Bumping version

For Field with several FieldValues, with a
value that contained no token at all, the token position
was reinitialized to 0.

As a result, PhraseQueries can show some false positives.
In addition, after the computation of the position delta, we can
underflow u32, and end up with gigantic delta.

We haven't been able to actually explain the bug in 1629, but it
is assumed that in some corner case these delta can cause a panic.

Closes #1629

.gitattributes

@@ -0,0 +1 @@
+cpp/* linguist-vendored

.github/workflows/coverage.yml

@@ -2,9 +2,9 @@ name: Coverage
 
 on:
   push:
-    branches: [main]
+    branches: [ main ]
   pull_request:
-    branches: [main]
+    branches: [ main ]
 
 jobs:
   coverage:
@@ -12,14 +12,12 @@ jobs:
     steps:
       - uses: actions/checkout@v3
       - name: Install Rust
-        run: rustup toolchain install nightly --profile minimal --component llvm-tools-preview
-      - uses: Swatinem/rust-cache@v2
+        run: rustup toolchain install nightly --component llvm-tools-preview
       - uses: taiki-e/install-action@cargo-llvm-cov
       - name: Generate code coverage
-        run: cargo +nightly llvm-cov --all-features --workspace --doctests --lcov --output-path lcov.info
+        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

@@ -9,20 +9,16 @@ env:
   NUM_FUNCTIONAL_TEST_ITERATIONS: 20000
 
 jobs:
-  test:
-
+  functional_test_unsorted:
     runs-on: ubuntu-latest
-
     steps:
     - uses: actions/checkout@v3
-    - name: Install stable
-      uses: actions-rs/toolchain@v1
-      with:
-          toolchain: stable
-          profile: minimal
-          override: true
-
     - name: Run indexing_unsorted
       run: cargo test indexing_unsorted -- --ignored
+  functional_test_sorted:
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v3
     - name: Run indexing_sorted
       run: cargo test indexing_sorted -- --ignored
+

.github/workflows/test.yml

@@ -10,27 +10,33 @@ env:
   CARGO_TERM_COLOR: always
 
 jobs:
-  check:
+  test:
 
     runs-on: ubuntu-latest
 
     steps:
     - uses: actions/checkout@v3
-
-    - name: Install nightly
+    - name: Build
+      run: cargo build --verbose --workspace
+    - name: Install latest nightly to test also against unstable feature flag
       uses: actions-rs/toolchain@v1
       with:
             toolchain: nightly
-            profile: minimal
+            override: true
             components: rustfmt
-    - name: Install stable
+
+    - name: Install latest nightly to test also against unstable feature flag
       uses: actions-rs/toolchain@v1
       with:
             toolchain: stable
-            profile: minimal
-            components: clippy
+            override: true
+            components: rustfmt, clippy
 
-    - uses: Swatinem/rust-cache@v2
+    - 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
 
     - name: Check Formatting
       run: cargo +nightly fmt --all -- --check
@@ -41,34 +47,3 @@ jobs:
         token: ${{ secrets.GITHUB_TOKEN }}
         args: --tests
 
-  test:
-
-    runs-on: ubuntu-latest
-
-    strategy:
-      matrix:
-        features: [
-            { label: "all", flags: "mmap,stopwords,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

.gitignore

@@ -9,6 +9,7 @@ target/release
 Cargo.lock
 benchmark
 .DS_Store
+cpp/simdcomp/bitpackingbenchmark
 *.bk
 .idea
 trace.dat

ARCHITECTURE.md

@@ -10,7 +10,6 @@ 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,
@@ -23,28 +22,27 @@ 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.
 
@@ -54,15 +52,17 @@ 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```.
+On commit, tantivy will find all of the segments with documents matching this existing term and create a [tombstone file](src/fastfield/delete.rs) that represents the bitset of the document that are deleted.
+Like all segment files, this file is immutable. Because it is possible to have more than one tombstone file at a given instant, the tombstone filename has the format ``` segment_id . commit_opstamp . del```.
 
