Introducing a column trait

And renaming the `Serializer` trait `FastFieldCodec`.

.github/workflows/coverage.yml

@@ -12,12 +12,14 @@ jobs:
     steps:
       - uses: actions/checkout@v3
       - name: Install Rust
-        run: rustup toolchain install nightly --component llvm-tools-preview
+        run: rustup toolchain install nightly --profile minimal --component llvm-tools-preview
+      - uses: Swatinem/rust-cache@v2
       - uses: taiki-e/install-action@cargo-llvm-cov
       - name: Generate code coverage
         run: cargo +nightly llvm-cov --all-features --workspace --lcov --output-path lcov.info
       - name: Upload coverage to Codecov
         uses: codecov/codecov-action@v3
+        continue-on-error: true
         with:
           token: ${{ secrets.CODECOV_TOKEN }} # not required for public repos
           files: lcov.info

.github/workflows/long_running.yml

@@ -19,11 +19,10 @@ jobs:
       uses: actions-rs/toolchain@v1
       with:
           toolchain: stable
+          profile: minimal
           override: true
-          components: rustfmt, clippy
 
     - name: Run indexing_unsorted
       run: cargo test indexing_unsorted -- --ignored
     - name: Run indexing_sorted
       run: cargo test indexing_sorted -- --ignored
-

.github/workflows/test.yml

@@ -10,34 +10,27 @@ env:
   CARGO_TERM_COLOR: always
 
 jobs:
-  test:
+  check:
 
     runs-on: ubuntu-latest
 
     steps:
     - uses: actions/checkout@v3
-    - name: Install latest nightly to test also against unstable feature flag
+
+    - name: Install nightly
       uses: actions-rs/toolchain@v1
       with:
             toolchain: nightly
-            override: true
+            profile: minimal
             components: rustfmt
-
     - name: Install stable
       uses: actions-rs/toolchain@v1
       with:
             toolchain: stable
-            override: true
-            components: rustfmt, clippy
+            profile: minimal
+            components: clippy
 
-    - name: Build
-      run: cargo build --verbose --workspace
-
-    - name: Run tests
-      run: cargo +stable test --features mmap,brotli-compression,lz4-compression,snappy-compression,zstd-compression,failpoints --verbose --workspace
-
-    - name: Run tests quickwit feature
-      run: cargo +stable test --features mmap,quickwit,failpoints --verbose --workspace
+    - uses: Swatinem/rust-cache@v2
 
     - name: Check Formatting
       run: cargo +nightly fmt --all -- --check
@@ -48,3 +41,34 @@ jobs:
         token: ${{ secrets.GITHUB_TOKEN }}
         args: --tests
 
+  test:
+
+    runs-on: ubuntu-latest
+
+    strategy:
+      matrix:
+        features: [
+            { label: "all", flags: "mmap,brotli-compression,lz4-compression,snappy-compression,zstd-compression,failpoints" },
+            { label: "quickwit", flags: "mmap,quickwit,failpoints" }
+        ]
+
+    name: test-${{ matrix.features.label}}
+
+    steps:
+    - uses: actions/checkout@v3
+
+    - name: Install stable
+      uses: actions-rs/toolchain@v1
+      with:
+            toolchain: stable
+            profile: minimal
+            override: true
+
+    - uses: taiki-e/install-action@nextest
+    - uses: Swatinem/rust-cache@v2
+
+    - name: Run tests
+      run: cargo +stable nextest run --features ${{ matrix.features.flags }} --verbose --workspace
+
+    - name: Run doctests
+      run: cargo +stable test --doc --features ${{ matrix.features.flags }} --verbose --workspace

ARCHITECTURE.md

@@ -10,6 +10,7 @@ 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,
@@ -22,27 +23,28 @@ rapidly select all documents matching a given predicate (also known as a query)
 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/](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...
+### Index and Segments
 
-A tantivy index is a collection of smaller independent immutable 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 ```
+ ```segment-id . ext```
 
 The extension signals which data structure (or [`SegmentComponent`](src/core/segment_component.rs)) is stored in the file.
 
@@ -52,17 +54,15 @@ On commit, one segment per indexing thread is written to disk, and the `meta.jso
 
 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 remove from [alive bitset file](src/fastfield/alive_bitset.rs) that represents the bitset of the alive document ids.
-Like all segment files, this file is immutable. Because it is possible to have more than one alive bitset file at a given instant, the alive bitset filename has the format ``` segment_id . commit_opstamp . del```.
+Like all segment files, this file is immutable. Because it is possible to have more than one alive bitset file at a given instant, the alive bitset 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)`.
@@ -74,6 +74,7 @@ The DocIds are simply allocated in the order documents are added to the index.
 
 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.
 
@@ -104,6 +105,7 @@ Tantivy's document follows a very strict schema, decided before building any ind
 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
@@ -117,9 +119,10 @@ As of today, tantivy's schema imposes a 1:1 relationship between a field that is
 
 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.
+## General information about these data structures
 
 All data structures in tantivy, have:
+
 - a writer
 - a serializer
 - a reader
@@ -132,7 +135,7 @@ 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!
+## [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
@@ -146,6 +149,7 @@ Once the top 10 documents have been identified, we fetch them from the store, an
 **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.
@@ -154,8 +158,7 @@ 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!
+## [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.
@@ -163,7 +166,7 @@ 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
 
-```
+```rust
 min_value + fetch_bits(num_bits * doc_id..num_bits * (doc_id+1))
 ```
 
@@ -190,7 +193,7 @@ For advanced search engine, it is possible to store all of the features required
 
 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 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).
@@ -215,19 +218,18 @@ The inverted index actually consists of two data structures chained together.
 
 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)!
+## [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!
+## [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.
@@ -257,7 +259,6 @@ we advance the position reader by the number of term frequencies of the current
 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.
@@ -268,7 +269,6 @@ Text processing can be configured by selecting an off-the-shelf [`Tokenizer`](./
 
 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.

CHANGELOG.md

@@ -1,5 +1,6 @@
 Tantivy 0.19
 ================================
+
 - Updated [Date Field Type](https://github.com/quickwit-oss/tantivy/pull/1396)
   The `DateTime` type has been updated to hold timestamps with microseconds precision.
   `DateOptions` and `DatePrecision` have been added to configure Date fields. The precision is used to hint on fast values compression. Otherwise, seconds precision is used everywhere else (i.e terms, indexing).
@@ -7,6 +8,7 @@ Tantivy 0.19
 
 Tantivy 0.18
 ================================
+
 - For date values `chrono` has been replaced with `time` (@uklotzde) #1304 :
   - The `time` crate is re-exported as `tantivy::time` instead of `tantivy::chrono`.
   - The type alias `tantivy::DateTime` has been removed.
@@ -22,6 +24,7 @@ Tantivy 0.18
 
 Tantivy 0.17
 ================================
+
 - LogMergePolicy now triggers merges if the ratio of deleted documents reaches a threshold (@shikhar @fulmicoton) [#115](https://github.com/quickwit-oss/tantivy/issues/115)
 - Adds a searcher Warmer API (@shikhar @fulmicoton)
 - Change to non-strict schema. Ignore fields in data which are not defined in schema. Previously this returned an error. #1211
@@ -36,33 +39,39 @@ Tantivy 0.17
 
 Tantivy 0.16.2
 ================================
+
 - Bugfix in FuzzyTermQuery. (transposition_cost_one was not doing anything)
 
 Tantivy 0.16.1
 ========================
+
 - Major Bugfix on multivalued fastfield.  #1151
 - Demux operation (@PSeitz)
 
 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
 
+- 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
@@ -84,9 +93,9 @@ Tantivy 0.15.0
 - 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.
@@ -105,16 +114,19 @@ 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
@@ -129,6 +141,7 @@ so that we can discuss possible solutions.
 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 {
@@ -136,7 +149,9 @@ 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
@@ -144,6 +159,7 @@ 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)
@@ -154,30 +170,32 @@ Tantivy 0.12.0
 ## How to update?
 
 Crates relying on custom tokenizer, or registering tokenizer in the manager will require some
-minor changes. Check https://github.com/quickwit-oss/tantivy/blob/main/examples/custom_tokenizer.rs
+minor changes. Check <https://github.com/quickwit-oss/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
 
+- 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.
+  - 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)
@@ -208,7 +226,6 @@ Tantivy 0.10.1
 Avoid watching the mmap directory until someone effectively creates a reader that uses
 this functionality.
 
-
 Tantivy 0.10.0
 =====================
 
@@ -224,6 +241,7 @@ Tantivy 0.10.0
 
 Minor
 ---------
+
 - Switched to Rust 2018 (@uvd)
 - Small simplification of the code.
 Calling .freq() or .doc() when .advance() has never been called
@@ -231,8 +249,7 @@ 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
+- `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)
@@ -248,16 +265,17 @@ Your program should be usable as is.
 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)
@@ -301,37 +319,40 @@ To update from tantivy 0.8, you will need to go through the following steps.
 
     ```
 
-
 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
 
 Tantivy 0.7
 =====================
+
 - Skip data for doc ids and positions (@fulmicoton),
   greatly improving performance
 - Tantivy error now rely on the failure crate (@drusellers)
@@ -341,15 +362,15 @@ Tantivy 0.7
 
 Tantivy 0.6.1
 =========================
+
 - Bugfix #324. GC removing was removing file that were still in useful
 - Added support for parsing AllQuery and RangeQuery via QueryParser
-    - AllQuery: `*`
-    - RangeQuery:
-        - Inclusive `field:[startIncl to endIncl]`
-        - Exclusive `field:{startExcl to endExcl}`
-        - Mixed `field:[startIncl to endExcl}` and vice versa
-        - Unbounded `field:[start to *]`, `field:[* to end]`
-
+  - AllQuery: `*`
+  - RangeQuery:
+    - Inclusive `field:[startIncl to endIncl]`
+    - Exclusive `field:{startExcl to endExcl}`
+    - Mixed `field:[startIncl to endExcl}` and vice versa
+    - Unbounded `field:[start to *]`, `field:[* to end]`
 
 Tantivy 0.6
 ==========================
@@ -362,58 +383,53 @@ to this release!
 - 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.
+  - Completely uncompressed
+  - Internally: One u64 fast field for indexes, one fast field for the bytes themselves.
 - Add NGram token support (@drusellers)
 - Add Stopword Filter support (@drusellers)
 - Add a FuzzyTermQuery (@drusellers)
 - Add a RegexQuery (@drusellers)
 - Various performance improvements (@fulmicoton)_
 
-
 Tantivy 0.5.2
 ===========================
+
 - bugfix #274
 - bugfix #280
 - bugfix #289
 
-
 Tantivy 0.5.1
 ==========================
-- bugfix #254 : tantivy failed if no documents in a segment contained a specific field.
 
+- bugfix #254 : tantivy failed if no documents in a segment contained a specific field.
 
 Tantivy 0.5
 ==========================
+
 - Faceting
 - RangeQuery
 - Configurable tokenization pipeline
 - Bugfix in PhraseQuery
 - Various query optimisation
 - Allowing very large indexes
-    - 64 bits file address
-    - Smarter encoding of the `TermInfo` objects
-
-
+  - 64 bits file address
+  - Smarter encoding of the `TermInfo` objects
 
 Tantivy 0.4.3
 ==========================
 
 - Bugfix race condition when deleting files. (#198)
 
-
 Tantivy 0.4.2
 ==========================
 
 - Prevent usage of AVX2 instructions (#201)
 
-
 Tantivy 0.4.1
 ==========================
 
 - Bugfix for non-indexed fields. (#199)
 
-
 Tantivy 0.4.0
 ==========================
 
@@ -428,37 +444,31 @@ Tantivy 0.4.0
   - Searching for a non-indexed field returns an explicit Error
   - Phrase query for non-tokenized field are not tokenized by the query parser.
 - Faster/Better indexing (@fulmicoton)
-    - using murmurhash2
-    - faster merging
-    - more memory efficient fast field writer (@lnicola )
-    - better handling of collisions
-    - lesser memory usage
+  - using murmurhash2
+  - faster merging
+  - more memory efficient fast field writer (@lnicola )
+  - better handling of collisions
+  - lesser memory usage
 - Added API, most notably to iterate over ranges of terms (@fulmicoton)
 - Bugfix that was preventing to unmap segment files, on index drop (@fulmicoton)
 - Made the doc! macro public (@fulmicoton)
 - Added an alternative implementation of the streaming dictionary (@fulmicoton)
 
-
-
 Tantivy 0.3.1
 ==========================
 
 - Expose a method to trigger files garbage collection
 
-
-
 Tantivy 0.3
 ==========================
 
-
 Special thanks to @Kodraus @lnicola @Ameobea @manuel-woelker @celaus
 for their contribution to this release.
 
 Thanks also to everyone in tantivy gitter chat
 for their advise and company :)
 
-https://gitter.im/tantivy-search/tantivy
-
+<https://gitter.im/tantivy-search/tantivy>
 
 Warning:
 
@@ -467,19 +477,16 @@ code and index format.
 You should not expect backward compatibility before
 tantivy 1.0.
 
-
-
 New Features
 ------------
 
 - Delete. You can now delete documents from an index.
 - Support for windows (Thanks to @lnicola)
 
-
 Various Bugfixes & small improvements
 ----------------------------------------
 
-- Added CI for Windows (https://ci.appveyor.com/project/fulmicoton/tantivy)
+- Added CI for Windows (<https://ci.appveyor.com/project/fulmicoton/tantivy>)
 Thanks to @KodrAus ! (#108)
 - Various dependy version update (Thanks to @Ameobea) #76
 - Fixed several race conditions in `Index.wait_merge_threads`
@@ -491,7 +498,3 @@ Thanks to @KodrAus ! (#108)
 - Building binary targets for tantivy-cli (Thanks to @KodrAus)
 - Misc invisible bug fixes, and code cleanup.
 - Use
-
-
-
-

Cargo.toml

@@ -60,7 +60,6 @@ pretty_assertions = "1.2.1"
 serde_cbor = { version = "0.11.2", optional = true }
 async-trait = "0.1.53"
 arc-swap = "1.5.0"
-gcd = "2.1.0"
 
 [target.'cfg(windows)'.dependencies]
 winapi = "0.3.9"

README.md

@@ -5,7 +5,6 @@
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![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)
 
 **Tantivy** is a **full-text search engine library** written in Rust.
@@ -16,7 +15,7 @@ to build such a search engine.
 
 Tantivy is, in fact, strongly inspired by Lucene's design.
 
-If you are looking for an alternative to Elasticsearch or Apache Solr, check out [Quickwit](https://github.com/quickwit-oss/quickwit), our search engine built on top of Tantivy. 
+If you are looking for an alternative to Elasticsearch or Apache Solr, check out [Quickwit](https://github.com/quickwit-oss/quickwit), our search engine built on top of Tantivy.
 
 # Benchmark
 
@@ -57,7 +56,6 @@ Your mileage WILL vary depending on the nature of queries and their load.
 
 Distributed search is out of the scope of Tantivy, but if you are looking for this feature, check out [Quickwit](https://github.com/quickwit-oss/quickwit/).
 
-
 # Getting started
 
 Tantivy works on stable Rust (>= 1.27) and supports Linux, macOS, and Windows.
@@ -125,7 +123,8 @@ By default, `rustc` compiles everything in the `examples/` directory in debug mo
 rust-gdb target/debug/examples/$EXAMPLE_NAME
 $ gdb run
 ```
-# Companies Using Tantivy 
+
+# Companies Using Tantivy
 
 <p align="left">
 <img align="center" src="doc/assets/images/Nuclia.png#gh-light-mode-only" alt="Nuclia" height="25" width="auto" /> &nbsp;
@@ -134,11 +133,12 @@ $ gdb run
 <img align="center" src="doc/assets/images/nuclia-dark-theme.png#gh-dark-mode-only" alt="Nuclia" height="35" width="auto" /> &nbsp;
 <img align="center" src="doc/assets/images/humanfirst.ai-dark-theme.png#gh-dark-mode-only" alt="Humanfirst.ai" height="25" width="auto" />&nbsp; &nbsp;
 <img align="center" src="doc/assets/images/element-dark-theme.png#gh-dark-mode-only" alt="Element.io" height="25" width="auto" />
-</p> 
-
+</p>
 
 # FAQ
+
 ### Can I use Tantivy in other languages?
+
 - Python → [tantivy-py](https://github.com/quickwit-oss/tantivy-py)
 - Ruby → [tantiny](https://github.com/baygeldin/tantiny)
 
@@ -152,13 +152,17 @@ You can also find other bindings on [GitHub](https://github.com/search?q=tantivy
 - and [more](https://github.com/search?q=tantivy)!
 
 ### On average, how much faster is Tantivy compared to Lucene?
+
 - According to our [search latency benchmark](https://tantivy-search.github.io/bench/), Tantivy is approximately 2x faster than Lucene.
 
 ### Does tantivy support incremental indexing?
+
 - Yes.
 
 ### How can I edit documents?
+
 - Data in tantivy is immutable. To edit a document, the document needs to be deleted and reindexed.
 
 ### When will my documents be searchable during indexing?
+
 - Documents will be searchable after a `commit` is called on an `IndexWriter`. Existing `IndexReader`s will also need to be reloaded in order to reflect the changes. Finally, changes are only visible to newly acquired `Searcher`.

bitpacker/src/bitpacker.rs

@@ -82,14 +82,16 @@ impl BitUnpacker {
         }
     }
 
+    pub fn bit_width(&self) -> u8 {
+        self.num_bits as u8
+    }
+
     #[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_in_bits = idx * self.num_bits;
         let addr = addr_in_bits >> 3;
         let bit_shift = addr_in_bits & 7;
         debug_assert!(
@@ -101,7 +103,7 @@ impl BitUnpacker {
             .unwrap();
         let val_unshifted_unmasked: u64 = u64::from_le_bytes(bytes);
         let val_shifted = (val_unshifted_unmasked >> bit_shift) as u64;
-        val_shifted & mask
+        val_shifted & self.mask
     }
 }
 

bitpacker/src/blocked_bitpacker.rs

@@ -58,6 +58,10 @@ fn metadata_test() {
     assert_eq!(meta.num_bits(), 6);
 }
 
+fn mem_usage<T>(items: &Vec<T>) -> usize {
+    items.capacity() * std::mem::size_of::<T>()
+}
+
 impl BlockedBitpacker {
     pub fn new() -> Self {
         let mut compressed_blocks = vec![];
@@ -73,10 +77,8 @@ impl BlockedBitpacker {
     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())
+            + mem_usage(&self.offset_and_bits)
+            + mem_usage(&self.buffer)
     }
 
     #[inline]

common/src/serialize.rs

@@ -19,7 +19,7 @@ pub trait DeserializeFrom<T: BinarySerializable> {
 
 /// Implement deserialize from &[u8] for all types which implement BinarySerializable.
 ///
-/// TryFrom would actually be preferrable, but not possible because of the orphan
+/// TryFrom would actually be preferable, 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> {

doc/src/SUMMARY.md

@@ -1,7 +1,5 @@
 # Summary
 
-
-
 [Avant Propos](./avant-propos.md)
 
 - [Segments](./basis.md)

doc/src/avant-propos.md

@@ -3,7 +3,7 @@
 > Tantivy is a **search** engine **library** for Rust.
 
 If you are familiar with Lucene, it's an excellent approximation to consider tantivy as Lucene for rust. tantivy is heavily inspired by Lucene's design and
-they both have the same scope and targetted use cases.
+they both have the same scope and targeted use cases.
 
 If you are not familiar with Lucene, let's break down our little tagline.
 
@@ -31,4 +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/basis.md

@@ -11,7 +11,7 @@ directory shipped with tantivy is the `MmapDirectory`.
 While this design has some downsides, this greatly simplifies the source code of
 tantivy. Caching is also entirely delegated to the OS.
 
-`tantivy` works entirely (or almost) by directly reading the datastructures as they are layed on disk. As a result, the act of opening an indexing does not involve loading different datastructures from the disk into random access memory : starting a process, opening an index, and performing your first query can typically be done in a matter of milliseconds.
+`tantivy` works entirely (or almost) by directly reading the datastructures as they are laid on disk. As a result, the act of opening an indexing does not involve loading different datastructures from the disk into random access memory : starting a process, opening an index, and performing your first query can typically be done in a matter of milliseconds.
 
 This is an interesting property for a command line search engine, or for some multi-tenant log search engine : spawning a new process for each new query can be a perfectly sensible solution in some use case.
 
@@ -22,7 +22,6 @@ Of course this is crucial to reduce IO, and ensure that as much of our index can
 Also, whenever possible its data is accessed sequentially. Of course, this is an amazing property when tantivy needs to access the data from your spinning hard disk, but this is also
 critical for performance, if your data is read from and an `SSD` or even already in your pagecache.
 
-
 ## Segments, and the log method
 
 That kind of compact layout comes at one cost: it prevents our datastructures from being dynamic.
@@ -53,11 +52,7 @@ to get tantivy to fit your use case:
 
 *Example 2* You could also disable your merge policy and enforce daily segments. Removing data after one week can then be done very efficiently by just editing the `meta.json` and deleting the files associated to segment `D-7`.
 
-
-
-
-
-# Merging
+## Merging
 
 As you index more and more data, your index will accumulate more and more segments.
 Having a lot of small segments is not really optimal. There is a bit of redundancy in having
@@ -66,11 +61,7 @@ all these term dictionary. Also when searching, we will need to do term lookups
 That's where merging or compacting comes into place. Tantivy will continuously consider merge
 opportunities and start merging segments in the background.
 
-
-# Indexing throughput, number of indexing threads
-
-
-
+## Indexing throughput, number of indexing threads
 
 [^1]: This may eventually change.
 

doc/src/examples.md

@@ -1,3 +1,3 @@
 # Examples
 
-- [Basic search](/examples/basic_search.html)
+- [Basic search](/examples/basic_search.html)

doc/src/index_sorting.md

@@ -1,11 +1,11 @@
 
 - [Index Sorting](#index-sorting)
-    + [Why Sorting](#why-sorting)
-        * [Compression](#compression)
-        * [Top-N Optimization](#top-n-optimization)
-        * [Pruning](#pruning)
-        * [Other](#other)
-    + [Usage](#usage)
+  - [Why Sorting](#why-sorting)
+    - [Compression](#compression)
+    - [Top-N Optimization](#top-n-optimization)
+    - [Pruning](#pruning)
+    - [Other](#other)
+  - [Usage](#usage)
 
 # Index Sorting
 
@@ -15,32 +15,34 @@ Tantivy allows you to sort the index according to a property.
 
 Presorting an index has several advantages:
 
-###### Compression
+### 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]. 
+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
+Compression ratio is mainly affected on the fast field of the sorted property, every thing else is likely unaffected.
 
-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. 
+### 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
+### 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?
+### 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/quickwit-oss/tantivy/blob/000d76b11a139a84b16b9b95060a1c93e8b9851c/src/core/index_meta.rs#L238) on `IndexSettings` and passing it to a `IndexBuilder`. As of Tantivy 0.16 only fast fields are allowed to be used.
 
-```
+```rust
 let settings = IndexSettings {
     sort_by_field: Some(IndexSortByField {
         field: "intval".to_string(),
@@ -58,4 +60,3 @@ let index = index_builder.create_in_ram().unwrap();
 Sorting an index is applied in the serialization step. In general there are two serialization steps: [Finishing a single segment](https://github.com/quickwit-oss/tantivy/blob/000d76b11a139a84b16b9b95060a1c93e8b9851c/src/indexer/segment_writer.rs#L338) and [merging multiple segments](https://github.com/quickwit-oss/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/json.md

@@ -21,16 +21,17 @@ For instance,  if user is a json field, the following document:
 ```
 
 emits the following tokens:
--  ("name", Text, "Paul")
--  ("name", Text, "Masurel")
--  ("address.city", Text, "Tokyo")
--  ("address.country", Text, "Japan")
--  ("created_at", Date, 15420648505)
 
+- ("name", Text, "Paul")
+- ("name", Text, "Masurel")
+- ("address.city", Text, "Tokyo")
+- ("address.country", Text, "Japan")
+- ("created_at", Date, 15420648505)
 
-# Bytes-encoding and lexicographical sort.
+## Bytes-encoding and lexicographical sort
 
 Like any other terms, these triplets are encoded into a binary format as follows.
+
 - `json_path`: the json path is a sequence of "segments". In the example above, `address.city`
 is just a debug representation of the json path `["address", "city"]`.
 Its representation is done by separating segments by a unicode char `\x01`, and ending the path by `\x00`.
@@ -41,16 +42,16 @@ This representation is designed to align the natural sort of Terms with the lexi
 of their binary representation (Tantivy's dictionary (whether fst or sstable) is sorted and does prefix encoding).
 
 In the example above, the terms will be sorted as
--  ("address.city", Text, "Tokyo")
--  ("address.country", Text, "Japan")
--  ("name", Text, "Masurel")
--  ("name", Text, "Paul")
--  ("created_at", Date, 15420648505)
+
+- ("address.city", Text, "Tokyo")
+- ("address.country", Text, "Japan")
+- ("name", Text, "Masurel")
+- ("name", Text, "Paul")
+- ("created_at", Date, 15420648505)
 
 As seen in "pitfalls", we may end up having to search for a value for a same path in several different fields. Putting the field code after the path makes it maximizes compression opportunities but also increases the chances for the two terms to end up in the actual same term dictionary block.
 
-
-# Pitfalls, limitation and corner cases.
+## Pitfalls, limitation and corner cases
 
 Json gives very little information about the type of the literals it stores.
 All numeric types end up mapped as a "Number" and there are no types for dates.
@@ -70,19 +71,21 @@ For instance, we do not even know if the type is a number or string based.
 
 So the query
 
-```
+```rust
 my_path.my_segment:233
 ```
 
 Will be interpreted as
-`(my_path.my_segment, String, 233) or (my_path.my_segment, u64, 233)`
+
+```rust
+(my_path.my_segment, String, 233) or (my_path.my_segment, u64, 233)
+```
 
 Likewise, we need to emit two tokens if the query contains an rfc3999 date.
 Indeed the date could have been actually a single token inside the text of a document at ingestion time. Generally speaking, we will always at least emit a string token in query parsing, and sometimes more.
 
 If one more json field is defined, things get even more complicated.
 
-
 ## Default json field
 
 If the schema contains a text field called "text" and a json field that is set as a default field:
@@ -96,11 +99,11 @@ This is a product decision.
 The user can still target the JSON field by specifying its name explicitly:
 `json_dynamic.text:hello`.
 
-## Range queries are not supported.
+## Range queries are not supported
 
 Json field do not support range queries.
 
-## Arrays do not work like nested object.
+## Arrays do not work like nested object
 
 If json object contains an array, a search query might return more documents
 than what might be expected.
@@ -120,9 +123,8 @@ Let's take an example.
 Despite the array structure, a document in tantivy is a bag of terms.
 The query:
 
-```
+```rust
 cart.product_type:sneakers AND cart.attributes.color:red
 ```
 
 Actually match the document above.
-

examples/custom_collector.rs

@@ -7,10 +7,11 @@
 // Of course, you can have a look at the tantivy's built-in collectors
 // such as the `CountCollector` for more examples.
 
+use fastfield_codecs::Column;
 // ---
 // Importing tantivy...
 use tantivy::collector::{Collector, SegmentCollector};
-use tantivy::fastfield::{DynamicFastFieldReader, FastFieldReader};
+use tantivy::fastfield::DynamicFastFieldReader;
 use tantivy::query::QueryParser;
 use tantivy::schema::{Field, Schema, FAST, INDEXED, TEXT};
 use tantivy::{doc, Index, Score, SegmentReader};
@@ -103,7 +104,7 @@ 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;
+        let value = self.fast_field_reader.get_val(doc as u64) as f64;
         self.stats.count += 1;
         self.stats.sum += value;
         self.stats.squared_sum += value * value;

examples/custom_tokenizer.rs

@@ -50,7 +50,7 @@ fn main() -> tantivy::Result<()> {
     // for your unit tests... Or this example.
     let index = Index::create_in_ram(schema.clone());
 
-    // here we are registering our custome tokenizer
+    // here we are registering our custom tokenizer
     // this will store tokens of 3 characters each
     index
         .tokenizers()

examples/warmer.rs

@@ -2,8 +2,8 @@ use std::cmp::Reverse;
 use std::collections::{HashMap, HashSet};
 use std::sync::{Arc, RwLock, Weak};
 
+use fastfield_codecs::Column;
 use tantivy::collector::TopDocs;
-use tantivy::fastfield::FastFieldReader;
 use tantivy::query::QueryParser;
 use tantivy::schema::{Field, Schema, FAST, TEXT};
 use tantivy::{
@@ -52,7 +52,7 @@ impl Warmer for DynamicPriceColumn {
             let product_id_reader = segment.fast_fields().u64(self.field)?;
             let product_ids: Vec<ProductId> = segment
                 .doc_ids_alive()
-                .map(|doc| product_id_reader.get(doc))
+                .map(|doc| product_id_reader.get_val(doc as u64))
                 .collect();
             let mut prices_it = self.price_fetcher.fetch_prices(&product_ids).into_iter();
             let mut price_vals: Vec<Price> = Vec::new();

fastfield_codecs/Cargo.toml

@@ -11,11 +11,13 @@ description = "Fast field codecs used by tantivy"
 [dependencies]
 common = { version = "0.3", path = "../common/", package = "tantivy-common" }
 tantivy-bitpacker = { version="0.2", path = "../bitpacker/" }
-prettytable-rs = {version="0.8.0", optional= true}
+ownedbytes = { version = "0.3.0", path = "../ownedbytes" }
+prettytable-rs = {version="0.9.0", optional= true}
 rand = {version="0.8.3", optional= true}
 
 [dev-dependencies]
 more-asserts = "0.3.0"
+proptest = "1.0.0"
 rand = "0.8.3"
 
 [features]

fastfield_codecs/benches/bench.rs

@@ -4,13 +4,9 @@ extern crate test;
 
 #[cfg(test)]
 mod tests {
-    use fastfield_codecs::bitpacked::{BitpackedFastFieldReader, BitpackedFastFieldSerializer};
-    use fastfield_codecs::linearinterpol::{
-        LinearInterpolFastFieldReader, LinearInterpolFastFieldSerializer,
-    };
-    use fastfield_codecs::multilinearinterpol::{
-        MultiLinearInterpolFastFieldReader, MultiLinearInterpolFastFieldSerializer,
-    };
+    use fastfield_codecs::bitpacked::BitpackedCodec;
+    use fastfield_codecs::blockwise_linear::BlockwiseLinearCodec;
+    use fastfield_codecs::linear::LinearCodec;
     use fastfield_codecs::*;
 
     fn get_data() -> Vec<u64> {
@@ -29,72 +25,59 @@ mod tests {
     fn value_iter() -> impl Iterator<Item = u64> {
         0..20_000
     }
-    fn bench_get<S: FastFieldCodecSerializer, R: FastFieldCodecReader>(
-        b: &mut Bencher,
-        data: &[u64],
-    ) {
+    fn bench_get<Codec: FastFieldCodec>(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();
+        Codec::serialize(&mut bytes, &data).unwrap();
+        let reader = Codec::open_from_bytes(OwnedBytes::new(bytes)).unwrap();
         b.iter(|| {
+            let mut sum = 0u64;
             for pos in value_iter() {
-                reader.get_u64(pos as u64, &bytes);
+                let val = reader.get_val(pos as u64);
+                debug_assert_eq!(data[pos as usize], val);
+                sum = sum.wrapping_add(val);
             }
+            sum
         });
     }
-    fn bench_create<S: FastFieldCodecSerializer>(b: &mut Bencher, data: &[u64]) {
-        let mut bytes = vec![];
+    fn bench_create<Codec: FastFieldCodec>(b: &mut Bencher, data: &[u64]) {
+        let mut bytes = Vec::new();
         b.iter(|| {
-            S::serialize(
-                &mut bytes,
-                &data,
-                stats_from_vec(data),
-                data.iter().cloned(),
-                data.iter().cloned(),
-            )
-            .unwrap();
+            bytes.clear();
+            Codec::serialize(&mut bytes, &data).unwrap();
         });
     }
 
+    use ownedbytes::OwnedBytes;
     use test::Bencher;
     #[bench]
     fn bench_fastfield_bitpack_create(b: &mut Bencher) {
         let data: Vec<_> = get_data();
-        bench_create::<BitpackedFastFieldSerializer>(b, &data);
+        bench_create::<BitpackedCodec>(b, &data);
     }
     #[bench]
     fn bench_fastfield_linearinterpol_create(b: &mut Bencher) {
         let data: Vec<_> = get_data();
-        bench_create::<LinearInterpolFastFieldSerializer>(b, &data);
+        bench_create::<LinearCodec>(b, &data);
     }
     #[bench]
     fn bench_fastfield_multilinearinterpol_create(b: &mut Bencher) {
         let data: Vec<_> = get_data();
-        bench_create::<MultiLinearInterpolFastFieldSerializer>(b, &data);
+        bench_create::<BlockwiseLinearCodec>(b, &data);
     }
     #[bench]
     fn bench_fastfield_bitpack_get(b: &mut Bencher) {
         let data: Vec<_> = get_data();
-        bench_get::<BitpackedFastFieldSerializer, BitpackedFastFieldReader>(b, &data);
+        bench_get::<BitpackedCodec>(b, &data);
     }
     #[bench]
     fn bench_fastfield_linearinterpol_get(b: &mut Bencher) {
         let data: Vec<_> = get_data();
-        bench_get::<LinearInterpolFastFieldSerializer, LinearInterpolFastFieldReader>(b, &data);
+        bench_get::<LinearCodec>(b, &data);
     }
     #[bench]
     fn bench_fastfield_multilinearinterpol_get(b: &mut Bencher) {
         let data: Vec<_> = get_data();
-        bench_get::<MultiLinearInterpolFastFieldSerializer, MultiLinearInterpolFastFieldReader>(
-            b, &data,
-        );
+        bench_get::<BlockwiseLinearCodec>(b, &data);
     }
     pub fn stats_from_vec(data: &[u64]) -> FastFieldStats {
         let min_value = data.iter().cloned().min().unwrap_or(0);

fastfield_codecs/src/bitpacked.rs

@@ -1,37 +1,26 @@
 use std::io::{self, Write};
 
 use common::BinarySerializable;
+use ownedbytes::OwnedBytes;
 use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
 
-use crate::{FastFieldCodecReader, FastFieldCodecSerializer, FastFieldDataAccess, FastFieldStats};
+use crate::{Column, FastFieldCodec, FastFieldCodecType};
 
