [0u8; 10];
+        let num_bytes = self.serialize_into(&mut buffer);
+        output.extend(&buffer[0..num_bytes]);
+    }
+
+    pub fn serialize_into(&self, buffer: &mut [u8; 10]) -> usize {
+        let mut remaining = self.0;
+        for (i, b) in buffer.iter_mut().enumerate() {
+            let next_byte: u8 = (remaining % 128u64) as u8;
+            remaining /= 128u64;
+            if remaining == 0u64 {
+                *b = next_byte | STOP_BIT;
+                return i + 1;
+            } else {
+                *b = next_byte;
+            }
+        }
+        unreachable!();
+    }
+}
+
+impl BinarySerializable for VInt {
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
+        let mut buffer = [0u8; 10];
+        let num_bytes = self.serialize_into(&mut buffer);
+        writer.write_all(&buffer[0..num_bytes])
+    }
+
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
+        let mut bytes = reader.bytes();
+        let mut result = 0u64;
+        let mut shift = 0u64;
+        loop {
+            match bytes.next() {
+                Some(Ok(b)) => {
+                    result |= u64::from(b % 128u8) << shift;
+                    if b >= STOP_BIT {
+                        return Ok(VInt(result));
+                    }
+                    shift += 7;
+                }
+                _ => {
+                    return Err(io::Error::new(
+                        io::ErrorKind::InvalidData,
+                        "Reach end of buffer while reading VInt",
+                    ));
+                }
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+
+    use super::serialize_vint_u32;
+    use super::BinarySerializable;
+    use super::VInt;
+
+    fn aux_test_vint(val: u64) {
+        let mut v = [14u8; 10];
+        let num_bytes = VInt(val).serialize_into(&mut v);
+        for el in &v[num_bytes..10] {
+            assert_eq!(el, &14u8);
+        }
+        assert!(num_bytes > 0);
+        if num_bytes < 10 {
+            assert!(1u64 << (7 * num_bytes) > val);
+        }
+        if num_bytes > 1 {
+            assert!(1u64 << (7 * (num_bytes - 1)) <= val);
+        }
+        let serdeser_val = VInt::deserialize(&mut &v[..]).unwrap();
+        assert_eq!(val, serdeser_val.0);
+    }
+
+    #[test]
+    fn test_vint() {
+        aux_test_vint(0);
+        aux_test_vint(1);
+        aux_test_vint(5);
+        aux_test_vint(u64::max_value());
+        for i in 1..9 {
+            let power_of_128 = 1u64 << (7 * i);
+            aux_test_vint(power_of_128 - 1u64);
+            aux_test_vint(power_of_128);
+            aux_test_vint(power_of_128 + 1u64);
+        }
+        aux_test_vint(10);
+    }
+
+    fn aux_test_serialize_vint_u32(val: u32) {
+        let mut buffer = [0u8; 10];
+        let mut buffer2 = [0u8; 8];
+        let len_vint = VInt(val as u64).serialize_into(&mut buffer);
+        let res2 = serialize_vint_u32(val, &mut buffer2);
+        assert_eq!(&buffer[..len_vint], res2, "array wrong for {}", val);
+    }
+
+    #[test]
+    fn test_vint_u32() {
+        aux_test_serialize_vint_u32(0);
+        aux_test_serialize_vint_u32(1);
+        aux_test_serialize_vint_u32(5);
+        for i in 1..3 {
+            let power_of_128 = 1u32 << (7 * i);
+            aux_test_serialize_vint_u32(power_of_128 - 1u32);
+            aux_test_serialize_vint_u32(power_of_128);
+            aux_test_serialize_vint_u32(power_of_128 + 1u32);
+        }
+        aux_test_serialize_vint_u32(u32::max_value());
+    }
+}

common/src/writer.rs

@@ -0,0 +1,114 @@
+use std::io::{self, BufWriter, Write};
+
+pub struct CountingWriter<W> {
+    underlying: W,
+    written_bytes: u64,
+}
+
+impl<W: Write> CountingWriter<W> {
+    pub fn wrap(underlying: W) -> CountingWriter<W> {
+        CountingWriter {
+            underlying,
+            written_bytes: 0,
+        }
+    }
+
+    #[inline]
+    pub fn written_bytes(&self) -> u64 {
+        self.written_bytes
+    }
+
+    /// Returns the underlying write object.
+    /// Note that this method does not trigger any flushing.
+    #[inline]
+    pub fn finish(self) -> W {
+        self.underlying
+    }
+}
+
+impl<W: Write> Write for CountingWriter<W> {
+    #[inline]
+    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
+        let written_size = self.underlying.write(buf)?;
+        self.written_bytes += written_size as u64;
+        Ok(written_size)
+    }
+
+    #[inline]
+    fn write_all(&mut self, buf: &[u8]) -> io::Result<()> {
+        self.underlying.write_all(buf)?;
+        self.written_bytes += buf.len() as u64;
+        Ok(())
+    }
+
+    #[inline]
+    fn flush(&mut self) -> io::Result<()> {
+        self.underlying.flush()
+    }
+}
+
+impl<W: TerminatingWrite> TerminatingWrite for CountingWriter<W> {
+    #[inline]
+    fn terminate_ref(&mut self, token: AntiCallToken) -> io::Result<()> {
+        self.underlying.terminate_ref(token)
+    }
+}
+
+/// Struct used to prevent from calling [`terminate_ref`](trait.TerminatingWrite#method.terminate_ref) directly
+///
+/// The point is that while the type is public, it cannot be built by anyone
+/// outside of this module.
+pub struct AntiCallToken(());
+
+/// Trait used to indicate when no more write need to be done on a writer
+pub trait TerminatingWrite: Write {
+    /// Indicate that the writer will no longer be used. Internally call terminate_ref.
+    fn terminate(mut self) -> io::Result<()>
+    where
+        Self: Sized,
+    {
+        self.terminate_ref(AntiCallToken(()))
+    }
+
+    /// You should implement this function to define custom behavior.
+    /// This function should flush any buffer it may hold.
+    fn terminate_ref(&mut self, _: AntiCallToken) -> io::Result<()>;
+}
+
+impl<W: TerminatingWrite + ?Sized> TerminatingWrite for Box<W> {
+    fn terminate_ref(&mut self, token: AntiCallToken) -> io::Result<()> {
+        self.as_mut().terminate_ref(token)
+    }
+}
+
+impl<W: TerminatingWrite> TerminatingWrite for BufWriter<W> {
+    fn terminate_ref(&mut self, a: AntiCallToken) -> io::Result<()> {
+        self.flush()?;
+        self.get_mut().terminate_ref(a)
+    }
+}
+
+impl<'a> TerminatingWrite for &'a mut Vec<u8> {
+    fn terminate_ref(&mut self, _a: AntiCallToken) -> io::Result<()> {
+        self.flush()
+    }
+}
+
+#[cfg(test)]
+mod test {
+
+    use super::CountingWriter;
+    use std::io::Write;
+
+    #[test]
+    fn test_counting_writer() {
+        let buffer: Vec<u8> = vec![];
+        let mut counting_writer = CountingWriter::wrap(buffer);
+        let bytes = (0u8..10u8).collect::<Vec<u8>>();
+        counting_writer.write_all(&bytes).unwrap();
+        let len = counting_writer.written_bytes();
+        let buffer_restituted: Vec<u8> = counting_writer.finish();
+        assert_eq!(len, 10u64);
+        assert_eq!(buffer_restituted.len(), 10);
+    }
+}

doc/src/SUMMARY.md

@@ -4,10 +4,10 @@
 
 [Avant Propos](./avant-propos.md)
 
-- [Schema](./schema.md)
-- [Indexing](./indexing.md)
 - [Segments](./basis.md)
+- [Defining your schema](./schema.md)
 - [Facetting](./facetting.md)
+- [Index Sorting](./index_sorting.md)
 - [Innerworkings](./innerworkings.md)
   - [Inverted index](./inverted_index.md)
 - [Best practise](./inverted_index.md)

doc/src/avant-propos.md

@@ -31,3 +31,4 @@ relevancy, collapsing, highlighting, spatial search.
   index from a different format.
 
   Tantivy exposes a lot of low level API to do all of these things.
+  

doc/src/index_sorting.md

@@ -0,0 +1,61 @@
+
+- [Index Sorting](#index-sorting)
+    + [Why Sorting](#why-sorting)
+        * [Compression](#compression)
+        * [Top-N Optimization](#top-n-optimization)
+        * [Pruning](#pruning)
+        * [Other](#other)
+    + [Usage](#usage)
+
+# Index Sorting
+
+Tantivy allows you to sort the index according to a property.
+
+## Why Sorting
+
+Presorting an index has several advantages:
+
+###### Compression
+
+When data is sorted it is easier to compress the data. E.g. the numbers sequence [5, 2, 3, 1, 4] would be sorted to [1, 2, 3, 4, 5]. 
+If we apply delta encoding this list would be unsorted [5, -3, 1, -2, 3] vs. [1, 1, 1, 1, 1].
+Compression ratio is mainly affected on the fast field of the sorted property, every thing else is likely unaffected. 
+###### Top-N Optimization
+
+When data is presorted by a field and search queries request sorting by the same field, we can leverage the natural order of the documents. 
+E.g. if the data is sorted by timestamp and want the top n newest docs containing a term, we can simply leveraging the order of the docids.
+
+Note: Tantivy 0.16 does not do this optimization yet.
+
+###### Pruning
+
+Let's say we want all documents and want to apply the filter `>= 2010-08-11`. When the data is sorted, we could make a lookup in the fast field to find the docid range and use this as the filter.
+
+Note: Tantivy 0.16 does not do this optimization yet.
+
+###### Other?
+
+In principle there are many algorithms possible that exploit the monotonically increasing nature. (aggregations maybe?)
+
+## Usage
+The index sorting can be configured setting [`sort_by_field`](https://github.com/tantivy-search/tantivy/blob/000d76b11a139a84b16b9b95060a1c93e8b9851c/src/core/index_meta.rs#L238) on `IndexSettings` and passing it to a `IndexBuilder`. As of tantvy 0.16 only fast fields are allowed to be used.
+
+```
+let settings = IndexSettings {
+    sort_by_field: Some(IndexSortByField {
+        field: "intval".to_string(),
+        order: Order::Desc,
+    }),
+    ..Default::default()
+};
+let mut index_builder = Index::builder().schema(schema);
+index_builder = index_builder.settings(settings);
+let index = index_builder.create_in_ram().unwrap();
+```
+
+## Implementation details
+
+Sorting an index is applied in the serialization step. In general there are two serialization steps: [Finishing a single segment](https://github.com/tantivy-search/tantivy/blob/000d76b11a139a84b16b9b95060a1c93e8b9851c/src/indexer/segment_writer.rs#L338) and [merging multiple segments](https://github.com/tantivy-search/tantivy/blob/000d76b11a139a84b16b9b95060a1c93e8b9851c/src/indexer/merger.rs#L1073).
+
+In both cases we generate a docid mapping reflecting the sort. This mapping is used when serializing the different components (doc store, fastfields, posting list, normfield, facets).
+

doc/src/schema.md

@@ -1,50 +1 @@
-# Schema
-
-When starting a new project using tantivy, your first step will be to your schema. Be aware that changing it will probably require you to reindex all of your data.
-It is strongly recommended you keep the means to iterate through your original data when this happens.
-
-If not specified otherwise, tantivy does not keep a raw version of your data,
-so the good practise is to rely on a distinct storage to store your
-raw documents.
-
-The schema defines both the type of the fields you are indexing, but also the type of indexing you want to apply to them. The set of search operations that you will be able to perform depends on the way you set up your schema.
-
-Here is what defining your schema could look like.
-
-```Rust
-use tantivy::schema::{Schema, TEXT, STORED, INT_INDEXED};
-
-let mut schema_builder = SchemaBuilder::default();
-let text_field = schema_builder.add_text_field("name", TEXT | STORED);
-let tag_field = schema_builder.add_facet_field("tags");
-let timestamp_field = schema_buider.add_u64_field("timestamp", INT_INDEXED)
-let schema = schema_builder.build();
-```
-
-Notice how adding a new field to your schema builder
-follows the following pattern :
-
-```verbatim
-  schema_builder.add_<fieldtype>_field("<fieldname>", <field_configuration>);
-```
-
-This method returns a `Field` handle that will be used for all kind of 
-
-# Field types
-
-Tantivy currently supports only 4 types.
-
-- `text` (understand `&str`)
-- `u64` and `i64`
-- `HierarchicalFacet`
-
-Let's go into their specificities.
-
-# Text
-
-Full-text search is the bread and butter of search engine.
-The key idea is fairly simple. Your text is broken apart into tokens (that's
-what we call tokenization). Tantivy then keeps track of the list of the documents containing each token.
-
-In order to increase recall you might want to normalize tokens. For instance,
-you most likely want to lowercase your tokens so that documents match the query `cat` regardless of whether your they contain the token `cat` or `Cat`.
+# Defining your schema

examples/basic_search.rs

@@ -5,26 +5,23 @@
 //
 // We will :
 // - define our schema
-// = create an index in a directory
-// - index few documents in our index
-// - search for the best document matchings "sea whale"
-// - retrieve the best document original content.
-
-extern crate tempdir;
+// - create an index in a directory
+// - index a few documents into our index
+// - search for the best document matching a basic query
+// - retrieve the best document's original content.
 
 // ---
 // Importing tantivy...
-#[macro_use]
-extern crate tantivy;
-use tantivy::collector::TopCollector;
+use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
 use tantivy::schema::*;
-use tantivy::Index;
+use tantivy::{doc, Index, ReloadPolicy};
+use tempfile::TempDir;
 
 fn main() -> tantivy::Result<()> {
     // Let's create a temporary directory for the
     // sake of this example
-    let index_path = TempDir::new("tantivy_example_dir")?;
+    let index_path = TempDir::new()?;
 
     // # Defining the schema
     //
@@ -33,8 +30,8 @@ fn main() -> tantivy::Result<()> {
     // and for each field, its type and "the way it should
     // be indexed".
 
-    // first we need to define a schema ...
-    let mut schema_builder = SchemaBuilder::default();
+    // First we need to define a schema ...
+    let mut schema_builder = Schema::builder();
 
     // Our first field is title.
     // We want full-text search for it, and we also want
@@ -48,7 +45,7 @@ fn main() -> tantivy::Result<()> {
     //
     // `STORED` means that the field will also be saved
     // in a compressed, row-oriented key-value store.
-    // This store is useful to reconstruct the
+    // This store is useful for reconstructing the
     // documents that were selected during the search phase.
     schema_builder.add_text_field("title", TEXT | STORED);
 
@@ -57,8 +54,7 @@ fn main() -> tantivy::Result<()> {
     // need to be able to be able to retrieve it
     // for our application.
     //
-    // We can make our index lighter and
-    // by omitting `STORED` flag.
+    // We can make our index lighter by omitting the `STORED` flag.
     schema_builder.add_text_field("body", TEXT);
 
     let schema = schema_builder.build();
@@ -71,7 +67,7 @@ fn main() -> tantivy::Result<()> {
     // with our schema in the directory.
     let index = Index::create_in_dir(&index_path, schema.clone())?;
 
-    // To insert document we need an index writer.
+    // To insert a document we will need an index writer.
     // There must be only one writer at a time.
     // This single `IndexWriter` is already
     // multithreaded.
@@ -105,37 +101,25 @@ fn main() -> tantivy::Result<()> {
     // For convenience, tantivy also comes with a macro to
     // reduce the boilerplate above.
     index_writer.add_document(doc!(
-        title => "Of Mice and Men",
-        body => "A few miles south of Soledad, the Salinas River drops in close to the hillside \
-                bank and runs deep and green. The water is warm too, for it has slipped twinkling \
-                over the yellow sands in the sunlight before reaching the narrow pool. On one \
-                side of the river the golden foothill slopes curve up to the strong and rocky \
-                Gabilan Mountains, but on the valley side the water is lined with trees—willows \
-                fresh and green with every spring, carrying in their lower leaf junctures the \
-                debris of the winter’s flooding; and sycamores with mottled, white, recumbent \
-                limbs and branches that arch over the pool"
-    ));
-
-    index_writer.add_document(doc!(
-        title => "Of Mice and Men",
-        body => "A few miles south of Soledad, the Salinas River drops in close to the hillside \
-                bank and runs deep and green. The water is warm too, for it has slipped twinkling \
-                over the yellow sands in the sunlight before reaching the narrow pool. On one \
-                side of the river the golden foothill slopes curve up to the strong and rocky \
-                Gabilan Mountains, but on the valley side the water is lined with trees—willows \
-                fresh and green with every spring, carrying in their lower leaf junctures the \
-                debris of the winter’s flooding; and sycamores with mottled, white, recumbent \
-                limbs and branches that arch over the pool"
+    title => "Of Mice and Men",
+    body => "A few miles south of Soledad, the Salinas River drops in close to the hillside \
+            bank and runs deep and green. The water is warm too, for it has slipped twinkling \
+            over the yellow sands in the sunlight before reaching the narrow pool. On one \
+            side of the river the golden foothill slopes curve up to the strong and rocky \
+            Gabilan Mountains, but on the valley side the water is lined with trees—willows \
+            fresh and green with every spring, carrying in their lower leaf junctures the \
+            debris of the winter’s flooding; and sycamores with mottled, white, recumbent \
+            limbs and branches that arch over the pool"
     ));
 
     // Multivalued field just need to be repeated.
     index_writer.add_document(doc!(
-       title => "Frankenstein",
-       title => "The Modern Prometheus",
-       body => "You will rejoice to hear that no disaster has accompanied the commencement of an \
-                enterprise which you have regarded with such evil forebodings.  I arrived here \
-                yesterday, and my first task is to assure my dear sister of my welfare and \
-                increasing confidence in the success of my undertaking."
+    title => "Frankenstein",
+    title => "The Modern Prometheus",
+    body => "You will rejoice to hear that no disaster has accompanied the commencement of an \
+             enterprise which you have regarded with such evil forebodings.  I arrived here \
+             yesterday, and my first task is to assure my dear sister of my welfare and \
+             increasing confidence in the success of my undertaking."
     ));
 
     // This is an example, so we will only index 3 documents
@@ -149,8 +133,8 @@ fn main() -> tantivy::Result<()> {
     // At this point our documents are not searchable.
     //
     //
-    // We need to call .commit() explicitly to force the
-    // index_writer to finish processing the documents in the queue,
+    // We need to call `.commit()` explicitly to force the
+    // `index_writer` to finish processing the documents in the queue,
     // flush the current index to the disk, and advertise
     // the existence of new documents.
     //
@@ -162,31 +146,40 @@ fn main() -> tantivy::Result<()> {
     // persistently indexed.
     //
     // In the scenario of a crash or a power failure,
-    // tantivy behaves as if has rolled back to its last
+    // tantivy behaves as if it has rolled back to its last
     // commit.
 
     // # Searching
     //
     // ### Searcher
     //
-    // Let's search our index. Start by reloading
-    // searchers in the index. This should be done
-    // after every `commit()`.
-    index.load_searchers()?;
+    // A reader is required first in order to search an index.
+    // It acts as a `Searcher` pool that reloads itself,
+    // depending on a `ReloadPolicy`.
+    //
+    // For a search server you will typically create one reader for the entire lifetime of your
+    // program, and acquire a new searcher for every single request.
+    //
+    // In the code below, we rely on the 'ON_COMMIT' policy: the reader
+    // will reload the index automatically after each commit.
+    let reader = index
+        .reader_builder()
+        .reload_policy(ReloadPolicy::OnCommit)
+        .try_into()?;
 
     // We now need to acquire a searcher.
-    // Some search experience might require more than
-    // one query.
     //
-    // The searcher ensure that we get to work
-    // with a consistent version of the index.
+    // A searcher points to a snapshotted, immutable version of the index.
+    //
+    // Some search experience might require more than
+    // one query. Using the same searcher ensures that all of these queries will run on the
+    // same version of the index.
     //
     // Acquiring a `searcher` is very cheap.
     //
-    // You should acquire a searcher every time you
-    // start processing a request and
+    // You should acquire a searcher every time you start processing a request and
     // and release it right after your query is finished.
-    let searcher = index.searcher();
+    let searcher = reader.searcher();
 
     // ### Query
 
@@ -196,7 +189,7 @@ fn main() -> tantivy::Result<()> {
     // in both title and body.
     let query_parser = QueryParser::for_index(&index, vec![title, body]);
 
-    // QueryParser may fail if the query is not in the right
+    // `QueryParser` may fail if the query is not in the right
     // format. For user facing applications, this can be a problem.
     // A ticket has been opened regarding this problem.
     let query = query_parser.parse_query("sea whale")?;
@@ -212,15 +205,10 @@ fn main() -> tantivy::Result<()> {
     //
     // We are not interested in all of the documents but
     // only in the top 10. Keeping track of our top 10 best documents
-    // is the role of the TopCollector.
-    let mut top_collector = TopCollector::with_limit(10);
+    // is the role of the `TopDocs` collector.
 
     // We can now perform our query.
-    searcher.search(&*query, &mut top_collector)?;
-
-    // Our top collector now contains the 10
-    // most relevant doc ids...
-    let doc_addresses = top_collector.docs();
+    let top_docs = searcher.search(&query, &TopDocs::with_limit(10))?;
 
     // The actual documents still need to be
     // retrieved from Tantivy's store.
@@ -228,13 +216,10 @@ fn main() -> tantivy::Result<()> {
     // Since the body field was not configured as stored,
     // the document returned will only contain
     // a title.
-
-    for doc_address in doc_addresses {
+    for (_score, doc_address) in top_docs {
         let retrieved_doc = searcher.doc(doc_address)?;
         println!("{}", schema.to_json(&retrieved_doc));
     }
 
     Ok(())
 }
-
-use tempdir::TempDir;

examples/custom_collector.rs

@@ -0,0 +1,180 @@
+// # Custom collector example
+//
+// This example shows how you can implement your own
+// collector. As an example, we will compute a collector
+// that computes the standard deviation of a given fast field.
+//
+// Of course, you can have a look at the tantivy's built-in collectors
+// such as the `CountCollector` for more examples.
+
+// ---
+// Importing tantivy...
+use tantivy::collector::{Collector, SegmentCollector};
+use tantivy::fastfield::{DynamicFastFieldReader, FastFieldReader};
+use tantivy::query::QueryParser;
+use tantivy::schema::Field;
+use tantivy::schema::{Schema, FAST, INDEXED, TEXT};
+use tantivy::{doc, Index, Score, SegmentReader};
+
+#[derive(Default)]
+struct Stats {
+    count: usize,
+    sum: f64,
+    squared_sum: f64,
+}
+
+impl Stats {
+    pub fn count(&self) -> usize {
+        self.count
+    }
+
+    pub fn mean(&self) -> f64 {
+        self.sum / (self.count as f64)
+    }
+
+    fn square_mean(&self) -> f64 {
+        self.squared_sum / (self.count as f64)
+    }
+
+    pub fn standard_deviation(&self) -> f64 {
+        let mean = self.mean();
+        (self.square_mean() - mean * mean).sqrt()
+    }
+
+    fn non_zero_count(self) -> Option<Stats> {
+        if self.count == 0 {
+            None
+        } else {
+            Some(self)
+        }
+    }
+}
+
+struct StatsCollector {
+    field: Field,
+}
+
+impl StatsCollector {
+    fn with_field(field: Field) -> StatsCollector {
+        StatsCollector { field }
+    }
+}
+
+impl Collector for StatsCollector {
+    // That's the type of our result.
+    // Our standard deviation will be a float.
+    type Fruit = Option<Stats>;
+
+    type Child = StatsSegmentCollector;
+
+    fn for_segment(
+        &self,
+        _segment_local_id: u32,
+        segment_reader: &SegmentReader,
+    ) -> tantivy::Result<StatsSegmentCollector> {
+        let fast_field_reader = segment_reader.fast_fields().u64(self.field)?;
+        Ok(StatsSegmentCollector {
+            fast_field_reader,
+            stats: Stats::default(),
+        })
+    }
+
+    fn requires_scoring(&self) -> bool {
+        // this collector does not care about score.
+        false
+    }
+
+    fn merge_fruits(&self, segment_stats: Vec<Option<Stats>>) -> tantivy::Result<Option<Stats>> {
+        let mut stats = Stats::default();
+        for segment_stats in segment_stats.into_iter().flatten() {
+            stats.count += segment_stats.count;
+            stats.sum += segment_stats.sum;
+            stats.squared_sum += segment_stats.squared_sum;
+        }
+        Ok(stats.non_zero_count())
+    }
+}
+
+struct StatsSegmentCollector {
+    fast_field_reader: DynamicFastFieldReader<u64>,
+    stats: Stats,
+}
+
+impl SegmentCollector for StatsSegmentCollector {
+    type Fruit = Option<Stats>;
+
+    fn collect(&mut self, doc: u32, _score: Score) {
+        let value = self.fast_field_reader.get(doc) as f64;
+        self.stats.count += 1;
+        self.stats.sum += value;
+        self.stats.squared_sum += value * value;
+    }
+
+    fn harvest(self) -> <Self as SegmentCollector>::Fruit {
+        self.stats.non_zero_count()
+    }
+}
+
+fn main() -> tantivy::Result<()> {
+    // # Defining the schema
+    //
+    // The Tantivy index requires a very strict schema.
+    // The schema declares which fields are in the index,
+    // and for each field, its type and "the way it should
+    // be indexed".
+
+    // first we need to define a schema ...
+    let mut schema_builder = Schema::builder();
+
+    // We'll assume a fictional index containing
+    // products, and with a name, a description, and a price.
+    let product_name = schema_builder.add_text_field("name", TEXT);
+    let product_description = schema_builder.add_text_field("description", TEXT);
+    let price = schema_builder.add_u64_field("price", INDEXED | FAST);
+    let schema = schema_builder.build();
+
+    // # Indexing documents
+    //
+    // Lets index a bunch of fake documents for the sake of
+    // this example.
+    let index = Index::create_in_ram(schema);
+
+    let mut index_writer = index.writer(50_000_000)?;
+    index_writer.add_document(doc!(
+        product_name => "Super Broom 2000",
+        product_description => "While it is ok for short distance travel, this broom \
+        was designed quiditch. It will up your game.",
+        price => 30_200u64
+    ));
+    index_writer.add_document(doc!(
+        product_name => "Turbulobroom",
+        product_description => "You might have heard of this broom before : it is the sponsor of the Wales team.\
+            You'll enjoy its sharp turns, and rapid acceleration",
+        price => 29_240u64
+    ));
+    index_writer.add_document(doc!(
+        product_name => "Broomio",
+        product_description => "Great value for the price. This broom is a market favorite",
+        price => 21_240u64
+    ));
+    index_writer.add_document(doc!(
+        product_name => "Whack a Mole",
+        product_description => "Prime quality bat.",
+        price => 5_200u64
+    ));
+    index_writer.commit()?;
+
+    let reader = index.reader()?;
+    let searcher = reader.searcher();
+    let query_parser = QueryParser::for_index(&index, vec![product_name, product_description]);
+
+    // here we want to get a hit on the 'ken' in Frankenstein
+    let query = query_parser.parse_query("broom")?;
+    if let Some(stats) = searcher.search(&query, &StatsCollector::with_field(price))? {
+        println!("count: {}", stats.count());
+        println!("mean: {}", stats.mean());
+        println!("standard deviation: {}", stats.standard_deviation());
+    }
+
+    Ok(())
+}

examples/custom_tokenizer.rs

@@ -2,14 +2,11 @@
 //
 // In this example, we'll see how to define a tokenizer pipeline
 // by aligning a bunch of `TokenFilter`.
-
-#[macro_use]
-extern crate tantivy;
-use tantivy::collector::TopCollector;
+use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
 use tantivy::schema::*;
 use tantivy::tokenizer::NgramTokenizer;
-use tantivy::Index;
+use tantivy::{doc, Index};
 
 fn main() -> tantivy::Result<()> {
     // # Defining the schema
@@ -20,7 +17,7 @@ fn main() -> tantivy::Result<()> {
     // be indexed".
 
     // first we need to define a schema ...
-    let mut schema_builder = SchemaBuilder::default();
+    let mut schema_builder = Schema::builder();
 
     // Our first field is title.
     // In this example we want to use NGram searching
@@ -68,12 +65,12 @@ fn main() -> tantivy::Result<()> {
     // heap for the indexer can increase its throughput.
     let mut index_writer = index.writer(50_000_000)?;
     index_writer.add_document(doc!(
-        title => "The Old Man and the Sea",
-        body => "He was an old man who fished alone in a skiff in the Gulf Stream and \
-         he had gone eighty-four days now without taking a fish."
+    title => "The Old Man and the Sea",
+    body => "He was an old man who fished alone in a skiff in the Gulf Stream and \
+     he had gone eighty-four days now without taking a fish."
     ));
     index_writer.add_document(doc!(
-       title => "Of Mice and Men",
+    title => "Of Mice and Men",
        body => r#"A few miles south of Soledad, the Salinas River drops in close to the hillside
                 bank and runs deep and green. The water is warm too, for it has slipped twinkling
                 over the yellow sands in the sunlight before reaching the narrow pool. On one
@@ -84,16 +81,16 @@ fn main() -> tantivy::Result<()> {
                 limbs and branches that arch over the pool"#
     ));
     index_writer.add_document(doc!(
-        title => "Frankenstein",
+    title => "Frankenstein",
         body => r#"You will rejoice to hear that no disaster has accompanied the commencement of an
                 enterprise which you have regarded with such evil forebodings.  I arrived here
                 yesterday, and my first task is to assure my dear sister of my welfare and
                 increasing confidence in the success of my undertaking."#
     ));
     index_writer.commit()?;
-    index.load_searchers()?;
 
-    let searcher = index.searcher();
+    let reader = index.reader()?;
+    let searcher = reader.searcher();
 
     // The query parser can interpret human queries.
     // Here, if the user does not specify which
@@ -104,11 +101,9 @@ fn main() -> tantivy::Result<()> {
     // here we want to get a hit on the 'ken' in Frankenstein
     let query = query_parser.parse_query("ken")?;
 
-    let mut top_collector = TopCollector::with_limit(10);
-    searcher.search(&*query, &mut top_collector)?;
+    let top_docs = searcher.search(&query, &TopDocs::with_limit(10))?;
 
-    let doc_addresses = top_collector.docs();
-    for doc_address in doc_addresses {
+    for (_, doc_address) in top_docs {
         let retrieved_doc = searcher.doc(doc_address)?;
         println!("{}", schema.to_json(&retrieved_doc));
     }

examples/deleting_updating_documents.rs

@@ -8,18 +8,19 @@
 //
 // ---
 // Importing tantivy...
-#[macro_use]
-extern crate tantivy;
-use tantivy::collector::TopCollector;
+use tantivy::collector::TopDocs;
 use tantivy::query::TermQuery;
 use tantivy::schema::*;
-use tantivy::Index;
+use tantivy::{doc, Index, IndexReader};
 
 // A simple helper function to fetch a single document
 // given its id from our index.
 // It will be helpful to check our work.
-fn extract_doc_given_isbn(index: &Index, isbn_term: &Term) -> tantivy::Result<Option<Document>> {
-    let searcher = index.searcher();
+fn extract_doc_given_isbn(
+    reader: &IndexReader,
+    isbn_term: &Term,
+) -> tantivy::Result<Option<Document>> {
+    let searcher = reader.searcher();
 
     // This is the simplest query you can think of.
     // It matches all of the documents containing a specific term.
@@ -27,10 +28,9 @@ fn extract_doc_given_isbn(index: &Index, isbn_term: &Term) -> tantivy::Result<Op
     // The second argument is here to tell we don't care about decoding positions,
     // or term frequencies.
     let term_query = TermQuery::new(isbn_term.clone(), IndexRecordOption::Basic);
-    let mut top_collector = TopCollector::with_limit(1);
-    searcher.search(&term_query, &mut top_collector)?;
+    let top_docs = searcher.search(&term_query, &TopDocs::with_limit(1))?;
 
-    if let Some(doc_address) = top_collector.docs().first() {
+    if let Some((_score, doc_address)) = top_docs.first() {
         let doc = searcher.doc(*doc_address)?;
         Ok(Some(doc))
     } else {
@@ -44,7 +44,7 @@ fn main() -> tantivy::Result<()> {
     //
     // Check out the *basic_search* example if this makes
     // small sense to you.
-    let mut schema_builder = SchemaBuilder::default();
+    let mut schema_builder = Schema::builder();
 
     // Tantivy does not really have a notion of primary id.
     // This may change in the future.
@@ -86,12 +86,12 @@ fn main() -> tantivy::Result<()> {
        isbn => "978-9176370711",
     ));
     index_writer.commit()?;
-    index.load_searchers()?;
+    let reader = index.reader()?;
 
     let frankenstein_isbn = Term::from_field_text(isbn, "978-9176370711");
 
-    // Oops our frankenstein doc seems mispelled
-    let frankenstein_doc_misspelled = extract_doc_given_isbn(&index, &frankenstein_isbn)?.unwrap();
+    // Oops our frankenstein doc seems misspelled
+    let frankenstein_doc_misspelled = extract_doc_given_isbn(&reader, &frankenstein_isbn)?.unwrap();
     assert_eq!(
         schema.to_json(&frankenstein_doc_misspelled),
         r#"{"isbn":["978-9176370711"],"title":["Frankentein"]}"#,
@@ -130,10 +130,10 @@ fn main() -> tantivy::Result<()> {
     // Everything happened as if the document was updated.
     index_writer.commit()?;
     // We reload our searcher to make our change available to clients.
-    index.load_searchers()?;
+    reader.reload()?;
 
     // No more typo!
-    let frankenstein_new_doc = extract_doc_given_isbn(&index, &frankenstein_isbn)?.unwrap();
+    let frankenstein_new_doc = extract_doc_given_isbn(&reader, &frankenstein_isbn)?.unwrap();
     assert_eq!(
         schema.to_json(&frankenstein_new_doc),
         r#"{"isbn":["978-9176370711"],"title":["Frankenstein"]}"#,

examples/faceted_search.rs

@@ -10,72 +10,103 @@
 // - search for the best document matchings "sea whale"
 // - retrieve the best document original content.
 
-extern crate tempdir;
-
 // ---
 // Importing tantivy...
-#[macro_use]
-extern crate tantivy;
 use tantivy::collector::FacetCollector;
-use tantivy::query::AllQuery;
+use tantivy::query::{AllQuery, TermQuery};
 use tantivy::schema::*;
-use tantivy::Index;
+use tantivy::{doc, Index};
 
 fn main() -> tantivy::Result<()> {
-    // Let's create a temporary directory for the
-    // sake of this example
-    let index_path = TempDir::new("tantivy_facet_example_dir")?;
-    let mut schema_builder = SchemaBuilder::default();
+    // Let's create a temporary directory for the sake of this example
+    let mut schema_builder = Schema::builder();
 
-    schema_builder.add_text_field("name", TEXT | STORED);
-
-    // this is our faceted field
-    schema_builder.add_facet_field("tags");
+    let name = schema_builder.add_text_field("felin_name", TEXT | STORED);
+    // this is our faceted field: its scientific classification
+    let classification = schema_builder.add_facet_field("classification", INDEXED);
 
     let schema = schema_builder.build();
+    let index = Index::create_in_ram(schema);
 
-    let index = Index::create_in_dir(&index_path, schema.clone())?;
-
-    let mut index_writer = index.writer(50_000_000)?;
-
-    let name = schema.get_field("name").unwrap();
-    let tags = schema.get_field("tags").unwrap();
+    let mut index_writer = index.writer(30_000_000)?;
 
     // For convenience, tantivy also comes with a macro to
     // reduce the boilerplate above.
     index_writer.add_document(doc!(
-        name => "the ditch",
-        tags => Facet::from("/pools/north")
+        name => "Cat",
+        classification => Facet::from("/Felidae/Felinae/Felis")
     ));
-
     index_writer.add_document(doc!(
-        name => "little stacey",
-        tags => Facet::from("/pools/south")
+        name => "Canada lynx",
+        classification => Facet::from("/Felidae/Felinae/Lynx")
+    ));
+    index_writer.add_document(doc!(
+        name => "Cheetah",
+        classification => Facet::from("/Felidae/Felinae/Acinonyx")
+    ));
+    index_writer.add_document(doc!(
+        name => "Tiger",
+        classification => Facet::from("/Felidae/Pantherinae/Panthera")
+    ));
+    index_writer.add_document(doc!(
+        name => "Lion",
+        classification => Facet::from("/Felidae/Pantherinae/Panthera")
+    ));
+    index_writer.add_document(doc!(
+        name => "Jaguar",
+        classification => Facet::from("/Felidae/Pantherinae/Panthera")
+    ));
+    index_writer.add_document(doc!(
+        name => "Sunda clouded leopard",
+        classification => Facet::from("/Felidae/Pantherinae/Neofelis")
+    ));
+    index_writer.add_document(doc!(
+        name => "Fossa",
+        classification => Facet::from("/Eupleridae/Cryptoprocta")
     ));
-
     index_writer.commit()?;
 
-    index.load_searchers()?;
+    let reader = index.reader()?;
+    let searcher = reader.searcher();
+    {
+        let mut facet_collector = FacetCollector::for_field(classification);
+        facet_collector.add_facet("/Felidae");
+        let facet_counts = searcher.search(&AllQuery, &facet_collector)?;
+        // This lists all of the facet counts, right below "/Felidae".
+        let facets: Vec<(&Facet, u64)> = facet_counts.get("/Felidae").collect();
+        assert_eq!(
+            facets,
+            vec![
+                (&Facet::from("/Felidae/Felinae"), 3),
+                (&Facet::from("/Felidae/Pantherinae"), 4),
+            ]
+        );
+    }
 
-    let searcher = index.searcher();
+    // Facets are also searchable.
+    //
+    // For instance a common UI pattern is to allow the user someone to click on a facet link
+    // (e.g: `Pantherinae`) to drill down and filter the current result set with this subfacet.
+    //
+    // The search would then look as follows.
 
-    let mut facet_collector = FacetCollector::for_field(tags);
-    facet_collector.add_facet("/pools");
-
-    searcher.search(&AllQuery, &mut facet_collector).unwrap();
-
-    let counts = facet_collector.harvest();
-    // This lists all of the facet counts
-    let facets: Vec<(&Facet, u64)> = counts.get("/pools").collect();
-    assert_eq!(
-        facets,
-        vec![
-            (&Facet::from("/pools/north"), 1),
-            (&Facet::from("/pools/south"), 1),
-        ]
-    );
+    // Check the reference doc for different ways to create a `Facet` object.
+    {
+        let facet = Facet::from("/Felidae/Pantherinae");
+        let facet_term = Term::from_facet(classification, &facet);
+        let facet_term_query = TermQuery::new(facet_term, IndexRecordOption::Basic);
+        let mut facet_collector = FacetCollector::for_field(classification);
+        facet_collector.add_facet("/Felidae/Pantherinae");
+        let facet_counts = searcher.search(&facet_term_query, &facet_collector)?;
+        let facets: Vec<(&Facet, u64)> = facet_counts.get("/Felidae/Pantherinae").collect();
+        assert_eq!(
+            facets,
+            vec![
+                (&Facet::from("/Felidae/Pantherinae/Neofelis"), 1),
+                (&Facet::from("/Felidae/Pantherinae/Panthera"), 3),
+            ]
+        );
+    }
 
     Ok(())
 }
-
-use tempdir::TempDir;

examples/faceted_search_with_tweaked_score.rs

@@ -0,0 +1,98 @@
+use std::collections::HashSet;
+use tantivy::collector::TopDocs;
+use tantivy::doc;
+use tantivy::query::BooleanQuery;
+use tantivy::schema::*;
+use tantivy::{DocId, Index, Score, SegmentReader};
+
+fn main() -> tantivy::Result<()> {
+    let mut schema_builder = Schema::builder();
+
+    let title = schema_builder.add_text_field("title", STORED);
+    let ingredient = schema_builder.add_facet_field("ingredient", INDEXED);
+
+    let schema = schema_builder.build();
+    let index = Index::create_in_ram(schema);
+
+    let mut index_writer = index.writer(30_000_000)?;
+
+    index_writer.add_document(doc!(
+        title => "Fried egg",
+        ingredient => Facet::from("/ingredient/egg"),
+        ingredient => Facet::from("/ingredient/oil"),
+    ));
+    index_writer.add_document(doc!(
+        title => "Scrambled egg",
+        ingredient => Facet::from("/ingredient/egg"),
+        ingredient => Facet::from("/ingredient/butter"),
+        ingredient => Facet::from("/ingredient/milk"),
+        ingredient => Facet::from("/ingredient/salt"),
+    ));
+    index_writer.add_document(doc!(
+        title => "Egg rolls",
+        ingredient => Facet::from("/ingredient/egg"),
+        ingredient => Facet::from("/ingredient/garlic"),
+        ingredient => Facet::from("/ingredient/salt"),
+        ingredient => Facet::from("/ingredient/oil"),
+        ingredient => Facet::from("/ingredient/tortilla-wrap"),
+        ingredient => Facet::from("/ingredient/mushroom"),
+    ));
+    index_writer.commit()?;
+
+    let reader = index.reader()?;
+    let searcher = reader.searcher();
+    {
+        let facets = vec![
+            Facet::from("/ingredient/egg"),
+            Facet::from("/ingredient/oil"),
+            Facet::from("/ingredient/garlic"),
+            Facet::from("/ingredient/mushroom"),
+        ];
+        let query = BooleanQuery::new_multiterms_query(
+            facets
+                .iter()
+                .map(|key| Term::from_facet(ingredient, key))
+                .collect(),
+        );
+        let top_docs_by_custom_score =
+            TopDocs::with_limit(2).tweak_score(move |segment_reader: &SegmentReader| {
+                let ingredient_reader = segment_reader.facet_reader(ingredient).unwrap();
+                let facet_dict = ingredient_reader.facet_dict();
+
+                let query_ords: HashSet<u64> = facets
+                    .iter()
+                    .filter_map(|key| facet_dict.term_ord(key.encoded_str()).unwrap())
+                    .collect();
+
+                let mut facet_ords_buffer: Vec<u64> = Vec::with_capacity(20);
+
+                move |doc: DocId, original_score: Score| {
+                    ingredient_reader.facet_ords(doc, &mut facet_ords_buffer);
+                    let missing_ingredients = facet_ords_buffer
+                        .iter()
+                        .filter(|ord| !query_ords.contains(ord))
+                        .count();
+                    let tweak = 1.0 / 4_f32.powi(missing_ingredients as i32);
+
+                    original_score * tweak
+                }
+            });
+        let top_docs = searcher.search(&query, &top_docs_by_custom_score)?;
+
+        let titles: Vec<String> = top_docs
+            .iter()
+            .map(|(_, doc_id)| {
+                searcher
+                    .doc(*doc_id)
+                    .unwrap()
+                    .get_first(title)
+                    .unwrap()
+                    .text()
+                    .unwrap()
+                    .to_owned()
+            })
+            .collect();
+        assert_eq!(titles, vec!["Fried egg", "Egg rolls"]);
+    }
+    Ok(())
+}

examples/integer_range_search.rs

@@ -0,0 +1,39 @@
+// # Searching a range on an indexed int field.
+//
+// Below is an example of creating an indexed integer field in your schema
+// You can use RangeQuery to get a Count of all occurrences in a given range.
+use tantivy::collector::Count;
+use tantivy::query::RangeQuery;
+use tantivy::schema::{Schema, INDEXED};
+use tantivy::{doc, Index, Result};
+
+fn run() -> Result<()> {
+    // For the sake of simplicity, this schema will only have 1 field
+    let mut schema_builder = Schema::builder();
+
+    // `INDEXED` is a short-hand to indicate that our field should be "searchable".
+    let year_field = schema_builder.add_u64_field("year", INDEXED);
+    let schema = schema_builder.build();
+    let index = Index::create_in_ram(schema);
+    let reader = index.reader()?;
+    {
+        let mut index_writer = index.writer_with_num_threads(1, 6_000_000)?;
+        for year in 1950u64..2019u64 {
+            index_writer.add_document(doc!(year_field => year));
+        }
+        index_writer.commit()?;
+        // The index will be a range of years
+    }
+    reader.reload()?;
+    let searcher = reader.searcher();
+    // The end is excluded i.e. here we are searching up to 1969
+    let docs_in_the_sixties = RangeQuery::new_u64(year_field, 1960..1970);
+    // Uses a Count collector to sum the total number of docs in the range
+    let num_60s_books = searcher.search(&docs_in_the_sixties, &Count)?;
+    assert_eq!(num_60s_books, 10);
+    Ok(())
+}
+
+fn main() {
+    run().unwrap()
+}

examples/iterating_docs_and_positions.rs

@@ -1,4 +1,4 @@
-// # Iterating docs and positioms.
+// # Iterating docs and positions.
 //
 // At its core of tantivy, relies on a data structure
 // called an inverted index.
@@ -9,23 +9,20 @@
 
 // ---
 // Importing tantivy...
-#[macro_use]
-extern crate tantivy;
 use tantivy::schema::*;
-use tantivy::Index;
-use tantivy::{DocId, DocSet, Postings};
+use tantivy::{doc, DocSet, Index, Postings, TERMINATED};
 
 fn main() -> tantivy::Result<()> {
     // We first create a schema for the sake of the
     // example. Check the `basic_search` example for more information.
-    let mut schema_builder = SchemaBuilder::default();
+    let mut schema_builder = Schema::builder();
 
     // For this example, we need to make sure to index positions for our title
     // field. `TEXT` precisely does this.
     let title = schema_builder.add_text_field("title", TEXT | STORED);
     let schema = schema_builder.build();
 
-    let index = Index::create_in_ram(schema.clone());
+    let index = Index::create_in_ram(schema);
 
     let mut index_writer = index.writer_with_num_threads(1, 50_000_000)?;
     index_writer.add_document(doc!(title => "The Old Man and the Sea"));
@@ -33,9 +30,9 @@ fn main() -> tantivy::Result<()> {
     index_writer.add_document(doc!(title => "The modern Promotheus"));
     index_writer.commit()?;
 
-    index.load_searchers()?;
+    let reader = index.reader()?;
 
-    let searcher = index.searcher();
+    let searcher = reader.searcher();
 
     // A tantivy index is actually a collection of segments.
     // Similarly, a searcher just wraps a list `segment_reader`.
@@ -48,7 +45,7 @@ fn main() -> tantivy::Result<()> {
         // Inverted index stands for the combination of
         // - the term dictionary
         // - the inverted lists associated to each terms and their positions
-        let inverted_index = segment_reader.inverted_index(title);
+        let inverted_index = segment_reader.inverted_index(title)?;
 
         // A `Term` is a text token associated with a field.
         // Let's go through all docs containing the term `title:the` and access their position
@@ -61,16 +58,15 @@ fn main() -> tantivy::Result<()> {
         // If you don't need all this information, you may get better performance by decompressing less
         // information.
         if let Some(mut segment_postings) =
-            inverted_index.read_postings(&term_the, IndexRecordOption::WithFreqsAndPositions)
+            inverted_index.read_postings(&term_the, IndexRecordOption::WithFreqsAndPositions)?
         {
             // this buffer will be used to request for positions
             let mut positions: Vec<u32> = Vec::with_capacity(100);
-            while segment_postings.advance() {
-                // the number of time the term appears in the document.
-                let doc_id: DocId = segment_postings.doc(); //< do not try to access this before calling advance once.
-
+            let mut doc_id = segment_postings.doc();
+            while doc_id != TERMINATED {
                 // This MAY contains deleted documents as well.
                 if segment_reader.is_deleted(doc_id) {
+                    doc_id = segment_postings.advance();
                     continue;
                 }
 
@@ -89,6 +85,7 @@ fn main() -> tantivy::Result<()> {
                 // Doc 2: TermFreq 1: [0]
                 // ```
                 println!("Doc {}: TermFreq {}: {:?}", doc_id, term_freq, positions);
+                doc_id = segment_postings.advance();
             }
         }
     }
@@ -109,7 +106,7 @@ fn main() -> tantivy::Result<()> {
         // Inverted index stands for the combination of
         // - the term dictionary
         // - the inverted lists associated to each terms and their positions
-        let inverted_index = segment_reader.inverted_index(title);
+        let inverted_index = segment_reader.inverted_index(title)?;
 
         // This segment posting object is like a cursor over the documents matching the term.
         // The `IndexRecordOption` arguments tells tantivy we will be interested in both term frequencies
@@ -118,13 +115,18 @@ fn main() -> tantivy::Result<()> {
         // If you don't need all this information, you may get better performance by decompressing less
         // information.
         if let Some(mut block_segment_postings) =
-            inverted_index.read_block_postings(&term_the, IndexRecordOption::Basic)
+            inverted_index.read_block_postings(&term_the, IndexRecordOption::Basic)?
         {
-            while block_segment_postings.advance() {
+            loop {
+                let docs = block_segment_postings.docs();
+                if docs.is_empty() {
+                    break;
+                }
                 // Once again these docs MAY contains deleted documents as well.
                 let docs = block_segment_postings.docs();
                 // Prints `Docs [0, 2].`
                 println!("Docs {:?}", docs);
+                block_segment_postings.advance();
             }
         }
     }

examples/multiple_producer.rs

@@ -0,0 +1,100 @@
+// # Indexing from different threads.
+//
+// It is fairly common to have to index from different threads.
+// Tantivy forbids to create more than one `IndexWriter` at a time.
+//
+// This `IndexWriter` itself has its own multithreaded layer, so managing your own
+// indexing threads will not help. However, it can still be useful for some applications.
+//
+// For instance, if preparing documents to send to tantivy before indexing is the bottleneck of
+// your application, it is reasonable to have multiple threads.
+//
+// Another very common reason to want to index from multiple threads, is implementing a webserver
+// with CRUD capabilities. The server framework will most likely handle request from
+// different threads.
+//
+// The recommended way to address both of these use case is to wrap your `IndexWriter` into a
+// `Arc<RwLock<IndexWriter>>`.
+//
+// While this is counterintuitive, adding and deleting documents do not require mutability
+// over the `IndexWriter`, so several threads will be able to do this operation concurrently.
+//
+// The example below does not represent an actual real-life use case (who would spawn thread to
+// index a single document?), but aims at demonstrating the mechanism that makes indexing
+// from several threads possible.
+
+// ---
+// Importing tantivy...
+use std::sync::{Arc, RwLock};
+use std::thread;
+use std::time::Duration;
+use tantivy::schema::{Schema, STORED, TEXT};
+use tantivy::{doc, Index, IndexWriter, Opstamp};
+
+fn main() -> tantivy::Result<()> {
+    // # Defining the schema
+    let mut schema_builder = Schema::builder();
+    let title = schema_builder.add_text_field("title", TEXT | STORED);
+    let body = schema_builder.add_text_field("body", TEXT);
+    let schema = schema_builder.build();
+
+    let index = Index::create_in_ram(schema);
+    let index_writer: Arc<RwLock<IndexWriter>> = Arc::new(RwLock::new(index.writer(50_000_000)?));
+
+    // # First indexing thread.
+    let index_writer_clone_1 = index_writer.clone();
+    thread::spawn(move || {
+        // we index 100 times the document... for the sake of the example.
+        for i in 0..100 {
+            let opstamp = index_writer_clone_1
+                .read().unwrap() //< A read lock is sufficient here.
+                .add_document(
+                    doc!(
+                        title => "Of Mice and Men",
+                        body => "A few miles south of Soledad, the Salinas River drops in close to the hillside \
+                        bank and runs deep and green. The water is warm too, for it has slipped twinkling \
+                        over the yellow sands in the sunlight before reaching the narrow pool. On one \
+                        side of the river the golden foothill slopes curve up to the strong and rocky \
+                        Gabilan Mountains, but on the valley side the water is lined with trees—willows \
+                        fresh and green with every spring, carrying in their lower leaf junctures the \
+                        debris of the winter’s flooding; and sycamores with mottled, white, recumbent \
+                        limbs and branches that arch over the pool"
+                    ));
+            println!("add doc {} from thread 1 - opstamp {}", i, opstamp);
+            thread::sleep(Duration::from_millis(20));
+        }
+    });
+
+    // # Second indexing thread.
+    let index_writer_clone_2 = index_writer.clone();
+    // For convenience, tantivy also comes with a macro to
+    // reduce the boilerplate above.
+    thread::spawn(move || {
+        // we index 100 times the document... for the sake of the example.
+        for i in 0..100 {
+            // A read lock is sufficient here.
+            let opstamp = {
+                let index_writer_rlock = index_writer_clone_2.read().unwrap();
+                index_writer_rlock.add_document(doc!(
+                    title => "Manufacturing consent",
+                    body => "Some great book description..."
+                ))
+            };
+            println!("add doc {} from thread 2 - opstamp {}", i, opstamp);
+            thread::sleep(Duration::from_millis(10));
+        }
+    });
+
+    // # In the main thread, we commit 10 times, once every 500ms.
+    for _ in 0..10 {
+        let opstamp: Opstamp = {
+            // Committing or rollbacking on the other hand requires write lock. This will block other threads.
+            let mut index_writer_wlock = index_writer.write().unwrap();
+            index_writer_wlock.commit().unwrap()
+        };
+        println!("committed with opstamp {}", opstamp);
+        thread::sleep(Duration::from_millis(500));
+    }
+
+    Ok(())
+}

examples/pre_tokenized_text.rs

@@ -0,0 +1,139 @@
+// # Pre-tokenized text example
+//
+// This example shows how to use pre-tokenized text. Sometimes you might
+// want to index and search through text which is already split into
+// tokens by some external tool.
+//
+// In this example we will:
+// - use tantivy tokenizer to create tokens and load them directly into tantivy,
+// - import tokenized text straight from json,
+// - perform a search on documents with pre-tokenized text
+
+use tantivy::collector::{Count, TopDocs};
+use tantivy::query::TermQuery;
+use tantivy::schema::*;
+use tantivy::tokenizer::{PreTokenizedString, SimpleTokenizer, Token, Tokenizer};
+use tantivy::{doc, Index, ReloadPolicy};
+use tempfile::TempDir;
+
+fn pre_tokenize_text(text: &str) -> Vec<Token> {
+    let mut token_stream = SimpleTokenizer.token_stream(text);
+    let mut tokens = vec![];
+    while token_stream.advance() {
+        tokens.push(token_stream.token().clone());
+    }
+    tokens
+}
+
+fn main() -> tantivy::Result<()> {
+    let index_path = TempDir::new()?;
+
+    let mut schema_builder = Schema::builder();
+
+    schema_builder.add_text_field("title", TEXT | STORED);
+    schema_builder.add_text_field("body", TEXT);
+
+    let schema = schema_builder.build();
+
+    let index = Index::create_in_dir(&index_path, schema.clone())?;
+
+    let mut index_writer = index.writer(50_000_000)?;
+
+    // We can create a document manually, by setting the fields
+    // one by one in a Document object.
+    let title = schema.get_field("title").unwrap();
+    let body = schema.get_field("body").unwrap();
+
+    let title_text = "The Old Man and the Sea";
+    let body_text = "He was an old man who fished alone in a skiff in the Gulf Stream";
+
+    // Content of our first document
+    // We create `PreTokenizedString` which contains original text and vector of tokens
+    let title_tok = PreTokenizedString {
+        text: String::from(title_text),
+        tokens: pre_tokenize_text(title_text),
+    };
+
+    println!(
+        "Original text: \"{}\" and tokens: {:?}",
+        title_tok.text, title_tok.tokens
+    );
+
+    let body_tok = PreTokenizedString {
+        text: String::from(body_text),
+        tokens: pre_tokenize_text(body_text),
+    };
+
+    // Now lets create a document and add our `PreTokenizedString`
+    let old_man_doc = doc!(title => title_tok, body => body_tok);
+
+    // ... now let's just add it to the IndexWriter
+    index_writer.add_document(old_man_doc);
+
+    // Pretokenized text can also be fed as JSON
+    let short_man_json = r#"{
+        "title":[{
+            "text":"The Old Man",
+            "tokens":[
+                {"offset_from":0,"offset_to":3,"position":0,"text":"The","position_length":1},
+                {"offset_from":4,"offset_to":7,"position":1,"text":"Old","position_length":1},
+                {"offset_from":8,"offset_to":11,"position":2,"text":"Man","position_length":1}
+            ]
+        }]
+    }"#;
+
+    let short_man_doc = schema.parse_document(short_man_json)?;
+
+    index_writer.add_document(short_man_doc);
+
+    // Let's commit changes
+    index_writer.commit()?;
+
+    // ... and now is the time to query our index
+
+    let reader = index
+        .reader_builder()
+        .reload_policy(ReloadPolicy::OnCommit)
+        .try_into()?;
+
+    let searcher = reader.searcher();
+
+    // We want to get documents with token "Man", we will use TermQuery to do it
+    // Using PreTokenizedString means the tokens are stored as is avoiding stemming
+    // and lowercasing, which preserves full words in their original form
+    let query = TermQuery::new(
+        Term::from_field_text(title, "Man"),
+        IndexRecordOption::Basic,
+    );
+
+    let (top_docs, count) = searcher
+        .search(&query, &(TopDocs::with_limit(2), Count))
+        .unwrap();
+
+    assert_eq!(count, 2);
+
+    // Now let's print out the results.
+    // Note that the tokens are not stored along with the original text
+    // in the document store
+    for (_score, doc_address) in top_docs {
+        let retrieved_doc = searcher.doc(doc_address)?;
+        println!("Document: {}", schema.to_json(&retrieved_doc));
+    }
+
+    // In contrary to the previous query, when we search for the "man" term we
+    // should get no results, as it's not one of the indexed tokens. SimpleTokenizer
+    // only splits text on whitespace / punctuation.
+
+    let query = TermQuery::new(
+        Term::from_field_text(title, "man"),
+        IndexRecordOption::Basic,
+    );
+
+    let (_top_docs, count) = searcher
+        .search(&query, &(TopDocs::with_limit(2), Count))
+        .unwrap();
+
+    assert_eq!(count, 0);
+
+    Ok(())
+}

examples/snippet.rs

@@ -4,68 +4,79 @@
 // your hit result.
 // Snippet are an extracted of a target document, and returned in HTML format.
 // The keyword searched by the user are highlighted with a `<b>` tag.
-extern crate tempdir;
 
 // ---
 // Importing tantivy...
-#[macro_use]
-extern crate tantivy;
-use tantivy::collector::TopCollector;
+use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
 use tantivy::schema::*;
-use tantivy::Index;
-use tantivy::SnippetGenerator;
-use tempdir::TempDir;
+use tantivy::{doc, Index, Snippet, SnippetGenerator};
+use tempfile::TempDir;
 
 fn main() -> tantivy::Result<()> {
     // Let's create a temporary directory for the
     // sake of this example
-    let index_path = TempDir::new("tantivy_example_dir")?;
+    let index_path = TempDir::new()?;
 
     // # Defining the schema
-    let mut schema_builder = SchemaBuilder::default();
+    let mut schema_builder = Schema::builder();
     let title = schema_builder.add_text_field("title", TEXT | STORED);
     let body = schema_builder.add_text_field("body", TEXT | STORED);
     let schema = schema_builder.build();
 
     // # Indexing documents
-    let index = Index::create_in_dir(&index_path, schema.clone())?;
+    let index = Index::create_in_dir(&index_path, schema)?;
 
     let mut index_writer = index.writer(50_000_000)?;
 
     // we'll only need one doc for this example.
     index_writer.add_document(doc!(
-        title => "Of Mice and Men",
-        body => "A few miles south of Soledad, the Salinas River drops in close to the hillside \
-                bank and runs deep and green. The water is warm too, for it has slipped twinkling \
-                over the yellow sands in the sunlight before reaching the narrow pool. On one \
-                side of the river the golden foothill slopes curve up to the strong and rocky \
-                Gabilan Mountains, but on the valley side the water is lined with trees—willows \
-                fresh and green with every spring, carrying in their lower leaf junctures the \
-                debris of the winter’s flooding; and sycamores with mottled, white, recumbent \
-                limbs and branches that arch over the pool"
+    title => "Of Mice and Men",
+    body => "A few miles south of Soledad, the Salinas River drops in close to the hillside \
+            bank and runs deep and green. The water is warm too, for it has slipped twinkling \
+            over the yellow sands in the sunlight before reaching the narrow pool. On one \
+            side of the river the golden foothill slopes curve up to the strong and rocky \
+            Gabilan Mountains, but on the valley side the water is lined with trees—willows \
+            fresh and green with every spring, carrying in their lower leaf junctures the \
+            debris of the winter’s flooding; and sycamores with mottled, white, recumbent \
+            limbs and branches that arch over the pool"
     ));
     // ...
     index_writer.commit()?;
 
-    index.load_searchers()?;
-
-    let searcher = index.searcher();
+    let reader = index.reader()?;
+    let searcher = reader.searcher();
     let query_parser = QueryParser::for_index(&index, vec![title, body]);
     let query = query_parser.parse_query("sycamore spring")?;
 
-    let mut top_collector = TopCollector::with_limit(10);
-    searcher.search(&*query, &mut top_collector)?;
+    let top_docs = searcher.search(&query, &TopDocs::with_limit(10))?;
 
-    let snippet_generator = SnippetGenerator::new(&searcher, &*query, body)?;
+    let snippet_generator = SnippetGenerator::create(&searcher, &*query, body)?;
 
-    let doc_addresses = top_collector.docs();
-    for doc_address in doc_addresses {
+    for (score, doc_address) in top_docs {
         let doc = searcher.doc(doc_address)?;
         let snippet = snippet_generator.snippet_from_doc(&doc);
+        println!("Document score {}:", score);
         println!("title: {}", doc.get_first(title).unwrap().text().unwrap());
         println!("snippet: {}", snippet.to_html());
+        println!("custom highlighting: {}", highlight(snippet));
     }
 