 An opstamp is simply an incremental id that identifies any operation applied to the index. For instance, performing a commit or adding a document.
 
+
 ### DocId
 
 Within a segment, all documents are identified by a DocId that ranges within `[0, max_doc)`.
@@ -74,7 +74,6 @@ 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.
 
@@ -95,7 +94,7 @@ called [`Directory`](src/directory/directory.rs).
 Contrary to Lucene however, "files" are quite different from some kind of `io::Read` object.
 Check out [`src/directory/directory.rs`](src/directory/directory.rs) trait for more details.
 
-Tantivy ships two main directory implementation: the `MmapDirectory` and the `RamDirectory`,
+Tantivy ships two main directory implementation: the `MMapDirectory` and the `RAMDirectory`,
 but users can extend tantivy with their own implementation.
 
 ## [schema/](src/schema): What are documents?
@@ -105,7 +104,6 @@ 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
@@ -119,10 +117,9 @@ 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
@@ -135,7 +132,7 @@ 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
@@ -149,7 +146,6 @@ 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.
@@ -158,7 +154,8 @@ 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.
@@ -166,7 +163,7 @@ 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))
 ```
 
@@ -193,7 +190,7 @@ 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).
@@ -218,18 +215,19 @@ 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.
@@ -251,7 +249,7 @@ For instance, when the phrase query "the art of war" does not match "the war of
 To make it possible, it is possible to specify in the schema that a field should store positions in addition to being indexed.
 
 The token positions of all of the terms are then stored in a separate file with the extension `.pos`.
-The [TermInfo](src/postings/term_info.rs) gives an offset (expressed in position this time) in this file. As we iterate through the docset,
+The [TermInfo](src/postings/term_info.rs) gives an offset (expressed in position this time) in this file. As we iterate throught the docset,
 we advance the position reader by the number of term frequencies of the current document.
 
 ## [fieldnorms/](src/fieldnorms): Here is my doc, how many tokens in this field?
@@ -259,6 +257,7 @@ 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.
@@ -269,6 +268,7 @@ 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

@@ -1,41 +1,9 @@
-Tantivy 0.19
+Tantivy 0.18.1
 ================================
-#### Bugfixes
-- Fix missing fieldnorms for u64, i64, f64, bool, bytes and date [#1620](https://github.com/quickwit-oss/tantivy/pull/1620) (@PSeitz)
-- Fix interpolation overflow in linear interpolation fastfield codec [#1480](https://github.com/quickwit-oss/tantivy/pull/1480) (@PSeitz @fulmicoton)
-
-#### Features/Improvements
-- Add support for `IN` in queryparser , e.g. `field: IN [val1 val2 val3]` [#1683](https://github.com/quickwit-oss/tantivy/pull/1683) (@trinity-1686a)
-- Skip score calculation, when no scoring is required [#1646](https://github.com/quickwit-oss/tantivy/pull/1646) (@PSeitz)
-- Limit fast fields to u32 (`get_val(u32)`) [#1644](https://github.com/quickwit-oss/tantivy/pull/1644) (@PSeitz)
-- 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) [#1396](https://github.com/quickwit-oss/tantivy/pull/1396) (@evanxg852000)
-- Add IP address field type [#1553](https://github.com/quickwit-oss/tantivy/pull/1553) (@PSeitz)
-- Add boolean field type [#1382](https://github.com/quickwit-oss/tantivy/pull/1382) (@boraarslan)
-- Remove Searcher pool and make `Searcher` cloneable. (@PSeitz)
-- Validate settings on create [#1570](https://github.com/quickwit-oss/tantivy/pull/1570) (@PSeitz)
-- Detect and apply gcd on fastfield codecs [#1418](https://github.com/quickwit-oss/tantivy/pull/1418) (@PSeitz)
-- Doc store
-  - use separate thread to compress block store [#1389](https://github.com/quickwit-oss/tantivy/pull/1389) [#1510](https://github.com/quickwit-oss/tantivy/pull/1510) (@PSeitz @fulmicoton)