 /// Depending on the field type, a different
 /// fast field is required.
 #[derive(Clone)]
-pub struct BitpackedFastFieldReader {
+pub struct BitpackedReader {
+    data: OwnedBytes,
     bit_unpacker: BitUnpacker,
-    pub min_value_u64: u64,
-    pub max_value_u64: u64,
+    min_value_u64: u64,
+    max_value_u64: u64,
+    num_vals: u64,
 }
 
-impl 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,
-        })
-    }
+impl Column for BitpackedReader {
     #[inline]
-    fn get_u64(&self, doc: u64, data: &[u8]) -> u64 {
-        self.min_value_u64 + self.bit_unpacker.get(doc, data)
+    fn get_val(&self, doc: u64) -> u64 {
+        self.min_value_u64 + self.bit_unpacker.get(doc, &self.data)
     }
     #[inline]
     fn min_value(&self) -> u64 {
@@ -41,16 +30,21 @@ impl FastFieldCodecReader for BitpackedFastFieldReader {
     fn max_value(&self) -> u64 {
         self.max_value_u64
     }
+    #[inline]
+    fn num_vals(&self) -> u64 {
+        self.num_vals
+    }
 }
-pub struct BitpackedFastFieldSerializerLegacy<'a, W: 'a + Write> {
+pub struct BitpackedSerializerLegacy<'a, W: 'a + Write> {
     bit_packer: BitPacker,
     write: &'a mut W,
     min_value: u64,
+    num_vals: u64,
     amplitude: u64,
     num_bits: u8,
 }
 
-impl<'a, W: Write> BitpackedFastFieldSerializerLegacy<'a, W> {
+impl<'a, W: Write> BitpackedSerializerLegacy<'a, W> {
     /// Creates a new fast field serializer.
     ///
     /// The serializer in fact encode the values by bitpacking
@@ -63,15 +57,16 @@ impl<'a, W: Write> BitpackedFastFieldSerializerLegacy<'a, W> {
         write: &'a mut W,
         min_value: u64,
         max_value: u64,
-    ) -> io::Result<BitpackedFastFieldSerializerLegacy<'a, W>> {
+    ) -> io::Result<BitpackedSerializerLegacy<'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 {
+        Ok(BitpackedSerializerLegacy {
             bit_packer,
             write,
             min_value,
+            num_vals: 0,
             amplitude,
             num_bits,
         })
@@ -82,21 +77,45 @@ impl<'a, W: Write> BitpackedFastFieldSerializerLegacy<'a, W> {
         let val_to_write: u64 = val - self.min_value;
         self.bit_packer
             .write(val_to_write, self.num_bits, &mut self.write)?;
+        self.num_vals += 1;
         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)?;
+        self.num_vals.serialize(&mut self.write)?;
         Ok(())
     }
 }
 
-pub struct BitpackedFastFieldSerializer {}
+pub struct BitpackedCodec;
+
+impl FastFieldCodec for BitpackedCodec {
+    /// The CODEC_TYPE is an enum value used for serialization.
+    const CODEC_TYPE: FastFieldCodecType = FastFieldCodecType::Bitpacked;
+
+    type Reader = BitpackedReader;
+
+    /// Opens a fast field given a file.
+    fn open_from_bytes(bytes: OwnedBytes) -> io::Result<Self::Reader> {
+        let footer_offset = bytes.len() - 24;
+        let (data, mut footer) = bytes.split(footer_offset);
+        let min_value = u64::deserialize(&mut footer)?;
+        let amplitude = u64::deserialize(&mut footer)?;
+        let num_vals = 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(BitpackedReader {
+            data,
+            bit_unpacker,
+            min_value_u64: min_value,
+            max_value_u64: max_value,
+            num_vals,
+        })
+    }
 
-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
@@ -105,51 +124,41 @@ impl FastFieldCodecSerializer for BitpackedFastFieldSerializer {
     /// 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: &dyn 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)?;
+    fn serialize(write: &mut impl Write, fastfield_accessor: &dyn Column) -> io::Result<()> {
+        let mut serializer = BitpackedSerializerLegacy::open(
+            write,
+            fastfield_accessor.min_value(),
+            fastfield_accessor.max_value(),
+        )?;
 
-        for val in data_iter {
+        for val in fastfield_accessor.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;
+
+    fn estimate(fastfield_accessor: &impl Column) -> Option<f32> {
+        let amplitude = fastfield_accessor.max_value() - fastfield_accessor.min_value();
         let num_bits = compute_num_bits(amplitude);
         let num_bits_uncompressed = 64;
-        num_bits as f32 / num_bits_uncompressed as f32
+        Some(num_bits as f32 / num_bits_uncompressed as f32)
     }
 }
 
 #[cfg(test)]
 mod tests {
     use super::*;
-    use crate::tests::get_codec_test_data_sets;
+    use crate::tests::get_codec_test_datasets;
 
     fn create_and_validate(data: &[u64], name: &str) {
-        crate::tests::create_and_validate::<BitpackedFastFieldSerializer, BitpackedFastFieldReader>(
-            data, name,
-        );
+        crate::tests::create_and_validate::<BitpackedCodec>(data, name);
     }
 
     #[test]
     fn test_with_codec_data_sets() {
-        let data_sets = get_codec_test_data_sets();
+        let data_sets = get_codec_test_datasets();
         for (mut data, name) in data_sets {
             create_and_validate(&data, name);
             data.reverse();

fastfield_codecs/src/blockwise_linear.rs

@@ -1,4 +1,4 @@
-//! MultiLinearInterpol compressor uses linear interpolation to guess a values and stores the
+//! The BlockwiseLinear codec 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 /
@@ -14,22 +14,25 @@ use std::io::{self, Read, Write};
 use std::ops::Sub;
 
 use common::{BinarySerializable, CountingWriter, DeserializeFrom};
+use ownedbytes::OwnedBytes;
 use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
 
-use crate::{FastFieldCodecReader, FastFieldCodecSerializer, FastFieldDataAccess, FastFieldStats};
+use crate::linear::{get_calculated_value, get_slope};
+use crate::{Column, FastFieldCodec, FastFieldCodecType};
 
 const CHUNK_SIZE: u64 = 512;
 
 /// Depending on the field type, a different
 /// fast field is required.
 #[derive(Clone)]
-pub struct MultiLinearInterpolFastFieldReader {
-    pub footer: MultiLinearInterpolFooter,
+pub struct BlockwiseLinearReader {
+    data: OwnedBytes,
+    pub footer: BlockwiseLinearFooter,
 }
 
 #[derive(Clone, Debug, Default)]
 struct Function {
-    // The offset in the data is required, because we have diffrent bit_widths per block
+    // The offset in the data is required, because we have different bit_widths per block
     data_start_offset: u64,
     // start_pos in the block will be CHUNK_SIZE * BLOCK_NUM
     start_pos: u64,
@@ -99,14 +102,14 @@ impl BinarySerializable for Function {
 }
 
 #[derive(Clone, Debug)]
-pub struct MultiLinearInterpolFooter {
+pub struct BlockwiseLinearFooter {
     pub num_vals: u64,
     pub min_value: u64,
     pub max_value: u64,
     interpolations: Vec<Function>,
 }
 
-impl BinarySerializable for MultiLinearInterpolFooter {
+impl BinarySerializable for BlockwiseLinearFooter {
     fn serialize<W: Write>(&self, write: &mut W) -> io::Result<()> {
         let mut out = vec![];
         self.num_vals.serialize(&mut out)?;
@@ -118,8 +121,8 @@ impl BinarySerializable for MultiLinearInterpolFooter {
         Ok(())
     }
 
-    fn deserialize<R: Read>(reader: &mut R) -> io::Result<MultiLinearInterpolFooter> {
-        let mut footer = MultiLinearInterpolFooter {
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<BlockwiseLinearFooter> {
+        let mut footer = BlockwiseLinearFooter {
             num_vals: u64::deserialize(reader)?,
             min_value: u64::deserialize(reader)?,
             max_value: u64::deserialize(reader)?,
@@ -143,26 +146,20 @@ fn get_interpolation_function(doc: u64, interpolations: &[Function]) -> &Functio
     &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 })
-    }
-
+impl Column for BlockwiseLinearReader {
     #[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..]);
+    fn get_val(&self, idx: u64) -> u64 {
+        let interpolation = get_interpolation_function(idx, &self.footer.interpolations);
+        let in_block_idx = idx - interpolation.start_pos;
+        let calculated_value = get_calculated_value(
+            interpolation.value_start_pos,
+            in_block_idx,
+            interpolation.slope,
+        );
+        let diff = interpolation.bit_unpacker.get(
+            in_block_idx,
+            &self.data[interpolation.data_start_offset as usize..],
+        );
         (calculated_value + diff) - interpolation.positive_val_offset
     }
 
@@ -174,39 +171,38 @@ impl FastFieldCodecReader for MultiLinearInterpolFastFieldReader {
     fn max_value(&self) -> u64 {
         self.footer.max_value
     }
+    #[inline]
+    fn num_vals(&self) -> u64 {
+        self.footer.num_vals
+    }
 }
 
-#[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
-}
+/// Same as LinearSerializer, but working on chunks of CHUNK_SIZE elements.
+pub struct BlockwiseLinearCodec;
 
-#[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
-}
+impl FastFieldCodec for BlockwiseLinearCodec {
+    const CODEC_TYPE: FastFieldCodecType = FastFieldCodecType::BlockwiseLinear;
 
-/// Same as LinearInterpolFastFieldSerializer, but working on chunks of CHUNK_SIZE elements.
-pub struct MultiLinearInterpolFastFieldSerializer {}
+    type Reader = BlockwiseLinearReader;
+
+    /// Opens a fast field given a file.
+    fn open_from_bytes(bytes: OwnedBytes) -> io::Result<Self::Reader> {
+        let footer_len: u32 = (&bytes[bytes.len() - 4..]).deserialize()?;
+        let footer_offset = bytes.len() - 4 - footer_len as usize;
+        let (data, mut footer) = bytes.split(footer_offset);
+        let footer = BlockwiseLinearFooter::deserialize(&mut footer)?;
+        Ok(BlockwiseLinearReader { data, footer })
+    }
 
-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: &dyn FastFieldDataAccess,
-        stats: FastFieldStats,
-        data_iter: impl Iterator<Item = u64>,
-        _data_iter1: impl Iterator<Item = u64>,
-    ) -> io::Result<()> {
-        assert!(stats.min_value <= stats.max_value);
+    fn serialize(write: &mut impl Write, fastfield_accessor: &dyn Column) -> io::Result<()> {
+        assert!(fastfield_accessor.min_value() <= fastfield_accessor.max_value());
 
         let first_val = fastfield_accessor.get_val(0);
-        let last_val = fastfield_accessor.get_val(stats.num_vals as u64 - 1);
+        let last_val = fastfield_accessor.get_val(fastfield_accessor.num_vals() as u64 - 1);
 
         let mut first_function = Function {
-            end_pos: stats.num_vals,
+            end_pos: fastfield_accessor.num_vals(),
             value_start_pos: first_val,
             value_end_pos: last_val,
             ..Default::default()
@@ -217,7 +213,7 @@ impl FastFieldCodecSerializer for MultiLinearInterpolFastFieldSerializer {
         // 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 data = fastfield_accessor.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) {
@@ -280,49 +276,46 @@ impl FastFieldCodecSerializer for MultiLinearInterpolFastFieldSerializer {
         }
         bit_packer.close(write)?;
 
-        let footer = MultiLinearInterpolFooter {
-            num_vals: stats.num_vals,
-            min_value: stats.min_value,
-            max_value: stats.max_value,
+        let footer = BlockwiseLinearFooter {
+            num_vals: fastfield_accessor.num_vals(),
+            min_value: fastfield_accessor.min_value(),
+            max_value: fastfield_accessor.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 {
+    fn estimate(fastfield_accessor: &impl Column) -> Option<f32> {
+        if fastfield_accessor.num_vals() < 10 * CHUNK_SIZE {
+            return None;
+        }
+
+        // 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 algorithm should be fine
+        let theorethical_maximum_offset =
+            fastfield_accessor.max_value() - fastfield_accessor.min_value();
+        if fastfield_accessor
+            .max_value()
+            .checked_add(theorethical_maximum_offset)
+            .is_none()
+        {
+            return None;
+        }
+
         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_elem_in_first_chunk = CHUNK_SIZE.min(fastfield_accessor.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,
+            fastfield_accessor.num_vals(),
         );
 
         // let's sample at 0%, 5%, 10% .. 95%, 100%, but for the first block only
@@ -349,11 +342,11 @@ impl FastFieldCodecSerializer for MultiLinearInterpolFastFieldSerializer {
         //
         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
+        let num_bits = compute_num_bits(relative_max_value as u64) as u64 * fastfield_accessor.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
+            + 29 * (fastfield_accessor.num_vals() / CHUNK_SIZE);
+        let num_bits_uncompressed = 64 * fastfield_accessor.num_vals();
+        Some(num_bits as f32 / num_bits_uncompressed as f32)
     }
 }
 
@@ -368,20 +361,35 @@ fn distance<T: Sub<Output = T> + Ord>(x: T, y: T) -> T {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use crate::tests::get_codec_test_data_sets;
+    use crate::tests::get_codec_test_datasets;
 
-    fn create_and_validate(data: &[u64], name: &str) -> (f32, f32) {
-        crate::tests::create_and_validate::<
-            MultiLinearInterpolFastFieldSerializer,
-            MultiLinearInterpolFastFieldReader,
-        >(data, name)
+    fn create_and_validate(data: &[u64], name: &str) -> Option<(f32, f32)> {
+        crate::tests::create_and_validate::<BlockwiseLinearCodec>(data, name)
+    }
+
+    const HIGHEST_BIT: u64 = 1 << 63;
+    pub fn i64_to_u64(val: i64) -> u64 {
+        (val as u64) ^ HIGHEST_BIT
+    }
+
+    #[test]
+    fn test_compression_i64() {
+        let data = (i64::MAX - 600_000..=i64::MAX - 550_000)
+            .map(i64_to_u64)
+            .collect::<Vec<_>>();
+        let (estimate, actual_compression) =
+            create_and_validate(&data, "simple monotonically large i64").unwrap();
+        assert!(actual_compression < 0.2);
+        assert!(estimate < 0.20);
+        assert!(estimate > 0.15);
+        assert!(actual_compression > 0.01);
     }
 
     #[test]
     fn test_compression() {
         let data = (10..=6_000_u64).collect::<Vec<_>>();
         let (estimate, actual_compression) =
-            create_and_validate(&data, "simple monotonically large");
+            create_and_validate(&data, "simple monotonically large").unwrap();
         assert!(actual_compression < 0.2);
         assert!(estimate < 0.20);
         assert!(estimate > 0.15);
@@ -390,7 +398,7 @@ mod tests {
 
     #[test]
     fn test_with_codec_data_sets() {
-        let data_sets = get_codec_test_data_sets();
+        let data_sets = get_codec_test_datasets();
         for (mut data, name) in data_sets {
             create_and_validate(&data, name);
             data.reverse();

fastfield_codecs/src/column.rs

@@ -0,0 +1,49 @@
+pub trait Column<T = u64> {
+    /// Return the value associated to the given idx.
+    ///
+    /// This accessor should return as fast as possible.
+    ///
+    /// # Panics
+    ///
+    /// May panic if `idx` is greater than the column length.
+    fn get_val(&self, idx: u64) -> T;
+
+    /// Fills an output buffer with the fast field values
+    /// associated with the `DocId` going from
+    /// `start` to `start + output.len()`.
+    ///
+    /// Regardless of the type of `Item`, this method works
+    /// - transmuting the output array
+    /// - extracting the `Item`s as if they were `u64`
+    /// - possibly converting the `u64` value to the right type.
+    ///
+    /// # Panics
+    ///
+    /// May panic if `start + output.len()` is greater than
+    /// the segment's `maxdoc`.
+    fn get_range(&self, start: u64, output: &mut [T]) {
+        for (out, idx) in output.iter_mut().zip(start..) {
+            *out = self.get_val(idx);
+        }
+    }
+
+    /// Returns the minimum value for this fast field.
+    ///
+    /// The min value does not take in account of possible
+    /// deleted document, and should be considered as a lower bound
+    /// of the actual minimum value.
+    fn min_value(&self) -> T;
+
+    /// Returns the maximum value for this fast field.
+    ///
+    /// The max value does not take in account of possible
+    /// deleted document, and should be considered as an upper bound
+    /// of the actual maximum value
+    fn max_value(&self) -> T;
+
+    fn num_vals(&self) -> u64;
+    /// Returns a iterator over the data
+    fn iter<'a>(&'a self) -> Box<dyn Iterator<Item = T> + 'a> {
+        Box::new((0..self.num_vals()).map(|idx| self.get_val(idx)))
+    }
+}

fastfield_codecs/src/lib.rs

@@ -5,61 +5,82 @@ extern crate more_asserts;
 use std::io;
 use std::io::Write;
 
+use common::BinarySerializable;
+use ownedbytes::OwnedBytes;
+
 pub mod bitpacked;
-pub mod linearinterpol;
-pub mod multilinearinterpol;
+pub mod blockwise_linear;
+pub mod linear;
 
-pub trait FastFieldCodecReader: Sized {
-    /// reads the metadata and returns the CodecReader
-    fn open_from_bytes(bytes: &[u8]) -> std::io::Result<Self>;
+mod column;
 
-    fn get_u64(&self, doc: u64, data: &[u8]) -> u64;
+pub use self::column::Column;
 
-    fn min_value(&self) -> u64;
-    fn max_value(&self) -> u64;
+
+#[derive(PartialEq, Eq, PartialOrd, Ord, Debug, Clone, Copy)]
+#[repr(u8)]
+pub enum FastFieldCodecType {
+    Bitpacked = 1,
+    Linear = 2,
+    BlockwiseLinear = 3,
+    Gcd = 4,
+}
+
+impl BinarySerializable for FastFieldCodecType {
+    fn serialize<W: Write>(&self, wrt: &mut W) -> io::Result<()> {
+        self.to_code().serialize(wrt)
+    }
+
+    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
+        let code = u8::deserialize(reader)?;
+        let codec_type: Self = Self::from_code(code)
+            .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidData, "Unknown code `{code}.`"))?;
+        Ok(codec_type)
+    }
+}
+
+impl FastFieldCodecType {
+    pub fn to_code(self) -> u8 {
+        self as u8
+    }
+
+    pub fn from_code(code: u8) -> Option<Self> {
+        match code {
+            1 => Some(Self::Bitpacked),
+            2 => Some(Self::Linear),
+            3 => Some(Self::BlockwiseLinear),
+            4 => Some(Self::Gcd),
+            _ => None,
+        }
+    }
 }
 
 /// The FastFieldSerializerEstimate trait is required on all variants
 /// of fast field compressions, to decide which one to choose.
-pub trait FastFieldCodecSerializer {
+pub trait FastFieldCodec {
     /// 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;
+    const CODEC_TYPE: FastFieldCodecType;
 
-    /// Check if the Codec is able to compress the data
-    fn is_applicable(fastfield_accessor: &impl FastFieldDataAccess, stats: FastFieldStats) -> bool;
+    type Reader: Column<u64>;
+
+    /// Reads the metadata and returns the CodecReader
+    fn open_from_bytes(bytes: OwnedBytes) -> io::Result<Self::Reader>;
+
+    /// Serializes the data using the serializer into write.
+    ///
+    /// The fastfield_accessor iterator should be preferred over using fastfield_accessor for
+    /// performance reasons.
+    fn serialize(write: &mut impl Write, fastfield_accessor: &dyn Column<u64>) -> io::Result<()>;
 
     /// Returns an estimate of the compression ratio.
+    /// If the codec is not applicable, returns `None`.
+    ///
     /// 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: &dyn 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;
+    fn estimate(fastfield_accessor: &impl Column) -> Option<f32>;
 }
 
 #[derive(Debug, Clone)]
@@ -70,61 +91,102 @@ pub struct FastFieldStats {
     pub num_vals: u64,
 }
 
-impl<'a> FastFieldDataAccess for &'a [u64] {
+impl<'a> Column for &'a [u64] {
     fn get_val(&self, position: u64) -> u64 {
         self[position as usize]
     }
+
+    fn iter<'b>(&'b self) -> Box<dyn Iterator<Item = u64> + 'b> {
+        Box::new((self as &[u64]).iter().cloned())
+    }
+
+    fn min_value(&self) -> u64 {
+        self.iter().min().unwrap_or(0)
+    }
+
+    fn max_value(&self) -> u64 {
+        self.iter().max().unwrap_or(0)
+    }
+
+    fn num_vals(&self) -> u64 {
+        self.len() as u64
+    }
 }
 
-impl FastFieldDataAccess for Vec<u64> {
+impl Column for Vec<u64> {
     fn get_val(&self, position: u64) -> u64 {
         self[position as usize]
     }
+    fn iter<'b>(&'b self) -> Box<dyn Iterator<Item = u64> + 'b> {
+        Box::new((self as &[u64]).iter().cloned())
+    }
+    fn min_value(&self) -> u64 {
+        self.iter().min().unwrap_or(0)
+    }
+
+    fn max_value(&self) -> u64 {
+        self.iter().max().unwrap_or(0)
+    }
+
+    fn num_vals(&self) -> u64 {
+        self.len() as u64
+    }
 }
 
 #[cfg(test)]
 mod tests {
-    use crate::bitpacked::{BitpackedFastFieldReader, BitpackedFastFieldSerializer};
-    use crate::linearinterpol::{LinearInterpolFastFieldReader, LinearInterpolFastFieldSerializer};
-    use crate::multilinearinterpol::{
-        MultiLinearInterpolFastFieldReader, MultiLinearInterpolFastFieldSerializer,
-    };
+    use proptest::arbitrary::any;
+    use proptest::proptest;
 
-    pub fn create_and_validate<S: FastFieldCodecSerializer, R: FastFieldCodecReader>(
+    use crate::bitpacked::BitpackedCodec;
+    use crate::blockwise_linear::BlockwiseLinearCodec;
+    use crate::linear::LinearCodec;
+
+    pub fn create_and_validate<Codec: FastFieldCodec>(
         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();
+    ) -> Option<(f32, f32)> {
+        let estimation = Codec::estimate(&data)?;
+
+        let mut out: Vec<u8> = Vec::new();
+        Codec::serialize(&mut out, &data).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)
+
+        let reader = Codec::open_from_bytes(OwnedBytes::new(out)).unwrap();
+        assert_eq!(reader.num_vals(), data.len() as u64);
+        for (doc, orig_val) in data.iter().copied().enumerate() {
+            let val = reader.get_val(doc as u64);
+            assert_eq!(
+                val, orig_val,
+                "val `{val}` does not match orig_val {orig_val:?}, in data set {name}, data \
+                 `{data:?}`",
+            );
+        }
+        Some((estimation, actual_compression))
     }
-    pub fn get_codec_test_data_sets() -> Vec<(Vec<u64>, &'static str)> {
+
+    proptest! {
+        #[test]
+        fn test_proptest_small(data in proptest::collection::vec(any::<u64>(), 1..10)) {
+            create_and_validate::<LinearCodec>(&data, "proptest linearinterpol");
+            create_and_validate::<BlockwiseLinearCodec>(&data, "proptest multilinearinterpol");
+            create_and_validate::<BitpackedCodec>(&data, "proptest bitpacked");
+        }
+
+        #[test]
+        fn test_proptest_large(data in proptest::collection::vec(any::<u64>(), 1..6000)) {
+            create_and_validate::<LinearCodec>(&data, "proptest linearinterpol");
+            create_and_validate::<BlockwiseLinearCodec>(&data, "proptest multilinearinterpol");
+            create_and_validate::<BitpackedCodec>(&data, "proptest bitpacked");
+        }
+
+    }
+
+    pub fn get_codec_test_datasets() -> Vec<(Vec<u64>, &'static str)> {
         let mut data_and_names = vec![];
 
-        let data = (10..=20_u64).collect::<Vec<_>>();
+        let data = (10..=10_000_u64).collect::<Vec<_>>();
         data_and_names.push((data, "simple monotonically increasing"));
 
         data_and_names.push((
@@ -137,89 +199,82 @@ mod tests {
         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()
+    fn test_codec<C: FastFieldCodec>() {
+        let codec_name = format!("{:?}", C::CODEC_TYPE);
+        for (data, dataset_name) in get_codec_test_datasets() {
+            let estimate_actual_opt: Option<(f32, f32)> =
+                crate::tests::create_and_validate::<C>(&data, dataset_name);
+            let result = if let Some((estimate, actual)) = estimate_actual_opt {
+                format!("Estimate `{estimate}` Actual `{actual}`")
             } else {
-                format!("Estimate {:?} Actual {:?} ", estimate, actual)
+                "Disabled".to_string()
             };
-            println!(
-                "Codec {}, DataSet {}, {}",
-                codec_name, data_set_name, result
-            );
+            println!("Codec {codec_name}, DataSet {dataset_name}, {result}");
         }
     }
     #[test]
     fn test_codec_bitpacking() {
-        test_codec::<BitpackedFastFieldSerializer, BitpackedFastFieldReader>();
+        test_codec::<BitpackedCodec>();
     }
     #[test]
     fn test_codec_interpolation() {
-        test_codec::<LinearInterpolFastFieldSerializer, LinearInterpolFastFieldReader>();
+        test_codec::<LinearCodec>();
     }
     #[test]
     fn test_codec_multi_interpolation() {
-        test_codec::<MultiLinearInterpolFastFieldSerializer, MultiLinearInterpolFastFieldReader>();
+        test_codec::<BlockwiseLinearCodec>();
     }
 
     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));
+        let linear_interpol_estimation = LinearCodec::estimate(&data).unwrap();
         assert_le!(linear_interpol_estimation, 0.01);
 
-        let multi_linear_interpol_estimation =
-            MultiLinearInterpolFastFieldSerializer::estimate(&data, stats_from_vec(&data));
+        let multi_linear_interpol_estimation = BlockwiseLinearCodec::estimate(&data).unwrap();
         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));
+        let bitpacked_estimation = BitpackedCodec::estimate(&data).unwrap();
         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));
+        let linear_interpol_estimation = LinearCodec::estimate(&data).unwrap();
         assert_le!(linear_interpol_estimation, 0.32);
 
-        let bitpacked_estimation =
-            BitpackedFastFieldSerializer::estimate(&data, stats_from_vec(&data));
+        let bitpacked_estimation = BitpackedCodec::estimate(&data).unwrap();
         assert_le!(bitpacked_estimation, linear_interpol_estimation);
     }
     #[test]
     fn estimation_test_bad_interpolation_case_monotonically_increasing() {
-        let mut data = (200..=20000_u64).collect::<Vec<_>>();
+        let mut data: Vec<u64> = (200..=20000_u64).collect();
         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));
+        let linear_interpol_estimation = LinearCodec::estimate(&data).unwrap();
         assert_le!(linear_interpol_estimation, 0.35);
 
-        let bitpacked_estimation =
-            BitpackedFastFieldSerializer::estimate(&data, stats_from_vec(&data));
+        let bitpacked_estimation = BitpackedCodec::estimate(&data).unwrap();
         assert_le!(bitpacked_estimation, 0.32);
         assert_le!(bitpacked_estimation, linear_interpol_estimation);
     }
+
+    #[test]
+    fn test_fast_field_codec_type_to_code() {
+        let mut count_codec = 0;
+        for code in 0..=255 {
+            if let Some(codec_type) = FastFieldCodecType::from_code(code) {
+                assert_eq!(codec_type.to_code(), code);
+                count_codec += 1;
+            }
+        }
+        assert_eq!(count_codec, 4);
+    }
 }

fastfield_codecs/src/linear.rs

@@ -2,21 +2,23 @@ use std::io::{self, Read, Write};
 use std::ops::Sub;
 
 use common::{BinarySerializable, FixedSize};
+use ownedbytes::OwnedBytes;
 use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
 
-use crate::{FastFieldCodecReader, FastFieldCodecSerializer, FastFieldDataAccess, FastFieldStats};
+use crate::{Column, FastFieldCodec, FastFieldCodecType};
 
 /// Depending on the field type, a different
 /// fast field is required.
 #[derive(Clone)]
-pub struct LinearInterpolFastFieldReader {
+pub struct LinearReader {
+    data: OwnedBytes,
     bit_unpacker: BitUnpacker,
-    pub footer: LinearInterpolFooter,
+    pub footer: LinearFooter,
     pub slope: f32,
 }
 
 #[derive(Clone, Debug)]
-pub struct LinearInterpolFooter {
+pub struct LinearFooter {
     pub relative_max_value: u64,
     pub offset: u64,
     pub first_val: u64,
@@ -26,7 +28,7 @@ pub struct LinearInterpolFooter {
     pub max_value: u64,
 }
 
-impl BinarySerializable for LinearInterpolFooter {
+impl BinarySerializable for LinearFooter {
     fn serialize<W: Write>(&self, write: &mut W) -> io::Result<()> {
         self.relative_max_value.serialize(write)?;
         self.offset.serialize(write)?;
@@ -38,8 +40,8 @@ impl BinarySerializable for LinearInterpolFooter {
         Ok(())
     }
 
-    fn deserialize<R: Read>(reader: &mut R) -> io::Result<LinearInterpolFooter> {
-        Ok(LinearInterpolFooter {
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<LinearFooter> {
+        Ok(LinearFooter {
             relative_max_value: u64::deserialize(reader)?,
             offset: u64::deserialize(reader)?,
             first_val: u64::deserialize(reader)?,
@@ -51,29 +53,15 @@ impl BinarySerializable for LinearInterpolFooter {
     }
 }
 
-impl FixedSize for LinearInterpolFooter {
+impl FixedSize for LinearFooter {
     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,
-        })
-    }
+impl Column for LinearReader {
     #[inline]
-    fn get_u64(&self, doc: u64, data: &[u8]) -> u64 {
+    fn get_val(&self, doc: u64) -> 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
+        (calculated_value + self.bit_unpacker.get(doc, &self.data)) - self.footer.offset
     }
 
     #[inline]
@@ -84,47 +72,87 @@ impl FastFieldCodecReader for LinearInterpolFastFieldReader {
     fn max_value(&self) -> u64 {
         self.footer.max_value
     }
+    #[inline]
+    fn num_vals(&self) -> u64 {
+        self.footer.num_vals
+    }
 }
 
 /// Fastfield serializer, which tries to guess values by linear interpolation
 /// and stores the difference bitpacked.
-pub struct LinearInterpolFastFieldSerializer {}
+pub struct LinearCodec;
 
 #[inline]
-fn get_slope(first_val: u64, last_val: u64, num_vals: u64) -> f32 {
+pub(crate) 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
+    let diff = diff(last_val, first_val);
+    (diff / (num_vals - 1) as f64) as f32
+}
+
+/// Delay the cast, to improve precision for very large u64 values.
+///
+/// Since i64 is mapped monotonically to u64 space, 0i64 is after the mapping i64::MAX.
+/// So very large values are not uncommon.
+///
+/// ```rust
+///     let val1 = i64::MAX;
+///     let val2 = i64::MAX - 100;
+///     assert_eq!(val1 - val2, 100);
+///     assert_eq!(val1 as f64 - val2 as f64, 0.0);
+/// ```
+fn diff(val1: u64, val2: u64) -> f64 {
+    if val1 >= val2 {
+        (val1 - val2) as f64
+    } else {
+        (val2 - val1) as f64 * -1.0
+    }
 }
 
 #[inline]
-fn get_calculated_value(first_val: u64, pos: u64, slope: f32) -> u64 {
-    first_val + (pos as f32 * slope) as u64
+pub fn get_calculated_value(first_val: u64, pos: u64, slope: f32) -> u64 {
+    if slope < 0.0 {
+        first_val - (pos as f32 * -slope) as u64
+    } else {
+        first_val + (pos as f32 * slope) as u64
+    }
 }
 
-impl FastFieldCodecSerializer for LinearInterpolFastFieldSerializer {
-    const NAME: &'static str = "LinearInterpol";
-    const ID: u8 = 2;
+impl FastFieldCodec for LinearCodec {
+    const CODEC_TYPE: FastFieldCodecType = FastFieldCodecType::Linear;
+
+    type Reader = LinearReader;
+
+    /// Opens a fast field given a file.
+    fn open_from_bytes(bytes: OwnedBytes) -> io::Result<Self::Reader> {
+        let footer_offset = bytes.len() - LinearFooter::SIZE_IN_BYTES;
+        let (data, mut footer) = bytes.split(footer_offset);
+        let footer = LinearFooter::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(LinearReader {
+            data,
+            bit_unpacker,
+            footer,
+            slope,
+        })
+    }
+
     /// Creates a new fast field serializer.
-    fn serialize(
-        write: &mut impl Write,
-        fastfield_accessor: &dyn FastFieldDataAccess,
-        stats: FastFieldStats,
-        data_iter: impl Iterator<Item = u64>,
-        data_iter1: impl Iterator<Item = u64>,
-    ) -> io::Result<()> {
-        assert!(stats.min_value <= stats.max_value);
+    fn serialize(write: &mut impl Write, fastfield_accessor: &dyn Column) -> io::Result<()> {
+        assert!(fastfield_accessor.min_value() <= fastfield_accessor.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);
+        let last_val = fastfield_accessor.get_val(fastfield_accessor.num_vals() as u64 - 1);
+        let slope = get_slope(first_val, last_val, fastfield_accessor.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() {
+        for (pos, actual_value) in fastfield_accessor.iter().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
@@ -142,56 +170,54 @@ impl FastFieldCodecSerializer for LinearInterpolFastFieldSerializer {
 
         let num_bits = compute_num_bits(relative_max_value);
         let mut bit_packer = BitPacker::new();
-        for (pos, val) in data_iter.enumerate() {
+        for (pos, val) in fastfield_accessor.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 {
+        let footer = LinearFooter {
             relative_max_value,
             offset,
             first_val,
             last_val,
-            num_vals: stats.num_vals,
-            min_value: stats.min_value,
-            max_value: stats.max_value,
+            num_vals: fastfield_accessor.num_vals(),
+            min_value: fastfield_accessor.min_value(),
+            max_value: fastfield_accessor.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 {
+    fn estimate(fastfield_accessor: &impl Column) -> Option<f32> {
+        if fastfield_accessor.num_vals() < 3 {
+            return None; // 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 algorithm should be fine
+        let theorethical_maximum_offset =
+            fastfield_accessor.max_value() - fastfield_accessor.min_value();
+        if fastfield_accessor
+            .max_value()
+            .checked_add(theorethical_maximum_offset)
+            .is_none()
+        {
+            return None;
+        }
+
         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 last_val = fastfield_accessor.get_val(fastfield_accessor.num_vals() as u64 - 1);
+        let slope = get_slope(first_val, last_val, fastfield_accessor.num_vals());
 