     Ok(())
 }
+
+fn highlight(snippet: Snippet) -> String {
+    let mut result = String::new();
+    let mut start_from = 0;
+
+    for fragment_range in snippet.highlighted() {
+        result.push_str(&snippet.fragments()[start_from..fragment_range.start]);
+        result.push_str(" --> ");
+        result.push_str(&snippet.fragments()[fragment_range.clone()]);
+        result.push_str(" <-- ");
+        start_from = fragment_range.end;
+    }
+
+    result.push_str(&snippet.fragments()[start_from..]);
+    result
+}

examples/stop_words.rs

@@ -9,21 +9,17 @@
 // - add a few stop words
 // - index few documents in our index
 
-extern crate tempdir;
-
 // ---
 // Importing tantivy...
-#[macro_use]
-extern crate tantivy;
-use tantivy::collector::TopCollector;
+use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
 use tantivy::schema::*;
 use tantivy::tokenizer::*;
-use tantivy::Index;
+use tantivy::{doc, Index};
 
 fn main() -> tantivy::Result<()> {
     // this example assumes you understand the content in `basic_search`
-    let mut schema_builder = SchemaBuilder::default();
+    let mut schema_builder = Schema::builder();
 
     // This configures your custom options for how tantivy will
     // store and process your content in the index; The key
@@ -54,7 +50,7 @@ fn main() -> tantivy::Result<()> {
 
     // This tokenizer lowers all of the text (to help with stop word matching)
     // then removes all instances of `the` and `and` from the corpus
-    let tokenizer = SimpleTokenizer
+    let tokenizer = TextAnalyzer::from(SimpleTokenizer)
         .filter(LowerCaser)
         .filter(StopWordFilter::remove(vec![
             "the".to_string(),
@@ -72,48 +68,44 @@ fn main() -> tantivy::Result<()> {
     title => "The Old Man and the Sea",
     body => "He was an old man who fished alone in a skiff in the Gulf Stream and \
      he had gone eighty-four days now without taking a fish."
-  ));
+    ));
 
     index_writer.add_document(doc!(
-      title => "Of Mice and Men",
-      body => "A few miles south of Soledad, the Salinas River drops in close to the hillside \
-              bank and runs deep and green. The water is warm too, for it has slipped twinkling \
-              over the yellow sands in the sunlight before reaching the narrow pool. On one \
-              side of the river the golden foothill slopes curve up to the strong and rocky \
-              Gabilan Mountains, but on the valley side the water is lined with trees—willows \
-              fresh and green with every spring, carrying in their lower leaf junctures the \
-              debris of the winter’s flooding; and sycamores with mottled, white, recumbent \
-              limbs and branches that arch over the pool"
-  ));
+    title => "Of Mice and Men",
+    body => "A few miles south of Soledad, the Salinas River drops in close to the hillside \
+            bank and runs deep and green. The water is warm too, for it has slipped twinkling \
+            over the yellow sands in the sunlight before reaching the narrow pool. On one \
+            side of the river the golden foothill slopes curve up to the strong and rocky \
+            Gabilan Mountains, but on the valley side the water is lined with trees—willows \
+            fresh and green with every spring, carrying in their lower leaf junctures the \
+            debris of the winter’s flooding; and sycamores with mottled, white, recumbent \
+            limbs and branches that arch over the pool"
+    ));
 
     index_writer.add_document(doc!(
-       title => "Frankenstein",
-       body => "You will rejoice to hear that no disaster has accompanied the commencement of an \
-                enterprise which you have regarded with such evil forebodings.  I arrived here \
-                yesterday, and my first task is to assure my dear sister of my welfare and \
-                increasing confidence in the success of my undertaking."
+    title => "Frankenstein",
+    body => "You will rejoice to hear that no disaster has accompanied the commencement of an \
+             enterprise which you have regarded with such evil forebodings.  I arrived here \
+             yesterday, and my first task is to assure my dear sister of my welfare and \
+             increasing confidence in the success of my undertaking."
     ));
 
     index_writer.commit()?;
 
-    index.load_searchers()?;
+    let reader = index.reader()?;
 
-    let searcher = index.searcher();
+    let searcher = reader.searcher();
 
     let query_parser = QueryParser::for_index(&index, vec![title, body]);
 
     // stop words are applied on the query as well.
     // The following will be equivalent to `title:frankenstein`
     let query = query_parser.parse_query("title:\"the Frankenstein\"")?;
+    let top_docs = searcher.search(&query, &TopDocs::with_limit(10))?;
 
-    let mut top_collector = TopCollector::with_limit(10);
-
-    searcher.search(&*query, &mut top_collector)?;
-
-    let doc_addresses = top_collector.docs();
-
-    for doc_address in doc_addresses {
+    for (score, doc_address) in top_docs {
         let retrieved_doc = searcher.doc(doc_address)?;
+        println!("\n==\nDocument score {}:", score);
         println!("{}", schema.to_json(&retrieved_doc));
     }
 

examples/working_with_json.rs

@@ -1,4 +1,3 @@
-extern crate tantivy;
 use tantivy::schema::*;
 
 // # Document from json
@@ -9,10 +8,10 @@ fn main() -> tantivy::Result<()> {
     // Check out the basic example if this is confusing to you.
     //
     // first we need to define a schema ...
-    let mut schema_builder = SchemaBuilder::default();
+    let mut schema_builder = Schema::builder();
     schema_builder.add_text_field("title", TEXT | STORED);
     schema_builder.add_text_field("body", TEXT);
-    schema_builder.add_u64_field("year", INT_INDEXED);
+    schema_builder.add_u64_field("year", INDEXED);
     let schema = schema_builder.build();
 
     // Let's assume we have a json-serialized document.
@@ -22,7 +21,7 @@ fn main() -> tantivy::Result<()> {
     }"#;
 
     // We can parse our document
-    let _mice_and_men_doc = schema.parse_document(&mice_and_men_doc_json)?;
+    let _mice_and_men_doc = schema.parse_document(mice_and_men_doc_json)?;
 
     // Multi-valued field are allowed, they are
     // expressed in JSON by an array.
@@ -31,7 +30,7 @@ fn main() -> tantivy::Result<()> {
        "title": ["Frankenstein", "The Modern Prometheus"],
        "year": 1818
     }"#;
-    let _frankenstein_doc = schema.parse_document(&frankenstein_json)?;
+    let _frankenstein_doc = schema.parse_document(frankenstein_json)?;
 
     // Note that the schema is saved in your index directory.
     //

fastfield_codecs/Cargo.toml

@@ -0,0 +1,24 @@
+[package]
+name = "fastfield_codecs"
+version = "0.1.0"
+authors = ["Pascal Seitz <pascal@quickwit.io>"]
+license = "MIT"
+edition = "2018"
+description = "Fast field codecs used by tantivy"
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]
+common = { version = "0.1", path = "../common/", package = "tantivy-common" }
+tantivy-bitpacker = { version="0.1.1", path = "../bitpacker/" }
+prettytable-rs = {version="0.8.0", optional= true}
+rand = {version="0.8.3", optional= true}
+
+[dev-dependencies]
+more-asserts = "0.2.1"
+rand = "0.8.3"
+
+[features]
+bin = ["prettytable-rs", "rand"]
+default = ["bin"]
+

fastfield_codecs/README.md

@@ -0,0 +1,68 @@
+
+
+# Fast Field Codecs
+
+This crate contains various fast field codecs, used to compress/decompress fast field data in tantivy.
+
+## Contributing
+
+Contributing is pretty straightforward. Since the bitpacking is the simplest compressor, you can check it for reference.
+
+A codec needs to implement 2 traits:
+
+- A reader implementing `FastFieldCodecReader` to read the codec.
+- A serializer implementing `FastFieldCodecSerializer` for compression estimation and codec name + id.
+
+### Tests
+
+Once the traits are implemented test and benchmark integration is pretty easy (see `test_with_codec_data_sets` and `bench.rs`).
+
+Make sure to add the codec to the main.rs, which tests the compression ratio and estimation against different data sets. You can run it with:
+```
+cargo run --features bin
+```
+
+### TODO
+- Add real world data sets in comparison
+- Add codec to cover sparse data sets
+
+
+### Codec Comparison
+```
++----------------------------------+-------------------+------------------------+
+|                                  | Compression Ratio | Compression Estimation |
++----------------------------------+-------------------+------------------------+
+| Autoincrement                    |                   |                        |
++----------------------------------+-------------------+------------------------+
+| LinearInterpol                   | 0.000039572664    | 0.000004396963         |
++----------------------------------+-------------------+------------------------+
+| MultiLinearInterpol              | 0.1477348         | 0.17275847             |
++----------------------------------+-------------------+------------------------+
+| Bitpacked                        | 0.28126493        | 0.28125                |
++----------------------------------+-------------------+------------------------+
+| Monotonically increasing concave |                   |                        |
++----------------------------------+-------------------+------------------------+
+| LinearInterpol                   | 0.25003937        | 0.26562938             |
++----------------------------------+-------------------+------------------------+
+| MultiLinearInterpol              | 0.190665          | 0.1883836              |
++----------------------------------+-------------------+------------------------+
+| Bitpacked                        | 0.31251436        | 0.3125                 |
++----------------------------------+-------------------+------------------------+
+| Monotonically increasing convex  |                   |                        |
++----------------------------------+-------------------+------------------------+
+| LinearInterpol                   | 0.25003937        | 0.28125438             |
++----------------------------------+-------------------+------------------------+
+| MultiLinearInterpol              | 0.18676           | 0.2040086              |
++----------------------------------+-------------------+------------------------+
+| Bitpacked                        | 0.31251436        | 0.3125                 |
++----------------------------------+-------------------+------------------------+
+| Almost monotonically increasing  |                   |                        |
++----------------------------------+-------------------+------------------------+
+| LinearInterpol                   | 0.14066513        | 0.1562544              |
++----------------------------------+-------------------+------------------------+
+| MultiLinearInterpol              | 0.16335973        | 0.17275847             |
++----------------------------------+-------------------+------------------------+
+| Bitpacked                        | 0.28126493        | 0.28125                |
++----------------------------------+-------------------+------------------------+
+
+```

fastfield_codecs/benches/bench.rs

@@ -0,0 +1,108 @@
+#![feature(test)]
+
+extern crate test;
+
+#[cfg(test)]
+mod tests {
+    use fastfield_codecs::{
+        bitpacked::{BitpackedFastFieldReader, BitpackedFastFieldSerializer},
+        linearinterpol::{LinearInterpolFastFieldReader, LinearInterpolFastFieldSerializer},
+        multilinearinterpol::{
+            MultiLinearInterpolFastFieldReader, MultiLinearInterpolFastFieldSerializer,
+        },
+        *,
+    };
+
+    fn get_data() -> Vec<u64> {
+        let mut data: Vec<_> = (100..55000_u64)
+            .map(|num| num + rand::random::<u8>() as u64)
+            .collect();
+        data.push(99_000);
+        data.insert(1000, 2000);
+        data.insert(2000, 100);
+        data.insert(3000, 4100);
+        data.insert(4000, 100);
+        data.insert(5000, 800);
+        data
+    }
+
+    fn value_iter() -> impl Iterator<Item = u64> {
+        0..20_000
+    }
+    fn bench_get<S: FastFieldCodecSerializer, R: FastFieldCodecReader>(
+        b: &mut Bencher,
+        data: &[u64],
+    ) {
+        let mut bytes = vec![];
+        S::serialize(
+            &mut bytes,
+            &data,
+            stats_from_vec(data),
+            data.iter().cloned(),
+            data.iter().cloned(),
+        )
+        .unwrap();
+        let reader = R::open_from_bytes(&bytes).unwrap();
+        b.iter(|| {
+            for pos in value_iter() {
+                reader.get_u64(pos as u64, &bytes);
+            }
+        });
+    }
+    fn bench_create<S: FastFieldCodecSerializer>(b: &mut Bencher, data: &[u64]) {
+        let mut bytes = vec![];
+        b.iter(|| {
+            S::serialize(
+                &mut bytes,
+                &data,
+                stats_from_vec(data),
+                data.iter().cloned(),
+                data.iter().cloned(),
+            )
+            .unwrap();
+        });
+    }
+
+    use test::Bencher;
+    #[bench]
+    fn bench_fastfield_bitpack_create(b: &mut Bencher) {
+        let data: Vec<_> = get_data();
+        bench_create::<BitpackedFastFieldSerializer>(b, &data);
+    }
+    #[bench]
+    fn bench_fastfield_linearinterpol_create(b: &mut Bencher) {
+        let data: Vec<_> = get_data();
+        bench_create::<LinearInterpolFastFieldSerializer>(b, &data);
+    }
+    #[bench]
+    fn bench_fastfield_multilinearinterpol_create(b: &mut Bencher) {
+        let data: Vec<_> = get_data();
+        bench_create::<MultiLinearInterpolFastFieldSerializer>(b, &data);
+    }
+    #[bench]
+    fn bench_fastfield_bitpack_get(b: &mut Bencher) {
+        let data: Vec<_> = get_data();
+        bench_get::<BitpackedFastFieldSerializer, BitpackedFastFieldReader>(b, &data);
+    }
+    #[bench]
+    fn bench_fastfield_linearinterpol_get(b: &mut Bencher) {
+        let data: Vec<_> = get_data();
+        bench_get::<LinearInterpolFastFieldSerializer, LinearInterpolFastFieldReader>(b, &data);
+    }
+    #[bench]
+    fn bench_fastfield_multilinearinterpol_get(b: &mut Bencher) {
+        let data: Vec<_> = get_data();
+        bench_get::<MultiLinearInterpolFastFieldSerializer, MultiLinearInterpolFastFieldReader>(
+            b, &data,
+        );
+    }
+    pub fn stats_from_vec(data: &[u64]) -> FastFieldStats {
+        let min_value = data.iter().cloned().min().unwrap_or(0);
+        let max_value = data.iter().cloned().max().unwrap_or(0);
+        FastFieldStats {
+            min_value,
+            max_value,
+            num_vals: data.len() as u64,
+        }
+    }
+}

fastfield_codecs/src/bitpacked.rs

@@ -0,0 +1,176 @@
+use crate::FastFieldCodecReader;
+use crate::FastFieldCodecSerializer;
+use crate::FastFieldDataAccess;
+use crate::FastFieldStats;
+use common::BinarySerializable;
+use std::io::{self, Write};
+use tantivy_bitpacker::compute_num_bits;
+use tantivy_bitpacker::BitPacker;
+
+use tantivy_bitpacker::BitUnpacker;
+
+/// Depending on the field type, a different
+/// fast field is required.
+#[derive(Clone)]
+pub struct BitpackedFastFieldReader {
+    bit_unpacker: BitUnpacker,
+    pub min_value_u64: u64,
+    pub max_value_u64: u64,
+}
+
+impl<'data> FastFieldCodecReader for BitpackedFastFieldReader {
+    /// Opens a fast field given a file.
+    fn open_from_bytes(bytes: &[u8]) -> io::Result<Self> {
+        let (_data, mut footer) = bytes.split_at(bytes.len() - 16);
+        let min_value = u64::deserialize(&mut footer)?;
+        let amplitude = u64::deserialize(&mut footer)?;
+        let max_value = min_value + amplitude;
+        let num_bits = compute_num_bits(amplitude);
+        let bit_unpacker = BitUnpacker::new(num_bits);
+        Ok(BitpackedFastFieldReader {
+            min_value_u64: min_value,
+            max_value_u64: max_value,
+            bit_unpacker,
+        })
+    }
+    #[inline]
+    fn get_u64(&self, doc: u64, data: &[u8]) -> u64 {
+        self.min_value_u64 + self.bit_unpacker.get(doc, data)
+    }
+    #[inline]
+    fn min_value(&self) -> u64 {
+        self.min_value_u64
+    }
+    #[inline]
+    fn max_value(&self) -> u64 {
+        self.max_value_u64
+    }
+}
+pub struct BitpackedFastFieldSerializerLegacy<'a, W: 'a + Write> {
+    bit_packer: BitPacker,
+    write: &'a mut W,
+    min_value: u64,
+    amplitude: u64,
+    num_bits: u8,
+}
+
+impl<'a, W: Write> BitpackedFastFieldSerializerLegacy<'a, W> {
+    /// Creates a new fast field serializer.
+    ///
+    /// The serializer in fact encode the values by bitpacking
+    /// `(val - min_value)`.
+    ///
+    /// It requires a `min_value` and a `max_value` to compute
+    /// compute the minimum number of bits required to encode
+    /// values.
+    pub fn open(
+        write: &'a mut W,
+        min_value: u64,
+        max_value: u64,
+    ) -> io::Result<BitpackedFastFieldSerializerLegacy<'a, W>> {
+        assert!(min_value <= max_value);
+        let amplitude = max_value - min_value;
+        let num_bits = compute_num_bits(amplitude);
+        let bit_packer = BitPacker::new();
+        Ok(BitpackedFastFieldSerializerLegacy {
+            bit_packer,
+            write,
+            min_value,
+            amplitude,
+            num_bits,
+        })
+    }
+    /// Pushes a new value to the currently open u64 fast field.
+    #[inline]
+    pub fn add_val(&mut self, val: u64) -> io::Result<()> {
+        let val_to_write: u64 = val - self.min_value;
+        self.bit_packer
+            .write(val_to_write, self.num_bits, &mut self.write)?;
+        Ok(())
+    }
+    pub fn close_field(mut self) -> io::Result<()> {
+        self.bit_packer.close(&mut self.write)?;
+        self.min_value.serialize(&mut self.write)?;
+        self.amplitude.serialize(&mut self.write)?;
+        Ok(())
+    }
+}
+
+pub struct BitpackedFastFieldSerializer {}
+
+impl FastFieldCodecSerializer for BitpackedFastFieldSerializer {
+    const NAME: &'static str = "Bitpacked";
+    const ID: u8 = 1;
+    /// Serializes data with the BitpackedFastFieldSerializer.
+    ///
+    /// The serializer in fact encode the values by bitpacking
+    /// `(val - min_value)`.
+    ///
+    /// It requires a `min_value` and a `max_value` to compute
+    /// compute the minimum number of bits required to encode
+    /// values.
+    fn serialize(
+        write: &mut impl Write,
+        _fastfield_accessor: &impl FastFieldDataAccess,
+        stats: FastFieldStats,
+        data_iter: impl Iterator<Item = u64>,
+        _data_iter1: impl Iterator<Item = u64>,
+    ) -> io::Result<()> {
+        let mut serializer =
+            BitpackedFastFieldSerializerLegacy::open(write, stats.min_value, stats.max_value)?;
+
+        for val in data_iter {
+            serializer.add_val(val)?;
+        }
+        serializer.close_field()?;
+
+        Ok(())
+    }
+    fn is_applicable(
+        _fastfield_accessor: &impl FastFieldDataAccess,
+        _stats: FastFieldStats,
+    ) -> bool {
+        true
+    }
+    fn estimate(_fastfield_accessor: &impl FastFieldDataAccess, stats: FastFieldStats) -> f32 {
+        let amplitude = stats.max_value - stats.min_value;
+        let num_bits = compute_num_bits(amplitude);
+        let num_bits_uncompressed = 64;
+        num_bits as f32 / num_bits_uncompressed as f32
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::tests::get_codec_test_data_sets;
+
+    fn create_and_validate(data: &[u64], name: &str) {
+        crate::tests::create_and_validate::<BitpackedFastFieldSerializer, BitpackedFastFieldReader>(
+            data, name,
+        );
+    }
+
+    #[test]
+    fn test_with_codec_data_sets() {
+        let data_sets = get_codec_test_data_sets();
+        for (mut data, name) in data_sets {
+            create_and_validate(&data, name);
+            data.reverse();
+            create_and_validate(&data, name);
+        }
+    }
+
+    #[test]
+    fn bitpacked_fast_field_rand() {
+        for _ in 0..500 {
+            let mut data = (0..1 + rand::random::<u8>() as usize)
+                .map(|_| rand::random::<i64>() as u64 / 2)
+                .collect::<Vec<_>>();
+            create_and_validate(&data, "rand");
+
+            data.reverse();
+            create_and_validate(&data, "rand");
+        }
+    }
+}

fastfield_codecs/src/lib.rs

@@ -0,0 +1,225 @@
+#[cfg(test)]
+#[macro_use]
+extern crate more_asserts;
+
+use std::io;
+use std::io::Write;
+
+pub mod bitpacked;
+pub mod linearinterpol;
+pub mod multilinearinterpol;
+
+pub trait FastFieldCodecReader: Sized {
+    /// reads the metadata and returns the CodecReader
+    fn open_from_bytes(bytes: &[u8]) -> std::io::Result<Self>;
+
+    fn get_u64(&self, doc: u64, data: &[u8]) -> u64;
+
+    fn min_value(&self) -> u64;
+    fn max_value(&self) -> u64;
+}
+
+/// The FastFieldSerializerEstimate trait is required on all variants
+/// of fast field compressions, to decide which one to choose.
+pub trait FastFieldCodecSerializer {
+    /// A codex needs to provide a unique name and id, which is
+    /// used for debugging and de/serialization.
+    const NAME: &'static str;
+    const ID: u8;
+
+    /// Check if the Codec is able to compress the data
+    fn is_applicable(fastfield_accessor: &impl FastFieldDataAccess, stats: FastFieldStats) -> bool;
+
+    /// Returns an estimate of the compression ratio.
+    /// The baseline is uncompressed 64bit data.
+    ///
+    /// It could make sense to also return a value representing
+    /// computational complexity.
+    fn estimate(fastfield_accessor: &impl FastFieldDataAccess, stats: FastFieldStats) -> f32;
+
+    /// Serializes the data using the serializer into write.
+    /// There are multiple iterators, in case the codec needs to read the data multiple times.
+    /// The iterators should be preferred over using fastfield_accessor for performance reasons.
+    fn serialize(
+        write: &mut impl Write,
+        fastfield_accessor: &impl FastFieldDataAccess,
+        stats: FastFieldStats,
+        data_iter: impl Iterator<Item = u64>,
+        data_iter1: impl Iterator<Item = u64>,
+    ) -> io::Result<()>;
+}
+
+/// FastFieldDataAccess is the trait to access fast field data during serialization and estimation.
+pub trait FastFieldDataAccess {
+    /// Return the value associated to the given position.
+    ///
+    /// Whenever possible use the Iterator passed to the fastfield creation instead, for performance reasons.
+    ///
+    /// # Panics
+    ///
+    /// May panic if `position` is greater than the index.
+    fn get_val(&self, position: u64) -> u64;
+}
+
+#[derive(Debug, Clone)]
+pub struct FastFieldStats {
+    pub min_value: u64,
+    pub max_value: u64,
+    pub num_vals: u64,
+}
+
+impl<'a> FastFieldDataAccess for &'a [u64] {
+    fn get_val(&self, position: u64) -> u64 {
+        self[position as usize]
+    }
+}
+
+impl FastFieldDataAccess for Vec<u64> {
+    fn get_val(&self, position: u64) -> u64 {
+        self[position as usize]
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::{
+        bitpacked::{BitpackedFastFieldReader, BitpackedFastFieldSerializer},
+        linearinterpol::{LinearInterpolFastFieldReader, LinearInterpolFastFieldSerializer},
+        multilinearinterpol::{
+            MultiLinearInterpolFastFieldReader, MultiLinearInterpolFastFieldSerializer,
+        },
+    };
+
+    pub fn create_and_validate<S: FastFieldCodecSerializer, R: FastFieldCodecReader>(
+        data: &[u64],
+        name: &str,
+    ) -> (f32, f32) {
+        if !S::is_applicable(&data, crate::tests::stats_from_vec(data)) {
+            return (f32::MAX, 0.0);
+        }
+        let estimation = S::estimate(&data, crate::tests::stats_from_vec(data));
+        let mut out = vec![];
+        S::serialize(
+            &mut out,
+            &data,
+            crate::tests::stats_from_vec(data),
+            data.iter().cloned(),
+            data.iter().cloned(),
+        )
+        .unwrap();
+
+        let reader = R::open_from_bytes(&out).unwrap();
+        for (doc, orig_val) in data.iter().enumerate() {
+            let val = reader.get_u64(doc as u64, &out);
+            if val != *orig_val {
+                panic!(
+                    "val {:?} does not match orig_val {:?}, in data set {}, data {:?}",
+                    val, orig_val, name, data
+                );
+            }
+        }
+        let actual_compression = out.len() as f32 / (data.len() as f32 * 8.0);
+        (estimation, actual_compression)
+    }
+    pub fn get_codec_test_data_sets() -> Vec<(Vec<u64>, &'static str)> {
+        let mut data_and_names = vec![];
+
+        let data = (10..=20_u64).collect::<Vec<_>>();
+        data_and_names.push((data, "simple monotonically increasing"));
+
+        data_and_names.push((
+            vec![5, 6, 7, 8, 9, 10, 99, 100],
+            "offset in linear interpol",
+        ));
+        data_and_names.push((vec![5, 50, 3, 13, 1, 1000, 35], "rand small"));
+        data_and_names.push((vec![10], "single value"));
+
+        data_and_names
+    }
+
+    fn test_codec<S: FastFieldCodecSerializer, R: FastFieldCodecReader>() {
+        let codec_name = S::NAME;
+        for (data, data_set_name) in get_codec_test_data_sets() {
+            let (estimate, actual) =
+                crate::tests::create_and_validate::<S, R>(&data, data_set_name);
+            let result = if estimate == f32::MAX {
+                "Disabled".to_string()
+            } else {
+                format!("Estimate {:?} Actual {:?} ", estimate, actual)
+            };
+            println!(
+                "Codec {}, DataSet {}, {}",
+                codec_name, data_set_name, result
+            );
+        }
+    }
+    #[test]
+    fn test_codec_bitpacking() {
+        test_codec::<BitpackedFastFieldSerializer, BitpackedFastFieldReader>();
+    }
+    #[test]
+    fn test_codec_interpolation() {
+        test_codec::<LinearInterpolFastFieldSerializer, LinearInterpolFastFieldReader>();
+    }
+    #[test]
+    fn test_codec_multi_interpolation() {
+        test_codec::<MultiLinearInterpolFastFieldSerializer, MultiLinearInterpolFastFieldReader>();
+    }
+
+    use super::*;
+    pub fn stats_from_vec(data: &[u64]) -> FastFieldStats {
+        let min_value = data.iter().cloned().min().unwrap_or(0);
+        let max_value = data.iter().cloned().max().unwrap_or(0);
+        FastFieldStats {
+            min_value,
+            max_value,
+            num_vals: data.len() as u64,
+        }
+    }
+
+    #[test]
+    fn estimation_good_interpolation_case() {
+        let data = (10..=20000_u64).collect::<Vec<_>>();
+
+        let linear_interpol_estimation =
+            LinearInterpolFastFieldSerializer::estimate(&data, stats_from_vec(&data));
+        assert_le!(linear_interpol_estimation, 0.01);
+
+        let multi_linear_interpol_estimation =
+            MultiLinearInterpolFastFieldSerializer::estimate(&data, stats_from_vec(&data));
+        assert_le!(multi_linear_interpol_estimation, 0.2);
+        assert_le!(linear_interpol_estimation, multi_linear_interpol_estimation);
+
+        let bitpacked_estimation =
+            BitpackedFastFieldSerializer::estimate(&data, stats_from_vec(&data));
+        assert_le!(linear_interpol_estimation, bitpacked_estimation);
+    }
+    #[test]
+    fn estimation_test_bad_interpolation_case() {
+        let data = vec![200, 10, 10, 10, 10, 1000, 20];
+
+        let linear_interpol_estimation =
+            LinearInterpolFastFieldSerializer::estimate(&data, stats_from_vec(&data));
+        assert_le!(linear_interpol_estimation, 0.32);
+
+        let bitpacked_estimation =
+            BitpackedFastFieldSerializer::estimate(&data, stats_from_vec(&data));
+        assert_le!(bitpacked_estimation, linear_interpol_estimation);
+    }
+    #[test]
+    fn estimation_test_bad_interpolation_case_monotonically_increasing() {
+        let mut data = (200..=20000_u64).collect::<Vec<_>>();
+        data.push(1_000_000);
+
+        // in this case the linear interpolation can't in fact not be worse than bitpacking,
+        // but the estimator adds some threshold, which leads to estimated worse behavior
+        let linear_interpol_estimation =
+            LinearInterpolFastFieldSerializer::estimate(&data, stats_from_vec(&data));
+        assert_le!(linear_interpol_estimation, 0.35);
+
+        let bitpacked_estimation =
+            BitpackedFastFieldSerializer::estimate(&data, stats_from_vec(&data));
+        assert_le!(bitpacked_estimation, 0.32);
+        assert_le!(bitpacked_estimation, linear_interpol_estimation);
+    }
+}

fastfield_codecs/src/linearinterpol.rs

@@ -0,0 +1,305 @@
+use crate::FastFieldCodecReader;
+use crate::FastFieldCodecSerializer;
+use crate::FastFieldDataAccess;
+use crate::FastFieldStats;
+use std::io::{self, Read, Write};
+use std::ops::Sub;
+use tantivy_bitpacker::compute_num_bits;
+use tantivy_bitpacker::BitPacker;
+
+use common::BinarySerializable;
+use common::FixedSize;
+use tantivy_bitpacker::BitUnpacker;
+
+/// Depending on the field type, a different
+/// fast field is required.
+#[derive(Clone)]
+pub struct LinearInterpolFastFieldReader {
+    bit_unpacker: BitUnpacker,
+    pub footer: LinearInterpolFooter,
+    pub slope: f32,
+}
+
+#[derive(Clone, Debug)]
+pub struct LinearInterpolFooter {
+    pub relative_max_value: u64,
+    pub offset: u64,
+    pub first_val: u64,
+    pub last_val: u64,
+    pub num_vals: u64,
+    pub min_value: u64,
+    pub max_value: u64,
+}
+
+impl BinarySerializable for LinearInterpolFooter {
+    fn serialize<W: Write>(&self, write: &mut W) -> io::Result<()> {
+        self.relative_max_value.serialize(write)?;
+        self.offset.serialize(write)?;
+        self.first_val.serialize(write)?;
+        self.last_val.serialize(write)?;
+        self.num_vals.serialize(write)?;
+        self.min_value.serialize(write)?;
+        self.max_value.serialize(write)?;
+        Ok(())
+    }
+
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<LinearInterpolFooter> {
+        Ok(LinearInterpolFooter {
+            relative_max_value: u64::deserialize(reader)?,
+            offset: u64::deserialize(reader)?,
+            first_val: u64::deserialize(reader)?,
+            last_val: u64::deserialize(reader)?,
+            num_vals: u64::deserialize(reader)?,
+            min_value: u64::deserialize(reader)?,
+            max_value: u64::deserialize(reader)?,
+        })
+    }
+}
+
+impl FixedSize for LinearInterpolFooter {
+    const SIZE_IN_BYTES: usize = 56;
+}
+
+impl FastFieldCodecReader for LinearInterpolFastFieldReader {
+    /// Opens a fast field given a file.
+    fn open_from_bytes(bytes: &[u8]) -> io::Result<Self> {
+        let (_data, mut footer) = bytes.split_at(bytes.len() - LinearInterpolFooter::SIZE_IN_BYTES);
+        let footer = LinearInterpolFooter::deserialize(&mut footer)?;
+        let slope = get_slope(footer.first_val, footer.last_val, footer.num_vals);
+
+        let num_bits = compute_num_bits(footer.relative_max_value);
+        let bit_unpacker = BitUnpacker::new(num_bits);
+        Ok(LinearInterpolFastFieldReader {
+            bit_unpacker,
+            footer,
+            slope,
+        })
+    }
+    #[inline]
+    fn get_u64(&self, doc: u64, data: &[u8]) -> u64 {
+        let calculated_value = get_calculated_value(self.footer.first_val, doc, self.slope);
+        (calculated_value + self.bit_unpacker.get(doc, data)) - self.footer.offset
+    }
+
+    #[inline]
+    fn min_value(&self) -> u64 {
+        self.footer.min_value
+    }
+    #[inline]
+    fn max_value(&self) -> u64 {
+        self.footer.max_value
+    }
+}
+
+/// Fastfield serializer, which tries to guess values by linear interpolation
+/// and stores the difference bitpacked.
+pub struct LinearInterpolFastFieldSerializer {}
+
+#[inline]
+fn get_slope(first_val: u64, last_val: u64, num_vals: u64) -> f32 {
+    if num_vals <= 1 {
+        return 0.0;
+    }
+    //  We calculate the slope with f64 high precision and use the result in lower precision f32
+    //  This is done in order to handle estimations for very large values like i64::MAX
+    ((last_val as f64 - first_val as f64) / (num_vals as u64 - 1) as f64) as f32
+}
+
+#[inline]
+fn get_calculated_value(first_val: u64, pos: u64, slope: f32) -> u64 {
+    first_val + (pos as f32 * slope) as u64
+}
+
+impl FastFieldCodecSerializer for LinearInterpolFastFieldSerializer {
+    const NAME: &'static str = "LinearInterpol";
+    const ID: u8 = 2;
+    /// Creates a new fast field serializer.
+    fn serialize(
+        write: &mut impl Write,
+        fastfield_accessor: &impl FastFieldDataAccess,
+        stats: FastFieldStats,
+        data_iter: impl Iterator<Item = u64>,
+        data_iter1: impl Iterator<Item = u64>,
+    ) -> io::Result<()> {
+        assert!(stats.min_value <= stats.max_value);
+
+        let first_val = fastfield_accessor.get_val(0);
+        let last_val = fastfield_accessor.get_val(stats.num_vals as u64 - 1);
+        let slope = get_slope(first_val, last_val, stats.num_vals);
+        // calculate offset to ensure all values are positive
+        let mut offset = 0;
+        let mut rel_positive_max = 0;
+        for (pos, actual_value) in data_iter1.enumerate() {
+            let calculated_value = get_calculated_value(first_val, pos as u64, slope);
+            if calculated_value > actual_value {
+                // negative value we need to apply an offset
+                // we ignore negative values in the max value calculation, because negative values
+                // will be offset to 0
+                offset = offset.max(calculated_value - actual_value);
+            } else {
+                //positive value no offset reuqired
+                rel_positive_max = rel_positive_max.max(actual_value - calculated_value);
+            }
+        }
+
+        // rel_positive_max will be adjusted by offset
+        let relative_max_value = rel_positive_max + offset;
+
+        let num_bits = compute_num_bits(relative_max_value);
+        let mut bit_packer = BitPacker::new();
+        for (pos, val) in data_iter.enumerate() {
+            let calculated_value = get_calculated_value(first_val, pos as u64, slope);
+            let diff = (val + offset) - calculated_value;
+            bit_packer.write(diff, num_bits, write)?;
+        }
+        bit_packer.close(write)?;
+
+        let footer = LinearInterpolFooter {
+            relative_max_value,
+            offset,
+            first_val,
+            last_val,
+            num_vals: stats.num_vals,
+            min_value: stats.min_value,
+            max_value: stats.max_value,
+        };
+        footer.serialize(write)?;
+        Ok(())
+    }
+    fn is_applicable(
+        _fastfield_accessor: &impl FastFieldDataAccess,
+        stats: FastFieldStats,
+    ) -> bool {
+        if stats.num_vals < 3 {
+            return false; //disable compressor for this case
+        }
+        // On serialisation the offset is added to the actual value.
+        // We need to make sure this won't run into overflow calculation issues.
+        // For this we take the maximum theroretical offset and add this to the max value.
+        // If this doesn't overflow the algortihm should be fine
+        let theorethical_maximum_offset = stats.max_value - stats.min_value;
+        if stats
+            .max_value
+            .checked_add(theorethical_maximum_offset)
+            .is_none()
+        {
+            return false;
+        }
+        true
+    }
+    /// estimation for linear interpolation is hard because, you don't know
+    /// where the local maxima for the deviation of the calculated value are and
+    /// the offset to shift all values to >=0 is also unknown.
+    fn estimate(fastfield_accessor: &impl FastFieldDataAccess, stats: FastFieldStats) -> f32 {
+        let first_val = fastfield_accessor.get_val(0);
+        let last_val = fastfield_accessor.get_val(stats.num_vals as u64 - 1);
+        let slope = get_slope(first_val, last_val, stats.num_vals);
+
+        // let's sample at 0%, 5%, 10% .. 95%, 100%
+        let num_vals = stats.num_vals as f32 / 100.0;
+        let sample_positions = (0..20)
+            .map(|pos| (num_vals * pos as f32 * 5.0) as usize)
+            .collect::<Vec<_>>();
+
+        let max_distance = sample_positions
+            .iter()
+            .map(|pos| {
+                let calculated_value = get_calculated_value(first_val, *pos as u64, slope);
+                let actual_value = fastfield_accessor.get_val(*pos as u64);
+                distance(calculated_value, actual_value)
+            })
+            .max()
+            .unwrap_or(0);
+
+        // the theory would be that we don't have the actual max_distance, but we are close within 50%
+        // threshold.
+        // It is multiplied by 2 because in a log case scenario the line would be as much above as
+        // below. So the offset would = max_distance
+        //
+        let relative_max_value = (max_distance as f32 * 1.5) * 2.0;
+
+        let num_bits = compute_num_bits(relative_max_value as u64) as u64 * stats.num_vals as u64
+            + LinearInterpolFooter::SIZE_IN_BYTES as u64;
+        let num_bits_uncompressed = 64 * stats.num_vals;
+        num_bits as f32 / num_bits_uncompressed as f32
+    }
+}
+
+#[inline]
+fn distance<T: Sub<Output = T> + Ord>(x: T, y: T) -> T {
+    if x < y {
+        y - x
+    } else {
+        x - y
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::tests::get_codec_test_data_sets;
+
+    fn create_and_validate(data: &[u64], name: &str) -> (f32, f32) {
+        crate::tests::create_and_validate::<
+            LinearInterpolFastFieldSerializer,
+            LinearInterpolFastFieldReader,
+        >(data, name)
+    }
+
+    #[test]
+    fn test_compression() {
+        let data = (10..=6_000_u64).collect::<Vec<_>>();
+        let (estimate, actual_compression) =
+            create_and_validate(&data, "simple monotonically large");
+
+        assert!(actual_compression < 0.01);
+        assert!(estimate < 0.01);
+    }
+
+    #[test]
+    fn test_with_codec_data_sets() {
+        let data_sets = get_codec_test_data_sets();
+        for (mut data, name) in data_sets {
+            create_and_validate(&data, name);
+            data.reverse();
+            create_and_validate(&data, name);
+        }
+    }
+    #[test]
+    fn linear_interpol_fast_field_test_large_amplitude() {
+        let data = vec![
+            i64::MAX as u64 / 2,
+            i64::MAX as u64 / 3,
+            i64::MAX as u64 / 2,
+        ];
+
+        create_and_validate(&data, "large amplitude");
+    }
+    #[test]
+    fn linear_interpol_fast_concave_data() {
+        let data = vec![0, 1, 2, 5, 8, 10, 20, 50];
+        create_and_validate(&data, "concave data");
+    }
+    #[test]
+    fn linear_interpol_fast_convex_data() {
+        let data = vec![0, 40, 60, 70, 75, 77];
+        create_and_validate(&data, "convex data");
+    }
+    #[test]
+    fn linear_interpol_fast_field_test_simple() {
+        let data = (10..=20_u64).collect::<Vec<_>>();
+
+        create_and_validate(&data, "simple monotonically");
+    }
+
+    #[test]
+    fn linear_interpol_fast_field_rand() {
+        for _ in 0..5000 {
+            let mut data = (0..50).map(|_| rand::random::<u64>()).collect::<Vec<_>>();
+            create_and_validate(&data, "random");
+
+            data.reverse();
+            create_and_validate(&data, "random");
+        }
+    }
+}

fastfield_codecs/src/main.rs

@@ -0,0 +1,127 @@
+#[macro_use]
+extern crate prettytable;
+use fastfield_codecs::{
+    linearinterpol::LinearInterpolFastFieldSerializer,
+    multilinearinterpol::MultiLinearInterpolFastFieldSerializer, FastFieldCodecSerializer,
+    FastFieldStats,
+};
+use prettytable::{Cell, Row, Table};
+
+fn main() {
+    let mut table = Table::new();
+
+    // Add a row per time
+    table.add_row(row!["", "Compression Ratio", "Compression Estimation"]);
+
+    for (data, data_set_name) in get_codec_test_data_sets() {
+        let mut results = vec![];
+        let res = serialize_with_codec::<LinearInterpolFastFieldSerializer>(&data);
+        results.push(res);
+        let res = serialize_with_codec::<MultiLinearInterpolFastFieldSerializer>(&data);
+        results.push(res);
+        let res = serialize_with_codec::<fastfield_codecs::bitpacked::BitpackedFastFieldSerializer>(
+            &data,
+        );
+        results.push(res);
+
+        //let best_estimation_codec = results
+        //.iter()
+        //.min_by(|res1, res2| res1.partial_cmp(&res2).unwrap())
+        //.unwrap();
+        let best_compression_ratio_codec = results
+            .iter()
+            .min_by(|res1, res2| res1.partial_cmp(res2).unwrap())
+            .cloned()
+            .unwrap();
+
+        table.add_row(Row::new(vec![Cell::new(data_set_name).style_spec("Bbb")]));
+        for (is_applicable, est, comp, name) in results {
+            let (est_cell, ratio_cell) = if !is_applicable {
+                ("Codec Disabled".to_string(), "".to_string())
+            } else {
+                (est.to_string(), comp.to_string())
+            };
+            #[allow(clippy::all)]
+            let style = if comp == best_compression_ratio_codec.1 {
+                "Fb"
+            } else {
+                ""
+            };
+
+            table.add_row(Row::new(vec![
+                Cell::new(&name.to_string()).style_spec("bFg"),
+                Cell::new(&ratio_cell).style_spec(style),
+                Cell::new(&est_cell).style_spec(""),
+            ]));
+        }
+    }
+
+    table.printstd();
+}
+
+pub fn get_codec_test_data_sets() -> Vec<(Vec<u64>, &'static str)> {
+    let mut data_and_names = vec![];
+
+    let data = (1000..=200_000_u64).collect::<Vec<_>>();
+    data_and_names.push((data, "Autoincrement"));
+
+    let mut current_cumulative = 0;
+    let data = (1..=200_000_u64)
+        .map(|num| {
+            let num = (num as f32 + num as f32).log10() as u64;
+            current_cumulative += num;
+            current_cumulative
+        })
+        .collect::<Vec<_>>();
+    //let data = (1..=200000_u64).map(|num| num + num).collect::<Vec<_>>();
+    data_and_names.push((data, "Monotonically increasing concave"));
+
+    let mut current_cumulative = 0;
+    let data = (1..=200_000_u64)
+        .map(|num| {
+            let num = (200_000.0 - num as f32).log10() as u64;
+            current_cumulative += num;
+            current_cumulative
+        })
+        .collect::<Vec<_>>();
+    data_and_names.push((data, "Monotonically increasing convex"));
+
+    let data = (1000..=200_000_u64)
+        .map(|num| num + rand::random::<u8>() as u64)
+        .collect::<Vec<_>>();
+    data_and_names.push((data, "Almost monotonically increasing"));
+
+    data_and_names
+}
+
+pub fn serialize_with_codec<S: FastFieldCodecSerializer>(
+    data: &[u64],
+) -> (bool, f32, f32, &'static str) {
+    let is_applicable = S::is_applicable(&data, stats_from_vec(data));
+    if !is_applicable {
+        return (false, 0.0, 0.0, S::NAME);
+    }
+    let estimation = S::estimate(&data, stats_from_vec(data));
+    let mut out = vec![];
+    S::serialize(
+        &mut out,
+        &data,
+        stats_from_vec(data),
+        data.iter().cloned(),
+        data.iter().cloned(),
+    )
+    .unwrap();
+
+    let actual_compression = out.len() as f32 / (data.len() * 8) as f32;
+    (true, estimation, actual_compression, S::NAME)
+}
+
+pub fn stats_from_vec(data: &[u64]) -> FastFieldStats {
+    let min_value = data.iter().cloned().min().unwrap_or(0);
+    let max_value = data.iter().cloned().max().unwrap_or(0);
+    FastFieldStats {
+        min_value,
+        max_value,
+        num_vals: data.len() as u64,
+    }
+}

fastfield_codecs/src/multilinearinterpol.rs

@@ -0,0 +1,438 @@
+/*!
+
+MultiLinearInterpol compressor uses linear interpolation to guess a values and stores the offset, but in blocks of 512.
+
+With a CHUNK_SIZE of 512 and 29 byte metadata per block, we get a overhead for metadata of 232 / 512 = 0,45 bits per element.
+The additional space required per element in a block is the the maximum deviation of the linear interpolation estimation function.
+
+E.g. if the maximum deviation of an element is 12, all elements cost 4bits.
+
+Size per block:
+Num Elements * Maximum Deviation from Interpolation + 29 Byte Metadata
+
+*/
+
+use crate::FastFieldCodecReader;
+use crate::FastFieldCodecSerializer;
+use crate::FastFieldDataAccess;
+use crate::FastFieldStats;
+use common::CountingWriter;
+use std::io::{self, Read, Write};
+use std::ops::Sub;
+use tantivy_bitpacker::compute_num_bits;
+use tantivy_bitpacker::BitPacker;
+
+use common::BinarySerializable;
+use common::DeserializeFrom;
+use tantivy_bitpacker::BitUnpacker;
+
+const CHUNK_SIZE: u64 = 512;
+
+/// Depending on the field type, a different
+/// fast field is required.
+#[derive(Clone)]
+pub struct MultiLinearInterpolFastFieldReader {
+    pub footer: MultiLinearInterpolFooter,
+}
+
+#[derive(Clone, Debug, Default)]
+struct Function {
+    // The offset in the data is required, because we have diffrent bit_widths per block
+    data_start_offset: u64,
+    // start_pos in the block will be CHUNK_SIZE * BLOCK_NUM
+    start_pos: u64,
+    // only used during serialization, 0 after deserialization
+    end_pos: u64,
+    // only used during serialization, 0 after deserialization
+    value_start_pos: u64,
+    // only used during serialization, 0 after deserialization
+    value_end_pos: u64,
+    slope: f32,
+    // The offset so that all values are positive when writing them
+    positive_val_offset: u64,
+    num_bits: u8,
+    bit_unpacker: BitUnpacker,
+}
+
+impl Function {
+    fn calc_slope(&mut self) {
+        let num_vals = self.end_pos - self.start_pos;
+        self.slope = get_slope(self.value_start_pos, self.value_end_pos, num_vals);
+    }
+    // split the interpolation into two function, change self and return the second split
+    fn split(&mut self, split_pos: u64, split_pos_value: u64) -> Function {
+        let mut new_function = Function {
+            start_pos: split_pos,
+            end_pos: self.end_pos,
+            value_start_pos: split_pos_value,
+            value_end_pos: self.value_end_pos,
+            ..Default::default()
+        };
+        new_function.calc_slope();
+        self.end_pos = split_pos;
+        self.value_end_pos = split_pos_value;
+        self.calc_slope();
+        new_function
+    }
+}
+
+impl BinarySerializable for Function {
+    fn serialize<W: Write>(&self, write: &mut W) -> io::Result<()> {
+        self.data_start_offset.serialize(write)?;
+        self.value_start_pos.serialize(write)?;
+        self.positive_val_offset.serialize(write)?;
+        self.slope.serialize(write)?;
+        self.num_bits.serialize(write)?;
+        Ok(())
+    }
+
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Function> {
+        let data_start_offset = u64::deserialize(reader)?;
+        let value_start_pos = u64::deserialize(reader)?;
+        let offset = u64::deserialize(reader)?;
+        let slope = f32::deserialize(reader)?;
+        let num_bits = u8::deserialize(reader)?;
+        let interpolation = Function {
+            data_start_offset,
+            value_start_pos,
+            positive_val_offset: offset,
+            num_bits,
+            bit_unpacker: BitUnpacker::new(num_bits),
+            slope,
+            ..Default::default()
+        };
+
+        Ok(interpolation)
+    }
+}
+
+#[derive(Clone, Debug)]
+pub struct MultiLinearInterpolFooter {
+    pub num_vals: u64,
+    pub min_value: u64,
+    pub max_value: u64,
+    interpolations: Vec<Function>,
+}
+
+impl BinarySerializable for MultiLinearInterpolFooter {
+    fn serialize<W: Write>(&self, write: &mut W) -> io::Result<()> {
+        let mut out = vec![];
+        self.num_vals.serialize(&mut out)?;
+        self.min_value.serialize(&mut out)?;
+        self.max_value.serialize(&mut out)?;
+        self.interpolations.serialize(&mut out)?;
+        write.write_all(&out)?;
+        (out.len() as u32).serialize(write)?;
+        Ok(())
+    }
+
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<MultiLinearInterpolFooter> {
+        let mut footer = MultiLinearInterpolFooter {
+            num_vals: u64::deserialize(reader)?,
+            min_value: u64::deserialize(reader)?,
+            max_value: u64::deserialize(reader)?,
+            interpolations: Vec::<Function>::deserialize(reader)?,
+        };
+        for (num, interpol) in footer.interpolations.iter_mut().enumerate() {
+            interpol.start_pos = CHUNK_SIZE * num as u64;
+        }
+        Ok(footer)
+    }
+}
+
+#[inline]
+fn get_interpolation_position(doc: u64) -> usize {
+    let index = doc / CHUNK_SIZE;
+    index as usize
+}
+
+#[inline]
+fn get_interpolation_function(doc: u64, interpolations: &[Function]) -> &Function {
+    &interpolations[get_interpolation_position(doc)]
+}
+
+impl FastFieldCodecReader for MultiLinearInterpolFastFieldReader {
+    /// Opens a fast field given a file.
+    fn open_from_bytes(bytes: &[u8]) -> io::Result<Self> {
+        let footer_len: u32 = (&bytes[bytes.len() - 4..]).deserialize()?;
+
+        let (_data, mut footer) = bytes.split_at(bytes.len() - (4 + footer_len) as usize);
+        let footer = MultiLinearInterpolFooter::deserialize(&mut footer)?;
+
+        Ok(MultiLinearInterpolFastFieldReader { footer })
+    }
+
+    #[inline]
+    fn get_u64(&self, doc: u64, data: &[u8]) -> u64 {
+        let interpolation = get_interpolation_function(doc, &self.footer.interpolations);
+        let doc = doc - interpolation.start_pos;
+        let calculated_value =
+            get_calculated_value(interpolation.value_start_pos, doc, interpolation.slope);
+        let diff = interpolation
+            .bit_unpacker
+            .get(doc, &data[interpolation.data_start_offset as usize..]);
+        (calculated_value + diff) - interpolation.positive_val_offset
+    }
+
+    #[inline]
+    fn min_value(&self) -> u64 {
+        self.footer.min_value
+    }
+    #[inline]
+    fn max_value(&self) -> u64 {
+        self.footer.max_value
+    }
+}
+
+#[inline]
+fn get_slope(first_val: u64, last_val: u64, num_vals: u64) -> f32 {
+    ((last_val as f64 - first_val as f64) / (num_vals as u64 - 1) as f64) as f32
+}
+
+#[inline]
+fn get_calculated_value(first_val: u64, pos: u64, slope: f32) -> u64 {
+    (first_val as i64 + (pos as f32 * slope) as i64) as u64
+}
+
+/// Same as LinearInterpolFastFieldSerializer, but working on chunks of CHUNK_SIZE elements.
+pub struct MultiLinearInterpolFastFieldSerializer {}
+
+impl FastFieldCodecSerializer for MultiLinearInterpolFastFieldSerializer {
+    const NAME: &'static str = "MultiLinearInterpol";
+    const ID: u8 = 3;
+    /// Creates a new fast field serializer.
+    fn serialize(
+        write: &mut impl Write,
+        fastfield_accessor: &impl FastFieldDataAccess,
+        stats: FastFieldStats,
+        data_iter: impl Iterator<Item = u64>,
+        _data_iter1: impl Iterator<Item = u64>,
+    ) -> io::Result<()> {
+        assert!(stats.min_value <= stats.max_value);
+
+        let first_val = fastfield_accessor.get_val(0);
+        let last_val = fastfield_accessor.get_val(stats.num_vals as u64 - 1);
+
+        let mut first_function = Function {
+            end_pos: stats.num_vals,
+            value_start_pos: first_val,
+            value_end_pos: last_val,
+            ..Default::default()
+        };
+        first_function.calc_slope();
+        let mut interpolations = vec![first_function];
+
+        // Since we potentially apply multiple passes over the data, the data is cached.
+        // Multiple iteration can be expensive (merge with index sorting can add lot of overhead per
+        // iteration)
+        let data = data_iter.collect::<Vec<_>>();
+
+        //// let's split this into chunks of CHUNK_SIZE
+        for data_pos in (0..data.len() as u64).step_by(CHUNK_SIZE as usize).skip(1) {
+            let new_fun = {
+                let current_interpolation = interpolations.last_mut().unwrap();
+                current_interpolation.split(data_pos, data[data_pos as usize])
+            };
+            interpolations.push(new_fun);
+        }
+        // calculate offset and max (-> numbits) for each function
+        for interpolation in &mut interpolations {
+            let mut offset = 0;
+            let mut rel_positive_max = 0;
+            for (pos, actual_value) in data
+                [interpolation.start_pos as usize..interpolation.end_pos as usize]
+                .iter()
+                .cloned()
+                .enumerate()
+            {
+                let calculated_value = get_calculated_value(
+                    interpolation.value_start_pos,
+                    pos as u64,
+                    interpolation.slope,
+                );
+                if calculated_value > actual_value {
+                    // negative value we need to apply an offset
+                    // we ignore negative values in the max value calculation, because negative values
+                    // will be offset to 0
+                    offset = offset.max(calculated_value - actual_value);
+                } else {
+                    //positive value no offset reuqired
+                    rel_positive_max = rel_positive_max.max(actual_value - calculated_value);
+                }
+            }
+
+            interpolation.positive_val_offset = offset;
+            interpolation.num_bits = compute_num_bits(rel_positive_max + offset);
+        }
+        let mut bit_packer = BitPacker::new();
+
+        let write = &mut CountingWriter::wrap(write);
+        for interpolation in &mut interpolations {
+            interpolation.data_start_offset = write.written_bytes();
+            let num_bits = interpolation.num_bits;
+            for (pos, actual_value) in data
+                [interpolation.start_pos as usize..interpolation.end_pos as usize]
+                .iter()
+                .cloned()
+                .enumerate()
+            {
+                let calculated_value = get_calculated_value(
+                    interpolation.value_start_pos,
+                    pos as u64,
+                    interpolation.slope,
+                );
+                let diff = (actual_value + interpolation.positive_val_offset) - calculated_value;
+                bit_packer.write(diff, num_bits, write)?;
+            }
+            bit_packer.flush(write)?;
+        }
+        bit_packer.close(write)?;
+
+        let footer = MultiLinearInterpolFooter {
+            num_vals: stats.num_vals,
+            min_value: stats.min_value,
+            max_value: stats.max_value,
+            interpolations,
+        };
+        footer.serialize(write)?;
+        Ok(())
+    }
+
+    fn is_applicable(
+        _fastfield_accessor: &impl FastFieldDataAccess,
+        stats: FastFieldStats,
+    ) -> bool {
+        if stats.num_vals < 5_000 {
+            return false;
+        }
+        // On serialization the offset is added to the actual value.
+        // We need to make sure this won't run into overflow calculation issues.
+        // For this we take the maximum theroretical offset and add this to the max value.
+        // If this doesn't overflow the algortihm should be fine
+        let theorethical_maximum_offset = stats.max_value - stats.min_value;
+        if stats
+            .max_value
+            .checked_add(theorethical_maximum_offset)
+            .is_none()
+        {
+            return false;
+        }
+        true
+    }
+    /// estimation for linear interpolation is hard because, you don't know
+    /// where the local maxima are for the deviation of the calculated value and
+    /// the offset is also unknown.
+    fn estimate(fastfield_accessor: &impl FastFieldDataAccess, stats: FastFieldStats) -> f32 {
+        let first_val_in_first_block = fastfield_accessor.get_val(0);
+        let last_elem_in_first_chunk = CHUNK_SIZE.min(stats.num_vals);
+        let last_val_in_first_block =
+            fastfield_accessor.get_val(last_elem_in_first_chunk as u64 - 1);
+        let slope = get_slope(
+            first_val_in_first_block,
+            last_val_in_first_block,
+            stats.num_vals,
+        );
+
+        // let's sample at 0%, 5%, 10% .. 95%, 100%, but for the first block only
+        let sample_positions = (0..20)
+            .map(|pos| (last_elem_in_first_chunk as f32 / 100.0 * pos as f32 * 5.0) as usize)
+            .collect::<Vec<_>>();
+
+        let max_distance = sample_positions
+            .iter()
+            .map(|pos| {
+                let calculated_value =
+                    get_calculated_value(first_val_in_first_block, *pos as u64, slope);
+                let actual_value = fastfield_accessor.get_val(*pos as u64);
+                distance(calculated_value, actual_value)
+            })
+            .max()
+            .unwrap();
+
+        // Estimate one block and extrapolate the cost to all blocks.
+        // the theory would be that we don't have the actual max_distance, but we are close within 50%
+        // threshold.
+        // It is multiplied by 2 because in a log case scenario the line would be as much above as
+        // below. So the offset would = max_distance
+        //
+        let relative_max_value = (max_distance as f32 * 1.5) * 2.0;
+
+        let num_bits = compute_num_bits(relative_max_value as u64) as u64 * stats.num_vals as u64
+            // function metadata per block
+            + 29 * (stats.num_vals / CHUNK_SIZE);
+        let num_bits_uncompressed = 64 * stats.num_vals;
+        num_bits as f32 / num_bits_uncompressed as f32
+    }
+}
+
+fn distance<T: Sub<Output = T> + Ord>(x: T, y: T) -> T {
+    if x < y {
+        y - x
+    } else {
+        x - y
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::tests::get_codec_test_data_sets;
+
+    fn create_and_validate(data: &[u64], name: &str) -> (f32, f32) {
+        crate::tests::create_and_validate::<
+            MultiLinearInterpolFastFieldSerializer,
+            MultiLinearInterpolFastFieldReader,
+        >(data, name)
+    }
+
+    #[test]
+    fn test_compression() {
+        let data = (10..=6_000_u64).collect::<Vec<_>>();
+        let (estimate, actual_compression) =
+            create_and_validate(&data, "simple monotonically large");
+        assert!(actual_compression < 0.2);
+        assert!(estimate < 0.20);
+        assert!(estimate > 0.15);
+        assert!(actual_compression > 0.01);
+    }
+
+    #[test]
+    fn test_with_codec_data_sets() {
+        let data_sets = get_codec_test_data_sets();
+        for (mut data, name) in data_sets {
+            create_and_validate(&data, name);
+            data.reverse();
+            create_and_validate(&data, name);
+        }
+    }
+    #[test]
+    fn test_simple() {
+        let data = (10..=20_u64).collect::<Vec<_>>();
+        create_and_validate(&data, "simple monotonically");
+    }
+
+    #[test]
+    fn border_cases_1() {
+        let data = (0..1024).collect::<Vec<_>>();
+        create_and_validate(&data, "border case");
+    }
+    #[test]
+    fn border_case_2() {
+        let data = (0..1025).collect::<Vec<_>>();
+        create_and_validate(&data, "border case");
+    }
+    #[test]
+    fn rand() {
+        for _ in 0..10 {
+            let mut data = (5_000..20_000)
+                .map(|_| rand::random::<u32>() as u64)
+                .collect::<Vec<_>>();
+            let (estimate, actual_compression) = create_and_validate(&data, "random");
+            dbg!(estimate);
+            dbg!(actual_compression);
+
+            data.reverse();
+            create_and_validate(&data, "random");
+        }
+    }
+}

ownedbytes/Cargo.toml

@@ -0,0 +1,11 @@
+[package]
+authors = ["Paul Masurel <paul@quickwit.io>", "Pascal Seitz <pascal@quickwit.io>"]
+name = "ownedbytes"
+version = "0.1.0"
+edition = "2018"
+description = "Expose data as static slice"
+license = "MIT"
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]
+stable_deref_trait = "1.2.0"

ownedbytes/src/lib.rs

@@ -0,0 +1,290 @@
+use stable_deref_trait::StableDeref;
+use std::convert::TryInto;
+use std::mem;
+use std::ops::{Deref, Range};
+use std::sync::Arc;
+use std::{fmt, io};
+
+/// An OwnedBytes simply wraps an object that owns a slice of data and exposes
+/// this data as a static slice.
+///
+/// The backing object is required to be `StableDeref`.
+#[derive(Clone)]
+pub struct OwnedBytes {
+    data: &'static [u8],
+    box_stable_deref: Arc<dyn Deref<Target = [u8]> + Sync + Send>,
+}
+
+impl OwnedBytes {
+    /// Creates an empty `OwnedBytes`.
+    pub fn empty() -> OwnedBytes {
+        OwnedBytes::new(&[][..])
+    }
+
+    /// Creates an `OwnedBytes` intance given a `StableDeref` object.
+    pub fn new<T: StableDeref + Deref<Target = [u8]> + 'static + Send + Sync>(
+        data_holder: T,
+    ) -> OwnedBytes {
+        let box_stable_deref = Arc::new(data_holder);
+        let bytes: &[u8] = box_stable_deref.as_ref();
+        let data = unsafe { mem::transmute::<_, &'static [u8]>(bytes.deref()) };
+        OwnedBytes {
+            data,
+            box_stable_deref,
+        }
+    }
+
+    /// creates a fileslice that is just a view over a slice of the data.
+    pub fn slice(&self, range: Range<usize>) -> Self {
+        OwnedBytes {
+            data: &self.data[range],
+            box_stable_deref: self.box_stable_deref.clone(),
+        }
+    }
+
+    /// Returns the underlying slice of data.
+    /// `Deref` and `AsRef` are also available.
+    #[inline]
+    pub fn as_slice(&self) -> &[u8] {
+        self.data
+    }
+
+    /// Returns the len of the slice.
+    #[inline]
+    pub fn len(&self) -> usize {
+        self.data.len()
+    }
+
+    /// Splits the OwnedBytes into two OwnedBytes `(left, right)`.
+    ///
+    /// Left will hold `split_len` bytes.
+    ///
+    /// This operation is cheap and does not require to copy any memory.
+    /// On the other hand, both `left` and `right` retain a handle over
+    /// the entire slice of memory. In other words, the memory will only
+    /// be released when both left and right are dropped.
+    pub fn split(self, split_len: usize) -> (OwnedBytes, OwnedBytes) {
+        let right_box_stable_deref = self.box_stable_deref.clone();
+        let left = OwnedBytes {
+            data: &self.data[..split_len],
+            box_stable_deref: self.box_stable_deref,
+        };
+        let right = OwnedBytes {
+            data: &self.data[split_len..],
+            box_stable_deref: right_box_stable_deref,
+        };
+        (left, right)
+    }
+
+    /// Returns true iff this `OwnedBytes` is empty.
+    #[inline]
+    pub fn is_empty(&self) -> bool {
+        self.as_slice().is_empty()
+    }
+
+    /// Drops the left most `advance_len` bytes.
+    ///
+    /// See also [.clip(clip_len: usize))](#method.clip).
+    #[inline]
+    pub fn advance(&mut self, advance_len: usize) {
+        self.data = &self.data[advance_len..]
+    }
+
+    /// Reads an `u8` from the `OwnedBytes` and advance by one byte.
+    #[inline]
+    pub fn read_u8(&mut self) -> u8 {
+        assert!(!self.is_empty());
+
+        let byte = self.as_slice()[0];
+        self.advance(1);
+        byte
+    }
+
+    /// Reads an `u64` encoded as little-endian from the `OwnedBytes` and advance by 8 bytes.
+    #[inline]
+    pub fn read_u64(&mut self) -> u64 {
+        assert!(self.len() > 7);
+
+        let octlet: [u8; 8] = self.as_slice()[..8].try_into().unwrap();
+        self.advance(8);
+        u64::from_le_bytes(octlet)
+    }
+}
+
+impl fmt::Debug for OwnedBytes {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        // We truncate the bytes in order to make sure the debug string
+        // is not too long.
+        let bytes_truncated: &[u8] = if self.len() > 8 {
+            &self.as_slice()[..10]
+        } else {
+            self.as_slice()
+        };
+        write!(f, "OwnedBytes({:?}, len={})", bytes_truncated, self.len())
+    }
+}
+
+impl Deref for OwnedBytes {
+    type Target = [u8];
+
+    #[inline]
+    fn deref(&self) -> &Self::Target {
+        self.as_slice()
+    }
+}
+
+impl io::Read for OwnedBytes {
+    #[inline]
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        let read_len = {
+            let data = self.as_slice();
+            if data.len() >= buf.len() {
+                let buf_len = buf.len();
+                buf.copy_from_slice(&data[..buf_len]);
+                buf.len()
+            } else {
+                let data_len = data.len();
+                buf[..data_len].copy_from_slice(data);
+                data_len
+            }
+        };
+        self.advance(read_len);
+        Ok(read_len)
+    }
+    #[inline]
+    fn read_to_end(&mut self, buf: &mut Vec<u8>) -> io::Result<usize> {
+        let read_len = {
+            let data = self.as_slice();
+            buf.extend(data);
+            data.len()
+        };
+        self.advance(read_len);
+        Ok(read_len)
+    }
+    #[inline]
+    fn read_exact(&mut self, buf: &mut [u8]) -> io::Result<()> {
+        let read_len = self.read(buf)?;
+        if read_len != buf.len() {
+            return Err(io::Error::new(
+                io::ErrorKind::UnexpectedEof,
+                "failed to fill whole buffer",
+            ));
+        }
+        Ok(())
+    }
+}
+
+impl AsRef<[u8]> for OwnedBytes {
+    #[inline]
+    fn as_ref(&self) -> &[u8] {
+        self.as_slice()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::io::{self, Read};
+
+    use super::OwnedBytes;
+
+    #[test]
+    fn test_owned_bytes_debug() {
+        let short_bytes = OwnedBytes::new(b"abcd".as_ref());
+        assert_eq!(
+            format!("{:?}", short_bytes),
+            "OwnedBytes([97, 98, 99, 100], len=4)"
+        );
+        let long_bytes = OwnedBytes::new(b"abcdefghijklmnopq".as_ref());
+        assert_eq!(
+            format!("{:?}", long_bytes),
+            "OwnedBytes([97, 98, 99, 100, 101, 102, 103, 104, 105, 106], len=17)"
+        );
+    }
+
+    #[test]
+    fn test_owned_bytes_read() -> io::Result<()> {
+        let mut bytes = OwnedBytes::new(b"abcdefghiklmnopqrstuvwxyz".as_ref());
+        {
+            let mut buf = [0u8; 5];
+            bytes.read_exact(&mut buf[..]).unwrap();
+            assert_eq!(&buf, b"abcde");
+            assert_eq!(bytes.as_slice(), b"fghiklmnopqrstuvwxyz")
+        }
+        {
+            let mut buf = [0u8; 2];
+            bytes.read_exact(&mut buf[..]).unwrap();
+            assert_eq!(&buf, b"fg");
+            assert_eq!(bytes.as_slice(), b"hiklmnopqrstuvwxyz")
+        }
+        Ok(())
+    }
+
+    #[test]
+    fn test_owned_bytes_read_right_at_the_end() -> io::Result<()> {
+        let mut bytes = OwnedBytes::new(b"abcde".as_ref());
+        let mut buf = [0u8; 5];
+        assert_eq!(bytes.read(&mut buf[..]).unwrap(), 5);
+        assert_eq!(&buf, b"abcde");
+        assert_eq!(bytes.as_slice(), b"");
+        assert_eq!(bytes.read(&mut buf[..]).unwrap(), 0);
+        assert_eq!(&buf, b"abcde");
+        Ok(())
+    }
+    #[test]
+    fn test_owned_bytes_read_incomplete() -> io::Result<()> {
+        let mut bytes = OwnedBytes::new(b"abcde".as_ref());
+        let mut buf = [0u8; 7];
+        assert_eq!(bytes.read(&mut buf[..]).unwrap(), 5);
+        assert_eq!(&buf[..5], b"abcde");
+        assert_eq!(bytes.read(&mut buf[..]).unwrap(), 0);
+        Ok(())
+    }
+
+    #[test]
+    fn test_owned_bytes_read_to_end() -> io::Result<()> {
+        let mut bytes = OwnedBytes::new(b"abcde".as_ref());
+        let mut buf = Vec::new();
+        bytes.read_to_end(&mut buf)?;
+        assert_eq!(buf.as_slice(), b"abcde".as_ref());
+        Ok(())
+    }
+
+    #[test]
+    fn test_owned_bytes_read_u8() -> io::Result<()> {
+        let mut bytes = OwnedBytes::new(b"\xFF".as_ref());
+        assert_eq!(bytes.read_u8(), 255);
+        assert_eq!(bytes.len(), 0);
+        Ok(())
+    }
+
+    #[test]
+    fn test_owned_bytes_read_u64() -> io::Result<()> {
+        let mut bytes = OwnedBytes::new(b"\0\xFF\xFF\xFF\xFF\xFF\xFF\xFF".as_ref());
+        assert_eq!(bytes.read_u64(), u64::MAX - 255);
+        assert_eq!(bytes.len(), 0);
+        Ok(())
+    }
+
+    #[test]
+    fn test_owned_bytes_split() {
+        let bytes = OwnedBytes::new(b"abcdefghi".as_ref());
+        let (left, right) = bytes.split(3);
+        assert_eq!(left.as_slice(), b"abc");
+        assert_eq!(right.as_slice(), b"defghi");
+    }
+
+    #[test]
+    fn test_owned_bytes_split_boundary() {
+        let bytes = OwnedBytes::new(b"abcdefghi".as_ref());
+        {
+            let (left, right) = bytes.clone().split(0);
+            assert_eq!(left.as_slice(), b"");
+            assert_eq!(right.as_slice(), b"abcdefghi");
+        }
+        {
+            let (left, right) = bytes.split(9);
+            assert_eq!(left.as_slice(), b"abcdefghi");
+            assert_eq!(right.as_slice(), b"");
+        }
+    }
+}

query-grammar/Cargo.toml

@@ -0,0 +1,18 @@
+[package]
+name = "tantivy-query-grammar"
+version = "0.15.0"
+authors = ["Paul Masurel <paul.masurel@gmail.com>"]
+license = "MIT"
+categories = ["database-implementations", "data-structures"]
+description = """Search engine library"""
+documentation = "https://tantivy-search.github.io/tantivy/tantivy/index.html"
+homepage = "https://github.com/tantivy-search/tantivy"
+repository = "https://github.com/tantivy-search/tantivy"
+readme = "README.md"
+keywords = ["search", "information", "retrieval"]
+edition = "2018"
+
+[dependencies]
+combine = {version="4", default-features=false, features=[] }
+once_cell = "1.7.2"
+regex ={ version = "1.5.4", default-features = false, features = ["std"] }

query-grammar/README.md

@@ -0,0 +1,3 @@
+# Tantivy Query Grammar
+
+This crate is used by tantivy to parse queries.