         // let's sample at 0%, 5%, 10% .. 95%, 100%
-        let num_vals = stats.num_vals as f32 / 100.0;
+        let num_vals = fastfield_accessor.num_vals() as f32 / 100.0;
         let sample_positions = (0..20)
             .map(|pos| (num_vals * pos as f32 * 5.0) as usize)
             .collect::<Vec<_>>();
@@ -213,10 +239,11 @@ impl FastFieldCodecSerializer for LinearInterpolFastFieldSerializer {
         //
         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
+        let num_bits = compute_num_bits(relative_max_value as u64) as u64
+            * fastfield_accessor.num_vals()
+            + LinearFooter::SIZE_IN_BYTES as u64;
+        let num_bits_uncompressed = 64 * fastfield_accessor.num_vals();
+        Some(num_bits as f32 / num_bits_uncompressed as f32)
     }
 }
 
@@ -232,28 +259,45 @@ fn distance<T: Sub<Output = T> + Ord>(x: T, y: T) -> T {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use crate::tests::get_codec_test_data_sets;
+    use crate::tests::get_codec_test_datasets;
 
-    fn create_and_validate(data: &[u64], name: &str) -> (f32, f32) {
-        crate::tests::create_and_validate::<
-            LinearInterpolFastFieldSerializer,
-            LinearInterpolFastFieldReader,
-        >(data, name)
+    fn create_and_validate(data: &[u64], name: &str) -> Option<(f32, f32)> {
+        crate::tests::create_and_validate::<LinearCodec>(data, name)
+    }
+
+    #[test]
+    fn get_calculated_value_test() {
+        // pos slope
+        assert_eq!(get_calculated_value(100, 10, 5.0), 150);
+
+        // neg slope
+        assert_eq!(get_calculated_value(100, 10, -5.0), 50);
+
+        // pos slope, very high values
+        assert_eq!(
+            get_calculated_value(i64::MAX as u64, 10, 5.0),
+            i64::MAX as u64 + 50
+        );
+        // neg slope, very high values
+        assert_eq!(
+            get_calculated_value(i64::MAX as u64, 10, -5.0),
+            i64::MAX as u64 - 50
+        );
     }
 
     #[test]
     fn test_compression() {
         let data = (10..=6_000_u64).collect::<Vec<_>>();
         let (estimate, actual_compression) =
-            create_and_validate(&data, "simple monotonically large");
+            create_and_validate(&data, "simple monotonically large").unwrap();
 
         assert!(actual_compression < 0.01);
         assert!(estimate < 0.01);
     }
 
     #[test]
-    fn test_with_codec_data_sets() {
-        let data_sets = get_codec_test_data_sets();
+    fn test_with_codec_datasets() {
+        let data_sets = get_codec_test_datasets();
         for (mut data, name) in data_sets {
             create_and_validate(&data, name);
             data.reverse();
@@ -290,9 +334,10 @@ mod tests {
     #[test]
     fn linear_interpol_fast_field_rand() {
         for _ in 0..5000 {
-            let mut data = (0..50).map(|_| rand::random::<u64>()).collect::<Vec<_>>();
+            let mut data = (0..10_000)
+                .map(|_| rand::random::<u64>())
+                .collect::<Vec<_>>();
             create_and_validate(&data, "random");
-
             data.reverse();
             create_and_validate(&data, "random");
         }

fastfield_codecs/src/main.rs

@@ -1,8 +1,9 @@
 #[macro_use]
 extern crate prettytable;
-use fastfield_codecs::linearinterpol::LinearInterpolFastFieldSerializer;
-use fastfield_codecs::multilinearinterpol::MultiLinearInterpolFastFieldSerializer;
-use fastfield_codecs::{FastFieldCodecSerializer, FastFieldStats};
+use fastfield_codecs::bitpacked::BitpackedCodec;
+use fastfield_codecs::blockwise_linear::BlockwiseLinearCodec;
+use fastfield_codecs::linear::LinearCodec;
+use fastfield_codecs::{FastFieldCodec, FastFieldCodecType, FastFieldStats};
 use prettytable::{Cell, Row, Table};
 
 fn main() {
@@ -12,41 +13,32 @@ fn main() {
     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 results: Vec<(f32, f32, FastFieldCodecType)> = [
+            serialize_with_codec::<LinearCodec>(&data),
+            serialize_with_codec::<BlockwiseLinearCodec>(&data),
+            serialize_with_codec::<BlockwiseLinearCodec>(&data),
+            serialize_with_codec::<BitpackedCodec>(&data),
+        ]
+        .into_iter()
+        .flatten()
+        .collect();
         let best_compression_ratio_codec = results
             .iter()
-            .min_by(|res1, res2| res1.partial_cmp(res2).unwrap())
+            .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())
-            };
+        for (est, comp, codec_type) in results {
+            let est_cell = est.to_string();
+            let ratio_cell = comp.to_string();
             let style = if comp == best_compression_ratio_codec.1 {
                 "Fb"
             } else {
                 ""
             };
-
             table.add_row(Row::new(vec![
-                Cell::new(name).style_spec("bFg"),
+                Cell::new(&format!("{codec_type:?}")).style_spec("bFg"),
                 Cell::new(&ratio_cell).style_spec(style),
                 Cell::new(&est_cell).style_spec(""),
             ]));
@@ -91,26 +83,14 @@ pub fn get_codec_test_data_sets() -> Vec<(Vec<u64>, &'static str)> {
     data_and_names
 }
 
-pub fn serialize_with_codec<S: FastFieldCodecSerializer>(
+pub fn serialize_with_codec<C: FastFieldCodec>(
     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();
-
+) -> Option<(f32, f32, FastFieldCodecType)> {
+    let estimation = C::estimate(&data)?;
+    let mut out = Vec::new();
+    C::serialize(&mut out, &data).unwrap();
     let actual_compression = out.len() as f32 / (data.len() * 8) as f32;
-    (true, estimation, actual_compression, S::NAME)
+    Some((estimation, actual_compression, C::CODEC_TYPE))
 }
 
 pub fn stats_from_vec(data: &[u64]) -> FastFieldStats {

ownedbytes/src/lib.rs

@@ -21,7 +21,7 @@ impl OwnedBytes {
         OwnedBytes::new(&[][..])
     }
 
-    /// Creates an `OwnedBytes` intance given a `StableDeref` object.
+    /// Creates an `OwnedBytes` instance given a `StableDeref` object.
     pub fn new<T: StableDeref + Deref<Target = [u8]> + 'static + Send + Sync>(
         data_holder: T,
     ) -> OwnedBytes {

query-grammar/src/lib.rs

@@ -1,3 +1,5 @@
+#![allow(clippy::derive_partial_eq_without_eq)]
+
 mod occur;
 mod query_grammar;
 mod user_input_ast;

query-grammar/src/query_grammar.rs

@@ -67,7 +67,7 @@ fn word<'a>() -> impl Parser<&'a str, Output = String> {
 /// 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
+/// We delegate rejecting such invalid dates to the logical AST computation code
 /// which invokes time::OffsetDateTime::parse(..., &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> {

src/aggregation/agg_result.rs

@@ -57,8 +57,7 @@ impl AggregationResult {
         match self {
             AggregationResult::BucketResult(_bucket) => Err(TantivyError::InternalError(
                 "Tried to retrieve value from bucket aggregation. This is not supported and \
-                 should not happen during collection phase, but should be catched during \
-                 validation"
+                 should not happen during collection phase, but should be caught during validation"
                     .to_string(),
             )),
             AggregationResult::MetricResult(metric) => metric.get_value(agg_property),

src/aggregation/bucket/histogram/histogram.rs

@@ -1,6 +1,7 @@
 use std::cmp::Ordering;
 use std::fmt::Display;
 
+use fastfield_codecs::Column;
 use itertools::Itertools;
 use serde::{Deserialize, Serialize};
 
@@ -14,7 +15,7 @@ use crate::aggregation::intermediate_agg_result::{
     IntermediateAggregationResults, IntermediateBucketResult, IntermediateHistogramBucketEntry,
 };
 use crate::aggregation::segment_agg_result::SegmentAggregationResultsCollector;
-use crate::fastfield::{DynamicFastFieldReader, FastFieldReader};
+use crate::fastfield::DynamicFastFieldReader;
 use crate::schema::Type;
 use crate::{DocId, TantivyError};
 
@@ -70,7 +71,7 @@ pub struct HistogramAggregation {
     /// The interval to chunk your data range. Each bucket spans a value range of [0..interval).
     /// Must be a positive value.
     pub interval: f64,
-    /// Intervals implicitely defines an absolute grid of buckets `[interval * k, interval * (k +
+    /// Intervals implicitly defines an absolute grid of buckets `[interval * k, interval * (k +
     /// 1))`.
     ///
     /// Offset makes it possible to shift this grid into
@@ -331,10 +332,10 @@ impl SegmentHistogramCollector {
             .expect("unexpected fast field cardinatility");
         let mut iter = doc.chunks_exact(4);
         for docs in iter.by_ref() {
-            let val0 = self.f64_from_fastfield_u64(accessor.get(docs[0]));
-            let val1 = self.f64_from_fastfield_u64(accessor.get(docs[1]));
-            let val2 = self.f64_from_fastfield_u64(accessor.get(docs[2]));
-            let val3 = self.f64_from_fastfield_u64(accessor.get(docs[3]));
+            let val0 = self.f64_from_fastfield_u64(accessor.get_val(docs[0] as u64));
+            let val1 = self.f64_from_fastfield_u64(accessor.get_val(docs[1] as u64));
+            let val2 = self.f64_from_fastfield_u64(accessor.get_val(docs[2] as u64));
+            let val3 = self.f64_from_fastfield_u64(accessor.get_val(docs[3] as u64));
 
             let bucket_pos0 = get_bucket_num(val0);
             let bucket_pos1 = get_bucket_num(val1);
@@ -370,8 +371,8 @@ impl SegmentHistogramCollector {
                 &bucket_with_accessor.sub_aggregation,
             )?;
         }
-        for doc in iter.remainder() {
-            let val = f64_from_fastfield_u64(accessor.get(*doc), &self.field_type);
+        for &doc in iter.remainder() {
+            let val = f64_from_fastfield_u64(accessor.get_val(doc as u64), &self.field_type);
             if !bounds.contains(val) {
                 continue;
             }
@@ -382,7 +383,7 @@ impl SegmentHistogramCollector {
                 self.buckets[bucket_pos].key,
                 get_bucket_val(val, self.interval, self.offset) as f64
             );
-            self.increment_bucket(bucket_pos, *doc, &bucket_with_accessor.sub_aggregation)?;
+            self.increment_bucket(bucket_pos, doc, &bucket_with_accessor.sub_aggregation)?;
         }
         if force_flush {
             if let Some(sub_aggregations) = self.sub_aggregations.as_mut() {

src/aggregation/bucket/range.rs

@@ -1,6 +1,7 @@
 use std::fmt::Debug;
 use std::ops::Range;
 
+use fastfield_codecs::Column;
 use fnv::FnvHashMap;
 use serde::{Deserialize, Serialize};
 
@@ -12,7 +13,6 @@ use crate::aggregation::intermediate_agg_result::{
 };
 use crate::aggregation::segment_agg_result::{BucketCount, SegmentAggregationResultsCollector};
 use crate::aggregation::{f64_from_fastfield_u64, f64_to_fastfield_u64, Key, SerializedKey};
-use crate::fastfield::FastFieldReader;
 use crate::schema::Type;
 use crate::{DocId, TantivyError};
 
@@ -210,8 +210,8 @@ impl SegmentRangeCollector {
                 let key = range
                     .key
                     .clone()
-                    .map(|key| Key::Str(key))
-                    .unwrap_or(range_to_key(&range.range, &field_type));
+                    .map(Key::Str)
+                    .unwrap_or_else(|| range_to_key(&range.range, &field_type));
                 let to = if range.range.end == u64::MAX {
                     None
                 } else {
@@ -264,10 +264,10 @@ impl SegmentRangeCollector {
             .as_single()
             .expect("unexpected fast field cardinatility");
         for docs in iter.by_ref() {
-            let val1 = accessor.get(docs[0]);
-            let val2 = accessor.get(docs[1]);
-            let val3 = accessor.get(docs[2]);
-            let val4 = accessor.get(docs[3]);
+            let val1 = accessor.get_val(docs[0] as u64);
+            let val2 = accessor.get_val(docs[1] as u64);
+            let val3 = accessor.get_val(docs[2] as u64);
+            let val4 = accessor.get_val(docs[3] as u64);
             let bucket_pos1 = self.get_bucket_pos(val1);
             let bucket_pos2 = self.get_bucket_pos(val2);
             let bucket_pos3 = self.get_bucket_pos(val3);
@@ -278,10 +278,10 @@ impl SegmentRangeCollector {
             self.increment_bucket(bucket_pos3, docs[2], &bucket_with_accessor.sub_aggregation)?;
             self.increment_bucket(bucket_pos4, docs[3], &bucket_with_accessor.sub_aggregation)?;
         }
-        for doc in iter.remainder() {
-            let val = accessor.get(*doc);
+        for &doc in iter.remainder() {
+            let val = accessor.get_val(doc as u64);
             let bucket_pos = self.get_bucket_pos(val);
-            self.increment_bucket(bucket_pos, *doc, &bucket_with_accessor.sub_aggregation)?;
+            self.increment_bucket(bucket_pos, doc, &bucket_with_accessor.sub_aggregation)?;
         }
         if force_flush {
             for bucket in &mut self.buckets {

src/aggregation/bucket/term_agg.rs

@@ -110,8 +110,8 @@ pub struct TermsAggregation {
     /// Set the order. `String` is here a target, which is either "_count", "_key", or the name of
     /// a metric sub_aggregation.
     ///
-    /// Single value metrics like average can be adressed by its name.
-    /// Multi value metrics like stats are required to adress their field by name e.g.
+    /// Single value metrics like average can be addressed by its name.
+    /// Multi value metrics like stats are required to address their field by name e.g.
     /// "stats.avg"
     ///
     /// Examples in JSON format:

src/aggregation/collector.rs

@@ -39,7 +39,7 @@ impl AggregationCollector {
 ///
 /// # Purpose
 /// AggregationCollector returns `IntermediateAggregationResults` and not the final
-/// `AggregationResults`, so that results from differenct indices can be merged and then converted
+/// `AggregationResults`, so that results from different indices can be merged and then converted
 /// into the final `AggregationResults` via the `into_final_result()` method.
 pub struct DistributedAggregationCollector {
     agg: Aggregations,

src/aggregation/intermediate_agg_result.rs

@@ -43,7 +43,7 @@ impl IntermediateAggregationResults {
     /// Convert intermediate result and its aggregation request to the final result.
     ///
     /// Internal function, AggregationsInternal is used instead Aggregations, which is optimized
-    /// for internal processing, by splitting metric and buckets into seperate groups.
+    /// for internal processing, by splitting metric and buckets into separate groups.
     pub(crate) fn into_final_bucket_result_internal(
         self,
         req: &AggregationsInternal,

src/aggregation/metric/average.rs

@@ -1,9 +1,10 @@
 use std::fmt::Debug;
 
+use fastfield_codecs::Column;
 use serde::{Deserialize, Serialize};
 
 use crate::aggregation::f64_from_fastfield_u64;
-use crate::fastfield::{DynamicFastFieldReader, FastFieldReader};
+use crate::fastfield::DynamicFastFieldReader;
 use crate::schema::Type;
 use crate::DocId;
 
@@ -60,10 +61,10 @@ impl SegmentAverageCollector {
     pub(crate) fn collect_block(&mut self, doc: &[DocId], field: &DynamicFastFieldReader<u64>) {
         let mut iter = doc.chunks_exact(4);
         for docs in iter.by_ref() {
-            let val1 = field.get(docs[0]);
-            let val2 = field.get(docs[1]);
-            let val3 = field.get(docs[2]);
-            let val4 = field.get(docs[3]);
+            let val1 = field.get_val(docs[0] as u64);
+            let val2 = field.get_val(docs[1] as u64);
+            let val3 = field.get_val(docs[2] as u64);
+            let val4 = field.get_val(docs[3] as u64);
             let val1 = f64_from_fastfield_u64(val1, &self.field_type);
             let val2 = f64_from_fastfield_u64(val2, &self.field_type);
             let val3 = f64_from_fastfield_u64(val3, &self.field_type);
@@ -73,8 +74,8 @@ impl SegmentAverageCollector {
             self.data.collect(val3);
             self.data.collect(val4);
         }
-        for doc in iter.remainder() {
-            let val = field.get(*doc);
+        for &doc in iter.remainder() {
+            let val = field.get_val(doc as u64);
             let val = f64_from_fastfield_u64(val, &self.field_type);
             self.data.collect(val);
         }

src/aggregation/metric/stats.rs

@@ -1,7 +1,8 @@
+use fastfield_codecs::Column;
 use serde::{Deserialize, Serialize};
 
 use crate::aggregation::f64_from_fastfield_u64;
-use crate::fastfield::{DynamicFastFieldReader, FastFieldReader};
+use crate::fastfield::DynamicFastFieldReader;
 use crate::schema::Type;
 use crate::{DocId, TantivyError};
 
@@ -166,10 +167,10 @@ impl SegmentStatsCollector {
     pub(crate) fn collect_block(&mut self, doc: &[DocId], field: &DynamicFastFieldReader<u64>) {
         let mut iter = doc.chunks_exact(4);
         for docs in iter.by_ref() {
-            let val1 = field.get(docs[0]);
-            let val2 = field.get(docs[1]);
-            let val3 = field.get(docs[2]);
-            let val4 = field.get(docs[3]);
+            let val1 = field.get_val(docs[0] as u64);
+            let val2 = field.get_val(docs[1] as u64);
+            let val3 = field.get_val(docs[2] as u64);
+            let val4 = field.get_val(docs[3] as u64);
             let val1 = f64_from_fastfield_u64(val1, &self.field_type);
             let val2 = f64_from_fastfield_u64(val2, &self.field_type);
             let val3 = f64_from_fastfield_u64(val3, &self.field_type);
@@ -179,8 +180,8 @@ impl SegmentStatsCollector {
             self.stats.collect(val3);
             self.stats.collect(val4);
         }
-        for doc in iter.remainder() {
-            let val = field.get(*doc);
+        for &doc in iter.remainder() {
+            let val = field.get_val(doc as u64);
             let val = f64_from_fastfield_u64(val, &self.field_type);
             self.stats.collect(val);
         }

src/collector/filter_collector_wrapper.rs

@@ -11,8 +11,10 @@
 // Importing tantivy...
 use std::marker::PhantomData;
 
+use fastfield_codecs::Column;
+
 use crate::collector::{Collector, SegmentCollector};
-use crate::fastfield::{DynamicFastFieldReader, FastFieldReader, FastValue};
+use crate::fastfield::{DynamicFastFieldReader, FastValue};
 use crate::schema::Field;
 use crate::{Score, SegmentReader, TantivyError};
 
@@ -174,7 +176,7 @@ where
     type Fruit = TSegmentCollector::Fruit;
 
     fn collect(&mut self, doc: u32, score: Score) {
-        let value = self.fast_field_reader.get(doc);
+        let value = self.fast_field_reader.get_val(doc as u64);
         if (self.predicate)(value) {
             self.segment_collector.collect(doc, score)
         }

src/collector/histogram_collector.rs

@@ -1,7 +1,8 @@
 use fastdivide::DividerU64;
+use fastfield_codecs::Column;
 
 use crate::collector::{Collector, SegmentCollector};
-use crate::fastfield::{DynamicFastFieldReader, FastFieldReader, FastValue};
+use crate::fastfield::{DynamicFastFieldReader, FastValue};
 use crate::schema::{Field, Type};
 use crate::{DocId, Score};
 
@@ -91,7 +92,7 @@ impl SegmentCollector for SegmentHistogramCollector {
     type Fruit = Vec<u64>;
 
     fn collect(&mut self, doc: DocId, _score: Score) {
-        let value = self.ff_reader.get(doc);
+        let value = self.ff_reader.get_val(doc as u64);
         self.histogram_computer.add_value(value);
     }
 

src/collector/tests.rs

@@ -1,7 +1,9 @@
+use fastfield_codecs::Column;
+
 use super::*;
 use crate::collector::{Count, FilterCollector, TopDocs};
 use crate::core::SegmentReader;
-use crate::fastfield::{BytesFastFieldReader, DynamicFastFieldReader, FastFieldReader};
+use crate::fastfield::{BytesFastFieldReader, DynamicFastFieldReader};
 use crate::query::{AllQuery, QueryParser};
 use crate::schema::{Field, Schema, FAST, TEXT};
 use crate::time::format_description::well_known::Rfc3339;
@@ -197,7 +199,7 @@ impl SegmentCollector for FastFieldSegmentCollector {
     type Fruit = Vec<u64>;
 
     fn collect(&mut self, doc: DocId, _score: Score) {
-        let val = self.reader.get(doc);
+        let val = self.reader.get_val(doc as u64);
         self.vals.push(val);
     }
 

src/collector/top_score_collector.rs

@@ -2,6 +2,8 @@ use std::collections::BinaryHeap;
 use std::fmt;
 use std::marker::PhantomData;
 
+use fastfield_codecs::Column;
+
 use super::Collector;
 use crate::collector::custom_score_top_collector::CustomScoreTopCollector;
 use crate::collector::top_collector::{ComparableDoc, TopCollector, TopSegmentCollector};
@@ -9,7 +11,7 @@ use crate::collector::tweak_score_top_collector::TweakedScoreTopCollector;
 use crate::collector::{
     CustomScorer, CustomSegmentScorer, ScoreSegmentTweaker, ScoreTweaker, SegmentCollector,
 };
-use crate::fastfield::{DynamicFastFieldReader, FastFieldReader, FastValue};
+use crate::fastfield::{DynamicFastFieldReader, FastValue};
 use crate::query::Weight;
 use crate::schema::Field;
 use crate::{DocAddress, DocId, Score, SegmentOrdinal, SegmentReader, TantivyError};
@@ -134,7 +136,7 @@ struct ScorerByFastFieldReader {
 
 impl CustomSegmentScorer<u64> for ScorerByFastFieldReader {
     fn score(&mut self, doc: DocId) -> u64 {
-        self.ff_reader.get(doc)
+        self.ff_reader.get_val(doc as u64)
     }
 }
 
@@ -499,7 +501,7 @@ impl TopDocs {
     ///
     /// This method only makes it possible to compute the score from a given
     /// `DocId`, fastfield values for the doc and any information you could
-    /// have precomputed beforehands. It does not make it possible for instance
+    /// have precomputed beforehand. It does not make it possible for instance
     /// to compute something like TfIdf as it does not have access to the list of query
     /// terms present in the document, nor the term frequencies for the different terms.
     ///

src/core/index_meta.rs

@@ -311,7 +311,7 @@ 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.
+    /// List of `SegmentMeta` information associated to each finalized segment of the index.
     pub segments: Vec<SegmentMeta>,
     /// Index `Schema`
     pub schema: Schema,

src/core/inverted_index_reader.rs

@@ -230,4 +230,13 @@ impl InvertedIndexReader {
         }
         Ok(())
     }
+
+    /// Returns the number of documents containing the term asynchronously.
+    pub async fn doc_freq_async(&self, term: &Term) -> crate::AsyncIoResult<u32> {
+        Ok(self
+            .get_term_info_async(term)
+            .await?
+            .map(|term_info| term_info.doc_freq)
+            .unwrap_or(0u32))
+    }
 }

src/core/searcher.rs

@@ -134,6 +134,19 @@ impl Searcher {
         Ok(total_doc_freq)
     }
 
+    /// Return the overall number of documents containing
+    /// the given term in an asynchronous manner.
+    #[cfg(feature = "quickwit")]
+    pub async fn doc_freq_async(&self, term: &Term) -> crate::Result<u64> {
+        let mut total_doc_freq = 0;
+        for segment_reader in &self.inner.segment_readers {
+            let inverted_index = segment_reader.inverted_index(term.field())?;
+            let doc_freq = inverted_index.doc_freq_async(term).await?;
+            total_doc_freq += u64::from(doc_freq);
+        }
+        Ok(total_doc_freq)
+    }
+
     /// Return the list of segment readers
     pub fn segment_readers(&self) -> &[SegmentReader] {
         &self.inner.segment_readers
@@ -234,6 +247,14 @@ impl SearcherInner {
         generation: TrackedObject<SearcherGeneration>,
         doc_store_cache_size: usize,
     ) -> io::Result<SearcherInner> {
+        assert_eq!(
+            &segment_readers
+                .iter()
+                .map(|reader| (reader.segment_id(), reader.delete_opstamp()))
+                .collect::<BTreeMap<_, _>>(),
+            generation.segments(),
+            "Set of segments referenced by this Searcher and its SearcherGeneration must match"
+        );
         let store_readers: Vec<StoreReader> = segment_readers
             .iter()
             .map(|segment_reader| segment_reader.get_store_reader(doc_store_cache_size))

src/core/segment_id.rs

@@ -16,7 +16,7 @@ use uuid::Uuid;
 /// by a UUID which is used to prefix the filenames
 /// of all of the file associated with the segment.
 ///
-/// In unit test, for reproducability, the `SegmentId` are
+/// In unit test, for reproducibility, the `SegmentId` are
 /// simply generated in an autoincrement fashion.
 #[derive(Clone, Copy, PartialEq, Eq, Hash, Serialize, Deserialize)]
 pub struct SegmentId(Uuid);

src/directory/composite_file.rs

@@ -38,7 +38,7 @@ impl BinarySerializable for FileAddr {
 /// A `CompositeWrite` is used to write a `CompositeFile`.
 pub struct CompositeWrite<W = WritePtr> {
     write: CountingWriter<W>,
-    offsets: HashMap<FileAddr, u64>,
+    offsets: Vec<(FileAddr, u64)>,
 }
 
 impl<W: TerminatingWrite + Write> CompositeWrite<W> {
@@ -47,7 +47,7 @@ impl<W: TerminatingWrite + Write> CompositeWrite<W> {
     pub fn wrap(w: W) -> CompositeWrite<W> {
         CompositeWrite {
             write: CountingWriter::wrap(w),
-            offsets: HashMap::new(),
+            offsets: Vec::new(),
         }
     }
 
@@ -60,8 +60,8 @@ impl<W: TerminatingWrite + Write> CompositeWrite<W> {
     pub fn for_field_with_idx(&mut self, field: Field, idx: usize) -> &mut CountingWriter<W> {
         let offset = self.write.written_bytes();
         let file_addr = FileAddr::new(field, idx);
-        assert!(!self.offsets.contains_key(&file_addr));
-        self.offsets.insert(file_addr, offset);
+        assert!(!self.offsets.iter().any(|el| el.0 == file_addr));
+        self.offsets.push((file_addr, offset));
         &mut self.write
     }
 
@@ -73,16 +73,8 @@ impl<W: TerminatingWrite + Write> CompositeWrite<W> {
         let footer_offset = self.write.written_bytes();
         VInt(self.offsets.len() as u64).serialize(&mut self.write)?;
 
-        let mut offset_fields: Vec<_> = self
-            .offsets
-            .iter()
-            .map(|(file_addr, offset)| (*offset, *file_addr))
-            .collect();
-
-        offset_fields.sort();
-
         let mut prev_offset = 0;
-        for (offset, file_addr) in offset_fields {
+        for (file_addr, offset) in self.offsets {
             VInt((offset - prev_offset) as u64).serialize(&mut self.write)?;
             file_addr.serialize(&mut self.write)?;
             prev_offset = offset;
@@ -106,6 +98,14 @@ pub struct CompositeFile {
     offsets_index: HashMap<FileAddr, Range<usize>>,
 }
 
+impl std::fmt::Debug for CompositeFile {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("CompositeFile")
+            .field("offsets_index", &self.offsets_index)
+            .finish()
+    }
+}
+
 impl CompositeFile {
     /// Opens a composite file stored in a given
     /// `FileSlice`.
@@ -233,4 +233,56 @@ mod test {
         }
         Ok(())
     }
+
+    #[test]
+    fn test_composite_file_bug() -> crate::Result<()> {
+        let path = Path::new("test_path");
+        let directory = RamDirectory::create();
+        {
+            let w = directory.open_write(path).unwrap();
+            let mut composite_write = CompositeWrite::wrap(w);
+            let mut write = composite_write.for_field_with_idx(Field::from_field_id(1u32), 0);
+            VInt(32431123u64).serialize(&mut write)?;
+            write.flush()?;
+            let write = composite_write.for_field_with_idx(Field::from_field_id(1u32), 1);
+            write.flush()?;
+
+            let mut write = composite_write.for_field_with_idx(Field::from_field_id(0u32), 0);
+            VInt(1_000_000).serialize(&mut write)?;
+            write.flush()?;
+
+            composite_write.close()?;
+        }
+        {
+            let r = directory.open_read(path)?;
+            let composite_file = CompositeFile::open(&r)?;
+            {
+                let file = composite_file
+                    .open_read_with_idx(Field::from_field_id(1u32), 0)
+                    .unwrap()
+                    .read_bytes()?;
+                let mut file0_buf = file.as_slice();
+                let payload_0 = VInt::deserialize(&mut file0_buf)?.0;
+                assert_eq!(file0_buf.len(), 0);
+                assert_eq!(payload_0, 32431123u64);
+            }
+            {
+                let file = composite_file
+                    .open_read_with_idx(Field::from_field_id(1u32), 1)
+                    .unwrap()
+                    .read_bytes()?;
+                let file = file.as_slice();
+                assert_eq!(file.len(), 0);
+            }
+            {
+                let file = composite_file
+                    .open_read_with_idx(Field::from_field_id(0u32), 0)
+                    .unwrap()
+                    .read_bytes()?;
+                let file = file.as_slice();
+                assert_eq!(file.len(), 3);
+            }
+        }
+        Ok(())
+    }
 }

src/directory/directory_lock.rs

@@ -45,7 +45,7 @@ pub static INDEX_WRITER_LOCK: Lazy<Lock> = Lazy::new(|| Lock {
 /// 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
+/// our index in-writing. Ideally, we may have preferred `RWLock` semantics
 /// here, but it is difficult to achieve on Windows.
 ///
 /// Opening segment readers is a very fast process.

src/directory/file_slice.rs

@@ -112,7 +112,7 @@ impl FileSlice {
 
     /// Returns a `OwnedBytes` with all of the data in the `FileSlice`.
     ///
-    /// The behavior is strongly dependant on the implementation of the underlying
+    /// The behavior is strongly dependent 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.

src/directory/managed_directory.rs

@@ -114,7 +114,7 @@ impl ManagedDirectory {
         let mut files_to_delete = vec![];
 
         // It is crucial to get the living files after acquiring the
-        // read lock of meta informations. That way, we
+        // read lock of meta information. That way, we
         // avoid the following scenario.
         //
         // 1) we get the list of living files.

src/directory/ram_directory.rs

@@ -40,7 +40,7 @@ impl Drop for VecWriter {
     fn drop(&mut self) {
         if !self.is_flushed {
             warn!(
-                "You forgot to flush {:?} before its writter got Drop. Do not rely on drop. This \
+                "You forgot to flush {:?} before its writer got Drop. Do not rely on drop. This \
                  also occurs when the indexer crashed, so you may want to check the logs for the \
                  root cause.",
                 self.path

src/directory/tests.rs

@@ -247,7 +247,7 @@ fn test_lock_blocking(directory: &dyn Directory) {
         //< lock_a_res is sent to the thread.
         in_thread_clone.store(true, SeqCst);
         let _just_sync = receiver.recv();
-        // explicitely dropping lock_a_res. It would have been sufficient to just force it
+        // explicitly dropping lock_a_res. It would have been sufficient to just force it
         // to be part of the move, but the intent seems clearer that way.
         drop(lock_a_res);
     });

src/fastfield/bytes/reader.rs

@@ -1,5 +1,7 @@
+use fastfield_codecs::Column;
+
 use crate::directory::{FileSlice, OwnedBytes};
-use crate::fastfield::{DynamicFastFieldReader, FastFieldReader, MultiValueLength};
+use crate::fastfield::{DynamicFastFieldReader, MultiValueLength};
 use crate::DocId;
 
 /// Reader for byte array fast fields
@@ -28,8 +30,9 @@ impl BytesFastFieldReader {
     }
 
     fn range(&self, doc: DocId) -> (usize, usize) {
-        let start = self.idx_reader.get(doc) as usize;
-        let stop = self.idx_reader.get(doc + 1) as usize;
+        let idx = doc as u64;
+        let start = self.idx_reader.get_val(idx) as usize;
+        let stop = self.idx_reader.get_val(idx + 1) as usize;
         (start, stop)
     }
 

src/fastfield/gcd.rs

@@ -1,80 +1,130 @@
 use std::io::{self, Write};
+use std::num::NonZeroU64;
 
 use common::BinarySerializable;
 use fastdivide::DividerU64;
-use fastfield_codecs::FastFieldCodecReader;
-use gcd::Gcd;
+use fastfield_codecs::{Column, FastFieldCodec};
+use ownedbytes::OwnedBytes;
 
 pub const GCD_DEFAULT: u64 = 1;
-pub const GCD_CODEC_ID: u8 = 4;
 
 /// Wrapper for accessing a fastfield.
 ///
 /// Holds the data and the codec to the read the data.
 #[derive(Clone)]
-pub struct GCDFastFieldCodec<CodecReader> {
-    gcd: u64,
-    min_value: u64,
+pub struct GCDReader<CodecReader: Column> {
+    gcd_params: GCDParams,
     reader: CodecReader,
 }
-impl<C: FastFieldCodecReader + Clone> FastFieldCodecReader for GCDFastFieldCodec<C> {
-    /// Opens a fast field given the bytes.
-    fn open_from_bytes(bytes: &[u8]) -> std::io::Result<Self> {
-        let (header, mut footer) = bytes.split_at(bytes.len() - 16);
-        let gcd = u64::deserialize(&mut footer)?;
-        let min_value = u64::deserialize(&mut footer)?;
-        let reader = C::open_from_bytes(header)?;
 
-        Ok(GCDFastFieldCodec {
-            gcd,
-            min_value,
-            reader,
-        })
+#[derive(Debug, Clone, Copy)]
+struct GCDParams {
+    gcd: u64,
+    min_value: u64,
+    num_vals: u64,
+}
+
+impl GCDParams {
+    pub fn eval(&self, val: u64) -> u64 {
+        self.min_value + self.gcd * val
+    }
+}
+
+impl BinarySerializable for GCDParams {
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
+        self.gcd.serialize(writer)?;
+        self.min_value.serialize(writer)?;
+        self.num_vals.serialize(writer)?;
+        Ok(())
     }
 
+    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
+        let gcd: u64 = u64::deserialize(reader)?;
+        let min_value: u64 = u64::deserialize(reader)?;
+        let num_vals: u64 = u64::deserialize(reader)?;
+        Ok(Self {
+            gcd,
+            min_value,
+            num_vals,
+        })
+    }
+}
+
+pub fn open_gcd_from_bytes<WrappedCodec: FastFieldCodec>(
+    bytes: OwnedBytes,
+) -> io::Result<GCDReader<WrappedCodec::Reader>> {
+    let footer_offset = bytes.len() - 24;
+    let (body, mut footer) = bytes.split(footer_offset);
+    let gcd_params = GCDParams::deserialize(&mut footer)?;
+    let reader: WrappedCodec::Reader = WrappedCodec::open_from_bytes(body)?;
+    Ok(GCDReader { gcd_params, reader })
+}
+
+impl<C: Column + Clone> Column for GCDReader<C> {
     #[inline]
-    fn get_u64(&self, doc: u64, data: &[u8]) -> u64 {
-        let mut data = self.reader.get_u64(doc, data);
-        data *= self.gcd;
-        data += self.min_value;
-        data
+    fn get_val(&self, doc: u64) -> u64 {
+        let val = self.reader.get_val(doc);
+        self.gcd_params.eval(val)
     }
 
     fn min_value(&self) -> u64 {
-        self.min_value + self.reader.min_value() * self.gcd
+        self.gcd_params.eval(self.reader.min_value())
     }
 
     fn max_value(&self) -> u64 {
-        self.min_value + self.reader.max_value() * self.gcd
+        self.gcd_params.eval(self.reader.max_value())
+    }
+    fn num_vals(&self) -> u64 {
+        self.gcd_params.num_vals
     }
 }
 
-pub fn write_gcd_header<W: Write>(field_write: &mut W, min_value: u64, gcd: u64) -> io::Result<()> {
+pub fn write_gcd_header<W: Write>(
+    field_write: &mut W,
+    min_value: u64,
+    gcd: u64,
+    num_vals: u64,
+) -> io::Result<()> {
     gcd.serialize(field_write)?;
     min_value.serialize(field_write)?;
+    num_vals.serialize(field_write)?;
     Ok(())
 }
 
+/// Compute the gcd of two non null numbers.
+///
+/// It is recommended, but not required, to feed values such that `large >= small`.
+fn compute_gcd(mut large: NonZeroU64, mut small: NonZeroU64) -> NonZeroU64 {
+    loop {
+        let rem: u64 = large.get() % small;
+        if let Some(new_small) = NonZeroU64::new(rem) {
+            (large, small) = (small, new_small);
+        } else {
+            return small;
+        }
+    }
+}
+
 // Find GCD for iterator of numbers
-pub fn find_gcd(numbers: impl Iterator<Item = u64>) -> Option<u64> {
-    let mut numbers = numbers.filter(|n| *n != 0);
-    let mut gcd = numbers.next()?;
-    if gcd == 1 {
-        return Some(1);
+pub fn find_gcd(numbers: impl Iterator<Item = u64>) -> Option<NonZeroU64> {
+    let mut numbers = numbers.flat_map(NonZeroU64::new);
+    let mut gcd: NonZeroU64 = numbers.next()?;
+    if gcd.get() == 1 {
+        return Some(gcd);
     }
 
-    let mut gcd_divider = DividerU64::divide_by(gcd);
+    let mut gcd_divider = DividerU64::divide_by(gcd.get());
     for val in numbers {
-        let remainder = val - (gcd_divider.divide(val)) * gcd;
+        let remainder = val.get() - (gcd_divider.divide(val.get())) * gcd.get();
         if remainder == 0 {
             continue;
         }
-        gcd = gcd.gcd(val);
-        if gcd == 1 {
-            return Some(1);
+        gcd = compute_gcd(val, gcd);
+        if gcd.get() == 1 {
+            return Some(gcd);
         }
 
-        gcd_divider = DividerU64::divide_by(gcd);
+        gcd_divider = DividerU64::divide_by(gcd.get());
     }
     Some(gcd)
 }
@@ -82,19 +132,23 @@ pub fn find_gcd(numbers: impl Iterator<Item = u64>) -> Option<u64> {
 #[cfg(test)]
 mod tests {
     use std::collections::HashMap;
+    use std::num::NonZeroU64;
     use std::path::Path;
+    use std::time::{Duration, SystemTime};
 
     use common::HasLen;
+    use fastfield_codecs::Column;
 
     use crate::directory::{CompositeFile, RamDirectory, WritePtr};
+    use crate::fastfield::gcd::compute_gcd;
     use crate::fastfield::serializer::FastFieldCodecEnableCheck;
     use crate::fastfield::tests::{FIELD, FIELDI64, SCHEMA, SCHEMAI64};
     use crate::fastfield::{
-        find_gcd, CompositeFastFieldSerializer, DynamicFastFieldReader, FastFieldCodecName,
-        FastFieldReader, FastFieldsWriter, ALL_CODECS,
+        find_gcd, CompositeFastFieldSerializer, DynamicFastFieldReader, FastFieldCodecType,
+        FastFieldsWriter, ALL_CODECS,
     };
-    use crate::schema::Schema;
-    use crate::Directory;
+    use crate::schema::{Cardinality, Schema};
+    use crate::{DateOptions, DatePrecision, DateTime, Directory};
 
     fn get_index(
         docs: &[crate::Document],
@@ -120,33 +174,33 @@ mod tests {
     }
 
     fn test_fastfield_gcd_i64_with_codec(
-        codec_name: FastFieldCodecName,
+        code_type: FastFieldCodecType,
         num_vals: usize,
     ) -> crate::Result<()> {
         let path = Path::new("test");
         let mut docs = vec![];
         for i in 1..=num_vals {
-            let val = i as i64 * 1000i64;
+            let val = (i as i64 - 5) * 1000i64;
             docs.push(doc!(*FIELDI64=>val));
         }
-        let directory = get_index(&docs, &SCHEMAI64, codec_name.clone().into())?;
+        let directory = get_index(&docs, &SCHEMAI64, code_type.into())?;
         let file = directory.open_read(path).unwrap();
-        // assert_eq!(file.len(), 118);
         let composite_file = CompositeFile::open(&file)?;
         let file = composite_file.open_read(*FIELD).unwrap();
         let fast_field_reader = DynamicFastFieldReader::<i64>::open(file)?;
-        assert_eq!(fast_field_reader.get(0), 1000i64);
-        assert_eq!(fast_field_reader.get(1), 2000i64);
-        assert_eq!(fast_field_reader.get(2), 3000i64);
-        assert_eq!(fast_field_reader.max_value(), num_vals as i64 * 1000);
-        assert_eq!(fast_field_reader.min_value(), 1000i64);
+
+        assert_eq!(fast_field_reader.get_val(0), -4000i64);
+        assert_eq!(fast_field_reader.get_val(1), -3000i64);
+        assert_eq!(fast_field_reader.get_val(2), -2000i64);
+        assert_eq!(fast_field_reader.max_value(), (num_vals as i64 - 5) * 1000);
+        assert_eq!(fast_field_reader.min_value(), -4000i64);
         let file = directory.open_read(path).unwrap();
 
         // Can't apply gcd
         let path = Path::new("test");
         docs.pop();
         docs.push(doc!(*FIELDI64=>2001i64));
-        let directory = get_index(&docs, &SCHEMAI64, codec_name.into())?;
+        let directory = get_index(&docs, &SCHEMAI64, code_type.into())?;
         let file2 = directory.open_read(path).unwrap();
         assert!(file2.len() > file.len());
 
@@ -155,14 +209,14 @@ mod tests {
 
     #[test]
     fn test_fastfield_gcd_i64() -> crate::Result<()> {
-        for codec_name in ALL_CODECS {
-            test_fastfield_gcd_i64_with_codec(codec_name.clone(), 5005)?;
+        for &code_type in ALL_CODECS {
+            test_fastfield_gcd_i64_with_codec(code_type, 5005)?;
         }
         Ok(())
     }
 
     fn test_fastfield_gcd_u64_with_codec(
-        codec_name: FastFieldCodecName,
+        code_type: FastFieldCodecType,
         num_vals: usize,
     ) -> crate::Result<()> {
         let path = Path::new("test");
@@ -171,15 +225,14 @@ mod tests {
             let val = i as u64 * 1000u64;
             docs.push(doc!(*FIELD=>val));
         }
-        let directory = get_index(&docs, &SCHEMA, codec_name.clone().into())?;
+        let directory = get_index(&docs, &SCHEMA, code_type.into())?;
         let file = directory.open_read(path).unwrap();
-        // assert_eq!(file.len(), 118);
         let composite_file = CompositeFile::open(&file)?;
         let file = composite_file.open_read(*FIELD).unwrap();
         let fast_field_reader = DynamicFastFieldReader::<u64>::open(file)?;
-        assert_eq!(fast_field_reader.get(0), 1000u64);
-        assert_eq!(fast_field_reader.get(1), 2000u64);
-        assert_eq!(fast_field_reader.get(2), 3000u64);
+        assert_eq!(fast_field_reader.get_val(0), 1000u64);
+        assert_eq!(fast_field_reader.get_val(1), 2000u64);
+        assert_eq!(fast_field_reader.get_val(2), 3000u64);
         assert_eq!(fast_field_reader.max_value(), num_vals as u64 * 1000);
         assert_eq!(fast_field_reader.min_value(), 1000u64);
         let file = directory.open_read(path).unwrap();
@@ -188,7 +241,7 @@ mod tests {
         let path = Path::new("test");
         docs.pop();
         docs.push(doc!(*FIELDI64=>2001u64));
-        let directory = get_index(&docs, &SCHEMA, codec_name.into())?;
+        let directory = get_index(&docs, &SCHEMA, code_type.into())?;
         let file2 = directory.open_read(path).unwrap();
         assert!(file2.len() > file.len());
 
@@ -197,8 +250,8 @@ mod tests {
 
     #[test]
     fn test_fastfield_gcd_u64() -> crate::Result<()> {
-        for codec_name in ALL_CODECS {
-            test_fastfield_gcd_u64_with_codec(codec_name.clone(), 5005)?;
+        for &code_type in ALL_CODECS {
+            test_fastfield_gcd_u64_with_codec(code_type, 5005)?;
         }
         Ok(())
     }
@@ -206,19 +259,103 @@ mod tests {
     #[test]
     pub fn test_fastfield2() {
         let test_fastfield = DynamicFastFieldReader::<u64>::from(vec![100, 200, 300]);
-        assert_eq!(test_fastfield.get(0), 100);
-        assert_eq!(test_fastfield.get(1), 200);
-        assert_eq!(test_fastfield.get(2), 300);
+        assert_eq!(test_fastfield.get_val(0), 100);
+        assert_eq!(test_fastfield.get_val(1), 200);
+        assert_eq!(test_fastfield.get_val(2), 300);
+    }
+
+    #[test]
+    pub fn test_gcd_date() -> crate::Result<()> {
+        let size_prec_sec =
+            test_gcd_date_with_codec(FastFieldCodecType::Bitpacked, DatePrecision::Seconds)?;
+        let size_prec_micro =
+            test_gcd_date_with_codec(FastFieldCodecType::Bitpacked, DatePrecision::Microseconds)?;
+        assert!(size_prec_sec < size_prec_micro);
+
+        let size_prec_sec =
+            test_gcd_date_with_codec(FastFieldCodecType::Linear, DatePrecision::Seconds)?;
+        let size_prec_micro =
+            test_gcd_date_with_codec(FastFieldCodecType::Linear, DatePrecision::Microseconds)?;
+        assert!(size_prec_sec < size_prec_micro);
+
+        Ok(())
+    }
+
+    fn test_gcd_date_with_codec(
+        codec_type: FastFieldCodecType,
+        precision: DatePrecision,
+    ) -> crate::Result<usize> {
+        let time1 = DateTime::from_timestamp_micros(
+            SystemTime::now()
+                .duration_since(SystemTime::UNIX_EPOCH)
+                .unwrap()
+                .as_secs() as i64,
+        );
+        let time2 = DateTime::from_timestamp_micros(
+            SystemTime::now()
+                .checked_sub(Duration::from_micros(4111))
+                .unwrap()
+                .duration_since(SystemTime::UNIX_EPOCH)
+                .unwrap()
+                .as_secs() as i64,
+        );
+
+        let time3 = DateTime::from_timestamp_micros(
+            SystemTime::now()
+                .checked_sub(Duration::from_millis(2000))
+                .unwrap()
+                .duration_since(SystemTime::UNIX_EPOCH)
+                .unwrap()
+                .as_secs() as i64,
+        );
+
+        let mut schema_builder = Schema::builder();
+        let date_options = DateOptions::default()
+            .set_fast(Cardinality::SingleValue)
+            .set_precision(precision);
+        let field = schema_builder.add_date_field("field", date_options);
+        let schema = schema_builder.build();
+
+        let docs = vec![doc!(field=>time1), doc!(field=>time2), doc!(field=>time3)];
+
+        let directory = get_index(&docs, &schema, codec_type.into())?;
+        let path = Path::new("test");
+        let file = directory.open_read(path).unwrap();
+        let composite_file = CompositeFile::open(&file)?;
+        let file = composite_file.open_read(*FIELD).unwrap();
+        let len = file.len();
+        let test_fastfield = DynamicFastFieldReader::<DateTime>::open(file)?;
+
+        assert_eq!(test_fastfield.get_val(0), time1.truncate(precision));
+        assert_eq!(test_fastfield.get_val(1), time2.truncate(precision));
+        assert_eq!(test_fastfield.get_val(2), time3.truncate(precision));
+        Ok(len)
+    }
+
+    #[test]
+    fn test_compute_gcd() {
+        let test_compute_gcd_aux = |large, small, expected| {
+            let large = NonZeroU64::new(large).unwrap();
+            let small = NonZeroU64::new(small).unwrap();
+            let expected = NonZeroU64::new(expected).unwrap();
+            assert_eq!(compute_gcd(small, large), expected);
+            assert_eq!(compute_gcd(large, small), expected);
+        };
+        test_compute_gcd_aux(1, 4, 1);
+        test_compute_gcd_aux(2, 4, 2);
+        test_compute_gcd_aux(10, 25, 5);
+        test_compute_gcd_aux(25, 25, 25);
     }
 
     #[test]
     fn find_gcd_test() {
         assert_eq!(find_gcd([0].into_iter()), None);
-        assert_eq!(find_gcd([0, 10].into_iter()), Some(10));
-        assert_eq!(find_gcd([10, 0].into_iter()), Some(10));
+        assert_eq!(find_gcd([0, 10].into_iter()), NonZeroU64::new(10));
+        assert_eq!(find_gcd([10, 0].into_iter()), NonZeroU64::new(10));
         assert_eq!(find_gcd([].into_iter()), None);
-        assert_eq!(find_gcd([15, 30, 5, 10].into_iter()), Some(5));
-        assert_eq!(find_gcd([15, 16, 10].into_iter()), Some(1));
-        assert_eq!(find_gcd([0, 5, 5, 5].into_iter()), Some(5));
+        assert_eq!(find_gcd([15, 30, 5, 10].into_iter()), NonZeroU64::new(5));
+        assert_eq!(find_gcd([15, 16, 10].into_iter()), NonZeroU64::new(1));
+        assert_eq!(find_gcd([0, 5, 5, 5].into_iter()), NonZeroU64::new(5));
+        assert_eq!(find_gcd([0, 0].into_iter()), None);
     }
 }

src/fastfield/mod.rs

@@ -20,16 +20,18 @@
 //!
 //! Read access performance is comparable to that of an array lookup.
 
+use fastfield_codecs::FastFieldCodecType;
+
 pub use self::alive_bitset::{intersect_alive_bitsets, write_alive_bitset, AliveBitSet};
 pub use self::bytes::{BytesFastFieldReader, BytesFastFieldWriter};
 pub use self::error::{FastFieldNotAvailableError, Result};
 pub use self::facet_reader::FacetReader;
-pub(crate) use self::gcd::{find_gcd, GCDFastFieldCodec, GCD_CODEC_ID, GCD_DEFAULT};
+pub(crate) use self::gcd::{find_gcd, GCDReader, GCD_DEFAULT};
 pub use self::multivalued::{MultiValuedFastFieldReader, MultiValuedFastFieldWriter};
-pub use self::reader::{DynamicFastFieldReader, FastFieldReader};
+pub use self::reader::DynamicFastFieldReader;
 pub use self::readers::FastFieldReaders;
 pub(crate) use self::readers::{type_and_cardinality, FastType};
-pub use self::serializer::{CompositeFastFieldSerializer, FastFieldDataAccess, FastFieldStats};
+pub use self::serializer::{Column, CompositeFastFieldSerializer, FastFieldStats};
 pub use self::writer::{FastFieldsWriter, IntFastFieldWriter};
 use crate::schema::{Cardinality, FieldType, Type, Value};
 use crate::{DateTime, DocId};
@@ -45,16 +47,10 @@ mod readers;
 mod serializer;
 mod writer;
 
-#[derive(PartialEq, Eq, PartialOrd, Ord, Debug, Clone)]
-pub(crate) enum FastFieldCodecName {
-    Bitpacked,
-    LinearInterpol,
-    BlockwiseLinearInterpol,
-}
-pub(crate) const ALL_CODECS: &[FastFieldCodecName; 3] = &[
-    FastFieldCodecName::Bitpacked,
-    FastFieldCodecName::LinearInterpol,
-    FastFieldCodecName::BlockwiseLinearInterpol,
+pub(crate) const ALL_CODECS: &[FastFieldCodecType; 3] = &[
+    FastFieldCodecType::Bitpacked,
+    FastFieldCodecType::Linear,
+    FastFieldCodecType::BlockwiseLinear,
 ];
 
 /// Trait for `BytesFastFieldReader` and `MultiValuedFastFieldReader` to return the length of data
@@ -302,9 +298,9 @@ mod tests {
     #[test]
     pub fn test_fastfield() {
         let test_fastfield = DynamicFastFieldReader::<u64>::from(vec![100, 200, 300]);
-        assert_eq!(test_fastfield.get(0), 100);
-        assert_eq!(test_fastfield.get(1), 200);
-        assert_eq!(test_fastfield.get(2), 300);
+        assert_eq!(test_fastfield.get_val(0u64), 100);
+        assert_eq!(test_fastfield.get_val(1u64), 200);
+        assert_eq!(test_fastfield.get_val(2u64), 300);
     }
 
     #[test]
@@ -330,13 +326,13 @@ mod tests {
             serializer.close().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 37);
+        assert_eq!(file.len(), 45);
         let composite_file = CompositeFile::open(&file)?;
         let file = composite_file.open_read(*FIELD).unwrap();
         let fast_field_reader = DynamicFastFieldReader::<u64>::open(file)?;
-        assert_eq!(fast_field_reader.get(0), 13u64);
-        assert_eq!(fast_field_reader.get(1), 14u64);
-        assert_eq!(fast_field_reader.get(2), 2u64);
+        assert_eq!(fast_field_reader.get_val(0), 13u64);
+        assert_eq!(fast_field_reader.get_val(1), 14u64);
+        assert_eq!(fast_field_reader.get_val(2), 2u64);
         Ok(())
     }
 
@@ -361,20 +357,20 @@ mod tests {
             serializer.close()?;
         }
         let file = directory.open_read(path)?;
-        assert_eq!(file.len(), 62);
+        assert_eq!(file.len(), 70);
         {
             let fast_fields_composite = CompositeFile::open(&file)?;
             let data = fast_fields_composite.open_read(*FIELD).unwrap();
             let fast_field_reader = DynamicFastFieldReader::<u64>::open(data)?;
-            assert_eq!(fast_field_reader.get(0), 4u64);
-            assert_eq!(fast_field_reader.get(1), 14_082_001u64);
-            assert_eq!(fast_field_reader.get(2), 3_052u64);
-            assert_eq!(fast_field_reader.get(3), 9002u64);
-            assert_eq!(fast_field_reader.get(4), 15_001u64);
-            assert_eq!(fast_field_reader.get(5), 777u64);
-            assert_eq!(fast_field_reader.get(6), 1_002u64);
-            assert_eq!(fast_field_reader.get(7), 1_501u64);
-            assert_eq!(fast_field_reader.get(8), 215u64);
+            assert_eq!(fast_field_reader.get_val(0), 4u64);
+            assert_eq!(fast_field_reader.get_val(1), 14_082_001u64);
+            assert_eq!(fast_field_reader.get_val(2), 3_052u64);
+            assert_eq!(fast_field_reader.get_val(3), 9002u64);
+            assert_eq!(fast_field_reader.get_val(4), 15_001u64);
+            assert_eq!(fast_field_reader.get_val(5), 777u64);
+            assert_eq!(fast_field_reader.get_val(6), 1_002u64);
+            assert_eq!(fast_field_reader.get_val(7), 1_501u64);
+            assert_eq!(fast_field_reader.get_val(8), 215u64);
         }
         Ok(())
     }
@@ -397,13 +393,13 @@ mod tests {
             serializer.close().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 35);
+        assert_eq!(file.len(), 43);
         {
             let fast_fields_composite = CompositeFile::open(&file).unwrap();
             let data = fast_fields_composite.open_read(*FIELD).unwrap();
             let fast_field_reader = DynamicFastFieldReader::<u64>::open(data)?;
             for doc in 0..10_000 {
-                assert_eq!(fast_field_reader.get(doc), 100_000u64);
+                assert_eq!(fast_field_reader.get_val(doc), 100_000u64);
             }
         }
         Ok(())
@@ -429,15 +425,15 @@ mod tests {
             serializer.close().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 80043);
+        assert_eq!(file.len(), 80051);
         {
             let fast_fields_composite = CompositeFile::open(&file)?;
             let data = fast_fields_composite.open_read(*FIELD).unwrap();
             let fast_field_reader = DynamicFastFieldReader::<u64>::open(data)?;
-            assert_eq!(fast_field_reader.get(0), 0u64);
+            assert_eq!(fast_field_reader.get_val(0), 0u64);
             for doc in 1..10_001 {
                 assert_eq!(
-                    fast_field_reader.get(doc),
+                    fast_field_reader.get_val(doc),
                     5_000_000_000_000_000_000u64 + doc as u64 - 1u64
                 );
             }
@@ -469,7 +465,8 @@ mod tests {
         }
         let file = directory.open_read(path).unwrap();
         // assert_eq!(file.len(), 17710 as usize); //bitpacked size
-        assert_eq!(file.len(), 10175_usize); // linear interpol size
+        // assert_eq!(file.len(), 10175_usize); // linear interpol size
+        assert_eq!(file.len(), 75_usize); // linear interpol size after calc improvement
         {
             let fast_fields_composite = CompositeFile::open(&file)?;
             let data = fast_fields_composite.open_read(i64_field).unwrap();
@@ -478,7 +475,7 @@ mod tests {
             assert_eq!(fast_field_reader.min_value(), -100i64);
             assert_eq!(fast_field_reader.max_value(), 9_999i64);
             for (doc, i) in (-100i64..10_000i64).enumerate() {
-                assert_eq!(fast_field_reader.get(doc as u32), i);
+                assert_eq!(fast_field_reader.get_val(doc as u64), i);
             }
             let mut buffer = vec![0i64; 100];
             fast_field_reader.get_range(53, &mut buffer[..]);
@@ -514,7 +511,7 @@ mod tests {
             let fast_fields_composite = CompositeFile::open(&file).unwrap();
             let data = fast_fields_composite.open_read(i64_field).unwrap();
             let fast_field_reader = DynamicFastFieldReader::<i64>::open(data)?;
-            assert_eq!(fast_field_reader.get(0u32), 0i64);
+            assert_eq!(fast_field_reader.get_val(0), 0i64);
         }
         Ok(())
     }
@@ -554,7 +551,7 @@ mod tests {
             let fast_field_reader = DynamicFastFieldReader::<u64>::open(data)?;
 
             for a in 0..n {
-                assert_eq!(fast_field_reader.get(a as u32), permutation[a as usize]);
+                assert_eq!(fast_field_reader.get_val(a as u64), permutation[a as usize]);
             }
         }
         Ok(())
@@ -845,19 +842,19 @@ mod tests {
         let dates_fast_field = fast_fields.dates(multi_date_field).unwrap();
         let mut dates = vec![];
         {
-            assert_eq!(date_fast_field.get(0u32).into_timestamp_micros(), 1i64);
+            assert_eq!(date_fast_field.get_val(0).into_timestamp_micros(), 1i64);
             dates_fast_field.get_vals(0u32, &mut dates);
             assert_eq!(dates.len(), 2);
             assert_eq!(dates[0].into_timestamp_micros(), 2i64);
             assert_eq!(dates[1].into_timestamp_micros(), 3i64);
         }
         {
-            assert_eq!(date_fast_field.get(1u32).into_timestamp_micros(), 4i64);
+            assert_eq!(date_fast_field.get_val(1).into_timestamp_micros(), 4i64);
             dates_fast_field.get_vals(1u32, &mut dates);
             assert!(dates.is_empty());
         }
         {
-            assert_eq!(date_fast_field.get(2u32).into_timestamp_micros(), 0i64);
+            assert_eq!(date_fast_field.get_val(2).into_timestamp_micros(), 0i64);
             dates_fast_field.get_vals(2u32, &mut dates);
             assert_eq!(dates.len(), 2);
             assert_eq!(dates[0].into_timestamp_micros(), 5i64);
@@ -869,10 +866,10 @@ mod tests {
     #[test]
     pub fn test_fastfield_bool() {
         let test_fastfield = DynamicFastFieldReader::<bool>::from(vec![true, false, true, false]);
-        assert_eq!(test_fastfield.get(0), true);
-        assert_eq!(test_fastfield.get(1), false);
-        assert_eq!(test_fastfield.get(2), true);
-        assert_eq!(test_fastfield.get(3), false);
+        assert_eq!(test_fastfield.get_val(0), true);
+        assert_eq!(test_fastfield.get_val(1), false);
+        assert_eq!(test_fastfield.get_val(2), true);
+        assert_eq!(test_fastfield.get_val(3), false);
     }
 
     #[test]
@@ -899,14 +896,14 @@ mod tests {
             serializer.close().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 36);
+        assert_eq!(file.len(), 44);
         let composite_file = CompositeFile::open(&file)?;
         let file = composite_file.open_read(field).unwrap();
         let fast_field_reader = DynamicFastFieldReader::<bool>::open(file)?;
-        assert_eq!(fast_field_reader.get(0), true);
-        assert_eq!(fast_field_reader.get(1), false);
-        assert_eq!(fast_field_reader.get(2), true);
-        assert_eq!(fast_field_reader.get(3), false);
+        assert_eq!(fast_field_reader.get_val(0), true);
+        assert_eq!(fast_field_reader.get_val(1), false);
+        assert_eq!(fast_field_reader.get_val(2), true);
+        assert_eq!(fast_field_reader.get_val(3), false);
 
         Ok(())
     }
@@ -935,13 +932,13 @@ mod tests {
             serializer.close().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 48);
+        assert_eq!(file.len(), 56);
         let composite_file = CompositeFile::open(&file)?;
         let file = composite_file.open_read(field).unwrap();
         let fast_field_reader = DynamicFastFieldReader::<bool>::open(file)?;
         for i in 0..25 {
-            assert_eq!(fast_field_reader.get(i * 2), true);
-            assert_eq!(fast_field_reader.get(i * 2 + 1), false);
+            assert_eq!(fast_field_reader.get_val(i * 2), true);
+            assert_eq!(fast_field_reader.get_val(i * 2 + 1), false);
         }
 
         Ok(())
@@ -969,11 +966,11 @@ mod tests {
             serializer.close().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 35);
+        assert_eq!(file.len(), 43);
         let composite_file = CompositeFile::open(&file)?;
         let file = composite_file.open_read(field).unwrap();
         let fast_field_reader = DynamicFastFieldReader::<bool>::open(file)?;
-        assert_eq!(fast_field_reader.get(0), false);
+        assert_eq!(fast_field_reader.get_val(0), false);
 
         Ok(())
     }

src/fastfield/multivalued/reader.rs

@@ -1,6 +1,8 @@
 use std::ops::Range;
 
-use crate::fastfield::{DynamicFastFieldReader, FastFieldReader, FastValue, MultiValueLength};
+use fastfield_codecs::Column;
+
+use crate::fastfield::{DynamicFastFieldReader, FastValue, MultiValueLength};
 use crate::DocId;
 
 /// Reader for a multivalued `u64` fast field.
@@ -31,8 +33,9 @@ impl<Item: FastValue> MultiValuedFastFieldReader<Item> {
     /// to the given document are `start..end`.
     #[inline]
     fn range(&self, doc: DocId) -> Range<u64> {
-        let start = self.idx_reader.get(doc);
-        let end = self.idx_reader.get(doc + 1);
+        let idx = doc as u64;
+        let start = self.idx_reader.get_val(idx);
+        let end = self.idx_reader.get_val(idx + 1);
         start..end
     }
 
@@ -55,7 +58,7 @@ impl<Item: FastValue> MultiValuedFastFieldReader<Item> {
     ///
     /// The min value does not take in account of possible
     /// deleted document, and should be considered as a lower bound
-    /// of the actual mimimum value.
+    /// of the actual minimum value.
     pub fn min_value(&self) -> Item {
         self.vals_reader.min_value()
     }

src/fastfield/multivalued/writer.rs

@@ -3,7 +3,7 @@ use std::io;
 use fnv::FnvHashMap;
 use tantivy_bitpacker::minmax;
 
-use crate::fastfield::serializer::BitpackedFastFieldSerializerLegacy;
+use crate::fastfield::serializer::BitpackedSerializerLegacy;
 use crate::fastfield::{value_to_u64, CompositeFastFieldSerializer, FastFieldType, FastValue};
 use crate::indexer::doc_id_mapping::DocIdMapping;
 use crate::postings::UnorderedTermId;
@@ -171,7 +171,7 @@ impl MultiValuedFastFieldWriter {
         }
         {
             // writing the values themselves.
-            let mut value_serializer: BitpackedFastFieldSerializerLegacy<'_, _>;
+            let mut value_serializer: BitpackedSerializerLegacy<'_, _>;
             if let Some(mapping) = mapping_opt {
                 value_serializer = serializer.new_u64_fast_field_with_idx(
                     self.field,

src/fastfield/reader.rs

@@ -2,63 +2,18 @@ use std::collections::HashMap;
 use std::marker::PhantomData;
 use std::path::Path;
 
-use fastfield_codecs::bitpacked::{
-    BitpackedFastFieldReader as BitpackedReader, BitpackedFastFieldSerializer,
-};
-use fastfield_codecs::linearinterpol::{
-    LinearInterpolFastFieldReader, LinearInterpolFastFieldSerializer,
-};
-use fastfield_codecs::multilinearinterpol::{
-    MultiLinearInterpolFastFieldReader, MultiLinearInterpolFastFieldSerializer,
-};
-use fastfield_codecs::{FastFieldCodecReader, FastFieldCodecSerializer};
+use common::BinarySerializable;
+use fastfield_codecs::bitpacked::{BitpackedCodec, BitpackedReader};
+use fastfield_codecs::blockwise_linear::{BlockwiseLinearCodec, BlockwiseLinearReader};
+use fastfield_codecs::linear::{LinearCodec, LinearReader};
+use fastfield_codecs::{Column, FastFieldCodec, FastFieldCodecType};
 
-use super::{FastValue, GCDFastFieldCodec, GCD_CODEC_ID};
+use super::gcd::open_gcd_from_bytes;
+use super::FastValue;
 use crate::directory::{CompositeFile, Directory, FileSlice, OwnedBytes, RamDirectory, WritePtr};
-use crate::fastfield::{CompositeFastFieldSerializer, FastFieldsWriter};
+use crate::error::DataCorruption;
+use crate::fastfield::{CompositeFastFieldSerializer, FastFieldsWriter, GCDReader};
 use crate::schema::{Schema, FAST};
-use crate::DocId;
-
-/// FastFieldReader is the trait to access fast field data.
-pub trait FastFieldReader<Item: FastValue>: Clone {
-    /// Return the value associated to the given document.
-    ///
-    /// This accessor should return as fast as possible.
-    ///
-    /// # Panics
-    ///
-    /// May panic if `doc` is greater than the segment
-    fn get(&self, doc: DocId) -> Item;
-
-    /// Fills an output buffer with the fast field values
-    /// associated with the `DocId` going from
-    /// `start` to `start + output.len()`.
-    ///
-    /// Regardless of the type of `Item`, this method works
-    /// - transmuting the output array
-    /// - extracting the `Item`s as if they were `u64`
-    /// - possibly converting the `u64` value to the right type.
-    ///
-    /// # Panics
-    ///
-    /// May panic if `start + output.len()` is greater than
-    /// the segment's `maxdoc`.
-    fn get_range(&self, start: u64, output: &mut [Item]);
-
-    /// Returns the minimum value for this fast field.
-    ///
-    /// The min value does not take in account of possible
-    /// deleted document, and should be considered as a lower bound
-    /// of the actual mimimum value.
-    fn min_value(&self) -> Item;
-
-    /// Returns the maximum value for this fast field.
-    ///
-    /// The max value does not take in account of possible
-    /// deleted document, and should be considered as an upper bound
-    /// of the actual maximum value.
-    fn max_value(&self) -> Item;
-}
 
 #[derive(Clone)]
 /// DynamicFastFieldReader wraps different readers to access
@@ -67,145 +22,121 @@ pub enum DynamicFastFieldReader<Item: FastValue> {
     /// Bitpacked compressed fastfield data.
     Bitpacked(FastFieldReaderCodecWrapper<Item, BitpackedReader>),
     /// Linear interpolated values + bitpacked
-    LinearInterpol(FastFieldReaderCodecWrapper<Item, LinearInterpolFastFieldReader>),
+    Linear(FastFieldReaderCodecWrapper<Item, LinearReader>),
     /// Blockwise linear interpolated values + bitpacked
-    MultiLinearInterpol(FastFieldReaderCodecWrapper<Item, MultiLinearInterpolFastFieldReader>),
+    BlockwiseLinear(FastFieldReaderCodecWrapper<Item, BlockwiseLinearReader>),
 