query-grammar/src/lib.rs

@@ -0,0 +1,15 @@
+mod occur;
+mod query_grammar;
+mod user_input_ast;
+use combine::parser::Parser;
+
+pub use crate::occur::Occur;
+use crate::query_grammar::parse_to_ast;
+pub use crate::user_input_ast::{UserInputAst, UserInputBound, UserInputLeaf, UserInputLiteral};
+
+pub struct Error;
+
+pub fn parse_query(query: &str) -> Result<UserInputAst, Error> {
+    let (user_input_ast, _remaining) = parse_to_ast().parse(query).map_err(|_| Error)?;
+    Ok(user_input_ast)
+}

query-grammar/src/occur.rs

@@ -0,0 +1,72 @@
+use std::fmt;
+use std::fmt::Write;
+
+/// Defines whether a term in a query must be present,
+/// should be present or must be not present.
+#[derive(Debug, Clone, Hash, Copy, Eq, PartialEq)]
+pub enum Occur {
+    /// For a given document to be considered for scoring,
+    /// at least one of the document with the Should or the Must
+    /// Occur constraint must be within the document.
+    Should,
+    /// Document without the term are excluded from the search.
+    Must,
+    /// Document that contain the term are excluded from the
+    /// search.
+    MustNot,
+}
+
+impl Occur {
+    /// Returns the one-char prefix symbol for this `Occur`.
+    /// - `Should` => '?',
+    /// - `Must` => '+'
+    /// - `Not` => '-'
+    fn to_char(self) -> char {
+        match self {
+            Occur::Should => '?',
+            Occur::Must => '+',
+            Occur::MustNot => '-',
+        }
+    }
+
+    /// Compose two occur values.
+    pub fn compose(left: Occur, right: Occur) -> Occur {
+        match (left, right) {
+            (Occur::Should, _) => right,
+            (Occur::Must, Occur::MustNot) => Occur::MustNot,
+            (Occur::Must, _) => Occur::Must,
+            (Occur::MustNot, Occur::MustNot) => Occur::Must,
+            (Occur::MustNot, _) => Occur::MustNot,
+        }
+    }
+}
+
+impl fmt::Display for Occur {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        f.write_char(self.to_char())
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use crate::Occur;
+
+    #[test]
+    fn test_occur_compose() {
+        assert_eq!(Occur::compose(Occur::Should, Occur::Should), Occur::Should);
+        assert_eq!(Occur::compose(Occur::Should, Occur::Must), Occur::Must);
+        assert_eq!(
+            Occur::compose(Occur::Should, Occur::MustNot),
+            Occur::MustNot
+        );
+        assert_eq!(Occur::compose(Occur::Must, Occur::Should), Occur::Must);
+        assert_eq!(Occur::compose(Occur::Must, Occur::Must), Occur::Must);
+        assert_eq!(Occur::compose(Occur::Must, Occur::MustNot), Occur::MustNot);
+        assert_eq!(
+            Occur::compose(Occur::MustNot, Occur::Should),
+            Occur::MustNot
+        );
+        assert_eq!(Occur::compose(Occur::MustNot, Occur::Must), Occur::MustNot);
+        assert_eq!(Occur::compose(Occur::MustNot, Occur::MustNot), Occur::Must);
+    }
+}

query-grammar/src/query_grammar.rs

@@ -0,0 +1,693 @@
+use super::user_input_ast::{UserInputAst, UserInputBound, UserInputLeaf, UserInputLiteral};
+use crate::Occur;
+use combine::parser::char::{char, digit, space, spaces, string};
+use combine::parser::range::{take_while, take_while1};
+use combine::parser::repeat::escaped;
+use combine::parser::Parser;
+use combine::{
+    attempt, choice, eof, many, many1, one_of, optional, parser, satisfy, skip_many1, value,
+};
+use combine::{error::StringStreamError, parser::combinator::recognize};
+use once_cell::sync::Lazy;
+use regex::Regex;
+
+// Note: '-' char is only forbidden at the beginning of a field name, would be clearer to add it to special characters.
+const SPECIAL_CHARS: &[char] = &[
+    '+', '^', '`', ':', '{', '}', '"', '[', ']', '(', ')', '~', '!', '\\', '*', ' ',
+];
+const ESCAPED_SPECIAL_CHARS_PATTERN: &str = r#"\\(\+|\^|`|:|\{|\}|"|\[|\]|\(|\)|\~|!|\\|\*| )"#;
+
+/// Parses a field_name
+/// A field name must have at least one character and be followed by a colon.
+/// All characters are allowed including special characters `SPECIAL_CHARS`, but these
+/// need to be escaped with a backslack character '\'.
+fn field_name<'a>() -> impl Parser<&'a str, Output = String> {
+    static ESCAPED_SPECIAL_CHARS_RE: Lazy<Regex> =
+        Lazy::new(|| Regex::new(ESCAPED_SPECIAL_CHARS_PATTERN).unwrap());
+
+    recognize::<String, _, _>(escaped(
+        (
+            take_while1(|c| !SPECIAL_CHARS.contains(&c) && c != '-'),
+            take_while(|c| !SPECIAL_CHARS.contains(&c)),
+        ),
+        '\\',
+        satisfy(|c| SPECIAL_CHARS.contains(&c)),
+    ))
+    .skip(char(':'))
+    .map(|s| ESCAPED_SPECIAL_CHARS_RE.replace_all(&s, "$1").to_string())
+    .and_then(|s: String| match s.is_empty() {
+        true => Err(StringStreamError::UnexpectedParse),
+        _ => Ok(s),
+    })
+}
+
+fn word<'a>() -> impl Parser<&'a str, Output = String> {
+    (
+        satisfy(|c: char| {
+            !c.is_whitespace()
+                && !['-', '^', '`', ':', '{', '}', '"', '[', ']', '(', ')'].contains(&c)
+        }),
+        many(satisfy(|c: char| {
+            !c.is_whitespace() && ![':', '^', '{', '}', '"', '[', ']', '(', ')'].contains(&c)
+        })),
+    )
+        .map(|(s1, s2): (char, String)| format!("{}{}", s1, s2))
+        .and_then(|s: String| match s.as_str() {
+            "OR" | "AND " | "NOT" => Err(StringStreamError::UnexpectedParse),
+            _ => Ok(s),
+        })
+}
+
+/// Parses a date time according to rfc3339
+/// 2015-08-02T18:54:42+02
+/// 2021-04-13T19:46:26.266051969+00:00
+///
+/// NOTE: also accepts 999999-99-99T99:99:99.266051969+99:99
+/// We delegate rejecting such invalid dates to the logical AST compuation code
+/// which invokes chrono::DateTime::parse_from_rfc3339 on the value to actually parse
+/// it (instead of merely extracting the datetime value as string as done here).
+fn date_time<'a>() -> impl Parser<&'a str, Output = String> {
+    let two_digits = || recognize::<String, _, _>((digit(), digit()));
+
+    // Parses a time zone
+    // -06:30
+    // Z
+    let time_zone = {
+        let utc = recognize::<String, _, _>(char('Z'));
+        let offset = recognize((
+            choice([char('-'), char('+')]),
+            two_digits(),
+            char(':'),
+            two_digits(),
+        ));
+
+        utc.or(offset)
+    };
+
+    // Parses a date
+    // 2010-01-30
+    let date = {
+        recognize::<String, _, _>((
+            many1::<String, _, _>(digit()),
+            char('-'),
+            two_digits(),
+            char('-'),
+            two_digits(),
+        ))
+    };
+
+    // Parses a time
+    // 12:30:02
+    // 19:46:26.266051969
+    let time = {
+        recognize::<String, _, _>((
+            two_digits(),
+            char(':'),
+            two_digits(),
+            char(':'),
+            two_digits(),
+            optional((char('.'), many1::<String, _, _>(digit()))),
+            time_zone,
+        ))
+    };
+
+    recognize((date, char('T'), time))
+}
+
+fn term_val<'a>() -> impl Parser<&'a str, Output = String> {
+    let phrase = char('"').with(many1(satisfy(|c| c != '"'))).skip(char('"'));
+    phrase.or(word())
+}
+
+fn term_query<'a>() -> impl Parser<&'a str, Output = UserInputLiteral> {
+    let term_val_with_field = negative_number().or(term_val());
+    (field_name(), term_val_with_field).map(|(field_name, phrase)| UserInputLiteral {
+        field_name: Some(field_name),
+        phrase,
+    })
+}
+
+fn literal<'a>() -> impl Parser<&'a str, Output = UserInputLeaf> {
+    let term_default_field = term_val().map(|phrase| UserInputLiteral {
+        field_name: None,
+        phrase,
+    });
+    attempt(term_query())
+        .or(term_default_field)
+        .map(UserInputLeaf::from)
+}
+
+fn negative_number<'a>() -> impl Parser<&'a str, Output = String> {
+    (
+        char('-'),
+        many1(digit()),
+        optional((char('.'), many1(digit()))),
+    )
+        .map(|(s1, s2, s3): (char, String, Option<(char, String)>)| {
+            if let Some(('.', s3)) = s3 {
+                format!("{}{}.{}", s1, s2, s3)
+            } else {
+                format!("{}{}", s1, s2)
+            }
+        })
+}
+
+fn spaces1<'a>() -> impl Parser<&'a str, Output = ()> {
+    skip_many1(space())
+}
+
+/// Function that parses a range out of a Stream
+/// Supports ranges like:
+/// [5 TO 10], {5 TO 10}, [* TO 10], [10 TO *], {10 TO *], >5, <=10
+/// [a TO *], [a TO c], [abc TO bcd}
+fn range<'a>() -> impl Parser<&'a str, Output = UserInputLeaf> {
+    let range_term_val = || {
+        attempt(date_time())
+            .or(word())
+            .or(negative_number())
+            .or(char('*').with(value("*".to_string())))
+    };
+
+    // check for unbounded range in the form of <5, <=10, >5, >=5
+    let elastic_unbounded_range = (
+        choice([
+            attempt(string(">=")),
+            attempt(string("<=")),
+            attempt(string("<")),
+            attempt(string(">")),
+        ])
+        .skip(spaces()),
+        range_term_val(),
+    )
+        .map(
+            |(comparison_sign, bound): (&str, String)| match comparison_sign {
+                ">=" => (UserInputBound::Inclusive(bound), UserInputBound::Unbounded),
+                "<=" => (UserInputBound::Unbounded, UserInputBound::Inclusive(bound)),
+                "<" => (UserInputBound::Unbounded, UserInputBound::Exclusive(bound)),
+                ">" => (UserInputBound::Exclusive(bound), UserInputBound::Unbounded),
+                // default case
+                _ => (UserInputBound::Unbounded, UserInputBound::Unbounded),
+            },
+        );
+    let lower_bound = (one_of("{[".chars()), range_term_val()).map(
+        |(boundary_char, lower_bound): (char, String)| {
+            if lower_bound == "*" {
+                UserInputBound::Unbounded
+            } else if boundary_char == '{' {
+                UserInputBound::Exclusive(lower_bound)
+            } else {
+                UserInputBound::Inclusive(lower_bound)
+            }
+        },
+    );
+    let upper_bound = (range_term_val(), one_of("}]".chars())).map(
+        |(higher_bound, boundary_char): (String, char)| {
+            if higher_bound == "*" {
+                UserInputBound::Unbounded
+            } else if boundary_char == '}' {
+                UserInputBound::Exclusive(higher_bound)
+            } else {
+                UserInputBound::Inclusive(higher_bound)
+            }
+        },
+    );
+    // return only lower and upper
+    let lower_to_upper = (
+        lower_bound.skip((spaces(), string("TO"), spaces())),
+        upper_bound,
+    );
+
+    (
+        optional(field_name()).skip(spaces()),
+        // try elastic first, if it matches, the range is unbounded
+        attempt(elastic_unbounded_range).or(lower_to_upper),
+    )
+        .map(|(field, (lower, upper))|
+             // Construct the leaf from extracted field (optional)
+             // and bounds
+             UserInputLeaf::Range {
+                 field,
+                 lower,
+                 upper
+    })
+}
+
+fn negate(expr: UserInputAst) -> UserInputAst {
+    expr.unary(Occur::MustNot)
+}
+
+fn leaf<'a>() -> impl Parser<&'a str, Output = UserInputAst> {
+    parser(|input| {
+        char('(')
+            .with(ast())
+            .skip(char(')'))
+            .or(char('*').map(|_| UserInputAst::from(UserInputLeaf::All)))
+            .or(attempt(
+                string("NOT").skip(spaces1()).with(leaf()).map(negate),
+            ))
+            .or(attempt(range().map(UserInputAst::from)))
+            .or(literal().map(UserInputAst::from))
+            .parse_stream(input)
+            .into_result()
+    })
+}
+
+fn occur_symbol<'a>() -> impl Parser<&'a str, Output = Occur> {
+    char('-')
+        .map(|_| Occur::MustNot)
+        .or(char('+').map(|_| Occur::Must))
+}
+
+fn occur_leaf<'a>() -> impl Parser<&'a str, Output = (Option<Occur>, UserInputAst)> {
+    (optional(occur_symbol()), boosted_leaf())
+}
+
+fn positive_float_number<'a>() -> impl Parser<&'a str, Output = f64> {
+    (many1(digit()), optional((char('.'), many1(digit())))).map(
+        |(int_part, decimal_part_opt): (String, Option<(char, String)>)| {
+            let mut float_str = int_part;
+            if let Some((chr, decimal_str)) = decimal_part_opt {
+                float_str.push(chr);
+                float_str.push_str(&decimal_str);
+            }
+            float_str.parse::<f64>().unwrap()
+        },
+    )
+}
+
+fn boost<'a>() -> impl Parser<&'a str, Output = f64> {
+    (char('^'), positive_float_number()).map(|(_, boost)| boost)
+}
+
+fn boosted_leaf<'a>() -> impl Parser<&'a str, Output = UserInputAst> {
+    (leaf(), optional(boost())).map(|(leaf, boost_opt)| match boost_opt {
+        Some(boost) if (boost - 1.0).abs() > std::f64::EPSILON => {
+            UserInputAst::Boost(Box::new(leaf), boost)
+        }
+        _ => leaf,
+    })
+}
+
+#[derive(Clone, Copy)]
+enum BinaryOperand {
+    Or,
+    And,
+}
+
+fn binary_operand<'a>() -> impl Parser<&'a str, Output = BinaryOperand> {
+    string("AND")
+        .with(value(BinaryOperand::And))
+        .or(string("OR").with(value(BinaryOperand::Or)))
+}
+
+fn aggregate_binary_expressions(
+    left: UserInputAst,
+    others: Vec<(BinaryOperand, UserInputAst)>,
+) -> UserInputAst {
+    let mut dnf: Vec<Vec<UserInputAst>> = vec![vec![left]];
+    for (operator, operand_ast) in others {
+        match operator {
+            BinaryOperand::And => {
+                if let Some(last) = dnf.last_mut() {
+                    last.push(operand_ast);
+                }
+            }
+            BinaryOperand::Or => {
+                dnf.push(vec![operand_ast]);
+            }
+        }
+    }
+    if dnf.len() == 1 {
+        UserInputAst::and(dnf.into_iter().next().unwrap()) //< safe
+    } else {
+        let conjunctions = dnf.into_iter().map(UserInputAst::and).collect();
+        UserInputAst::or(conjunctions)
+    }
+}
+
+fn operand_leaf<'a>() -> impl Parser<&'a str, Output = (BinaryOperand, UserInputAst)> {
+    (
+        binary_operand().skip(spaces()),
+        boosted_leaf().skip(spaces()),
+    )
+}
+
+pub fn ast<'a>() -> impl Parser<&'a str, Output = UserInputAst> {
+    let boolean_expr = (boosted_leaf().skip(spaces()), many1(operand_leaf()))
+        .map(|(left, right)| aggregate_binary_expressions(left, right));
+    let whitespace_separated_leaves = many1(occur_leaf().skip(spaces().silent())).map(
+        |subqueries: Vec<(Option<Occur>, UserInputAst)>| {
+            if subqueries.len() == 1 {
+                let (occur_opt, ast) = subqueries.into_iter().next().unwrap();
+                match occur_opt.unwrap_or(Occur::Should) {
+                    Occur::Must | Occur::Should => ast,
+                    Occur::MustNot => UserInputAst::Clause(vec![(Some(Occur::MustNot), ast)]),
+                }
+            } else {
+                UserInputAst::Clause(subqueries.into_iter().collect())
+            }
+        },
+    );
+    let expr = attempt(boolean_expr).or(whitespace_separated_leaves);
+    spaces().with(expr).skip(spaces())
+}
+
+pub fn parse_to_ast<'a>() -> impl Parser<&'a str, Output = UserInputAst> {
+    spaces()
+        .with(optional(ast()).skip(eof()))
+        .map(|opt_ast| opt_ast.unwrap_or_else(UserInputAst::empty_query))
+}
+
+#[cfg(test)]
+mod test {
+
+    type TestParseResult = Result<(), StringStreamError>;
+
+    use super::*;
+    use combine::parser::Parser;
+
+    pub fn nearly_equals(a: f64, b: f64) -> bool {
+        (a - b).abs() < 0.0005 * (a + b).abs()
+    }
+
+    fn assert_nearly_equals(expected: f64, val: f64) {
+        assert!(
+            nearly_equals(val, expected),
+            "Got {}, expected {}.",
+            val,
+            expected
+        );
+    }
+
+    #[test]
+    fn test_occur_symbol() -> TestParseResult {
+        assert_eq!(super::occur_symbol().parse("-")?, (Occur::MustNot, ""));
+        assert_eq!(super::occur_symbol().parse("+")?, (Occur::Must, ""));
+        Ok(())
+    }
+
+    #[test]
+    fn test_positive_float_number() {
+        fn valid_parse(float_str: &str, expected_val: f64, expected_remaining: &str) {
+            let (val, remaining) = positive_float_number().parse(float_str).unwrap();
+            assert_eq!(remaining, expected_remaining);
+            assert_nearly_equals(val, expected_val);
+        }
+        fn error_parse(float_str: &str) {
+            assert!(positive_float_number().parse(float_str).is_err());
+        }
+        valid_parse("1.0", 1.0, "");
+        valid_parse("1", 1.0, "");
+        valid_parse("0.234234 aaa", 0.234234f64, " aaa");
+        error_parse(".3332");
+        error_parse("1.");
+        error_parse("-1.");
+    }
+
+    #[test]
+    fn test_date_time() {
+        let (val, remaining) = date_time()
+            .parse("2015-08-02T18:54:42+02:30")
+            .expect("cannot parse date");
+        assert_eq!(val, "2015-08-02T18:54:42+02:30");
+        assert_eq!(remaining, "");
+        assert!(date_time().parse("2015-08-02T18:54:42+02").is_err());
+
+        let (val, remaining) = date_time()
+            .parse("2021-04-13T19:46:26.266051969+00:00")
+            .expect("cannot parse fractional date");
+        assert_eq!(val, "2021-04-13T19:46:26.266051969+00:00");
+        assert_eq!(remaining, "");
+    }
+
+    fn test_parse_query_to_ast_helper(query: &str, expected: &str) {
+        let query = parse_to_ast().parse(query).unwrap().0;
+        let query_str = format!("{:?}", query);
+        assert_eq!(query_str, expected);
+    }
+
+    fn test_is_parse_err(query: &str) {
+        assert!(parse_to_ast().parse(query).is_err());
+    }
+
+    #[test]
+    fn test_parse_empty_to_ast() {
+        test_parse_query_to_ast_helper("", "<emptyclause>");
+    }
+
+    #[test]
+    fn test_parse_query_to_ast_hyphen() {
+        test_parse_query_to_ast_helper("\"www-form-encoded\"", "\"www-form-encoded\"");
+        test_parse_query_to_ast_helper("www-form-encoded", "\"www-form-encoded\"");
+        test_parse_query_to_ast_helper("www-form-encoded", "\"www-form-encoded\"");
+    }
+
+    #[test]
+    fn test_parse_query_to_ast_not_op() {
+        assert_eq!(
+            format!("{:?}", parse_to_ast().parse("NOT")),
+            "Err(UnexpectedParse)"
+        );
+        test_parse_query_to_ast_helper("NOTa", "\"NOTa\"");
+        test_parse_query_to_ast_helper("NOT a", "(-\"a\")");
+    }
+
+    #[test]
+    fn test_boosting() {
+        assert!(parse_to_ast().parse("a^2^3").is_err());
+        assert!(parse_to_ast().parse("a^2^").is_err());
+        test_parse_query_to_ast_helper("a^3", "(\"a\")^3");
+        test_parse_query_to_ast_helper("a^3 b^2", "(*(\"a\")^3 *(\"b\")^2)");
+        test_parse_query_to_ast_helper("a^1", "\"a\"");
+    }
+
+    #[test]
+    fn test_parse_query_to_ast_binary_op() {
+        test_parse_query_to_ast_helper("a AND b", "(+\"a\" +\"b\")");
+        test_parse_query_to_ast_helper("a OR b", "(?\"a\" ?\"b\")");
+        test_parse_query_to_ast_helper("a OR b AND c", "(?\"a\" ?(+\"b\" +\"c\"))");
+        test_parse_query_to_ast_helper("a AND b         AND c", "(+\"a\" +\"b\" +\"c\")");
+        assert_eq!(
+            format!("{:?}", parse_to_ast().parse("a OR b aaa")),
+            "Err(UnexpectedParse)"
+        );
+        assert_eq!(
+            format!("{:?}", parse_to_ast().parse("a AND b aaa")),
+            "Err(UnexpectedParse)"
+        );
+        assert_eq!(
+            format!("{:?}", parse_to_ast().parse("aaa a OR b ")),
+            "Err(UnexpectedParse)"
+        );
+        assert_eq!(
+            format!("{:?}", parse_to_ast().parse("aaa ccc a OR b ")),
+            "Err(UnexpectedParse)"
+        );
+    }
+
+    #[test]
+    fn test_parse_elastic_query_ranges() {
+        test_parse_query_to_ast_helper("title: >a", "\"title\":{\"a\" TO \"*\"}");
+        test_parse_query_to_ast_helper("title:>=a", "\"title\":[\"a\" TO \"*\"}");
+        test_parse_query_to_ast_helper("title: <a", "\"title\":{\"*\" TO \"a\"}");
+        test_parse_query_to_ast_helper("title:<=a", "\"title\":{\"*\" TO \"a\"]");
+        test_parse_query_to_ast_helper("title:<=bsd", "\"title\":{\"*\" TO \"bsd\"]");
+
+        test_parse_query_to_ast_helper("weight: >70", "\"weight\":{\"70\" TO \"*\"}");
+        test_parse_query_to_ast_helper("weight:>=70", "\"weight\":[\"70\" TO \"*\"}");
+        test_parse_query_to_ast_helper("weight: <70", "\"weight\":{\"*\" TO \"70\"}");
+        test_parse_query_to_ast_helper("weight:<=70", "\"weight\":{\"*\" TO \"70\"]");
+        test_parse_query_to_ast_helper("weight: >60.7", "\"weight\":{\"60.7\" TO \"*\"}");
+
+        test_parse_query_to_ast_helper("weight: <= 70", "\"weight\":{\"*\" TO \"70\"]");
+
+        test_parse_query_to_ast_helper("weight: <= 70.5", "\"weight\":{\"*\" TO \"70.5\"]");
+    }
+
+    #[test]
+    fn test_occur_leaf() {
+        let ((occur, ast), _) = super::occur_leaf().parse("+abc").unwrap();
+        assert_eq!(occur, Some(Occur::Must));
+        assert_eq!(format!("{:?}", ast), "\"abc\"");
+    }
+
+    #[test]
+    fn test_field_name() -> TestParseResult {
+        assert_eq!(
+            super::field_name().parse(".my.field.name:a"),
+            Ok((".my.field.name".to_string(), "a"))
+        );
+        assert_eq!(
+            super::field_name().parse("my\\ field\\ name:a"),
+            Ok(("my field name".to_string(), "a"))
+        );
+        assert!(super::field_name().parse("my field:a").is_err());
+        assert_eq!(
+            super::field_name().parse("\\(1\\+1\\):2"),
+            Ok(("(1+1)".to_string(), "2"))
+        );
+        assert_eq!(
+            super::field_name().parse("my_field_name:a"),
+            Ok(("my_field_name".to_string(), "a"))
+        );
+        assert!(super::field_name().parse("my_field_name").is_err());
+        assert!(super::field_name().parse(":a").is_err());
+        assert!(super::field_name().parse("-my_field:a").is_err());
+        assert_eq!(
+            super::field_name().parse("_my_field:a")?,
+            ("_my_field".to_string(), "a")
+        );
+        Ok(())
+    }
+
+    #[test]
+    fn test_field_name_re() {
+        let escaped_special_chars_re = Regex::new(ESCAPED_SPECIAL_CHARS_PATTERN).unwrap();
+        for special_char in SPECIAL_CHARS.iter() {
+            assert_eq!(
+                escaped_special_chars_re.replace_all(&format!("\\{}", special_char), "$1"),
+                special_char.to_string()
+            );
+        }
+    }
+
+    #[test]
+    fn test_range_parser() {
+        // testing the range() parser separately
+        let res = range()
+            .parse("title: <hello")
+            .expect("Cannot parse felxible bound word")
+            .0;
+        let expected = UserInputLeaf::Range {
+            field: Some("title".to_string()),
+            lower: UserInputBound::Unbounded,
+            upper: UserInputBound::Exclusive("hello".to_string()),
+        };
+        let res2 = range()
+            .parse("title:{* TO hello}")
+            .expect("Cannot parse ununbounded to word")
+            .0;
+        assert_eq!(res, expected);
+        assert_eq!(res2, expected);
+
+        let expected_weight = UserInputLeaf::Range {
+            field: Some("weight".to_string()),
+            lower: UserInputBound::Inclusive("71.2".to_string()),
+            upper: UserInputBound::Unbounded,
+        };
+        let res3 = range()
+            .parse("weight: >=71.2")
+            .expect("Cannot parse flexible bound float")
+            .0;
+        let res4 = range()
+            .parse("weight:[71.2 TO *}")
+            .expect("Cannot parse float to unbounded")
+            .0;
+        assert_eq!(res3, expected_weight);
+        assert_eq!(res4, expected_weight);
+
+        let expected_dates = UserInputLeaf::Range {
+            field: Some("date_field".to_string()),
+            lower: UserInputBound::Exclusive("2015-08-02T18:54:42Z".to_string()),
+            upper: UserInputBound::Inclusive("2021-08-02T18:54:42+02:30".to_string()),
+        };
+        let res5 = range()
+            .parse("date_field:{2015-08-02T18:54:42Z TO 2021-08-02T18:54:42+02:30]")
+            .expect("Cannot parse date range")
+            .0;
+        assert_eq!(res5, expected_dates);
+
+        let expected_flexible_dates = UserInputLeaf::Range {
+            field: Some("date_field".to_string()),
+            lower: UserInputBound::Unbounded,
+            upper: UserInputBound::Inclusive("2021-08-02T18:54:42.12345+02:30".to_string()),
+        };
+
+        let res6 = range()
+            .parse("date_field: <=2021-08-02T18:54:42.12345+02:30")
+            .expect("Cannot parse date range")
+            .0;
+        assert_eq!(res6, expected_flexible_dates);
+    }
+
+    #[test]
+    fn test_parse_query_to_triming_spaces() {
+        test_parse_query_to_ast_helper("   abc", "\"abc\"");
+        test_parse_query_to_ast_helper("abc ", "\"abc\"");
+        test_parse_query_to_ast_helper("(  a OR abc)", "(?\"a\" ?\"abc\")");
+        test_parse_query_to_ast_helper("(a  OR abc)", "(?\"a\" ?\"abc\")");
+        test_parse_query_to_ast_helper("(a OR  abc)", "(?\"a\" ?\"abc\")");
+        test_parse_query_to_ast_helper("a OR abc ", "(?\"a\" ?\"abc\")");
+        test_parse_query_to_ast_helper("(a OR abc )", "(?\"a\" ?\"abc\")");
+        test_parse_query_to_ast_helper("(a OR  abc) ", "(?\"a\" ?\"abc\")");
+    }
+
+    #[test]
+    fn test_parse_query_single_term() {
+        test_parse_query_to_ast_helper("abc", "\"abc\"");
+    }
+
+    #[test]
+    fn test_parse_query_default_clause() {
+        test_parse_query_to_ast_helper("a b", "(*\"a\" *\"b\")");
+    }
+
+    #[test]
+    fn test_parse_query_must_default_clause() {
+        test_parse_query_to_ast_helper("+(a b)", "(*\"a\" *\"b\")");
+    }
+
+    #[test]
+    fn test_parse_query_must_single_term() {
+        test_parse_query_to_ast_helper("+d", "\"d\"");
+    }
+
+    #[test]
+    fn test_single_term_with_field() {
+        test_parse_query_to_ast_helper("abc:toto", "\"abc\":\"toto\"");
+    }
+
+    #[test]
+    fn test_single_term_with_float() {
+        test_parse_query_to_ast_helper("abc:1.1", "\"abc\":\"1.1\"");
+        test_parse_query_to_ast_helper("a.b.c:1.1", "\"a.b.c\":\"1.1\"");
+        test_parse_query_to_ast_helper("a\\ b\\ c:1.1", "\"a b c\":\"1.1\"");
+    }
+
+    #[test]
+    fn test_must_clause() {
+        test_parse_query_to_ast_helper("(+a +b)", "(+\"a\" +\"b\")");
+    }
+
+    #[test]
+    fn test_parse_test_query_plus_a_b_plus_d() {
+        test_parse_query_to_ast_helper("+(a b) +d", "(+(*\"a\" *\"b\") +\"d\")");
+    }
+
+    #[test]
+    fn test_parse_test_query_other() {
+        test_parse_query_to_ast_helper("(+a +b) d", "(*(+\"a\" +\"b\") *\"d\")");
+        test_parse_query_to_ast_helper("+abc:toto", "\"abc\":\"toto\"");
+        test_parse_query_to_ast_helper("+a\\+b\\+c:toto", "\"a+b+c\":\"toto\"");
+        test_parse_query_to_ast_helper("(+abc:toto -titi)", "(+\"abc\":\"toto\" -\"titi\")");
+        test_parse_query_to_ast_helper("-abc:toto", "(-\"abc\":\"toto\")");
+        test_is_parse_err("--abc:toto");
+        test_parse_query_to_ast_helper("abc:a b", "(*\"abc\":\"a\" *\"b\")");
+        test_parse_query_to_ast_helper("abc:\"a b\"", "\"abc\":\"a b\"");
+        test_parse_query_to_ast_helper("foo:[1 TO 5]", "\"foo\":[\"1\" TO \"5\"]");
+    }
+
+    #[test]
+    fn test_parse_query_with_range() {
+        test_parse_query_to_ast_helper("[1 TO 5]", "[\"1\" TO \"5\"]");
+        test_parse_query_to_ast_helper("foo:{a TO z}", "\"foo\":{\"a\" TO \"z\"}");
+        test_parse_query_to_ast_helper("foo:[1 TO toto}", "\"foo\":[\"1\" TO \"toto\"}");
+        test_parse_query_to_ast_helper("foo:[* TO toto}", "\"foo\":{\"*\" TO \"toto\"}");
+        test_parse_query_to_ast_helper("foo:[1 TO *}", "\"foo\":[\"1\" TO \"*\"}");
+        test_parse_query_to_ast_helper(
+            "1.2.foo.bar:[1.1 TO *}",
+            "\"1.2.foo.bar\":[\"1.1\" TO \"*\"}",
+        );
+        test_is_parse_err("abc +    ");
+    }
+}

query-grammar/src/user_input_ast.rs

@@ -0,0 +1,171 @@
+use std::fmt;
+use std::fmt::{Debug, Formatter};
+
+use crate::Occur;
+
+#[derive(PartialEq)]
+pub enum UserInputLeaf {
+    Literal(UserInputLiteral),
+    All,
+    Range {
+        field: Option<String>,
+        lower: UserInputBound,
+        upper: UserInputBound,
+    },
+}
+
+impl Debug for UserInputLeaf {
+    fn fmt(&self, formatter: &mut Formatter<'_>) -> Result<(), fmt::Error> {
+        match self {
+            UserInputLeaf::Literal(literal) => literal.fmt(formatter),
+            UserInputLeaf::Range {
+                ref field,
+                ref lower,
+                ref upper,
+            } => {
+                if let Some(ref field) = field {
+                    write!(formatter, "\"{}\":", field)?;
+                }
+                lower.display_lower(formatter)?;
+                write!(formatter, " TO ")?;
+                upper.display_upper(formatter)?;
+                Ok(())
+            }
+            UserInputLeaf::All => write!(formatter, "*"),
+        }
+    }
+}
+
+#[derive(PartialEq)]
+pub struct UserInputLiteral {
+    pub field_name: Option<String>,
+    pub phrase: String,
+}
+
+impl fmt::Debug for UserInputLiteral {
+    fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> Result<(), fmt::Error> {
+        match self.field_name {
+            Some(ref field_name) => write!(formatter, "\"{}\":\"{}\"", field_name, self.phrase),
+            None => write!(formatter, "\"{}\"", self.phrase),
+        }
+    }
+}
+
+#[derive(PartialEq)]
+pub enum UserInputBound {
+    Inclusive(String),
+    Exclusive(String),
+    Unbounded,
+}
+
+impl UserInputBound {
+    fn display_lower(&self, formatter: &mut fmt::Formatter<'_>) -> Result<(), fmt::Error> {
+        match *self {
+            UserInputBound::Inclusive(ref word) => write!(formatter, "[\"{}\"", word),
+            UserInputBound::Exclusive(ref word) => write!(formatter, "{{\"{}\"", word),
+            UserInputBound::Unbounded => write!(formatter, "{{\"*\""),
+        }
+    }
+
+    fn display_upper(&self, formatter: &mut fmt::Formatter<'_>) -> Result<(), fmt::Error> {
+        match *self {
+            UserInputBound::Inclusive(ref word) => write!(formatter, "\"{}\"]", word),
+            UserInputBound::Exclusive(ref word) => write!(formatter, "\"{}\"}}", word),
+            UserInputBound::Unbounded => write!(formatter, "\"*\"}}"),
+        }
+    }
+
+    pub fn term_str(&self) -> &str {
+        match *self {
+            UserInputBound::Inclusive(ref contents) => contents,
+            UserInputBound::Exclusive(ref contents) => contents,
+            UserInputBound::Unbounded => "*",
+        }
+    }
+}
+
+pub enum UserInputAst {
+    Clause(Vec<(Option<Occur>, UserInputAst)>),
+    Leaf(Box<UserInputLeaf>),
+    Boost(Box<UserInputAst>, f64),
+}
+
+impl UserInputAst {
+    pub fn unary(self, occur: Occur) -> UserInputAst {
+        UserInputAst::Clause(vec![(Some(occur), self)])
+    }
+
+    fn compose(occur: Occur, asts: Vec<UserInputAst>) -> UserInputAst {
+        assert_ne!(occur, Occur::MustNot);
+        assert!(!asts.is_empty());
+        if asts.len() == 1 {
+            asts.into_iter().next().unwrap() //< safe
+        } else {
+            UserInputAst::Clause(
+                asts.into_iter()
+                    .map(|ast: UserInputAst| (Some(occur), ast))
+                    .collect::<Vec<_>>(),
+            )
+        }
+    }
+
+    pub fn empty_query() -> UserInputAst {
+        UserInputAst::Clause(Vec::default())
+    }
+
+    pub fn and(asts: Vec<UserInputAst>) -> UserInputAst {
+        UserInputAst::compose(Occur::Must, asts)
+    }
+
+    pub fn or(asts: Vec<UserInputAst>) -> UserInputAst {
+        UserInputAst::compose(Occur::Should, asts)
+    }
+}
+
+impl From<UserInputLiteral> for UserInputLeaf {
+    fn from(literal: UserInputLiteral) -> UserInputLeaf {
+        UserInputLeaf::Literal(literal)
+    }
+}
+
+impl From<UserInputLeaf> for UserInputAst {
+    fn from(leaf: UserInputLeaf) -> UserInputAst {
+        UserInputAst::Leaf(Box::new(leaf))
+    }
+}
+
+fn print_occur_ast(
+    occur_opt: Option<Occur>,
+    ast: &UserInputAst,
+    formatter: &mut fmt::Formatter,
+) -> fmt::Result {
+    if let Some(occur) = occur_opt {
+        write!(formatter, "{}{:?}", occur, ast)?;
+    } else {
+        write!(formatter, "*{:?}", ast)?;
+    }
+    Ok(())
+}
+
+impl fmt::Debug for UserInputAst {
+    fn fmt(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
+        match *self {
+            UserInputAst::Clause(ref subqueries) => {
+                if subqueries.is_empty() {
+                    write!(formatter, "<emptyclause>")?;
+                } else {
+                    write!(formatter, "(")?;
+                    print_occur_ast(subqueries[0].0, &subqueries[0].1, formatter)?;
+                    for subquery in &subqueries[1..] {
+                        write!(formatter, " ")?;
+                        print_occur_ast(subquery.0, &subquery.1, formatter)?;
+                    }
+                    write!(formatter, ")")?;
+                }
+                Ok(())
+            }
+            UserInputAst::Leaf(ref subquery) => write!(formatter, "{:?}", subquery),
+            UserInputAst::Boost(ref leaf, boost) => write!(formatter, "({:?})^{}", leaf, boost),
+        }
+    }
+}

run-tests.sh

@@ -1,2 +1,2 @@
 #!/bin/bash
-cargo test --no-default-features --features mmap -- --test-threads 1
+cargo test

src/collector/chained_collector.rs

@@ -1,142 +0,0 @@
-use collector::Collector;
-use DocId;
-use Result;
-use Score;
-use SegmentLocalId;
-use SegmentReader;
-
-/// Collector that does nothing.
-/// This is used in the chain Collector and will hopefully
-/// be optimized away by the compiler.
-pub struct DoNothingCollector;
-impl Collector for DoNothingCollector {
-    #[inline]
-    fn set_segment(&mut self, _: SegmentLocalId, _: &SegmentReader) -> Result<()> {
-        Ok(())
-    }
-    #[inline]
-    fn collect(&mut self, _doc: DocId, _score: Score) {}
-    #[inline]
-    fn requires_scoring(&self) -> bool {
-        false
-    }
-}
-
-/// Zero-cost abstraction used to collect on multiple collectors.
-/// This contraption is only usable if the type of your collectors
-/// are known at compile time.
-///
-/// ```rust
-/// #[macro_use]
-/// extern crate tantivy;
-/// use tantivy::schema::{SchemaBuilder, TEXT};
-/// use tantivy::{Index, Result};
-/// use tantivy::collector::{CountCollector, TopCollector, chain};
-/// use tantivy::query::QueryParser;
-///
-/// # fn main() { example().unwrap(); }
-/// fn example() -> Result<()> {
-///     let mut schema_builder = SchemaBuilder::new();
-///     let title = schema_builder.add_text_field("title", TEXT);
-///     let schema = schema_builder.build();
-///     let index = Index::create_in_ram(schema);
-///     {
-///         let mut index_writer = index.writer(3_000_000)?;
-///         index_writer.add_document(doc!(
-///             title => "The Name of the Wind",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of Muadib",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "A Dairy Cow",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of a Young Girl",
-///         ));
-///         index_writer.commit().unwrap();
-///     }
-///
-///     index.load_searchers()?;
-///     let searcher = index.searcher();
-///
-///     {
-///         let mut top_collector = TopCollector::with_limit(2);
-///         let mut count_collector = CountCollector::default();
-///         {
-///             let mut collectors = chain().push(&mut top_collector).push(&mut count_collector);
-///             let query_parser = QueryParser::for_index(&index, vec![title]);
-///             let query = query_parser.parse_query("diary")?;
-///             searcher.search(&*query, &mut collectors).unwrap();
-///         }
-///         assert_eq!(count_collector.count(), 2);
-///         assert!(top_collector.at_capacity());
-///     }
-///
-///     Ok(())
-/// }
-/// ```
-pub struct ChainedCollector<Left: Collector, Right: Collector> {
-    left: Left,
-    right: Right,
-}
-
-impl<Left: Collector, Right: Collector> ChainedCollector<Left, Right> {
-    /// Adds a collector
-    pub fn push<C: Collector>(self, new_collector: &mut C) -> ChainedCollector<Self, &mut C> {
-        ChainedCollector {
-            left: self,
-            right: new_collector,
-        }
-    }
-}
-
-impl<Left: Collector, Right: Collector> Collector for ChainedCollector<Left, Right> {
-    fn set_segment(
-        &mut self,
-        segment_local_id: SegmentLocalId,
-        segment: &SegmentReader,
-    ) -> Result<()> {
-        self.left.set_segment(segment_local_id, segment)?;
-        self.right.set_segment(segment_local_id, segment)?;
-        Ok(())
-    }
-
-    fn collect(&mut self, doc: DocId, score: Score) {
-        self.left.collect(doc, score);
-        self.right.collect(doc, score);
-    }
-
-    fn requires_scoring(&self) -> bool {
-        self.left.requires_scoring() || self.right.requires_scoring()
-    }
-}
-
-/// Creates a `ChainedCollector`
-pub fn chain() -> ChainedCollector<DoNothingCollector, DoNothingCollector> {
-    ChainedCollector {
-        left: DoNothingCollector,
-        right: DoNothingCollector,
-    }
-}
-
-#[cfg(test)]
-mod tests {
-
-    use super::*;
-    use collector::{Collector, CountCollector, TopCollector};
-
-    #[test]
-    fn test_chained_collector() {
-        let mut top_collector = TopCollector::with_limit(2);
-        let mut count_collector = CountCollector::default();
-        {
-            let mut collectors = chain().push(&mut top_collector).push(&mut count_collector);
-            collectors.collect(1, 0.2);
-            collectors.collect(2, 0.1);
-            collectors.collect(3, 0.5);
-        }
-        assert_eq!(count_collector.count(), 3);
-        assert!(top_collector.at_capacity());
-    }
-}

src/collector/count_collector.rs

@@ -1,101 +1,114 @@
 use super::Collector;
-use DocId;
-use Result;
-use Score;
-use SegmentLocalId;
-use SegmentReader;
+use crate::collector::SegmentCollector;
+use crate::DocId;
+use crate::Score;
+use crate::SegmentOrdinal;
+use crate::SegmentReader;
 
 /// `CountCollector` collector only counts how many
 /// documents match the query.
 ///
 /// ```rust
-/// #[macro_use]
-/// extern crate tantivy;
-/// use tantivy::schema::{SchemaBuilder, TEXT};
-/// use tantivy::{Index, Result};
-/// use tantivy::collector::CountCollector;
+/// use tantivy::collector::Count;
 /// use tantivy::query::QueryParser;
+/// use tantivy::schema::{Schema, TEXT};
+/// use tantivy::{doc, Index};
 ///
-/// # fn main() { example().unwrap(); }
-/// fn example() -> Result<()> {
-///     let mut schema_builder = SchemaBuilder::new();
-///     let title = schema_builder.add_text_field("title", TEXT);
-///     let schema = schema_builder.build();
-///     let index = Index::create_in_ram(schema);
-///     {
-///         let mut index_writer = index.writer(3_000_000)?;
-///         index_writer.add_document(doc!(
-///             title => "The Name of the Wind",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of Muadib",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "A Dairy Cow",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of a Young Girl",
-///         ));
-///         index_writer.commit().unwrap();
-///     }
+/// let mut schema_builder = Schema::builder();
+/// let title = schema_builder.add_text_field("title", TEXT);
+/// let schema = schema_builder.build();
+/// let index = Index::create_in_ram(schema);
 ///
-///     index.load_searchers()?;
-///     let searcher = index.searcher();
+/// let mut index_writer = index.writer(3_000_000).unwrap();
+/// index_writer.add_document(doc!(title => "The Name of the Wind"));
+/// index_writer.add_document(doc!(title => "The Diary of Muadib"));
+/// index_writer.add_document(doc!(title => "A Dairy Cow"));
+/// index_writer.add_document(doc!(title => "The Diary of a Young Girl"));
+/// assert!(index_writer.commit().is_ok());
 ///
-///     {
-///	        let mut count_collector = CountCollector::default();
-///         let query_parser = QueryParser::for_index(&index, vec![title]);
-///         let query = query_parser.parse_query("diary")?;
-///         searcher.search(&*query, &mut count_collector).unwrap();
+/// let reader = index.reader().unwrap();
+/// let searcher = reader.searcher();
 ///
-///         assert_eq!(count_collector.count(), 2);
-///     }
+/// // Here comes the important part
+/// let query_parser = QueryParser::for_index(&index, vec![title]);
+/// let query = query_parser.parse_query("diary").unwrap();
+/// let count = searcher.search(&query, &Count).unwrap();
 ///
-///     Ok(())
-/// }
+/// assert_eq!(count, 2);
 /// ```
-#[derive(Default)]
-pub struct CountCollector {
-    count: usize,
-}
+pub struct Count;
 
-impl CountCollector {
-    /// Returns the count of documents that were
-    /// collected.
-    pub fn count(&self) -> usize {
-        self.count
-    }
-}
+impl Collector for Count {
+    type Fruit = usize;
 
-impl Collector for CountCollector {
-    fn set_segment(&mut self, _: SegmentLocalId, _: &SegmentReader) -> Result<()> {
-        Ok(())
-    }
+    type Child = SegmentCountCollector;
 
-    fn collect(&mut self, _: DocId, _: Score) {
-        self.count += 1;
+    fn for_segment(
+        &self,
+        _: SegmentOrdinal,
+        _: &SegmentReader,
+    ) -> crate::Result<SegmentCountCollector> {
+        Ok(SegmentCountCollector::default())
     }
 
     fn requires_scoring(&self) -> bool {
         false
     }
+
+    fn merge_fruits(&self, segment_counts: Vec<usize>) -> crate::Result<usize> {
+        Ok(segment_counts.into_iter().sum())
+    }
+}
+
+#[derive(Default)]
+pub struct SegmentCountCollector {
+    count: usize,
+}
+
+impl SegmentCollector for SegmentCountCollector {
+    type Fruit = usize;
+
+    fn collect(&mut self, _: DocId, _: Score) {
+        self.count += 1;
+    }
+
+    fn harvest(self) -> usize {
+        self.count
+    }
 }
 
 #[cfg(test)]
 mod tests {
-
-    use collector::{Collector, CountCollector};
+    use super::{Count, SegmentCountCollector};
+    use crate::collector::Collector;
+    use crate::collector::SegmentCollector;
 
     #[test]
-    fn test_count_collector() {
-        let mut count_collector = CountCollector::default();
-        assert_eq!(count_collector.count(), 0);
-        count_collector.collect(0u32, 1f32);
-        assert_eq!(count_collector.count(), 1);
-        assert_eq!(count_collector.count(), 1);
-        count_collector.collect(1u32, 1f32);
-        assert_eq!(count_collector.count(), 2);
-        assert!(!count_collector.requires_scoring());
+    fn test_count_collect_does_not_requires_scoring() {
+        assert!(!Count.requires_scoring());
     }
 
+    #[test]
+    fn test_segment_count_collector() {
+        {
+            let count_collector = SegmentCountCollector::default();
+            assert_eq!(count_collector.harvest(), 0);
+        }
+        {
+            let mut count_collector = SegmentCountCollector::default();
+            count_collector.collect(0u32, 1.0);
+            assert_eq!(count_collector.harvest(), 1);
+        }
+        {
+            let mut count_collector = SegmentCountCollector::default();
+            count_collector.collect(0u32, 1.0);
+            assert_eq!(count_collector.harvest(), 1);
+        }
+        {
+            let mut count_collector = SegmentCountCollector::default();
+            count_collector.collect(0u32, 1.0);
+            count_collector.collect(1u32, 1.0);
+            assert_eq!(count_collector.harvest(), 2);
+        }
+    }
 }

src/collector/custom_score_top_collector.rs

@@ -0,0 +1,123 @@
+use crate::collector::top_collector::{TopCollector, TopSegmentCollector};
+use crate::collector::{Collector, SegmentCollector};
+use crate::{DocAddress, DocId, Score, SegmentReader};
+
+pub(crate) struct CustomScoreTopCollector<TCustomScorer, TScore = Score> {
+    custom_scorer: TCustomScorer,
+    collector: TopCollector<TScore>,
+}
+
+impl<TCustomScorer, TScore> CustomScoreTopCollector<TCustomScorer, TScore>
+where
+    TScore: Clone + PartialOrd,
+{
+    pub(crate) fn new(
+        custom_scorer: TCustomScorer,
+        collector: TopCollector<TScore>,
+    ) -> CustomScoreTopCollector<TCustomScorer, TScore> {
+        CustomScoreTopCollector {
+            custom_scorer,
+            collector,
+        }
+    }
+}
+
+/// A custom segment scorer makes it possible to define any kind of score
+/// for a given document belonging to a specific segment.
+///
+/// It is the segment local version of the [`CustomScorer`](./trait.CustomScorer.html).
+pub trait CustomSegmentScorer<TScore>: 'static {
+    /// Computes the score of a specific `doc`.
+    fn score(&mut self, doc: DocId) -> TScore;
+}
+
+/// `CustomScorer` makes it possible to define any kind of score.
+///
+/// The `CustomerScorer` itself does not make much of the computation itself.
+/// Instead, it helps constructing `Self::Child` instances that will compute
+/// the score at a segment scale.
+pub trait CustomScorer<TScore>: Sync {
+    /// Type of the associated [`CustomSegmentScorer`](./trait.CustomSegmentScorer.html).
+    type Child: CustomSegmentScorer<TScore>;
+    /// Builds a child scorer for a specific segment. The child scorer is associated to
+    /// a specific segment.
+    fn segment_scorer(&self, segment_reader: &SegmentReader) -> crate::Result<Self::Child>;
+}
+
+impl<TCustomScorer, TScore> Collector for CustomScoreTopCollector<TCustomScorer, TScore>
+where
+    TCustomScorer: CustomScorer<TScore> + Send + Sync,
+    TScore: 'static + PartialOrd + Clone + Send + Sync,
+{
+    type Fruit = Vec<(TScore, DocAddress)>;
+
+    type Child = CustomScoreTopSegmentCollector<TCustomScorer::Child, TScore>;
+
+    fn for_segment(
+        &self,
+        segment_local_id: u32,
+        segment_reader: &SegmentReader,
+    ) -> crate::Result<Self::Child> {
+        let segment_collector = self.collector.for_segment(segment_local_id, segment_reader);
+        let segment_scorer = self.custom_scorer.segment_scorer(segment_reader)?;
+        Ok(CustomScoreTopSegmentCollector {
+            segment_collector,
+            segment_scorer,
+        })
+    }
+
+    fn requires_scoring(&self) -> bool {
+        false
+    }
+
+    fn merge_fruits(&self, segment_fruits: Vec<Self::Fruit>) -> crate::Result<Self::Fruit> {
+        self.collector.merge_fruits(segment_fruits)
+    }
+}
+
+pub struct CustomScoreTopSegmentCollector<T, TScore>
+where
+    TScore: 'static + PartialOrd + Clone + Send + Sync + Sized,
+    T: CustomSegmentScorer<TScore>,
+{
+    segment_collector: TopSegmentCollector<TScore>,
+    segment_scorer: T,
+}
+
+impl<T, TScore> SegmentCollector for CustomScoreTopSegmentCollector<T, TScore>
+where
+    TScore: 'static + PartialOrd + Clone + Send + Sync,
+    T: 'static + CustomSegmentScorer<TScore>,
+{
+    type Fruit = Vec<(TScore, DocAddress)>;
+
+    fn collect(&mut self, doc: DocId, _score: Score) {
+        let score = self.segment_scorer.score(doc);
+        self.segment_collector.collect(doc, score);
+    }
+
+    fn harvest(self) -> Vec<(TScore, DocAddress)> {
+        self.segment_collector.harvest()
+    }
+}
+
+impl<F, TScore, T> CustomScorer<TScore> for F
+where
+    F: 'static + Send + Sync + Fn(&SegmentReader) -> T,
+    T: CustomSegmentScorer<TScore>,
+{
+    type Child = T;
+
+    fn segment_scorer(&self, segment_reader: &SegmentReader) -> crate::Result<Self::Child> {
+        Ok((self)(segment_reader))
+    }
+}
+
+impl<F, TScore> CustomSegmentScorer<TScore> for F
+where
+    F: 'static + FnMut(DocId) -> TScore,
+{
+    fn score(&mut self, doc: DocId) -> TScore {
+        (self)(doc)
+    }
+}

src/collector/docset_collector.rs

@@ -0,0 +1,61 @@
+use std::collections::HashSet;
+
+use crate::{DocAddress, DocId, Score};
+
+use super::{Collector, SegmentCollector};
+
+/// Collectors that returns the set of DocAddress that matches the query.
+///
+/// This collector is mostly useful for tests.
+pub struct DocSetCollector;
+
+impl Collector for DocSetCollector {
+    type Fruit = HashSet<DocAddress>;
+    type Child = DocSetChildCollector;
+
+    fn for_segment(
+        &self,
+        segment_local_id: crate::SegmentOrdinal,
+        _segment: &crate::SegmentReader,
+    ) -> crate::Result<Self::Child> {
+        Ok(DocSetChildCollector {
+            segment_local_id,
+            docs: HashSet::new(),
+        })
+    }
+
+    fn requires_scoring(&self) -> bool {
+        false
+    }
+
+    fn merge_fruits(
+        &self,
+        segment_fruits: Vec<(u32, HashSet<DocId>)>,
+    ) -> crate::Result<Self::Fruit> {
+        let len: usize = segment_fruits.iter().map(|(_, docset)| docset.len()).sum();
+        let mut result = HashSet::with_capacity(len);
+        for (segment_local_id, docs) in segment_fruits {
+            for doc in docs {
+                result.insert(DocAddress::new(segment_local_id, doc));
+            }
+        }
+        Ok(result)
+    }
+}
+
+pub struct DocSetChildCollector {
+    segment_local_id: u32,
+    docs: HashSet<DocId>,
+}
+
+impl SegmentCollector for DocSetChildCollector {
+    type Fruit = (u32, HashSet<DocId>);
+
+    fn collect(&mut self, doc: crate::DocId, _score: Score) {
+        self.docs.insert(doc);
+    }
+
+    fn harvest(self) -> (u32, HashSet<DocId>) {
+        (self.segment_local_id, self.docs)
+    }
+}

src/collector/facet_collector.rs

@@ -1,25 +1,20 @@
-use collector::Collector;
-use docset::SkipResult;
-use fastfield::FacetReader;
-use schema::Facet;
-use schema::Field;
-use std::cell::UnsafeCell;
+use crate::collector::Collector;
+use crate::collector::SegmentCollector;
+use crate::fastfield::FacetReader;
+use crate::schema::Facet;
+use crate::schema::Field;
+use crate::DocId;
+use crate::Score;
+use crate::SegmentOrdinal;
+use crate::SegmentReader;
+use std::cmp::Ordering;
 use std::collections::btree_map;
 use std::collections::BTreeMap;
 use std::collections::BTreeSet;
 use std::collections::BinaryHeap;
-use std::collections::Bound;
 use std::iter::Peekable;
-use std::mem;
+use std::ops::Bound;
 use std::{u64, usize};
-use termdict::TermMerger;
-
-use std::cmp::Ordering;
-use DocId;
-use Result;
-use Score;
-use SegmentLocalId;
-use SegmentReader;
 
 struct Hit<'a> {
     count: u64,
@@ -29,29 +24,26 @@ struct Hit<'a> {
 impl<'a> Eq for Hit<'a> {}
 
 impl<'a> PartialEq<Hit<'a>> for Hit<'a> {
-    fn eq(&self, other: &Hit) -> bool {
+    fn eq(&self, other: &Hit<'_>) -> bool {
         self.count == other.count
     }
 }
 
 impl<'a> PartialOrd<Hit<'a>> for Hit<'a> {
-    fn partial_cmp(&self, other: &Hit) -> Option<Ordering> {
+    fn partial_cmp(&self, other: &Hit<'_>) -> Option<Ordering> {
         Some(self.cmp(other))
     }
 }
 
 impl<'a> Ord for Hit<'a> {
     fn cmp(&self, other: &Self) -> Ordering {
-        other.count.cmp(&self.count)
+        other
+            .count
+            .cmp(&self.count)
+            .then(self.facet.cmp(other.facet))
     }
 }
 