     /// GCD and Bitpacked compressed fastfield data.
-    BitpackedGCD(FastFieldReaderCodecWrapper<Item, GCDFastFieldCodec<BitpackedReader>>),
+    BitpackedGCD(FastFieldReaderCodecWrapper<Item, GCDReader<BitpackedReader>>),
     /// GCD and Linear interpolated values + bitpacked
-    LinearInterpolGCD(
-        FastFieldReaderCodecWrapper<Item, GCDFastFieldCodec<LinearInterpolFastFieldReader>>,
-    ),
+    LinearGCD(FastFieldReaderCodecWrapper<Item, GCDReader<LinearReader>>),
     /// GCD and Blockwise linear interpolated values + bitpacked
-    MultiLinearInterpolGCD(
-        FastFieldReaderCodecWrapper<Item, GCDFastFieldCodec<MultiLinearInterpolFastFieldReader>>,
-    ),
+    BlockwiseLinearGCD(FastFieldReaderCodecWrapper<Item, GCDReader<BlockwiseLinearReader>>),
 }
 
 impl<Item: FastValue> DynamicFastFieldReader<Item> {
     /// Returns correct the reader wrapped in the `DynamicFastFieldReader` enum for the data.
     pub fn open_from_id(
         mut bytes: OwnedBytes,
-        codec_id: u8,
+        codec_type: FastFieldCodecType,
     ) -> crate::Result<DynamicFastFieldReader<Item>> {
-        let reader = match codec_id {
-            BitpackedFastFieldSerializer::ID => {
-                DynamicFastFieldReader::Bitpacked(FastFieldReaderCodecWrapper::<
-                    Item,
-                    BitpackedReader,
-                >::open_from_bytes(bytes)?)
+        let reader = match codec_type {
+            FastFieldCodecType::Bitpacked => {
+                DynamicFastFieldReader::Bitpacked(BitpackedCodec::open_from_bytes(bytes)?.into())
             }
-            LinearInterpolFastFieldSerializer::ID => {
-                DynamicFastFieldReader::LinearInterpol(FastFieldReaderCodecWrapper::<
-                    Item,
-                    LinearInterpolFastFieldReader,
-                >::open_from_bytes(bytes)?)
+            FastFieldCodecType::Linear => {
+                DynamicFastFieldReader::Linear(LinearCodec::open_from_bytes(bytes)?.into())
             }
-            MultiLinearInterpolFastFieldSerializer::ID => {
-                DynamicFastFieldReader::MultiLinearInterpol(FastFieldReaderCodecWrapper::<
-                    Item,
-                    MultiLinearInterpolFastFieldReader,
-                >::open_from_bytes(
-                    bytes
-                )?)
-            }
-            _ if codec_id == GCD_CODEC_ID => {
-                let codec_id = bytes.read_u8();
-
-                match codec_id {
-                    BitpackedFastFieldSerializer::ID => {
-                        DynamicFastFieldReader::BitpackedGCD(FastFieldReaderCodecWrapper::<
-                            Item,
-                            GCDFastFieldCodec<BitpackedReader>,
-                        >::open_from_bytes(
-                            bytes
-                        )?)
-                    }
-                    LinearInterpolFastFieldSerializer::ID => {
-                        DynamicFastFieldReader::LinearInterpolGCD(FastFieldReaderCodecWrapper::<
-                            Item,
-                            GCDFastFieldCodec<LinearInterpolFastFieldReader>,
-                        >::open_from_bytes(
-                            bytes
-                        )?)
-                    }
-                    MultiLinearInterpolFastFieldSerializer::ID => {
-                        DynamicFastFieldReader::MultiLinearInterpolGCD(
-                            FastFieldReaderCodecWrapper::<
-                                Item,
-                                GCDFastFieldCodec<MultiLinearInterpolFastFieldReader>,
-                            >::open_from_bytes(bytes)?,
+            FastFieldCodecType::BlockwiseLinear => DynamicFastFieldReader::BlockwiseLinear(
+                BlockwiseLinearCodec::open_from_bytes(bytes)?.into(),
+            ),
+            FastFieldCodecType::Gcd => {
+                let codec_type = FastFieldCodecType::deserialize(&mut bytes)?;
+                match codec_type {
+                    FastFieldCodecType::Bitpacked => DynamicFastFieldReader::BitpackedGCD(
+                        open_gcd_from_bytes::<BitpackedCodec>(bytes)?.into(),
+                    ),
+                    FastFieldCodecType::Linear => DynamicFastFieldReader::LinearGCD(
+                        open_gcd_from_bytes::<LinearCodec>(bytes)?.into(),
+                    ),
+                    FastFieldCodecType::BlockwiseLinear => {
+                        DynamicFastFieldReader::BlockwiseLinearGCD(
+                            open_gcd_from_bytes::<BlockwiseLinearCodec>(bytes)?.into(),
                         )
                     }
-                    _ => {
-                        panic!(
-                            "unknown fastfield codec id {:?}. Data corrupted or using old tantivy \
-                             version.",
-                            codec_id
+                    FastFieldCodecType::Gcd => {
+                        return Err(DataCorruption::comment_only(
+                            "Gcd codec wrapped into another gcd codec. This combination is not \
+                             allowed.",
                         )
+                        .into())
                     }
                 }
             }
-            _ => {
-                panic!(
-                    "unknown fastfield codec id {:?}. Data corrupted or using old tantivy version.",
-                    codec_id
-                )
-            }
         };
         Ok(reader)
     }
+
     /// Returns correct the reader wrapped in the `DynamicFastFieldReader` enum for the data.
     pub fn open(file: FileSlice) -> crate::Result<DynamicFastFieldReader<Item>> {
         let mut bytes = file.read_bytes()?;
-        let codec_id = bytes.read_u8();
-
-        Self::open_from_id(bytes, codec_id)
+        let codec_type = FastFieldCodecType::deserialize(&mut bytes)?;
+        Self::open_from_id(bytes, codec_type)
     }
 }
 
-impl<Item: FastValue> FastFieldReader<Item> for DynamicFastFieldReader<Item> {
+impl<Item: FastValue> Column<Item> for DynamicFastFieldReader<Item> {
     #[inline]
-    fn get(&self, doc: DocId) -> Item {
+    fn get_val(&self, idx: u64) -> Item {
         match self {
-            Self::Bitpacked(reader) => reader.get(doc),
-            Self::LinearInterpol(reader) => reader.get(doc),
-            Self::MultiLinearInterpol(reader) => reader.get(doc),
-            Self::BitpackedGCD(reader) => reader.get(doc),
-            Self::LinearInterpolGCD(reader) => reader.get(doc),
-            Self::MultiLinearInterpolGCD(reader) => reader.get(doc),
+            Self::Bitpacked(reader) => reader.get_val(idx),
+            Self::Linear(reader) => reader.get_val(idx),
+            Self::BlockwiseLinear(reader) => reader.get_val(idx),
+            Self::BitpackedGCD(reader) => reader.get_val(idx),
+            Self::LinearGCD(reader) => reader.get_val(idx),
+            Self::BlockwiseLinearGCD(reader) => reader.get_val(idx),
         }
     }
     #[inline]
     fn get_range(&self, start: u64, output: &mut [Item]) {
         match self {
             Self::Bitpacked(reader) => reader.get_range(start, output),
-            Self::LinearInterpol(reader) => reader.get_range(start, output),
-            Self::MultiLinearInterpol(reader) => reader.get_range(start, output),
+            Self::Linear(reader) => reader.get_range(start, output),
+            Self::BlockwiseLinear(reader) => reader.get_range(start, output),
             Self::BitpackedGCD(reader) => reader.get_range(start, output),
-            Self::LinearInterpolGCD(reader) => reader.get_range(start, output),
-            Self::MultiLinearInterpolGCD(reader) => reader.get_range(start, output),
+            Self::LinearGCD(reader) => reader.get_range(start, output),
+            Self::BlockwiseLinearGCD(reader) => reader.get_range(start, output),
         }
     }
     fn min_value(&self) -> Item {
         match self {
             Self::Bitpacked(reader) => reader.min_value(),
-            Self::LinearInterpol(reader) => reader.min_value(),
-            Self::MultiLinearInterpol(reader) => reader.min_value(),
+            Self::Linear(reader) => reader.min_value(),
+            Self::BlockwiseLinear(reader) => reader.min_value(),
             Self::BitpackedGCD(reader) => reader.min_value(),
-            Self::LinearInterpolGCD(reader) => reader.min_value(),
-            Self::MultiLinearInterpolGCD(reader) => reader.min_value(),
+            Self::LinearGCD(reader) => reader.min_value(),
+            Self::BlockwiseLinearGCD(reader) => reader.min_value(),
         }
     }
     fn max_value(&self) -> Item {
         match self {
             Self::Bitpacked(reader) => reader.max_value(),
-            Self::LinearInterpol(reader) => reader.max_value(),
-            Self::MultiLinearInterpol(reader) => reader.max_value(),
+            Self::Linear(reader) => reader.max_value(),
+            Self::BlockwiseLinear(reader) => reader.max_value(),
             Self::BitpackedGCD(reader) => reader.max_value(),
-            Self::LinearInterpolGCD(reader) => reader.max_value(),
-            Self::MultiLinearInterpolGCD(reader) => reader.max_value(),
+            Self::LinearGCD(reader) => reader.max_value(),
+            Self::BlockwiseLinearGCD(reader) => reader.max_value(),
+        }
+    }
+
+    fn num_vals(&self) -> u64 {
+        match self {
+            Self::Bitpacked(reader) => reader.num_vals(),
+            Self::Linear(reader) => reader.num_vals(),
+            Self::BlockwiseLinear(reader) => reader.num_vals(),
+            Self::BitpackedGCD(reader) => reader.num_vals(),
+            Self::LinearGCD(reader) => reader.num_vals(),
+            Self::BlockwiseLinearGCD(reader) => reader.num_vals(),
         }
     }
 }
@@ -216,35 +147,24 @@ impl<Item: FastValue> FastFieldReader<Item> for DynamicFastFieldReader<Item> {
 #[derive(Clone)]
 pub struct FastFieldReaderCodecWrapper<Item: FastValue, CodecReader> {
     reader: CodecReader,
-    bytes: OwnedBytes,
     _phantom: PhantomData<Item>,
 }
 
-impl<Item: FastValue, C: FastFieldCodecReader> FastFieldReaderCodecWrapper<Item, C> {
-    /// Opens a fast field given a file.
-    pub fn open(file: FileSlice) -> crate::Result<Self> {
-        let mut bytes = file.read_bytes()?;
-        let codec_id = bytes.read_u8();
-        assert_eq!(
-            BitpackedFastFieldSerializer::ID,
-            codec_id,
-            "Tried to open fast field as bitpacked encoded (id=1), but got serializer with \
-             different id"
-        );
-        Self::open_from_bytes(bytes)
-    }
-    /// Opens a fast field given the bytes.
-    pub fn open_from_bytes(bytes: OwnedBytes) -> crate::Result<Self> {
-        let reader = C::open_from_bytes(bytes.as_slice())?;
-        Ok(FastFieldReaderCodecWrapper {
+impl<Item: FastValue, CodecReader> From<CodecReader>
+    for FastFieldReaderCodecWrapper<Item, CodecReader>
+{
+    fn from(reader: CodecReader) -> Self {
+        FastFieldReaderCodecWrapper {
             reader,
-            bytes,
             _phantom: PhantomData,
-        })
+        }
     }
+}
+
+impl<Item: FastValue, D: Column> FastFieldReaderCodecWrapper<Item, D> {
     #[inline]
-    pub(crate) fn get_u64(&self, doc: u64) -> Item {
-        let data = self.reader.get_u64(doc, self.bytes.as_slice());
+    pub(crate) fn get_u64(&self, idx: u64) -> Item {
+        let data = self.reader.get_val(idx);
         Item::from_u64(data)
     }
 
@@ -267,9 +187,7 @@ impl<Item: FastValue, C: FastFieldCodecReader> FastFieldReaderCodecWrapper<Item,
     }
 }
 
-impl<Item: FastValue, C: FastFieldCodecReader + Clone> FastFieldReader<Item>
-    for FastFieldReaderCodecWrapper<Item, C>
-{
+impl<Item: FastValue, C: Column + Clone> Column<Item> for FastFieldReaderCodecWrapper<Item, C> {
     /// Return the value associated to the given document.
     ///
     /// This accessor should return as fast as possible.
@@ -278,8 +196,8 @@ impl<Item: FastValue, C: FastFieldCodecReader + Clone> FastFieldReader<Item>
     ///
     /// May panic if `doc` is greater than the segment
     // `maxdoc`.
-    fn get(&self, doc: DocId) -> Item {
-        self.get_u64(u64::from(doc))
+    fn get_val(&self, idx: u64) -> Item {
+        self.get_u64(idx)
     }
 
     /// Fills an output buffer with the fast field values
@@ -316,6 +234,10 @@ impl<Item: FastValue, C: FastFieldCodecReader + Clone> FastFieldReader<Item>
     fn max_value(&self) -> Item {
         Item::from_u64(self.reader.max_value())
     }
+
+    fn num_vals(&self) -> u64 {
+        self.reader.num_vals()
+    }
 }
 
 impl<Item: FastValue> From<Vec<Item>> for DynamicFastFieldReader<Item> {

src/fastfield/serializer/mod.rs

@@ -1,17 +1,17 @@
 use std::io::{self, Write};
+use std::num::NonZeroU64;
 
 use common::{BinarySerializable, CountingWriter};
-pub use fastfield_codecs::bitpacked::{
-    BitpackedFastFieldSerializer, BitpackedFastFieldSerializerLegacy,
-};
-use fastfield_codecs::linearinterpol::LinearInterpolFastFieldSerializer;
-use fastfield_codecs::multilinearinterpol::MultiLinearInterpolFastFieldSerializer;
-pub use fastfield_codecs::{FastFieldCodecSerializer, FastFieldDataAccess, FastFieldStats};
+use fastdivide::DividerU64;
+pub use fastfield_codecs::bitpacked::{BitpackedCodec, BitpackedSerializerLegacy};
+use fastfield_codecs::blockwise_linear::BlockwiseLinearCodec;
+use fastfield_codecs::linear::LinearCodec;
+use fastfield_codecs::FastFieldCodecType;
+pub use fastfield_codecs::{Column, FastFieldCodec, FastFieldStats};
 
-use super::{find_gcd, FastFieldCodecName, ALL_CODECS, GCD_DEFAULT};
+use super::{find_gcd, ALL_CODECS, GCD_DEFAULT};
 use crate::directory::{CompositeWrite, WritePtr};
 use crate::fastfield::gcd::write_gcd_header;
-use crate::fastfield::GCD_CODEC_ID;
 use crate::schema::Field;
 
 /// `CompositeFastFieldSerializer` is in charge of serializing
@@ -41,7 +41,7 @@ pub struct CompositeFastFieldSerializer {
 
 #[derive(Debug, Clone)]
 pub struct FastFieldCodecEnableCheck {
-    enabled_codecs: Vec<FastFieldCodecName>,
+    enabled_codecs: Vec<FastFieldCodecType>,
 }
 impl FastFieldCodecEnableCheck {
     fn allow_all() -> Self {
@@ -49,31 +49,28 @@ impl FastFieldCodecEnableCheck {
             enabled_codecs: ALL_CODECS.to_vec(),
         }
     }
-    fn is_enabled(&self, codec_name: FastFieldCodecName) -> bool {
-        self.enabled_codecs.contains(&codec_name)
+    fn is_enabled(&self, code_type: FastFieldCodecType) -> bool {
+        self.enabled_codecs.contains(&code_type)
     }
 }
 
-impl From<FastFieldCodecName> for FastFieldCodecEnableCheck {
-    fn from(codec_name: FastFieldCodecName) -> Self {
+impl From<FastFieldCodecType> for FastFieldCodecEnableCheck {
+    fn from(code_type: FastFieldCodecType) -> Self {
         FastFieldCodecEnableCheck {
-            enabled_codecs: vec![codec_name],
+            enabled_codecs: vec![code_type],
         }
     }
 }
 
 // use this, when this is merged and stabilized explicit_generic_args_with_impl_trait
 // https://github.com/rust-lang/rust/pull/86176
-fn codec_estimation<T: FastFieldCodecSerializer, A: FastFieldDataAccess>(
-    stats: FastFieldStats,
-    fastfield_accessor: &A,
-    estimations: &mut Vec<(f32, &str, u8)>,
+fn codec_estimation<C: FastFieldCodec>(
+    fastfield_accessor: &impl Column,
+    estimations: &mut Vec<(f32, FastFieldCodecType)>,
 ) {
-    if !T::is_applicable(fastfield_accessor, stats.clone()) {
-        return;
+    if let Some(ratio) = C::estimate(fastfield_accessor) {
+        estimations.push((ratio, C::CODEC_TYPE));
     }
-    let (ratio, name, id) = (T::estimate(fastfield_accessor, stats), T::NAME, T::ID);
-    estimations.push((ratio, name, id));
 }
 
 impl CompositeFastFieldSerializer {
@@ -97,99 +94,100 @@ impl CompositeFastFieldSerializer {
 
     /// Serialize data into a new u64 fast field. The best compression codec will be chosen
     /// automatically.
-    pub fn create_auto_detect_u64_fast_field<F, I>(
+    pub fn create_auto_detect_u64_fast_field(
         &mut self,
         field: Field,
-        stats: FastFieldStats,
-        fastfield_accessor: impl FastFieldDataAccess,
-        iter_gen: F,
-    ) -> io::Result<()>
-    where
-        F: Fn() -> I,
-        I: Iterator<Item = u64>,
-    {
-        self.create_auto_detect_u64_fast_field_with_idx(
-            field,
-            stats,
-            fastfield_accessor,
-            iter_gen,
-            0,
-        )
+        fastfield_accessor: impl Column,
+    ) -> io::Result<()> {
+        self.create_auto_detect_u64_fast_field_with_idx(field, fastfield_accessor, 0)
     }
 
     /// Serialize data into a new u64 fast field. The best compression codec will be chosen
     /// automatically.
-    pub fn write_header<W: Write>(field_write: &mut W, codec_id: u8) -> io::Result<()> {
-        codec_id.serialize(field_write)?;
-
+    pub fn write_header<W: Write>(
+        field_write: &mut W,
+        codec_type: FastFieldCodecType,
+    ) -> io::Result<()> {
+        codec_type.to_code().serialize(field_write)?;
         Ok(())
     }
 
     /// Serialize data into a new u64 fast field. The best compression codec will be chosen
     /// automatically.
-    pub fn create_auto_detect_u64_fast_field_with_idx<F, I>(
+    pub fn create_auto_detect_u64_fast_field_with_idx(
         &mut self,
         field: Field,
-        stats: FastFieldStats,
-        fastfield_accessor: impl FastFieldDataAccess,
-        iter_gen: F,
+        fastfield_accessor: impl Column,
         idx: usize,
-    ) -> io::Result<()>
-    where
-        F: Fn() -> I,
-        I: Iterator<Item = u64>,
-    {
+    ) -> io::Result<()> {
+        let min_value = fastfield_accessor.min_value();
         let field_write = self.composite_write.for_field_with_idx(field, idx);
-        let gcd = find_gcd(iter_gen().map(|val| val - stats.min_value)).unwrap_or(GCD_DEFAULT);
+        let gcd = find_gcd(fastfield_accessor.iter().map(|val| val - min_value))
+            .map(NonZeroU64::get)
+            .unwrap_or(GCD_DEFAULT);
 
         if gcd == 1 {
             return Self::create_auto_detect_u64_fast_field_with_idx_gcd(
                 self.codec_enable_checker.clone(),
                 field,
                 field_write,
-                stats,
                 fastfield_accessor,
-                iter_gen(),
-                iter_gen(),
             );
         }
 
-        Self::write_header(field_write, GCD_CODEC_ID)?;
-        struct GCDWrappedFFAccess<T: FastFieldDataAccess> {
+        Self::write_header(field_write, FastFieldCodecType::Gcd)?;
+        struct GCDWrappedFFAccess<T: Column> {
             fastfield_accessor: T,
-            min_value: u64,
-            gcd: u64,
+            base_value: u64,
+            max_value: u64,
+            num_vals: u64,
+            gcd: DividerU64,
         }
-        impl<T: FastFieldDataAccess> FastFieldDataAccess for GCDWrappedFFAccess<T> {
+
+        impl<T: Column> Column for GCDWrappedFFAccess<T> {
             fn get_val(&self, position: u64) -> u64 {
-                (self.fastfield_accessor.get_val(position) - self.min_value) / self.gcd
+                self.gcd
+                    .divide(self.fastfield_accessor.get_val(position) - self.base_value)
+            }
+            fn iter(&self) -> Box<dyn Iterator<Item = u64> + '_> {
+                Box::new(
+                    self.fastfield_accessor
+                        .iter()
+                        .map(|val| self.gcd.divide(val - self.base_value)),
+                )
+            }
+            fn min_value(&self) -> u64 {
+                0
+            }
+
+            fn max_value(&self) -> u64 {
+                self.max_value
+            }
+
+            fn num_vals(&self) -> u64 {
+                self.num_vals
             }
         }
 
+        let num_vals = fastfield_accessor.num_vals();
+        let base_value = fastfield_accessor.min_value();
+        let max_value = (fastfield_accessor.max_value() - fastfield_accessor.min_value()) / gcd;
+
         let fastfield_accessor = GCDWrappedFFAccess {
             fastfield_accessor,
-            min_value: stats.min_value,
-            gcd,
+            base_value,
+            max_value,
+            num_vals,
+            gcd: DividerU64::divide_by(gcd),
         };
 
-        let min_value = stats.min_value;
-        let stats = FastFieldStats {
-            min_value: 0,
-            max_value: (stats.max_value - stats.min_value) / gcd,
-            num_vals: stats.num_vals,
-        };
-        let iter1 = iter_gen().map(|val| (val - min_value) / gcd);
-        let iter2 = iter_gen().map(|val| (val - min_value) / gcd);
         Self::create_auto_detect_u64_fast_field_with_idx_gcd(
             self.codec_enable_checker.clone(),
             field,
             field_write,
-            stats,
             fastfield_accessor,
-            iter1,
-            iter2,
         )?;
-        write_gcd_header(field_write, min_value, gcd)?;
+        write_gcd_header(field_write, base_value, gcd, num_vals)?;
         Ok(())
     }
 
@@ -199,38 +197,23 @@ impl CompositeFastFieldSerializer {
         codec_enable_checker: FastFieldCodecEnableCheck,
         field: Field,
         field_write: &mut CountingWriter<W>,
-        stats: FastFieldStats,
-        fastfield_accessor: impl FastFieldDataAccess,
-        iter1: impl Iterator<Item = u64>,
-        iter2: impl Iterator<Item = u64>,
+        fastfield_accessor: impl Column,
     ) -> io::Result<()> {
         let mut estimations = vec![];
 
-        if codec_enable_checker.is_enabled(FastFieldCodecName::Bitpacked) {
-            codec_estimation::<BitpackedFastFieldSerializer, _>(
-                stats.clone(),
-                &fastfield_accessor,
-                &mut estimations,
-            );
+        if codec_enable_checker.is_enabled(FastFieldCodecType::Bitpacked) {
+            codec_estimation::<BitpackedCodec>(&fastfield_accessor, &mut estimations);
         }
-        if codec_enable_checker.is_enabled(FastFieldCodecName::LinearInterpol) {
-            codec_estimation::<LinearInterpolFastFieldSerializer, _>(
-                stats.clone(),
-                &fastfield_accessor,
-                &mut estimations,
-            );
+        if codec_enable_checker.is_enabled(FastFieldCodecType::Linear) {
+            codec_estimation::<LinearCodec>(&fastfield_accessor, &mut estimations);
         }
-        if codec_enable_checker.is_enabled(FastFieldCodecName::BlockwiseLinearInterpol) {
-            codec_estimation::<MultiLinearInterpolFastFieldSerializer, _>(
-                stats.clone(),
-                &fastfield_accessor,
-                &mut estimations,
-            );
+        if codec_enable_checker.is_enabled(FastFieldCodecType::BlockwiseLinear) {
+            codec_estimation::<BlockwiseLinearCodec>(&fastfield_accessor, &mut estimations);
         }
         if let Some(broken_estimation) = estimations.iter().find(|estimation| estimation.0.is_nan())
         {
             warn!(
-                "broken estimation for fast field codec {}",
+                "broken estimation for fast field codec {:?}",
                 broken_estimation.1
             );
         }
@@ -238,43 +221,25 @@ impl CompositeFastFieldSerializer {
         // codecs
         estimations.retain(|estimation| !estimation.0.is_nan() && estimation.0 != f32::MAX);
         estimations.sort_by(|a, b| a.0.partial_cmp(&b.0).unwrap());
-        let (_ratio, name, id) = estimations[0];
-        debug!(
-            "choosing fast field codec {} for field_id {:?}",
-            name, field
-        ); // todo print actual field name
+        let (_ratio, codec_type) = estimations[0];
+        debug!("choosing fast field codec {codec_type:?} for field_id {field:?}"); // todo print actual field name
 
-        Self::write_header(field_write, id)?;
-        match name {
-            BitpackedFastFieldSerializer::NAME => {
-                BitpackedFastFieldSerializer::serialize(
-                    field_write,
-                    &fastfield_accessor,
-                    stats,
-                    iter1,
-                    iter2,
-                )?;
+        Self::write_header(field_write, codec_type)?;
+        match codec_type {
+            FastFieldCodecType::Bitpacked => {
+                BitpackedCodec::serialize(field_write, &fastfield_accessor)?;
             }
-            LinearInterpolFastFieldSerializer::NAME => {
-                LinearInterpolFastFieldSerializer::serialize(
-                    field_write,
-                    &fastfield_accessor,
-                    stats,
-                    iter1,
-                    iter2,
-                )?;
+            FastFieldCodecType::Linear => {
+                LinearCodec::serialize(field_write, &fastfield_accessor)?;
             }
-            MultiLinearInterpolFastFieldSerializer::NAME => {
-                MultiLinearInterpolFastFieldSerializer::serialize(
-                    field_write,
-                    &fastfield_accessor,
-                    stats,
-                    iter1,
-                    iter2,
-                )?;
+            FastFieldCodecType::BlockwiseLinear => {
+                BlockwiseLinearCodec::serialize(field_write, &fastfield_accessor)?;
             }
-            _ => {
-                panic!("unknown fastfield serializer {}", name)
+            FastFieldCodecType::Gcd => {
+                return Err(io::Error::new(
+                    io::ErrorKind::InvalidData,
+                    "GCD codec not supported.",
+                ));
             }
         }
         field_write.flush()?;
@@ -288,7 +253,7 @@ impl CompositeFastFieldSerializer {
         field: Field,
         min_value: u64,
         max_value: u64,
-    ) -> io::Result<BitpackedFastFieldSerializerLegacy<'_, CountingWriter<WritePtr>>> {
+    ) -> io::Result<BitpackedSerializerLegacy<'_, CountingWriter<WritePtr>>> {
         self.new_u64_fast_field_with_idx(field, min_value, max_value, 0)
     }
 
@@ -298,7 +263,7 @@ impl CompositeFastFieldSerializer {
         field: Field,
         min_value: u64,
         max_value: u64,
-    ) -> io::Result<BitpackedFastFieldSerializerLegacy<'_, CountingWriter<WritePtr>>> {
+    ) -> io::Result<BitpackedSerializerLegacy<'_, CountingWriter<WritePtr>>> {
         self.new_u64_fast_field_with_idx(field, min_value, max_value, 0)
     }
 
@@ -309,12 +274,11 @@ impl CompositeFastFieldSerializer {
         min_value: u64,
         max_value: u64,
         idx: usize,
-    ) -> io::Result<BitpackedFastFieldSerializerLegacy<'_, CountingWriter<WritePtr>>> {
+    ) -> io::Result<BitpackedSerializerLegacy<'_, CountingWriter<WritePtr>>> {
         let field_write = self.composite_write.for_field_with_idx(field, idx);
         // Prepend codec id to field data for compatibility with DynamicFastFieldReader.
-        let id = BitpackedFastFieldSerializer::ID;
-        id.serialize(field_write)?;
-        BitpackedFastFieldSerializerLegacy::open(field_write, min_value, max_value)
+        FastFieldCodecType::Bitpacked.serialize(field_write)?;
+        BitpackedSerializerLegacy::open(field_write, min_value, max_value)
     }
 
     /// Start serializing a new [u8] fast field

src/fastfield/writer.rs

@@ -2,12 +2,13 @@ use std::collections::HashMap;
 use std::io;
 
 use common;
+use fastfield_codecs::Column;
 use fnv::FnvHashMap;
 use tantivy_bitpacker::BlockedBitpacker;
 
 use super::multivalued::MultiValuedFastFieldWriter;
 use super::serializer::FastFieldStats;
-use super::{FastFieldDataAccess, FastFieldType, FastValue};
+use super::{FastFieldType, FastValue};
 use crate::fastfield::{BytesFastFieldWriter, CompositeFastFieldSerializer};
 use crate::indexer::doc_id_mapping::DocIdMapping;
 use crate::postings::UnorderedTermId;
@@ -293,7 +294,7 @@ impl IntFastFieldWriter {
 
     /// Records a new value.
     ///
-    /// The n-th value being recorded is implicitely
+    /// The n-th value being recorded is implicitly
     /// associated to the document with the `DocId` n.
     /// (Well, `n-1` actually because of 0-indexing)
     pub fn add_val(&mut self, val: u64) {
@@ -359,38 +360,20 @@ impl IntFastFieldWriter {
             (self.val_min, self.val_max)
         };
 
-        let fastfield_accessor = WriterFastFieldAccessProvider {
-            doc_id_map,
-            vals: &self.vals,
-        };
         let stats = FastFieldStats {
             min_value: min,
             max_value: max,
             num_vals: self.val_count as u64,
         };
 
-        if let Some(doc_id_map) = doc_id_map {
-            let iter_gen = || {
-                doc_id_map
-                    .iter_old_doc_ids()
-                    .map(|doc_id| self.vals.get(doc_id as usize))
-            };
-            serializer.create_auto_detect_u64_fast_field(
-                self.field,
-                stats,
-                fastfield_accessor,
-                iter_gen,
-            )?;
-        } else {
-            let iter_gen = || self.vals.iter();
-
-            serializer.create_auto_detect_u64_fast_field(
-                self.field,
-                stats,
-                fastfield_accessor,
-                iter_gen,
-            )?;
+        let fastfield_accessor = WriterFastFieldAccessProvider {
+            doc_id_map,
+            vals: &self.vals,
+            stats,
         };
+
+        serializer.create_auto_detect_u64_fast_field(self.field, fastfield_accessor)?;
+
         Ok(())
     }
 }
@@ -399,8 +382,9 @@ impl IntFastFieldWriter {
 struct WriterFastFieldAccessProvider<'map, 'bitp> {
     doc_id_map: Option<&'map DocIdMapping>,
     vals: &'bitp BlockedBitpacker,
+    stats: FastFieldStats,
 }
-impl<'map, 'bitp> FastFieldDataAccess for WriterFastFieldAccessProvider<'map, 'bitp> {
+impl<'map, 'bitp> Column for WriterFastFieldAccessProvider<'map, 'bitp> {
     /// Return the value associated to the given doc.
     ///
     /// Whenever possible use the Iterator passed to the fastfield creation instead, for performance
@@ -419,4 +403,28 @@ impl<'map, 'bitp> FastFieldDataAccess for WriterFastFieldAccessProvider<'map, 'b
             self.vals.get(doc as usize)
         }
     }
+
+    fn iter(&self) -> Box<dyn Iterator<Item = u64> + '_> {
+        if let Some(doc_id_map) = self.doc_id_map {
+            Box::new(
+                doc_id_map
+                    .iter_old_doc_ids()
+                    .map(|doc_id| self.vals.get(doc_id as usize)),
+            )
+        } else {
+            Box::new(self.vals.iter())
+        }
+    }
+
+    fn min_value(&self) -> u64 {
+        self.stats.min_value
+    }
+
+    fn max_value(&self) -> u64 {
+        self.stats.max_value
+    }
+
+    fn num_vals(&self) -> u64 {
+        self.stats.num_vals
+    }
 }

src/indexer/doc_id_mapping.rs

@@ -2,35 +2,42 @@
 //! to get mappings from old doc_id to new doc_id and vice versa, after sorting
 
 use std::cmp::Reverse;
-use std::ops::Index;
 
 use super::SegmentWriter;
 use crate::schema::{Field, Schema};
-use crate::{DocId, IndexSortByField, Order, SegmentOrdinal, TantivyError};
+use crate::{DocAddress, DocId, IndexSortByField, Order, TantivyError};
 
 /// Struct to provide mapping from new doc_id to old doc_id and segment.
 #[derive(Clone)]
 pub(crate) struct SegmentDocIdMapping {
-    new_doc_id_to_old_and_segment: Vec<(DocId, SegmentOrdinal)>,
+    new_doc_id_to_old_doc_addr: Vec<DocAddress>,
     is_trivial: bool,
 }
 
 impl SegmentDocIdMapping {
-    pub(crate) fn new(
-        new_doc_id_to_old_and_segment: Vec<(DocId, SegmentOrdinal)>,
-        is_trivial: bool,
-    ) -> Self {
+    pub(crate) fn new(new_doc_id_to_old_and_segment: Vec<DocAddress>, is_trivial: bool) -> Self {
         Self {
-            new_doc_id_to_old_and_segment,
+            new_doc_id_to_old_doc_addr: new_doc_id_to_old_and_segment,
             is_trivial,
         }
     }
-    pub(crate) fn iter(&self) -> impl Iterator<Item = &(DocId, SegmentOrdinal)> {
-        self.new_doc_id_to_old_and_segment.iter()
+
+    /// Returns an iterator over the old document addresses, ordered by the new document ids.
+    ///
+    /// In the returned `DocAddress`, the `segment_ord` is the ordinal of targetted segment
+    /// in the list of merged segments.
+    pub(crate) fn iter_old_doc_addrs(&self) -> impl Iterator<Item = DocAddress> + '_ {
+        self.new_doc_id_to_old_doc_addr.iter().copied()
     }
+
     pub(crate) fn len(&self) -> usize {
-        self.new_doc_id_to_old_and_segment.len()
+        self.new_doc_id_to_old_doc_addr.len()
     }
+
+    pub(crate) fn get_old_doc_addr(&self, new_doc_id: DocId) -> DocAddress {
+        self.new_doc_id_to_old_doc_addr[new_doc_id as usize]
+    }
+
     /// This flags means the segments are simply stacked in the order of their ordinal.
     /// e.g. [(0, 1), .. (n, 1), (0, 2)..., (m, 2)]
     ///
@@ -39,21 +46,6 @@ impl SegmentDocIdMapping {
         self.is_trivial
     }
 }
-impl Index<usize> for SegmentDocIdMapping {
-    type Output = (DocId, SegmentOrdinal);
-
-    fn index(&self, idx: usize) -> &Self::Output {
-        &self.new_doc_id_to_old_and_segment[idx]
-    }
-}
-impl IntoIterator for SegmentDocIdMapping {
-    type Item = (DocId, SegmentOrdinal);
-    type IntoIter = std::vec::IntoIter<Self::Item>;
-
-    fn into_iter(self) -> Self::IntoIter {
-        self.new_doc_id_to_old_and_segment.into_iter()
-    }
-}
 
 /// Struct to provide mapping from old doc_id to new doc_id and vice versa within a segment.
 pub struct DocIdMapping {
@@ -151,8 +143,9 @@ pub(crate) fn get_doc_id_mapping_from_field(
 
 #[cfg(test)]
 mod tests_indexsorting {
+    use fastfield_codecs::Column;
+
     use crate::collector::TopDocs;
-    use crate::fastfield::FastFieldReader;
     use crate::indexer::doc_id_mapping::DocIdMapping;
     use crate::query::QueryParser;
     use crate::schema::{Schema, *};
@@ -472,9 +465,9 @@ mod tests_indexsorting {
         let my_number = index.schema().get_field("my_number").unwrap();
 
         let fast_field = fast_fields.u64(my_number).unwrap();
-        assert_eq!(fast_field.get(0u32), 10u64);
-        assert_eq!(fast_field.get(1u32), 20u64);
-        assert_eq!(fast_field.get(2u32), 30u64);
+        assert_eq!(fast_field.get_val(0), 10u64);
+        assert_eq!(fast_field.get_val(1), 20u64);
+        assert_eq!(fast_field.get_val(2), 30u64);
 
         let multi_numbers = index.schema().get_field("multi_numbers").unwrap();
         let multifield = fast_fields.u64s(multi_numbers).unwrap();

src/indexer/index_writer.rs

@@ -31,7 +31,7 @@ pub const MARGIN_IN_BYTES: usize = 1_000_000;
 pub const MEMORY_ARENA_NUM_BYTES_MIN: usize = ((MARGIN_IN_BYTES as u32) * 3u32) as usize;
 pub const MEMORY_ARENA_NUM_BYTES_MAX: usize = u32::MAX as usize - MARGIN_IN_BYTES;
 
-// We impose the number of index writter thread to be at most this.
+// We impose the number of index writer thread to be at most this.
 pub const MAX_NUM_THREAD: usize = 8;
 
 // Add document will block if the number of docs waiting in the queue to be indexed
@@ -710,7 +710,7 @@ impl IndexWriter {
     }
 
     /// Runs a group of document operations ensuring that the operations are
-    /// assigned contigous u64 opstamps and that add operations of the same
+    /// assigned contiguous u64 opstamps and that add operations of the same
     /// group are flushed into the same segment.
     ///
     /// If the indexing pipeline is full, this call may block.
@@ -777,6 +777,7 @@ impl Drop for IndexWriter {
 mod tests {
     use std::collections::{HashMap, HashSet};
 
+    use fastfield_codecs::Column;
     use proptest::prelude::*;
     use proptest::prop_oneof;
     use proptest::strategy::Strategy;
@@ -785,7 +786,6 @@ mod tests {
     use crate::collector::TopDocs;
     use crate::directory::error::LockError;
     use crate::error::*;
-    use crate::fastfield::FastFieldReader;
     use crate::indexer::NoMergePolicy;
     use crate::query::{QueryParser, TermQuery};
     use crate::schema::{
@@ -1327,7 +1327,7 @@ mod tests {
         let fast_field_reader = segment_reader.fast_fields().u64(id_field)?;
         let in_order_alive_ids: Vec<u64> = segment_reader
             .doc_ids_alive()
-            .map(|doc| fast_field_reader.get(doc))
+            .map(|doc| fast_field_reader.get_val(doc as u64))
             .collect();
         assert_eq!(&in_order_alive_ids[..], &[9, 8, 7, 6, 5, 4, 1, 0]);
         Ok(())
@@ -1493,7 +1493,7 @@ mod tests {
                 let ff_reader = segment_reader.fast_fields().u64(id_field).unwrap();
                 segment_reader
                     .doc_ids_alive()
-                    .map(move |doc| ff_reader.get(doc))
+                    .map(move |doc| ff_reader.get_val(doc as u64))
             })
             .collect();
 
@@ -1504,7 +1504,7 @@ mod tests {
                 let ff_reader = segment_reader.fast_fields().u64(id_field).unwrap();
                 segment_reader
                     .doc_ids_alive()
-                    .map(move |doc| ff_reader.get(doc))
+                    .map(move |doc| ff_reader.get_val(doc as u64))
             })
             .collect();
 
@@ -1622,7 +1622,7 @@ mod tests {
                 facet_reader
                     .facet_from_ord(facet_ords[0], &mut facet)
                     .unwrap();
-                let id = ff_reader.get(doc_id);
+                let id = ff_reader.get_val(doc_id as u64);
                 let facet_expected = Facet::from(&("/cola/".to_string() + &id.to_string()));
 
                 assert_eq!(facet, facet_expected);

src/indexer/json_term_writer.rs

@@ -38,10 +38,10 @@ use crate::{DatePrecision, DateTime, DocId, Term};
 /// of values, with a position gap. Here we would like `The` and `Who` to get indexed at
 /// position 2 and 3 respectively.
 ///
-/// With regular fields, we sort the fields beforehands, so that all terms with the same
+/// With regular fields, we sort the fields beforehand, so that all terms with the same
 /// path are indexed consecutively.
 ///
-/// In JSON object, we do not have this confort, so we need to record these position offsets in
+/// In JSON object, we do not have this comfort, so we need to record these position offsets in
 /// a map.
 ///
 /// Note that using a single position for the entire object would not hurt correctness.

src/indexer/merge_policy.rs

@@ -43,7 +43,7 @@ pub mod tests {
 
     /// `MergePolicy` useful for test purposes.
     ///
-    /// Everytime there is more than one segment,
+    /// Every time there is more than one segment,
     /// it will suggest to merge them.
     #[derive(Debug, Clone)]
     pub struct MergeWheneverPossible;

src/indexer/merger.rs

@@ -4,14 +4,13 @@ use std::sync::Arc;
 
 use itertools::Itertools;
 use measure_time::debug_time;
-use tantivy_bitpacker::minmax;
 
 use crate::core::{Segment, SegmentReader};
 use crate::docset::{DocSet, TERMINATED};
 use crate::error::DataCorruption;
 use crate::fastfield::{
-    AliveBitSet, CompositeFastFieldSerializer, DynamicFastFieldReader, FastFieldDataAccess,
-    FastFieldReader, FastFieldStats, MultiValueLength, MultiValuedFastFieldReader,
+    AliveBitSet, Column, CompositeFastFieldSerializer, DynamicFastFieldReader, FastFieldStats,
+    MultiValueLength, MultiValuedFastFieldReader,
 };
 use crate::fieldnorm::{FieldNormReader, FieldNormReaders, FieldNormsSerializer, FieldNormsWriter};
 use crate::indexer::doc_id_mapping::{expect_field_id_for_sort_field, SegmentDocIdMapping};
@@ -21,8 +20,8 @@ use crate::schema::{Cardinality, Field, FieldType, Schema};
 use crate::store::StoreWriter;
 use crate::termdict::{TermMerger, TermOrdinal};
 use crate::{
-    DocId, IndexSettings, IndexSortByField, InvertedIndexReader, Order, SegmentComponent,
-    SegmentOrdinal,
+    DocAddress, DocId, IndexSettings, IndexSortByField, InvertedIndexReader, Order,
+    SegmentComponent, SegmentOrdinal,
 };
 
 /// Segment's max doc must be `< MAX_DOC_LIMIT`.
@@ -88,7 +87,7 @@ pub struct IndexMerger {
 }
 
 fn compute_min_max_val(
-    u64_reader: &impl FastFieldReader<u64>,
+    u64_reader: &impl Column<u64>,
     segment_reader: &SegmentReader,
 ) -> Option<(u64, u64)> {
     if segment_reader.max_doc() == 0 {
@@ -102,11 +101,11 @@ fn compute_min_max_val(
     }
     // some deleted documents,
     // we need to recompute the max / min
-    minmax(
-        segment_reader
-            .doc_ids_alive()
-            .map(|doc_id| u64_reader.get(doc_id)),
-    )
+    segment_reader
+        .doc_ids_alive()
+        .map(|doc_id| u64_reader.get_val(doc_id as u64))
+        .minmax()
+        .into_option()
 }
 
 struct TermOrdinalMapping {
@@ -134,7 +133,7 @@ impl TermOrdinalMapping {
     fn max_term_ord(&self) -> TermOrdinal {
         self.per_segment_new_term_ordinals
             .iter()
-            .flat_map(|term_ordinals| term_ordinals.iter().cloned().max())
+            .flat_map(|term_ordinals| term_ordinals.iter().max())
             .max()
             .unwrap_or_default()
     }
@@ -260,9 +259,9 @@ impl IndexMerger {
                 .iter()
                 .map(|reader| reader.get_fieldnorms_reader(field))
                 .collect::<Result<_, _>>()?;
-            for (doc_id, reader_ordinal) in doc_id_mapping.iter() {
-                let fieldnorms_reader = &fieldnorms_readers[*reader_ordinal as usize];
-                let fieldnorm_id = fieldnorms_reader.fieldnorm_id(*doc_id);
+            for old_doc_addr in doc_id_mapping.iter_old_doc_addrs() {
+                let fieldnorms_reader = &fieldnorms_readers[old_doc_addr.segment_ord as usize];
+                let fieldnorm_id = fieldnorms_reader.fieldnorm_id(old_doc_addr.doc_id);
                 fieldnorms_data.push(fieldnorm_id);
             }
 
@@ -374,29 +373,46 @@ impl IndexMerger {
         struct SortedDocIdFieldAccessProvider<'a> {
             doc_id_mapping: &'a SegmentDocIdMapping,
             fast_field_readers: &'a Vec<DynamicFastFieldReader<u64>>,
+            stats: FastFieldStats,
         }
-        impl<'a> FastFieldDataAccess for SortedDocIdFieldAccessProvider<'a> {
+        impl<'a> Column for SortedDocIdFieldAccessProvider<'a> {
             fn get_val(&self, doc: u64) -> u64 {
-                let (doc_id, reader_ordinal) = self.doc_id_mapping[doc as usize];
-                self.fast_field_readers[reader_ordinal as usize].get(doc_id)
+                let DocAddress {
+                    doc_id,
+                    segment_ord,
+                } = self.doc_id_mapping.get_old_doc_addr(doc as u32);
+                self.fast_field_readers[segment_ord as usize].get_val(doc_id as u64)
+            }
+
+            fn iter(&self) -> Box<dyn Iterator<Item = u64> + '_> {
+                Box::new(
+                    self.doc_id_mapping
+                        .iter_old_doc_addrs()
+                        .map(|old_doc_addr| {
+                            let fast_field_reader =
+                                &self.fast_field_readers[old_doc_addr.segment_ord as usize];
+                            fast_field_reader.get_val(old_doc_addr.doc_id as u64)
+                        }),
+                )
+            }
+            fn min_value(&self) -> u64 {
+                self.stats.min_value
+            }
+
+            fn max_value(&self) -> u64 {
+                self.stats.max_value
+            }
+
+            fn num_vals(&self) -> u64 {
+                self.stats.num_vals
             }
         }
         let fastfield_accessor = SortedDocIdFieldAccessProvider {
             doc_id_mapping,
             fast_field_readers: &fast_field_readers,
-        };
-        let iter_gen = || {
-            doc_id_mapping.iter().map(|(doc_id, reader_ordinal)| {
-                let fast_field_reader = &fast_field_readers[*reader_ordinal as usize];
-                fast_field_reader.get(*doc_id)
-            })
-        };
-        fast_field_serializer.create_auto_detect_u64_fast_field(
-            field,
             stats,
-            fastfield_accessor,
-            iter_gen,
-        )?;
+        };
+        fast_field_serializer.create_auto_detect_u64_fast_field(field, fastfield_accessor)?;
 
         Ok(())
     }
@@ -412,7 +428,7 @@ impl IndexMerger {
 
         let everything_is_in_order = reader_ordinal_and_field_accessors
             .into_iter()
-            .map(|reader| reader.1)
+            .map(|(_, col)| Arc::new(col))
             .tuple_windows()
             .all(|(field_accessor1, field_accessor2)| {
                 if sort_by_field.order.is_asc() {
@@ -427,7 +443,7 @@ impl IndexMerger {
     pub(crate) fn get_sort_field_accessor(
         reader: &SegmentReader,
         sort_by_field: &IndexSortByField,
-    ) -> crate::Result<impl FastFieldReader<u64>> {
+    ) -> crate::Result<impl Column> {
         let field_id = expect_field_id_for_sort_field(reader.schema(), sort_by_field)?; // for now expect fastfield, but not strictly required
         let value_accessor = reader.fast_fields().u64_lenient(field_id)?;
         Ok(value_accessor)
@@ -436,7 +452,7 @@ impl IndexMerger {
     pub(crate) fn get_reader_with_sort_field_accessor(
         &self,
         sort_by_field: &IndexSortByField,
-    ) -> crate::Result<Vec<(SegmentOrdinal, impl FastFieldReader<u64> + Clone)>> {
+    ) -> crate::Result<Vec<(SegmentOrdinal, impl Column)>> {
         let reader_ordinal_and_field_accessors = self
             .readers
             .iter()
@@ -469,15 +485,11 @@ impl IndexMerger {
         let doc_id_reader_pair =
             reader_ordinal_and_field_accessors
                 .iter()
-                .map(|reader_and_field_accessor| {
-                    let reader = &self.readers[reader_and_field_accessor.0 as usize];
-                    reader.doc_ids_alive().map(move |doc_id| {
-                        (
-                            doc_id,
-                            reader_and_field_accessor.0,
-                            &reader_and_field_accessor.1,
-                        )
-                    })
+                .map(|(reader_ord, ff_reader)| {
+                    let reader = &self.readers[*reader_ord as usize];
+                    reader
+                        .doc_ids_alive()
+                        .map(move |doc_id| (doc_id, reader_ord, ff_reader))
                 });
 
         let total_num_new_docs = self
@@ -486,22 +498,25 @@ impl IndexMerger {
             .map(|reader| reader.num_docs() as usize)
             .sum();
 
-        let mut sorted_doc_ids = Vec::with_capacity(total_num_new_docs);
+        let mut sorted_doc_ids: Vec<DocAddress> = Vec::with_capacity(total_num_new_docs);
 
         // create iterator tuple of (old doc_id, reader) in order of the new doc_ids
         sorted_doc_ids.extend(
             doc_id_reader_pair
                 .into_iter()
                 .kmerge_by(|a, b| {
-                    let val1 = a.2.get(a.0);
-                    let val2 = b.2.get(b.0);
+                    let val1 = a.2.get_val(a.0 as u64);
+                    let val2 = b.2.get_val(b.0 as u64);
                     if sort_by_field.order == Order::Asc {
                         val1 < val2
                     } else {
                         val1 > val2
                     }
                 })
-                .map(|(doc_id, reader_with_id, _)| (doc_id, reader_with_id)),
+                .map(|(doc_id, &segment_ord, _)| DocAddress {
+                    doc_id,
+                    segment_ord,
+                }),
         );
         Ok(SegmentDocIdMapping::new(sorted_doc_ids, false))
     }
@@ -545,25 +560,49 @@ impl IndexMerger {
 
         // copying into a temp vec is not ideal, but the fast field codec api requires random
         // access, which is used in the estimation. It's possible to 1. calculate random
-        // acccess on the fly or 2. change the codec api to make random access optional, but
+        // access on the fly or 2. change the codec api to make random access optional, but
         // they both have also major drawbacks.
 
         let mut offsets = Vec::with_capacity(doc_id_mapping.len());
         let mut offset = 0;
-        for (doc_id, reader) in doc_id_mapping.iter() {
-            let reader = &reader_and_field_accessors[*reader as usize].1;
+        for old_doc_addr in doc_id_mapping.iter_old_doc_addrs() {
+            let reader = &reader_and_field_accessors[old_doc_addr.segment_ord as usize].1;
             offsets.push(offset);
-            offset += reader.get_len(*doc_id) as u64;
+            offset += reader.get_len(old_doc_addr.doc_id) as u64;
         }
         offsets.push(offset);
 
-        let iter_gen = || offsets.iter().cloned();
-        fast_field_serializer.create_auto_detect_u64_fast_field(
-            field,
+        #[derive(Clone)]
+        struct FieldIndexAccessProvider<'a> {
+            offsets: &'a [u64],
+            stats: FastFieldStats,
+        }
+        impl<'a> Column for FieldIndexAccessProvider<'a> {
+            fn get_val(&self, doc: u64) -> u64 {
+                self.offsets[doc as usize]
+            }
+
+            fn iter(&self) -> Box<dyn Iterator<Item = u64> + '_> {
+                Box::new(self.offsets.iter().cloned())
+            }
+            fn min_value(&self) -> u64 {
+                self.stats.min_value
+            }
+
+            fn max_value(&self) -> u64 {
+                self.stats.max_value
+            }
+
+            fn num_vals(&self) -> u64 {
+                self.stats.num_vals
+            }
+        }
+        let fastfield_accessor = FieldIndexAccessProvider {
+            offsets: &offsets,
             stats,
-            &offsets[..],
-            iter_gen,
-        )?;
+        };
+
+        fast_field_serializer.create_auto_detect_u64_fast_field(field, fastfield_accessor)?;
         Ok(offsets)
     }
     /// Returns the fastfield index (index for the data, not the data).
@@ -606,7 +645,7 @@ impl IndexMerger {
         debug_time!("write-term-id-fast-field");
 
         // Multifastfield consists of 2 fastfields.
-        // The first serves as an index into the second one and is stricly increasing.
+        // The first serves as an index into the second one and is strictly increasing.
         // The second contains the actual values.
 
         // First we merge the idx fast field.
@@ -631,12 +670,12 @@ impl IndexMerger {
                 fast_field_serializer.new_u64_fast_field_with_idx(field, 0u64, max_term_ord, 1)?;
             let mut vals = Vec::with_capacity(100);
 
-            for (old_doc_id, reader_ordinal) in doc_id_mapping.iter() {
+            for old_doc_addr in doc_id_mapping.iter_old_doc_addrs() {
                 let term_ordinal_mapping: &[TermOrdinal] =
-                    term_ordinal_mappings.get_segment(*reader_ordinal as usize);
+                    term_ordinal_mappings.get_segment(old_doc_addr.segment_ord as usize);
 
-                let ff_reader = &fast_field_reader[*reader_ordinal as usize];
-                ff_reader.get_vals(*old_doc_id, &mut vals);
+                let ff_reader = &fast_field_reader[old_doc_addr.segment_ord as usize];
+                ff_reader.get_vals(old_doc_addr.doc_id, &mut vals);
                 for &prev_term_ord in &vals {
                     let new_term_ord = term_ordinal_mapping[prev_term_ord as usize];
                     serialize_vals.add_val(new_term_ord)?;
@@ -657,16 +696,17 @@ impl IndexMerger {
             .map(|reader| reader.num_docs() as usize)
             .sum();
 
-        let mut mapping = Vec::with_capacity(total_num_new_docs);
+        let mut mapping: Vec<DocAddress> = Vec::with_capacity(total_num_new_docs);
 
         mapping.extend(
             self.readers
                 .iter()
                 .enumerate()
-                .flat_map(|(reader_ordinal, reader)| {
-                    reader
-                        .doc_ids_alive()
-                        .map(move |doc_id| (doc_id, reader_ordinal as SegmentOrdinal))
+                .flat_map(|(segment_ord, reader)| {
+                    reader.doc_ids_alive().map(move |doc_id| DocAddress {
+                        segment_ord: segment_ord as u32,
+                        doc_id,
+                    })
                 }),
         );
         Ok(SegmentDocIdMapping::new(mapping, true))
@@ -678,7 +718,7 @@ impl IndexMerger {
         doc_id_mapping: &SegmentDocIdMapping,
     ) -> crate::Result<()> {
         // Multifastfield consists in 2 fastfields.
-        // The first serves as an index into the second one and is stricly increasing.
+        // The first serves as an index into the second one and is strictly increasing.
         // The second contains the actual values.
 
         // First we merge the idx fast field.
@@ -735,49 +775,68 @@ impl IndexMerger {
             doc_id_mapping: &'a SegmentDocIdMapping,
             fast_field_readers: &'a Vec<MultiValuedFastFieldReader<u64>>,
             offsets: Vec<u64>,
+            stats: FastFieldStats,
         }
-        impl<'a> FastFieldDataAccess for SortedDocIdMultiValueAccessProvider<'a> {
+        impl<'a> Column for SortedDocIdMultiValueAccessProvider<'a> {
             fn get_val(&self, pos: u64) -> u64 {
                 // use the offsets index to find the doc_id which will contain the position.
-                // the offsets are stricly increasing so we can do a simple search on it.
-                let new_doc_id = self
-                    .offsets
-                    .iter()
-                    .position(|&offset| offset > pos)
-                    .expect("pos is out of bounds")
-                    - 1;
+                // the offsets are strictly increasing so we can do a simple search on it.
+                let new_doc_id: DocId =
+                    self.offsets
+                        .iter()
+                        .position(|offset| offset > pos)
+                        .expect("pos is out of bounds") as DocId
+                        - 1u32;
 
                 // now we need to find the position of `pos` in the multivalued bucket
-                let num_pos_covered_until_now = self.offsets[new_doc_id];
+                let num_pos_covered_until_now = self.offsets[new_doc_id as usize];
                 let pos_in_values = pos - num_pos_covered_until_now;
 
-                let (old_doc_id, reader_ordinal) = self.doc_id_mapping[new_doc_id as usize];
-                let num_vals = self.fast_field_readers[reader_ordinal as usize].get_len(old_doc_id);
+                let old_doc_addr = self.doc_id_mapping.get_old_doc_addr(new_doc_id);
+                let num_vals = self.fast_field_readers[old_doc_addr.segment_ord as usize]
+                    .get_len(old_doc_addr.doc_id);
                 assert!(num_vals >= pos_in_values);
-                let mut vals = vec![];
-                self.fast_field_readers[reader_ordinal as usize].get_vals(old_doc_id, &mut vals);
+                let mut vals = Vec::new();
+                self.fast_field_readers[old_doc_addr.segment_ord as usize]
+                    .get_vals(old_doc_addr.doc_id, &mut vals);
 
                 vals[pos_in_values as usize]
             }
+
+            fn iter(&self) -> Box<dyn Iterator<Item = u64> + '_> {
+                Box::new(
+                    self.doc_id_mapping
+                        .iter_old_doc_addrs()
+                        .flat_map(|old_doc_addr| {
+                            let ff_reader =
+                                &self.fast_field_readers[old_doc_addr.segment_ord as usize];
+                            let mut vals = Vec::new();
+                            ff_reader.get_vals(old_doc_addr.doc_id, &mut vals);
+                            vals.into_iter()
+                        }),
+                )
+            }
+            fn min_value(&self) -> u64 {
+                self.stats.min_value
+            }
+
+            fn max_value(&self) -> u64 {
+                self.stats.max_value
+            }
+
+            fn num_vals(&self) -> u64 {
+                self.stats.num_vals
+            }
         }
         let fastfield_accessor = SortedDocIdMultiValueAccessProvider {
             doc_id_mapping,
             fast_field_readers: &ff_readers,
             offsets,
-        };
-        let iter_gen = || {
-            doc_id_mapping.iter().flat_map(|(doc_id, reader_ordinal)| {
-                let ff_reader = &ff_readers[*reader_ordinal as usize];
-                let mut vals = vec![];
-                ff_reader.get_vals(*doc_id, &mut vals);
-                vals.into_iter()
-            })
+            stats,
         };
         fast_field_serializer.create_auto_detect_u64_fast_field_with_idx(
             field,
-            stats,
             fastfield_accessor,
-            iter_gen,
             1,
         )?;
 
@@ -810,9 +869,9 @@ impl IndexMerger {
         )?;
         let mut serialize_vals = fast_field_serializer.new_bytes_fast_field_with_idx(field, 1);
 
-        for (doc_id, reader_ordinal) in doc_id_mapping.iter() {
-            let bytes_reader = &reader_and_field_accessors[*reader_ordinal as usize].1;
-            let val = bytes_reader.get_bytes(*doc_id);
+        for old_doc_addr in doc_id_mapping.iter_old_doc_addrs() {
+            let bytes_reader = &reader_and_field_accessors[old_doc_addr.segment_ord as usize].1;
+            let val = bytes_reader.get_bytes(old_doc_addr.doc_id);
             serialize_vals.write_all(val)?;
         }
 
@@ -868,9 +927,9 @@ impl IndexMerger {
                 segment_local_map
             })
             .collect();
-        for (new_doc_id, (old_doc_id, segment_ord)) in doc_id_mapping.iter().enumerate() {
-            let segment_map = &mut merged_doc_id_map[*segment_ord as usize];
-            segment_map[*old_doc_id as usize] = Some(new_doc_id as DocId);
+        for (new_doc_id, old_doc_addr) in doc_id_mapping.iter_old_doc_addrs().enumerate() {
+            let segment_map = &mut merged_doc_id_map[old_doc_addr.segment_ord as usize];
+            segment_map[old_doc_addr.doc_id as usize] = Some(new_doc_id as DocId);
         }
 
         // Note that the total number of tokens is not exact.
@@ -1045,15 +1104,15 @@ impl IndexMerger {
                 .map(|(i, store)| store.iter_raw(self.readers[i].alive_bitset()))
                 .collect();
 
-            for (old_doc_id, reader_ordinal) in doc_id_mapping.iter() {
-                let doc_bytes_it = &mut document_iterators[*reader_ordinal as usize];
+            for old_doc_addr in doc_id_mapping.iter_old_doc_addrs() {
+                let doc_bytes_it = &mut document_iterators[old_doc_addr.segment_ord as usize];
                 if let Some(doc_bytes_res) = doc_bytes_it.next() {
                     let doc_bytes = doc_bytes_res?;
                     store_writer.store_bytes(&doc_bytes)?;
                 } else {
                     return Err(DataCorruption::comment_only(&format!(
-                        "unexpected missing document in docstore on merge, doc id {:?}",
-                        old_doc_id
+                        "unexpected missing document in docstore on merge, doc address \
+                         {old_doc_addr:?}",
                     ))
                     .into());
                 }
@@ -1140,6 +1199,7 @@ impl IndexMerger {
 #[cfg(test)]
 mod tests {
     use byteorder::{BigEndian, ReadBytesExt};
+    use fastfield_codecs::Column;
     use schema::FAST;
 
     use crate::collector::tests::{
@@ -1147,7 +1207,6 @@ mod tests {
     };
     use crate::collector::{Count, FacetCollector};
     use crate::core::Index;
-    use crate::fastfield::FastFieldReader;
     use crate::query::{AllQuery, BooleanQuery, Scorer, TermQuery};
     use crate::schema::{
         Cardinality, Document, Facet, FacetOptions, IndexRecordOption, NumericOptions, Term,
@@ -2078,7 +2137,7 @@ mod tests {
             let mut term_scorer = term_query
                 .specialized_weight(&searcher, true)?
                 .specialized_scorer(segment_reader, 1.0)?;
-            // the difference compared to before is instrinsic to the bm25 formula. no worries
+            // the difference compared to before is intrinsic to the bm25 formula. no worries
             // there.
             for doc in segment_reader.doc_ids_alive() {
                 assert_eq!(term_scorer.doc(), doc);
@@ -2103,7 +2162,7 @@ mod tests {
         let mut term_scorer = term_query
             .specialized_weight(&searcher, true)?
             .specialized_scorer(segment_reader, 1.0)?;
-        // the difference compared to before is instrinsic to the bm25 formula. no worries there.
+        // the difference compared to before is intrinsic to the bm25 formula. no worries there.
         for doc in segment_reader.doc_ids_alive() {
             assert_eq!(term_scorer.doc(), doc);
             assert_nearly_equals!(term_scorer.block_max_score(), 0.003478312);

src/indexer/merger_sorted_index_test.rs

@@ -1,8 +1,10 @@
 #[cfg(test)]
 mod tests {
+    use fastfield_codecs::Column;
+
     use crate::collector::TopDocs;
     use crate::core::Index;
-    use crate::fastfield::{AliveBitSet, FastFieldReader, MultiValuedFastFieldReader};
+    use crate::fastfield::{AliveBitSet, MultiValuedFastFieldReader};
     use crate::query::QueryParser;
     use crate::schema::{
         self, BytesOptions, Cardinality, Facet, FacetOptions, IndexRecordOption, NumericOptions,
@@ -186,17 +188,17 @@ mod tests {
 
         let fast_fields = segment_reader.fast_fields();
         let fast_field = fast_fields.u64(int_field).unwrap();
-        assert_eq!(fast_field.get(5u32), 1u64);
-        assert_eq!(fast_field.get(4u32), 2u64);
-        assert_eq!(fast_field.get(3u32), 3u64);
+        assert_eq!(fast_field.get_val(5), 1u64);
+        assert_eq!(fast_field.get_val(4), 2u64);
+        assert_eq!(fast_field.get_val(3), 3u64);
         if force_disjunct_segment_sort_values {
-            assert_eq!(fast_field.get(2u32), 20u64);
-            assert_eq!(fast_field.get(1u32), 100u64);
+            assert_eq!(fast_field.get_val(2u64), 20u64);
+            assert_eq!(fast_field.get_val(1u64), 100u64);
         } else {
-            assert_eq!(fast_field.get(2u32), 10u64);
-            assert_eq!(fast_field.get(1u32), 20u64);
+            assert_eq!(fast_field.get_val(2u64), 10u64);
+            assert_eq!(fast_field.get_val(1u64), 20u64);
         }
-        assert_eq!(fast_field.get(0u32), 1_000u64);
+        assert_eq!(fast_field.get_val(0u64), 1_000u64);
 
         // test new field norm mapping
         {
@@ -373,12 +375,12 @@ mod tests {
 
         let fast_fields = segment_reader.fast_fields();
         let fast_field = fast_fields.u64(int_field).unwrap();
-        assert_eq!(fast_field.get(0u32), 1u64);
-        assert_eq!(fast_field.get(1u32), 2u64);
-        assert_eq!(fast_field.get(2u32), 3u64);
-        assert_eq!(fast_field.get(3u32), 10u64);
-        assert_eq!(fast_field.get(4u32), 20u64);
-        assert_eq!(fast_field.get(5u32), 1_000u64);
+        assert_eq!(fast_field.get_val(0), 1u64);
+        assert_eq!(fast_field.get_val(1), 2u64);
+        assert_eq!(fast_field.get_val(2), 3u64);
+        assert_eq!(fast_field.get_val(3), 10u64);
+        assert_eq!(fast_field.get_val(4), 20u64);
+        assert_eq!(fast_field.get_val(5), 1_000u64);
 
         let get_vals = |fast_field: &MultiValuedFastFieldReader<u64>, doc_id: u32| -> Vec<u64> {
             let mut vals = vec![];
@@ -484,7 +486,7 @@ mod bench_sorted_index_merge {
     // use cratedoc_id, readerdoc_id_mappinglet vals = reader.fate::schema;
     use crate::fastfield::{DynamicFastFieldReader, FastFieldReader};
     use crate::indexer::merger::IndexMerger;
-    use crate::schema::{Cardinality, Document, NumericOptions, Schema};
+    use crate::schema::{Cardinality, NumericOptions, Schema};
     use crate::{IndexSettings, IndexSortByField, IndexWriter, Order};
     fn create_index(sort_by_field: Option<IndexSortByField>) -> Index {
         let mut schema_builder = Schema::builder();
@@ -503,9 +505,7 @@ mod bench_sorted_index_merge {
         {
             let mut index_writer = index.writer_for_tests().unwrap();
             let index_doc = |index_writer: &mut IndexWriter, val: u64| {
-                let mut doc = Document::default();
-                doc.add_u64(int_field, val);
-                index_writer.add_document(doc).unwrap();
+                index_writer.add_document(doc!(int_field=>val)).unwrap();
             };
             // 3 segments with 10_000 values in the fast fields
             for _ in 0..3 {
@@ -518,6 +518,7 @@ mod bench_sorted_index_merge {
         }
         index
     }
+
     #[bench]
     fn create_sorted_index_walk_overkmerge_on_merge_fastfield(
         b: &mut Bencher,
@@ -533,19 +534,19 @@ mod bench_sorted_index_merge {
             IndexMerger::open(index.schema(), index.settings().clone(), &segments[..])?;
         let doc_id_mapping = merger.generate_doc_id_mapping(&sort_by_field).unwrap();
         b.iter(|| {
-            let sorted_doc_ids = doc_id_mapping.iter().map(|(doc_id, ordinal)| {
-                let reader = &merger.readers[*ordinal as usize];
+            let sorted_doc_ids = doc_id_mapping.iter_old_doc_addrs().map(|doc_addr| {
+                let reader = &merger.readers[doc_addr.segment_ord as usize];
                 let u64_reader: DynamicFastFieldReader<u64> =
                     reader.fast_fields().typed_fast_field_reader(field).expect(
                         "Failed to find a reader for single fast field. This is a tantivy bug and \
                          it should never happen.",
                     );
-                (doc_id, reader, u64_reader)
+                (doc_addr.doc_id, reader, u64_reader)
             });
             // add values in order of the new doc_ids
             let mut val = 0;
             for (doc_id, _reader, field_reader) in sorted_doc_ids {
-                val = field_reader.get(*doc_id);
+                val = field_reader.get(doc_id);
             }
 
             val

src/indexer/segment_manager.rs

@@ -21,7 +21,7 @@ pub(crate) enum SegmentsStatus {
 }
 
 impl SegmentRegisters {
-    /// Check if all the segments are committed or uncommited.
+    /// Check if all the segments are committed or uncommitted.
     ///
     /// If some segment is missing or segments are in a different state (this should not happen
     /// if tantivy is used correctly), returns `None`.
@@ -168,8 +168,8 @@ impl SegmentManager {
                 segment_entries.push(segment_entry);
             }
         } else {
-            let error_msg = "Merge operation sent for segments that are not all uncommited or \
-                             commited."
+            let error_msg = "Merge operation sent for segments that are not all uncommitted or \
+                             committed."
                 .to_string();
             return Err(TantivyError::InvalidArgument(error_msg));
         }
@@ -182,7 +182,7 @@ impl SegmentManager {
     }
     // Replace a list of segments for their equivalent merged segment.
     //
-    // Returns true if these segments are committed, false if the merge segments are uncommited.
+    // Returns true if these segments are committed, false if the merge segments are uncommitted.
     pub(crate) fn end_merge(
         &self,
         before_merge_segment_ids: &[SegmentId],

src/indexer/segment_updater.rs

@@ -171,7 +171,7 @@ pub fn merge_indices<T: Into<Box<dyn Directory>>>(
     if indices.is_empty() {
         // If there are no indices to merge, there is no need to do anything.
         return Err(crate::TantivyError::InvalidArgument(
-            "No indices given to marge".to_string(),
+            "No indices given to merge".to_string(),
         ));
     }
 