-struct SegmentFacetCounter {
-    pub facet_reader: FacetReader,
-    pub facet_ords: Vec<u64>,
-    pub facet_counts: Vec<u64>,
-}
-
 fn facet_depth(facet_bytes: &[u8]) -> usize {
     if facet_bytes.is_empty() {
         0
@@ -89,21 +81,18 @@ fn facet_depth(facet_bytes: &[u8]) -> usize {
 ///
 ///
 /// ```rust
-/// #[macro_use]
-/// extern crate tantivy;
-/// use tantivy::schema::{Facet, SchemaBuilder, TEXT};
-/// use tantivy::{Index, Result};
 /// use tantivy::collector::FacetCollector;
 /// use tantivy::query::AllQuery;
+/// use tantivy::schema::{Facet, Schema, INDEXED, TEXT};
+/// use tantivy::{doc, Index};
 ///
-/// # fn main() { example().unwrap(); }
-/// fn example() -> Result<()> {
-///     let mut schema_builder = SchemaBuilder::new();
+/// fn example() -> tantivy::Result<()> {
+///     let mut schema_builder = Schema::builder();
 ///
 ///     // Facet have their own specific type.
 ///     // It is not a bad practise to put all of your
 ///     // facet information in the same field.
-///     let facet = schema_builder.add_facet_field("facet");
+///     let facet = schema_builder.add_facet_field("facet", INDEXED);
 ///     let title = schema_builder.add_text_field("title", TEXT);
 ///     let schema = schema_builder.build();
 ///     let index = Index::create_in_ram(schema);
@@ -131,23 +120,19 @@ fn facet_depth(facet_bytes: &[u8]) -> usize {
 ///             facet => Facet::from("/lang/en"),
 ///             facet => Facet::from("/category/biography")
 ///         ));
-///         index_writer.commit().unwrap();
+///         index_writer.commit()?;
 ///     }
-///
-///     index.load_searchers()?;
-///     let searcher = index.searcher();
+///     let reader = index.reader()?;
+///     let searcher = reader.searcher();
 ///
 ///     {
-///			let mut facet_collector = FacetCollector::for_field(facet);
+///         let mut facet_collector = FacetCollector::for_field(facet);
 ///         facet_collector.add_facet("/lang");
 ///         facet_collector.add_facet("/category");
-///         searcher.search(&AllQuery, &mut facet_collector).unwrap();
-///
-///         // this object contains count aggregate for all of the facets.
-///         let counts = facet_collector.harvest();
+///         let facet_counts = searcher.search(&AllQuery, &facet_collector)?;
 ///
 ///         // This lists all of the facet counts
-///         let facets: Vec<(&Facet, u64)> = counts
+///         let facets: Vec<(&Facet, u64)> = facet_counts
 ///             .get("/category")
 ///             .collect();
 ///         assert_eq!(facets, vec![
@@ -157,15 +142,12 @@ fn facet_depth(facet_bytes: &[u8]) -> usize {
 ///     }
 ///
 ///     {
-///			let mut facet_collector = FacetCollector::for_field(facet);
+///         let mut facet_collector = FacetCollector::for_field(facet);
 ///         facet_collector.add_facet("/category/fiction");
-///         searcher.search(&AllQuery, &mut facet_collector).unwrap();
-///
-///         // this object contains count aggregate for all of the facets.
-///         let counts = facet_collector.harvest();
+///         let facet_counts = searcher.search(&AllQuery, &facet_collector)?;
 ///
 ///         // This lists all of the facet counts
-///         let facets: Vec<(&Facet, u64)> = counts
+///         let facets: Vec<(&Facet, u64)> = facet_counts
 ///             .get("/category/fiction")
 ///             .collect();
 ///         assert_eq!(facets, vec![
@@ -175,16 +157,13 @@ fn facet_depth(facet_bytes: &[u8]) -> usize {
 ///         ]);
 ///     }
 ///
-///    {
-///			let mut facet_collector = FacetCollector::for_field(facet);
+///     {
+///         let mut facet_collector = FacetCollector::for_field(facet);
 ///         facet_collector.add_facet("/category/fiction");
-///         searcher.search(&AllQuery, &mut facet_collector).unwrap();
-///
-///         // this object contains count aggregate for all of the facets.
-///         let counts = facet_collector.harvest();
+///         let facet_counts = searcher.search(&AllQuery, &facet_collector)?;
 ///
 ///         // This lists all of the facet counts
-///         let facets: Vec<(&Facet, u64)> = counts.top_k("/category/fiction", 1);
+///         let facets: Vec<(&Facet, u64)> = facet_counts.top_k("/category/fiction", 1);
 ///         assert_eq!(facets, vec![
 ///             (&Facet::from("/category/fiction/fantasy"), 2)
 ///         ]);
@@ -192,40 +171,46 @@ fn facet_depth(facet_bytes: &[u8]) -> usize {
 ///
 ///     Ok(())
 /// }
+/// # assert!(example().is_ok());
 /// ```
 pub struct FacetCollector {
-    facet_ords: Vec<u64>,
     field: Field,
-    ff_reader: Option<UnsafeCell<FacetReader>>,
-    segment_counters: Vec<SegmentFacetCounter>,
-
-    // facet_ord -> collapse facet_id
-    current_segment_collapse_mapping: Vec<usize>,
-    // collapse facet_id -> count
-    current_segment_counts: Vec<u64>,
-    // collapse facet_id -> facet_ord
-    current_collapse_facet_ords: Vec<u64>,
-
     facets: BTreeSet<Facet>,
 }
 
+pub struct FacetSegmentCollector {
+    reader: FacetReader,
+    facet_ords_buf: Vec<u64>,
+    // facet_ord -> collapse facet_id
+    collapse_mapping: Vec<usize>,
+    // collapse facet_id -> count
+    counts: Vec<u64>,
+    // collapse facet_id -> facet_ord
+    collapse_facet_ords: Vec<u64>,
+}
+
+enum SkipResult {
+    Found,
+    NotFound,
+}
+
 fn skip<'a, I: Iterator<Item = &'a Facet>>(
     target: &[u8],
     collapse_it: &mut Peekable<I>,
 ) -> SkipResult {
     loop {
         match collapse_it.peek() {
-            Some(facet_bytes) => match facet_bytes.encoded_bytes().cmp(target) {
+            Some(facet_bytes) => match facet_bytes.encoded_str().as_bytes().cmp(target) {
                 Ordering::Less => {}
                 Ordering::Greater => {
-                    return SkipResult::OverStep;
+                    return SkipResult::NotFound;
                 }
                 Ordering::Equal => {
-                    return SkipResult::Reached;
+                    return SkipResult::Found;
                 }
             },
             None => {
-                return SkipResult::End;
+                return SkipResult::NotFound;
             }
         }
         collapse_it.next();
@@ -240,15 +225,8 @@ impl FacetCollector {
     /// is of the proper type.
     pub fn for_field(field: Field) -> FacetCollector {
         FacetCollector {
-            facet_ords: Vec::with_capacity(255),
-            segment_counters: Vec::new(),
             field,
-            ff_reader: None,
-            facets: BTreeSet::new(),
-
-            current_segment_collapse_mapping: Vec::new(),
-            current_collapse_facet_ords: Vec::new(),
-            current_segment_counts: Vec::new(),
+            facets: BTreeSet::default(),
         }
     }
 
@@ -278,143 +256,98 @@ impl FacetCollector {
         }
         self.facets.insert(facet);
     }
-
-    fn set_collapse_mapping(&mut self, facet_reader: &FacetReader) {
-        self.current_segment_collapse_mapping.clear();
-        self.current_collapse_facet_ords.clear();
-        self.current_segment_counts.clear();
-        let mut collapse_facet_it = self.facets.iter().peekable();
-        self.current_collapse_facet_ords.push(0);
-        let mut facet_streamer = facet_reader.facet_dict().range().into_stream();
-        if !facet_streamer.advance() {
-            return;
-        }
-        'outer: loop {
-            // at the begining of this loop, facet_streamer
-            // is positionned on a term that has not been processed yet.
-            let skip_result = skip(facet_streamer.key(), &mut collapse_facet_it);
-            match skip_result {
-                SkipResult::Reached => {
-                    // we reach a facet we decided to collapse.
-                    let collapse_depth = facet_depth(facet_streamer.key());
-                    let mut collapsed_id = 0;
-                    self.current_segment_collapse_mapping.push(0);
-                    while facet_streamer.advance() {
-                        let depth = facet_depth(facet_streamer.key());
-                        if depth <= collapse_depth {
-                            continue 'outer;
-                        }
-                        if depth == collapse_depth + 1 {
-                            collapsed_id = self.current_collapse_facet_ords.len();
-                            self.current_collapse_facet_ords
-                                .push(facet_streamer.term_ord());
-                            self.current_segment_collapse_mapping.push(collapsed_id);
-                        } else {
-                            self.current_segment_collapse_mapping.push(collapsed_id);
-                        }
-                    }
-                    break;
-                }
-                SkipResult::End | SkipResult::OverStep => {
-                    self.current_segment_collapse_mapping.push(0);
-                    if !facet_streamer.advance() {
-                        break;
-                    }
-                }
-            }
-        }
-    }
-
-    fn finalize_segment(&mut self) {
-        if self.ff_reader.is_some() {
-            self.segment_counters.push(SegmentFacetCounter {
-                facet_reader: self.ff_reader.take().unwrap().into_inner(),
-                facet_ords: mem::replace(&mut self.current_collapse_facet_ords, Vec::new()),
-                facet_counts: mem::replace(&mut self.current_segment_counts, Vec::new()),
-            });
-        }
-    }
-
-    /// Returns the results of the collection.
-    ///
-    /// This method does not just return the counters,
-    /// it also translates the facet ordinals of the last segment.
-    pub fn harvest(mut self) -> FacetCounts {
-        self.finalize_segment();
-
-        let collapsed_facet_ords: Vec<&[u64]> = self
-            .segment_counters
-            .iter()
-            .map(|segment_counter| &segment_counter.facet_ords[..])
-            .collect();
-        let collapsed_facet_counts: Vec<&[u64]> = self
-            .segment_counters
-            .iter()
-            .map(|segment_counter| &segment_counter.facet_counts[..])
-            .collect();
-
-        let facet_streams = self
-            .segment_counters
-            .iter()
-            .map(|seg_counts| seg_counts.facet_reader.facet_dict().range().into_stream())
-            .collect::<Vec<_>>();
-
-        let mut facet_merger = TermMerger::new(facet_streams);
-        let mut facet_counts = BTreeMap::new();
-
-        while facet_merger.advance() {
-            let count = facet_merger
-                .current_kvs()
-                .iter()
-                .map(|it| {
-                    let seg_ord = it.segment_ord;
-                    let term_ord = it.streamer.term_ord();
-                    collapsed_facet_ords[seg_ord]
-                        .binary_search(&term_ord)
-                        .map(|collapsed_term_id| {
-                            if collapsed_term_id == 0 {
-                                0
-                            } else {
-                                collapsed_facet_counts[seg_ord][collapsed_term_id]
-                            }
-                        }).unwrap_or(0)
-                }).sum();
-            if count > 0u64 {
-                let bytes: Vec<u8> = facet_merger.key().to_owned();
-                // may create an corrupted facet if the term dicitonary is corrupted
-                let facet = unsafe { Facet::from_encoded(bytes) };
-                facet_counts.insert(facet, count);
-            }
-        }
-        FacetCounts { facet_counts }
-    }
 }
 
 impl Collector for FacetCollector {
-    fn set_segment(&mut self, _: SegmentLocalId, reader: &SegmentReader) -> Result<()> {
-        self.finalize_segment();
+    type Fruit = FacetCounts;
+
+    type Child = FacetSegmentCollector;
+
+    fn for_segment(
+        &self,
+        _: SegmentOrdinal,
+        reader: &SegmentReader,
+    ) -> crate::Result<FacetSegmentCollector> {
         let facet_reader = reader.facet_reader(self.field)?;
-        self.set_collapse_mapping(&facet_reader);
-        self.current_segment_counts
-            .resize(self.current_collapse_facet_ords.len(), 0);
-        self.ff_reader = Some(UnsafeCell::new(facet_reader));
-        Ok(())
+
+        let mut collapse_mapping = Vec::new();
+        let mut counts = Vec::new();
+        let mut collapse_facet_ords = Vec::new();
+
+        let mut collapse_facet_it = self.facets.iter().peekable();
+        collapse_facet_ords.push(0);
+        {
+            let mut facet_streamer = facet_reader.facet_dict().range().into_stream()?;
+            if facet_streamer.advance() {
+                'outer: loop {
+                    // at the begining of this loop, facet_streamer
+                    // is positionned on a term that has not been processed yet.
+                    let skip_result = skip(facet_streamer.key(), &mut collapse_facet_it);
+                    match skip_result {
+                        SkipResult::Found => {
+                            // we reach a facet we decided to collapse.
+                            let collapse_depth = facet_depth(facet_streamer.key());
+                            let mut collapsed_id = 0;
+                            collapse_mapping.push(0);
+                            while facet_streamer.advance() {
+                                let depth = facet_depth(facet_streamer.key());
+                                if depth <= collapse_depth {
+                                    continue 'outer;
+                                }
+                                if depth == collapse_depth + 1 {
+                                    collapsed_id = collapse_facet_ords.len();
+                                    collapse_facet_ords.push(facet_streamer.term_ord());
+                                }
+                                collapse_mapping.push(collapsed_id);
+                            }
+                            break;
+                        }
+                        SkipResult::NotFound => {
+                            collapse_mapping.push(0);
+                            if !facet_streamer.advance() {
+                                break;
+                            }
+                        }
+                    }
+                }
+            }
+        }
+
+        counts.resize(collapse_facet_ords.len(), 0);
+
+        Ok(FacetSegmentCollector {
+            reader: facet_reader,
+            facet_ords_buf: Vec::with_capacity(255),
+            collapse_mapping,
+            counts,
+            collapse_facet_ords,
+        })
     }
 
+    fn requires_scoring(&self) -> bool {
+        false
+    }
+
+    fn merge_fruits(&self, segments_facet_counts: Vec<FacetCounts>) -> crate::Result<FacetCounts> {
+        let mut facet_counts: BTreeMap<Facet, u64> = BTreeMap::new();
+        for segment_facet_counts in segments_facet_counts {
+            for (facet, count) in segment_facet_counts.facet_counts {
+                *(facet_counts.entry(facet).or_insert(0)) += count;
+            }
+        }
+        Ok(FacetCounts { facet_counts })
+    }
+}
+
+impl SegmentCollector for FacetSegmentCollector {
+    type Fruit = FacetCounts;
+
     fn collect(&mut self, doc: DocId, _: Score) {
-        let facet_reader: &mut FacetReader = unsafe {
-            &mut *self
-                .ff_reader
-                .as_ref()
-                .expect("collect() was called before set_segment. This should never happen.")
-                .get()
-        };
-        facet_reader.facet_ords(doc, &mut self.facet_ords);
+        self.reader.facet_ords(doc, &mut self.facet_ords_buf);
         let mut previous_collapsed_ord: usize = usize::MAX;
-        for &facet_ord in &self.facet_ords {
-            let collapsed_ord = self.current_segment_collapse_mapping[facet_ord as usize];
-            self.current_segment_counts[collapsed_ord] += if collapsed_ord == previous_collapsed_ord
-            {
+        for &facet_ord in &self.facet_ords_buf {
+            let collapsed_ord = self.collapse_mapping[facet_ord as usize];
+            self.counts[collapsed_ord] += if collapsed_ord == previous_collapsed_ord {
                 0
             } else {
                 1
@@ -423,8 +356,27 @@ impl Collector for FacetCollector {
         }
     }
 
-    fn requires_scoring(&self) -> bool {
-        false
+    /// Returns the results of the collection.
+    ///
+    /// This method does not just return the counters,
+    /// it also translates the facet ordinals of the last segment.
+    fn harvest(self) -> FacetCounts {
+        let mut facet_counts = BTreeMap::new();
+        let facet_dict = self.reader.facet_dict();
+        for (collapsed_facet_ord, count) in self.counts.iter().cloned().enumerate() {
+            if count == 0 {
+                continue;
+            }
+            let mut facet = vec![];
+            let facet_ord = self.collapse_facet_ords[collapsed_facet_ord];
+            // TODO handle errors.
+            if facet_dict.ord_to_term(facet_ord as u64, &mut facet).is_ok() {
+                if let Ok(facet) = Facet::from_encoded(facet) {
+                    facet_counts.insert(facet, count);
+                }
+            }
+        }
+        FacetCounts { facet_counts }
     }
 }
 
@@ -447,7 +399,9 @@ impl<'a> Iterator for FacetChildIterator<'a> {
 }
 
 impl FacetCounts {
-    pub fn get<T>(&self, facet_from: T) -> FacetChildIterator
+    /// Returns an iterator over all of the facet count pairs inside this result.
+    /// See the documentation for `FacetCollector` for a usage example.
+    pub fn get<T>(&self, facet_from: T) -> FacetChildIterator<'_>
     where
         Facet: From<T>,
     {
@@ -456,15 +410,18 @@ impl FacetCounts {
         let right_bound = if facet.is_root() {
             Bound::Unbounded
         } else {
-            let mut facet_after_bytes: Vec<u8> = facet.encoded_bytes().to_owned();
-            facet_after_bytes.push(1u8);
-            let facet_after = unsafe { Facet::from_encoded(facet_after_bytes) }; // ok logic
+            let mut facet_after_bytes: String = facet.encoded_str().to_owned();
+            facet_after_bytes.push('\u{1}');
+            let facet_after = Facet::from_encoded_string(facet_after_bytes);
             Bound::Excluded(facet_after)
         };
-        let underlying: btree_map::Range<_, _> = self.facet_counts.range((left_bound, right_bound));
+        let underlying: btree_map::Range<'_, _, _> =
+            self.facet_counts.range((left_bound, right_bound));
         FacetChildIterator { underlying }
     }
 
+    /// Returns a vector of top `k` facets with their counts, sorted highest-to-lowest by counts.
+    /// See the documentation for `FacetCollector` for a usage example.
     pub fn top_k<T>(&self, facet: T, k: usize) -> Vec<(&Facet, u64)>
     where
         Facet: From<T>,
@@ -502,22 +459,24 @@ impl FacetCounts {
 #[cfg(test)]
 mod tests {
     use super::{FacetCollector, FacetCounts};
-    use core::Index;
-    use query::AllQuery;
+    use crate::collector::Count;
+    use crate::core::Index;
+    use crate::query::{AllQuery, QueryParser, TermQuery};
+    use crate::schema::{Document, Facet, Field, IndexRecordOption, Schema, INDEXED};
+    use crate::Term;
     use rand::distributions::Uniform;
+    use rand::prelude::SliceRandom;
     use rand::{thread_rng, Rng};
-    use schema::Field;
-    use schema::{Document, Facet, SchemaBuilder};
     use std::iter;
 
     #[test]
     fn test_facet_collector_drilldown() {
-        let mut schema_builder = SchemaBuilder::new();
-        let facet_field = schema_builder.add_facet_field("facet");
+        let mut schema_builder = Schema::builder();
+        let facet_field = schema_builder.add_facet_field("facet", INDEXED);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
 
-        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        let mut index_writer = index.writer_for_tests().unwrap();
         let num_facets: usize = 3 * 4 * 5;
         let facets: Vec<Facet> = (0..num_facets)
             .map(|mut n| {
@@ -527,21 +486,20 @@ mod tests {
                 n /= 4;
                 let leaf = n % 5;
                 Facet::from(&format!("/top{}/mid{}/leaf{}", top, mid, leaf))
-            }).collect();
+            })
+            .collect();
         for i in 0..num_facets * 10 {
             let mut doc = Document::new();
             doc.add_facet(facet_field, facets[i % num_facets].clone());
             index_writer.add_document(doc);
         }
         index_writer.commit().unwrap();
-        index.load_searchers().unwrap();
-        let searcher = index.searcher();
-
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
         let mut facet_collector = FacetCollector::for_field(facet_field);
         facet_collector.add_facet(Facet::from("/top1"));
-        searcher.search(&AllQuery, &mut facet_collector).unwrap();
+        let counts = searcher.search(&AllQuery, &facet_collector).unwrap();
 
-        let counts: FacetCounts = facet_collector.harvest();
         {
             let facets: Vec<(String, u64)> = counts
                 .get("/top1")
@@ -555,60 +513,107 @@ mod tests {
                     ("/top1/mid2", 50),
                     ("/top1/mid3", 50),
                 ]
-                    .iter()
-                    .map(|&(facet_str, count)| (String::from(facet_str), count))
-                    .collect::<Vec<_>>()
+                .iter()
+                .map(|&(facet_str, count)| (String::from(facet_str), count))
+                .collect::<Vec<_>>()
             );
         }
     }
 
     #[test]
-    #[should_panic(
-        expected = "Tried to add a facet which is a descendant of \
-                    an already added facet."
-    )]
+    #[should_panic(expected = "Tried to add a facet which is a descendant of \
+                               an already added facet.")]
     fn test_misused_facet_collector() {
-        let mut facet_collector = FacetCollector::for_field(Field(0));
+        let mut facet_collector = FacetCollector::for_field(Field::from_field_id(0));
         facet_collector.add_facet(Facet::from("/country"));
         facet_collector.add_facet(Facet::from("/country/europe"));
     }
 
     #[test]
     fn test_doc_unsorted_multifacet() {
-        let mut schema_builder = SchemaBuilder::new();
-        let facet_field = schema_builder.add_facet_field("facets");
+        let mut schema_builder = Schema::builder();
+        let facet_field = schema_builder.add_facet_field("facets", INDEXED);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        let mut index_writer = index.writer_for_tests().unwrap();
         index_writer.add_document(doc!(
-            facet_field => Facet::from_text(&"/subjects/A/a"),
-            facet_field => Facet::from_text(&"/subjects/B/a"),
-            facet_field => Facet::from_text(&"/subjects/A/b"),
-            facet_field => Facet::from_text(&"/subjects/B/b"),
+            facet_field => Facet::from_text(&"/subjects/A/a").unwrap(),
+            facet_field => Facet::from_text(&"/subjects/B/a").unwrap(),
+            facet_field => Facet::from_text(&"/subjects/A/b").unwrap(),
+            facet_field => Facet::from_text(&"/subjects/B/b").unwrap(),
         ));
         index_writer.commit().unwrap();
-        index.load_searchers().unwrap();
-        let searcher = index.searcher();
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
         assert_eq!(searcher.num_docs(), 1);
         let mut facet_collector = FacetCollector::for_field(facet_field);
         facet_collector.add_facet("/subjects");
-        searcher.search(&AllQuery, &mut facet_collector).unwrap();
-        let counts = facet_collector.harvest();
+        let counts = searcher.search(&AllQuery, &facet_collector).unwrap();
         let facets: Vec<(&Facet, u64)> = counts.get("/subjects").collect();
         assert_eq!(facets[0].1, 1);
     }
 
+    #[test]
+    fn test_doc_search_by_facet() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        let facet_field = schema_builder.add_facet_field("facet", INDEXED);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_for_tests()?;
+        index_writer.add_document(doc!(
+            facet_field => Facet::from_text(&"/A/A").unwrap(),
+        ));
+        index_writer.add_document(doc!(
+            facet_field => Facet::from_text(&"/A/B").unwrap(),
+        ));
+        index_writer.add_document(doc!(
+            facet_field => Facet::from_text(&"/A/C/A").unwrap(),
+        ));
+        index_writer.add_document(doc!(
+            facet_field => Facet::from_text(&"/D/C/A").unwrap(),
+        ));
+        index_writer.commit()?;
+        let reader = index.reader()?;
+        let searcher = reader.searcher();
+        assert_eq!(searcher.num_docs(), 4);
+
+        let count_facet = |facet_str: &str| {
+            let term = Term::from_facet(facet_field, &Facet::from_text(facet_str).unwrap());
+            searcher
+                .search(&TermQuery::new(term, IndexRecordOption::Basic), &Count)
+                .unwrap()
+        };
+
+        assert_eq!(count_facet("/"), 4);
+        assert_eq!(count_facet("/A"), 3);
+        assert_eq!(count_facet("/A/B"), 1);
+        assert_eq!(count_facet("/A/C"), 1);
+        assert_eq!(count_facet("/A/C/A"), 1);
+        assert_eq!(count_facet("/C/A"), 0);
+
+        let query_parser = QueryParser::for_index(&index, vec![]);
+        {
+            let query = query_parser.parse_query("facet:/A/B")?;
+            assert_eq!(1, searcher.search(&query, &Count).unwrap());
+        }
+        {
+            let query = query_parser.parse_query("facet:/A")?;
+            assert_eq!(3, searcher.search(&query, &Count)?);
+        }
+        Ok(())
+    }
+
     #[test]
     fn test_non_used_facet_collector() {
-        let mut facet_collector = FacetCollector::for_field(Field(0));
+        let mut facet_collector = FacetCollector::for_field(Field::from_field_id(0));
         facet_collector.add_facet(Facet::from("/country"));
         facet_collector.add_facet(Facet::from("/countryeurope"));
     }
 
     #[test]
     fn test_facet_collector_topk() {
-        let mut schema_builder = SchemaBuilder::new();
-        let facet_field = schema_builder.add_facet_field("facet");
+        let mut schema_builder = Schema::builder();
+        let facet_field = schema_builder.add_facet_field("facet", INDEXED);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
 
@@ -619,29 +624,28 @@ mod tests {
                 let facet = Facet::from(&format!("/facet/{}", c));
                 let doc = doc!(facet_field => facet);
                 iter::repeat(doc).take(count)
-            }).map(|mut doc| {
+            })
+            .map(|mut doc| {
                 doc.add_facet(
                     facet_field,
                     &format!("/facet/{}", thread_rng().sample(&uniform)),
                 );
                 doc
-            }).collect();
-        thread_rng().shuffle(&mut docs[..]);
+            })
+            .collect();
+        docs[..].shuffle(&mut thread_rng());
 
-        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        let mut index_writer = index.writer_for_tests().unwrap();
         for doc in docs {
             index_writer.add_document(doc);
         }
         index_writer.commit().unwrap();
-        index.load_searchers().unwrap();
-
-        let searcher = index.searcher();
+        let searcher = index.reader().unwrap().searcher();
 
         let mut facet_collector = FacetCollector::for_field(facet_field);
         facet_collector.add_facet("/facet");
-        searcher.search(&AllQuery, &mut facet_collector).unwrap();
+        let counts: FacetCounts = searcher.search(&AllQuery, &facet_collector).unwrap();
 
-        let counts: FacetCounts = facet_collector.harvest();
         {
             let facets: Vec<(&Facet, u64)> = counts.top_k("/facet", 3);
             assert_eq!(
@@ -655,23 +659,57 @@ mod tests {
         }
     }
 
+    #[test]
+    fn test_facet_collector_topk_tie_break() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        let facet_field = schema_builder.add_facet_field("facet", INDEXED);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+
+        let docs: Vec<Document> = vec![("b", 2), ("a", 2), ("c", 4)]
+            .into_iter()
+            .flat_map(|(c, count)| {
+                let facet = Facet::from(&format!("/facet/{}", c));
+                let doc = doc!(facet_field => facet);
+                iter::repeat(doc).take(count)
+            })
+            .collect();
+
+        let mut index_writer = index.writer_for_tests()?;
+        for doc in docs {
+            index_writer.add_document(doc);
+        }
+        index_writer.commit()?;
+
+        let searcher = index.reader()?.searcher();
+        let mut facet_collector = FacetCollector::for_field(facet_field);
+        facet_collector.add_facet("/facet");
+        let counts: FacetCounts = searcher.search(&AllQuery, &facet_collector)?;
+
+        let facets: Vec<(&Facet, u64)> = counts.top_k("/facet", 2);
+        assert_eq!(
+            facets,
+            vec![(&Facet::from("/facet/c"), 4), (&Facet::from("/facet/a"), 2)]
+        );
+        Ok(())
+    }
 }
 
 #[cfg(all(test, feature = "unstable"))]
 mod bench {
 
-    use collector::FacetCollector;
-    use query::AllQuery;
-    use rand::{thread_rng, Rng};
-    use schema::Facet;
-    use schema::SchemaBuilder;
+    use crate::collector::FacetCollector;
+    use crate::query::AllQuery;
+    use crate::schema::{Facet, Schema, INDEXED};
+    use crate::Index;
+    use rand::seq::SliceRandom;
+    use rand::thread_rng;
     use test::Bencher;
-    use Index;
 
     #[bench]
     fn bench_facet_collector(b: &mut Bencher) {
-        let mut schema_builder = SchemaBuilder::new();
-        let facet_field = schema_builder.add_facet_field("facet");
+        let mut schema_builder = Schema::builder();
+        let facet_field = schema_builder.add_facet_field("facet", INDEXED);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
 
@@ -683,19 +721,18 @@ mod bench {
             }
         }
         // 40425 docs
-        thread_rng().shuffle(&mut docs[..]);
+        docs[..].shuffle(&mut thread_rng());
 
-        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        let mut index_writer = index.writer_for_tests().unwrap();
         for doc in docs {
             index_writer.add_document(doc);
         }
         index_writer.commit().unwrap();
-        index.load_searchers().unwrap();
-
+        let reader = index.reader().unwrap();
         b.iter(|| {
-            let searcher = index.searcher();
-            let mut facet_collector = FacetCollector::for_field(facet_field);
-            searcher.search(&AllQuery, &mut facet_collector).unwrap();
+            let searcher = reader.searcher();
+            let facet_collector = FacetCollector::for_field(facet_field);
+            searcher.search(&AllQuery, &facet_collector).unwrap();
         });
     }
 }

src/collector/filter_collector_wrapper.rs

@@ -0,0 +1,183 @@
+// # Custom collector example
+//
+// This example shows how you can implement your own
+// collector. As an example, we will compute a collector
+// that computes the standard deviation of a given fast field.
+//
+// Of course, you can have a look at the tantivy's built-in collectors
+// such as the `CountCollector` for more examples.
+
+// ---
+// Importing tantivy...
+use std::marker::PhantomData;
+
+use crate::collector::{Collector, SegmentCollector};
+use crate::fastfield::{DynamicFastFieldReader, FastFieldReader, FastValue};
+use crate::schema::Field;
+use crate::{Score, SegmentReader, TantivyError};
+
+/// The `FilterCollector` collector filters docs using a fast field value and a predicate.
+/// Only the documents for which the predicate returned "true" will be passed on to the next collector.
+///
+/// ```rust
+/// use tantivy::collector::{TopDocs, FilterCollector};
+/// use tantivy::query::QueryParser;
+/// use tantivy::schema::{Schema, TEXT, INDEXED, FAST};
+/// use tantivy::{doc, DocAddress, Index};
+///
+/// let mut schema_builder = Schema::builder();
+/// let title = schema_builder.add_text_field("title", TEXT);
+/// let price = schema_builder.add_u64_field("price", INDEXED | FAST);
+/// let schema = schema_builder.build();
+/// let index = Index::create_in_ram(schema);
+///
+/// let mut index_writer = index.writer_with_num_threads(1, 10_000_000).unwrap();
+/// index_writer.add_document(doc!(title => "The Name of the Wind", price => 30_200u64));
+/// index_writer.add_document(doc!(title => "The Diary of Muadib", price => 29_240u64));
+/// index_writer.add_document(doc!(title => "A Dairy Cow", price => 21_240u64));
+/// index_writer.add_document(doc!(title => "The Diary of a Young Girl", price => 20_120u64));
+/// assert!(index_writer.commit().is_ok());
+///
+/// let reader = index.reader().unwrap();
+/// let searcher = reader.searcher();
+///
+/// let query_parser = QueryParser::for_index(&index, vec![title]);
+/// let query = query_parser.parse_query("diary").unwrap();
+/// let no_filter_collector = FilterCollector::new(price, &|value: u64| value > 20_120u64, TopDocs::with_limit(2));
+/// let top_docs = searcher.search(&query, &no_filter_collector).unwrap();
+///
+/// assert_eq!(top_docs.len(), 1);
+/// assert_eq!(top_docs[0].1, DocAddress::new(0, 1));
+///
+/// let filter_all_collector: FilterCollector<_, _, u64> = FilterCollector::new(price, &|value| value < 5u64, TopDocs::with_limit(2));
+/// let filtered_top_docs = searcher.search(&query, &filter_all_collector).unwrap();
+///
+/// assert_eq!(filtered_top_docs.len(), 0);
+/// ```
+pub struct FilterCollector<TCollector, TPredicate, TPredicateValue: FastValue>
+where
+    TPredicate: 'static + Clone,
+{
+    field: Field,
+    collector: TCollector,
+    predicate: TPredicate,
+    t_predicate_value: PhantomData<TPredicateValue>,
+}
+
+impl<TCollector, TPredicate, TPredicateValue: FastValue>
+    FilterCollector<TCollector, TPredicate, TPredicateValue>
+where
+    TCollector: Collector + Send + Sync,
+    TPredicate: Fn(TPredicateValue) -> bool + Send + Sync + Clone,
+{
+    /// Create a new FilterCollector.
+    pub fn new(
+        field: Field,
+        predicate: TPredicate,
+        collector: TCollector,
+    ) -> FilterCollector<TCollector, TPredicate, TPredicateValue> {
+        FilterCollector {
+            field,
+            predicate,
+            collector,
+            t_predicate_value: PhantomData,
+        }
+    }
+}
+
+impl<TCollector, TPredicate, TPredicateValue: FastValue> Collector
+    for FilterCollector<TCollector, TPredicate, TPredicateValue>
+where
+    TCollector: Collector + Send + Sync,
+    TPredicate: 'static + Fn(TPredicateValue) -> bool + Send + Sync + Clone,
+    TPredicateValue: FastValue,
+{
+    // That's the type of our result.
+    // Our standard deviation will be a float.
+    type Fruit = TCollector::Fruit;
+
+    type Child = FilterSegmentCollector<TCollector::Child, TPredicate, TPredicateValue>;
+
+    fn for_segment(
+        &self,
+        segment_local_id: u32,
+        segment_reader: &SegmentReader,
+    ) -> crate::Result<FilterSegmentCollector<TCollector::Child, TPredicate, TPredicateValue>> {
+        let schema = segment_reader.schema();
+        let field_entry = schema.get_field_entry(self.field);
+        if !field_entry.is_fast() {
+            return Err(TantivyError::SchemaError(format!(
+                "Field {:?} is not a fast field.",
+                field_entry.name()
+            )));
+        }
+        let requested_type = TPredicateValue::to_type();
+        let field_schema_type = field_entry.field_type().value_type();
+        if requested_type != field_schema_type {
+            return Err(TantivyError::SchemaError(format!(
+                "Field {:?} is of type {:?}!={:?}",
+                field_entry.name(),
+                requested_type,
+                field_schema_type
+            )));
+        }
+
+        let fast_field_reader = segment_reader
+            .fast_fields()
+            .typed_fast_field_reader(self.field)?;
+
+        let segment_collector = self
+            .collector
+            .for_segment(segment_local_id, segment_reader)?;
+
+        Ok(FilterSegmentCollector {
+            fast_field_reader,
+            segment_collector,
+            predicate: self.predicate.clone(),
+            t_predicate_value: PhantomData,
+        })
+    }
+
+    fn requires_scoring(&self) -> bool {
+        self.collector.requires_scoring()
+    }
+
+    fn merge_fruits(
+        &self,
+        segment_fruits: Vec<<TCollector::Child as SegmentCollector>::Fruit>,
+    ) -> crate::Result<TCollector::Fruit> {
+        self.collector.merge_fruits(segment_fruits)
+    }
+}
+
+pub struct FilterSegmentCollector<TSegmentCollector, TPredicate, TPredicateValue>
+where
+    TPredicate: 'static,
+    TPredicateValue: FastValue,
+{
+    fast_field_reader: DynamicFastFieldReader<TPredicateValue>,
+    segment_collector: TSegmentCollector,
+    predicate: TPredicate,
+    t_predicate_value: PhantomData<TPredicateValue>,
+}
+
+impl<TSegmentCollector, TPredicate, TPredicateValue> SegmentCollector
+    for FilterSegmentCollector<TSegmentCollector, TPredicate, TPredicateValue>
+where
+    TSegmentCollector: SegmentCollector,
+    TPredicate: 'static + Fn(TPredicateValue) -> bool + Send + Sync,
+    TPredicateValue: FastValue,
+{
+    type Fruit = TSegmentCollector::Fruit;
+
+    fn collect(&mut self, doc: u32, score: Score) {
+        let value = self.fast_field_reader.get(doc);
+        if (self.predicate)(value) {
+            self.segment_collector.collect(doc, score)
+        }
+    }
+
+    fn harvest(self) -> <TSegmentCollector as SegmentCollector>::Fruit {
+        self.segment_collector.harvest()
+    }
+}

src/collector/histogram_collector.rs

@@ -0,0 +1,291 @@
+use crate::collector::{Collector, SegmentCollector};
+use crate::fastfield::{DynamicFastFieldReader, FastFieldReader, FastValue};
+use crate::schema::{Field, Type};
+use crate::{DocId, Score};
+use fastdivide::DividerU64;
+
+/// Histogram builds an histogram of the values of a fastfield for the
+/// collected DocSet.
+///
+/// At construction, it is given parameters that define a partition of an interval
+/// [min_val, max_val) into N buckets with the same width.
+/// The ith bucket is then defined by `[min_val + i * bucket_width, min_val + (i+1) * bucket_width)`
+///
+/// An histogram is then defined as a `Vec<u64>` of length `num_buckets`, that contains a count of
+/// documents for each value bucket.
+///
+/// See also [`HistogramCollector::new()`].
+///
+/// # Warning
+///
+/// f64 field. are not supported.
+#[derive(Clone)]
+pub struct HistogramCollector {
+    min_value: u64,
+    num_buckets: usize,
+    divider: DividerU64,
+    field: Field,
+}
+
+impl HistogramCollector {
+    /// Builds a new HistogramCollector.
+    ///
+    /// The scale/range of the histogram is not dynamic. It is required to
+    /// define it by supplying following parameter:
+    ///  - `min_value`: the minimum value that can be recorded in the histogram.
+    ///  - `bucket_width`: the length of the interval that is associated to each buckets.
+    ///  - `num_buckets`: The overall number of buckets.
+    ///
+    /// Together, this parameters define a partition of `[min_value, min_value + num_buckets * bucket_width)`
+    /// into `num_buckets` intervals of width bucket that we call `bucket`.
+    ///
+    /// # Disclaimer
+    /// This function panics if the field given is of type f64.
+    pub fn new<TFastValue: FastValue>(
+        field: Field,
+        min_value: TFastValue,
+        bucket_width: u64,
+        num_buckets: usize,
+    ) -> HistogramCollector {
+        let fast_type = TFastValue::to_type();
+        assert!(fast_type == Type::U64 || fast_type == Type::I64 || fast_type == Type::Date);
+        HistogramCollector {
+            min_value: min_value.to_u64(),
+            num_buckets,
+            field,
+            divider: DividerU64::divide_by(bucket_width),
+        }
+    }
+}
+
+struct HistogramComputer {
+    counts: Vec<u64>,
+    min_value: u64,
+    divider: DividerU64,
+}
+
+impl HistogramComputer {
+    #[inline]
+    pub(crate) fn add_value(&mut self, value: u64) {
+        if value < self.min_value {
+            return;
+        }
+        let delta = value - self.min_value;
+        let delta_u64 = delta.to_u64();
+        let bucket_id: usize = self.divider.divide(delta_u64) as usize;
+        if bucket_id < self.counts.len() {
+            self.counts[bucket_id] += 1;
+        }
+    }
+
+    fn harvest(self) -> Vec<u64> {
+        self.counts
+    }
+}
+pub struct SegmentHistogramCollector {
+    histogram_computer: HistogramComputer,
+    ff_reader: DynamicFastFieldReader<u64>,
+}
+
+impl SegmentCollector for SegmentHistogramCollector {
+    type Fruit = Vec<u64>;
+
+    fn collect(&mut self, doc: DocId, _score: Score) {
+        let value = self.ff_reader.get(doc);
+        self.histogram_computer.add_value(value);
+    }
+
+    fn harvest(self) -> Self::Fruit {
+        self.histogram_computer.harvest()
+    }
+}
+
+impl Collector for HistogramCollector {
+    type Fruit = Vec<u64>;
+    type Child = SegmentHistogramCollector;
+
+    fn for_segment(
+        &self,
+        _segment_local_id: crate::SegmentOrdinal,
+        segment: &crate::SegmentReader,
+    ) -> crate::Result<Self::Child> {
+        let ff_reader = segment.fast_fields().u64_lenient(self.field)?;
+        Ok(SegmentHistogramCollector {
+            histogram_computer: HistogramComputer {
+                counts: vec![0; self.num_buckets],
+                min_value: self.min_value,
+                divider: self.divider,
+            },
+            ff_reader,
+        })
+    }
+
+    fn requires_scoring(&self) -> bool {
+        false
+    }
+
+    fn merge_fruits(&self, child_histograms: Vec<Vec<u64>>) -> crate::Result<Vec<u64>> {
+        Ok(add_vecs(child_histograms, self.num_buckets))
+    }
+}
+
+pub fn add_arrays_into(acc: &mut [u64], add: &[u64]) {
+    assert_eq!(acc.len(), add.len());
+    for (dest_bucket, bucket_count) in acc.iter_mut().zip(add) {
+        *dest_bucket += bucket_count;
+    }
+}
+
+fn add_vecs(mut vals_list: Vec<Vec<u64>>, len: usize) -> Vec<u64> {
+    let mut acc = vals_list.pop().unwrap_or_else(|| vec![0u64; len]);
+    assert_eq!(acc.len(), len);
+    for vals in vals_list {
+        add_arrays_into(&mut acc, &vals);
+    }
+    acc
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{add_vecs, HistogramCollector, HistogramComputer};
+    use crate::chrono::{TimeZone, Utc};
+    use crate::schema::{Schema, FAST};
+    use crate::{doc, query, Index};
+    use fastdivide::DividerU64;
+    use query::AllQuery;
+
+    #[test]
+    fn test_add_histograms_simple() {
+        assert_eq!(
+            add_vecs(vec![vec![1, 0, 3], vec![11, 2, 3], vec![0, 0, 1]], 3),
+            vec![12, 2, 7]
+        )
+    }
+
+    #[test]
+    fn test_add_histograms_empty() {
+        assert_eq!(add_vecs(vec![], 3), vec![0, 0, 0])
+    }
+
+    #[test]
+    fn test_histogram_builder_simple() {
+        // [1..3)
+        // [3..5)
+        // ..
+        // [9..11)
+        let mut histogram_computer = HistogramComputer {
+            counts: vec![0; 5],
+            min_value: 1,
+            divider: DividerU64::divide_by(2),
+        };
+        histogram_computer.add_value(1);
+        histogram_computer.add_value(7);
+        assert_eq!(histogram_computer.harvest(), vec![1, 0, 0, 1, 0]);
+    }
+
+    #[test]
+    fn test_histogram_too_low_is_ignored() {
+        let mut histogram_computer = HistogramComputer {
+            counts: vec![0; 5],
+            min_value: 2,
+            divider: DividerU64::divide_by(2),
+        };
+        histogram_computer.add_value(0);
+        assert_eq!(histogram_computer.harvest(), vec![0, 0, 0, 0, 0]);
+    }
+
+    #[test]
+    fn test_histogram_too_high_is_ignored() {
+        let mut histogram_computer = HistogramComputer {
+            counts: vec![0u64; 5],
+            min_value: 0,
+            divider: DividerU64::divide_by(2),
+        };
+        histogram_computer.add_value(10);
+        assert_eq!(histogram_computer.harvest(), vec![0, 0, 0, 0, 0]);
+    }
+    #[test]
+    fn test_no_segments() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        let val_field = schema_builder.add_u64_field("val_field", FAST);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let reader = index.reader()?;
+        let searcher = reader.searcher();
+        let all_query = AllQuery;
+        let histogram_collector = HistogramCollector::new(val_field, 0u64, 2, 5);
+        let histogram = searcher.search(&all_query, &histogram_collector)?;
+        assert_eq!(histogram, vec![0; 5]);
+        Ok(())
+    }
+
+    #[test]
+    fn test_histogram_i64() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        let val_field = schema_builder.add_i64_field("val_field", FAST);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut writer = index.writer_with_num_threads(1, 4_000_000)?;
+        writer.add_document(doc!(val_field=>12i64));
+        writer.add_document(doc!(val_field=>-30i64));
+        writer.add_document(doc!(val_field=>-12i64));
+        writer.add_document(doc!(val_field=>-10i64));
+        writer.commit()?;
+        let reader = index.reader()?;
+        let searcher = reader.searcher();
+        let all_query = AllQuery;
+        let histogram_collector = HistogramCollector::new(val_field, -20i64, 10u64, 4);
+        let histogram = searcher.search(&all_query, &histogram_collector)?;
+        assert_eq!(histogram, vec![1, 1, 0, 1]);
+        Ok(())
+    }
+
+    #[test]
+    fn test_histogram_merge() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        let val_field = schema_builder.add_i64_field("val_field", FAST);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut writer = index.writer_with_num_threads(1, 4_000_000)?;
+        writer.add_document(doc!(val_field=>12i64));
+        writer.commit()?;
+        writer.add_document(doc!(val_field=>-30i64));
+        writer.commit()?;
+        writer.add_document(doc!(val_field=>-12i64));
+        writer.commit()?;
+        writer.add_document(doc!(val_field=>-10i64));
+        writer.commit()?;
+        let reader = index.reader()?;
+        let searcher = reader.searcher();
+        let all_query = AllQuery;
+        let histogram_collector = HistogramCollector::new(val_field, -20i64, 10u64, 4);
+        let histogram = searcher.search(&all_query, &histogram_collector)?;
+        assert_eq!(histogram, vec![1, 1, 0, 1]);
+        Ok(())
+    }
+
+    #[test]
+    fn test_histogram_dates() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        let date_field = schema_builder.add_date_field("date_field", FAST);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut writer = index.writer_with_num_threads(1, 4_000_000)?;
+        writer.add_document(doc!(date_field=>Utc.ymd(1982, 9, 17).and_hms(0, 0,0)));
+        writer.add_document(doc!(date_field=>Utc.ymd(1986, 3, 9).and_hms(0, 0, 0)));
+        writer.add_document(doc!(date_field=>Utc.ymd(1983, 9, 27).and_hms(0, 0, 0)));
+        writer.commit()?;
+        let reader = index.reader()?;
+        let searcher = reader.searcher();
+        let all_query = AllQuery;
+        let week_histogram_collector = HistogramCollector::new(
+            date_field,
+            Utc.ymd(1980, 1, 1).and_hms(0, 0, 0),
+            3600 * 24 * 365, // it is just for a unit test... sorry leap years.
+            10,
+        );
+        let week_histogram = searcher.search(&all_query, &week_histogram_collector)?;
+        assert_eq!(week_histogram, vec![0, 0, 1, 1, 0, 0, 1, 0, 0, 0]);
+        Ok(())
+    }
+}

src/collector/int_facet_collector.rs

@@ -1,123 +0,0 @@
-use std::cmp::Eq;
-use std::collections::HashMap;
-use std::hash::Hash;
-
-use collector::Collector;
-use fastfield::FastFieldReader;
-use schema::Field;
-
-use DocId;
-use Result;
-use Score;
-use SegmentReader;
-use SegmentLocalId;
-
-
-/// Facet collector  for i64/u64 fast field
-pub struct IntFacetCollector<T>
-where
-    T: FastFieldReader,
-    T::ValueType: Eq + Hash,
-{
-    counters: HashMap<T::ValueType, u64>,
-    field: Field,
-    ff_reader: Option<T>,
-}
-
-
-impl<T> IntFacetCollector<T>
-where
-    T: FastFieldReader,
-    T::ValueType: Eq + Hash,
-{
-    /// Creates a new facet collector for aggregating a given field.
-    pub fn new(field: Field) -> IntFacetCollector<T> {
-        IntFacetCollector {
-            counters: HashMap::new(),
-            field: field,
-            ff_reader: None,
-        }
-    }
-}
-
-
-impl<T> Collector for IntFacetCollector<T>
-where
-    T: FastFieldReader,
-    T::ValueType: Eq + Hash,
-{
-    fn set_segment(&mut self, _: SegmentLocalId, reader: &SegmentReader) -> Result<()> {
-        self.ff_reader = Some(reader.get_fast_field_reader(self.field)?);
-        Ok(())
-    }
-
-    fn collect(&mut self, doc: DocId, _: Score) {
-        let val = self.ff_reader
-            .as_ref()
-            .expect(
-                "collect() was called before set_segment. \
-                This should never happen.",
-            )
-            .get(doc);
-        *(self.counters.entry(val).or_insert(0)) += 1;
-    }
-}
-
-
-
-#[cfg(test)]
-mod tests {
-
-    use collector::{chain, IntFacetCollector};
-    use query::QueryParser;
-    use fastfield::{I64FastFieldReader, U64FastFieldReader};
-    use schema::{self, FAST, STRING};
-    use Index;
-
-    #[test]
-    // create 10 documents, set num field value to 0 or 1 for even/odd ones
-    // make sure we have facet counters correctly filled
-    fn test_facet_collector_results() {
-
-        let mut schema_builder = schema::SchemaBuilder::new();
-        let num_field_i64 = schema_builder.add_i64_field("num_i64", FAST);
-        let num_field_u64 = schema_builder.add_u64_field("num_u64", FAST);
-        let text_field = schema_builder.add_text_field("text", STRING);
-        let schema = schema_builder.build();
-
-        let index = Index::create_in_ram(schema.clone());
-
-        {
-            let mut index_writer = index.writer_with_num_threads(1, 40_000_000).unwrap();
-            {
-                for i in 0u64..10u64 {
-                    index_writer.add_document(doc!(
-                        num_field_i64 => ((i as i64) % 3i64) as i64,
-                        num_field_u64 => (i % 2u64) as u64,
-                        text_field => "text"
-                    ));
-                }
-            }
-            assert_eq!(index_writer.commit().unwrap(), 10u64);
-        }
-
-        index.load_searchers().unwrap();
-        let searcher = index.searcher();
-        let mut ffvf_i64: IntFacetCollector<I64FastFieldReader> = IntFacetCollector::new(num_field_i64);
-        let mut ffvf_u64: IntFacetCollector<U64FastFieldReader> = IntFacetCollector::new(num_field_u64);
-
-        {
-            // perform the query
-            let mut facet_collectors = chain().push(&mut ffvf_i64).push(&mut ffvf_u64);
-            let mut query_parser = QueryParser::for_index(index, vec![text_field]);
-            let query = query_parser.parse_query("text:text").unwrap();
-            query.search(&searcher, &mut facet_collectors).unwrap();
-        }
-
-        assert_eq!(ffvf_u64.counters[&0], 5);
-        assert_eq!(ffvf_u64.counters[&1], 5);
-        assert_eq!(ffvf_i64.counters[&0], 4);
-        assert_eq!(ffvf_i64.counters[&1], 3);
-
-    }
-}

src/collector/mod.rs

@@ -1,15 +1,100 @@
 /*!
-Defines how the documents matching a search query should be processed.
+
+# Collectors
+
+Collectors define the information you want to extract from the documents matching the queries.
+In tantivy jargon, we call this information your search "fruit".
+
+Your fruit could for instance be :
+- [the count of matching documents](./struct.Count.html)
+- [the top 10 documents, by relevancy or by a fast field](./struct.TopDocs.html)
+- [facet counts](./struct.FacetCollector.html)
+
+At one point in your code, you will trigger the actual search operation by calling
+[the `search(...)` method of your `Searcher` object](../struct.Searcher.html#method.search).
+This call will look like this.
+
+```verbatim
+let fruit = searcher.search(&query, &collector)?;
+```
+
+Here the type of fruit is actually determined as an associated type of the collector (`Collector::Fruit`).
+
+
+# Combining several collectors
+
+A rich search experience often requires to run several collectors on your search query.
+For instance,
+- selecting the top-K products matching your query
+- counting the matching documents
+- computing several facets
+- computing statistics about the matching product prices
+
+A simple and efficient way to do that is to pass your collectors as one tuple.
+The resulting `Fruit` will then be a typed tuple with each collector's original fruits
+in their respective position.
+
+```rust
+# use tantivy::schema::*;
+# use tantivy::*;
+# use tantivy::query::*;
+use tantivy::collector::{Count, TopDocs};
+#
+# fn main() -> tantivy::Result<()> {
+# let mut schema_builder = Schema::builder();
+#     let title = schema_builder.add_text_field("title", TEXT);
+#     let schema = schema_builder.build();
+#     let index = Index::create_in_ram(schema);
+#     let mut index_writer = index.writer(3_000_000)?;
+#       index_writer.add_document(doc!(
+#       title => "The Name of the Wind",
+#      ));
+#     index_writer.add_document(doc!(
+#        title => "The Diary of Muadib",
+#     ));
+#     index_writer.commit()?;
+#     let reader = index.reader()?;
+#     let searcher = reader.searcher();
+#     let query_parser = QueryParser::for_index(&index, vec![title]);
+#     let query = query_parser.parse_query("diary")?;
+let (doc_count, top_docs): (usize, Vec<(Score, DocAddress)>) =
+    searcher.search(&query, &(Count, TopDocs::with_limit(2)))?;
+#     Ok(())
+# }
+```
+
+The `Collector` trait is implemented for up to 4 collectors.
+If you have more than 4 collectors, you can either group them into
+tuples of tuples `(a,(b,(c,d)))`, or rely on [`MultiCollector`](./struct.MultiCollector.html).
+
+# Combining several collectors dynamically
+
+Combining collectors into a tuple is a zero-cost abstraction: everything
+happens as if you had manually implemented a single collector
+combining all of our features.
+
+Unfortunately it requires you to know at compile time your collector types.
+If on the other hand, the collectors depend on some query parameter,
+you can rely on `MultiCollector`'s.
+
+
+# Implementing your own collectors.
+
+See the `custom_collector` example.
+
 */
 
-use DocId;
-use Result;
-use Score;
-use SegmentLocalId;
-use SegmentReader;
+use crate::DocId;
+use crate::Score;
+use crate::SegmentOrdinal;
+use crate::SegmentReader;
+use downcast_rs::impl_downcast;
 
 mod count_collector;
-pub use self::count_collector::CountCollector;
+pub use self::count_collector::Count;
+
+mod histogram_collector;
+pub use histogram_collector::HistogramCollector;
 
 mod multi_collector;
 pub use self::multi_collector::MultiCollector;
@@ -17,237 +102,374 @@ pub use self::multi_collector::MultiCollector;
 mod top_collector;
 
 mod top_score_collector;
-pub use self::top_score_collector::TopScoreCollector;
-#[deprecated]
-pub use self::top_score_collector::TopScoreCollector as TopCollector;
+pub use self::top_score_collector::TopDocs;
 
-mod top_field_collector;
-pub use self::top_field_collector::TopFieldCollector;
+mod custom_score_top_collector;
+pub use self::custom_score_top_collector::{CustomScorer, CustomSegmentScorer};
+
+mod tweak_score_top_collector;
+pub use self::tweak_score_top_collector::{ScoreSegmentTweaker, ScoreTweaker};
 
 mod facet_collector;
 pub use self::facet_collector::FacetCollector;
+pub use self::facet_collector::FacetCounts;
+use crate::query::Weight;
 
-mod chained_collector;
-pub use self::chained_collector::{chain, ChainedCollector};
+mod docset_collector;
+pub use self::docset_collector::DocSetCollector;
+
+mod filter_collector_wrapper;
+pub use self::filter_collector_wrapper::FilterCollector;
+
+/// `Fruit` is the type for the result of our collection.
+/// e.g. `usize` for the `Count` collector.
+pub trait Fruit: Send + downcast_rs::Downcast {}
+
+impl<T> Fruit for T where T: Send + downcast_rs::Downcast {}
 
 /// Collectors are in charge of collecting and retaining relevant
 /// information from the document found and scored by the query.
 ///
-///
 /// For instance,
 ///
 /// - keeping track of the top 10 best documents
 /// - computing a breakdown over a fast field
 /// - computing the number of documents matching the query
 ///
-/// Queries are in charge of pushing the `DocSet` to the collector.
+/// Our search index is in fact a collection of segments, so
+/// a `Collector` trait is actually more of a factory to instance
+/// `SegmentCollector`s for each segments.
 ///
-/// As they work on multiple segments, they first inform
-/// the collector of a change in a segment and then
-/// call the `collect` method to push the document to the collector.
-///
-/// Temporally, our collector will receive calls
-/// - `.set_segment(0, segment_reader_0)`
-/// - `.collect(doc0_of_segment_0)`
-/// - `.collect(...)`
-/// - `.collect(last_doc_of_segment_0)`
-/// - `.set_segment(1, segment_reader_1)`
-/// - `.collect(doc0_of_segment_1)`
-/// - `.collect(...)`
-/// - `.collect(last_doc_of_segment_1)`
-/// - `...`
-/// - `.collect(last_doc_of_last_segment)`
+/// The collection logic itself is in the `SegmentCollector`.
 ///
 /// Segments are not guaranteed to be visited in any specific order.
-pub trait Collector {
+pub trait Collector: Sync + Send {
+    /// `Fruit` is the type for the result of our collection.
+    /// e.g. `usize` for the `Count` collector.
+    type Fruit: Fruit;
+
+    /// Type of the `SegmentCollector` associated to this collector.
+    type Child: SegmentCollector;
+
     /// `set_segment` is called before beginning to enumerate
     /// on this segment.
-    fn set_segment(
-        &mut self,
-        segment_local_id: SegmentLocalId,
+    fn for_segment(
+        &self,
+        segment_local_id: SegmentOrdinal,
         segment: &SegmentReader,
-    ) -> Result<()>;
-    /// The query pushes the scored document to the collector via this method.
-    fn collect(&mut self, doc: DocId, score: Score);
+    ) -> crate::Result<Self::Child>;
 
     /// Returns true iff the collector requires to compute scores for documents.
     fn requires_scoring(&self) -> bool;
+
+    /// Combines the fruit associated to the collection of each segments
+    /// into one fruit.
+    fn merge_fruits(
+        &self,
+        segment_fruits: Vec<<Self::Child as SegmentCollector>::Fruit>,
+    ) -> crate::Result<Self::Fruit>;
+
+    /// Created a segment collector and
+    fn collect_segment(
+        &self,
+        weight: &dyn Weight,
+        segment_ord: u32,
+        reader: &SegmentReader,
+    ) -> crate::Result<<Self::Child as SegmentCollector>::Fruit> {
+        let mut segment_collector = self.for_segment(segment_ord as u32, reader)?;
+
+        if let Some(alive_bitset) = reader.alive_bitset() {
+            weight.for_each(reader, &mut |doc, score| {
+                if alive_bitset.is_alive(doc) {
+                    segment_collector.collect(doc, score);
+                }
+            })?;
+        } else {
+            weight.for_each(reader, &mut |doc, score| {
+                segment_collector.collect(doc, score);
+            })?;
+        }
+        Ok(segment_collector.harvest())
+    }
 }
 
-impl<'a, C: Collector> Collector for &'a mut C {
-    fn set_segment(
-        &mut self,
-        segment_local_id: SegmentLocalId,
-        segment: &SegmentReader,
-    ) -> Result<()> {
-        (*self).set_segment(segment_local_id, segment)
-    }
-    /// The query pushes the scored document to the collector via this method.
+impl<TSegmentCollector: SegmentCollector> SegmentCollector for Option<TSegmentCollector> {
+    type Fruit = Option<TSegmentCollector::Fruit>;
+
     fn collect(&mut self, doc: DocId, score: Score) {
-        C::collect(self, doc, score)
+        if let Some(segment_collector) = self {
+            segment_collector.collect(doc, score);
+        }
+    }
+
+    fn harvest(self) -> Self::Fruit {
+        self.map(|segment_collector| segment_collector.harvest())
+    }
+}
+
+impl<TCollector: Collector> Collector for Option<TCollector> {
+    type Fruit = Option<TCollector::Fruit>;
+
+    type Child = Option<<TCollector as Collector>::Child>;
+
+    fn for_segment(
+        &self,
+        segment_local_id: SegmentOrdinal,
+        segment: &SegmentReader,
+    ) -> crate::Result<Self::Child> {
+        Ok(if let Some(inner) = self {
+            let inner_segment_collector = inner.for_segment(segment_local_id, segment)?;
+            Some(inner_segment_collector)
+        } else {
+            None
+        })
     }
 
     fn requires_scoring(&self) -> bool {
-        C::requires_scoring(self)
+        self.as_ref()
+            .map(|inner| inner.requires_scoring())
+            .unwrap_or(false)
+    }
+
+    fn merge_fruits(
+        &self,
+        segment_fruits: Vec<<Self::Child as SegmentCollector>::Fruit>,
+    ) -> crate::Result<Self::Fruit> {
+        if let Some(inner) = self.as_ref() {
+            let inner_segment_fruits: Vec<_> = segment_fruits
+                .into_iter()
+                .flat_map(|fruit_opt| fruit_opt.into_iter())
+                .collect();
+            let fruit = inner.merge_fruits(inner_segment_fruits)?;
+            Ok(Some(fruit))
+        } else {
+            Ok(None)
+        }
     }
 }
 
+/// The `SegmentCollector` is the trait in charge of defining the
+/// collect operation at the scale of the segment.
+///
+/// `.collect(doc, score)` will be called for every documents
+/// matching the query.
+pub trait SegmentCollector: 'static {
+    /// `Fruit` is the type for the result of our collection.
+    /// e.g. `usize` for the `Count` collector.
+    type Fruit: Fruit;
+
+    /// The query pushes the scored document to the collector via this method.
+    fn collect(&mut self, doc: DocId, score: Score);
+
+    /// Extract the fruit of the collection from the `SegmentCollector`.
+    fn harvest(self) -> Self::Fruit;
+}
+
+// -----------------------------------------------
+// Tuple implementations.
+
+impl<Left, Right> Collector for (Left, Right)
+where
+    Left: Collector,
+    Right: Collector,
+{
+    type Fruit = (Left::Fruit, Right::Fruit);
+    type Child = (Left::Child, Right::Child);
+
+    fn for_segment(
+        &self,
+        segment_local_id: u32,
+        segment: &SegmentReader,
+    ) -> crate::Result<Self::Child> {
+        let left = self.0.for_segment(segment_local_id, segment)?;
+        let right = self.1.for_segment(segment_local_id, segment)?;
+        Ok((left, right))
+    }
+
+    fn requires_scoring(&self) -> bool {
+        self.0.requires_scoring() || self.1.requires_scoring()
+    }
+
+    fn merge_fruits(
+        &self,
+        segment_fruits: Vec<<Self::Child as SegmentCollector>::Fruit>,
+    ) -> crate::Result<(Left::Fruit, Right::Fruit)> {
+        let mut left_fruits = vec![];
+        let mut right_fruits = vec![];
+        for (left_fruit, right_fruit) in segment_fruits {
+            left_fruits.push(left_fruit);
+            right_fruits.push(right_fruit);
+        }
+        Ok((
+            self.0.merge_fruits(left_fruits)?,
+            self.1.merge_fruits(right_fruits)?,
+        ))
+    }
+}
+
+impl<Left, Right> SegmentCollector for (Left, Right)
+where
+    Left: SegmentCollector,
+    Right: SegmentCollector,
+{
+    type Fruit = (Left::Fruit, Right::Fruit);
+
+    fn collect(&mut self, doc: DocId, score: Score) {
+        self.0.collect(doc, score);
+        self.1.collect(doc, score);
+    }
+
+    fn harvest(self) -> <Self as SegmentCollector>::Fruit {
+        (self.0.harvest(), self.1.harvest())
+    }
+}
+
+// 3-Tuple
+
+impl<One, Two, Three> Collector for (One, Two, Three)
+where
+    One: Collector,
+    Two: Collector,
+    Three: Collector,
+{
+    type Fruit = (One::Fruit, Two::Fruit, Three::Fruit);
+    type Child = (One::Child, Two::Child, Three::Child);
+
+    fn for_segment(
+        &self,
+        segment_local_id: u32,
+        segment: &SegmentReader,
+    ) -> crate::Result<Self::Child> {
+        let one = self.0.for_segment(segment_local_id, segment)?;
+        let two = self.1.for_segment(segment_local_id, segment)?;
+        let three = self.2.for_segment(segment_local_id, segment)?;
+        Ok((one, two, three))
+    }
+
+    fn requires_scoring(&self) -> bool {
+        self.0.requires_scoring() || self.1.requires_scoring() || self.2.requires_scoring()
+    }
+
+    fn merge_fruits(
+        &self,
+        children: Vec<<Self::Child as SegmentCollector>::Fruit>,
+    ) -> crate::Result<Self::Fruit> {
+        let mut one_fruits = vec![];
+        let mut two_fruits = vec![];
+        let mut three_fruits = vec![];
+        for (one_fruit, two_fruit, three_fruit) in children {
+            one_fruits.push(one_fruit);
+            two_fruits.push(two_fruit);
+            three_fruits.push(three_fruit);
+        }
+        Ok((
+            self.0.merge_fruits(one_fruits)?,
+            self.1.merge_fruits(two_fruits)?,
+            self.2.merge_fruits(three_fruits)?,
+        ))
+    }
+}
+
+impl<One, Two, Three> SegmentCollector for (One, Two, Three)
+where
+    One: SegmentCollector,
+    Two: SegmentCollector,
+    Three: SegmentCollector,
+{
+    type Fruit = (One::Fruit, Two::Fruit, Three::Fruit);
+
+    fn collect(&mut self, doc: DocId, score: Score) {
+        self.0.collect(doc, score);
+        self.1.collect(doc, score);
+        self.2.collect(doc, score);
+    }
+
+    fn harvest(self) -> <Self as SegmentCollector>::Fruit {
+        (self.0.harvest(), self.1.harvest(), self.2.harvest())
+    }
+}
+
+// 4-Tuple
+
+impl<One, Two, Three, Four> Collector for (One, Two, Three, Four)
+where
+    One: Collector,
+    Two: Collector,
+    Three: Collector,
+    Four: Collector,
+{
+    type Fruit = (One::Fruit, Two::Fruit, Three::Fruit, Four::Fruit);
+    type Child = (One::Child, Two::Child, Three::Child, Four::Child);
+
+    fn for_segment(
+        &self,
+        segment_local_id: u32,
+        segment: &SegmentReader,
+    ) -> crate::Result<Self::Child> {
+        let one = self.0.for_segment(segment_local_id, segment)?;
+        let two = self.1.for_segment(segment_local_id, segment)?;
+        let three = self.2.for_segment(segment_local_id, segment)?;
+        let four = self.3.for_segment(segment_local_id, segment)?;
+        Ok((one, two, three, four))
+    }
+
+    fn requires_scoring(&self) -> bool {
+        self.0.requires_scoring()
+            || self.1.requires_scoring()
+            || self.2.requires_scoring()
+            || self.3.requires_scoring()
+    }
+
+    fn merge_fruits(
+        &self,
+        children: Vec<<Self::Child as SegmentCollector>::Fruit>,
+    ) -> crate::Result<Self::Fruit> {
+        let mut one_fruits = vec![];
+        let mut two_fruits = vec![];
+        let mut three_fruits = vec![];
+        let mut four_fruits = vec![];
+        for (one_fruit, two_fruit, three_fruit, four_fruit) in children {
+            one_fruits.push(one_fruit);
+            two_fruits.push(two_fruit);
+            three_fruits.push(three_fruit);
+            four_fruits.push(four_fruit);
+        }
+        Ok((
+            self.0.merge_fruits(one_fruits)?,
+            self.1.merge_fruits(two_fruits)?,
+            self.2.merge_fruits(three_fruits)?,
+            self.3.merge_fruits(four_fruits)?,
+        ))
+    }
+}
+
+impl<One, Two, Three, Four> SegmentCollector for (One, Two, Three, Four)
+where
+    One: SegmentCollector,
+    Two: SegmentCollector,
+    Three: SegmentCollector,
+    Four: SegmentCollector,
+{
+    type Fruit = (One::Fruit, Two::Fruit, Three::Fruit, Four::Fruit);
+
+    fn collect(&mut self, doc: DocId, score: Score) {
+        self.0.collect(doc, score);
+        self.1.collect(doc, score);
+        self.2.collect(doc, score);
+        self.3.collect(doc, score);
+    }
+
+    fn harvest(self) -> <Self as SegmentCollector>::Fruit {
+        (
+            self.0.harvest(),
+            self.1.harvest(),
+            self.2.harvest(),
+            self.3.harvest(),
+        )
+    }
+}
+
+impl_downcast!(Fruit);
+
 #[cfg(test)]
-pub mod tests {
-
-    use super::*;
-    use core::SegmentReader;
-    use fastfield::BytesFastFieldReader;
-    use fastfield::FastFieldReader;
-    use schema::Field;
-    use DocId;
-    use Score;
-    use SegmentLocalId;
-
-    /// Stores all of the doc ids.
-    /// This collector is only used for tests.
-    /// It is unusable in practise, as it does not store
-    /// the segment ordinals
-    pub struct TestCollector {
-        offset: DocId,
-        segment_max_doc: DocId,
-        docs: Vec<DocId>,
-        scores: Vec<Score>,
-    }
-
-    impl TestCollector {
-        /// Return the exhalist of documents.
-        pub fn docs(self) -> Vec<DocId> {
-            self.docs
-        }
-
-        pub fn scores(self) -> Vec<Score> {
-            self.scores
-        }
-    }
-
-    impl Default for TestCollector {
-        fn default() -> TestCollector {
-            TestCollector {
-                offset: 0,
-                segment_max_doc: 0,
-                docs: Vec::new(),
-                scores: Vec::new(),
-            }
-        }
-    }
-
-    impl Collector for TestCollector {
-        fn set_segment(&mut self, _: SegmentLocalId, reader: &SegmentReader) -> Result<()> {
-            self.offset += self.segment_max_doc;
-            self.segment_max_doc = reader.max_doc();
-            Ok(())
-        }
-
-        fn collect(&mut self, doc: DocId, score: Score) {
-            self.docs.push(doc + self.offset);
-            self.scores.push(score);
-        }
-
-        fn requires_scoring(&self) -> bool {
-            true
-        }
-    }
-
-    /// Collects in order all of the fast fields for all of the
-    /// doc in the `DocSet`
-    ///
-    /// This collector is mainly useful for tests.
-    pub struct FastFieldTestCollector {
-        vals: Vec<u64>,
-        field: Field,
-        ff_reader: Option<FastFieldReader<u64>>,
-    }
-
-    impl FastFieldTestCollector {
-        pub fn for_field(field: Field) -> FastFieldTestCollector {
-            FastFieldTestCollector {
-                vals: Vec::new(),
-                field,
-                ff_reader: None,
-            }
-        }
-
-        pub fn vals(self) -> Vec<u64> {
-            self.vals
-        }
-    }
-
-    impl Collector for FastFieldTestCollector {
-        fn set_segment(&mut self, _: SegmentLocalId, reader: &SegmentReader) -> Result<()> {
-            self.ff_reader = Some(reader.fast_field_reader(self.field)?);
-            Ok(())
-        }
-
-        fn collect(&mut self, doc: DocId, _score: Score) {
-            let val = self.ff_reader.as_ref().unwrap().get(doc);
-            self.vals.push(val);
-        }
-        fn requires_scoring(&self) -> bool {
-            false
-        }
-    }
-
-    /// Collects in order all of the fast field bytes for all of the
-    /// docs in the `DocSet`
-    ///
-    /// This collector is mainly useful for tests.
-    pub struct BytesFastFieldTestCollector {
-        vals: Vec<u8>,
-        field: Field,
-        ff_reader: Option<BytesFastFieldReader>,
-    }
-
-    impl BytesFastFieldTestCollector {
-        pub fn for_field(field: Field) -> BytesFastFieldTestCollector {
-            BytesFastFieldTestCollector {
-                vals: Vec::new(),
-                field,
-                ff_reader: None,
-            }
-        }
-
-        pub fn vals(self) -> Vec<u8> {
-            self.vals
-        }
-    }
-
-    impl Collector for BytesFastFieldTestCollector {
-        fn set_segment(&mut self, _segment_local_id: u32, segment: &SegmentReader) -> Result<()> {
-            self.ff_reader = Some(segment.bytes_fast_field_reader(self.field)?);
-            Ok(())
-        }
-
-        fn collect(&mut self, doc: u32, _score: f32) {
-            let val = self.ff_reader.as_ref().unwrap().get_val(doc);
-            self.vals.extend(val);
-        }
-
-        fn requires_scoring(&self) -> bool {
-            false
-        }
-    }
-}
-
-#[cfg(all(test, feature = "unstable"))]
-mod bench {
-    use collector::{Collector, CountCollector};
-    use test::Bencher;
-
-    #[bench]
-    fn build_collector(b: &mut Bencher) {
-        b.iter(|| {
-            let mut count_collector = CountCollector::default();
-            let docs: Vec<u32> = (0..1_000_000).collect();
-            for doc in docs {
-                count_collector.collect(doc, 1f32);
-            }
-            count_collector.count()
-        });
-    }
-}
+pub mod tests;

src/collector/multi_collector.rs

@@ -1,98 +1,242 @@
 use super::Collector;
-use DocId;
-use Result;
-use Score;
-use SegmentLocalId;
-use SegmentReader;
+use super::SegmentCollector;
+use crate::collector::Fruit;
+use crate::DocId;
+use crate::Score;
+use crate::SegmentOrdinal;
+use crate::SegmentReader;
+use crate::TantivyError;
+use std::marker::PhantomData;
+use std::ops::Deref;
+
+pub struct MultiFruit {
+    sub_fruits: Vec<Option<Box<dyn Fruit>>>,
+}
+
+pub struct CollectorWrapper<TCollector: Collector>(TCollector);
+
+impl<TCollector: Collector> Collector for CollectorWrapper<TCollector> {
+    type Fruit = Box<dyn Fruit>;
+    type Child = Box<dyn BoxableSegmentCollector>;
+
+    fn for_segment(
+        &self,
+        segment_local_id: u32,
+        reader: &SegmentReader,
+    ) -> crate::Result<Box<dyn BoxableSegmentCollector>> {
+        let child = self.0.for_segment(segment_local_id, reader)?;
+        Ok(Box::new(SegmentCollectorWrapper(child)))
+    }
+
+    fn requires_scoring(&self) -> bool {
+        self.0.requires_scoring()
+    }
+
+    fn merge_fruits(
+        &self,
+        children: Vec<<Self::Child as SegmentCollector>::Fruit>,
+    ) -> crate::Result<Box<dyn Fruit>> {
+        let typed_fruit: Vec<<TCollector::Child as SegmentCollector>::Fruit> = children
+            .into_iter()
+            .map(|untyped_fruit| {
+                untyped_fruit
+                    .downcast::<<TCollector::Child as SegmentCollector>::Fruit>()
+                    .map(|boxed_but_typed| *boxed_but_typed)
+                    .map_err(|_| {
+                        TantivyError::InvalidArgument("Failed to cast child fruit.".to_string())
+                    })
+            })
+            .collect::<crate::Result<_>>()?;
+        let merged_fruit = self.0.merge_fruits(typed_fruit)?;
+        Ok(Box::new(merged_fruit))
+    }
+}
+
+impl SegmentCollector for Box<dyn BoxableSegmentCollector> {
+    type Fruit = Box<dyn Fruit>;
+
+    fn collect(&mut self, doc: u32, score: Score) {
+        self.as_mut().collect(doc, score);
+    }
+
+    fn harvest(self) -> Box<dyn Fruit> {
+        BoxableSegmentCollector::harvest_from_box(self)
+    }
+}
+
+pub trait BoxableSegmentCollector {
+    fn collect(&mut self, doc: u32, score: Score);
+    fn harvest_from_box(self: Box<Self>) -> Box<dyn Fruit>;
+}
+
+pub struct SegmentCollectorWrapper<TSegmentCollector: SegmentCollector>(TSegmentCollector);
+
+impl<TSegmentCollector: SegmentCollector> BoxableSegmentCollector
+    for SegmentCollectorWrapper<TSegmentCollector>
+{
+    fn collect(&mut self, doc: u32, score: Score) {
+        self.0.collect(doc, score);
+    }
+
+    fn harvest_from_box(self: Box<Self>) -> Box<dyn Fruit> {
+        Box::new(self.0.harvest())
+    }
+}
+
+pub struct FruitHandle<TFruit: Fruit> {
+    pos: usize,
+    _phantom: PhantomData<TFruit>,
+}
+
+impl<TFruit: Fruit> FruitHandle<TFruit> {
+    pub fn extract(self, fruits: &mut MultiFruit) -> TFruit {
+        let boxed_fruit = fruits.sub_fruits[self.pos].take().expect("");
+        *boxed_fruit
+            .downcast::<TFruit>()
+            .map_err(|_| ())
+            .expect("Failed to downcast collector fruit.")
+    }
+}
 
 /// Multicollector makes it possible to collect on more than one collector.
 /// It should only be used for use cases where the Collector types is unknown
 /// at compile time.
-/// If the type of the collectors is known, you should prefer to use `ChainedCollector`.
+///
+/// If the type of the collectors is known, you can just group yours collectors
+/// in a tuple. See the
+/// [Combining several collectors section of the collector documentation](./index.html#combining-several-collectors).
 ///
 /// ```rust
-/// #[macro_use]
-/// extern crate tantivy;
-/// use tantivy::schema::{SchemaBuilder, TEXT};
-/// use tantivy::{Index, Result};
-/// use tantivy::collector::{CountCollector, TopCollector, MultiCollector};
+/// use tantivy::collector::{Count, TopDocs, MultiCollector};
 /// use tantivy::query::QueryParser;
+/// use tantivy::schema::{Schema, TEXT};
+/// use tantivy::{doc, Index};
 ///
-/// # fn main() { example().unwrap(); }
-/// fn example() -> Result<()> {
-///     let mut schema_builder = SchemaBuilder::new();
-///     let title = schema_builder.add_text_field("title", TEXT);
-///     let schema = schema_builder.build();
-///     let index = Index::create_in_ram(schema);
-///     {
-///         let mut index_writer = index.writer(3_000_000)?;
-///         index_writer.add_document(doc!(
-///             title => "The Name of the Wind",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of Muadib",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "A Dairy Cow",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of a Young Girl",
-///         ));
-///         index_writer.commit().unwrap();
-///     }
+/// let mut schema_builder = Schema::builder();
+/// let title = schema_builder.add_text_field("title", TEXT);
+/// let schema = schema_builder.build();
+/// let index = Index::create_in_ram(schema);
 ///
-///     index.load_searchers()?;
-///     let searcher = index.searcher();
+/// let mut index_writer = index.writer(3_000_000).unwrap();
+/// index_writer.add_document(doc!(title => "The Name of the Wind"));
+/// index_writer.add_document(doc!(title => "The Diary of Muadib"));
+/// index_writer.add_document(doc!(title => "A Dairy Cow"));
+/// index_writer.add_document(doc!(title => "The Diary of a Young Girl"));
+/// assert!(index_writer.commit().is_ok());
 ///
-///     {
-///         let mut top_collector = TopCollector::with_limit(2);
-///         let mut count_collector = CountCollector::default();
-///         {
-///             let mut collectors =
-///                 MultiCollector::from(vec![&mut top_collector, &mut count_collector]);
-///             let query_parser = QueryParser::for_index(&index, vec![title]);
-///             let query = query_parser.parse_query("diary")?;
-///             searcher.search(&*query, &mut collectors).unwrap();
-///         }
-///         assert_eq!(count_collector.count(), 2);
-///         assert!(top_collector.at_capacity());
-///     }
+/// let reader = index.reader().unwrap();
+/// let searcher = reader.searcher();
 ///
-///     Ok(())
-/// }
+/// let mut collectors = MultiCollector::new();
+/// let top_docs_handle = collectors.add_collector(TopDocs::with_limit(2));
+/// let count_handle = collectors.add_collector(Count);
+/// let query_parser = QueryParser::for_index(&index, vec![title]);
+/// let query = query_parser.parse_query("diary").unwrap();
+/// let mut multi_fruit = searcher.search(&query, &collectors).unwrap();
+///
+/// let count = count_handle.extract(&mut multi_fruit);
+/// let top_docs = top_docs_handle.extract(&mut multi_fruit);
+///
+/// assert_eq!(count, 2);
+/// assert_eq!(top_docs.len(), 2);
 /// ```
+#[allow(clippy::type_complexity)]
+#[derive(Default)]
 pub struct MultiCollector<'a> {
-    collectors: Vec<&'a mut Collector>,
+    collector_wrappers: Vec<
+        Box<dyn Collector<Child = Box<dyn BoxableSegmentCollector>, Fruit = Box<dyn Fruit>> + 'a>,
+    >,
 }
 
 impl<'a> MultiCollector<'a> {
-    /// Constructor
-    pub fn from(collectors: Vec<&'a mut Collector>) -> MultiCollector {
-        MultiCollector { collectors }
+    /// Create a new `MultiCollector`
+    pub fn new() -> Self {
+        Default::default()
+    }
+
+    /// Add a new collector to our `MultiCollector`.
+    pub fn add_collector<'b: 'a, TCollector: Collector + 'b>(
+        &mut self,
+        collector: TCollector,
+    ) -> FruitHandle<TCollector::Fruit> {
+        let pos = self.collector_wrappers.len();
+        self.collector_wrappers
+            .push(Box::new(CollectorWrapper(collector)));
+        FruitHandle {
+            pos,
+            _phantom: PhantomData,
+        }
     }
 }
 
 impl<'a> Collector for MultiCollector<'a> {
-    fn set_segment(
-        &mut self,
-        segment_local_id: SegmentLocalId,
+    type Fruit = MultiFruit;
+    type Child = MultiCollectorChild;
+
+    fn for_segment(
+        &self,
+        segment_local_id: SegmentOrdinal,
         segment: &SegmentReader,
-    ) -> Result<()> {
-        for collector in &mut self.collectors {
-            collector.set_segment(segment_local_id, segment)?;
-        }
-        Ok(())
+    ) -> crate::Result<MultiCollectorChild> {
+        let children = self
+            .collector_wrappers
+            .iter()
+            .map(|collector_wrapper| collector_wrapper.for_segment(segment_local_id, segment))
+            .collect::<crate::Result<Vec<_>>>()?;
+        Ok(MultiCollectorChild { children })
     }
 
+    fn requires_scoring(&self) -> bool {
+        self.collector_wrappers
+            .iter()
+            .map(Deref::deref)
+            .any(Collector::requires_scoring)
+    }
+
+    fn merge_fruits(&self, segments_multifruits: Vec<MultiFruit>) -> crate::Result<MultiFruit> {
+        let mut segment_fruits_list: Vec<Vec<Box<dyn Fruit>>> = (0..self.collector_wrappers.len())
+            .map(|_| Vec::with_capacity(segments_multifruits.len()))
+            .collect::<Vec<_>>();
+        for segment_multifruit in segments_multifruits {
+            for (idx, segment_fruit_opt) in segment_multifruit.sub_fruits.into_iter().enumerate() {
+                if let Some(segment_fruit) = segment_fruit_opt {
+                    segment_fruits_list[idx].push(segment_fruit);
+                }
+            }
+        }
+        let sub_fruits = self
+            .collector_wrappers
+            .iter()
+            .zip(segment_fruits_list)
+            .map(|(child_collector, segment_fruits)| {
+                Ok(Some(child_collector.merge_fruits(segment_fruits)?))
+            })
+            .collect::<crate::Result<_>>()?;
+        Ok(MultiFruit { sub_fruits })
+    }
+}
+
+pub struct MultiCollectorChild {
+    children: Vec<Box<dyn BoxableSegmentCollector>>,
+}
+
+impl SegmentCollector for MultiCollectorChild {
+    type Fruit = MultiFruit;
+
     fn collect(&mut self, doc: DocId, score: Score) {
-        for collector in &mut self.collectors {
-            collector.collect(doc, score);
+        for child in &mut self.children {
+            child.collect(doc, score);
         }
     }
-    fn requires_scoring(&self) -> bool {
-        self.collectors
-            .iter()
-            .any(|collector| collector.requires_scoring())
+
+    fn harvest(self) -> MultiFruit {
+        MultiFruit {
+            sub_fruits: self
+                .children
+                .into_iter()
+                .map(|child| Some(child.harvest()))
+                .collect(),
+        }
     }
 }
 
@@ -100,20 +244,41 @@ impl<'a> Collector for MultiCollector<'a> {
 mod tests {
 
     use super::*;
-    use collector::{Collector, CountCollector, TopScoreCollector};
+    use crate::collector::{Count, TopDocs};
+    use crate::query::TermQuery;
+    use crate::schema::IndexRecordOption;
+    use crate::schema::{Schema, TEXT};
+    use crate::Index;
+    use crate::Term;
 
     #[test]
     fn test_multi_collector() {
-        let mut top_collector = TopScoreCollector::with_limit(2);
-        let mut count_collector = CountCollector::default();
+        let mut schema_builder = Schema::builder();
+        let text = schema_builder.add_text_field("text", TEXT);
+        let schema = schema_builder.build();
+
+        let index = Index::create_in_ram(schema);
         {
-            let mut collectors =
-                MultiCollector::from(vec![&mut top_collector, &mut count_collector]);
-            collectors.collect(1, 0.2);
-            collectors.collect(2, 0.1);
-            collectors.collect(3, 0.5);
+            let mut index_writer = index.writer_for_tests().unwrap();
+            index_writer.add_document(doc!(text=>"abc"));
+            index_writer.add_document(doc!(text=>"abc abc abc"));
+            index_writer.add_document(doc!(text=>"abc abc"));
+            index_writer.commit().unwrap();
+            index_writer.add_document(doc!(text=>""));
+            index_writer.add_document(doc!(text=>"abc abc abc abc"));
+            index_writer.add_document(doc!(text=>"abc"));
+            index_writer.commit().unwrap();
         }
-        assert_eq!(count_collector.count(), 3);
-        assert!(top_collector.at_capacity());
+        let searcher = index.reader().unwrap().searcher();
+        let term = Term::from_field_text(text, "abc");
+        let query = TermQuery::new(term, IndexRecordOption::Basic);
+
+        let mut collectors = MultiCollector::new();
+        let topdocs_handler = collectors.add_collector(TopDocs::with_limit(2));
+        let count_handler = collectors.add_collector(Count);
+        let mut multifruits = searcher.search(&query, &collectors).unwrap();
+
+        assert_eq!(count_handler.extract(&mut multifruits), 5);
+        assert_eq!(topdocs_handler.extract(&mut multifruits).len(), 2);
     }
 }

src/collector/tests.rs

@@ -0,0 +1,298 @@
+use super::*;
+use crate::core::SegmentReader;
+use crate::fastfield::BytesFastFieldReader;
+use crate::fastfield::DynamicFastFieldReader;
+use crate::fastfield::FastFieldReader;
+use crate::schema::Field;
+use crate::DocId;
+use crate::Score;
+use crate::SegmentOrdinal;
+use crate::{DocAddress, Document, Searcher};
+
+use crate::collector::{Count, FilterCollector, TopDocs};
+use crate::query::{AllQuery, QueryParser};
+use crate::schema::{Schema, FAST, TEXT};
+use crate::DateTime;
+use crate::{doc, Index};
+use std::str::FromStr;
+
+pub const TEST_COLLECTOR_WITH_SCORE: TestCollector = TestCollector {
+    compute_score: true,
+};
+
+pub const TEST_COLLECTOR_WITHOUT_SCORE: TestCollector = TestCollector {
+    compute_score: true,
+};
+
+#[test]
+pub fn test_filter_collector() {
+    let mut schema_builder = Schema::builder();
+    let title = schema_builder.add_text_field("title", TEXT);
+    let price = schema_builder.add_u64_field("price", FAST);
+    let date = schema_builder.add_date_field("date", FAST);
+    let schema = schema_builder.build();
+    let index = Index::create_in_ram(schema);
+
+    let mut index_writer = index.writer_with_num_threads(1, 10_000_000).unwrap();
+    index_writer.add_document(doc!(title => "The Name of the Wind", price => 30_200u64, date => DateTime::from_str("1898-04-09T00:00:00+00:00").unwrap()));
+    index_writer.add_document(doc!(title => "The Diary of Muadib", price => 29_240u64, date => DateTime::from_str("2020-04-09T00:00:00+00:00").unwrap()));
+    index_writer.add_document(doc!(title => "The Diary of Anne Frank", price => 18_240u64, date => DateTime::from_str("2019-04-20T00:00:00+00:00").unwrap()));
+    index_writer.add_document(doc!(title => "A Dairy Cow", price => 21_240u64, date => DateTime::from_str("2019-04-09T00:00:00+00:00").unwrap()));
+    index_writer.add_document(doc!(title => "The Diary of a Young Girl", price => 20_120u64, date => DateTime::from_str("2018-04-09T00:00:00+00:00").unwrap()));
+    assert!(index_writer.commit().is_ok());
+
+    let reader = index.reader().unwrap();
+    let searcher = reader.searcher();
+
+    let query_parser = QueryParser::for_index(&index, vec![title]);
+    let query = query_parser.parse_query("diary").unwrap();
+    let filter_some_collector = FilterCollector::new(
+        price,
+        &|value: u64| value > 20_120u64,
+        TopDocs::with_limit(2),
+    );
+    let top_docs = searcher.search(&query, &filter_some_collector).unwrap();
+
+    assert_eq!(top_docs.len(), 1);
+    assert_eq!(top_docs[0].1, DocAddress::new(0, 1));
+
+    let filter_all_collector: FilterCollector<_, _, u64> =
+        FilterCollector::new(price, &|value| value < 5u64, TopDocs::with_limit(2));
+    let filtered_top_docs = searcher.search(&query, &filter_all_collector).unwrap();
+
+    assert_eq!(filtered_top_docs.len(), 0);
+
+    fn date_filter(value: DateTime) -> bool {
+        (value - DateTime::from_str("2019-04-09T00:00:00+00:00").unwrap()).num_weeks() > 0
+    }
+
+    let filter_dates_collector = FilterCollector::new(date, &date_filter, TopDocs::with_limit(5));
+    let filtered_date_docs = searcher.search(&query, &filter_dates_collector).unwrap();
+
+    assert_eq!(filtered_date_docs.len(), 2);
+}
+
+/// Stores all of the doc ids.
+/// This collector is only used for tests.
+/// It is unusable in pr
+///
+/// actise, as it does not store
+/// the segment ordinals
+pub struct TestCollector {
+    pub compute_score: bool,
+}
+
+pub struct TestSegmentCollector {
+    segment_id: SegmentOrdinal,
+    fruit: TestFruit,
+}
+
+#[derive(Default)]
+pub struct TestFruit {
+    docs: Vec<DocAddress>,
+    scores: Vec<Score>,
+}
+
+impl TestFruit {
+    /// Return the list of matching documents exhaustively.
+    pub fn docs(&self) -> &[DocAddress] {
+        &self.docs[..]
+    }
+    pub fn scores(&self) -> &[Score] {
+        &self.scores[..]
+    }
+}
+
+impl Collector for TestCollector {
+    type Fruit = TestFruit;
+    type Child = TestSegmentCollector;
+
+    fn for_segment(
+        &self,
+        segment_id: SegmentOrdinal,
+        _reader: &SegmentReader,
+    ) -> crate::Result<TestSegmentCollector> {
+        Ok(TestSegmentCollector {
+            segment_id,
+            fruit: TestFruit::default(),
+        })
+    }
+
+    fn requires_scoring(&self) -> bool {
+        self.compute_score
+    }
+
+    fn merge_fruits(&self, mut children: Vec<TestFruit>) -> crate::Result<TestFruit> {
+        children.sort_by_key(|fruit| {
+            if fruit.docs().is_empty() {
+                0
+            } else {
+                fruit.docs()[0].segment_ord
+            }
+        });
+        let mut docs = vec![];
+        let mut scores = vec![];
+        for child in children {
+            docs.extend(child.docs());
+            scores.extend(child.scores);
+        }
+        Ok(TestFruit { docs, scores })
+    }
+}
+
+impl SegmentCollector for TestSegmentCollector {
+    type Fruit = TestFruit;
+
+    fn collect(&mut self, doc: DocId, score: Score) {
+        self.fruit.docs.push(DocAddress::new(self.segment_id, doc));
+        self.fruit.scores.push(score);
+    }
+
+    fn harvest(self) -> <Self as SegmentCollector>::Fruit {
+        self.fruit
+    }
+}
+
+/// Collects in order all of the fast fields for all of the
+/// doc in the `DocSet`
+///
+/// This collector is mainly useful for tests.
+pub struct FastFieldTestCollector {
+    field: Field,
+}
+
+pub struct FastFieldSegmentCollector {
+    vals: Vec<u64>,
+    reader: DynamicFastFieldReader<u64>,
+}
+
+impl FastFieldTestCollector {
+    pub fn for_field(field: Field) -> FastFieldTestCollector {
+        FastFieldTestCollector { field }
+    }
+}
+
+impl Collector for FastFieldTestCollector {
+    type Fruit = Vec<u64>;
+    type Child = FastFieldSegmentCollector;
+
+    fn for_segment(
+        &self,
+        _: SegmentOrdinal,
+        segment_reader: &SegmentReader,
+    ) -> crate::Result<FastFieldSegmentCollector> {
+        let reader = segment_reader
+            .fast_fields()
+            .u64(self.field)
+            .expect("Requested field is not a fast field.");
+        Ok(FastFieldSegmentCollector {
+            vals: Vec::new(),
+            reader,
+        })
+    }
+
+    fn requires_scoring(&self) -> bool {
+        false
+    }
+
+    fn merge_fruits(&self, children: Vec<Vec<u64>>) -> crate::Result<Vec<u64>> {
+        Ok(children.into_iter().flat_map(|v| v.into_iter()).collect())
+    }
+}
+
+impl SegmentCollector for FastFieldSegmentCollector {
+    type Fruit = Vec<u64>;
+
+    fn collect(&mut self, doc: DocId, _score: Score) {
+        let val = self.reader.get(doc);
+        self.vals.push(val);
+    }
+
+    fn harvest(self) -> Vec<u64> {
+        self.vals
+    }
+}
+
+/// Collects in order all of the fast field bytes for all of the
+/// docs in the `DocSet`
+///
+/// This collector is mainly useful for tests.
+pub struct BytesFastFieldTestCollector {
+    field: Field,
+}
+
+pub struct BytesFastFieldSegmentCollector {
+    vals: Vec<u8>,
+    reader: BytesFastFieldReader,
+}
+
+impl BytesFastFieldTestCollector {
+    pub fn for_field(field: Field) -> BytesFastFieldTestCollector {
+        BytesFastFieldTestCollector { field }
+    }
+}
+
+impl Collector for BytesFastFieldTestCollector {
+    type Fruit = Vec<u8>;
+    type Child = BytesFastFieldSegmentCollector;
+
+    fn for_segment(
+        &self,
+        _segment_local_id: u32,
+        segment_reader: &SegmentReader,
+    ) -> crate::Result<BytesFastFieldSegmentCollector> {
+        let reader = segment_reader.fast_fields().bytes(self.field)?;
+        Ok(BytesFastFieldSegmentCollector {
+            vals: Vec::new(),
+            reader,
+        })
+    }
+
+    fn requires_scoring(&self) -> bool {
+        false
+    }
+
+    fn merge_fruits(&self, children: Vec<Vec<u8>>) -> crate::Result<Vec<u8>> {
+        Ok(children.into_iter().flat_map(|c| c.into_iter()).collect())
+    }
+}
+
+impl SegmentCollector for BytesFastFieldSegmentCollector {
+    type Fruit = Vec<u8>;
+
+    fn collect(&mut self, doc: u32, _score: Score) {
+        let data = self.reader.get_bytes(doc);
+        self.vals.extend(data);
+    }
+
+    fn harvest(self) -> <Self as SegmentCollector>::Fruit {
+        self.vals
+    }
+}
+
+fn make_test_searcher() -> crate::Result<crate::LeasedItem<Searcher>> {
+    let schema = Schema::builder().build();
+    let index = Index::create_in_ram(schema);
+    let mut index_writer = index.writer_for_tests()?;
+    index_writer.add_document(Document::default());
+    index_writer.add_document(Document::default());
+    index_writer.commit()?;
+    Ok(index.reader()?.searcher())
+}
+
+#[test]
+fn test_option_collector_some() -> crate::Result<()> {
+    let searcher = make_test_searcher()?;
+    let counts = searcher.search(&AllQuery, &Some(Count))?;
+    assert_eq!(counts, Some(2));
+    Ok(())
+}
+
+#[test]
+fn test_option_collector_none() -> crate::Result<()> {
+    let searcher = make_test_searcher()?;
+    let none_collector: Option<Count> = None;
+    let counts = searcher.search(&AllQuery, &none_collector)?;
+    assert_eq!(counts, None);
+    Ok(())
+}

src/collector/top_collector.rs

@@ -1,56 +1,70 @@
+use crate::DocAddress;
+use crate::DocId;
+use crate::SegmentOrdinal;
+use crate::SegmentReader;
 use std::cmp::Ordering;
 use std::collections::BinaryHeap;
-use DocAddress;
-use DocId;
-use SegmentLocalId;
+use std::marker::PhantomData;
 
 /// Contains a feature (field, score, etc.) of a document along with the document address.
 ///
 /// It has a custom implementation of `PartialOrd` that reverses the order. This is because the
 /// default Rust heap is a max heap, whereas a min heap is needed.
-#[derive(Clone, Copy)]
-pub struct ComparableDoc<T> {
-    feature: T,
-    doc_address: DocAddress,
+///
+/// Additionally, it guarantees stable sorting: in case of a tie on the feature, the document
+/// address is used.
+///
+/// WARNING: equality is not what you would expect here.
+/// Two elements are equal if their feature is equal, and regardless of whether `doc`
+/// is equal. This should be perfectly fine for this usage, but let's make sure this
+/// struct is never public.
+pub(crate) struct ComparableDoc<T, D> {
+    pub feature: T,
+    pub doc: D,
 }
 
-impl<T: PartialOrd> PartialOrd for ComparableDoc<T> {
+impl<T: PartialOrd, D: PartialOrd> PartialOrd for ComparableDoc<T, D> {
     fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
         Some(self.cmp(other))
     }
 }
 
-impl<T: PartialOrd> Ord for ComparableDoc<T> {
+impl<T: PartialOrd, D: PartialOrd> Ord for ComparableDoc<T, D> {
     #[inline]
     fn cmp(&self, other: &Self) -> Ordering {
-        other
+        // Reversed to make BinaryHeap work as a min-heap
+        let by_feature = other
             .feature
             .partial_cmp(&self.feature)
-            .unwrap_or_else(|| other.doc_address.cmp(&self.doc_address))
+            .unwrap_or(Ordering::Equal);
+
+        let lazy_by_doc_address = || self.doc.partial_cmp(&other.doc).unwrap_or(Ordering::Equal);
+
+        // In case of a tie on the feature, we sort by ascending
+        // `DocAddress` in order to ensure a stable sorting of the
+        // documents.
+        by_feature.then_with(lazy_by_doc_address)
     }
 }
 
-impl<T: PartialOrd> PartialEq for ComparableDoc<T> {
+impl<T: PartialOrd, D: PartialOrd> PartialEq for ComparableDoc<T, D> {
     fn eq(&self, other: &Self) -> bool {
         self.cmp(other) == Ordering::Equal
     }
 }
 
-impl<T: PartialOrd> Eq for ComparableDoc<T> {}
+impl<T: PartialOrd, D: PartialOrd> Eq for ComparableDoc<T, D> {}
 
-/// The Top Collector keeps track of the K documents
-/// sorted by type `T`.
-///
-/// The implementation is based on a `BinaryHeap`.
-/// The theorical complexity for collecting the top `K` out of `n` documents
-/// is `O(n log K)`.
-pub struct TopCollector<T> {
-    limit: usize,
-    heap: BinaryHeap<ComparableDoc<T>>,
-    segment_id: u32,
+pub(crate) struct TopCollector<T> {
+    pub limit: usize,
+    pub offset: usize,
+    _marker: PhantomData<T>,
 }
 
-impl<T: PartialOrd + Clone> TopCollector<T> {
+impl<T> TopCollector<T>
+where
+    T: PartialOrd + Clone,
+{
     /// Creates a top collector, with a number of documents equal to "limit".
     ///
     /// # Panics
@@ -59,135 +73,317 @@ impl<T: PartialOrd + Clone> TopCollector<T> {
         if limit < 1 {
             panic!("Limit must be strictly greater than 0.");
         }
-        TopCollector {
+        Self {
             limit,
-            heap: BinaryHeap::with_capacity(limit),
-            segment_id: 0,
+            offset: 0,
+            _marker: PhantomData,
         }
     }
 
-    /// Returns K best documents sorted in decreasing order.
+    /// Skip the first "offset" documents when collecting.
     ///
-    /// Calling this method triggers the sort.
-    /// The result of the sort is not cached.
-    pub fn docs(&self) -> Vec<DocAddress> {
-        self.top_docs()
-            .into_iter()
-            .map(|(_feature, doc)| doc)
-            .collect()
+    /// This is equivalent to `OFFSET` in MySQL or PostgreSQL and `start` in
+    /// Lucene's TopDocsCollector.
+    pub fn and_offset(mut self, offset: usize) -> TopCollector<T> {
+        self.offset = offset;
+        self
     }
 
-    /// Returns K best FeatureDocuments sorted in decreasing order.
-    ///
-    /// Calling this method triggers the sort.
-    /// The result of the sort is not cached.
-    pub fn top_docs(&self) -> Vec<(T, DocAddress)> {
-        let mut feature_docs: Vec<ComparableDoc<T>> = self.heap.iter().cloned().collect();
-        feature_docs.sort();
-        feature_docs
+    pub fn merge_fruits(
+        &self,
+        children: Vec<Vec<(T, DocAddress)>>,
+    ) -> crate::Result<Vec<(T, DocAddress)>> {
+        if self.limit == 0 {
+            return Ok(Vec::new());
+        }
+        let mut top_collector = BinaryHeap::new();
+        for child_fruit in children {
+            for (feature, doc) in child_fruit {
+                if top_collector.len() < (self.limit + self.offset) {
+                    top_collector.push(ComparableDoc { feature, doc });
+                } else if let Some(mut head) = top_collector.peek_mut() {
+                    if head.feature < feature {
+                        *head = ComparableDoc { feature, doc };
+                    }
+                }
+            }
+        }
+        Ok(top_collector
+            .into_sorted_vec()
             .into_iter()
-            .map(
-                |ComparableDoc {
-                     feature,
-                     doc_address,
-                 }| (feature, doc_address),
-            ).collect()
+            .skip(self.offset)
+            .map(|cdoc| (cdoc.feature, cdoc.doc))
+            .collect())
+    }
+
+    pub(crate) fn for_segment<F: PartialOrd>(
+        &self,
+        segment_id: SegmentOrdinal,
+        _: &SegmentReader,
+    ) -> TopSegmentCollector<F> {
+        TopSegmentCollector::new(segment_id, self.limit + self.offset)
+    }
+
+    /// Create a new TopCollector with the same limit and offset.
+    ///
+    /// Ideally we would use Into but the blanket implementation seems to cause the Scorer traits
+    /// to fail.
+    #[doc(hidden)]
+    pub(crate) fn into_tscore<TScore: PartialOrd + Clone>(self) -> TopCollector<TScore> {
+        TopCollector {
+            limit: self.limit,
+            offset: self.offset,
+            _marker: PhantomData,
+        }
+    }
+}
+
+/// The Top Collector keeps track of the K documents
+/// sorted by type `T`.
+///
+/// The implementation is based on a `BinaryHeap`.
+/// The theorical complexity for collecting the top `K` out of `n` documents
+/// is `O(n log K)`.
+pub(crate) struct TopSegmentCollector<T> {
+    limit: usize,
+    heap: BinaryHeap<ComparableDoc<T, DocId>>,
+    segment_ord: u32,
+}
+
+impl<T: PartialOrd> TopSegmentCollector<T> {
+    fn new(segment_ord: SegmentOrdinal, limit: usize) -> TopSegmentCollector<T> {
+        TopSegmentCollector {
+            limit,
+            heap: BinaryHeap::with_capacity(limit),
+            segment_ord,
+        }
+    }
+}
+
+impl<T: PartialOrd + Clone> TopSegmentCollector<T> {
+    pub fn harvest(self) -> Vec<(T, DocAddress)> {
+        let segment_ord = self.segment_ord;
+        self.heap
+            .into_sorted_vec()
+            .into_iter()
+            .map(|comparable_doc| {
+                (
+                    comparable_doc.feature,
+                    DocAddress {
+                        segment_ord,
+                        doc_id: comparable_doc.doc,
+                    },
+                )
+            })
+            .collect()
     }
 
     /// Return true iff at least K documents have gone through
     /// the collector.
     #[inline]
-    pub fn at_capacity(&self) -> bool {
+    pub(crate) fn at_capacity(&self) -> bool {
         self.heap.len() >= self.limit
     }
 
-    /// Sets the segment local ID for the collector
-    pub fn set_segment_id(&mut self, segment_id: SegmentLocalId) {
-        self.segment_id = segment_id;
-    }
-
     /// Collects a document scored by the given feature
     ///
     /// It collects documents until it has reached the max capacity. Once it reaches capacity, it
     /// will compare the lowest scoring item with the given one and keep whichever is greater.
+    #[inline]
     pub fn collect(&mut self, doc: DocId, feature: T) {
         if self.at_capacity() {
             // It's ok to unwrap as long as a limit of 0 is forbidden.
-            let limit_doc: ComparableDoc<T> = self
-                .heap
-                .peek()
-                .expect("Top collector with size 0 is forbidden")
-                .clone();
-            if limit_doc.feature < feature {
-                let mut mut_head = self
-                    .heap
-                    .peek_mut()
-                    .expect("Top collector with size 0 is forbidden");
-                mut_head.feature = feature;
-                mut_head.doc_address = DocAddress(self.segment_id, doc);
+            if let Some(limit_feature) = self.heap.peek().map(|head| head.feature.clone()) {
+                if limit_feature < feature {
+                    if let Some(mut head) = self.heap.peek_mut() {
+                        head.feature = feature;
+                        head.doc = doc;
+                    }
+                }
             }
         } else {
-            let wrapped_doc = ComparableDoc {
-                feature,
-                doc_address: DocAddress(self.segment_id, doc),
-            };
-            self.heap.push(wrapped_doc);
+            // we have not reached capacity yet, so we can just push the
+            // element.
+            self.heap.push(ComparableDoc { feature, doc });
         }
     }
 }
 
 #[cfg(test)]
 mod tests {
-    use super::*;
-    use DocId;
-    use Score;
+    use super::{TopCollector, TopSegmentCollector};
+    use crate::DocAddress;
 
     #[test]
     fn test_top_collector_not_at_capacity() {
-        let mut top_collector = TopCollector::with_limit(4);
+        let mut top_collector = TopSegmentCollector::new(0, 4);
         top_collector.collect(1, 0.8);
         top_collector.collect(3, 0.2);
         top_collector.collect(5, 0.3);
-        assert!(!top_collector.at_capacity());
-        let score_docs: Vec<(Score, DocId)> = top_collector
-            .top_docs()
-            .into_iter()
-            .map(|(score, doc_address)| (score, doc_address.doc()))
-            .collect();
-        assert_eq!(score_docs, vec![(0.8, 1), (0.3, 5), (0.2, 3)]);
+        assert_eq!(
+            top_collector.harvest(),
+            vec![
+                (0.8, DocAddress::new(0, 1)),
+                (0.3, DocAddress::new(0, 5)),
+                (0.2, DocAddress::new(0, 3))
+            ]
+        );
     }
 
     #[test]
     fn test_top_collector_at_capacity() {
-        let mut top_collector = TopCollector::with_limit(4);
+        let mut top_collector = TopSegmentCollector::new(0, 4);
         top_collector.collect(1, 0.8);
         top_collector.collect(3, 0.2);
         top_collector.collect(5, 0.3);
         top_collector.collect(7, 0.9);
         top_collector.collect(9, -0.2);
-        assert!(top_collector.at_capacity());
-        {
-            let score_docs: Vec<(Score, DocId)> = top_collector
-                .top_docs()
-                .into_iter()
-                .map(|(score, doc_address)| (score, doc_address.doc()))
-                .collect();
-            assert_eq!(score_docs, vec![(0.9, 7), (0.8, 1), (0.3, 5), (0.2, 3)]);
-        }
-        {
-            let docs: Vec<DocId> = top_collector
-                .docs()
-                .into_iter()
-                .map(|doc_address| doc_address.doc())
-                .collect();
-            assert_eq!(docs, vec![7, 1, 5, 3]);
-        }
+        assert_eq!(
+            top_collector.harvest(),
+            vec![
+                (0.9, DocAddress::new(0, 7)),
+                (0.8, DocAddress::new(0, 1)),
+                (0.3, DocAddress::new(0, 5)),
+                (0.2, DocAddress::new(0, 3))
+            ]
+        );
     }
 
     #[test]
-    #[should_panic]
-    fn test_top_0() {
-        let _collector: TopCollector<Score> = TopCollector::with_limit(0);
+    fn test_top_segment_collector_stable_ordering_for_equal_feature() {
+        // given that the documents are collected in ascending doc id order,
+        // when harvesting we have to guarantee stable sorting in case of a tie
+        // on the score
+        let doc_ids_collection = [4, 5, 6];
+        let score = 3.14;
+
+        let mut top_collector_limit_2 = TopSegmentCollector::new(0, 2);
+        for id in &doc_ids_collection {
+            top_collector_limit_2.collect(*id, score);
+        }
+
+        let mut top_collector_limit_3 = TopSegmentCollector::new(0, 3);
+        for id in &doc_ids_collection {
+            top_collector_limit_3.collect(*id, score);
+        }
+
+        assert_eq!(
+            top_collector_limit_2.harvest(),
+            top_collector_limit_3.harvest()[..2].to_vec(),
+        );
     }
 
+    #[test]
+    fn test_top_collector_with_limit_and_offset() {
+        let collector = TopCollector::with_limit(2).and_offset(1);
+
+        let results = collector
+            .merge_fruits(vec![vec![
+                (0.9, DocAddress::new(0, 1)),
+                (0.8, DocAddress::new(0, 2)),
+                (0.7, DocAddress::new(0, 3)),
+                (0.6, DocAddress::new(0, 4)),
+                (0.5, DocAddress::new(0, 5)),
+            ]])
+            .unwrap();
+
+        assert_eq!(
+            results,
+            vec![(0.8, DocAddress::new(0, 2)), (0.7, DocAddress::new(0, 3)),]
+        );
+    }
+
+    #[test]
+    fn test_top_collector_with_limit_larger_than_set_and_offset() {
+        let collector = TopCollector::with_limit(2).and_offset(1);
+
+        let results = collector
+            .merge_fruits(vec![vec![
+                (0.9, DocAddress::new(0, 1)),
+                (0.8, DocAddress::new(0, 2)),
+            ]])
+            .unwrap();
+
+        assert_eq!(results, vec![(0.8, DocAddress::new(0, 2)),]);
+    }
+
+    #[test]
+    fn test_top_collector_with_limit_and_offset_larger_than_set() {
+        let collector = TopCollector::with_limit(2).and_offset(20);
+
+        let results = collector
+            .merge_fruits(vec![vec![
+                (0.9, DocAddress::new(0, 1)),
+                (0.8, DocAddress::new(0, 2)),
+            ]])
+            .unwrap();
+
+        assert_eq!(results, vec![]);
+    }
+}
+
+#[cfg(all(test, feature = "unstable"))]
+mod bench {
+    use super::TopSegmentCollector;
+    use test::Bencher;
+
+    #[bench]
+    fn bench_top_segment_collector_collect_not_at_capacity(b: &mut Bencher) {
+        let mut top_collector = TopSegmentCollector::new(0, 400);
+
+        b.iter(|| {
+            for i in 0..100 {
+                top_collector.collect(i, 0.8);
+            }
+        });
+    }
+
+    #[bench]
+    fn bench_top_segment_collector_collect_at_capacity(b: &mut Bencher) {
+        let mut top_collector = TopSegmentCollector::new(0, 100);
+
+        for i in 0..100 {
+            top_collector.collect(i, 0.8);
+        }
+
+        b.iter(|| {
+            for i in 0..100 {
+                top_collector.collect(i, 0.8);
+            }
+        });
+    }
+
+    #[bench]
+    fn bench_top_segment_collector_collect_and_harvest_many_ties(b: &mut Bencher) {
+        b.iter(|| {
+            let mut top_collector = TopSegmentCollector::new(0, 100);
+
+            for i in 0..100 {
+                top_collector.collect(i, 0.8);
+            }
+
+            // it would be nice to be able to do the setup N times but still
+            // measure only harvest(). We can't since harvest() consumes
+            // the top_collector.
+            top_collector.harvest()
+        });
+    }
+
+    #[bench]
+    fn bench_top_segment_collector_collect_and_harvest_no_tie(b: &mut Bencher) {
+        b.iter(|| {
+            let mut top_collector = TopSegmentCollector::new(0, 100);
+            let mut score = 1.0;
+
+            for i in 0..100 {
+                score += 1.0;
+                top_collector.collect(i, score);
+            }
+
+            // it would be nice to be able to do the setup N times but still
+            // measure only harvest(). We can't since harvest() consumes
+            // the top_collector.
+            top_collector.harvest()
+        });
+    }
 }

src/collector/top_field_collector.rs

@@ -1,263 +0,0 @@
-use super::Collector;
-use collector::top_collector::TopCollector;
-use fastfield::FastFieldReader;
-use fastfield::FastValue;
-use schema::Field;
-use DocAddress;
-use DocId;
-use Result;
-use Score;
-use SegmentReader;
-
-/// The Top Field Collector keeps track of the K documents
-/// sorted by a fast field in the index
-///
-/// The implementation is based on a `BinaryHeap`.
-/// The theorical complexity for collecting the top `K` out of `n` documents
-/// is `O(n log K)`.
-///
-/// ```rust
-/// #[macro_use]
-/// extern crate tantivy;
-/// use tantivy::schema::{SchemaBuilder, TEXT, FAST};
-/// use tantivy::{Index, Result, DocId};
-/// use tantivy::collector::TopFieldCollector;
-/// use tantivy::query::QueryParser;
-///
-/// # fn main() { example().unwrap(); }
-/// fn example() -> Result<()> {
-///     let mut schema_builder = SchemaBuilder::new();
-///     let title = schema_builder.add_text_field("title", TEXT);
-///     let rating = schema_builder.add_u64_field("rating", FAST);
-///     let schema = schema_builder.build();
-///     let index = Index::create_in_ram(schema);
-///     {
-///         let mut index_writer = index.writer_with_num_threads(1, 3_000_000)?;
-///         index_writer.add_document(doc!(
-///             title => "The Name of the Wind",
-///             rating => 92u64,
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of Muadib",
-///             rating => 97u64,
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "A Dairy Cow",
-///             rating => 63u64,
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of a Young Girl",
-///             rating => 80u64,
-///         ));
-///         index_writer.commit().unwrap();
-///     }
-///
-///     index.load_searchers()?;
-///     let searcher = index.searcher();
-///
-///     {
-///	        let mut top_collector = TopFieldCollector::with_limit(rating, 2);
-///         let query_parser = QueryParser::for_index(&index, vec![title]);
-///         let query = query_parser.parse_query("diary")?;
-///         searcher.search(&*query, &mut top_collector).unwrap();
-///
-///         let score_docs: Vec<(u64, DocId)> = top_collector
-///           .top_docs()
-///           .into_iter()
-///           .map(|(field, doc_address)| (field, doc_address.doc()))
-///           .collect();
-///
-///         assert_eq!(score_docs, vec![(97u64, 1), (80, 3)]);
-///     }
-///
-///     Ok(())
-/// }
-/// ```
-pub struct TopFieldCollector<T: FastValue> {
-    field: Field,
-    collector: TopCollector<T>,
-    fast_field: Option<FastFieldReader<T>>,
-}
-
-impl<T: FastValue + PartialOrd + Clone> TopFieldCollector<T> {
-    /// Creates a top field collector, with a number of documents equal to "limit".
-    ///
-    /// The given field name must be a fast field, otherwise the collector have an error while
-    /// collecting results.
-    ///
-    /// # Panics
-    /// The method panics if limit is 0
-    pub fn with_limit(field: Field, limit: usize) -> Self {
-        TopFieldCollector {
-            field,
-            collector: TopCollector::with_limit(limit),
-            fast_field: None,
-        }
-    }
-
-    /// Returns K best documents sorted the given field name in decreasing order.
-    ///
-    /// Calling this method triggers the sort.
-    /// The result of the sort is not cached.
-    pub fn docs(&self) -> Vec<DocAddress> {
-        self.collector.docs()
-    }
-
-    /// Returns K best FieldDocuments sorted in decreasing order.
-    ///
-    /// Calling this method triggers the sort.
-    /// The result of the sort is not cached.
-    pub fn top_docs(&self) -> Vec<(T, DocAddress)> {
-        self.collector.top_docs()
-    }
-
-    /// Return true iff at least K documents have gone through
-    /// the collector.
-    #[inline]
-    pub fn at_capacity(&self) -> bool {
-        self.collector.at_capacity()
-    }
-}
-
-impl<T: FastValue + PartialOrd + Clone> Collector for TopFieldCollector<T> {
-    fn set_segment(&mut self, segment_id: u32, segment: &SegmentReader) -> Result<()> {
-        self.collector.set_segment_id(segment_id);
-        self.fast_field = Some(segment.fast_field_reader(self.field)?);
-        Ok(())
-    }
-
-    fn collect(&mut self, doc: DocId, _score: Score) {
-        let field_value = self
-            .fast_field
-            .as_ref()
-            .expect("collect() was called before set_segment. This should never happen.")
-            .get(doc);
-        self.collector.collect(doc, field_value);
-    }
-
-    fn requires_scoring(&self) -> bool {
-        false
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use query::Query;
-    use query::QueryParser;
-    use schema::Field;
-    use schema::IntOptions;
-    use schema::Schema;
-    use schema::{SchemaBuilder, FAST, TEXT};
-    use Index;
-    use IndexWriter;
-    use TantivyError;
-
-    const TITLE: &str = "title";
-    const SIZE: &str = "size";
-
-    #[test]
-    fn test_top_collector_not_at_capacity() {
-        let mut schema_builder = SchemaBuilder::new();
-        let title = schema_builder.add_text_field(TITLE, TEXT);
-        let size = schema_builder.add_u64_field(SIZE, FAST);
-        let schema = schema_builder.build();
-        let (index, query) = index("beer", title, schema, |index_writer| {
-            index_writer.add_document(doc!(
-                title => "bottle of beer",
-                size => 12u64,
-            ));
-            index_writer.add_document(doc!(
-                title => "growler of beer",
-                size => 64u64,
-            ));
-            index_writer.add_document(doc!(
-                title => "pint of beer",
-                size => 16u64,
-            ));
-        });
-        let searcher = index.searcher();
-
-        let mut top_collector = TopFieldCollector::with_limit(size, 4);
-        searcher.search(&*query, &mut top_collector).unwrap();
-        assert!(!top_collector.at_capacity());
-
-        let score_docs: Vec<(u64, DocId)> = top_collector
-            .top_docs()
-            .into_iter()
-            .map(|(field, doc_address)| (field, doc_address.doc()))
-            .collect();
-        assert_eq!(score_docs, vec![(64, 1), (16, 2), (12, 0)]);
-    }
-
-    #[test]
-    #[should_panic]
-    fn test_field_does_not_exist() {
-        let mut schema_builder = SchemaBuilder::new();
-        let title = schema_builder.add_text_field(TITLE, TEXT);
-        let size = schema_builder.add_u64_field(SIZE, FAST);
-        let schema = schema_builder.build();
-        let (index, _) = index("beer", title, schema, |index_writer| {
-            index_writer.add_document(doc!(
-                title => "bottle of beer",
-                size => 12u64,
-            ));
-        });
-        let searcher = index.searcher();
-        let segment = searcher.segment_reader(0);
-        let mut top_collector: TopFieldCollector<u64> = TopFieldCollector::with_limit(Field(2), 4);
-        let _ = top_collector.set_segment(0, segment);
-    }
-
-    #[test]
-    fn test_field_not_fast_field() {
-        let mut schema_builder = SchemaBuilder::new();
-        let title = schema_builder.add_text_field(TITLE, TEXT);
-        let size = schema_builder.add_u64_field(SIZE, IntOptions::default());
-        let schema = schema_builder.build();
-        let (index, _) = index("beer", title, schema, |index_writer| {
-            index_writer.add_document(doc!(
-                title => "bottle of beer",
-                size => 12u64,
-            ));
-        });
-        let searcher = index.searcher();
-        let segment = searcher.segment_reader(0);
-        let mut top_collector: TopFieldCollector<u64> = TopFieldCollector::with_limit(size, 4);
-        assert_matches!(
-            top_collector.set_segment(0, segment),
-            Err(TantivyError::FastFieldError(_))
-        );
-    }
-
-    #[test]
-    #[should_panic]
-    fn test_collect_before_set_segment() {
-        let mut top_collector: TopFieldCollector<u64> = TopFieldCollector::with_limit(Field(0), 4);
-        top_collector.collect(0, 0f32);
-    }
-
-    #[test]
-    #[should_panic]
-    fn test_top_0() {
-        let _: TopFieldCollector<u64> = TopFieldCollector::with_limit(Field(0), 0);
-    }
-
-    fn index(
-        query: &str,
-        query_field: Field,
-        schema: Schema,
-        mut doc_adder: impl FnMut(&mut IndexWriter) -> (),
-    ) -> (Index, Box<Query>) {
-        let index = Index::create_in_ram(schema);
-
-        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-        doc_adder(&mut index_writer);
-        index_writer.commit().unwrap();
-        index.load_searchers().unwrap();
-
-        let query_parser = QueryParser::for_index(&index, vec![query_field]);
-        let query = query_parser.parse_query(query).unwrap();
-        (index, query)
-    }
-}