@@ -219,7 +219,7 @@ pub fn merge_filtered_segments<T: Into<Box<dyn Directory>>>(
     if segments.is_empty() {
         // If there are no indices to merge, there is no need to do anything.
         return Err(crate::TantivyError::InvalidArgument(
-            "No segments given to marge".to_string(),
+            "No segments given to merge".to_string(),
         ));
     }
 
@@ -282,7 +282,7 @@ pub fn merge_filtered_segments<T: Into<Box<dyn Directory>>>(
 
 pub(crate) struct InnerSegmentUpdater {
     // we keep a copy of the current active IndexMeta to
-    // avoid loading the file everytime we need it in the
+    // avoid loading the file every time we need it in the
     // `SegmentUpdater`.
     //
     // This should be up to date as all update happen through
@@ -500,7 +500,7 @@ impl SegmentUpdater {
     // It returns an error if for some reason the merge operation could not be started.
     //
     // At this point an error is not necessarily the sign of a malfunction.
-    // (e.g. A rollback could have happened, between the instant when the merge operaiton was
+    // (e.g. A rollback could have happened, between the instant when the merge operation was
     // suggested and the moment when it ended up being executed.)
     //
     // `segment_ids` is required to be non-empty.

src/indexer/segment_writer.rs

@@ -53,7 +53,7 @@ fn remap_doc_opstamps(
 /// set of documents.
 ///
 /// They creates the postings list in anonymous memory.
-/// The segment is layed on disk when the segment gets `finalized`.
+/// The segment is laid on disk when the segment gets `finalized`.
 pub struct SegmentWriter {
     pub(crate) max_doc: DocId,
     pub(crate) ctx: IndexingContext,

src/lib.rs

@@ -11,6 +11,7 @@
 #![doc(test(attr(allow(unused_variables), deny(warnings))))]
 #![warn(missing_docs)]
 #![allow(clippy::len_without_is_empty)]
+#![allow(clippy::derive_partial_eq_without_eq)]
 
 //! # `tantivy`
 //!
@@ -420,6 +421,7 @@ pub struct DocAddress {
 #[cfg(test)]
 pub mod tests {
     use common::{BinarySerializable, FixedSize};
+    use fastfield_codecs::Column;
     use rand::distributions::{Bernoulli, Uniform};
     use rand::rngs::StdRng;
     use rand::{Rng, SeedableRng};
@@ -428,7 +430,6 @@ pub mod tests {
     use crate::collector::tests::TEST_COLLECTOR_WITH_SCORE;
     use crate::core::SegmentReader;
     use crate::docset::{DocSet, TERMINATED};
-    use crate::fastfield::FastFieldReader;
     use crate::merge_policy::NoMergePolicy;
     use crate::query::BooleanQuery;
     use crate::schema::*;
@@ -1035,21 +1036,21 @@ pub mod tests {
             let fast_field_reader_opt = segment_reader.fast_fields().u64(fast_field_unsigned);
             assert!(fast_field_reader_opt.is_ok());
             let fast_field_reader = fast_field_reader_opt.unwrap();
-            assert_eq!(fast_field_reader.get(0), 4u64)
+            assert_eq!(fast_field_reader.get_val(0), 4u64)
         }
 
         {
             let fast_field_reader_res = segment_reader.fast_fields().i64(fast_field_signed);
             assert!(fast_field_reader_res.is_ok());
             let fast_field_reader = fast_field_reader_res.unwrap();
-            assert_eq!(fast_field_reader.get(0), 4i64)
+            assert_eq!(fast_field_reader.get_val(0), 4i64)
         }
 
         {
             let fast_field_reader_res = segment_reader.fast_fields().f64(fast_field_float);
             assert!(fast_field_reader_res.is_ok());
             let fast_field_reader = fast_field_reader_res.unwrap();
-            assert_eq!(fast_field_reader.get(0), 4f64)
+            assert_eq!(fast_field_reader.get_val(0), 4f64)
         }
         Ok(())
     }

src/postings/block_segment_postings.rs

@@ -199,7 +199,7 @@ impl BlockSegmentPostings {
         self.doc_decoder.output_array()
     }
 
-    /// Returns a full block, regardless of whetehr the block is complete or incomplete (
+    /// Returns a full block, regardless of whether the block is complete or incomplete (
     /// as it happens for the last block of the posting list).
     ///
     /// In the latter case, the block is guaranteed to be padded with the sentinel value:
@@ -494,7 +494,7 @@ mod tests {
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
         let mut index_writer = index.writer_for_tests()?;
-        // create two postings list, one containg even number,
+        // create two postings list, one containing even number,
         // the other containing odd numbers.
         for i in 0..6 {
             let doc = doc!(int_field=> (i % 2) as u64);

src/postings/mod.rs

@@ -500,7 +500,7 @@ pub mod tests {
         Ok(())
     }
 
-    /// Wraps a given docset, and forward alls call but the
+    /// Wraps a given docset, and forward all call but the
     /// `.skip_next(...)`. This is useful to test that a specialized
     /// implementation of `.skip_next(...)` is consistent
     /// with the default implementation.

src/postings/postings.rs

@@ -14,7 +14,7 @@ pub trait Postings: DocSet + 'static {
     /// The number of times the term appears in the document.
     fn term_freq(&self) -> u32;
 
-    /// Returns the positions offseted with a given value.
+    /// Returns the positions offsetted with a given value.
     /// The output vector will be resized to the `term_freq`.
     fn positions_with_offset(&mut self, offset: u32, output: &mut Vec<u32>);
 

src/postings/stacker/expull.rs

@@ -40,7 +40,7 @@ fn len_to_capacity(len: u32) -> CapacityResult {
 /// An exponential unrolled link.
 ///
 /// The use case is as follows. Tantivy's indexer conceptually acts like a
-/// `HashMap<Term, Vec<u32>>`. As we come accross a given term in document
+/// `HashMap<Term, Vec<u32>>`. As we come across a given term in document
 /// `D`, we lookup the term in the map and append the document id to its vector.
 ///
 /// The vector is then only read when it is serialized.

src/query/boolean_query/block_wand.rs

@@ -371,7 +371,7 @@ mod tests {
     fn compute_checkpoints_manual(term_scorers: Vec<TermScorer>, n: usize) -> Vec<(DocId, Score)> {
         let mut heap: BinaryHeap<Float> = BinaryHeap::with_capacity(n);
         let mut checkpoints: Vec<(DocId, Score)> = Vec::new();
-        let mut scorer: Union<TermScorer, SumCombiner> = Union::from(term_scorers);
+        let mut scorer = Union::build(term_scorers, SumCombiner::default);
 
         let mut limit = Score::MIN;
         loop {

src/query/boolean_query/boolean_query.rs

@@ -1,7 +1,5 @@
-use std::collections::BTreeMap;
-
 use super::boolean_weight::BooleanWeight;
-use crate::query::{Occur, Query, TermQuery, Weight};
+use crate::query::{Occur, Query, SumWithCoordsCombiner, TermQuery, Weight};
 use crate::schema::{IndexRecordOption, Term};
 use crate::Searcher;
 
@@ -153,12 +151,16 @@ impl Query for BooleanQuery {
                 Ok((*occur, subquery.weight(searcher, scoring_enabled)?))
             })
             .collect::<crate::Result<_>>()?;
-        Ok(Box::new(BooleanWeight::new(sub_weights, scoring_enabled)))
+        Ok(Box::new(BooleanWeight::new(
+            sub_weights,
+            scoring_enabled,
+            Box::new(SumWithCoordsCombiner::default),
+        )))
     }
 
-    fn query_terms(&self, terms: &mut BTreeMap<Term, bool>) {
+    fn query_terms<'a>(&'a self, visitor: &mut dyn FnMut(&'a Term, bool)) {
         for (_occur, subquery) in &self.subqueries {
-            subquery.query_terms(terms);
+            subquery.query_terms(visitor);
         }
     }
 }

src/query/boolean_query/boolean_weight.rs

@@ -3,7 +3,7 @@ use std::collections::HashMap;
 use crate::core::SegmentReader;
 use crate::postings::FreqReadingOption;
 use crate::query::explanation::does_not_match;
-use crate::query::score_combiner::{DoNothingCombiner, ScoreCombiner, SumWithCoordsCombiner};
+use crate::query::score_combiner::{DoNothingCombiner, ScoreCombiner};
 use crate::query::term_query::TermScorer;
 use crate::query::weight::{for_each_pruning_scorer, for_each_scorer};
 use crate::query::{
@@ -17,11 +17,16 @@ enum SpecializedScorer {
     Other(Box<dyn Scorer>),
 }
 
-fn scorer_union<TScoreCombiner>(scorers: Vec<Box<dyn Scorer>>) -> SpecializedScorer
-where TScoreCombiner: ScoreCombiner {
+fn scorer_union<TScoreCombiner>(
+    scorers: Vec<Box<dyn Scorer>>,
+    score_combiner_fn: impl Fn() -> TScoreCombiner,
+) -> SpecializedScorer
+where
+    TScoreCombiner: ScoreCombiner,
+{
     assert!(!scorers.is_empty());
     if scorers.len() == 1 {
-        return SpecializedScorer::Other(scorers.into_iter().next().unwrap()); //< we checked the size beforehands
+        return SpecializedScorer::Other(scorers.into_iter().next().unwrap()); //< we checked the size beforehand
     }
 
     {
@@ -38,35 +43,45 @@ where TScoreCombiner: ScoreCombiner {
                 // Block wand is only available if we read frequencies.
                 return SpecializedScorer::TermUnion(scorers);
             } else {
-                return SpecializedScorer::Other(Box::new(Union::<_, TScoreCombiner>::from(
+                return SpecializedScorer::Other(Box::new(Union::build(
                     scorers,
+                    score_combiner_fn,
                 )));
             }
         }
     }
-    SpecializedScorer::Other(Box::new(Union::<_, TScoreCombiner>::from(scorers)))
+    SpecializedScorer::Other(Box::new(Union::build(scorers, score_combiner_fn)))
 }
 
-fn into_box_scorer<TScoreCombiner: ScoreCombiner>(scorer: SpecializedScorer) -> Box<dyn Scorer> {
+fn into_box_scorer<TScoreCombiner: ScoreCombiner>(
+    scorer: SpecializedScorer,
+    score_combiner_fn: impl Fn() -> TScoreCombiner,
+) -> Box<dyn Scorer> {
     match scorer {
         SpecializedScorer::TermUnion(term_scorers) => {
-            let union_scorer = Union::<TermScorer, TScoreCombiner>::from(term_scorers);
+            let union_scorer = Union::build(term_scorers, score_combiner_fn);
             Box::new(union_scorer)
         }
         SpecializedScorer::Other(scorer) => scorer,
     }
 }
 
-pub struct BooleanWeight {
+pub struct BooleanWeight<TScoreCombiner: ScoreCombiner> {
     weights: Vec<(Occur, Box<dyn Weight>)>,
     scoring_enabled: bool,
+    score_combiner_fn: Box<dyn Fn() -> TScoreCombiner + Sync + Send>,
 }
 
-impl BooleanWeight {
-    pub fn new(weights: Vec<(Occur, Box<dyn Weight>)>, scoring_enabled: bool) -> BooleanWeight {
+impl<TScoreCombiner: ScoreCombiner> BooleanWeight<TScoreCombiner> {
+    pub fn new(
+        weights: Vec<(Occur, Box<dyn Weight>)>,
+        scoring_enabled: bool,
+        score_combiner_fn: Box<dyn Fn() -> TScoreCombiner + Sync + Send + 'static>,
+    ) -> BooleanWeight<TScoreCombiner> {
         BooleanWeight {
             weights,
             scoring_enabled,
+            score_combiner_fn,
         }
     }
 
@@ -86,21 +101,23 @@ impl BooleanWeight {
         Ok(per_occur_scorers)
     }
 
-    fn complex_scorer<TScoreCombiner: ScoreCombiner>(
+    fn complex_scorer<TComplexScoreCombiner: ScoreCombiner>(
         &self,
         reader: &SegmentReader,
         boost: Score,
+        score_combiner_fn: impl Fn() -> TComplexScoreCombiner,
     ) -> crate::Result<SpecializedScorer> {
         let mut per_occur_scorers = self.per_occur_scorers(reader, boost)?;
 
         let should_scorer_opt: Option<SpecializedScorer> = per_occur_scorers
             .remove(&Occur::Should)
-            .map(scorer_union::<TScoreCombiner>);
-
+            .map(|scorers| scorer_union(scorers, &score_combiner_fn));
         let exclude_scorer_opt: Option<Box<dyn Scorer>> = per_occur_scorers
             .remove(&Occur::MustNot)
-            .map(scorer_union::<DoNothingCombiner>)
-            .map(into_box_scorer::<DoNothingCombiner>);
+            .map(|scorers| scorer_union(scorers, DoNothingCombiner::default))
+            .map(|specialized_scorer| {
+                into_box_scorer(specialized_scorer, DoNothingCombiner::default)
+            });
 
         let must_scorer_opt: Option<Box<dyn Scorer>> = per_occur_scorers
             .remove(&Occur::Must)
@@ -112,10 +129,10 @@ impl BooleanWeight {
                     SpecializedScorer::Other(Box::new(RequiredOptionalScorer::<
                         Box<dyn Scorer>,
                         Box<dyn Scorer>,
-                        TScoreCombiner,
+                        TComplexScoreCombiner,
                     >::new(
                         must_scorer,
-                        into_box_scorer::<TScoreCombiner>(should_scorer),
+                        into_box_scorer(should_scorer, &score_combiner_fn),
                     )))
                 } else {
                     SpecializedScorer::Other(must_scorer)
@@ -129,8 +146,7 @@ impl BooleanWeight {
         };
 
         if let Some(exclude_scorer) = exclude_scorer_opt {
-            let positive_scorer_boxed: Box<dyn Scorer> =
-                into_box_scorer::<TScoreCombiner>(positive_scorer);
+            let positive_scorer_boxed = into_box_scorer(positive_scorer, &score_combiner_fn);
             Ok(SpecializedScorer::Other(Box::new(Exclude::new(
                 positive_scorer_boxed,
                 exclude_scorer,
@@ -141,7 +157,7 @@ impl BooleanWeight {
     }
 }
 
-impl Weight for BooleanWeight {
+impl<TScoreCombiner: ScoreCombiner + Sync> Weight for BooleanWeight<TScoreCombiner> {
     fn scorer(&self, reader: &SegmentReader, boost: Score) -> crate::Result<Box<dyn Scorer>> {
         if self.weights.is_empty() {
             Ok(Box::new(EmptyScorer))
@@ -153,13 +169,15 @@ impl Weight for BooleanWeight {
                 weight.scorer(reader, boost)
             }
         } else if self.scoring_enabled {
-            self.complex_scorer::<SumWithCoordsCombiner>(reader, boost)
+            self.complex_scorer(reader, boost, &self.score_combiner_fn)
                 .map(|specialized_scorer| {
-                    into_box_scorer::<SumWithCoordsCombiner>(specialized_scorer)
+                    into_box_scorer(specialized_scorer, &self.score_combiner_fn)
                 })
         } else {
-            self.complex_scorer::<DoNothingCombiner>(reader, boost)
-                .map(into_box_scorer::<DoNothingCombiner>)
+            self.complex_scorer(reader, boost, &DoNothingCombiner::default)
+                .map(|specialized_scorer| {
+                    into_box_scorer(specialized_scorer, &DoNothingCombiner::default)
+                })
         }
     }
 
@@ -188,11 +206,10 @@ impl Weight for BooleanWeight {
         reader: &SegmentReader,
         callback: &mut dyn FnMut(DocId, Score),
     ) -> crate::Result<()> {
-        let scorer = self.complex_scorer::<SumWithCoordsCombiner>(reader, 1.0)?;
+        let scorer = self.complex_scorer(reader, 1.0, &self.score_combiner_fn)?;
         match scorer {
             SpecializedScorer::TermUnion(term_scorers) => {
-                let mut union_scorer =
-                    Union::<TermScorer, SumWithCoordsCombiner>::from(term_scorers);
+                let mut union_scorer = Union::build(term_scorers, &self.score_combiner_fn);
                 for_each_scorer(&mut union_scorer, callback);
             }
             SpecializedScorer::Other(mut scorer) => {
@@ -218,7 +235,7 @@ impl Weight for BooleanWeight {
         reader: &SegmentReader,
         callback: &mut dyn FnMut(DocId, Score) -> Score,
     ) -> crate::Result<()> {
-        let scorer = self.complex_scorer::<SumWithCoordsCombiner>(reader, 1.0)?;
+        let scorer = self.complex_scorer(reader, 1.0, &self.score_combiner_fn)?;
         match scorer {
             SpecializedScorer::TermUnion(term_scorers) => {
                 super::block_wand(term_scorers, threshold, callback);

src/query/boolean_query/mod.rs

@@ -4,6 +4,7 @@ mod boolean_weight;
 
 pub(crate) use self::block_wand::{block_wand, block_wand_single_scorer};
 pub use self::boolean_query::BooleanQuery;
+pub(crate) use self::boolean_weight::BooleanWeight;
 
 #[cfg(test)]
 mod tests {

src/query/boost_query.rs

@@ -1,4 +1,3 @@
-use std::collections::BTreeMap;
 use std::fmt;
 
 use crate::fastfield::AliveBitSet;
@@ -49,8 +48,8 @@ impl Query for BoostQuery {
         Ok(boosted_weight)
     }
 
-    fn query_terms(&self, terms: &mut BTreeMap<Term, bool>) {
-        self.query.query_terms(terms)
+    fn query_terms<'a>(&'a self, visitor: &mut dyn FnMut(&'a Term, bool)) {
+        self.query.query_terms(visitor)
     }
 }
 

src/query/const_score_query.rs

@@ -0,0 +1,176 @@
+use std::fmt;
+
+use crate::query::{Explanation, Query, Scorer, Weight};
+use crate::{DocId, DocSet, Score, Searcher, SegmentReader, TantivyError, Term};
+
+/// `ConstScoreQuery` is a wrapper over a query to provide a constant score.
+/// It can avoid unnecessary score computation on the wrapped query.
+///
+/// The document set matched by the `ConstScoreQuery` is strictly the same as the underlying query.
+/// The configured score is used for each document.
+pub struct ConstScoreQuery {
+    query: Box<dyn Query>,
+    score: Score,
+}
+
+impl ConstScoreQuery {
+    /// Builds a const score query.
+    pub fn new(query: Box<dyn Query>, score: Score) -> ConstScoreQuery {
+        ConstScoreQuery { query, score }
+    }
+}
+
+impl Clone for ConstScoreQuery {
+    fn clone(&self) -> Self {
+        ConstScoreQuery {
+            query: self.query.box_clone(),
+            score: self.score,
+        }
+    }
+}
+
+impl fmt::Debug for ConstScoreQuery {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        write!(f, "Const(score={}, query={:?})", self.score, self.query)
+    }
+}
+
+impl Query for ConstScoreQuery {
+    fn weight(&self, searcher: &Searcher, scoring_enabled: bool) -> crate::Result<Box<dyn Weight>> {
+        let inner_weight = self.query.weight(searcher, scoring_enabled)?;
+        Ok(if scoring_enabled {
+            Box::new(ConstWeight::new(inner_weight, self.score))
+        } else {
+            inner_weight
+        })
+    }
+
+    fn query_terms<'a>(&'a self, visitor: &mut dyn FnMut(&'a Term, bool)) {
+        self.query.query_terms(visitor);
+    }
+}
+
+struct ConstWeight {
+    weight: Box<dyn Weight>,
+    score: Score,
+}
+
+impl ConstWeight {
+    pub fn new(weight: Box<dyn Weight>, score: Score) -> Self {
+        ConstWeight { weight, score }
+    }
+}
+
+impl Weight for ConstWeight {
+    fn scorer(&self, reader: &SegmentReader, boost: Score) -> crate::Result<Box<dyn Scorer>> {
+        let inner_scorer = self.weight.scorer(reader, boost)?;
+        Ok(Box::new(ConstScorer::new(inner_scorer, boost * self.score)))
+    }
+
+    fn explain(&self, reader: &SegmentReader, doc: u32) -> crate::Result<Explanation> {
+        let mut scorer = self.scorer(reader, 1.0)?;
+        if scorer.seek(doc) != doc {
+            return Err(TantivyError::InvalidArgument(format!(
+                "Document #({}) does not match",
+                doc
+            )));
+        }
+        let mut explanation = Explanation::new("Const", self.score);
+        let underlying_explanation = self.weight.explain(reader, doc)?;
+        explanation.add_detail(underlying_explanation);
+        Ok(explanation)
+    }
+
+    fn count(&self, reader: &SegmentReader) -> crate::Result<u32> {
+        self.weight.count(reader)
+    }
+}
+
+/// Wraps a `DocSet` and simply returns a constant `Scorer`.
+/// The `ConstScorer` is useful if you have a `DocSet` where
+/// you needed a scorer.
+///
+/// The `ConstScorer`'s constant score can be set
+/// by calling `.set_score(...)`.
+pub struct ConstScorer<TDocSet: DocSet> {
+    docset: TDocSet,
+    score: Score,
+}
+
+impl<TDocSet: DocSet> ConstScorer<TDocSet> {
+    /// Creates a new `ConstScorer`.
+    pub fn new(docset: TDocSet, score: Score) -> ConstScorer<TDocSet> {
+        ConstScorer { docset, score }
+    }
+}
+
+impl<TDocSet: DocSet> From<TDocSet> for ConstScorer<TDocSet> {
+    fn from(docset: TDocSet) -> Self {
+        ConstScorer::new(docset, 1.0)
+    }
+}
+
+impl<TDocSet: DocSet> DocSet for ConstScorer<TDocSet> {
+    fn advance(&mut self) -> DocId {
+        self.docset.advance()
+    }
+
+    fn seek(&mut self, target: DocId) -> DocId {
+        self.docset.seek(target)
+    }
+
+    fn fill_buffer(&mut self, buffer: &mut [DocId]) -> usize {
+        self.docset.fill_buffer(buffer)
+    }
+
+    fn doc(&self) -> DocId {
+        self.docset.doc()
+    }
+
+    fn size_hint(&self) -> u32 {
+        self.docset.size_hint()
+    }
+}
+
+impl<TDocSet: DocSet + 'static> Scorer for ConstScorer<TDocSet> {
+    fn score(&mut self) -> Score {
+        self.score
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::ConstScoreQuery;
+    use crate::query::{AllQuery, Query};
+    use crate::schema::Schema;
+    use crate::{DocAddress, Document, Index};
+
+    #[test]
+    fn test_const_score_query_explain() -> crate::Result<()> {
+        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::new())?;
+        index_writer.commit()?;
+        let reader = index.reader()?;
+        let searcher = reader.searcher();
+        let query = ConstScoreQuery::new(Box::new(AllQuery), 0.42);
+        let explanation = query.explain(&searcher, DocAddress::new(0, 0u32)).unwrap();
+        assert_eq!(
+            explanation.to_pretty_json(),
+            r#"{
+  "value": 0.42,
+  "description": "Const",
+  "details": [
+    {
+      "value": 1.0,
+      "description": "AllQuery",
+      "context": []
+    }
+  ],
+  "context": []
+}"#
+        );
+        Ok(())
+    }
+}

src/query/disjunction_max_query.rs

@@ -0,0 +1,131 @@
+use tantivy_query_grammar::Occur;
+
+use crate::query::{BooleanWeight, DisjunctionMaxCombiner, Query, Weight};
+use crate::{Score, Searcher, Term};
+
+/// The disjunction max query кeturns documents matching one or more wrapped queries,
+/// called query clauses or clauses.
+///
+/// If a returned document matches multiple query clauses,
+/// the `DisjunctionMaxQuery` assigns the document the highest relevance score from any matching
+/// clause, plus a tie breaking increment for any additional matching subqueries.
+///
+/// ```rust
+/// use tantivy::collector::TopDocs;
+/// use tantivy::doc;
+/// use tantivy::query::{DisjunctionMaxQuery, Query, QueryClone, TermQuery};
+/// use tantivy::schema::{IndexRecordOption, Schema, TEXT};
+/// use tantivy::Term;
+/// use tantivy::Index;
+///
+/// fn main() -> tantivy::Result<()> {
+///    let mut schema_builder = Schema::builder();
+///    let title = schema_builder.add_text_field("title", TEXT);
+///    let body = schema_builder.add_text_field("body", 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 Girl",
+///        ))?;
+///        index_writer.add_document(doc!(
+///            title => "The Diary of Muadib",
+///        ))?;
+///        index_writer.add_document(doc!(
+///            title => "The Diary of Girl",
+///        ))?;
+///        index_writer.commit()?;
+///    }
+///
+///    let reader = index.reader()?;
+///    let searcher = reader.searcher();
+///
+///    // Make TermQuery's for "girl" and "diary" in the title
+///    let girl_term_query: Box<dyn Query> = Box::new(TermQuery::new(
+///        Term::from_field_text(title, "girl"),
+///        IndexRecordOption::Basic,
+///    ));
+///    let diary_term_query: Box<dyn Query> = Box::new(TermQuery::new(
+///        Term::from_field_text(title, "diary"),
+///        IndexRecordOption::Basic,
+///    ));
+///
+///    // TermQuery "diary" and "girl" should be present and only one should be accounted in score
+///    let queries1 = vec![diary_term_query.box_clone(), girl_term_query.box_clone()];
+///    let diary_and_girl = DisjunctionMaxQuery::new(queries1);
+///    let documents = searcher.search(&diary_and_girl, &TopDocs::with_limit(3))?;
+///    assert_eq!(documents[0].0, documents[1].0);
+///    assert_eq!(documents[1].0, documents[2].0);
+///
+///    // TermQuery "diary" and "girl" should be present
+///    // and one should be accounted with multiplier 0.7
+///    let queries2 = vec![diary_term_query.box_clone(), girl_term_query.box_clone()];
+///    let tie_breaker = 0.7;
+///    let diary_and_girl_with_tie_breaker = DisjunctionMaxQuery::with_tie_breaker(queries2, tie_breaker);
+///    let documents = searcher.search(&diary_and_girl_with_tie_breaker, &TopDocs::with_limit(3))?;
+///    assert_eq!(documents[1].0, documents[2].0);
+///    // For this test all terms brings the same score. So we can do easy math and assume that
+///    // `DisjunctionMaxQuery` with tie breakers score should be equal
+///    // to term1 score + `tie_breaker` * term2 score or (1.0 + tie_breaker) * term score
+///    assert!(f32::abs(documents[0].0 - documents[1].0 * (1.0 + tie_breaker)) < 0.001);
+///    Ok(())
+/// }
+/// ```
+#[derive(Debug)]
+pub struct DisjunctionMaxQuery {
+    disjuncts: Vec<Box<dyn Query>>,
+    tie_breaker: Score,
+}
+
+impl Clone for DisjunctionMaxQuery {
+    fn clone(&self) -> Self {
+        DisjunctionMaxQuery::with_tie_breaker(
+            self.disjuncts
+                .iter()
+                .map(|disjunct| disjunct.box_clone())
+                .collect::<Vec<_>>(),
+            self.tie_breaker,
+        )
+    }
+}
+
+impl Query for DisjunctionMaxQuery {
+    fn weight(&self, searcher: &Searcher, scoring_enabled: bool) -> crate::Result<Box<dyn Weight>> {
+        let disjuncts = self
+            .disjuncts
+            .iter()
+            .map(|disjunct| Ok((Occur::Should, disjunct.weight(searcher, scoring_enabled)?)))
+            .collect::<crate::Result<_>>()?;
+        let tie_breaker = self.tie_breaker;
+        Ok(Box::new(BooleanWeight::new(
+            disjuncts,
+            scoring_enabled,
+            Box::new(move || DisjunctionMaxCombiner::with_tie_breaker(tie_breaker)),
+        )))
+    }
+
+    fn query_terms<'a>(&'a self, visitor: &mut dyn FnMut(&'a Term, bool)) {
+        for disjunct in &self.disjuncts {
+            disjunct.query_terms(visitor);
+        }
+    }
+}
+
+impl DisjunctionMaxQuery {
+    /// Creates a new `DisjunctionMaxQuery` with tie breaker.
+    pub fn with_tie_breaker(
+        disjuncts: Vec<Box<dyn Query>>,
+        tie_breaker: Score,
+    ) -> DisjunctionMaxQuery {
+        DisjunctionMaxQuery {
+            disjuncts,
+            tie_breaker,
+        }
+    }
+
+    /// Creates a new `DisjunctionMaxQuery` with no tie breaker.
+    pub fn new(disjuncts: Vec<Box<dyn Query>>) -> DisjunctionMaxQuery {
+        DisjunctionMaxQuery::with_tie_breaker(disjuncts, 0.0)
+    }
+}

src/query/mod.rs

@@ -6,6 +6,8 @@ mod bitset;
 mod bm25;
 mod boolean_query;
 mod boost_query;
+mod const_score_query;
+mod disjunction_max_query;
 mod empty_query;
 mod exclude;
 mod explanation;
@@ -34,7 +36,10 @@ pub use self::automaton_weight::AutomatonWeight;
 pub use self::bitset::BitSetDocSet;
 pub(crate) use self::bm25::Bm25Weight;
 pub use self::boolean_query::BooleanQuery;
+pub(crate) use self::boolean_query::BooleanWeight;
 pub use self::boost_query::BoostQuery;
+pub use self::const_score_query::{ConstScoreQuery, ConstScorer};
+pub use self::disjunction_max_query::DisjunctionMaxQuery;
 pub use self::empty_query::{EmptyQuery, EmptyScorer, EmptyWeight};
 pub use self::exclude::Exclude;
 pub use self::explanation::Explanation;
@@ -49,7 +54,10 @@ pub use self::query_parser::{QueryParser, QueryParserError};
 pub use self::range_query::RangeQuery;
 pub use self::regex_query::RegexQuery;
 pub use self::reqopt_scorer::RequiredOptionalScorer;
-pub use self::scorer::{ConstScorer, Scorer};
+pub use self::score_combiner::{
+    DisjunctionMaxCombiner, ScoreCombiner, SumCombiner, SumWithCoordsCombiner,
+};
+pub use self::scorer::Scorer;
 pub use self::term_query::TermQuery;
 pub use self::union::Union;
 #[cfg(test)]
@@ -58,8 +66,6 @@ pub use self::weight::Weight;
 
 #[cfg(test)]
 mod tests {
-    use std::collections::BTreeMap;
-
     use crate::query::QueryParser;
     use crate::schema::{Schema, TEXT};
     use crate::{Index, Term};
@@ -74,49 +80,34 @@ mod tests {
         let term_a = Term::from_field_text(text_field, "a");
         let term_b = Term::from_field_text(text_field, "b");
         {
-            let mut terms: BTreeMap<Term, bool> = Default::default();
-            query_parser
-                .parse_query("a")
-                .unwrap()
-                .query_terms(&mut terms);
-            let terms: Vec<(&Term, &bool)> = terms.iter().collect();
-            assert_eq!(vec![(&term_a, &false)], terms);
+            let query = query_parser.parse_query("a").unwrap();
+            let mut terms = Vec::new();
+            query.query_terms(&mut |term, pos| terms.push((term, pos)));
+            assert_eq!(vec![(&term_a, false)], terms);
         }
         {
-            let mut terms: BTreeMap<Term, bool> = Default::default();
-            query_parser
-                .parse_query("a b")
-                .unwrap()
-                .query_terms(&mut terms);
-            let terms: Vec<(&Term, &bool)> = terms.iter().collect();
-            assert_eq!(vec![(&term_a, &false), (&term_b, &false)], terms);
+            let query = query_parser.parse_query("a b").unwrap();
+            let mut terms = Vec::new();
+            query.query_terms(&mut |term, pos| terms.push((term, pos)));
+            assert_eq!(vec![(&term_a, false), (&term_b, false)], terms);
         }
         {
-            let mut terms: BTreeMap<Term, bool> = Default::default();
-            query_parser
-                .parse_query("\"a b\"")
-                .unwrap()
-                .query_terms(&mut terms);
-            let terms: Vec<(&Term, &bool)> = terms.iter().collect();
-            assert_eq!(vec![(&term_a, &true), (&term_b, &true)], terms);
+            let query = query_parser.parse_query("\"a b\"").unwrap();
+            let mut terms = Vec::new();
+            query.query_terms(&mut |term, pos| terms.push((term, pos)));
+            assert_eq!(vec![(&term_a, true), (&term_b, true)], terms);
         }
         {
-            let mut terms: BTreeMap<Term, bool> = Default::default();
-            query_parser
-                .parse_query("a a a a a")
-                .unwrap()
-                .query_terms(&mut terms);
-            let terms: Vec<(&Term, &bool)> = terms.iter().collect();
-            assert_eq!(vec![(&term_a, &false)], terms);
+            let query = query_parser.parse_query("a a a a a").unwrap();
+            let mut terms = Vec::new();
+            query.query_terms(&mut |term, pos| terms.push((term, pos)));
+            assert_eq!(vec![(&term_a, false); 5], terms);
         }
         {
-            let mut terms: BTreeMap<Term, bool> = Default::default();
-            query_parser
-                .parse_query("a -b")
-                .unwrap()
-                .query_terms(&mut terms);
-            let terms: Vec<(&Term, &bool)> = terms.iter().collect();
-            assert_eq!(vec![(&term_a, &false), (&term_b, &false)], terms);
+            let query = query_parser.parse_query("a -b").unwrap();
+            let mut terms = Vec::new();
+            query.query_terms(&mut |term, pos| terms.push((term, pos)));
+            assert_eq!(vec![(&term_a, false), (&term_b, false)], terms);
         }
     }
 }

src/query/more_like_this/more_like_this.rs

@@ -139,7 +139,7 @@ impl MoreLikeThis {
     }
 
     /// Finds terms for a more-like-this query.
-    /// field_to_field_values is a mapping from field to possible values of taht field.
+    /// field_to_field_values is a mapping from field to possible values of that field.
     fn retrieve_terms_from_doc_fields(
         &self,
         searcher: &Searcher,

src/query/phrase_query/phrase_query.rs

@@ -1,5 +1,3 @@
-use std::collections::BTreeMap;
-
 use super::PhraseWeight;
 use crate::core::searcher::Searcher;
 use crate::query::bm25::Bm25Weight;
@@ -68,7 +66,7 @@ impl PhraseQuery {
 
     /// Slop allowed for the phrase.
     ///
-    /// The query will match if its terms are seperated by `slop` terms at most.
+    /// The query will match if its terms are separated by `slop` terms at most.
     /// By default the slop is 0 meaning query terms need to be adjacent.  
     pub fn set_slop(&mut self, value: u32) {
         self.slop = value;
@@ -129,9 +127,9 @@ impl Query for PhraseQuery {
         Ok(Box::new(phrase_weight))
     }
 
-    fn query_terms(&self, terms: &mut BTreeMap<Term, bool>) {
+    fn query_terms<'a>(&'a self, visitor: &mut dyn FnMut(&'a Term, bool)) {
         for (_, term) in &self.phrase_terms {
-            terms.insert(term.clone(), true);
+            visitor(term, true);
         }
     }
 }

src/query/phrase_query/phrase_scorer.rs

@@ -428,7 +428,7 @@ mod tests {
     }
     #[test]
     fn test_slop() {
-        // The slop is not symetric. It does not allow for the phrase to be out of order.
+        // The slop is not symmetric. It does not allow for the phrase to be out of order.
         test_intersection_aux(&[1], &[2], &[2], 1);
         test_intersection_aux(&[1], &[3], &[], 1);
         test_intersection_aux(&[1], &[3], &[3], 2);

src/query/query.rs

@@ -1,4 +1,3 @@
-use std::collections::BTreeMap;
 use std::fmt;
 
 use downcast_rs::impl_downcast;
@@ -67,12 +66,15 @@ pub trait Query: QueryClone + Send + Sync + downcast_rs::Downcast + fmt::Debug {
         Ok(result)
     }
 
-    /// Extract all of the terms associated to the query and insert them in the
-    /// term set given in arguments.
+    /// Extract all of the terms associated to the query and pass them to the
+    /// given closure.
     ///
     /// Each term is associated with a boolean indicating whether
-    /// Positions are required or not.
-    fn query_terms(&self, _term_set: &mut BTreeMap<Term, bool>) {}
+    /// positions are required or not.
+    ///
+    /// Note that there can be multiple instances of any given term
+    /// in a query and deduplication must be handled by the visitor.
+    fn query_terms<'a>(&'a self, _visitor: &mut dyn FnMut(&'a Term, bool)) {}
 }
 
 /// Implements `box_clone`.
@@ -98,8 +100,8 @@ impl Query for Box<dyn Query> {
         self.as_ref().count(searcher)
     }
 
-    fn query_terms(&self, terms: &mut BTreeMap<Term, bool>) {
-        self.as_ref().query_terms(terms);
+    fn query_terms<'a>(&'a self, visitor: &mut dyn FnMut(&'a Term, bool)) {
+        self.as_ref().query_terms(visitor);
     }
 }
 

src/query/query_parser/query_parser.rs

@@ -184,7 +184,7 @@ pub struct QueryParser {
 fn all_negative(ast: &LogicalAst) -> bool {
     match ast {
         LogicalAst::Leaf(_) => false,
-        LogicalAst::Boost(ref child_ast, _) => all_negative(&*child_ast),
+        LogicalAst::Boost(ref child_ast, _) => all_negative(child_ast),
         LogicalAst::Clause(children) => children
             .iter()
             .all(|(ref occur, child)| (*occur == Occur::MustNot) || all_negative(child)),
@@ -577,7 +577,7 @@ impl QueryParser {
     /// object by naturally extending the json field name with a "." separated field_path
     /// - field_phrase: the phrase that is being searched.
     ///
-    /// The literal identifies the targetted field by a so-called *full field path*,
+    /// The literal identifies the targeted field by a so-called *full field path*,
     /// specified before the ":". (e.g. identity.username:fulmicoton).
     ///
     /// The way we split the full field path into (field_name, field_path) can be ambiguous,

src/query/reqopt_scorer.rs

@@ -51,6 +51,11 @@ where
         self.req_scorer.advance()
     }
 
+    fn seek(&mut self, target: DocId) -> DocId {
+        self.score_cache = None;
+        self.req_scorer.seek(target)
+    }
+
     fn doc(&self) -> DocId {
         self.req_scorer.doc()
     }
@@ -172,4 +177,23 @@ mod tests {
             skip_docs,
         );
     }
+
+    #[test]
+    fn test_reqopt_scorer_seek() {
+        let mut reqoptscorer: RequiredOptionalScorer<_, _, SumCombiner> =
+            RequiredOptionalScorer::new(
+                ConstScorer::new(VecDocSet::from(vec![1, 3, 7, 8, 9, 10, 13, 15]), 1.0),
+                ConstScorer::new(VecDocSet::from(vec![2, 7, 11, 12, 15]), 1.0),
+            );
+        {
+            assert_eq!(reqoptscorer.score(), 1.0);
+            assert_eq!(reqoptscorer.seek(7), 7);
+            assert_eq!(reqoptscorer.score(), 2.0);
+        }
+        {
+            assert_eq!(reqoptscorer.score(), 2.0);
+            assert_eq!(reqoptscorer.seek(12), 13);
+            assert_eq!(reqoptscorer.score(), 1.0);
+        }
+    }
 }

src/query/score_combiner.rs

@@ -77,3 +77,40 @@ impl ScoreCombiner for SumWithCoordsCombiner {
         self.score
     }
 }
+
+/// Take max score of different scorers
+/// and optionally sum it with other matches multiplied by `tie_breaker`
+#[derive(Default, Clone, Copy)]
+pub struct DisjunctionMaxCombiner {
+    max: Score,
+    sum: Score,
+    tie_breaker: Score,
+}
+
+impl DisjunctionMaxCombiner {
+    /// Creates `DisjunctionMaxCombiner` with tie breaker
+    pub fn with_tie_breaker(tie_breaker: Score) -> DisjunctionMaxCombiner {
+        DisjunctionMaxCombiner {
+            max: 0.0,
+            sum: 0.0,
+            tie_breaker,
+        }
+    }
+}
+
+impl ScoreCombiner for DisjunctionMaxCombiner {
+    fn update<TScorer: Scorer>(&mut self, scorer: &mut TScorer) {
+        let score = scorer.score();
+        self.max = Score::max(score, self.max);
+        self.sum += score;
+    }
+
+    fn clear(&mut self) {
+        self.max = 0.0;
+        self.sum = 0.0;
+    }
+
+    fn score(&self) -> Score {
+        self.max + (self.sum - self.max) * self.tie_breaker
+    }
+}

src/query/scorer.rs

@@ -3,7 +3,7 @@ use std::ops::DerefMut;
 use downcast_rs::impl_downcast;
 
 use crate::docset::DocSet;
-use crate::{DocId, Score};
+use crate::Score;
 
 /// Scored set of documents matching a query within a specific segment.
 ///
@@ -22,55 +22,3 @@ impl Scorer for Box<dyn Scorer> {
         self.deref_mut().score()
     }
 }
-
-/// Wraps a `DocSet` and simply returns a constant `Scorer`.
-/// The `ConstScorer` is useful if you have a `DocSet` where
-/// you needed a scorer.
-///
-/// The `ConstScorer`'s constant score can be set
-/// by calling `.set_score(...)`.
-pub struct ConstScorer<TDocSet: DocSet> {
-    docset: TDocSet,
-    score: Score,
-}
-
-impl<TDocSet: DocSet> ConstScorer<TDocSet> {
-    /// Creates a new `ConstScorer`.
-    pub fn new(docset: TDocSet, score: Score) -> ConstScorer<TDocSet> {
-        ConstScorer { docset, score }
-    }
-}
-
-impl<TDocSet: DocSet> From<TDocSet> for ConstScorer<TDocSet> {
-    fn from(docset: TDocSet) -> Self {
-        ConstScorer::new(docset, 1.0)
-    }
-}
-
-impl<TDocSet: DocSet> DocSet for ConstScorer<TDocSet> {
-    fn advance(&mut self) -> DocId {
-        self.docset.advance()
-    }
-
-    fn seek(&mut self, target: DocId) -> DocId {
-        self.docset.seek(target)
-    }
-
-    fn fill_buffer(&mut self, buffer: &mut [DocId]) -> usize {
-        self.docset.fill_buffer(buffer)
-    }
-
-    fn doc(&self) -> DocId {
-        self.docset.doc()
-    }
-
-    fn size_hint(&self) -> u32 {
-        self.docset.size_hint()
-    }
-}
-
-impl<TDocSet: DocSet + 'static> Scorer for ConstScorer<TDocSet> {
-    fn score(&mut self) -> Score {
-        self.score
-    }
-}

src/query/term_query/term_query.rs

@@ -1,4 +1,3 @@
-use std::collections::BTreeMap;
 use std::fmt;
 
 use super::term_weight::TermWeight;
@@ -121,7 +120,7 @@ impl Query for TermQuery {
             self.specialized_weight(searcher, scoring_enabled)?,
         ))
     }
-    fn query_terms(&self, terms: &mut BTreeMap<Term, bool>) {
-        terms.insert(self.term.clone(), false);
+    fn query_terms<'a>(&'a self, visitor: &mut dyn FnMut(&'a Term, bool)) {
+        visitor(&self.term, false);
     }
 }

src/query/union.rs

@@ -36,34 +36,6 @@ pub struct Union<TScorer, TScoreCombiner = DoNothingCombiner> {
     score: Score,
 }
 
-impl<TScorer, TScoreCombiner> From<Vec<TScorer>> for Union<TScorer, TScoreCombiner>
-where
-    TScoreCombiner: ScoreCombiner,
-    TScorer: Scorer,
-{
-    fn from(docsets: Vec<TScorer>) -> Union<TScorer, TScoreCombiner> {
-        let non_empty_docsets: Vec<TScorer> = docsets
-            .into_iter()
-            .filter(|docset| docset.doc() != TERMINATED)
-            .collect();
-        let mut union = Union {
-            docsets: non_empty_docsets,
-            bitsets: Box::new([TinySet::empty(); HORIZON_NUM_TINYBITSETS]),
-            scores: Box::new([TScoreCombiner::default(); HORIZON as usize]),
-            cursor: HORIZON_NUM_TINYBITSETS,
-            offset: 0,
-            doc: 0,
-            score: 0.0,
-        };
-        if union.refill() {
-            union.advance();
-        } else {
-            union.doc = TERMINATED;
-        }
-        union
-    }
-}
-
 fn refill<TScorer: Scorer, TScoreCombiner: ScoreCombiner>(
     scorers: &mut Vec<TScorer>,
     bitsets: &mut [TinySet; HORIZON_NUM_TINYBITSETS],
@@ -90,6 +62,31 @@ fn refill<TScorer: Scorer, TScoreCombiner: ScoreCombiner>(
 }
 
 impl<TScorer: Scorer, TScoreCombiner: ScoreCombiner> Union<TScorer, TScoreCombiner> {
+    pub(crate) fn build(
+        docsets: Vec<TScorer>,
+        score_combiner_fn: impl FnOnce() -> TScoreCombiner,
+    ) -> Union<TScorer, TScoreCombiner> {
+        let non_empty_docsets: Vec<TScorer> = docsets
+            .into_iter()
+            .filter(|docset| docset.doc() != TERMINATED)
+            .collect();
+        let mut union = Union {
+            docsets: non_empty_docsets,
+            bitsets: Box::new([TinySet::empty(); HORIZON_NUM_TINYBITSETS]),
+            scores: Box::new([score_combiner_fn(); HORIZON as usize]),
+            cursor: HORIZON_NUM_TINYBITSETS,
+            offset: 0,
+            doc: 0,
+            score: 0.0,
+        };
+        if union.refill() {
+            union.advance();
+        } else {
+            union.doc = TERMINATED;
+        }
+        union
+    }
+
     fn refill(&mut self) -> bool {
         if let Some(min_doc) = self.docsets.iter().map(DocSet::doc).min() {
             self.offset = min_doc;
@@ -179,7 +176,6 @@ where
 
             // The target is outside of the buffered horizon.
             // advance all docsets to a doc >= to the target.
-            #[cfg_attr(feature = "cargo-clippy", allow(clippy::clippy::collapsible_if))]
             unordered_drain_filter(&mut self.docsets, |docset| {
                 if docset.doc() < target {
                     docset.seek(target);
@@ -188,7 +184,7 @@ where
             });
 
             // at this point all of the docsets
-            // are positionned on a doc >= to the target.
+            // are positioned on a doc >= to the target.
             if !self.refill() {
                 self.doc = TERMINATED;
                 return TERMINATED;
@@ -266,12 +262,13 @@ mod tests {
         let union_vals: Vec<u32> = val_set.into_iter().collect();
         let mut union_expected = VecDocSet::from(union_vals);
         let make_union = || {
-            Union::from(
+            Union::build(
                 vals.iter()
                     .cloned()
                     .map(VecDocSet::from)
                     .map(|docset| ConstScorer::new(docset, 1.0))
                     .collect::<Vec<ConstScorer<VecDocSet>>>(),
+                DoNothingCombiner::default,
             )
         };
         let mut union: Union<_, DoNothingCombiner> = make_union();
@@ -312,13 +309,14 @@ mod tests {
             btree_set.extend(docs.iter().cloned());
         }
         let docset_factory = || {
-            let res: Box<dyn DocSet> = Box::new(Union::<_, DoNothingCombiner>::from(
+            let res: Box<dyn DocSet> = Box::new(Union::build(
                 docs_list
                     .iter()
                     .cloned()
                     .map(VecDocSet::from)
                     .map(|docset| ConstScorer::new(docset, 1.0))
                     .collect::<Vec<_>>(),
+                DoNothingCombiner::default,
             ));
             res
         };
@@ -346,10 +344,13 @@ mod tests {
 
     #[test]
     fn test_union_skip_corner_case3() {
-        let mut docset = Union::<_, DoNothingCombiner>::from(vec![
-            ConstScorer::from(VecDocSet::from(vec![0u32, 5u32])),
-            ConstScorer::from(VecDocSet::from(vec![1u32, 4u32])),
-        ]);
+        let mut docset = Union::build(
+            vec![
+                ConstScorer::from(VecDocSet::from(vec![0u32, 5u32])),
+                ConstScorer::from(VecDocSet::from(vec![1u32, 4u32])),
+            ],
+            DoNothingCombiner::default,
+        );
         assert_eq!(docset.doc(), 0u32);
         assert_eq!(docset.seek(0u32), 0u32);
         assert_eq!(docset.seek(0u32), 0u32);
@@ -405,12 +406,13 @@ mod bench {
             tests::sample_with_seed(100_000, 0.2, 1),
         ];
         bench.iter(|| {
-            let mut v = Union::<_, DoNothingCombiner>::from(
+            let mut v = Union::build(
                 union_docset
                     .iter()
                     .map(|doc_ids| VecDocSet::from(doc_ids.clone()))
                     .map(|docset| ConstScorer::new(docset, 1.0))
                     .collect::<Vec<_>>(),
+                DoNothingCombiner::default,
             );
             while v.doc() != TERMINATED {
                 v.advance();
@@ -425,12 +427,13 @@ mod bench {
             tests::sample_with_seed(100_000, 0.001, 2),
         ];
         bench.iter(|| {
-            let mut v = Union::<_, DoNothingCombiner>::from(
+            let mut v = Union::build(
                 union_docset
                     .iter()
                     .map(|doc_ids| VecDocSet::from(doc_ids.clone()))
                     .map(|docset| ConstScorer::new(docset, 1.0))
                     .collect::<Vec<_>>(),
+                DoNothingCombiner::default,
             );
             while v.doc() != TERMINATED {
                 v.advance();

src/reader/mod.rs

@@ -16,7 +16,7 @@ use crate::{Index, Inventory, Searcher, SegmentReader, TrackedObject};
 /// Defines when a new version of the index should be reloaded.
 ///
 /// Regardless of whether you search and index in the same process, tantivy does not necessarily
-/// reflects the change that are commited to your index. `ReloadPolicy` precisely helps you define
+/// reflects the change that are committed to your index. `ReloadPolicy` precisely helps you define
 /// when you want your index to be reloaded.
 #[derive(Clone, Copy)]
 pub enum ReloadPolicy {
@@ -164,21 +164,18 @@ impl InnerIndexReader {
         doc_store_cache_size: usize,
         index: Index,
         warming_state: WarmingState,
+        // The searcher_generation_inventory is not used as source, but as target to track the
+        // loaded segments.
         searcher_generation_inventory: Inventory<SearcherGeneration>,
     ) -> crate::Result<Self> {
         let searcher_generation_counter: Arc<AtomicU64> = Default::default();
-        let segment_readers = Self::open_segment_readers(&index)?;
-        let searcher_generation = Self::create_new_searcher_generation(
-            &segment_readers,
-            &searcher_generation_counter,
-            &searcher_generation_inventory,
-        );
 
         let searcher = Self::create_searcher(
             &index,
             doc_store_cache_size,
             &warming_state,
-            searcher_generation,
+            &searcher_generation_counter,
+            &searcher_generation_inventory,
         )?;
         Ok(InnerIndexReader {
             doc_store_cache_size,
@@ -204,12 +201,12 @@ impl InnerIndexReader {
         Ok(segment_readers)
     }
 
-    fn create_new_searcher_generation(
+    fn track_segment_readers_in_inventory(
         segment_readers: &[SegmentReader],
         searcher_generation_counter: &Arc<AtomicU64>,
         searcher_generation_inventory: &Inventory<SearcherGeneration>,
     ) -> TrackedObject<SearcherGeneration> {
-        let generation_id = searcher_generation_counter.fetch_add(1, atomic::Ordering::Relaxed);
+        let generation_id = searcher_generation_counter.fetch_add(1, atomic::Ordering::AcqRel);
         let searcher_generation =
             SearcherGeneration::from_segment_readers(segment_readers, generation_id);
         searcher_generation_inventory.track(searcher_generation)
@@ -219,9 +216,16 @@ impl InnerIndexReader {
         index: &Index,
         doc_store_cache_size: usize,
         warming_state: &WarmingState,
-        searcher_generation: TrackedObject<SearcherGeneration>,
+        searcher_generation_counter: &Arc<AtomicU64>,
+        searcher_generation_inventory: &Inventory<SearcherGeneration>,
     ) -> crate::Result<Arc<SearcherInner>> {
         let segment_readers = Self::open_segment_readers(index)?;
+        let searcher_generation = Self::track_segment_readers_in_inventory(
+            &segment_readers,
+            searcher_generation_counter,
+            searcher_generation_inventory,
+        );
+
         let schema = index.schema();
         let searcher = Arc::new(SearcherInner::new(
             schema,
@@ -236,17 +240,12 @@ impl InnerIndexReader {
     }
 
     fn reload(&self) -> crate::Result<()> {
-        let segment_readers = Self::open_segment_readers(&self.index)?;
-        let searcher_generation = Self::create_new_searcher_generation(
-            &segment_readers,
-            &self.searcher_generation_counter,
-            &self.searcher_generation_inventory,
-        );
         let searcher = Self::create_searcher(
             &self.index,
             self.doc_store_cache_size,
             &self.warming_state,
-            searcher_generation,
+            &self.searcher_generation_counter,
+            &self.searcher_generation_inventory,
         )?;
 
         self.searcher.store(searcher);

src/schema/bytes_options.rs

@@ -13,7 +13,7 @@ pub struct BytesOptions {
     stored: bool,
 }
 
-/// For backward compability we add an intermediary to interpret the
+/// For backward compatibility we add an intermediary to interpret the
 /// lack of fieldnorms attribute as "true" if and only if indexed.
 ///
 /// (Downstream, for the moment, this attribute is not used if not indexed...)

src/schema/facet.rs

@@ -7,6 +7,7 @@ use std::string::FromUtf8Error;
 use common::BinarySerializable;
 use once_cell::sync::Lazy;
 use regex::Regex;
+use serde::de::Error as _;
 use serde::{Deserialize, Deserializer, Serialize, Serializer};
 
 const SLASH_BYTE: u8 = b'/';
@@ -35,7 +36,7 @@ pub enum FacetParseError {
 /// have a `Facet` for `/electronics/tv_and_video/led_tv`.
 ///
 /// A document can be associated to any number of facets.
-/// The hierarchy implicitely imply that a document
+/// The hierarchy implicitly imply that a document
 /// belonging to a facet also belongs to the ancestor of
 /// its facet. In the example above, `/electronics/tv_and_video/`
 /// and `/electronics`.
@@ -150,13 +151,26 @@ impl Facet {
         self.0.push_str(facet_str);
     }
 
-    /// Returns `true` if other is a subfacet of `self`.
+    /// Returns `true` if other is a `strict` subfacet of `self`.
+    ///
+    /// Disclaimer: By strict we mean that the relation is not reflexive.
+    /// `/happy` is not a prefix of `/happy`.
     pub fn is_prefix_of(&self, other: &Facet) -> bool {
         let self_str = self.encoded_str();
         let other_str = other.encoded_str();
-        self_str.len() < other_str.len()
-            && other_str.starts_with(self_str)
-            && other_str.as_bytes()[self_str.len()] == FACET_SEP_BYTE
+
+        // Fast path, but also required to ensure that / is not a prefix of /.
+        if other_str.len() <= self_str.len() {
+            return false;
+        }
+
+        // Root is a prefix of every other path.
+        // This is not just an optimisation. It is necessary for correctness.
+        if self.is_root() {
+            return true;
+        }
+
+        other_str.starts_with(self_str) && other_str.as_bytes()[self_str.len()] == FACET_SEP_BYTE
     }
 
     /// Extract path from the `Facet`.
@@ -217,7 +231,9 @@ impl Serialize for Facet {
 impl<'de> Deserialize<'de> for Facet {
     fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
     where D: Deserializer<'de> {
-        <&'de str as Deserialize<'de>>::deserialize(deserializer).map(Facet::from)
+        <Cow<'de, str> as Deserialize<'de>>::deserialize(deserializer).and_then(|path| {
+            Facet::from_text(&*path).map_err(|err| D::Error::custom(err.to_string()))
+        })
     }
 }
 
@@ -301,4 +317,38 @@ mod tests {
             Facet::from_text("INVALID")
         );
     }
+
+    #[test]
+    fn only_proper_prefixes() {
+        assert!(Facet::from("/foo").is_prefix_of(&Facet::from("/foo/bar")));
+
+        assert!(!Facet::from("/foo/bar").is_prefix_of(&Facet::from("/foo/bar")));
+    }
+
+    #[test]
+    fn root_is_a_prefix() {
+        assert!(Facet::from("/").is_prefix_of(&Facet::from("/foobar")));
+        assert!(!Facet::from("/").is_prefix_of(&Facet::from("/")));
+    }
+
+    #[test]
+    fn deserialize_from_borrowed_string() {
+        let facet = serde_json::from_str::<Facet>(r#""/foo/bar""#).unwrap();
+        assert_eq!(facet, Facet::from_path(["foo", "bar"]));
+    }
+
+    #[test]
+    fn deserialize_from_owned_string() {
+        let facet = serde_json::from_str::<Facet>(r#""/foo/\u263A""#).unwrap();
+        assert_eq!(facet, Facet::from_path(["foo", "☺"]));
+    }
+
+    #[test]
+    fn deserialize_from_invalid_string() {
+        let error = serde_json::from_str::<Facet>(r#""foo/bar""#).unwrap_err();
+        assert_eq!(
+            error.to_string(),
+            "Failed to parse the facet string: 'foo/bar'"
+        );
+    }
 }

src/schema/field_type.rs

@@ -139,7 +139,7 @@ pub enum FieldType {
     Bool(NumericOptions),
     /// Signed 64-bits Date 64 field type configuration,
     Date(DateOptions),
-    /// Hierachical Facet
+    /// Hierarchical Facet
     Facet(FacetOptions),
     /// Bytes (one per document)
     Bytes(BytesOptions),

src/schema/numeric_options.rs

@@ -32,7 +32,7 @@ pub struct NumericOptions {
     stored: bool,
 }
 
-/// For backward compability we add an intermediary to interpret the
+/// For backward compatibility we add an intermediary to interpret the
 /// lack of fieldnorms attribute as "true" if and only if indexed.
 ///
 /// (Downstream, for the moment, this attribute is not used anyway if not indexed...)

src/schema/term.rs

@@ -16,7 +16,7 @@ use crate::{DatePrecision, DateTime};
 /// If this is a JSON term, the type is the type of the leaf of the json.
 ///
 /// - <value> is,  if this is not the json term, a binary representation specific to the type.
-/// If it is a JSON Term, then it is preprended with the path that leads to this leaf value.
+/// If it is a JSON Term, then it is prepended with the path that leads to this leaf value.
 const FAST_VALUE_TERM_LEN: usize = 4 + 1 + 8;
 
 /// Separates the different segments of

src/schema/value.rs

@@ -418,7 +418,7 @@ mod binary_serialize {
                         _ => Err(io::Error::new(
                             io::ErrorKind::InvalidData,
                             format!(
-                                "No extened field type is associated with code {:?}",
+                                "No extended field type is associated with code {:?}",
                                 ext_type_code
                             ),
                         )),

src/snippet/mod.rs

@@ -1,5 +1,5 @@
 use std::cmp::Ordering;
-use std::collections::BTreeMap;
+use std::collections::{BTreeMap, BTreeSet};
 use std::ops::Range;
 
 use htmlescape::encode_minimal;
@@ -7,7 +7,7 @@ use htmlescape::encode_minimal;
 use crate::query::Query;
 use crate::schema::{Field, Value};
 use crate::tokenizer::{TextAnalyzer, Token};
-use crate::{Document, Score, Searcher};
+use crate::{Document, Score, Searcher, Term};
 
 const DEFAULT_MAX_NUM_CHARS: usize = 150;
 
@@ -50,7 +50,7 @@ impl FragmentCandidate {
 }
 
 /// `Snippet`
-/// Contains a fragment of a document, and some highlighed parts inside it.
+/// Contains a fragment of a document, and some highlighted parts inside it.
 #[derive(Debug)]
 pub struct Snippet {
     fragment: String,
@@ -69,12 +69,17 @@ impl Snippet {
         }
     }
 
-    /// Returns a hignlightned html from the `Snippet`.
+    /// Returns `true` if the snippet is empty.
+    pub fn is_empty(&self) -> bool {
+        self.highlighted.len() == 0
+    }
+
+    /// Returns a highlighted html from the `Snippet`.
     pub fn to_html(&self) -> String {
         let mut html = String::new();
         let mut start_from: usize = 0;
 
-        for item in self.highlighted.iter() {
+        for item in collapse_overlapped_ranges(&self.highlighted) {
             html.push_str(&encode_minimal(&self.fragment[start_from..item.start]));
             html.push_str(HIGHLIGHTEN_PREFIX);
             html.push_str(&encode_minimal(&self.fragment[item.clone()]));
@@ -92,7 +97,7 @@ impl Snippet {
         &self.fragment
     }
 
-    /// Returns a list of higlighted positions from the `Snippet`.
+    /// Returns a list of highlighted positions from the `Snippet`.
     pub fn highlighted(&self) -> &[Range<usize>] {
         &self.highlighted
     }
@@ -181,6 +186,53 @@ fn select_best_fragment_combination(fragments: &[FragmentCandidate], text: &str)
     }
 }
 
+/// Returns ranges that are collapsed into non-overlapped ranges.
+///
+/// ## Examples
+/// - [0..1, 2..3] -> [0..1, 2..3]  # no overlap
+/// - [0..1, 1..2] -> [0..1, 1..2]  # no overlap
+/// - [0..2, 1..2] -> [0..2]  # collapsed
+/// - [0..2, 1..3] -> [0..3]  # collapsed
+/// - [0..3, 1..2] -> [0..3]  # second range's end is also inside of the first range
+///
+/// Note: This function assumes `ranges` is sorted by `Range.start` in ascending order.
+fn collapse_overlapped_ranges(ranges: &[Range<usize>]) -> Vec<Range<usize>> {
+    debug_assert!(is_sorted(ranges.iter().map(|range| range.start)));
+
+    let mut result = Vec::new();
+    let mut ranges_it = ranges.iter();
+
+    let mut current = match ranges_it.next() {
+        Some(range) => range.clone(),
+        None => return result,
+    };
+
+    for range in ranges {
+        if current.end > range.start {
+            current = current.start..std::cmp::max(current.end, range.end);
+        } else {
+            result.push(current);
+            current = range.clone();
+        }
+    }
+
+    result.push(current);
+    result
+}
+
+fn is_sorted(mut it: impl Iterator<Item = usize>) -> bool {
+    if let Some(first) = it.next() {
+        let mut prev = first;
+        for item in it {
+            if item < prev {
+                return false;
+            }
+            prev = item;
+        }
+    }
+    true
+}
+
 /// `SnippetGenerator`
 ///
 /// # Example
@@ -230,25 +282,40 @@ pub struct SnippetGenerator {
 }
 
 impl SnippetGenerator {
+    /// Creates a new snippet generator
+    pub fn new(
+        terms_text: BTreeMap<String, Score>,
+        tokenizer: TextAnalyzer,
+        field: Field,
+        max_num_chars: usize,
+    ) -> Self {
+        SnippetGenerator {
+            terms_text,
+            tokenizer,
+            field,
+            max_num_chars,
+        }
+    }
     /// Creates a new snippet generator
     pub fn create(
         searcher: &Searcher,
         query: &dyn Query,
         field: Field,
     ) -> crate::Result<SnippetGenerator> {
-        let mut terms = BTreeMap::new();
-        query.query_terms(&mut terms);
-        let mut terms_text: BTreeMap<String, Score> = Default::default();
-        for (term, _) in terms {
-            if term.field() != field {
-                continue;
+        let mut terms: BTreeSet<&Term> = BTreeSet::new();
+        query.query_terms(&mut |term, _| {
+            if term.field() == field {
+                terms.insert(term);
             }
+        });
+        let mut terms_text: BTreeMap<String, Score> = Default::default();
+        for term in terms {
             let term_str = if let Some(term_str) = term.as_str() {
                 term_str
             } else {
                 continue;
             };
-            let doc_freq = searcher.doc_freq(&term)?;
+            let doc_freq = searcher.doc_freq(term)?;
             if doc_freq > 0 {
                 let score = 1.0 / (1.0 + doc_freq as Score);
                 terms_text.insert(term_str.to_string(), score);
@@ -300,10 +367,10 @@ mod tests {
 
     use maplit::btreemap;
 
-    use super::{search_fragments, select_best_fragment_combination};
+    use super::{collapse_overlapped_ranges, search_fragments, select_best_fragment_combination};
     use crate::query::QueryParser;
     use crate::schema::{IndexRecordOption, Schema, TextFieldIndexing, TextOptions, TEXT};
-    use crate::tokenizer::SimpleTokenizer;
+    use crate::tokenizer::{NgramTokenizer, SimpleTokenizer};
     use crate::{Index, SnippetGenerator};
 
     const TEST_TEXT: &str = r#"Rust is a systems programming language sponsored by
@@ -460,6 +527,7 @@ Survey in 2016, 2017, and 2018."#;
         let snippet = select_best_fragment_combination(&fragments[..], text);
         assert_eq!(snippet.fragment, "");
         assert_eq!(snippet.to_html(), "");
+        assert!(snippet.is_empty());
     }
 
     #[test]
@@ -473,6 +541,7 @@ Survey in 2016, 2017, and 2018."#;
         let snippet = select_best_fragment_combination(&fragments[..], text);
         assert_eq!(snippet.fragment, "");
         assert_eq!(snippet.to_html(), "");
+        assert!(snippet.is_empty());
     }
 
     #[test]
@@ -566,4 +635,44 @@ Survey in 2016, 2017, and 2018."#;
         }
         Ok(())
     }
+
+    #[test]
+    fn test_collapse_overlapped_ranges() {
+        assert_eq!(&collapse_overlapped_ranges(&[0..1, 2..3,]), &[0..1, 2..3]);
+        assert_eq!(
+            collapse_overlapped_ranges(&vec![0..1, 1..2,]),
+            vec![0..1, 1..2]
+        );
+        assert_eq!(collapse_overlapped_ranges(&[0..2, 1..2,]), vec![0..2]);
+        assert_eq!(collapse_overlapped_ranges(&[0..2, 1..3,]), vec![0..3]);
+        assert_eq!(collapse_overlapped_ranges(&[0..3, 1..2,]), vec![0..3]);
+    }
+
+    #[test]
+    fn test_snippet_with_overlapped_highlighted_ranges() {
+        let text = "abc";
+
+        let mut terms = BTreeMap::new();
+        terms.insert(String::from("ab"), 0.9);
+        terms.insert(String::from("bc"), 1.0);
+
+        let fragments = search_fragments(
+            &From::from(NgramTokenizer::all_ngrams(2, 2)),
+            text,
+            &terms,
+            3,
+        );
+
+        assert_eq!(fragments.len(), 1);
+        {
+            let first = &fragments[0];
+            assert_eq!(first.score, 1.9);
+            assert_eq!(first.start_offset, 0);
+            assert_eq!(first.stop_offset, 3);
+        }
+
+        let snippet = select_best_fragment_combination(&fragments[..], text);
+        assert_eq!(snippet.fragment, "abc");
+        assert_eq!(snippet.to_html(), "<b>abc</b>");
+    }
 }