src/collector/tweak_score_top_collector.rs

@@ -0,0 +1,127 @@
+use crate::collector::top_collector::{TopCollector, TopSegmentCollector};
+use crate::collector::{Collector, SegmentCollector};
+use crate::DocAddress;
+use crate::{DocId, Result, Score, SegmentReader};
+
+pub(crate) struct TweakedScoreTopCollector<TScoreTweaker, TScore = Score> {
+    score_tweaker: TScoreTweaker,
+    collector: TopCollector<TScore>,
+}
+
+impl<TScoreTweaker, TScore> TweakedScoreTopCollector<TScoreTweaker, TScore>
+where
+    TScore: Clone + PartialOrd,
+{
+    pub fn new(
+        score_tweaker: TScoreTweaker,
+        collector: TopCollector<TScore>,
+    ) -> TweakedScoreTopCollector<TScoreTweaker, TScore> {
+        TweakedScoreTopCollector {
+            score_tweaker,
+            collector,
+        }
+    }
+}
+
+/// A `ScoreSegmentTweaker` makes it possible to modify the default score
+/// for a given document belonging to a specific segment.
+///
+/// It is the segment local version of the [`ScoreTweaker`](./trait.ScoreTweaker.html).
+pub trait ScoreSegmentTweaker<TScore>: 'static {
+    /// Tweak the given `score` for the document `doc`.
+    fn score(&mut self, doc: DocId, score: Score) -> TScore;
+}
+
+/// `ScoreTweaker` makes it possible to tweak the score
+/// emitted  by the scorer into another one.
+///
+/// The `ScoreTweaker` itself does not make much of the computation itself.
+/// Instead, it helps constructing `Self::Child` instances that will compute
+/// the score at a segment scale.
+pub trait ScoreTweaker<TScore>: Sync {
+    /// Type of the associated [`ScoreSegmentTweaker`](./trait.ScoreSegmentTweaker.html).
+    type Child: ScoreSegmentTweaker<TScore>;
+
+    /// Builds a child tweaker for a specific segment. The child scorer is associated to
+    /// a specific segment.
+    fn segment_tweaker(&self, segment_reader: &SegmentReader) -> Result<Self::Child>;
+}
+
+impl<TScoreTweaker, TScore> Collector for TweakedScoreTopCollector<TScoreTweaker, TScore>
+where
+    TScoreTweaker: ScoreTweaker<TScore> + Send + Sync,
+    TScore: 'static + PartialOrd + Clone + Send + Sync,
+{
+    type Fruit = Vec<(TScore, DocAddress)>;
+
+    type Child = TopTweakedScoreSegmentCollector<TScoreTweaker::Child, TScore>;
+
+    fn for_segment(
+        &self,
+        segment_local_id: u32,
+        segment_reader: &SegmentReader,
+    ) -> Result<Self::Child> {
+        let segment_scorer = self.score_tweaker.segment_tweaker(segment_reader)?;
+        let segment_collector = self.collector.for_segment(segment_local_id, segment_reader);
+        Ok(TopTweakedScoreSegmentCollector {
+            segment_collector,
+            segment_scorer,
+        })
+    }
+
+    fn requires_scoring(&self) -> bool {
+        true
+    }
+
+    fn merge_fruits(&self, segment_fruits: Vec<Self::Fruit>) -> Result<Self::Fruit> {
+        self.collector.merge_fruits(segment_fruits)
+    }
+}
+
+pub struct TopTweakedScoreSegmentCollector<TSegmentScoreTweaker, TScore>
+where
+    TScore: 'static + PartialOrd + Clone + Send + Sync + Sized,
+    TSegmentScoreTweaker: ScoreSegmentTweaker<TScore>,
+{
+    segment_collector: TopSegmentCollector<TScore>,
+    segment_scorer: TSegmentScoreTweaker,
+}
+
+impl<TSegmentScoreTweaker, TScore> SegmentCollector
+    for TopTweakedScoreSegmentCollector<TSegmentScoreTweaker, TScore>
+where
+    TScore: 'static + PartialOrd + Clone + Send + Sync,
+    TSegmentScoreTweaker: 'static + ScoreSegmentTweaker<TScore>,
+{
+    type Fruit = Vec<(TScore, DocAddress)>;
+
+    fn collect(&mut self, doc: DocId, score: Score) {
+        let score = self.segment_scorer.score(doc, score);
+        self.segment_collector.collect(doc, score);
+    }
+
+    fn harvest(self) -> Vec<(TScore, DocAddress)> {
+        self.segment_collector.harvest()
+    }
+}
+
+impl<F, TScore, TSegmentScoreTweaker> ScoreTweaker<TScore> for F
+where
+    F: 'static + Send + Sync + Fn(&SegmentReader) -> TSegmentScoreTweaker,
+    TSegmentScoreTweaker: ScoreSegmentTweaker<TScore>,
+{
+    type Child = TSegmentScoreTweaker;
+
+    fn segment_tweaker(&self, segment_reader: &SegmentReader) -> Result<Self::Child> {
+        Ok((self)(segment_reader))
+    }
+}
+
+impl<F, TScore> ScoreSegmentTweaker<TScore> for F
+where
+    F: 'static + FnMut(DocId, Score) -> TScore,
+{
+    fn score(&mut self, doc: DocId, score: Score) -> TScore {
+        (self)(doc, score)
+    }
+}

src/common/bitpacker.rs

@@ -1,188 +0,0 @@
-use common::serialize::BinarySerializable;
-use std::io;
-use std::io::Write;
-use std::mem;
-use std::ops::Deref;
-use std::ptr;
-
-pub(crate) struct BitPacker {
-    mini_buffer: u64,
-    mini_buffer_written: usize,
-}
-
-impl BitPacker {
-    pub fn new() -> BitPacker {
-        BitPacker {
-            mini_buffer: 0u64,
-            mini_buffer_written: 0,
-        }
-    }
-
-    pub fn write<TWrite: Write>(
-        &mut self,
-        val: u64,
-        num_bits: u8,
-        output: &mut TWrite,
-    ) -> io::Result<()> {
-        let val_u64 = val as u64;
-        let num_bits = num_bits as usize;
-        if self.mini_buffer_written + num_bits > 64 {
-            self.mini_buffer |= val_u64.wrapping_shl(self.mini_buffer_written as u32);
-            self.mini_buffer.serialize(output)?;
-            self.mini_buffer = val_u64.wrapping_shr((64 - self.mini_buffer_written) as u32);
-            self.mini_buffer_written = self.mini_buffer_written + num_bits - 64;
-        } else {
-            self.mini_buffer |= val_u64 << self.mini_buffer_written;
-            self.mini_buffer_written += num_bits;
-            if self.mini_buffer_written == 64 {
-                self.mini_buffer.serialize(output)?;
-                self.mini_buffer_written = 0;
-                self.mini_buffer = 0u64;
-            }
-        }
-        Ok(())
-    }
-
-    pub fn flush<TWrite: Write>(&mut self, output: &mut TWrite) -> io::Result<()> {
-        if self.mini_buffer_written > 0 {
-            let num_bytes = (self.mini_buffer_written + 7) / 8;
-            let arr: [u8; 8] = unsafe { mem::transmute::<u64, [u8; 8]>(self.mini_buffer.to_le()) };
-            output.write_all(&arr[..num_bytes])?;
-            self.mini_buffer_written = 0;
-        }
-        Ok(())
-    }
-
-    pub fn close<TWrite: Write>(&mut self, output: &mut TWrite) -> io::Result<()> {
-        self.flush(output)?;
-        // Padding the write file to simplify reads.
-        output.write_all(&[0u8; 7])?;
-        Ok(())
-    }
-}
-
-#[derive(Clone)]
-pub struct BitUnpacker<Data>
-where
-    Data: Deref<Target = [u8]>,
-{
-    num_bits: usize,
-    mask: u64,
-    data: Data,
-}
-
-impl<Data> BitUnpacker<Data>
-where
-    Data: Deref<Target = [u8]>,
-{
-    pub fn new(data: Data, num_bits: u8) -> BitUnpacker<Data> {
-        let mask: u64 = if num_bits == 64 {
-            !0u64
-        } else {
-            (1u64 << num_bits) - 1u64
-        };
-        BitUnpacker {
-            num_bits: num_bits as usize,
-            mask,
-            data,
-        }
-    }
-
-    pub fn get(&self, idx: usize) -> u64 {
-        if self.num_bits == 0 {
-            return 0u64;
-        }
-        let data: &[u8] = &*self.data;
-        let num_bits = self.num_bits;
-        let mask = self.mask;
-        let addr_in_bits = idx * num_bits;
-        let addr = addr_in_bits >> 3;
-        let bit_shift = addr_in_bits & 7;
-        debug_assert!(
-            addr + 8 <= data.len(),
-            "The fast field field should have been padded with 7 bytes."
-        );
-        #[cfg_attr(feature = "cargo-clippy", allow(clippy::cast_ptr_alignment))]
-        let val_unshifted_unmasked: u64 =
-            u64::from_le(unsafe { ptr::read_unaligned(data[addr..].as_ptr() as *const u64) });
-        let val_shifted = (val_unshifted_unmasked >> bit_shift) as u64;
-        val_shifted & mask
-    }
-
-    /// Reads a range of values from the fast field.
-    ///
-    /// The range of values read is from
-    /// `[start..start + output.len()[`
-    pub fn get_range(&self, start: u32, output: &mut [u64]) {
-        if self.num_bits == 0 {
-            for val in output.iter_mut() {
-                *val = 0u64;
-            }
-        } else {
-            let data: &[u8] = &*self.data;
-            let num_bits = self.num_bits;
-            let mask = self.mask;
-            let mut addr_in_bits = (start as usize) * num_bits;
-            for output_val in output.iter_mut() {
-                let addr = addr_in_bits >> 3;
-                let bit_shift = addr_in_bits & 7;
-                #[cfg_attr(feature = "cargo-clippy", allow(clippy::cast_ptr_alignment))]
-                let val_unshifted_unmasked: u64 =
-                    unsafe { ptr::read_unaligned(data[addr..].as_ptr() as *const u64) };
-                let val_shifted = (val_unshifted_unmasked >> bit_shift) as u64;
-                *output_val = val_shifted & mask;
-                addr_in_bits += num_bits;
-            }
-        }
-    }
-}
-
-#[cfg(test)]
-mod test {
-    use super::{BitPacker, BitUnpacker};
-
-    fn create_fastfield_bitpacker(len: usize, num_bits: u8) -> (BitUnpacker<Vec<u8>>, Vec<u64>) {
-        let mut data = Vec::new();
-        let mut bitpacker = BitPacker::new();
-        let max_val: u64 = (1u64 << num_bits as u64) - 1u64;
-        let vals: Vec<u64> = (0u64..len as u64)
-            .map(|i| if max_val == 0 { 0 } else { i % max_val })
-            .collect();
-        for &val in &vals {
-            bitpacker.write(val, num_bits, &mut data).unwrap();
-        }
-        bitpacker.close(&mut data).unwrap();
-        assert_eq!(data.len(), ((num_bits as usize) * len + 7) / 8 + 7);
-        let bitunpacker = BitUnpacker::new(data, num_bits);
-        (bitunpacker, vals)
-    }
-
-    fn test_bitpacker_util(len: usize, num_bits: u8) {
-        let (bitunpacker, vals) = create_fastfield_bitpacker(len, num_bits);
-        for (i, val) in vals.iter().enumerate() {
-            assert_eq!(bitunpacker.get(i), *val);
-        }
-    }
-
-    #[test]
-    fn test_bitpacker() {
-        test_bitpacker_util(10, 3);
-        test_bitpacker_util(10, 0);
-        test_bitpacker_util(10, 1);
-        test_bitpacker_util(6, 14);
-        test_bitpacker_util(1000, 14);
-    }
-
-    #[test]
-    fn test_bitpacker_range() {
-        let (bitunpacker, vals) = create_fastfield_bitpacker(100_000, 12);
-        let buffer_len = 100;
-        let mut buffer = vec![0u64; buffer_len];
-        for start in vec![0, 10, 20, 100, 1_000] {
-            bitunpacker.get_range(start as u32, &mut buffer[..]);
-            for i in 0..buffer_len {
-                assert_eq!(buffer[i], vals[start + i]);
-            }
-        }
-    }
-}

src/common/counting_writer.rs

@@ -1,55 +0,0 @@
-use std::io;
-use std::io::Write;
-
-pub struct CountingWriter<W> {
-    underlying: W,
-    written_bytes: usize,
-}
-
-impl<W: Write> CountingWriter<W> {
-    pub fn wrap(underlying: W) -> CountingWriter<W> {
-        CountingWriter {
-            underlying,
-            written_bytes: 0,
-        }
-    }
-
-    pub fn written_bytes(&self) -> usize {
-        self.written_bytes
-    }
-
-    pub fn finish(mut self) -> io::Result<(W, usize)> {
-        self.flush()?;
-        Ok((self.underlying, self.written_bytes))
-    }
-}
-
-impl<W: Write> Write for CountingWriter<W> {
-    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
-        let written_size = self.underlying.write(buf)?;
-        self.written_bytes += written_size;
-        Ok(written_size)
-    }
-
-    fn flush(&mut self) -> io::Result<()> {
-        self.underlying.flush()
-    }
-}
-
-#[cfg(test)]
-mod test {
-
-    use super::CountingWriter;
-    use std::io::Write;
-
-    #[test]
-    fn test_counting_writer() {
-        let buffer: Vec<u8> = vec![];
-        let mut counting_writer = CountingWriter::wrap(buffer);
-        let bytes = (0u8..10u8).collect::<Vec<u8>>();
-        counting_writer.write_all(&bytes).unwrap();
-        let (w, len): (Vec<u8>, usize) = counting_writer.finish().unwrap();
-        assert_eq!(len, 10);
-        assert_eq!(w.len(), 10);
-    }
-}

src/common/mod.rs

@@ -1,137 +0,0 @@
-pub mod bitpacker;
-mod bitset;
-mod composite_file;
-mod counting_writer;
-mod serialize;
-mod vint;
-
-pub use self::bitset::BitSet;
-pub(crate) use self::bitset::TinySet;
-pub(crate) use self::composite_file::{CompositeFile, CompositeWrite};
-pub use self::counting_writer::CountingWriter;
-pub use self::serialize::{BinarySerializable, FixedSize};
-pub use self::vint::VInt;
-pub use byteorder::LittleEndian as Endianness;
-
-use std::io;
-
-/// Computes the number of bits that will be used for bitpacking.
-///
-/// In general the target is the minimum number of bits
-/// required to express the amplitude given in argument.
-///
-/// e.g. If the amplitude is 10, we can store all ints on simply 4bits.
-///
-/// The logic is slightly more convoluted here as for optimization
-/// reasons, we want to ensure that a value spawns over at most 8 bytes
-/// of aligns bytes.
-///
-/// Spanning over 9 bytes is possible for instance, if we do
-/// bitpacking with an amplitude of 63 bits.
-/// In this case, the second int will start on bit
-/// 63 (which belongs to byte 7) and ends at byte 15;
-/// Hence 9 bytes (from byte 7 to byte 15 included).
-///
-/// To avoid this, we force the number of bits to 64bits
-/// when the result is greater than `64-8 = 56 bits`.
-///
-/// Note that this only affects rare use cases spawning over
-/// a very large range of values. Even in this case, it results
-/// in an extra cost of at most 12% compared to the optimal
-/// number of bits.
-pub(crate) fn compute_num_bits(n: u64) -> u8 {
-    let amplitude = (64u32 - n.leading_zeros()) as u8;
-    if amplitude <= 64 - 8 {
-        amplitude
-    } else {
-        64
-    }
-}
-
-pub(crate) fn is_power_of_2(n: usize) -> bool {
-    (n > 0) && (n & (n - 1) == 0)
-}
-
-/// Create a default io error given a string.
-pub(crate) fn make_io_err(msg: String) -> io::Error {
-    io::Error::new(io::ErrorKind::Other, msg)
-}
-
-/// Has length trait
-pub trait HasLen {
-    /// Return length
-    fn len(&self) -> usize;
-
-    /// Returns true iff empty.
-    fn is_empty(&self) -> bool {
-        self.len() == 0
-    }
-}
-
-const HIGHEST_BIT: u64 = 1 << 63;
-
-/// Maps a `i64` to `u64`
-///
-/// For simplicity, tantivy internally handles `i64` as `u64`.
-/// The mapping is defined by this function.
-///
-/// Maps `i64` to `u64` so that
-/// `-2^63 .. 2^63-1` is mapped
-///     to
-/// `0 .. 2^64-1`
-/// in that order.
-///
-/// This is more suited than simply casting (`val as u64`)
-/// because of bitpacking.
-///
-/// Imagine a list of `i64` ranging from -10 to 10.
-/// When casting negative values, the negative values are projected
-/// to values over 2^63, and all values end up requiring 64 bits.
-///
-/// # See also
-/// The [reverse mapping is `u64_to_i64`](./fn.u64_to_i64.html).
-#[inline(always)]
-pub fn i64_to_u64(val: i64) -> u64 {
-    (val as u64) ^ HIGHEST_BIT
-}
-
-/// Reverse the mapping given by [`i64_to_u64`](./fn.i64_to_u64.html).
-#[inline(always)]
-pub fn u64_to_i64(val: u64) -> i64 {
-    (val ^ HIGHEST_BIT) as i64
-}
-
-#[cfg(test)]
-pub(crate) mod test {
-
-    pub use super::serialize::test::fixed_size_test;
-    use super::{compute_num_bits, i64_to_u64, u64_to_i64};
-
-    fn test_i64_converter_helper(val: i64) {
-        assert_eq!(u64_to_i64(i64_to_u64(val)), val);
-    }
-
-    #[test]
-    fn test_i64_converter() {
-        assert_eq!(i64_to_u64(i64::min_value()), u64::min_value());
-        assert_eq!(i64_to_u64(i64::max_value()), u64::max_value());
-        test_i64_converter_helper(0i64);
-        test_i64_converter_helper(i64::min_value());
-        test_i64_converter_helper(i64::max_value());
-        for i in -1000i64..1000i64 {
-            test_i64_converter_helper(i);
-        }
-    }
-
-    #[test]
-    fn test_compute_num_bits() {
-        assert_eq!(compute_num_bits(1), 1u8);
-        assert_eq!(compute_num_bits(0), 0u8);
-        assert_eq!(compute_num_bits(2), 2u8);
-        assert_eq!(compute_num_bits(3), 2u8);
-        assert_eq!(compute_num_bits(4), 3u8);
-        assert_eq!(compute_num_bits(255), 8u8);
-        assert_eq!(compute_num_bits(256), 9u8);
-        assert_eq!(compute_num_bits(5_000_000_000), 33u8);
-    }
-}

src/common/vint.rs

@@ -1,111 +0,0 @@
-use super::BinarySerializable;
-use std::io;
-use std::io::Read;
-use std::io::Write;
-
-///   Wrapper over a `u64` that serializes as a variable int.
-#[derive(Debug, Eq, PartialEq)]
-pub struct VInt(pub u64);
-
-const STOP_BIT: u8 = 128;
-
-impl VInt {
-    pub fn val(&self) -> u64 {
-        self.0
-    }
-
-    pub fn deserialize_u64<R: Read>(reader: &mut R) -> io::Result<u64> {
-        VInt::deserialize(reader).map(|vint| vint.0)
-    }
-
-    pub fn serialize_into_vec(&self, output: &mut Vec<u8>) {
-        let mut buffer = [0u8; 10];
-        let num_bytes = self.serialize_into(&mut buffer);
-        output.extend(&buffer[0..num_bytes]);
-    }
-
-    fn serialize_into(&self, buffer: &mut [u8; 10]) -> usize {
-        let mut remaining = self.0;
-        for (i, b) in buffer.iter_mut().enumerate() {
-            let next_byte: u8 = (remaining % 128u64) as u8;
-            remaining /= 128u64;
-            if remaining == 0u64 {
-                *b = next_byte | STOP_BIT;
-                return i + 1;
-            } else {
-                *b = next_byte;
-            }
-        }
-        unreachable!();
-    }
-}
-
-impl BinarySerializable for VInt {
-    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
-        let mut buffer = [0u8; 10];
-        let num_bytes = self.serialize_into(&mut buffer);
-        writer.write_all(&buffer[0..num_bytes])
-    }
-
-    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
-        let mut bytes = reader.bytes();
-        let mut result = 0u64;
-        let mut shift = 0u64;
-        loop {
-            match bytes.next() {
-                Some(Ok(b)) => {
-                    result |= u64::from(b % 128u8) << shift;
-                    if b >= STOP_BIT {
-                        return Ok(VInt(result));
-                    }
-                    shift += 7;
-                }
-                _ => {
-                    return Err(io::Error::new(
-                        io::ErrorKind::InvalidData,
-                        "Reach end of buffer while reading VInt",
-                    ))
-                }
-            }
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-
-    use super::VInt;
-    use common::BinarySerializable;
-
-    fn aux_test_vint(val: u64) {
-        let mut v = [14u8; 10];
-        let num_bytes = VInt(val).serialize_into(&mut v);
-        for i in num_bytes..10 {
-            assert_eq!(v[i], 14u8);
-        }
-        assert!(num_bytes > 0);
-        if num_bytes < 10 {
-            assert!(1u64 << (7 * num_bytes) > val);
-        }
-        if num_bytes > 1 {
-            assert!(1u64 << (7 * (num_bytes - 1)) <= val);
-        }
-        let serdeser_val = VInt::deserialize(&mut &v[..]).unwrap();
-        assert_eq!(val, serdeser_val.0);
-    }
-
-    #[test]
-    fn test_vint() {
-        aux_test_vint(0);
-        aux_test_vint(1);
-        aux_test_vint(5);
-        aux_test_vint(u64::max_value());
-        for i in 1..9 {
-            let power_of_128 = 1u64 << (7 * i);
-            aux_test_vint(power_of_128 - 1u64);
-            aux_test_vint(power_of_128);
-            aux_test_vint(power_of_128 + 1u64);
-        }
-        aux_test_vint(10);
-    }
-}

src/core/executor.rs

@@ -0,0 +1,95 @@
+use crossbeam::channel;
+
+/// Search executor whether search request are single thread or multithread.
+///
+/// We don't expose Rayon thread pool directly here for several reasons.
+///
+/// First dependency hell. It is not a good idea to expose the
+/// API of a dependency, knowing it might conflict with a different version
+/// used by the client. Second, we may stop using rayon in the future.
+pub enum Executor {
+    /// Single thread variant of an Executor
+    SingleThread,
+}
+
+impl Executor {
+    /// Creates an Executor that performs all task in the caller thread.
+    pub fn single_thread() -> Executor {
+        Executor::SingleThread
+    }
+
+    /// Perform a map in the thread pool.
+    ///
+    /// Regardless of the executor (`SingleThread` or `ThreadPool`), panics in the task
+    /// will propagate to the caller.
+    pub fn map<
+        A: Send,
+        R: Send,
+        AIterator: Iterator<Item = A>,
+        F: Sized + Sync + Fn(A) -> crate::Result<R>,
+    >(
+        &self,
+        f: F,
+        args: AIterator,
+    ) -> crate::Result<Vec<R>> {
+        match self {
+            Executor::SingleThread => args.map(f).collect::<crate::Result<_>>(),
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+
+    use super::Executor;
+
+    #[test]
+    #[should_panic(expected = "panic should propagate")]
+    fn test_panic_propagates_single_thread() {
+        let _result: Vec<usize> = Executor::single_thread()
+            .map(
+                |_| {
+                    panic!("panic should propagate");
+                },
+                vec![0].into_iter(),
+            )
+            .unwrap();
+    }
+
+    #[test]
+    #[should_panic] //< unfortunately the panic message is not propagated
+    fn test_panic_propagates_multi_thread() {
+        let _result: Vec<usize> = Executor::multi_thread(1, "search-test")
+            .unwrap()
+            .map(
+                |_| {
+                    panic!("panic should propagate");
+                },
+                vec![0].into_iter(),
+            )
+            .unwrap();
+    }
+
+    #[test]
+    fn test_map_singlethread() {
+        let result: Vec<usize> = Executor::single_thread()
+            .map(|i| Ok(i * 2), 0..1_000)
+            .unwrap();
+        assert_eq!(result.len(), 1_000);
+        for i in 0..1_000 {
+            assert_eq!(result[i], i * 2);
+        }
+    }
+
+    #[test]
+    fn test_map_multithread() {
+        let result: Vec<usize> = Executor::multi_thread(3, "search-test")
+            .unwrap()
+            .map(|i| Ok(i * 2), 0..10)
+            .unwrap();
+        assert_eq!(result.len(), 10);
+        for i in 0..10 {
+            assert_eq!(result[i], i * 2);
+        }
+    }
+}

src/core/index.rs

@@ -1,97 +1,134 @@
-use super::pool::LeasedItem;
-use super::pool::Pool;
-use super::segment::create_segment;
-use super::segment::Segment;
-use core::searcher::Searcher;
-use core::IndexMeta;
-use core::SegmentId;
-use core::SegmentMeta;
-use core::SegmentReader;
-use core::META_FILEPATH;
-use directory::ManagedDirectory;
+use super::{segment::Segment, IndexSettings};
+use crate::core::Executor;
+use crate::core::IndexMeta;
+use crate::core::SegmentId;
+use crate::core::SegmentMeta;
+use crate::core::SegmentMetaInventory;
+use crate::core::META_FILEPATH;
+use crate::directory::error::OpenReadError;
+use crate::directory::ManagedDirectory;
 #[cfg(feature = "mmap")]
-use directory::MmapDirectory;
-use directory::{Directory, RAMDirectory};
-use error::TantivyError;
-use indexer::index_writer::open_index_writer;
-use indexer::index_writer::HEAP_SIZE_MIN;
-use indexer::segment_updater::save_new_metas;
-use indexer::LockType;
-use num_cpus;
-use schema::Field;
-use schema::FieldType;
-use schema::Schema;
-use serde_json;
-use std::borrow::BorrowMut;
+use crate::directory::MmapDirectory;
+use crate::directory::INDEX_WRITER_LOCK;
+use crate::directory::{Directory, RamDirectory};
+use crate::error::DataCorruption;
+use crate::error::TantivyError;
+use crate::indexer::index_writer::{HEAP_SIZE_MIN, MAX_NUM_THREAD};
+use crate::indexer::segment_updater::save_new_metas;
+use crate::reader::IndexReader;
+use crate::reader::IndexReaderBuilder;
+use crate::schema::Field;
+use crate::schema::FieldType;
+use crate::schema::Schema;
+use crate::tokenizer::{TextAnalyzer, TokenizerManager};
+use crate::IndexWriter;
+use std::collections::HashSet;
 use std::fmt;
+
+#[cfg(feature = "mmap")]
 use std::path::Path;
-use std::sync::atomic::{AtomicUsize, Ordering};
+use std::path::PathBuf;
 use std::sync::Arc;
-use tokenizer::BoxedTokenizer;
-use tokenizer::TokenizerManager;
-use IndexWriter;
-use Result;
 
-fn load_metas(directory: &Directory) -> Result<IndexMeta> {
+fn load_metas(
+    directory: &dyn Directory,
+    inventory: &SegmentMetaInventory,
+) -> crate::Result<IndexMeta> {
     let meta_data = directory.atomic_read(&META_FILEPATH)?;
-    let meta_string = String::from_utf8_lossy(&meta_data);
-    serde_json::from_str(&meta_string)
-        .map_err(|_| TantivyError::CorruptedFile(META_FILEPATH.clone()))
+    let meta_string = String::from_utf8(meta_data).map_err(|_utf8_err| {
+        error!("Meta data is not valid utf8.");
+        DataCorruption::new(
+            META_FILEPATH.to_path_buf(),
+            "Meta file does not contain valid utf8 file.".to_string(),
+        )
+    })?;
+    IndexMeta::deserialize(&meta_string, inventory)
+        .map_err(|e| {
+            DataCorruption::new(
+                META_FILEPATH.to_path_buf(),
+                format!(
+                    "Meta file cannot be deserialized. {:?}. Content: {:?}",
+                    e, meta_string
+                ),
+            )
+        })
+        .map_err(From::from)
 }
 
-/// Search Index
-pub struct Index {
-    directory: ManagedDirectory,
-    schema: Schema,
-    num_searchers: Arc<AtomicUsize>,
-    searcher_pool: Arc<Pool<Searcher>>,
-    tokenizers: TokenizerManager,
+/// IndexBuilder can be used to create an index.
+///
+/// Use in conjunction with `SchemaBuilder`. Global index settings
+/// can be configured with `IndexSettings`
+///
+/// # Examples
+///
+/// ```
+/// use tantivy::schema::*;
+/// use tantivy::{Index, IndexSettings, IndexSortByField, Order};
+///
+/// let mut schema_builder = Schema::builder();
+/// let id_field = schema_builder.add_text_field("id", STRING);
+/// let title_field = schema_builder.add_text_field("title", TEXT);
+/// let body_field = schema_builder.add_text_field("body", TEXT);
+/// let number_field = schema_builder.add_u64_field(
+///     "number",
+///     IntOptions::default().set_fast(Cardinality::SingleValue),
+/// );
+///
+/// let schema = schema_builder.build();
+/// let settings = IndexSettings{sort_by_field: Some(IndexSortByField{field:"number".to_string(), order:Order::Asc}), ..Default::default()};
+/// let index = Index::builder().schema(schema).settings(settings).create_in_ram();
+///
+/// ```
+pub struct IndexBuilder {
+    schema: Option<Schema>,
+    index_settings: IndexSettings,
 }
-
-impl Index {
-    /// Examines the director to see if it contains an index
-    pub fn exists<Dir: Directory>(dir: &Dir) -> bool {
-        dir.exists(&META_FILEPATH)
+impl Default for IndexBuilder {
+    fn default() -> Self {
+        IndexBuilder::new()
+    }
+}
+impl IndexBuilder {
+    /// Creates a new `IndexBuilder`
+    pub fn new() -> Self {
+        Self {
+            schema: None,
+            index_settings: IndexSettings::default(),
+        }
+    }
+    /// Set the settings
+    pub fn settings(mut self, settings: IndexSettings) -> Self {
+        self.index_settings = settings;
+        self
+    }
+    /// Set the schema
+    pub fn schema(mut self, schema: Schema) -> Self {
+        self.schema = Some(schema);
+        self
     }
-
     /// Creates a new index using the `RAMDirectory`.
     ///
     /// The index will be allocated in anonymous memory.
     /// This should only be used for unit tests.
-    pub fn create_in_ram(schema: Schema) -> Index {
-        let ram_directory = RAMDirectory::create();
-        Index::create(ram_directory, schema).expect("Creating a RAMDirectory should never fail")
+    pub fn create_in_ram(self) -> Result<Index, TantivyError> {
+        let ram_directory = RamDirectory::create();
+        Ok(self
+            .create(ram_directory)
+            .expect("Creating a RAMDirectory should never fail"))
     }
-
     /// Creates a new index in a given filepath.
     /// The index will use the `MMapDirectory`.
     ///
-    /// If a previous index was in this directory, then its meta file will be destroyed.
+    /// If a previous index was in this directory, it returns an `IndexAlreadyExists` error.
     #[cfg(feature = "mmap")]
-    pub fn create_in_dir<P: AsRef<Path>>(directory_path: P, schema: Schema) -> Result<Index> {
+    pub fn create_in_dir<P: AsRef<Path>>(self, directory_path: P) -> crate::Result<Index> {
         let mmap_directory = MmapDirectory::open(directory_path)?;
-        if Index::exists(&mmap_directory) {
+        if Index::exists(&mmap_directory)? {
             return Err(TantivyError::IndexAlreadyExists);
         }
-
-        Index::create(mmap_directory, schema)
+        self.create(mmap_directory)
     }
-
-    /// Opens or creates a new index in the provided directory
-    #[cfg(feature = "mmap")]
-    pub fn open_or_create<Dir: Directory>(dir: Dir, schema: Schema) -> Result<Index> {
-        if Index::exists(&dir) {
-            let index = Index::open(dir)?;
-            if index.schema() == schema {
-                Ok(index)
-            } else {
-                Err(TantivyError::SchemaError("An index exists but the schema does not match.".to_string()))
-            }
-        } else {
-            Index::create(dir, schema)
-        }
-    }
-
     /// Creates a new index in a temp directory.
     ///
     /// The index will use the `MMapDirectory` in a newly created directory.
@@ -101,39 +138,163 @@ impl Index {
     /// The temp directory is only used for testing the `MmapDirectory`.
     /// For other unit tests, prefer the `RAMDirectory`, see: `create_in_ram`.
     #[cfg(feature = "mmap")]
-    pub fn create_from_tempdir(schema: Schema) -> Result<Index> {
+    pub fn create_from_tempdir(self) -> crate::Result<Index> {
         let mmap_directory = MmapDirectory::create_from_tempdir()?;
-        Index::create(mmap_directory, schema)
+        self.create(mmap_directory)
     }
-
-    /// Creates a new index given an implementation of the trait `Directory`
-    pub fn create<Dir: Directory>(dir: Dir, schema: Schema) -> Result<Index> {
-        let directory = ManagedDirectory::new(dir)?;
-        Index::from_directory(directory, schema)
+    fn get_expect_schema(&self) -> crate::Result<Schema> {
+        self.schema
+            .as_ref()
+            .cloned()
+            .ok_or(TantivyError::IndexBuilderMissingArgument("schema"))
     }
-
-    /// Create a new index from a directory.
+    /// Opens or creates a new index in the provided directory
+    pub fn open_or_create<Dir: Directory>(self, dir: Dir) -> crate::Result<Index> {
+        if !Index::exists(&dir)? {
+            return self.create(dir);
+        }
+        let index = Index::open(dir)?;
+        if index.schema() == self.get_expect_schema()? {
+            Ok(index)
+        } else {
+            Err(TantivyError::SchemaError(
+                "An index exists but the schema does not match.".to_string(),
+            ))
+        }
+    }
+    /// Creates a new index given an implementation of the trait `Directory`.
     ///
-    /// This will overwrite existing meta.json
-    fn from_directory(mut directory: ManagedDirectory, schema: Schema) -> Result<Index> {
-        save_new_metas(schema.clone(), 0, directory.borrow_mut())?;
-        let metas = IndexMeta::with_schema(schema);
-        Index::create_from_metas(directory, &metas)
+    /// If a directory previously existed, it will be erased.
+    fn create<Dir: Directory>(self, dir: Dir) -> crate::Result<Index> {
+        let directory = ManagedDirectory::wrap(dir)?;
+        save_new_metas(
+            self.get_expect_schema()?,
+            self.index_settings.clone(),
+            &directory,
+        )?;
+        let mut metas = IndexMeta::with_schema(self.get_expect_schema()?);
+        metas.index_settings = self.index_settings;
+        let index = Index::open_from_metas(directory, &metas, SegmentMetaInventory::default());
+        Ok(index)
+    }
+}
+
+/// Search Index
+#[derive(Clone)]
+pub struct Index {
+    directory: ManagedDirectory,
+    schema: Schema,
+    settings: IndexSettings,
+    executor: Arc<Executor>,
+    tokenizers: TokenizerManager,
+    inventory: SegmentMetaInventory,
+}
+
+impl Index {
+    /// Creates a new builder.
+    pub fn builder() -> IndexBuilder {
+        IndexBuilder::new()
+    }
+    /// Examines the directory to see if it contains an index.
+    ///
+    /// Effectively, it only checks for the presence of the `meta.json` file.
+    pub fn exists<Dir: Directory>(dir: &Dir) -> Result<bool, OpenReadError> {
+        dir.exists(&META_FILEPATH)
+    }
+
+    /// Accessor to the search executor.
+    ///
+    /// This pool is used by default when calling `searcher.search(...)`
+    /// to perform search on the individual segments.
+    ///
+    /// By default the executor is single thread, and simply runs in the calling thread.
+    pub fn search_executor(&self) -> &Executor {
+        self.executor.as_ref()
+    }
+
+    /// Replace the default single thread search executor pool
+    /// by a thread pool with a given number of threads.
+    pub fn set_multithread_executor(&mut self, num_threads: usize) -> crate::Result<()> {
+        self.executor = Arc::new(Executor::multi_thread(num_threads, "thrd-tantivy-search-")?);
+        Ok(())
+    }
+
+    /// Replace the default single thread search executor pool
+    /// by a thread pool with a given number of threads.
+    pub fn set_default_multithread_executor(&mut self) -> crate::Result<()> {
+        let default_num_threads = num_cpus::get();
+        self.set_multithread_executor(default_num_threads)
+    }
+
+    /// Creates a new index using the `RamDirectory`.
+    ///
+    /// The index will be allocated in anonymous memory.
+    /// This is useful for indexing small set of documents
+    /// for instances like unit test or temporary in memory index.
+    pub fn create_in_ram(schema: Schema) -> Index {
+        IndexBuilder::new().schema(schema).create_in_ram().unwrap()
+    }
+
+    /// Creates a new index in a given filepath.
+    /// The index will use the `MMapDirectory`.
+    ///
+    /// If a previous index was in this directory, then it returns  an `IndexAlreadyExists` error.
+    #[cfg(feature = "mmap")]
+    pub fn create_in_dir<P: AsRef<Path>>(
+        directory_path: P,
+        schema: Schema,
+    ) -> crate::Result<Index> {
+        IndexBuilder::new()
+            .schema(schema)
+            .create_in_dir(directory_path)
+    }
+
+    /// Opens or creates a new index in the provided directory
+    pub fn open_or_create<Dir: Directory>(dir: Dir, schema: Schema) -> crate::Result<Index> {
+        IndexBuilder::new().schema(schema).open_or_create(dir)
+    }
+
+    /// Creates a new index in a temp directory.
+    ///
+    /// The index will use the `MMapDirectory` in a newly created directory.
+    /// The temp directory will be destroyed automatically when the `Index` object
+    /// is destroyed.
+    ///
+    /// The temp directory is only used for testing the `MmapDirectory`.
+    /// For other unit tests, prefer the `RamDirectory`, see: `create_in_ram`.
+    #[cfg(feature = "mmap")]
+    pub fn create_from_tempdir(schema: Schema) -> crate::Result<Index> {
+        IndexBuilder::new().schema(schema).create_from_tempdir()
+    }
+
+    /// Creates a new index given an implementation of the trait `Directory`.
+    ///
+    /// If a directory previously existed, it will be erased.
+    pub fn create<Dir: Directory>(
+        dir: Dir,
+        schema: Schema,
+        settings: IndexSettings,
+    ) -> crate::Result<Index> {
+        let mut builder = IndexBuilder::new().schema(schema);
+        builder = builder.settings(settings);
+        builder.create(dir)
     }
 
     /// Creates a new index given a directory and an `IndexMeta`.
-    fn create_from_metas(directory: ManagedDirectory, metas: &IndexMeta) -> Result<Index> {
+    fn open_from_metas(
+        directory: ManagedDirectory,
+        metas: &IndexMeta,
+        inventory: SegmentMetaInventory,
+    ) -> Index {
         let schema = metas.schema.clone();
-        let n_cpus = num_cpus::get();
-        let index = Index {
+        Index {
+            settings: metas.index_settings.clone(),
             directory,
             schema,
-            num_searchers: Arc::new(AtomicUsize::new(n_cpus)),
-            searcher_pool: Arc::new(Pool::new()),
             tokenizers: TokenizerManager::default(),
-        };
-        index.load_searchers()?;
-        Ok(index)
+            executor: Arc::new(Executor::single_thread()),
+            inventory,
+        }
     }
 
     /// Accessor for the tokenizer manager.
@@ -142,11 +303,11 @@ impl Index {
     }
 
     /// Helper to access the tokenizer associated to a specific field.
-    pub fn tokenizer_for_field(&self, field: Field) -> Result<Box<BoxedTokenizer>> {
+    pub fn tokenizer_for_field(&self, field: Field) -> crate::Result<TextAnalyzer> {
         let field_entry = self.schema.get_field_entry(field);
         let field_type = field_entry.field_type();
         let tokenizer_manager: &TokenizerManager = self.tokenizers();
-        let tokenizer_name_opt: Option<Box<BoxedTokenizer>> = match field_type {
+        let tokenizer_name_opt: Option<TextAnalyzer> = match field_type {
             FieldType::Str(text_options) => text_options
                 .get_indexing_options()
                 .map(|text_indexing_options| text_indexing_options.tokenizer().to_string())
@@ -162,23 +323,59 @@ impl Index {
         }
     }
 
+    /// Create a default `IndexReader` for the given index.
+    ///
+    /// See [`Index.reader_builder()`](#method.reader_builder).
+    pub fn reader(&self) -> crate::Result<IndexReader> {
+        self.reader_builder().try_into()
+    }
+
+    /// Create a `IndexReader` for the given index.
+    ///
+    /// Most project should create at most one reader for a given index.
+    /// This method is typically called only once per `Index` instance,
+    /// over the lifetime of most problem.
+    pub fn reader_builder(&self) -> IndexReaderBuilder {
+        IndexReaderBuilder::new(self.clone())
+    }
+
     /// Opens a new directory from an index path.
     #[cfg(feature = "mmap")]
-    pub fn open_in_dir<P: AsRef<Path>>(directory_path: P) -> Result<Index> {
+    pub fn open_in_dir<P: AsRef<Path>>(directory_path: P) -> crate::Result<Index> {
         let mmap_directory = MmapDirectory::open(directory_path)?;
         Index::open(mmap_directory)
     }
 
+    /// Returns the list of the segment metas tracked by the index.
+    ///
+    /// Such segments can of course be part of the index,
+    /// but also they could be segments being currently built or in the middle of a merge
+    /// operation.
+    pub(crate) fn list_all_segment_metas(&self) -> Vec<SegmentMeta> {
+        self.inventory.all()
+    }
+
+    /// Creates a new segment_meta (Advanced user only).
+    ///
+    /// As long as the `SegmentMeta` lives, the files associated with the
+    /// `SegmentMeta` are guaranteed to not be garbage collected, regardless of
+    /// whether the segment is recorded as part of the index or not.
+    pub fn new_segment_meta(&self, segment_id: SegmentId, max_doc: u32) -> SegmentMeta {
+        self.inventory.new_segment_meta(segment_id, max_doc)
+    }
+
     /// Open the index using the provided directory
-    pub fn open<D: Directory>(directory: D) -> Result<Index> {
-        let directory = ManagedDirectory::new(directory)?;
-        let metas = load_metas(&directory)?;
-        Index::create_from_metas(directory, &metas)
+    pub fn open<D: Directory>(directory: D) -> crate::Result<Index> {
+        let directory = ManagedDirectory::wrap(directory)?;
+        let inventory = SegmentMetaInventory::default();
+        let metas = load_metas(&directory, &inventory)?;
+        let index = Index::open_from_metas(directory, &metas, inventory);
+        Ok(index)
     }
 
     /// Reads the index meta file from the directory.
-    pub fn load_metas(&self) -> Result<IndexMeta> {
-        load_metas(self.directory())
+    pub fn load_metas(&self) -> crate::Result<IndexMeta> {
+        load_metas(self.directory(), &self.inventory)
     }
 
     /// Open a new index writer. Attempts to acquire a lockfile.
@@ -197,17 +394,32 @@ impl Index {
     /// Each thread will receive a budget of  `overall_heap_size_in_bytes / num_threads`.
     ///
     /// # Errors
-    /// If the lockfile already exists, returns `Error::FileAlreadyExists`.
+    /// If the lockfile already exists, returns `Error::DirectoryLockBusy` or an `Error::IoError`.
+    ///
     /// # Panics
     /// If the heap size per thread is too small, panics.
     pub fn writer_with_num_threads(
         &self,
         num_threads: usize,
         overall_heap_size_in_bytes: usize,
-    ) -> Result<IndexWriter> {
-        let directory_lock = LockType::IndexWriterLock.acquire_lock(&self.directory)?;
+    ) -> crate::Result<IndexWriter> {
+        let directory_lock = self
+            .directory
+            .acquire_lock(&INDEX_WRITER_LOCK)
+            .map_err(|err| {
+                TantivyError::LockFailure(
+                    err,
+                    Some(
+                        "Failed to acquire index lock. If you are using \
+                         a regular directory, this means there is already an \
+                         `IndexWriter` working on this `Directory`, in this process \
+                         or in a different process."
+                            .to_string(),
+                    ),
+                )
+            })?;
         let heap_size_in_bytes_per_thread = overall_heap_size_in_bytes / num_threads;
-        open_index_writer(
+        IndexWriter::new(
             self,
             num_threads,
             heap_size_in_bytes_per_thread,
@@ -215,9 +427,19 @@ impl Index {
         )
     }
 
+    /// Helper to create an index writer for tests.
+    ///
+    /// That index writer only simply has a single thread and a heap of 10 MB.
+    /// Using a single thread gives us a deterministic allocation of DocId.
+    #[cfg(test)]
+    pub fn writer_for_tests(&self) -> crate::Result<IndexWriter> {
+        self.writer_with_num_threads(1, 10_000_000)
+    }
+
     /// Creates a multithreaded writer
     ///
-    /// Tantivy will automatically define the number of threads to use.
+    /// Tantivy will automatically define the number of threads to use, but
+    /// no more than [`MAX_NUM_THREAD`] threads.
     /// `overall_heap_size_in_bytes` is the total target memory usage that will be split
     /// between a given number of threads.
     ///
@@ -225,8 +447,8 @@ impl Index {
     /// If the lockfile already exists, returns `Error::FileAlreadyExists`.
     /// # Panics
     /// If the heap size per thread is too small, panics.
-    pub fn writer(&self, overall_heap_size_in_bytes: usize) -> Result<IndexWriter> {
-        let mut num_threads = num_cpus::get();
+    pub fn writer(&self, overall_heap_size_in_bytes: usize) -> crate::Result<IndexWriter> {
+        let mut num_threads = std::cmp::min(num_cpus::get(), MAX_NUM_THREAD);
         let heap_size_in_bytes_per_thread = overall_heap_size_in_bytes / num_threads;
         if heap_size_in_bytes_per_thread < HEAP_SIZE_MIN {
             num_threads = (overall_heap_size_in_bytes / HEAP_SIZE_MIN).max(1);
@@ -234,6 +456,18 @@ impl Index {
         self.writer_with_num_threads(num_threads, overall_heap_size_in_bytes)
     }
 
+    /// Accessor to the index settings
+    ///
+    pub fn settings(&self) -> &IndexSettings {
+        &self.settings
+    }
+
+    /// Accessor to the index settings
+    ///
+    pub fn settings_mut(&mut self) -> &mut IndexSettings {
+        &mut self.settings
+    }
+
     /// Accessor to the index schema
     ///
     /// The schema is actually cloned.
@@ -242,7 +476,7 @@ impl Index {
     }
 
     /// Returns the list of segments that are searchable
-    pub fn searchable_segments(&self) -> Result<Vec<Segment>> {
+    pub fn searchable_segments(&self) -> crate::Result<Vec<Segment>> {
         Ok(self
             .searchable_segment_metas()?
             .into_iter()
@@ -252,12 +486,14 @@ impl Index {
 
     #[doc(hidden)]
     pub fn segment(&self, segment_meta: SegmentMeta) -> Segment {
-        create_segment(self.clone(), segment_meta)
+        Segment::for_index(self.clone(), segment_meta)
     }
 
     /// Creates a new segment.
     pub fn new_segment(&self) -> Segment {
-        let segment_meta = SegmentMeta::new(SegmentId::generate_random(), 0);
+        let segment_meta = self
+            .inventory
+            .new_segment_meta(SegmentId::generate_random(), 0);
         self.segment(segment_meta)
     }
 
@@ -273,95 +509,62 @@ impl Index {
 
     /// Reads the meta.json and returns the list of
     /// `SegmentMeta` from the last commit.
-    pub fn searchable_segment_metas(&self) -> Result<Vec<SegmentMeta>> {
+    pub fn searchable_segment_metas(&self) -> crate::Result<Vec<SegmentMeta>> {
         Ok(self.load_metas()?.segments)
     }
 
     /// Returns the list of segment ids that are searchable.
-    pub fn searchable_segment_ids(&self) -> Result<Vec<SegmentId>> {
+    pub fn searchable_segment_ids(&self) -> crate::Result<Vec<SegmentId>> {
         Ok(self
             .searchable_segment_metas()?
             .iter()
-            .map(|segment_meta| segment_meta.id())
+            .map(SegmentMeta::id)
             .collect())
     }
 
-    /// Sets the number of searchers to use
-    ///
-    /// Only works after the next call to `load_searchers`
-    pub fn set_num_searchers(&mut self, num_searchers: usize) {
-        self.num_searchers.store(num_searchers, Ordering::Release);
-    }
-
-    /// Update searchers so that they reflect the state of the last
-    /// `.commit()`.
-    ///
-    /// If indexing happens in the same process as searching,
-    /// you most likely want to call `.load_searchers()` right after each
-    /// successful call to `.commit()`.
-    ///
-    /// If indexing and searching happen in different processes, the way to
-    /// get the freshest `index` at all time, is to watch `meta.json` and
-    /// call `load_searchers` whenever a changes happen.
-    pub fn load_searchers(&self) -> Result<()> {
-        let _meta_lock = LockType::MetaLock.acquire_lock(self.directory())?;
-        let searchable_segments = self.searchable_segments()?;
-        let segment_readers: Vec<SegmentReader> = searchable_segments
+    /// Returns the set of corrupted files
+    pub fn validate_checksum(&self) -> crate::Result<HashSet<PathBuf>> {
+        let managed_files = self.directory.list_managed_files();
+        let active_segments_files: HashSet<PathBuf> = self
+            .searchable_segment_metas()?
             .iter()
-            .map(SegmentReader::open)
-            .collect::<Result<_>>()?;
-        let schema = self.schema();
-        let num_searchers: usize = self.num_searchers.load(Ordering::Acquire);
-        let searchers = (0..num_searchers)
-            .map(|_| Searcher::new(schema.clone(), self.clone(), segment_readers.clone()))
+            .flat_map(|segment_meta| segment_meta.list_files())
             .collect();
-        self.searcher_pool.publish_new_generation(searchers);
-        Ok(())
-    }
+        let active_existing_files: HashSet<&PathBuf> =
+            active_segments_files.intersection(&managed_files).collect();
 
-    /// Returns a searcher
-    ///
-    /// This method should be called every single time a search
-    /// query is performed.
-    /// The searchers are taken from a pool of `num_searchers` searchers.
-    /// If no searcher is available
-    /// this may block.
-    ///
-    /// The same searcher must be used for a given query, as it ensures
-    /// the use of a consistent segment set.
-    pub fn searcher(&self) -> LeasedItem<Searcher> {
-        self.searcher_pool.acquire()
+        let mut damaged_files = HashSet::new();
+        for path in active_existing_files {
+            if !self.directory.validate_checksum(path)? {
+                damaged_files.insert((*path).clone());
+            }
+        }
+        Ok(damaged_files)
     }
 }
 
 impl fmt::Debug for Index {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         write!(f, "Index({:?})", self.directory)
     }
 }
 
-impl Clone for Index {
-    fn clone(&self) -> Index {
-        Index {
-            directory: self.directory.clone(),
-            schema: self.schema.clone(),
-            num_searchers: Arc::clone(&self.num_searchers),
-            searcher_pool: Arc::clone(&self.searcher_pool),
-            tokenizers: self.tokenizers.clone(),
-        }
-    }
-}
-
 #[cfg(test)]
 mod tests {
-    use schema::{Schema, SchemaBuilder, INT_INDEXED, TEXT};
-    use Index;
-    use directory::RAMDirectory;
+    use crate::schema::Field;
+    use crate::schema::{Schema, INDEXED, TEXT};
+    use crate::IndexReader;
+    use crate::ReloadPolicy;
+    use crate::{
+        directory::{RamDirectory, WatchCallback},
+        IndexSettings,
+    };
+    use crate::{Directory, Index};
 
     #[test]
     fn test_indexer_for_field() {
-        let mut schema_builder = SchemaBuilder::default();
-        let num_likes_field = schema_builder.add_u64_field("num_likes", INT_INDEXED);
+        let mut schema_builder = Schema::builder();
+        let num_likes_field = schema_builder.add_u64_field("num_likes", INDEXED);
         let body_field = schema_builder.add_text_field("body", TEXT);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
@@ -374,50 +577,230 @@ mod tests {
 
     #[test]
     fn test_index_exists() {
-        let directory = RAMDirectory::create();
-        assert!(!Index::exists(&directory));
-        assert!(Index::create(directory.clone(), throw_away_schema()).is_ok());
-        assert!(Index::exists(&directory));
+        let directory = RamDirectory::create();
+        assert!(!Index::exists(&directory).unwrap());
+        assert!(Index::create(
+            directory.clone(),
+            throw_away_schema(),
+            IndexSettings::default()
+        )
+        .is_ok());
+        assert!(Index::exists(&directory).unwrap());
     }
 
     #[test]
     fn open_or_create_should_create() {
-        let directory = RAMDirectory::create();
-        assert!(!Index::exists(&directory));
+        let directory = RamDirectory::create();
+        assert!(!Index::exists(&directory).unwrap());
         assert!(Index::open_or_create(directory.clone(), throw_away_schema()).is_ok());
-        assert!(Index::exists(&directory));
+        assert!(Index::exists(&directory).unwrap());
     }
 
-
     #[test]
     fn open_or_create_should_open() {
-        let directory = RAMDirectory::create();
-        assert!(Index::create(directory.clone(), throw_away_schema()).is_ok());
-        assert!(Index::exists(&directory));
+        let directory = RamDirectory::create();
+        assert!(Index::create(
+            directory.clone(),
+            throw_away_schema(),
+            IndexSettings::default()
+        )
+        .is_ok());
+        assert!(Index::exists(&directory).unwrap());
         assert!(Index::open_or_create(directory, throw_away_schema()).is_ok());
     }
 
     #[test]
     fn create_should_wipeoff_existing() {
-        let directory = RAMDirectory::create();
-        assert!(Index::create(directory.clone(), throw_away_schema()).is_ok());
-        assert!(Index::exists(&directory));
-        assert!(Index::create(directory.clone(), SchemaBuilder::default().build()).is_ok());
+        let directory = RamDirectory::create();
+        assert!(Index::create(
+            directory.clone(),
+            throw_away_schema(),
+            IndexSettings::default()
+        )
+        .is_ok());
+        assert!(Index::exists(&directory).unwrap());
+        assert!(Index::create(
+            directory,
+            Schema::builder().build(),
+            IndexSettings::default()
+        )
+        .is_ok());
     }
 
     #[test]
     fn open_or_create_exists_but_schema_does_not_match() {
-        let directory = RAMDirectory::create();
-        assert!(Index::create(directory.clone(), throw_away_schema()).is_ok());
-        assert!(Index::exists(&directory));
+        let directory = RamDirectory::create();
+        assert!(Index::create(
+            directory.clone(),
+            throw_away_schema(),
+            IndexSettings::default()
+        )
+        .is_ok());
+        assert!(Index::exists(&directory).unwrap());
         assert!(Index::open_or_create(directory.clone(), throw_away_schema()).is_ok());
-        let err = Index::open_or_create(directory, SchemaBuilder::default().build());
-        assert_eq!(format!("{:?}", err.unwrap_err()), "SchemaError(\"An index exists but the schema does not match.\")");
+        let err = Index::open_or_create(directory, Schema::builder().build());
+        assert_eq!(
+            format!("{:?}", err.unwrap_err()),
+            "SchemaError(\"An index exists but the schema does not match.\")"
+        );
     }
 
     fn throw_away_schema() -> Schema {
-        let mut schema_builder = SchemaBuilder::default();
-        let _ = schema_builder.add_u64_field("num_likes", INT_INDEXED);
+        let mut schema_builder = Schema::builder();
+        let _ = schema_builder.add_u64_field("num_likes", INDEXED);
         schema_builder.build()
     }
+
+    #[test]
+    fn test_index_on_commit_reload_policy() {
+        let schema = throw_away_schema();
+        let field = schema.get_field("num_likes").unwrap();
+        let index = Index::create_in_ram(schema);
+        let reader = index
+            .reader_builder()
+            .reload_policy(ReloadPolicy::OnCommit)
+            .try_into()
+            .unwrap();
+        assert_eq!(reader.searcher().num_docs(), 0);
+        test_index_on_commit_reload_policy_aux(field, &index, &reader);
+    }
+
+    #[cfg(feature = "mmap")]
+    mod mmap_specific {
+
+        use super::*;
+        use crate::Directory;
+        use std::path::PathBuf;
+        use tempfile::TempDir;
+
+        #[test]
+        fn test_index_on_commit_reload_policy_mmap() {
+            let schema = throw_away_schema();
+            let field = schema.get_field("num_likes").unwrap();
+            let tempdir = TempDir::new().unwrap();
+            let tempdir_path = PathBuf::from(tempdir.path());
+            let index = Index::create_in_dir(&tempdir_path, schema).unwrap();
+            let reader = index
+                .reader_builder()
+                .reload_policy(ReloadPolicy::OnCommit)
+                .try_into()
+                .unwrap();
+            assert_eq!(reader.searcher().num_docs(), 0);
+            test_index_on_commit_reload_policy_aux(field, &index, &reader);
+        }
+
+        #[test]
+        fn test_index_manual_policy_mmap() -> crate::Result<()> {
+            let schema = throw_away_schema();
+            let field = schema.get_field("num_likes").unwrap();
+            let mut index = Index::create_from_tempdir(schema)?;
+            let mut writer = index.writer_for_tests()?;
+            writer.commit()?;
+            let reader = index
+                .reader_builder()
+                .reload_policy(ReloadPolicy::Manual)
+                .try_into()?;
+            assert_eq!(reader.searcher().num_docs(), 0);
+            writer.add_document(doc!(field=>1u64));
+            let (sender, receiver) = crossbeam::channel::unbounded();
+            let _handle = index.directory_mut().watch(WatchCallback::new(move || {
+                let _ = sender.send(());
+            }));
+            writer.commit()?;
+            assert!(receiver.recv().is_ok());
+            assert_eq!(reader.searcher().num_docs(), 0);
+            reader.reload()?;
+            assert_eq!(reader.searcher().num_docs(), 1);
+            Ok(())
+        }
+
+        #[test]
+        fn test_index_on_commit_reload_policy_different_directories() {
+            let schema = throw_away_schema();
+            let field = schema.get_field("num_likes").unwrap();
+            let tempdir = TempDir::new().unwrap();
+            let tempdir_path = PathBuf::from(tempdir.path());
+            let write_index = Index::create_in_dir(&tempdir_path, schema).unwrap();
+            let read_index = Index::open_in_dir(&tempdir_path).unwrap();
+            let reader = read_index
+                .reader_builder()
+                .reload_policy(ReloadPolicy::OnCommit)
+                .try_into()
+                .unwrap();
+            assert_eq!(reader.searcher().num_docs(), 0);
+            test_index_on_commit_reload_policy_aux(field, &write_index, &reader);
+        }
+    }
+    fn test_index_on_commit_reload_policy_aux(field: Field, index: &Index, reader: &IndexReader) {
+        let mut reader_index = reader.index();
+        let (sender, receiver) = crossbeam::channel::unbounded();
+        let _watch_handle = reader_index
+            .directory_mut()
+            .watch(WatchCallback::new(move || {
+                let _ = sender.send(());
+            }));
+        let mut writer = index.writer_for_tests().unwrap();
+        assert_eq!(reader.searcher().num_docs(), 0);
+        writer.add_document(doc!(field=>1u64));
+        writer.commit().unwrap();
+        // We need a loop here because it is possible for notify to send more than
+        // one modify event. It was observed on CI on MacOS.
+        loop {
+            assert!(receiver.recv().is_ok());
+            if reader.searcher().num_docs() == 1 {
+                break;
+            }
+        }
+        writer.add_document(doc!(field=>2u64));
+        writer.commit().unwrap();
+        // ... Same as above
+        loop {
+            assert!(receiver.recv().is_ok());
+            if reader.searcher().num_docs() == 2 {
+                break;
+            }
+        }
+    }
+
+    // This test will not pass on windows, because windows
+    // prevent deleting files that are MMapped.
+    #[cfg(not(target_os = "windows"))]
+    #[test]
+    fn garbage_collect_works_as_intended() {
+        let directory = RamDirectory::create();
+        let schema = throw_away_schema();
+        let field = schema.get_field("num_likes").unwrap();
+        let index = Index::create(directory.clone(), schema, IndexSettings::default()).unwrap();
+
+        let mut writer = index.writer_with_num_threads(8, 24_000_000).unwrap();
+        for i in 0u64..8_000u64 {
+            writer.add_document(doc!(field => i));
+        }
+        let (sender, receiver) = crossbeam::channel::unbounded();
+        let _handle = directory.watch(WatchCallback::new(move || {
+            let _ = sender.send(());
+        }));
+        writer.commit().unwrap();
+        let mem_right_after_commit = directory.total_mem_usage();
+        assert!(receiver.recv().is_ok());
+        let reader = index
+            .reader_builder()
+            .reload_policy(ReloadPolicy::Manual)
+            .try_into()
+            .unwrap();
+
+        assert_eq!(reader.searcher().num_docs(), 8_000);
+        writer.wait_merging_threads().unwrap();
+        let mem_right_after_merge_finished = directory.total_mem_usage();
+
+        reader.reload().unwrap();
+        let searcher = reader.searcher();
+        assert_eq!(searcher.num_docs(), 8_000);
+        assert!(
+            mem_right_after_merge_finished < mem_right_after_commit,
+            "(mem after merge){} is expected < (mem before merge){}",
+            mem_right_after_merge_finished,
+            mem_right_after_commit
+        );
+    }
 }

src/core/index_meta.rs

@@ -1,7 +1,275 @@
-use core::SegmentMeta;
-use schema::Schema;
-use serde_json;
-use std::fmt;
+use super::SegmentComponent;
+use crate::schema::Schema;
+use crate::Opstamp;
+use crate::{core::SegmentId, store::Compressor};
+use census::{Inventory, TrackedObject};
+use serde::{Deserialize, Serialize};
+use std::path::PathBuf;
+use std::{collections::HashSet, sync::atomic::AtomicBool};
+use std::{fmt, sync::Arc};
+
+#[derive(Clone, Debug, Serialize, Deserialize)]
+struct DeleteMeta {
+    num_deleted_docs: u32,
+    opstamp: Opstamp,
+}
+
+#[derive(Clone, Default)]
+pub struct SegmentMetaInventory {
+    inventory: Inventory<InnerSegmentMeta>,
+}
+
+impl SegmentMetaInventory {
+    /// Lists all living `SegmentMeta` object at the time of the call.
+    pub fn all(&self) -> Vec<SegmentMeta> {
+        self.inventory
+            .list()
+            .into_iter()
+            .map(SegmentMeta::from)
+            .collect::<Vec<_>>()
+    }
+
+    pub fn new_segment_meta(&self, segment_id: SegmentId, max_doc: u32) -> SegmentMeta {
+        let inner = InnerSegmentMeta {
+            segment_id,
+            max_doc,
+            include_temp_doc_store: Arc::new(AtomicBool::new(true)),
+            deletes: None,
+        };
+        SegmentMeta::from(self.inventory.track(inner))
+    }
+}
+
+/// `SegmentMeta` contains simple meta information about a segment.
+///
+/// For instance the number of docs it contains,
+/// how many are deleted, etc.
+#[derive(Clone)]
+pub struct SegmentMeta {
+    tracked: TrackedObject<InnerSegmentMeta>,
+}
+
+impl fmt::Debug for SegmentMeta {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> Result<(), fmt::Error> {
+        self.tracked.fmt(f)
+    }
+}
+
+impl serde::Serialize for SegmentMeta {
+    fn serialize<S>(
+        &self,
+        serializer: S,
+    ) -> Result<<S as serde::Serializer>::Ok, <S as serde::Serializer>::Error>
+    where
+        S: serde::Serializer,
+    {
+        self.tracked.serialize(serializer)
+    }
+}
+
+impl From<TrackedObject<InnerSegmentMeta>> for SegmentMeta {
+    fn from(tracked: TrackedObject<InnerSegmentMeta>) -> SegmentMeta {
+        SegmentMeta { tracked }
+    }
+}
+
+impl SegmentMeta {
+    // Creates a new `SegmentMeta` object.
+
+    /// Returns the segment id.
+    pub fn id(&self) -> SegmentId {
+        self.tracked.segment_id
+    }
+
+    /// Removes the Component::TempStore from the alive list and
+    /// therefore marks the temp docstore file to be deleted by
+    /// the garbage collection.
+    pub fn untrack_temp_docstore(&self) {
+        self.tracked
+            .include_temp_doc_store
+            .store(false, std::sync::atomic::Ordering::Relaxed);
+    }
+
+    /// Returns the number of deleted documents.
+    pub fn num_deleted_docs(&self) -> u32 {
+        self.tracked
+            .deletes
+            .as_ref()
+            .map(|delete_meta| delete_meta.num_deleted_docs)
+            .unwrap_or(0u32)
+    }
+
+    /// Returns the list of files that
+    /// are required for the segment meta.
+    /// Note: Some of the returned files may not exist depending on the state of the segment.
+    ///
+    /// This is useful as the way tantivy removes files
+    /// is by removing all files that have been created by tantivy
+    /// and are not used by any segment anymore.
+    pub fn list_files(&self) -> HashSet<PathBuf> {
+        if self
+            .tracked
+            .include_temp_doc_store
+            .load(std::sync::atomic::Ordering::Relaxed)
+        {
+            SegmentComponent::iterator()
+                .map(|component| self.relative_path(*component))
+                .collect::<HashSet<PathBuf>>()
+        } else {
+            SegmentComponent::iterator()
+                .filter(|comp| *comp != &SegmentComponent::TempStore)
+                .map(|component| self.relative_path(*component))
+                .collect::<HashSet<PathBuf>>()
+        }
+    }
+
+    /// Returns the relative path of a component of our segment.
+    ///
+    /// It just joins the segment id with the extension
+    /// associated to a segment component.
+    pub fn relative_path(&self, component: SegmentComponent) -> PathBuf {
+        let mut path = self.id().uuid_string();
+        path.push_str(&*match component {
+            SegmentComponent::Postings => ".idx".to_string(),
+            SegmentComponent::Positions => ".pos".to_string(),
+            SegmentComponent::Terms => ".term".to_string(),
+            SegmentComponent::Store => ".store".to_string(),
+            SegmentComponent::TempStore => ".store.temp".to_string(),
+            SegmentComponent::FastFields => ".fast".to_string(),
+            SegmentComponent::FieldNorms => ".fieldnorm".to_string(),
+            SegmentComponent::Delete => format!(".{}.del", self.delete_opstamp().unwrap_or(0)),
+        });
+        PathBuf::from(path)
+    }
+
+    /// Return the highest doc id + 1
+    ///
+    /// If there are no deletes, then num_docs = max_docs
+    /// and all the doc ids contains in this segment
+    /// are exactly (0..max_doc).
+    pub fn max_doc(&self) -> u32 {
+        self.tracked.max_doc
+    }
+
+    /// Return the number of documents in the segment.
+    pub fn num_docs(&self) -> u32 {
+        self.max_doc() - self.num_deleted_docs()
+    }
+
+    /// Returns the `Opstamp` of the last delete operation
+    /// taken in account in this segment.
+    pub fn delete_opstamp(&self) -> Option<Opstamp> {
+        self.tracked
+            .deletes
+            .as_ref()
+            .map(|delete_meta| delete_meta.opstamp)
+    }
+
+    /// Returns true iff the segment meta contains
+    /// delete information.
+    pub fn has_deletes(&self) -> bool {
+        self.num_deleted_docs() > 0
+    }
+
+    /// Updates the max_doc value from the `SegmentMeta`.
+    ///
+    /// This method is only used when updating `max_doc` from 0
+    /// as we finalize a fresh new segment.
+    pub(crate) fn with_max_doc(self, max_doc: u32) -> SegmentMeta {
+        assert_eq!(self.tracked.max_doc, 0);
+        assert!(self.tracked.deletes.is_none());
+        let tracked = self.tracked.map(move |inner_meta| InnerSegmentMeta {
+            segment_id: inner_meta.segment_id,
+            max_doc,
+            deletes: None,
+            include_temp_doc_store: Arc::new(AtomicBool::new(true)),
+        });
+        SegmentMeta { tracked }
+    }
+
+    #[doc(hidden)]
+    pub fn with_delete_meta(self, num_deleted_docs: u32, opstamp: Opstamp) -> SegmentMeta {
+        let delete_meta = DeleteMeta {
+            num_deleted_docs,
+            opstamp,
+        };
+        let tracked = self.tracked.map(move |inner_meta| InnerSegmentMeta {
+            segment_id: inner_meta.segment_id,
+            max_doc: inner_meta.max_doc,
+            include_temp_doc_store: Arc::new(AtomicBool::new(true)),
+            deletes: Some(delete_meta),
+        });
+        SegmentMeta { tracked }
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+struct InnerSegmentMeta {
+    segment_id: SegmentId,
+    max_doc: u32,
+    deletes: Option<DeleteMeta>,
+    /// If you want to avoid the SegmentComponent::TempStore file to be covered by
+    /// garbage collection and deleted, set this to true. This is used during merge.
+    #[serde(skip)]
+    #[serde(default = "default_temp_store")]
+    pub(crate) include_temp_doc_store: Arc<AtomicBool>,
+}
+fn default_temp_store() -> Arc<AtomicBool> {
+    Arc::new(AtomicBool::new(false))
+}
+
+impl InnerSegmentMeta {
+    pub fn track(self, inventory: &SegmentMetaInventory) -> SegmentMeta {
+        SegmentMeta {
+            tracked: inventory.inventory.track(self),
+        }
+    }
+}
+
+/// Search Index Settings.
+///
+/// Contains settings which are applied on the whole
+/// index, like presort documents.
+#[derive(Clone, Default, Serialize, Deserialize, Eq, PartialEq)]
+pub struct IndexSettings {
+    /// Sorts the documents by information
+    /// provided in `IndexSortByField`
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub sort_by_field: Option<IndexSortByField>,
+    /// The `Compressor` used to compress the doc store.
+    #[serde(default)]
+    pub docstore_compression: Compressor,
+}
+/// Settings to presort the documents in an index
+///
+/// Presorting documents can greatly performance
+/// in some scenarios, by applying top n
+/// optimizations.
+#[derive(Clone, Serialize, Deserialize, Eq, PartialEq)]
+pub struct IndexSortByField {
+    /// The field to sort the documents by
+    pub field: String,
+    /// The order to sort the documents by
+    pub order: Order,
+}
+/// The order to sort by
+#[derive(Clone, Serialize, Deserialize, Eq, PartialEq)]
+pub enum Order {
+    /// Ascending Order
+    Asc,
+    /// Descending Order
+    Desc,
+}
+impl Order {
+    /// return if the Order is ascending
+    pub fn is_asc(&self) -> bool {
+        self == &Order::Asc
+    }
+    /// return if the Order is descending
+    pub fn is_desc(&self) -> bool {
+        self == &Order::Desc
+    }
+}
 
 /// Meta information about the `Index`.
 ///
@@ -11,28 +279,80 @@ use std::fmt;
 /// * the index `docstamp`
 /// * the schema
 ///
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Clone, Serialize)]
 pub struct IndexMeta {
+    /// `IndexSettings` to configure index options.
+    #[serde(default)]
+    pub index_settings: IndexSettings,
+    /// List of `SegmentMeta` informations associated to each finalized segment of the index.
     pub segments: Vec<SegmentMeta>,
+    /// Index `Schema`
     pub schema: Schema,
-    pub opstamp: u64,
+    /// Opstamp associated to the last `commit` operation.
+    pub opstamp: Opstamp,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    /// Payload associated to the last commit.
+    ///
+    /// Upon commit, clients can optionally add a small `String` payload to their commit
+    /// to help identify this commit.
+    /// This payload is entirely unused by tantivy.
+    pub payload: Option<String>,
+}
+
+#[derive(Deserialize)]
+struct UntrackedIndexMeta {
+    pub segments: Vec<InnerSegmentMeta>,
+    #[serde(default)]
+    pub index_settings: IndexSettings,
+    pub schema: Schema,
+    pub opstamp: Opstamp,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub payload: Option<String>,
 }
 
+impl UntrackedIndexMeta {
+    pub fn track(self, inventory: &SegmentMetaInventory) -> IndexMeta {
+        IndexMeta {
+            index_settings: self.index_settings,
+            segments: self
+                .segments
+                .into_iter()
+                .map(|inner_seg_meta| inner_seg_meta.track(inventory))
+                .collect::<Vec<SegmentMeta>>(),
+            schema: self.schema,
+            opstamp: self.opstamp,
+            payload: self.payload,
+        }
+    }
+}
+
 impl IndexMeta {
+    /// Create an `IndexMeta` object representing a brand new `Index`
+    /// with the given index.
+    ///
+    /// This new index does not contains any segments.
+    /// Opstamp will the value `0u64`.
     pub fn with_schema(schema: Schema) -> IndexMeta {
         IndexMeta {
+            index_settings: IndexSettings::default(),
             segments: vec![],
             schema,
             opstamp: 0u64,
             payload: None,
         }
     }
+
+    pub(crate) fn deserialize(
+        meta_json: &str,
+        inventory: &SegmentMetaInventory,
+    ) -> serde_json::Result<IndexMeta> {
+        let untracked_meta_json: UntrackedIndexMeta = serde_json::from_str(meta_json)?;
+        Ok(untracked_meta_json.track(inventory))
+    }
 }
 
 impl fmt::Debug for IndexMeta {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         write!(
             f,
             "{}",
@@ -46,23 +366,35 @@ impl fmt::Debug for IndexMeta {
 mod tests {
 
     use super::IndexMeta;
-    use schema::{SchemaBuilder, TEXT};
-    use serde_json;
+    use crate::{
+        schema::{Schema, TEXT},
+        IndexSettings, IndexSortByField, Order,
+    };
 
     #[test]
     fn test_serialize_metas() {
         let schema = {
-            let mut schema_builder = SchemaBuilder::new();
+            let mut schema_builder = Schema::builder();
             schema_builder.add_text_field("text", TEXT);
             schema_builder.build()
         };
         let index_metas = IndexMeta {
+            index_settings: IndexSettings {
+                sort_by_field: Some(IndexSortByField {
+                    field: "text".to_string(),
+                    order: Order::Asc,
+                }),
+                ..Default::default()
+            },
             segments: Vec::new(),
             schema,
             opstamp: 0u64,
             payload: None,
         };
         let json = serde_json::ser::to_string(&index_metas).expect("serialization failed");
-        assert_eq!(json, r#"{"segments":[],"schema":[{"name":"text","type":"text","options":{"indexing":{"record":"position","tokenizer":"default"},"stored":false}}],"opstamp":0}"#);
+        assert_eq!(
+            json,
+            r#"{"index_settings":{"sort_by_field":{"field":"text","order":"Asc"},"docstore_compression":"lz4"},"segments":[],"schema":[{"name":"text","type":"text","options":{"indexing":{"record":"position","tokenizer":"default"},"stored":false}}],"opstamp":0}"#
+        );
     }
 }

src/core/inverted_index_reader.rs

@@ -1,13 +1,13 @@
+use std::io;
+
+use crate::directory::FileSlice;
+use crate::positions::PositionReader;
+use crate::postings::TermInfo;
+use crate::postings::{BlockSegmentPostings, SegmentPostings};
+use crate::schema::IndexRecordOption;
+use crate::schema::Term;
+use crate::termdict::TermDictionary;
 use common::BinarySerializable;
-use directory::ReadOnlySource;
-use owned_read::OwnedRead;
-use positions::PositionReader;
-use postings::TermInfo;
-use postings::{BlockSegmentPostings, SegmentPostings};
-use schema::FieldType;
-use schema::IndexRecordOption;
-use schema::Term;
-use termdict::TermDictionary;
 
 /// The inverted index reader is in charge of accessing
 /// the inverted index associated to a specific field.
@@ -16,7 +16,7 @@ use termdict::TermDictionary;
 ///
 /// It is safe to delete the segment associated to
 /// an `InvertedIndexReader`. As long as it is open,
-/// the `ReadOnlySource` it is relying on should
+/// the `FileSlice` it is relying on should
 /// stay available.
 ///
 ///
@@ -24,56 +24,45 @@ use termdict::TermDictionary;
 /// the `SegmentReader`'s [`.inverted_index(...)`] method
 pub struct InvertedIndexReader {
     termdict: TermDictionary,
-    postings_source: ReadOnlySource,
-    positions_source: ReadOnlySource,
-    positions_idx_source: ReadOnlySource,
+    postings_file_slice: FileSlice,
+    positions_file_slice: FileSlice,
     record_option: IndexRecordOption,
     total_num_tokens: u64,
 }
 
 impl InvertedIndexReader {
-    #[cfg_attr(
-        feature = "cargo-clippy",
-        allow(clippy::needless_pass_by_value)
-    )] // for symetry
+    #[cfg_attr(feature = "cargo-clippy", allow(clippy::needless_pass_by_value))] // for symmetry
     pub(crate) fn new(
         termdict: TermDictionary,
-        postings_source: ReadOnlySource,
-        positions_source: ReadOnlySource,
-        positions_idx_source: ReadOnlySource,
+        postings_file_slice: FileSlice,
+        positions_file_slice: FileSlice,
         record_option: IndexRecordOption,
-    ) -> InvertedIndexReader {
-        let total_num_tokens_data = postings_source.slice(0, 8);
-        let mut total_num_tokens_cursor = total_num_tokens_data.as_slice();
-        let total_num_tokens = u64::deserialize(&mut total_num_tokens_cursor).unwrap_or(0u64);
-        InvertedIndexReader {
+    ) -> io::Result<InvertedIndexReader> {
+        let (total_num_tokens_slice, postings_body) = postings_file_slice.split(8);
+        let total_num_tokens = u64::deserialize(&mut total_num_tokens_slice.read_bytes()?)?;
+        Ok(InvertedIndexReader {
             termdict,
-            postings_source: postings_source.slice_from(8),
-            positions_source,
-            positions_idx_source,
+            postings_file_slice: postings_body,
+            positions_file_slice,
             record_option,
             total_num_tokens,
-        }
+        })
     }
 
     /// Creates an empty `InvertedIndexReader` object, which
     /// contains no terms at all.
-    pub fn empty(field_type: &FieldType) -> InvertedIndexReader {
-        let record_option = field_type
-            .get_index_record_option()
-            .unwrap_or(IndexRecordOption::Basic);
+    pub fn empty(record_option: IndexRecordOption) -> InvertedIndexReader {
         InvertedIndexReader {
-            termdict: TermDictionary::empty(&field_type),
-            postings_source: ReadOnlySource::empty(),
-            positions_source: ReadOnlySource::empty(),
-            positions_idx_source: ReadOnlySource::empty(),
+            termdict: TermDictionary::empty(),
+            postings_file_slice: FileSlice::empty(),
+            positions_file_slice: FileSlice::empty(),
             record_option,
             total_num_tokens: 0u64,
         }
     }
 
     /// Returns the term info associated with the term.
-    pub fn get_term_info(&self, term: &Term) -> Option<TermInfo> {
+    pub fn get_term_info(&self, term: &Term) -> io::Result<Option<TermInfo>> {
         self.termdict.get(term.value_bytes())
     }
 
@@ -96,12 +85,12 @@ impl InvertedIndexReader {
         &self,
         term_info: &TermInfo,
         block_postings: &mut BlockSegmentPostings,
-    ) {
-        let offset = term_info.postings_offset as usize;
-        let end_source = self.postings_source.len();
-        let postings_slice = self.postings_source.slice(offset, end_source);
-        let postings_reader = OwnedRead::new(postings_slice);
-        block_postings.reset(term_info.doc_freq, postings_reader);
+    ) -> io::Result<()> {
+        let postings_slice = self
+            .postings_file_slice
+            .slice(term_info.postings_range.clone());
+        block_postings.reset(term_info.doc_freq, postings_slice.read_bytes()?);
+        Ok(())
     }
 
     /// Returns a block postings given a `Term`.
@@ -112,9 +101,10 @@ impl InvertedIndexReader {
         &self,
         term: &Term,
         option: IndexRecordOption,
-    ) -> Option<BlockSegmentPostings> {
-        self.get_term_info(term)
+    ) -> io::Result<Option<BlockSegmentPostings>> {
+        self.get_term_info(term)?
             .map(move |term_info| self.read_block_postings_from_terminfo(&term_info, option))
+            .transpose()
     }
 
     /// Returns a block postings given a `term_info`.
@@ -125,12 +115,13 @@ impl InvertedIndexReader {
         &self,
         term_info: &TermInfo,
         requested_option: IndexRecordOption,
-    ) -> BlockSegmentPostings {
-        let offset = term_info.postings_offset as usize;
-        let postings_data = self.postings_source.slice_from(offset);
-        BlockSegmentPostings::from_data(
+    ) -> io::Result<BlockSegmentPostings> {
+        let postings_data = self
+            .postings_file_slice
+            .slice(term_info.postings_range.clone());
+        BlockSegmentPostings::open(
             term_info.doc_freq,
-            OwnedRead::new(postings_data),
+            postings_data,
             self.record_option,
             requested_option,
         )
@@ -144,20 +135,23 @@ impl InvertedIndexReader {
         &self,
         term_info: &TermInfo,
         option: IndexRecordOption,
-    ) -> SegmentPostings {
-        let block_postings = self.read_block_postings_from_terminfo(term_info, option);
-        let position_stream = {
+    ) -> io::Result<SegmentPostings> {
+        let block_postings = self.read_block_postings_from_terminfo(term_info, option)?;
+        let position_reader = {
             if option.has_positions() {
-                let position_reader = self.positions_source.clone();
-                let skip_reader = self.positions_idx_source.clone();
-                let position_reader =
-                    PositionReader::new(position_reader, skip_reader, term_info.positions_idx);
+                let positions_data = self
+                    .positions_file_slice
+                    .read_bytes_slice(term_info.positions_range.clone())?;
+                let position_reader = PositionReader::open(positions_data)?;
                 Some(position_reader)
             } else {
                 None
             }
         };
-        SegmentPostings::from_block_postings(block_postings, position_stream)
+        Ok(SegmentPostings::from_block_postings(
+            block_postings,
+            position_reader,
+        ))
     }
 
     /// Returns the total number of tokens recorded for all documents
@@ -176,24 +170,31 @@ impl InvertedIndexReader {
     /// For instance, requesting `IndexRecordOption::Freq` for a
     /// `TextIndexingOptions` that does not index position will return a `SegmentPostings`
     /// with `DocId`s and frequencies.
-    pub fn read_postings(&self, term: &Term, option: IndexRecordOption) -> Option<SegmentPostings> {
-        self.get_term_info(term)
+    pub fn read_postings(
+        &self,
+        term: &Term,
+        option: IndexRecordOption,
+    ) -> io::Result<Option<SegmentPostings>> {
+        self.get_term_info(term)?
             .map(move |term_info| self.read_postings_from_terminfo(&term_info, option))
+            .transpose()
     }
 
     pub(crate) fn read_postings_no_deletes(
         &self,
         term: &Term,
         option: IndexRecordOption,
-    ) -> Option<SegmentPostings> {
-        self.get_term_info(term)
+    ) -> io::Result<Option<SegmentPostings>> {
+        self.get_term_info(term)?
             .map(|term_info| self.read_postings_from_terminfo(&term_info, option))
+            .transpose()
     }
 
     /// Returns the number of documents containing the term.
-    pub fn doc_freq(&self, term: &Term) -> u32 {
-        self.get_term_info(term)
+    pub fn doc_freq(&self, term: &Term) -> io::Result<u32> {
+        Ok(self
+            .get_term_info(term)?
             .map(|term_info| term_info.doc_freq)
-            .unwrap_or(0u32)
+            .unwrap_or(0u32))
     }
 }

src/core/mod.rs

@@ -1,36 +1,35 @@
+mod executor;
 pub mod index;
 mod index_meta;
 mod inverted_index_reader;
-mod pool;
 pub mod searcher;
 mod segment;
 mod segment_component;
 mod segment_id;
-mod segment_meta;
 mod segment_reader;
 
-pub use self::index::Index;
-pub use self::index_meta::IndexMeta;
+pub use self::executor::Executor;
+pub use self::index::{Index, IndexBuilder};
+pub use self::index_meta::{
+    IndexMeta, IndexSettings, IndexSortByField, Order, SegmentMeta, SegmentMetaInventory,
+};
 pub use self::inverted_index_reader::InvertedIndexReader;
 pub use self::searcher::Searcher;
 pub use self::segment::Segment;
-pub use self::segment::SerializableSegment;
 pub use self::segment_component::SegmentComponent;
 pub use self::segment_id::SegmentId;
-pub use self::segment_meta::SegmentMeta;
 pub use self::segment_reader::SegmentReader;
 
-use std::path::PathBuf;
+use once_cell::sync::Lazy;
+use std::path::Path;
 
-lazy_static! {
-    /// The meta file contains all the information about the list of segments and the schema
-    /// of the index.
-    pub static ref META_FILEPATH: PathBuf = PathBuf::from("meta.json");
+/// The meta file contains all the information about the list of segments and the schema
+/// of the index.
+pub static META_FILEPATH: Lazy<&'static Path> = Lazy::new(|| Path::new("meta.json"));
 
-    /// The managed file contains a list of files that were created by the tantivy
-    /// and will therefore be garbage collected when they are deemed useless by tantivy.
-    ///
-    /// Removing this file is safe, but will prevent the garbage collection of all of the file that
-    /// are currently in the directory
-    pub static ref MANAGED_FILEPATH: PathBuf = PathBuf::from(".managed.json");
-}
+/// The managed file contains a list of files that were created by the tantivy
+/// and will therefore be garbage collected when they are deemed useless by tantivy.
+///
+/// Removing this file is safe, but will prevent the garbage collection of all of the file that
+/// are currently in the directory
+pub static MANAGED_FILEPATH: Lazy<&'static Path> = Lazy::new(|| Path::new(".managed.json"));

src/core/pool.rs

@@ -1,136 +0,0 @@
-use crossbeam::queue::MsQueue;
-use std::mem;
-use std::ops::{Deref, DerefMut};
-use std::sync::atomic::AtomicUsize;
-use std::sync::atomic::Ordering;
-use std::sync::Arc;
-
-pub struct GenerationItem<T> {
-    generation: usize,
-    item: T,
-}
-
-pub struct Pool<T> {
-    queue: Arc<MsQueue<GenerationItem<T>>>,
-    freshest_generation: AtomicUsize,
-    next_generation: AtomicUsize,
-}
-
-impl<T> Pool<T> {
-    pub fn new() -> Pool<T> {
-        let queue = Arc::new(MsQueue::new());
-        Pool {
-            queue,
-            freshest_generation: AtomicUsize::default(),
-            next_generation: AtomicUsize::default(),
-        }
-    }
-
-    pub fn publish_new_generation(&self, items: Vec<T>) {
-        let next_generation = self.next_generation.fetch_add(1, Ordering::SeqCst) + 1;
-        for item in items {
-            let gen_item = GenerationItem {
-                item,
-                generation: next_generation,
-            };
-            self.queue.push(gen_item);
-        }
-        self.advertise_generation(next_generation);
-    }
-
-    /// At the exit of this method,
-    /// - freshest_generation has a value greater or equal than generation
-    /// - freshest_generation has a value that has been advertised
-    /// - freshest_generation has)
-    fn advertise_generation(&self, generation: usize) {
-        // not optimal at all but the easiest to read proof.
-        loop {
-            let former_generation = self.freshest_generation.load(Ordering::Acquire);
-            if former_generation >= generation {
-                break;
-            }
-            self.freshest_generation.compare_and_swap(
-                former_generation,
-                generation,
-                Ordering::SeqCst,
-            );
-        }
-    }
-
-    fn generation(&self) -> usize {
-        self.freshest_generation.load(Ordering::Acquire)
-    }
-
-    pub fn acquire(&self) -> LeasedItem<T> {
-        let generation = self.generation();
-        loop {
-            let gen_item = self.queue.pop();
-            if gen_item.generation >= generation {
-                return LeasedItem {
-                    gen_item: Some(gen_item),
-                    recycle_queue: Arc::clone(&self.queue),
-                };
-            } else {
-                // this searcher is obsolete,
-                // removing it from the pool.
-            }
-        }
-    }
-}
-
-pub struct LeasedItem<T> {
-    gen_item: Option<GenerationItem<T>>,
-    recycle_queue: Arc<MsQueue<GenerationItem<T>>>,
-}
-
-impl<T> Deref for LeasedItem<T> {
-    type Target = T;
-
-    fn deref(&self) -> &T {
-        &self
-            .gen_item
-            .as_ref()
-            .expect("Unwrapping a leased item should never fail")
-            .item // unwrap is safe here
-    }
-}
-
-impl<T> DerefMut for LeasedItem<T> {
-    fn deref_mut(&mut self) -> &mut T {
-        &mut self
-            .gen_item
-            .as_mut()
-            .expect("Unwrapping a mut leased item should never fail")
-            .item // unwrap is safe here
-    }
-}
-
-impl<T> Drop for LeasedItem<T> {
-    fn drop(&mut self) {
-        let gen_item: GenerationItem<T> = mem::replace(&mut self.gen_item, None)
-            .expect("Unwrapping a leased item should never fail");
-        self.recycle_queue.push(gen_item);
-    }
-}
-
-#[cfg(test)]
-mod tests {
-
-    use super::Pool;
-    use std::iter;
-
-    #[test]
-    fn test_pool() {
-        let items10: Vec<usize> = iter::repeat(10).take(10).collect();
-        let pool = Pool::new();
-        pool.publish_new_generation(items10);
-        for _ in 0..20 {
-            assert_eq!(*pool.acquire(), 10);
-        }
-        let items11: Vec<usize> = iter::repeat(11).take(10).collect();
-        pool.publish_new_generation(items11);
-        for _ in 0..20 {
-            assert_eq!(*pool.acquire(), 11);
-        }
-    }
-}

src/core/searcher.rs

@@ -1,17 +1,17 @@
-use collector::Collector;
-use core::InvertedIndexReader;
-use core::SegmentReader;
-use query::Query;
-use schema::Document;
-use schema::Schema;
-use schema::{Field, Term};
-use space_usage::SearcherSpaceUsage;
-use std::fmt;
-use std::sync::Arc;
-use termdict::TermMerger;
-use DocAddress;
-use Index;
-use Result;
+use crate::collector::Collector;
+use crate::core::Executor;
+
+use crate::core::SegmentReader;
+use crate::query::Query;
+use crate::schema::Document;
+use crate::schema::Schema;
+use crate::schema::Term;
+use crate::space_usage::SearcherSpaceUsage;
+use crate::store::StoreReader;
+use crate::DocAddress;
+use crate::Index;
+
+use std::{fmt, io};
 
 /// Holds a list of `SegmentReader`s ready for search.
 ///
@@ -22,6 +22,7 @@ pub struct Searcher {
     schema: Schema,
     index: Index,
     segment_readers: Vec<SegmentReader>,
+    store_readers: Vec<StoreReader>,
 }
 
 impl Searcher {
@@ -30,12 +31,17 @@ impl Searcher {
         schema: Schema,
         index: Index,
         segment_readers: Vec<SegmentReader>,
-    ) -> Searcher {
-        Searcher {
+    ) -> io::Result<Searcher> {
+        let store_readers: Vec<StoreReader> = segment_readers
+            .iter()
+            .map(SegmentReader::get_store_reader)
+            .collect::<io::Result<Vec<_>>>()?;
+        Ok(Searcher {
             schema,
             index,
             segment_readers,
-        }
+            store_readers,
+        })
     }
 
     /// Returns the `Index` associated to the `Searcher`
@@ -47,10 +53,9 @@ impl Searcher {
     ///
     /// The searcher uses the segment ordinal to route the
     /// the request to the right `Segment`.
-    pub fn doc(&self, doc_address: DocAddress) -> Result<Document> {
-        let DocAddress(segment_local_id, doc_id) = doc_address;
-        let segment_reader = &self.segment_readers[segment_local_id as usize];
-        segment_reader.doc(doc_id)
+    pub fn doc(&self, doc_address: DocAddress) -> crate::Result<Document> {
+        let store_reader = &self.store_readers[doc_address.segment_ord as usize];
+        store_reader.get(doc_address.doc_id)
     }
 
     /// Access the schema associated to the index of this searcher.
@@ -68,12 +73,14 @@ impl Searcher {
 
     /// Return the overall number of documents containing
     /// the given term.
-    pub fn doc_freq(&self, term: &Term) -> u64 {
-        self.segment_readers
-            .iter()
-            .map(|segment_reader| {
-                u64::from(segment_reader.inverted_index(term.field()).doc_freq(term))
-            }).sum::<u64>()
+    pub fn doc_freq(&self, term: &Term) -> crate::Result<u64> {
+        let mut total_doc_freq = 0;
+        for segment_reader in &self.segment_readers {
+            let inverted_index = segment_reader.inverted_index(term.field())?;
+            let doc_freq = inverted_index.doc_freq(term)?;
+            total_doc_freq += u64::from(doc_freq);
+        }
+        Ok(total_doc_freq)
     }
 
     /// Return the list of segment readers
@@ -86,58 +93,75 @@ impl Searcher {
         &self.segment_readers[segment_ord as usize]
     }
 
-    /// Runs a query on the segment readers wrapped by the searcher
-    pub fn search<C: Collector>(&self, query: &Query, collector: &mut C) -> Result<()> {
-        query.search(self, collector)
+    /// Runs a query on the segment readers wrapped by the searcher.
+    ///
+    /// Search works as follows :
+    ///
+    ///  First the weight object associated to the query is created.
+    ///
+    ///  Then, the query loops over the segments and for each segment :
+    ///  - setup the collector and informs it that the segment being processed has changed.
+    ///  - creates a SegmentCollector for collecting documents associated to the segment
+    ///  - creates a `Scorer` object associated for this segment
+    ///  - iterate through the matched documents and push them to the segment collector.
+    ///
+    ///  Finally, the Collector merges each of the child collectors into itself for result usability
+    ///  by the caller.
+    pub fn search<C: Collector>(
+        &self,
+        query: &dyn Query,
+        collector: &C,
+    ) -> crate::Result<C::Fruit> {
+        let executor = self.index.search_executor();
+        self.search_with_executor(query, collector, executor)
     }
 
-    /// Return the field searcher associated to a `Field`.
-    pub fn field(&self, field: Field) -> FieldSearcher {
-        let inv_index_readers = self
-            .segment_readers
-            .iter()
-            .map(|segment_reader| segment_reader.inverted_index(field))
-            .collect::<Vec<_>>();
-        FieldSearcher::new(inv_index_readers)
+    /// Same as [`search(...)`](#method.search) but multithreaded.
+    ///
+    /// The current implementation is rather naive :
+    /// multithreading is by splitting search into as many task
+    /// as there are segments.
+    ///
+    /// It is powerless at making search faster if your index consists in
+    /// one large segment.
+    ///
+    /// Also, keep in my multithreading a single query on several
+    /// threads will not improve your throughput. It can actually
+    /// hurt it. It will however, decrease the average response time.
+    pub fn search_with_executor<C: Collector>(
+        &self,
+        query: &dyn Query,
+        collector: &C,
+        executor: &Executor,
+    ) -> crate::Result<C::Fruit> {
+        let scoring_enabled = collector.requires_scoring();
+        let weight = query.weight(self, scoring_enabled)?;
+        let segment_readers = self.segment_readers();
+        let fruits = executor.map(
+            |(segment_ord, segment_reader)| {
+                collector.collect_segment(weight.as_ref(), segment_ord as u32, segment_reader)
+            },
+            segment_readers.iter().enumerate(),
+        )?;
+        collector.merge_fruits(fruits)
     }
 
     /// Summarize total space usage of this searcher.
-    pub fn space_usage(&self) -> SearcherSpaceUsage {
+    pub fn space_usage(&self) -> io::Result<SearcherSpaceUsage> {
         let mut space_usage = SearcherSpaceUsage::new();
-        for segment_reader in self.segment_readers.iter() {
-            space_usage.add_segment(segment_reader.space_usage());
+        for segment_reader in &self.segment_readers {
+            space_usage.add_segment(segment_reader.space_usage()?);
         }
-        space_usage
-    }
-}
-
-pub struct FieldSearcher {
-    inv_index_readers: Vec<Arc<InvertedIndexReader>>,
-}
-
-impl FieldSearcher {
-    fn new(inv_index_readers: Vec<Arc<InvertedIndexReader>>) -> FieldSearcher {
-        FieldSearcher { inv_index_readers }
-    }
-
-    /// Returns a Stream over all of the sorted unique terms of
-    /// for the given field.
-    pub fn terms(&self) -> TermMerger {
-        let term_streamers: Vec<_> = self
-            .inv_index_readers
-            .iter()
-            .map(|inverted_index| inverted_index.terms().stream())
-            .collect();
-        TermMerger::new(term_streamers)
+        Ok(space_usage)
     }
 }
 
 impl fmt::Debug for Searcher {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         let segment_ids = self
             .segment_readers
             .iter()
-            .map(|segment_reader| segment_reader.segment_id())
+            .map(SegmentReader::segment_id)
             .collect::<Vec<_>>();
         write!(f, "Searcher({:?})", segment_ids)
     }

src/core/segment.rs

@@ -1,16 +1,14 @@
 use super::SegmentComponent;
-use core::Index;
-use core::SegmentId;
-use core::SegmentMeta;
-use directory::error::{OpenReadError, OpenWriteError};
-use directory::Directory;
-use directory::{ReadOnlySource, WritePtr};
-use indexer::segment_serializer::SegmentSerializer;
-use schema::Schema;
+use crate::core::Index;
+use crate::core::SegmentId;
+use crate::core::SegmentMeta;
+use crate::directory::error::{OpenReadError, OpenWriteError};
+use crate::directory::Directory;
+use crate::directory::{FileSlice, WritePtr};
+use crate::schema::Schema;
+use crate::Opstamp;
 use std::fmt;
 use std::path::PathBuf;
-use std::result;
-use Result;
 
 /// A segment is a piece of the index.
 #[derive(Clone)]
@@ -20,20 +18,17 @@ pub struct Segment {
 }
 
 impl fmt::Debug for Segment {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         write!(f, "Segment({:?})", self.id().uuid_string())
     }
 }
 
-/// Creates a new segment given an `Index` and a `SegmentId`
-///
-/// The function is here to make it private outside `tantivy`.
-/// #[doc(hidden)]
-pub fn create_segment(index: Index, meta: SegmentMeta) -> Segment {
-    Segment { index, meta }
-}
-
 impl Segment {
+    /// Creates a new segment given an `Index` and a `SegmentId`
+    pub(crate) fn for_index(index: Index, meta: SegmentMeta) -> Segment {
+        Segment { index, meta }
+    }
+
     /// Returns the index the segment belongs to.
     pub fn index(&self) -> &Index {
         &self.index
@@ -49,8 +44,19 @@ impl Segment {
         &self.meta
     }
 
+    /// Updates the max_doc value from the `SegmentMeta`.
+    ///
+    /// This method is only used when updating `max_doc` from 0
+    /// as we finalize a fresh new segment.
+    pub(crate) fn with_max_doc(self, max_doc: u32) -> Segment {
+        Segment {
+            index: self.index,
+            meta: self.meta.with_max_doc(max_doc),
+        }
+    }
+
     #[doc(hidden)]
-    pub fn with_delete_meta(self, num_deleted_docs: u32, opstamp: u64) -> Segment {
+    pub fn with_delete_meta(self, num_deleted_docs: u32, opstamp: Opstamp) -> Segment {
         Segment {
             index: self.index,
             meta: self.meta.with_delete_meta(num_deleted_docs, opstamp),
@@ -71,31 +77,15 @@ impl Segment {
     }
 
     /// Open one of the component file for a *regular* read.
-    pub fn open_read(
-        &self,
-        component: SegmentComponent,
-    ) -> result::Result<ReadOnlySource, OpenReadError> {
+    pub fn open_read(&self, component: SegmentComponent) -> Result<FileSlice, OpenReadError> {
         let path = self.relative_path(component);
-        let source = self.index.directory().open_read(&path)?;
-        Ok(source)
+        self.index.directory().open_read(&path)
     }
 
     /// Open one of the component file for *regular* write.
-    pub fn open_write(
-        &mut self,
-        component: SegmentComponent,
-    ) -> result::Result<WritePtr, OpenWriteError> {
+    pub fn open_write(&mut self, component: SegmentComponent) -> Result<WritePtr, OpenWriteError> {
         let path = self.relative_path(component);
         let write = self.index.directory_mut().open_write(&path)?;
         Ok(write)
     }
 }
-
-pub trait SerializableSegment {
-    /// Writes a view of a segment by pushing information
-    /// to the `SegmentSerializer`.
-    ///
-    /// # Returns
-    /// The number of documents in the segment.
-    fn write(&self, serializer: SegmentSerializer) -> Result<u32>;
-}

src/core/segment_component.rs

@@ -4,43 +4,43 @@ use std::slice;
 /// Each component is stored in its own file,
 /// using the pattern `segment_uuid`.`component_extension`,
 /// except the delete component that takes an `segment_uuid`.`delete_opstamp`.`component_extension`
-#[derive(Copy, Clone)]
+#[derive(Copy, Clone, Eq, PartialEq)]
 pub enum SegmentComponent {
     /// Postings (or inverted list). Sorted lists of document ids, associated to terms
-    POSTINGS,
+    Postings,
     /// Positions of terms in each document.
-    POSITIONS,
-    /// Index to seek within the position file
-    POSITIONSSKIP,
+    Positions,
     /// Column-oriented random-access storage of fields.
-    FASTFIELDS,
+    FastFields,
     /// Stores the sum  of the length (in terms) of each field for each document.
     /// Field norms are stored as a special u64 fast field.
-    FIELDNORMS,
+    FieldNorms,
     /// Dictionary associating `Term`s to `TermInfo`s which is
     /// simply an address into the `postings` file and the `positions` file.
-    TERMS,
-    /// Row-oriented, LZ4-compressed storage of the documents.
+    Terms,
+    /// Row-oriented, compressed storage of the documents.
     /// Accessing a document from the store is relatively slow, as it
     /// requires to decompress the entire block it belongs to.
-    STORE,
+    Store,
+    /// Temporary storage of the documents, before streamed to `Store`.
+    TempStore,
     /// Bitset describing which document of the segment is deleted.
-    DELETE,
+    Delete,
 }
 
 impl SegmentComponent {
     /// Iterates through the components.
     pub fn iterator() -> slice::Iter<'static, SegmentComponent> {
         static SEGMENT_COMPONENTS: [SegmentComponent; 8] = [
-            SegmentComponent::POSTINGS,
-            SegmentComponent::POSITIONS,
-            SegmentComponent::POSITIONSSKIP,
-            SegmentComponent::FASTFIELDS,
-            SegmentComponent::FIELDNORMS,
-            SegmentComponent::TERMS,
-            SegmentComponent::STORE,
-            SegmentComponent::DELETE,
+            SegmentComponent::Postings,
+            SegmentComponent::Positions,
+            SegmentComponent::FastFields,
+            SegmentComponent::FieldNorms,
+            SegmentComponent::Terms,
+            SegmentComponent::Store,
+            SegmentComponent::TempStore,
+            SegmentComponent::Delete,
         ];
-        SEGMENT_COMPONENTS.into_iter()
+        SEGMENT_COMPONENTS.iter()
     }
 }

src/core/segment_id.rs

@@ -2,6 +2,11 @@ use std::cmp::{Ord, Ordering};
 use std::fmt;
 use uuid::Uuid;
 
+#[cfg(test)]
+use once_cell::sync::Lazy;
+use serde::{Deserialize, Serialize};
+use std::error::Error;
+use std::str::FromStr;
 #[cfg(test)]
 use std::sync::atomic;
 
@@ -17,10 +22,10 @@ use std::sync::atomic;
 pub struct SegmentId(Uuid);
 
 #[cfg(test)]
-lazy_static! {
-    static ref AUTO_INC_COUNTER: atomic::AtomicUsize = atomic::AtomicUsize::default();
-    static ref EMPTY_ARR: [u8; 8] = [0u8; 8];
-}
+static AUTO_INC_COUNTER: Lazy<atomic::AtomicUsize> = Lazy::new(atomic::AtomicUsize::default);
+
+#[cfg(test)]
+const ZERO_ARRAY: [u8; 8] = [0u8; 8];
 
 // During tests, we generate the segment id in a autoincrement manner
 // for consistency of segment id between run.
@@ -30,7 +35,7 @@ lazy_static! {
 #[cfg(test)]
 fn create_uuid() -> Uuid {
     let new_auto_inc_id = (*AUTO_INC_COUNTER).fetch_add(1, atomic::Ordering::SeqCst);
-    Uuid::from_fields(new_auto_inc_id as u32, 0, 0, &*EMPTY_ARR).unwrap()
+    Uuid::from_fields(new_auto_inc_id as u32, 0, 0, &ZERO_ARRAY).unwrap()
 }
 
 #[cfg(not(test))]
@@ -50,19 +55,61 @@ impl SegmentId {
     /// and the rest is random.
     ///
     /// Picking the first 8 chars is ok to identify
-    /// segments in a display message.
+    /// segments in a display message (e.g. a5c4dfcb).
     pub fn short_uuid_string(&self) -> String {
         (&self.0.to_simple_ref().to_string()[..8]).to_string()
     }
 
     /// Returns a segment uuid string.
+    ///
+    /// It consists in 32 lowercase hexadecimal chars
+    /// (e.g. a5c4dfcbdfe645089129e308e26d5523)
     pub fn uuid_string(&self) -> String {
         self.0.to_simple_ref().to_string()
     }
+
+    /// Build a `SegmentId` string from the full uuid string.
+    ///
+    /// E.g. "a5c4dfcbdfe645089129e308e26d5523"
+    pub fn from_uuid_string(uuid_string: &str) -> Result<SegmentId, SegmentIdParseError> {
+        FromStr::from_str(uuid_string)
+    }
+}
+
+/// Error type used when parsing a `SegmentId` from a string fails.
+pub struct SegmentIdParseError(uuid::Error);
+
+impl Error for SegmentIdParseError {}
+
+impl fmt::Debug for SegmentIdParseError {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        self.0.fmt(f)
+    }
+}
+
+impl fmt::Display for SegmentIdParseError {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        self.0.fmt(f)
+    }
+}
+
+impl FromStr for SegmentId {
+    type Err = SegmentIdParseError;
+
+    fn from_str(uuid_string: &str) -> Result<Self, SegmentIdParseError> {
+        let uuid = Uuid::parse_str(uuid_string).map_err(SegmentIdParseError)?;
+        Ok(SegmentId(uuid))
+    }
 }
 
 impl fmt::Debug for SegmentId {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "Seg({:?})", self.short_uuid_string())
+    }
+}
+
+impl fmt::Display for SegmentId {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         write!(f, "Seg({:?})", self.short_uuid_string())
     }
 }
@@ -78,3 +125,18 @@ impl Ord for SegmentId {
         self.0.as_bytes().cmp(other.0.as_bytes())
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::SegmentId;
+
+    #[test]
+    fn test_to_uuid_string() {
+        let full_uuid = "a5c4dfcbdfe645089129e308e26d5523";
+        let segment_id = SegmentId::from_uuid_string(full_uuid).unwrap();
+        assert_eq!(segment_id.uuid_string(), full_uuid);
+        assert_eq!(segment_id.short_uuid_string(), "a5c4dfcb");
+        // one extra char
+        assert!(SegmentId::from_uuid_string("a5c4dfcbdfe645089129e308e26d5523b").is_err());
+    }
+}

src/core/segment_meta.rs

@@ -1,174 +0,0 @@
-use super::SegmentComponent;
-use census::{Inventory, TrackedObject};
-use core::SegmentId;
-use serde;
-use std::collections::HashSet;
-use std::fmt;
-use std::path::PathBuf;
-
-lazy_static! {
-    static ref INVENTORY: Inventory<InnerSegmentMeta> = { Inventory::new() };
-}
-
-#[derive(Clone, Debug, Serialize, Deserialize)]
-struct DeleteMeta {
-    num_deleted_docs: u32,
-    opstamp: u64,
-}
-
-/// `SegmentMeta` contains simple meta information about a segment.
-///
-/// For instance the number of docs it contains,
-/// how many are deleted, etc.
-#[derive(Clone)]
-pub struct SegmentMeta {
-    tracked: TrackedObject<InnerSegmentMeta>,
-}
-
-impl fmt::Debug for SegmentMeta {
-    fn fmt(&self, f: &mut fmt::Formatter) -> Result<(), fmt::Error> {
-        self.tracked.fmt(f)
-    }
-}
-
-impl serde::Serialize for SegmentMeta {
-    fn serialize<S>(
-        &self,
-        serializer: S,
-    ) -> Result<<S as serde::Serializer>::Ok, <S as serde::Serializer>::Error>
-    where
-        S: serde::Serializer,
-    {
-        self.tracked.serialize(serializer)
-    }
-}
-
-impl<'a> serde::Deserialize<'a> for SegmentMeta {
-    fn deserialize<D>(deserializer: D) -> Result<Self, <D as serde::Deserializer<'a>>::Error>
-    where
-        D: serde::Deserializer<'a>,
-    {
-        let inner = InnerSegmentMeta::deserialize(deserializer)?;
-        let tracked = INVENTORY.track(inner);
-        Ok(SegmentMeta { tracked })
-    }
-}
-
-impl SegmentMeta {
-    /// Lists all living `SegmentMeta` object at the time of the call.
-    pub fn all() -> Vec<SegmentMeta> {
-        INVENTORY
-            .list()
-            .into_iter()
-            .map(|inner| SegmentMeta { tracked: inner })
-            .collect::<Vec<_>>()
-    }
-
-    /// Creates a new `SegmentMeta` object.
-    #[doc(hidden)]
-    pub fn new(segment_id: SegmentId, max_doc: u32) -> SegmentMeta {
-        let inner = InnerSegmentMeta {
-            segment_id,
-            max_doc,
-            deletes: None,
-        };
-        SegmentMeta {
-            tracked: INVENTORY.track(inner),
-        }
-    }
-
-    /// Returns the segment id.
-    pub fn id(&self) -> SegmentId {
-        self.tracked.segment_id
-    }
-
-    /// Returns the number of deleted documents.
-    pub fn num_deleted_docs(&self) -> u32 {
-        self.tracked
-            .deletes
-            .as_ref()
-            .map(|delete_meta| delete_meta.num_deleted_docs)
-            .unwrap_or(0u32)
-    }
-
-    /// Returns the list of files that
-    /// are required for the segment meta.
-    ///
-    /// This is useful as the way tantivy removes files
-    /// is by removing all files that have been created by tantivy
-    /// and are not used by any segment anymore.
-    pub fn list_files(&self) -> HashSet<PathBuf> {
-        SegmentComponent::iterator()
-            .map(|component| self.relative_path(*component))
-            .collect::<HashSet<PathBuf>>()
-    }
-
-    /// Returns the relative path of a component of our segment.
-    ///
-    /// It just joins the segment id with the extension
-    /// associated to a segment component.
-    pub fn relative_path(&self, component: SegmentComponent) -> PathBuf {
-        let mut path = self.id().uuid_string();
-        path.push_str(&*match component {
-            SegmentComponent::POSTINGS => ".idx".to_string(),
-            SegmentComponent::POSITIONS => ".pos".to_string(),
-            SegmentComponent::POSITIONSSKIP => ".posidx".to_string(),
-            SegmentComponent::TERMS => ".term".to_string(),
-            SegmentComponent::STORE => ".store".to_string(),
-            SegmentComponent::FASTFIELDS => ".fast".to_string(),
-            SegmentComponent::FIELDNORMS => ".fieldnorm".to_string(),
-            SegmentComponent::DELETE => format!(".{}.del", self.delete_opstamp().unwrap_or(0)),
-        });
-        PathBuf::from(path)
-    }
-
-    /// Return the highest doc id + 1
-    ///
-    /// If there are no deletes, then num_docs = max_docs
-    /// and all the doc ids contains in this segment
-    /// are exactly (0..max_doc).
-    pub fn max_doc(&self) -> u32 {
-        self.tracked.max_doc
-    }
-
-    /// Return the number of documents in the segment.
-    pub fn num_docs(&self) -> u32 {
-        self.max_doc() - self.num_deleted_docs()
-    }
-
-    /// Returns the opstamp of the last delete operation
-    /// taken in account in this segment.
-    pub fn delete_opstamp(&self) -> Option<u64> {
-        self.tracked
-            .deletes
-            .as_ref()
-            .map(|delete_meta| delete_meta.opstamp)
-    }
-
-    /// Returns true iff the segment meta contains
-    /// delete information.
-    pub fn has_deletes(&self) -> bool {
-        self.num_deleted_docs() > 0
-    }
-
-    #[doc(hidden)]
-    pub fn with_delete_meta(self, num_deleted_docs: u32, opstamp: u64) -> SegmentMeta {
-        let delete_meta = DeleteMeta {
-            num_deleted_docs,
-            opstamp,
-        };
-        let tracked = self.tracked.map(move |inner_meta| InnerSegmentMeta {
-            segment_id: inner_meta.segment_id,
-            max_doc: inner_meta.max_doc,
-            deletes: Some(delete_meta),
-        });
-        SegmentMeta { tracked }
-    }
-}
-
-#[derive(Clone, Debug, Serialize, Deserialize)]
-struct InnerSegmentMeta {
-    segment_id: SegmentId,
-    max_doc: u32,
-    deletes: Option<DeleteMeta>,
-}

src/core/segment_reader.rs

@@ -1,30 +1,26 @@
-use common::CompositeFile;
-use common::HasLen;
-use core::InvertedIndexReader;
-use core::Segment;
-use core::SegmentComponent;
-use core::SegmentId;
-use error::TantivyError;
-use fastfield::DeleteBitSet;
-use fastfield::FacetReader;
-use fastfield::FastFieldReader;
-use fastfield::{self, FastFieldNotAvailableError};
-use fastfield::{BytesFastFieldReader, FastValue, MultiValueIntFastFieldReader};
-use fieldnorm::FieldNormReader;
-use schema::Cardinality;
-use schema::Document;
-use schema::Field;
-use schema::FieldType;
-use schema::Schema;
-use space_usage::SegmentSpaceUsage;
-use std::collections::HashMap;
+use crate::core::InvertedIndexReader;
+use crate::core::Segment;
+use crate::core::SegmentComponent;
+use crate::core::SegmentId;
+use crate::directory::CompositeFile;
+use crate::directory::FileSlice;
+use crate::error::DataCorruption;
+use crate::fastfield::AliveBitSet;
+use crate::fastfield::FacetReader;
+use crate::fastfield::FastFieldReaders;
+use crate::fieldnorm::{FieldNormReader, FieldNormReaders};
+use crate::schema::FieldType;
+use crate::schema::Schema;
+use crate::schema::{Field, IndexRecordOption};
+use crate::space_usage::SegmentSpaceUsage;
+use crate::store::StoreReader;
+use crate::termdict::TermDictionary;
+use crate::DocId;
+use fail::fail_point;
 use std::fmt;
 use std::sync::Arc;
 use std::sync::RwLock;
-use store::StoreReader;
-use termdict::TermDictionary;
-use DocId;
-use Result;
+use std::{collections::HashMap, io};
 
 /// Entry point to access all of the datastructures of the `Segment`
 ///
@@ -36,9 +32,6 @@ use Result;
 ///
 /// The segment reader has a very low memory footprint,
 /// as close to all of the memory data is mmapped.
-///
-///
-/// TODO fix not decoding docfreq
 #[derive(Clone)]
 pub struct SegmentReader {
     inv_idx_reader_cache: Arc<RwLock<HashMap<Field, Arc<InvertedIndexReader>>>>,
@@ -50,29 +43,23 @@ pub struct SegmentReader {
     termdict_composite: CompositeFile,
     postings_composite: CompositeFile,
     positions_composite: CompositeFile,
-    positions_idx_composite: CompositeFile,
-    fast_fields_composite: CompositeFile,
-    fieldnorms_composite: CompositeFile,
+    fast_fields_readers: Arc<FastFieldReaders>,
+    fieldnorm_readers: FieldNormReaders,
 
-    store_reader: StoreReader,
-    delete_bitset_opt: Option<DeleteBitSet>,
+    store_file: FileSlice,
+    alive_bitset_opt: Option<AliveBitSet>,
     schema: Schema,
 }
 
 impl SegmentReader {
     /// Returns the highest document id ever attributed in
     /// this segment + 1.
-    /// Today, `tantivy` does not handle deletes, so it happens
-    /// to also be the number of documents in the index.
     pub fn max_doc(&self) -> DocId {
         self.max_doc
     }
 
-    /// Returns the number of documents.
+    /// Returns the number of alive documents.
     /// Deleted documents are not counted.
-    ///
-    /// Today, `tantivy` does not handle deletes so max doc and
-    /// num_docs are the same.
     pub fn num_docs(&self) -> DocId {
         self.num_docs
     }
@@ -85,14 +72,12 @@ impl SegmentReader {
     /// Return the number of documents that have been
     /// deleted in the segment.
     pub fn num_deleted_docs(&self) -> DocId {
-        self.delete_bitset()
-            .map(|delete_set| delete_set.len() as DocId)
-            .unwrap_or(0u32)
+        self.max_doc - self.num_docs
     }
 
     /// Returns true iff some of the documents of the segment have been deleted.
     pub fn has_deletes(&self) -> bool {
-        self.delete_bitset().is_some()
+        self.num_deleted_docs() > 0
     }
 
     /// Accessor to a segment's fast field reader given a field.
@@ -105,177 +90,104 @@ impl SegmentReader {
     ///
     /// # Panics
     /// May panic if the index is corrupted.
-    pub fn fast_field_reader<Item: FastValue>(
-        &self,
-        field: Field,
-    ) -> fastfield::Result<FastFieldReader<Item>> {
-        let field_entry = self.schema.get_field_entry(field);
-        if Item::fast_field_cardinality(field_entry.field_type()) == Some(Cardinality::SingleValue)
-        {
-            self.fast_fields_composite
-                .open_read(field)
-                .ok_or_else(|| FastFieldNotAvailableError::new(field_entry))
-                .map(FastFieldReader::open)
-        } else {
-            Err(FastFieldNotAvailableError::new(field_entry))
-        }
-    }
-
-    pub(crate) fn fast_field_reader_with_idx<Item: FastValue>(
-        &self,
-        field: Field,
-        idx: usize,
-    ) -> fastfield::Result<FastFieldReader<Item>> {
-        if let Some(ff_source) = self.fast_fields_composite.open_read_with_idx(field, idx) {
-            Ok(FastFieldReader::open(ff_source))
-        } else {
-            let field_entry = self.schema.get_field_entry(field);
-            Err(FastFieldNotAvailableError::new(field_entry))
-        }
-    }
-
-    /// Accessor to the `MultiValueIntFastFieldReader` associated to a given `Field`.
-    /// May panick if the field is not a multivalued fastfield of the type `Item`.
-    pub fn multi_fast_field_reader<Item: FastValue>(
-        &self,
-        field: Field,
-    ) -> fastfield::Result<MultiValueIntFastFieldReader<Item>> {
-        let field_entry = self.schema.get_field_entry(field);
-        if Item::fast_field_cardinality(field_entry.field_type()) == Some(Cardinality::MultiValues)
-        {
-            let idx_reader = self.fast_field_reader_with_idx(field, 0)?;
-            let vals_reader = self.fast_field_reader_with_idx(field, 1)?;
-            Ok(MultiValueIntFastFieldReader::open(idx_reader, vals_reader))
-        } else {
-            Err(FastFieldNotAvailableError::new(field_entry))
-        }
-    }
-
-    /// Accessor to the `BytesFastFieldReader` associated to a given `Field`.
-    pub fn bytes_fast_field_reader(&self, field: Field) -> fastfield::Result<BytesFastFieldReader> {
-        let field_entry = self.schema.get_field_entry(field);
-        match *field_entry.field_type() {
-            FieldType::Bytes => {}
-            _ => return Err(FastFieldNotAvailableError::new(field_entry)),
-        }
-        let idx_reader = self
-            .fast_fields_composite
-            .open_read_with_idx(field, 0)
-            .ok_or_else(|| FastFieldNotAvailableError::new(field_entry))
-            .map(FastFieldReader::open)?;
-        let values = self
-            .fast_fields_composite
-            .open_read_with_idx(field, 1)
-            .ok_or_else(|| FastFieldNotAvailableError::new(field_entry))?;
-        Ok(BytesFastFieldReader::open(idx_reader, values))
+    pub fn fast_fields(&self) -> &FastFieldReaders {
+        &self.fast_fields_readers
     }
 
     /// Accessor to the `FacetReader` associated to a given `Field`.
-    pub fn facet_reader(&self, field: Field) -> Result<FacetReader> {
+    pub fn facet_reader(&self, field: Field) -> crate::Result<FacetReader> {
         let field_entry = self.schema.get_field_entry(field);
-        if field_entry.field_type() != &FieldType::HierarchicalFacet {
-            return Err(TantivyError::InvalidArgument(format!(
-                "The field {:?} is not a \
-                 hierarchical facet.",
-                field_entry
-            )));
-        }
-        let term_ords_reader = self.multi_fast_field_reader(field)?;
-        let termdict_source = self.termdict_composite.open_read(field).ok_or_else(|| {
-            TantivyError::InvalidArgument(format!(
-                "The field \"{}\" is a hierarchical \
-                 but this segment does not seem to have the field term \
-                 dictionary.",
+
+        match field_entry.field_type() {
+            FieldType::HierarchicalFacet(_) => {
+                let term_ords_reader = self.fast_fields().u64s(field)?;
+                let termdict = self
+                    .termdict_composite
+                    .open_read(field)
+                    .map(TermDictionary::open)
+                    .unwrap_or_else(|| Ok(TermDictionary::empty()))?;
+                Ok(FacetReader::new(term_ords_reader, termdict))
+            }
+            _ => Err(crate::TantivyError::InvalidArgument(format!(
+                "Field {:?} is not a facet field.",
                 field_entry.name()
-            ))
-        })?;
-        let termdict = TermDictionary::from_source(&termdict_source);
-        let facet_reader = FacetReader::new(term_ords_reader, termdict);
-        Ok(facet_reader)
+            ))),
+        }
     }
 
     /// Accessor to the segment's `Field norms`'s reader.
     ///
     /// Field norms are the length (in tokens) of the fields.
-    /// It is used in the computation of the [TfIdf]
-    /// (https://fulmicoton.gitbooks.io/tantivy-doc/content/tfidf.html).
+    /// It is used in the computation of the [TfIdf](https://fulmicoton.gitbooks.io/tantivy-doc/content/tfidf.html).
     ///
     /// They are simply stored as a fast field, serialized in
     /// the `.fieldnorm` file of the segment.
-    pub fn get_fieldnorms_reader(&self, field: Field) -> FieldNormReader {
-        if let Some(fieldnorm_source) = self.fieldnorms_composite.open_read(field) {
-            FieldNormReader::open(fieldnorm_source)
-        } else {
+    pub fn get_fieldnorms_reader(&self, field: Field) -> crate::Result<FieldNormReader> {
+        self.fieldnorm_readers.get_field(field)?.ok_or_else(|| {
             let field_name = self.schema.get_field_name(field);
             let err_msg = format!(
-                "Field norm not found for field {:?}. Was it market as indexed during indexing.",
+                "Field norm not found for field {:?}. Was it marked as indexed during indexing?",
                 field_name
             );
-            panic!(err_msg);
-        }
+            crate::TantivyError::SchemaError(err_msg)
+        })
     }
 
     /// Accessor to the segment's `StoreReader`.
-    pub fn get_store_reader(&self) -> &StoreReader {
-        &self.store_reader
+    pub fn get_store_reader(&self) -> io::Result<StoreReader> {
+        StoreReader::open(self.store_file.clone())
     }
 
     /// Open a new segment for reading.
-    pub fn open(segment: &Segment) -> Result<SegmentReader> {
-        let termdict_source = segment.open_read(SegmentComponent::TERMS)?;
-        let termdict_composite = CompositeFile::open(&termdict_source)?;
+    pub fn open(segment: &Segment) -> crate::Result<SegmentReader> {
+        let termdict_file = segment.open_read(SegmentComponent::Terms)?;
+        let termdict_composite = CompositeFile::open(&termdict_file)?;
 
-        let store_source = segment.open_read(SegmentComponent::STORE)?;
-        let store_reader = StoreReader::from_source(store_source);
+        let store_file = segment.open_read(SegmentComponent::Store)?;
 
         fail_point!("SegmentReader::open#middle");
 
-        let postings_source = segment.open_read(SegmentComponent::POSTINGS)?;
-        let postings_composite = CompositeFile::open(&postings_source)?;
+        let postings_file = segment.open_read(SegmentComponent::Postings)?;
+        let postings_composite = CompositeFile::open(&postings_file)?;
 
         let positions_composite = {
-            if let Ok(source) = segment.open_read(SegmentComponent::POSITIONS) {
-                CompositeFile::open(&source)?
+            if let Ok(positions_file) = segment.open_read(SegmentComponent::Positions) {
+                CompositeFile::open(&positions_file)?
             } else {
                 CompositeFile::empty()
             }
         };
 
-        let positions_idx_composite = {
-            if let Ok(source) = segment.open_read(SegmentComponent::POSITIONSSKIP) {
-                CompositeFile::open(&source)?
-            } else {
-                CompositeFile::empty()
-            }
-        };
+        let schema = segment.schema();
 
-        let fast_fields_data = segment.open_read(SegmentComponent::FASTFIELDS)?;
+        let fast_fields_data = segment.open_read(SegmentComponent::FastFields)?;
         let fast_fields_composite = CompositeFile::open(&fast_fields_data)?;
+        let fast_field_readers =
+            Arc::new(FastFieldReaders::new(schema.clone(), fast_fields_composite));
 
-        let fieldnorms_data = segment.open_read(SegmentComponent::FIELDNORMS)?;
-        let fieldnorms_composite = CompositeFile::open(&fieldnorms_data)?;
+        let fieldnorm_data = segment.open_read(SegmentComponent::FieldNorms)?;
+        let fieldnorm_readers = FieldNormReaders::open(fieldnorm_data)?;
 
-        let delete_bitset_opt = if segment.meta().has_deletes() {
-            let delete_data = segment.open_read(SegmentComponent::DELETE)?;
-            Some(DeleteBitSet::open(delete_data))
+        let alive_bitset_opt = if segment.meta().has_deletes() {
+            let alive_bitset_bytes = segment.open_read(SegmentComponent::Delete)?.read_bytes()?;
+            let alive_bitset = AliveBitSet::open(alive_bitset_bytes);
+            Some(alive_bitset)
         } else {
             None
         };
 
-        let schema = segment.schema();
         Ok(SegmentReader {
-            inv_idx_reader_cache: Arc::new(RwLock::new(HashMap::new())),
+            inv_idx_reader_cache: Default::default(),
             max_doc: segment.meta().max_doc(),
             num_docs: segment.meta().num_docs(),
             termdict_composite,
             postings_composite,
-            fast_fields_composite,
-            fieldnorms_composite,
+            fast_fields_readers: fast_field_readers,
+            fieldnorm_readers,
             segment_id: segment.id(),
-            store_reader,
-            delete_bitset_opt,
+            store_file,
+            alive_bitset_opt,
             positions_composite,
-            positions_idx_composite,
             schema,
         })
     }
@@ -287,59 +199,61 @@ impl SegmentReader {
     /// The field reader is in charge of iterating through the
     /// term dictionary associated to a specific field,
     /// and opening the posting list associated to any term.
-    pub fn inverted_index(&self, field: Field) -> Arc<InvertedIndexReader> {
+    ///
+    /// If the field is marked as index, a warn is logged and an empty `InvertedIndexReader`
+    /// is returned.
+    /// Similarly if the field is marked as indexed but no term has been indexed for the given
+    /// index. an empty `InvertedIndexReader` is returned (but no warning is logged).
+    pub fn inverted_index(&self, field: Field) -> crate::Result<Arc<InvertedIndexReader>> {
         if let Some(inv_idx_reader) = self
             .inv_idx_reader_cache
             .read()
             .expect("Lock poisoned. This should never happen")
             .get(&field)
         {
-            return Arc::clone(inv_idx_reader);
+            return Ok(Arc::clone(inv_idx_reader));
         }
         let field_entry = self.schema.get_field_entry(field);
         let field_type = field_entry.field_type();
         let record_option_opt = field_type.get_index_record_option();
 
         if record_option_opt.is_none() {
-            panic!("Field {:?} does not seem indexed.", field_entry.name());
+            warn!("Field {:?} does not seem indexed.", field_entry.name());
         }
 
-        let record_option = record_option_opt.unwrap();
+        let postings_file_opt = self.postings_composite.open_read(field);
 
-        let postings_source_opt = self.postings_composite.open_read(field);
-
-        if postings_source_opt.is_none() {
+        if postings_file_opt.is_none() || record_option_opt.is_none() {
             // no documents in the segment contained this field.
             // As a result, no data is associated to the inverted index.
             //
             // Returns an empty inverted index.
-            return Arc::new(InvertedIndexReader::empty(field_type));
+            let record_option = record_option_opt.unwrap_or(IndexRecordOption::Basic);
+            return Ok(Arc::new(InvertedIndexReader::empty(record_option)));
         }
 
-        let postings_source = postings_source_opt.unwrap();
+        let record_option = record_option_opt.unwrap();
+        let postings_file = postings_file_opt.unwrap();
 
-        let termdict_source = self
-            .termdict_composite
-            .open_read(field)
-            .expect("Failed to open field term dictionary in composite file. Is the field indexed");
+        let termdict_file: FileSlice = self.termdict_composite.open_read(field)
+            .ok_or_else(||
+               DataCorruption::comment_only(format!("Failed to open field {:?}'s term dictionary in the composite file. Has the schema been modified?", field_entry.name()))
+            )?;
 
-        let positions_source = self
+        let positions_file = self
             .positions_composite
             .open_read(field)
-            .expect("Index corrupted. Failed to open field positions in composite file.");
-
-        let positions_idx_source = self
-            .positions_idx_composite
-            .open_read(field)
-            .expect("Index corrupted. Failed to open field positions in composite file.");
+            .ok_or_else(|| {
+                let error_msg = format!("Failed to open field {:?}'s positions in the composite file. Has the schema been modified?", field_entry.name());
+               DataCorruption::comment_only(error_msg)
+            })?;
 
         let inv_idx_reader = Arc::new(InvertedIndexReader::new(
-            TermDictionary::from_source(&termdict_source),
-            postings_source,
-            positions_source,
-            positions_idx_source,
+            TermDictionary::open(termdict_file)?,
+            postings_file,
+            positions_file,
             record_option,
-        ));
+        )?);
 
         // by releasing the lock in between, we may end up opening the inverting index
         // twice, but this is fine.
@@ -348,15 +262,7 @@ impl SegmentReader {
             .expect("Field reader cache lock poisoned. This should never happen.")
             .insert(field, Arc::clone(&inv_idx_reader));
 
-        inv_idx_reader
-    }
-
-    /// Returns the document (or to be accurate, its stored field)
-    /// bearing the given doc id.
-    /// This method is slow and should seldom be called from
-    /// within a collector.
-    pub fn doc(&self, doc_id: DocId) -> Result<Document> {
-        self.store_reader.get(doc_id)
+        Ok(inv_idx_reader)
     }
 
     /// Returns the segment id
@@ -366,128 +272,112 @@ impl SegmentReader {
 
     /// Returns the bitset representing
     /// the documents that have been deleted.
-    pub fn delete_bitset(&self) -> Option<&DeleteBitSet> {
-        self.delete_bitset_opt.as_ref()
+    pub fn alive_bitset(&self) -> Option<&AliveBitSet> {
+        self.alive_bitset_opt.as_ref()
     }
 
     /// Returns true iff the `doc` is marked
     /// as deleted.
     pub fn is_deleted(&self, doc: DocId) -> bool {
-        self.delete_bitset()
+        self.alive_bitset()
             .map(|delete_set| delete_set.is_deleted(doc))
             .unwrap_or(false)
     }
 
     /// Returns an iterator that will iterate over the alive document ids
-    pub fn doc_ids_alive(&self) -> SegmentReaderAliveDocsIterator {
-        SegmentReaderAliveDocsIterator::new(&self)
+    pub fn doc_ids_alive(&self) -> Box<dyn Iterator<Item = DocId> + '_> {
+        if let Some(alive_bitset) = &self.alive_bitset_opt {
+            Box::new(alive_bitset.iter_alive())
+        } else {
+            Box::new(0u32..self.max_doc)
+        }
     }
 
     /// Summarize total space usage of this segment.
-    pub fn space_usage(&self) -> SegmentSpaceUsage {
-        SegmentSpaceUsage::new(
+    pub fn space_usage(&self) -> io::Result<SegmentSpaceUsage> {
+        Ok(SegmentSpaceUsage::new(
             self.num_docs(),
             self.termdict_composite.space_usage(),
             self.postings_composite.space_usage(),
             self.positions_composite.space_usage(),
-            self.positions_idx_composite.space_usage(),
-            self.fast_fields_composite.space_usage(),
-            self.fieldnorms_composite.space_usage(),
-            self.store_reader.space_usage(),
-            self.delete_bitset_opt.as_ref().map(|x| x.space_usage()).unwrap_or(0),
-        )
+            self.fast_fields_readers.space_usage(),
+            self.fieldnorm_readers.space_usage(),
+            self.get_store_reader()?.space_usage(),
+            self.alive_bitset_opt
+                .as_ref()
+                .map(AliveBitSet::space_usage)
+                .unwrap_or(0),
+        ))
     }
 }
 
 impl fmt::Debug for SegmentReader {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         write!(f, "SegmentReader({:?})", self.segment_id)
     }
 }
 
-/// Implements the iterator trait to allow easy iteration
-/// over non-deleted ("alive") DocIds in a SegmentReader
-pub struct SegmentReaderAliveDocsIterator<'a> {
-    reader: &'a SegmentReader,
-    max_doc: DocId,
-    current: DocId,
-}
-
-impl<'a> SegmentReaderAliveDocsIterator<'a> {
-    pub fn new(reader: &'a SegmentReader) -> SegmentReaderAliveDocsIterator<'a> {
-        SegmentReaderAliveDocsIterator {
-            reader,
-            max_doc: reader.max_doc(),
-            current: 0,
-        }
-    }
-}
-
-impl<'a> Iterator for SegmentReaderAliveDocsIterator<'a> {
-    type Item = DocId;
-
-    fn next(&mut self) -> Option<Self::Item> {
-        // TODO: Use TinySet (like in BitSetDocSet) to speed this process up
-        if self.current >= self.max_doc {
-            return None;
-        }
-
-        // find the next alive doc id
-        while self.reader.is_deleted(self.current) {
-            self.current += 1;
-
-            if self.current >= self.max_doc {
-                return None;
-            }
-        }
-
-        // capture the current alive DocId
-        let result = Some(self.current);
-
-        // move down the chain
-        self.current += 1;
-
-        result
-    }
-}
-
 #[cfg(test)]
 mod test {
-    use core::Index;
-    use schema::{SchemaBuilder, Term, STORED, TEXT};
-    use DocId;
+    use crate::core::Index;
+    use crate::schema::{Schema, Term, STORED, TEXT};
+    use crate::DocId;
 
     #[test]
-    fn test_alive_docs_iterator() {
-        let mut schema_builder = SchemaBuilder::new();
+    fn test_num_alive() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
         schema_builder.add_text_field("name", TEXT | STORED);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema.clone());
         let name = schema.get_field("name").unwrap();
 
         {
-            let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+            let mut index_writer = index.writer_for_tests()?;
             index_writer.add_document(doc!(name => "tantivy"));
             index_writer.add_document(doc!(name => "horse"));
             index_writer.add_document(doc!(name => "jockey"));
             index_writer.add_document(doc!(name => "cap"));
-
             // we should now have one segment with two docs
-            index_writer.commit().unwrap();
+            index_writer.delete_term(Term::from_field_text(name, "horse"));
+            index_writer.delete_term(Term::from_field_text(name, "cap"));
+
+            // ok, now we should have a deleted doc
+            index_writer.commit()?;
+        }
+        let searcher = index.reader()?.searcher();
+        assert_eq!(2, searcher.segment_reader(0).num_docs());
+        assert_eq!(4, searcher.segment_reader(0).max_doc());
+        Ok(())
+    }
+    #[test]
+    fn test_alive_docs_iterator() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        schema_builder.add_text_field("name", TEXT | STORED);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema.clone());
+        let name = schema.get_field("name").unwrap();
+
+        {
+            let mut index_writer = index.writer_for_tests()?;
+            index_writer.add_document(doc!(name => "tantivy"));
+            index_writer.add_document(doc!(name => "horse"));
+            index_writer.add_document(doc!(name => "jockey"));
+            index_writer.add_document(doc!(name => "cap"));
+            // we should now have one segment with two docs
+            index_writer.commit()?;
         }
 
         {
-            let mut index_writer2 = index.writer(50_000_000).unwrap();
+            let mut index_writer2 = index.writer(50_000_000)?;
             index_writer2.delete_term(Term::from_field_text(name, "horse"));
             index_writer2.delete_term(Term::from_field_text(name, "cap"));
 
             // ok, now we should have a deleted doc
-            index_writer2.commit().unwrap();
+            index_writer2.commit()?;
         }
-
-        index.load_searchers().unwrap();
-        let searcher = index.searcher();
+        let searcher = index.reader()?.searcher();
         let docs: Vec<DocId> = searcher.segment_reader(0).doc_ids_alive().collect();
         assert_eq!(vec![0u32, 2u32], docs);
+        Ok(())
     }
 }

src/directory/composite_file.rs

@@ -1,14 +1,16 @@
+use crate::directory::FileSlice;
+use crate::directory::{TerminatingWrite, WritePtr};
+use crate::schema::Field;
+use crate::space_usage::FieldUsage;
+use crate::space_usage::PerFieldSpaceUsage;
 use common::BinarySerializable;
 use common::CountingWriter;
+use common::HasLen;
 use common::VInt;
-use directory::ReadOnlySource;
-use directory::WritePtr;
-use schema::Field;
-use space_usage::PerFieldSpaceUsage;
-use space_usage::FieldUsage;
 use std::collections::HashMap;
-use std::io::Write;
-use std::io::{self, Read};
+use std::io::{self, Read, Write};
+use std::iter::ExactSizeIterator;
+use std::ops::Range;
 
 #[derive(Eq, PartialEq, Hash, Copy, Ord, PartialOrd, Clone, Debug)]
 pub struct FileAddr {
@@ -39,10 +41,10 @@ impl BinarySerializable for FileAddr {
 /// A `CompositeWrite` is used to write a `CompositeFile`.
 pub struct CompositeWrite<W = WritePtr> {
     write: CountingWriter<W>,
-    offsets: HashMap<FileAddr, usize>,
+    offsets: HashMap<FileAddr, u64>,
 }
 
-impl<W: Write> CompositeWrite<W> {
+impl<W: TerminatingWrite + Write> CompositeWrite<W> {
     /// Crate a new API writer that writes a composite file
     /// in a given write.
     pub fn wrap(w: W) -> CompositeWrite<W> {
@@ -91,8 +93,7 @@ impl<W: Write> CompositeWrite<W> {
 
         let footer_len = (self.write.written_bytes() - footer_offset) as u32;
         footer_len.serialize(&mut self.write)?;
-        self.write.flush()?;
-        Ok(())
+        self.write.terminate()
     }
 }
 
@@ -104,25 +105,26 @@ impl<W: Write> CompositeWrite<W> {
 /// for each field.
 #[derive(Clone)]
 pub struct CompositeFile {
-    data: ReadOnlySource,
-    offsets_index: HashMap<FileAddr, (usize, usize)>,
+    data: FileSlice,
+    offsets_index: HashMap<FileAddr, Range<usize>>,
 }
 
 impl CompositeFile {
     /// Opens a composite file stored in a given
-    /// `ReadOnlySource`.
-    pub fn open(data: &ReadOnlySource) -> io::Result<CompositeFile> {
+    /// `FileSlice`.
+    pub fn open(data: &FileSlice) -> io::Result<CompositeFile> {
         let end = data.len();
-        let footer_len_data = data.slice_from(end - 4);
+        let footer_len_data = data.slice_from(end - 4).read_bytes()?;
         let footer_len = u32::deserialize(&mut footer_len_data.as_slice())? as usize;
         let footer_start = end - 4 - footer_len;
-        let footer_data = data.slice(footer_start, footer_start + footer_len);
+        let footer_data = data
+            .slice(footer_start..footer_start + footer_len)
+            .read_bytes()?;
         let mut footer_buffer = footer_data.as_slice();
         let num_fields = VInt::deserialize(&mut footer_buffer)?.0 as usize;
 
         let mut file_addrs = vec![];
         let mut offsets = vec![];
-
         let mut field_index = HashMap::new();
 
         let mut offset = 0;
@@ -137,7 +139,7 @@ impl CompositeFile {
             let file_addr = file_addrs[i];
             let start_offset = offsets[i];
             let end_offset = offsets[i + 1];
-            field_index.insert(file_addr, (start_offset, end_offset));
+            field_index.insert(file_addr, start_offset..end_offset);
         }
 
         Ok(CompositeFile {
@@ -151,30 +153,31 @@ impl CompositeFile {
     pub fn empty() -> CompositeFile {
         CompositeFile {
             offsets_index: HashMap::new(),
-            data: ReadOnlySource::empty(),
+            data: FileSlice::empty(),
         }
     }
 
-    /// Returns the `ReadOnlySource` associated
+    /// Returns the `FileSlice` associated
     /// to a given `Field` and stored in a `CompositeFile`.
-    pub fn open_read(&self, field: Field) -> Option<ReadOnlySource> {
+    pub fn open_read(&self, field: Field) -> Option<FileSlice> {
         self.open_read_with_idx(field, 0)
     }
 
-    /// Returns the `ReadOnlySource` associated
+    /// Returns the `FileSlice` associated
     /// to a given `Field` and stored in a `CompositeFile`.
-    pub fn open_read_with_idx(&self, field: Field, idx: usize) -> Option<ReadOnlySource> {
+    pub fn open_read_with_idx(&self, field: Field, idx: usize) -> Option<FileSlice> {
         self.offsets_index
             .get(&FileAddr { field, idx })
-            .map(|&(from, to)| self.data.slice(from, to))
+            .map(|byte_range| self.data.slice(byte_range.clone()))
     }
 
     pub fn space_usage(&self) -> PerFieldSpaceUsage {
         let mut fields = HashMap::new();
-        for (&field_addr, &(start, end)) in self.offsets_index.iter() {
-            fields.entry(field_addr.field)
+        for (&field_addr, byte_range) in &self.offsets_index {
+            fields
+                .entry(field_addr.field)
                 .or_insert_with(|| FieldUsage::empty(field_addr.field))
-                .add_field_idx(field_addr.idx, end - start);
+                .add_field_idx(field_addr.idx, byte_range.len());
         }
         PerFieldSpaceUsage::new(fields)
     }
@@ -184,51 +187,52 @@ impl CompositeFile {
 mod test {
 
     use super::{CompositeFile, CompositeWrite};
+    use crate::directory::{Directory, RamDirectory};
+    use crate::schema::Field;
     use common::BinarySerializable;
     use common::VInt;
-    use directory::{Directory, RAMDirectory};
-    use schema::Field;
     use std::io::Write;
     use std::path::Path;
 
     #[test]
-    fn test_composite_file() {
+    fn test_composite_file() -> crate::Result<()> {
         let path = Path::new("test_path");
-        let mut directory = RAMDirectory::create();
+        let directory = RamDirectory::create();
         {
             let w = directory.open_write(path).unwrap();
             let mut composite_write = CompositeWrite::wrap(w);
-            {
-                let mut write_0 = composite_write.for_field(Field(0u32));
-                VInt(32431123u64).serialize(&mut write_0).unwrap();
-                write_0.flush().unwrap();
-            }
-
-            {
-                let mut write_4 = composite_write.for_field(Field(4u32));
-                VInt(2).serialize(&mut write_4).unwrap();
-                write_4.flush().unwrap();
-            }
-            composite_write.close().unwrap();
+            let mut write_0 = composite_write.for_field(Field::from_field_id(0u32));
+            VInt(32431123u64).serialize(&mut write_0)?;
+            write_0.flush()?;
+            let mut write_4 = composite_write.for_field(Field::from_field_id(4u32));
+            VInt(2).serialize(&mut write_4)?;
+            write_4.flush()?;
+            composite_write.close()?;
         }
         {
-            let r = directory.open_read(path).unwrap();
-            let composite_file = CompositeFile::open(&r).unwrap();
+            let r = directory.open_read(path)?;
+            let composite_file = CompositeFile::open(&r)?;
             {
-                let file0 = composite_file.open_read(Field(0u32)).unwrap();
+                let file0 = composite_file
+                    .open_read(Field::from_field_id(0u32))
+                    .unwrap()
+                    .read_bytes()?;
                 let mut file0_buf = file0.as_slice();
-                let payload_0 = VInt::deserialize(&mut file0_buf).unwrap().0;
+                let payload_0 = VInt::deserialize(&mut file0_buf)?.0;
                 assert_eq!(file0_buf.len(), 0);
                 assert_eq!(payload_0, 32431123u64);
             }
             {
-                let file4 = composite_file.open_read(Field(4u32)).unwrap();
+                let file4 = composite_file
+                    .open_read(Field::from_field_id(4u32))
+                    .unwrap()
+                    .read_bytes()?;
                 let mut file4_buf = file4.as_slice();
-                let payload_4 = VInt::deserialize(&mut file4_buf).unwrap().0;
+                let payload_4 = VInt::deserialize(&mut file4_buf)?.0;
                 assert_eq!(file4_buf.len(), 0);
                 assert_eq!(payload_4, 2u64);
             }
         }
+        Ok(())
     }
-
 }

src/directory/directory.rs

@@ -1,11 +1,103 @@
-use directory::error::{DeleteError, OpenReadError, OpenWriteError};
-use directory::{ReadOnlySource, WritePtr};
+use crate::directory::directory_lock::Lock;
+use crate::directory::error::LockError;
+use crate::directory::error::{DeleteError, OpenReadError, OpenWriteError};
+use crate::directory::WatchHandle;
+use crate::directory::{FileHandle, WatchCallback};
+use crate::directory::{FileSlice, WritePtr};
 use std::fmt;
 use std::io;
+use std::io::Write;
 use std::marker::Send;
 use std::marker::Sync;
 use std::path::Path;
-use std::result;
+use std::path::PathBuf;
+use std::thread;
+use std::time::Duration;
+
+/// Retry the logic of acquiring locks is pretty simple.
+/// We just retry `n` times after a given `duratio`, both
+/// depending on the type of lock.
+struct RetryPolicy {
+    num_retries: usize,
+    wait_in_ms: u64,
+}
+
+impl RetryPolicy {
+    fn no_retry() -> RetryPolicy {
+        RetryPolicy {
+            num_retries: 0,
+            wait_in_ms: 0,
+        }
+    }
+
+    fn wait_and_retry(&mut self) -> bool {
+        if self.num_retries == 0 {
+            false
+        } else {
+            self.num_retries -= 1;
+            let wait_duration = Duration::from_millis(self.wait_in_ms);
+            thread::sleep(wait_duration);
+            true
+        }
+    }
+}
+
+/// The `DirectoryLock` is an object that represents a file lock.
+/// See  [`LockType`](struct.LockType.html)
+///
+/// It is transparently associated to a lock file, that gets deleted
+/// on `Drop.` The lock is released automatically on `Drop`.
+pub struct DirectoryLock(Box<dyn Send + Sync + 'static>);
+
+struct DirectoryLockGuard {
+    directory: Box<dyn Directory>,
+    path: PathBuf,
+}
+
+impl<T: Send + Sync + 'static> From<Box<T>> for DirectoryLock {
+    fn from(underlying: Box<T>) -> Self {
+        DirectoryLock(underlying)
+    }
+}
+
+impl Drop for DirectoryLockGuard {
+    fn drop(&mut self) {
+        if let Err(e) = self.directory.delete(&*self.path) {
+            error!("Failed to remove the lock file. {:?}", e);
+        }
+    }
+}
+
+enum TryAcquireLockError {
+    FileExists,
+    IoError(io::Error),
+}
+
+fn try_acquire_lock(
+    filepath: &Path,
+    directory: &mut dyn Directory,
+) -> Result<DirectoryLock, TryAcquireLockError> {
+    let mut write = directory.open_write(filepath).map_err(|e| match e {
+        OpenWriteError::FileAlreadyExists(_) => TryAcquireLockError::FileExists,
+        OpenWriteError::IoError { io_error, .. } => TryAcquireLockError::IoError(io_error),
+    })?;
+    write.flush().map_err(TryAcquireLockError::IoError)?;
+    Ok(DirectoryLock::from(Box::new(DirectoryLockGuard {
+        directory: directory.box_clone(),
+        path: filepath.to_owned(),
+    })))
+}
+
+fn retry_policy(is_blocking: bool) -> RetryPolicy {
+    if is_blocking {
+        RetryPolicy {
+            num_retries: 100,
+            wait_in_ms: 100,
+        }
+    } else {
+        RetryPolicy::no_retry()
+    }
+}
 
 /// Write-once read many (WORM) abstraction for where
 /// tantivy's data should be stored.
@@ -14,37 +106,45 @@ use std::result;
 ///
 /// - The [`MMapDirectory`](struct.MmapDirectory.html), this
 /// should be your default choice.
-/// - The [`RAMDirectory`](struct.RAMDirectory.html), which
+/// - The [`RamDirectory`](struct.RamDirectory.html), which
 /// should be used mostly for tests.
-///
 pub trait Directory: DirectoryClone + fmt::Debug + Send + Sync + 'static {
-    /// Opens a virtual file for read.
+    /// Opens a file and returns a boxed `FileHandle`.
     ///
+    /// Users of `Directory` should typically call `Directory::open_read(...)`,
+    /// while `Directory` implementor should implement `get_file_handle()`.
+    fn get_file_handle(&self, path: &Path) -> Result<Box<dyn FileHandle>, OpenReadError>;
+
     /// Once a virtual file is open, its data may not
     /// change.
     ///
     /// Specifically, subsequent writes or flushes should
-    /// have no effect on the returned `ReadOnlySource` object.
-    fn open_read(&self, path: &Path) -> result::Result<ReadOnlySource, OpenReadError>;
+    /// have no effect on the returned `FileSlice` object.
+    ///
+    /// You should only use this to read files create with [Directory::open_write].
+    fn open_read(&self, path: &Path) -> Result<FileSlice, OpenReadError> {
+        let file_handle = self.get_file_handle(path)?;
+        Ok(FileSlice::new(file_handle))
+    }
 
     /// Removes a file
     ///
     /// Removing a file will not affect an eventual
-    /// existing ReadOnlySource pointing to it.
+    /// existing FileSlice pointing to it.
     ///
     /// Removing a nonexistent file, yields a
     /// `DeleteError::DoesNotExist`.
-    fn delete(&self, path: &Path) -> result::Result<(), DeleteError>;
+    fn delete(&self, path: &Path) -> Result<(), DeleteError>;
 
     /// Returns true iff the file exists
-    fn exists(&self, path: &Path) -> bool;
+    fn exists(&self, path: &Path) -> Result<bool, OpenReadError>;
 
     /// Opens a writer for the *virtual file* associated with
     /// a Path.
     ///
     /// Right after this call, the file should be created
     /// and any subsequent call to `open_read` for the
-    /// same path should return a `ReadOnlySource`.
+    /// same path should return a `FileSlice`.
     ///
     /// Write operations may be aggressively buffered.
     /// The client of this trait is responsible for calling flush
@@ -54,16 +154,18 @@ pub trait Directory: DirectoryClone + fmt::Debug + Send + Sync + 'static {
     /// Flush operation should also be persistent.
     ///
     /// The user shall not rely on `Drop` triggering `flush`.
-    /// Note that `RAMDirectory` will panic! if `flush`
+    /// Note that `RamDirectory` will panic! if `flush`
     /// was not called.
     ///
     /// The file may not previously exist.
-    fn open_write(&mut self, path: &Path) -> Result<WritePtr, OpenWriteError>;
+    fn open_write(&self, path: &Path) -> Result<WritePtr, OpenWriteError>;
 
     /// Reads the full content file that has been written using
     /// atomic_write.
     ///
     /// This should only be used for small files.
+    ///
+    /// You should only use this to read files create with [Directory::atomic_write].
     fn atomic_read(&self, path: &Path) -> Result<Vec<u8>, OpenReadError>;
 
     /// Atomically replace the content of a file with data.
@@ -72,20 +174,59 @@ pub trait Directory: DirectoryClone + fmt::Debug + Send + Sync + 'static {
     /// a partially written file.
     ///
     /// The file may or may not previously exist.
-    fn atomic_write(&mut self, path: &Path, data: &[u8]) -> io::Result<()>;
+    fn atomic_write(&self, path: &Path, data: &[u8]) -> io::Result<()>;
+
+    /// Acquire a lock in the given directory.
+    ///
+    /// The method is blocking or not depending on the `Lock` object.
+    fn acquire_lock(&self, lock: &Lock) -> Result<DirectoryLock, LockError> {
+        let mut box_directory = self.box_clone();
+        let mut retry_policy = retry_policy(lock.is_blocking);
+        loop {
+            match try_acquire_lock(&lock.filepath, &mut *box_directory) {
+                Ok(result) => {
+                    return Ok(result);
+                }
+                Err(TryAcquireLockError::FileExists) => {
+                    if !retry_policy.wait_and_retry() {
+                        return Err(LockError::LockBusy);
+                    }
+                }
+                Err(TryAcquireLockError::IoError(io_error)) => {
+                    return Err(LockError::IoError(io_error));
+                }
+            }
+        }
+    }
+
+    /// Registers a callback that will be called whenever a change on the `meta.json`
+    /// using the `atomic_write` API is detected.
+    ///
+    /// The behavior when using `.watch()` on a file using [Directory::open_write] is, on the other
+    /// hand, undefined.
+    ///
+    /// The file will be watched for the lifetime of the returned `WatchHandle`. The caller is
+    /// required to keep it.
+    /// It does not override previous callbacks. When the file is modified, all callback that are
+    /// registered (and whose `WatchHandle` is still alive) are triggered.
+    ///
+    /// Internally, tantivy only uses this API to detect new commits to implement the
+    /// `OnCommit` `ReloadPolicy`. Not implementing watch in a `Directory` only prevents the
+    /// `OnCommit` `ReloadPolicy` to work properly.
+    fn watch(&self, watch_callback: WatchCallback) -> crate::Result<WatchHandle>;
 }
 
 /// DirectoryClone
 pub trait DirectoryClone {
     /// Clones the directory and boxes the clone
-    fn box_clone(&self) -> Box<Directory>;
+    fn box_clone(&self) -> Box<dyn Directory>;
 }
 
 impl<T> DirectoryClone for T
 where
     T: 'static + Directory + Clone,
 {
-    fn box_clone(&self) -> Box<Directory> {
+    fn box_clone(&self) -> Box<dyn Directory> {
         Box::new(self.clone())
     }
 }

src/directory/directory_lock.rs

@@ -0,0 +1,55 @@
+use once_cell::sync::Lazy;
+use std::path::PathBuf;
+
+/// A directory lock.
+///
+/// A lock is associated to a specific path and some
+/// [`LockParams`](./enum.LockParams.html).
+/// Tantivy itself uses only two locks but client application
+/// can use the directory facility to define their own locks.
+/// - [INDEX_WRITER_LOCK](./struct.INDEX_WRITER_LOCK.html)
+/// - [META_LOCK](./struct.META_LOCK.html)
+///
+/// Check out these locks documentation for more information.
+///
+#[derive(Debug)]
+pub struct Lock {
+    /// The lock needs to be associated with its own file `path`.
+    /// Depending on the platform, the lock might rely on the creation
+    /// and deletion of this filepath.
+    pub filepath: PathBuf,
+    /// `lock_params` describes whether acquiring the lock is meant
+    /// to be a blocking operation or a non-blocking.
+    ///
+    /// Acquiring a blocking lock blocks until the lock is
+    /// available.
+    /// Acquiring a blocking lock returns rapidly, either successfully
+    /// or with an error signifying that someone is already holding
+    /// the lock.
+    pub is_blocking: bool,
+}
+
+/// Only one process should be able to write tantivy's index at a time.
+/// This lock file, when present, is in charge of preventing other processes to open an IndexWriter.
+///
+/// If the process is killed and this file remains, it is safe to remove it manually.
+///
+/// Failing to acquire this lock usually means a misuse of tantivy's API,
+/// (creating more than one instance of the `IndexWriter`), are a spurious
+/// lock file remaining after a crash. In the latter case, removing the file after
+/// checking no process running tantivy is running is safe.
+pub static INDEX_WRITER_LOCK: Lazy<Lock> = Lazy::new(|| Lock {
+    filepath: PathBuf::from(".tantivy-writer.lock"),
+    is_blocking: false,
+});
+/// The meta lock file is here to protect the segment files being opened by
+/// `IndexReader::reload()` from being garbage collected.
+/// It makes it possible for another process to safely consume
+/// our index in-writing. Ideally, we may have prefered `RWLock` semantics
+/// here, but it is difficult to achieve on Windows.
+///
+/// Opening segment readers is a very fast process.
+pub static META_LOCK: Lazy<Lock> = Lazy::new(|| Lock {
+    filepath: PathBuf::from(".tantivy-meta.lock"),
+    is_blocking: true,
+});

src/directory/error.rs

@@ -1,208 +1,173 @@
-use std::error::Error as StdError;
+use crate::Version;
 use std::fmt;
 use std::io;
 use std::path::PathBuf;
 
-/// General IO error with an optional path to the offending file.
-#[derive(Debug)]
-pub struct IOError {
-    path: Option<PathBuf>,
-    err: io::Error,
-}
-
-impl fmt::Display for IOError {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        match self.path {
-            Some(ref path) => write!(f, "io error occurred on path '{:?}': '{}'", path, self.err),
-            None => write!(f, "io error occurred: '{}'", self.err),
-        }
-    }
-}
-
-impl StdError for IOError {
-    fn description(&self) -> &str {
-        "io error occurred"
-    }
-
-    fn cause(&self) -> Option<&StdError> {
-        Some(&self.err)
-    }
-}
-
-impl IOError {
-    pub(crate) fn with_path(path: PathBuf, err: io::Error) -> Self {
-        IOError {
-            path: Some(path),
-            err,
-        }
-    }
-}
-
-impl From<io::Error> for IOError {
-    fn from(err: io::Error) -> IOError {
-        IOError { path: None, err }
-    }
+/// Error while trying to acquire a directory lock.
+#[derive(Debug, Error)]
+pub enum LockError {
+    /// Failed to acquired a lock as it is already held by another
+    /// client.
+    /// - In the context of a blocking lock, this means the lock was not released within some `timeout` period.
+    /// - In the context of a non-blocking lock, this means the lock was busy at the moment of the call.
+    #[error("Could not acquire lock as it is already held, possibly by a different process.")]
+    LockBusy,
+    /// Trying to acquire a lock failed with an `IoError`
+    #[error("Failed to acquire the lock due to an io:Error.")]
+    IoError(io::Error),
 }
 
 /// Error that may occur when opening a directory
-#[derive(Debug)]
+#[derive(Debug, Error)]
 pub enum OpenDirectoryError {
     /// The underlying directory does not exists.
+    #[error("Directory does not exist: '{0}'.")]
     DoesNotExist(PathBuf),
     /// The path exists but is not a directory.
+    #[error("Path exists but is not a directory: '{0}'.")]
     NotADirectory(PathBuf),
-}
-
-impl fmt::Display for OpenDirectoryError {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        match *self {
-            OpenDirectoryError::DoesNotExist(ref path) => {
-                write!(f, "the underlying directory '{:?}' does not exist", path)
-            }
-            OpenDirectoryError::NotADirectory(ref path) => {
-                write!(f, "the path '{:?}' exists but is not a directory", path)
-            }
-        }
-    }
-}
-
-impl StdError for OpenDirectoryError {
-    fn description(&self) -> &str {
-        "error occurred while opening a directory"
-    }
-
-    fn cause(&self) -> Option<&StdError> {
-        None
-    }
+    /// Failed to create a temp directory.
+    #[error("Failed to create a temporary directory: '{0}'.")]
+    FailedToCreateTempDir(io::Error),
+    /// IoError
+    #[error("IoError '{io_error:?}' while create directory in: '{directory_path:?}'.")]
+    IoError {
+        /// underlying io Error.
+        io_error: io::Error,
+        /// directory we tried to open.
+        directory_path: PathBuf,
+    },
 }
 
 /// Error that may occur when starting to write in a file
-#[derive(Debug)]
+#[derive(Debug, Error)]
 pub enum OpenWriteError {
     /// Our directory is WORM, writing an existing file is forbidden.
     /// Checkout the `Directory` documentation.
+    #[error("File already exists: '{0}'")]
     FileAlreadyExists(PathBuf),
     /// Any kind of IO error that happens when
     /// writing in the underlying IO device.
-    IOError(IOError),
+    #[error("IoError '{io_error:?}' while opening file for write: '{filepath}'.")]
+    IoError {
+        /// The underlying `io::Error`.
+        io_error: io::Error,
+        /// File path of the file that tantivy failed to open for write.
+        filepath: PathBuf,
+    },
 }
 
-impl From<IOError> for OpenWriteError {
-    fn from(err: IOError) -> OpenWriteError {
-        OpenWriteError::IOError(err)
+impl OpenWriteError {
+    /// Wraps an io error.
+    pub fn wrap_io_error(io_error: io::Error, filepath: PathBuf) -> Self {
+        Self::IoError { io_error, filepath }
     }
 }
+/// Type of index incompatibility between the library and the index found on disk
+/// Used to catch and provide a hint to solve this incompatibility issue
+pub enum Incompatibility {
+    /// This library cannot decompress the index found on disk
+    CompressionMismatch {
+        /// Compression algorithm used by the current version of tantivy
+        library_compression_format: String,
+        /// Compression algorithm that was used to serialise the index
+        index_compression_format: String,
+    },
+    /// The index format found on disk isn't supported by this version of the library
+    IndexMismatch {
+        /// Version used by the library
+        library_version: Version,
+        /// Version the index was built with
+        index_version: Version,
+    },
+}
 
-impl fmt::Display for OpenWriteError {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        match *self {
-            OpenWriteError::FileAlreadyExists(ref path) => {
-                write!(f, "the file '{:?}' already exists", path)
+impl fmt::Debug for Incompatibility {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> Result<(), fmt::Error> {
+        match self {
+            Incompatibility::CompressionMismatch {
+                library_compression_format,
+                index_compression_format,
+            } => {
+                let err = format!(
+                    "Library was compiled with {:?} compression, index was compressed with {:?}",
+                    library_compression_format, index_compression_format
+                );
+                let advice = format!(
+                    "Change the feature flag to {:?} and rebuild the library",
+                    index_compression_format
+                );
+                write!(f, "{}. {}", err, advice)?;
+            }
+            Incompatibility::IndexMismatch {
+                library_version,
+                index_version,
+            } => {
+                let err = format!(
+                    "Library version: {}, index version: {}",
+                    library_version.index_format_version, index_version.index_format_version
+                );
+                // TODO make a more useful error message
+                // include the version range that supports this index_format_version
+                let advice = format!(
+                    "Change tantivy to a version compatible with index format {} (e.g. {}.{}.x) \
+                     and rebuild your project.",
+                    index_version.index_format_version, index_version.major, index_version.minor
+                );
+                write!(f, "{}. {}", err, advice)?;
             }
-            OpenWriteError::IOError(ref err) => write!(
-                f,
-                "an io error occurred while opening a file for writing: '{}'",
-                err
-            ),
         }
-    }
-}
 
-impl StdError for OpenWriteError {
-    fn description(&self) -> &str {
-        "error occurred while opening a file for writing"
-    }
-
-    fn cause(&self) -> Option<&StdError> {
-        match *self {
-            OpenWriteError::FileAlreadyExists(_) => None,
-            OpenWriteError::IOError(ref err) => Some(err),
-        }
+        Ok(())
     }
 }
 
 /// Error that may occur when accessing a file read
-#[derive(Debug)]
+#[derive(Debug, Error)]
 pub enum OpenReadError {
     /// The file does not exists.
+    #[error("Files does not exists: {0:?}")]
     FileDoesNotExist(PathBuf),
-    /// Any kind of IO error that happens when
-    /// interacting with the underlying IO device.
-    IOError(IOError),
+    /// Any kind of io::Error.
+    #[error(
+        "IoError: '{io_error:?}' happened while opening the following file for Read: {filepath}."
+    )]
+    IoError {
+        /// The underlying `io::Error`.
+        io_error: io::Error,
+        /// File path of the file that tantivy failed to open for read.
+        filepath: PathBuf,
+    },
+    /// This library does not support the index version found in file footer.
+    #[error("Index version unsupported: {0:?}")]
+    IncompatibleIndex(Incompatibility),
 }
 
-impl From<IOError> for OpenReadError {
-    fn from(err: IOError) -> OpenReadError {
-        OpenReadError::IOError(err)
+impl OpenReadError {
+    /// Wraps an io error.
+    pub fn wrap_io_error(io_error: io::Error, filepath: PathBuf) -> Self {
+        Self::IoError { io_error, filepath }
     }
 }
-
-impl fmt::Display for OpenReadError {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        match *self {
-            OpenReadError::FileDoesNotExist(ref path) => {
-                write!(f, "the file '{:?}' does not exist", path)
-            }
-            OpenReadError::IOError(ref err) => write!(
-                f,
-                "an io error occurred while opening a file for reading: '{}'",
-                err
-            ),
-        }
-    }
-}
-
-impl StdError for OpenReadError {
-    fn description(&self) -> &str {
-        "error occurred while opening a file for reading"
-    }
-
-    fn cause(&self) -> Option<&StdError> {
-        match *self {
-            OpenReadError::FileDoesNotExist(_) => None,
-            OpenReadError::IOError(ref err) => Some(err),
-        }
-    }
-}
-
 /// Error that may occur when trying to delete a file
-#[derive(Debug)]
+#[derive(Debug, Error)]
 pub enum DeleteError {
     /// The file does not exists.
+    #[error("File does not exists: '{0}'.")]
     FileDoesNotExist(PathBuf),
     /// Any kind of IO error that happens when
     /// interacting with the underlying IO device.
-    IOError(IOError),
+    #[error("The following IO error happened while deleting file '{filepath}': '{io_error:?}'.")]
+    IoError {
+        /// The underlying `io::Error`.
+        io_error: io::Error,
+        /// File path of the file that tantivy failed to delete.
+        filepath: PathBuf,
+    },
 }
 
-impl From<IOError> for DeleteError {
-    fn from(err: IOError) -> DeleteError {
-        DeleteError::IOError(err)
-    }
-}
-
-impl fmt::Display for DeleteError {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        match *self {
-            DeleteError::FileDoesNotExist(ref path) => {
-                write!(f, "the file '{:?}' does not exist", path)
-            }
-            DeleteError::IOError(ref err) => {
-                write!(f, "an io error occurred while deleting a file: '{}'", err)
-            }
-        }
-    }
-}
-
-impl StdError for DeleteError {
-    fn description(&self) -> &str {
-        "error occurred while deleting a file"
-    }
-
-    fn cause(&self) -> Option<&StdError> {
-        match *self {
-            DeleteError::FileDoesNotExist(_) => None,
-            DeleteError::IOError(ref err) => Some(err),
-        }
+impl From<Incompatibility> for OpenReadError {
+    fn from(incompatibility: Incompatibility) -> Self {
+        OpenReadError::IncompatibleIndex(incompatibility)
     }
 }

src/directory/file_slice.rs

@@ -0,0 +1,247 @@
+use stable_deref_trait::StableDeref;
+
+use crate::directory::OwnedBytes;
+use common::HasLen;
+use std::fmt;
+use std::ops::Range;
+use std::sync::{Arc, Weak};
+use std::{io, ops::Deref};
+
+pub type ArcBytes = Arc<dyn Deref<Target = [u8]> + Send + Sync + 'static>;
+pub type WeakArcBytes = Weak<dyn Deref<Target = [u8]> + Send + Sync + 'static>;
+
+/// Objects that represents files sections in tantivy.
+///
+/// By contract, whatever happens to the directory file, as long as a FileHandle
+/// is alive, the data associated with it cannot be altered or destroyed.
+///
+/// The underlying behavior is therefore specific to the `Directory` that created it.
+/// Despite its name, a `FileSlice` may or may not directly map to an actual file
+/// on the filesystem.
+pub trait FileHandle: 'static + Send + Sync + HasLen + fmt::Debug {
+    /// Reads a slice of bytes.
+    ///
+    /// This method may panic if the range requested is invalid.
+    fn read_bytes(&self, range: Range<usize>) -> io::Result<OwnedBytes>;
+}
+
+impl FileHandle for &'static [u8] {
+    fn read_bytes(&self, range: Range<usize>) -> io::Result<OwnedBytes> {
+        let bytes = &self[range];
+        Ok(OwnedBytes::new(bytes))
+    }
+}
+
+impl<B> From<B> for FileSlice
+where
+    B: StableDeref + Deref<Target = [u8]> + 'static + Send + Sync,
+{
+    fn from(bytes: B) -> FileSlice {
+        FileSlice::new(Box::new(OwnedBytes::new(bytes)))
+    }
+}
+
+/// Logical slice of read only file in tantivy.
+///
+/// It can be cloned and sliced cheaply.
+///
+#[derive(Clone)]
+pub struct FileSlice {
+    data: Arc<dyn FileHandle>,
+    range: Range<usize>,
+}
+
+impl fmt::Debug for FileSlice {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        write!(f, "FileSlice({:?}, {:?})", &self.data, self.range)
+    }
+}
+
+impl FileSlice {
+    /// Wraps a FileHandle.
+    pub fn new(file_handle: Box<dyn FileHandle>) -> Self {
+        let num_bytes = file_handle.len();
+        FileSlice::new_with_num_bytes(file_handle, num_bytes)
+    }
+
+    /// Wraps a FileHandle.
+    #[doc(hidden)]
+    pub fn new_with_num_bytes(file_handle: Box<dyn FileHandle>, num_bytes: usize) -> Self {
+        FileSlice {
+            data: Arc::from(file_handle),
+            range: 0..num_bytes,
+        }
+    }
+
+    /// Creates a fileslice that is just a view over a slice of the data.
+    ///
+    /// # Panics
+    ///
+    /// Panics if `byte_range.end` exceeds the filesize.
+    pub fn slice(&self, byte_range: Range<usize>) -> FileSlice {
+        assert!(byte_range.end <= self.len());
+        FileSlice {
+            data: self.data.clone(),
+            range: self.range.start + byte_range.start..self.range.start + byte_range.end,
+        }
+    }
+
+    /// Creates an empty FileSlice
+    pub fn empty() -> FileSlice {
+        const EMPTY_SLICE: &[u8] = &[];
+        FileSlice::from(EMPTY_SLICE)
+    }
+
+    /// Returns a `OwnedBytes` with all of the data in the `FileSlice`.
+    ///
+    /// The behavior is strongly dependant on the implementation of the underlying
+    /// `Directory` and the `FileSliceTrait` it creates.
+    /// In particular, it is  up to the `Directory` implementation
+    /// to handle caching if needed.
+    pub fn read_bytes(&self) -> io::Result<OwnedBytes> {
+        self.data.read_bytes(self.range.clone())
+    }
+
+    /// Reads a specific slice of data.
+    ///
+    /// This is equivalent to running `file_slice.slice(from, to).read_bytes()`.
+    pub fn read_bytes_slice(&self, range: Range<usize>) -> io::Result<OwnedBytes> {
+        assert!(
+            range.end <= self.len(),
+            "end of requested range exceeds the fileslice length ({} > {})",
+            range.end,
+            self.len()
+        );
+        self.data
+            .read_bytes(self.range.start + range.start..self.range.start + range.end)
+    }
+
+    /// Splits the FileSlice at the given offset and return two file slices.
+    /// `file_slice[..split_offset]` and `file_slice[split_offset..]`.
+    ///
+    /// This operation is cheap and must not copy any underlying data.
+    pub fn split(self, left_len: usize) -> (FileSlice, FileSlice) {
+        let left = self.slice_to(left_len);
+        let right = self.slice_from(left_len);
+        (left, right)
+    }
+
+    /// Splits the file slice at the given offset and return two file slices.
+    /// `file_slice[..split_offset]` and `file_slice[split_offset..]`.
+    pub fn split_from_end(self, right_len: usize) -> (FileSlice, FileSlice) {
+        let left_len = self.len() - right_len;
+        self.split(left_len)
+    }
+
+    /// Like `.slice(...)` but enforcing only the `from`
+    /// boundary.
+    ///
+    /// Equivalent to `.slice(from_offset, self.len())`
+    pub fn slice_from(&self, from_offset: usize) -> FileSlice {
+        self.slice(from_offset..self.len())
+    }
+
+    /// Returns a slice from the end.
+    ///
+    /// Equivalent to `.slice(self.len() - from_offset, self.len())`
+    pub fn slice_from_end(&self, from_offset: usize) -> FileSlice {
+        self.slice(self.len() - from_offset..self.len())
+    }
+
+    /// Like `.slice(...)` but enforcing only the `to`
+    /// boundary.
+    ///
+    /// Equivalent to `.slice(0, to_offset)`
+    pub fn slice_to(&self, to_offset: usize) -> FileSlice {
+        self.slice(0..to_offset)
+    }
+}
+
+impl FileHandle for FileSlice {
+    fn read_bytes(&self, range: Range<usize>) -> io::Result<OwnedBytes> {
+        self.read_bytes_slice(range)
+    }
+}
+
+impl HasLen for FileSlice {
+    fn len(&self) -> usize {
+        self.range.len()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{FileHandle, FileSlice};
+    use common::HasLen;
+    use std::io;
+
+    #[test]
+    fn test_file_slice() -> io::Result<()> {
+        let file_slice = FileSlice::new(Box::new(b"abcdef".as_ref()));
+        assert_eq!(file_slice.len(), 6);
+        assert_eq!(file_slice.slice_from(2).read_bytes()?.as_slice(), b"cdef");
+        assert_eq!(file_slice.slice_to(2).read_bytes()?.as_slice(), b"ab");
+        assert_eq!(
+            file_slice
+                .slice_from(1)
+                .slice_to(2)
+                .read_bytes()?
+                .as_slice(),
+            b"bc"
+        );
+        {
+            let (left, right) = file_slice.clone().split(0);
+            assert_eq!(left.read_bytes()?.as_slice(), b"");
+            assert_eq!(right.read_bytes()?.as_slice(), b"abcdef");
+        }
+        {
+            let (left, right) = file_slice.clone().split(2);
+            assert_eq!(left.read_bytes()?.as_slice(), b"ab");
+            assert_eq!(right.read_bytes()?.as_slice(), b"cdef");
+        }
+        {
+            let (left, right) = file_slice.clone().split_from_end(0);
+            assert_eq!(left.read_bytes()?.as_slice(), b"abcdef");
+            assert_eq!(right.read_bytes()?.as_slice(), b"");
+        }
+        {
+            let (left, right) = file_slice.split_from_end(2);
+            assert_eq!(left.read_bytes()?.as_slice(), b"abcd");
+            assert_eq!(right.read_bytes()?.as_slice(), b"ef");
+        }
+        Ok(())
+    }
+
+    #[test]
+    fn test_file_slice_trait_slice_len() {
+        let blop: &'static [u8] = b"abc";
+        let owned_bytes: Box<dyn FileHandle> = Box::new(blop);
+        assert_eq!(owned_bytes.len(), 3);
+    }
+
+    #[test]
+    fn test_slice_simple_read() -> io::Result<()> {
+        let slice = FileSlice::new(Box::new(&b"abcdef"[..]));
+        assert_eq!(slice.len(), 6);
+        assert_eq!(slice.read_bytes()?.as_ref(), b"abcdef");
+        assert_eq!(slice.slice(1..4).read_bytes()?.as_ref(), b"bcd");
+        Ok(())
+    }
+
+    #[test]
+    fn test_slice_read_slice() -> io::Result<()> {
+        let slice_deref = FileSlice::new(Box::new(&b"abcdef"[..]));
+        assert_eq!(slice_deref.read_bytes_slice(1..4)?.as_ref(), b"bcd");
+        Ok(())
+    }
+
+    #[test]
+    #[should_panic(expected = "end of requested range exceeds the fileslice length (10 > 6)")]
+    fn test_slice_read_slice_invalid_range_exceeds() {
+        let slice_deref = FileSlice::new(Box::new(&b"abcdef"[..]));
+        assert_eq!(
+            slice_deref.read_bytes_slice(0..10).unwrap().as_ref(),
+            b"bcd"
+        );
+    }
+}

src/directory/file_watcher.rs

@@ -0,0 +1,182 @@
+use crate::directory::{WatchCallback, WatchCallbackList, WatchHandle};
+use crc32fast::Hasher;
+use std::fs;
+use std::io;
+use std::io::BufRead;
+use std::path::Path;
+use std::sync::atomic::{AtomicUsize, Ordering};
+use std::sync::Arc;
+use std::thread;
+use std::time::Duration;
+
+pub const POLLING_INTERVAL: Duration = Duration::from_millis(if cfg!(test) { 1 } else { 500 });
+
+// Watches a file and executes registered callbacks when the file is modified.
+pub struct FileWatcher {
+    path: Arc<Path>,
+    callbacks: Arc<WatchCallbackList>,
+    state: Arc<AtomicUsize>, // 0: new, 1: runnable, 2: terminated
+}
+
+impl FileWatcher {
+    pub fn new(path: &Path) -> FileWatcher {
+        FileWatcher {
+            path: Arc::from(path),
+            callbacks: Default::default(),
+            state: Default::default(),
+        }
+    }
+
+    pub fn spawn(&self) {
+        if self
+            .state
+            .compare_exchange(0, 1, Ordering::SeqCst, Ordering::SeqCst)
+            .is_err()
+        {
+            return;
+        }
+
+        let path = self.path.clone();
+        let callbacks = self.callbacks.clone();
+        let state = self.state.clone();
+
+        thread::Builder::new()
+            .name("thread-tantivy-meta-file-watcher".to_string())
+            .spawn(move || {
+                let mut current_checksum = None;
+
+                while state.load(Ordering::SeqCst) == 1 {
+                    if let Ok(checksum) = FileWatcher::compute_checksum(&path) {
+                        // `None.unwrap_or_else(|| !checksum) != checksum` evaluates to `true`
+                        if current_checksum.unwrap_or_else(|| !checksum) != checksum {
+                            info!("Meta file {:?} was modified", path);
+                            current_checksum = Some(checksum);
+                            futures::executor::block_on(callbacks.broadcast());
+                        }
+                    }
+
+                    thread::sleep(POLLING_INTERVAL);
+                }
+            })
+            .expect("Failed to spawn meta file watcher thread");
+    }
+
+    pub fn watch(&self, callback: WatchCallback) -> WatchHandle {
+        let handle = self.callbacks.subscribe(callback);
+        self.spawn();
+        handle
+    }
+
+    fn compute_checksum(path: &Path) -> Result<u32, io::Error> {
+        let reader = match fs::File::open(path) {
+            Ok(f) => io::BufReader::new(f),
+            Err(e) => {
+                warn!("Failed to open meta file {:?}: {:?}", path, e);
+                return Err(e);
+            }
+        };
+
+        let mut hasher = Hasher::new();
+
+        for line in reader.lines() {
+            hasher.update(line?.as_bytes())
+        }
+
+        Ok(hasher.finalize())
+    }
+}
+
+impl Drop for FileWatcher {
+    fn drop(&mut self) {
+        self.state.store(2, Ordering::SeqCst);
+    }
+}
+
+#[cfg(test)]
+mod tests {
+
+    use std::mem;
+
+    use crate::directory::mmap_directory::atomic_write;
+
+    use super::*;
+
+    #[test]
+    fn test_file_watcher_drop_watcher() -> crate::Result<()> {
+        let tmp_dir = tempfile::TempDir::new()?;
+        let tmp_file = tmp_dir.path().join("watched.txt");
+
+        let counter: Arc<AtomicUsize> = Default::default();
+        let (tx, rx) = crossbeam::channel::unbounded();
+        let timeout = Duration::from_millis(100);
+
+        let watcher = FileWatcher::new(&tmp_file);
+
+        let state = watcher.state.clone();
+        assert_eq!(state.load(Ordering::SeqCst), 0);
+
+        let counter_clone = counter.clone();
+
+        let _handle = watcher.watch(WatchCallback::new(move || {
+            let val = counter_clone.fetch_add(1, Ordering::SeqCst);
+            tx.send(val + 1).unwrap();
+        }));
+
+        assert_eq!(counter.load(Ordering::SeqCst), 0);
+        assert_eq!(state.load(Ordering::SeqCst), 1);
+
+        atomic_write(&tmp_file, b"foo")?;
+        assert_eq!(rx.recv_timeout(timeout), Ok(1));
+
+        atomic_write(&tmp_file, b"foo")?;
+        assert!(rx.recv_timeout(timeout).is_err());
+
+        atomic_write(&tmp_file, b"bar")?;
+        assert_eq!(rx.recv_timeout(timeout), Ok(2));
+
+        mem::drop(watcher);
+
+        atomic_write(&tmp_file, b"qux")?;
+        thread::sleep(Duration::from_millis(10));
+        assert_eq!(counter.load(Ordering::SeqCst), 2);
+        assert_eq!(state.load(Ordering::SeqCst), 2);
+
+        Ok(())
+    }
+
+    #[test]
+    fn test_file_watcher_drop_handle() -> crate::Result<()> {
+        let tmp_dir = tempfile::TempDir::new()?;
+        let tmp_file = tmp_dir.path().join("watched.txt");
+
+        let counter: Arc<AtomicUsize> = Default::default();
+        let (tx, rx) = crossbeam::channel::unbounded();
+        let timeout = Duration::from_millis(100);
+
+        let watcher = FileWatcher::new(&tmp_file);
+
+        let state = watcher.state.clone();
+        assert_eq!(state.load(Ordering::SeqCst), 0);
+
+        let counter_clone = counter.clone();
+
+        let handle = watcher.watch(WatchCallback::new(move || {
+            let val = counter_clone.fetch_add(1, Ordering::SeqCst);
+            tx.send(val + 1).unwrap();
+        }));
+
+        assert_eq!(counter.load(Ordering::SeqCst), 0);
+        assert_eq!(state.load(Ordering::SeqCst), 1);
+
+        atomic_write(&tmp_file, b"foo")?;
+        assert_eq!(rx.recv_timeout(timeout), Ok(1));
+
+        mem::drop(handle);
+
+        atomic_write(&tmp_file, b"qux")?;
+        assert_eq!(counter.load(Ordering::SeqCst), 1);
+        assert_eq!(state.load(Ordering::SeqCst), 1);
+
+        Ok(())
+    }
+}