-  - Expose doc store cache size [#1403](https://github.com/quickwit-oss/tantivy/pull/1403) (@PSeitz)
-  - Enable compression levels for doc store [#1378](https://github.com/quickwit-oss/tantivy/pull/1378) (@PSeitz)
-  - Make block size configurable [#1374](https://github.com/quickwit-oss/tantivy/pull/1374) (@kryesh)
-- Make `tantivy::TantivyError` cloneable [#1402](https://github.com/quickwit-oss/tantivy/pull/1402) (@PSeitz)
-- Add support for phrase slop in query language [#1393](https://github.com/quickwit-oss/tantivy/pull/1393) (@saroh)
-- Aggregation
-  - Add aggregation support for date type [#1693](https://github.com/quickwit-oss/tantivy/pull/1693)(@PSeitz)
-  - Add support for keyed parameter in range and histgram aggregations [#1424](https://github.com/quickwit-oss/tantivy/pull/1424) (@k-yomo)
-  - Add aggregation bucket limit [#1363](https://github.com/quickwit-oss/tantivy/pull/1363) (@PSeitz)
-- Faster indexing
-  - [#1610](https://github.com/quickwit-oss/tantivy/pull/1610) (@PSeitz)
-  - [#1594](https://github.com/quickwit-oss/tantivy/pull/1594) (@PSeitz)
-  - [#1582](https://github.com/quickwit-oss/tantivy/pull/1582) (@PSeitz)
-  - [#1611](https://github.com/quickwit-oss/tantivy/pull/1611) (@PSeitz)
-  - Added a pre-configured stop word filter for various language [#1666](https://github.com/quickwit-oss/tantivy/pull/1666) (@adamreichold)
+- Hotfix: positions computation.  #1629 (@fmassot, @fulmicoton, @PSeitz)
 
 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.
@@ -49,13 +17,8 @@ Tantivy 0.18
 - Add terms aggregation (@PSeitz)
 - Add support for zstd compression (@kryesh)
 
-Tantivy 0.18.1
-================================
-- Hotfix: positions computation.  #1629 (@fmassot, @fulmicoton, @PSeitz)
-
 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
@@ -70,39 +33,33 @@ Tantivy 0.17
 
 Tantivy 0.16.2
 ================================
-
-- Bugfix in FuzzyTermQuery. (transposition_cost_one was not doing anything)
+- Bugfix in FuzzyTermQuery. (tranposition_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
 
+
 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
@@ -124,9 +81,9 @@ 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.
@@ -145,19 +102,16 @@ 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
@@ -172,7 +126,6 @@ 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 {
@@ -180,9 +133,7 @@ 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
@@ -190,7 +141,6 @@ 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)
@@ -201,32 +151,30 @@ 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
 
+
 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)
@@ -257,6 +205,7 @@ Tantivy 0.10.1
 Avoid watching the mmap directory until someone effectively creates a reader that uses
 this functionality.
 
+
 Tantivy 0.10.0
 =====================
 
@@ -272,7 +221,6 @@ 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
@@ -280,7 +228,8 @@ 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)
@@ -296,17 +245,16 @@ 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)
@@ -350,40 +298,37 @@ 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)
@@ -393,15 +338,15 @@ 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
 ==========================
@@ -414,53 +359,58 @@ 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.
 
+
 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
 ==========================
 
@@ -475,31 +425,37 @@ 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:
 
@@ -508,16 +464,19 @@ 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`
@@ -529,3 +488,7 @@ Thanks to @KodrAus ! (#108)
 - Building binary targets for tantivy-cli (Thanks to @KodrAus)
 - Misc invisible bug fixes, and code cleanup.
 - Use
+
+
+
+

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "tantivy"
-version = "0.19.0"
+version = "0.18.1"
 authors = ["Paul Masurel <paul.masurel@gmail.com>"]
 license = "MIT"
 categories = ["database-implementations", "data-structures"]
@@ -11,56 +11,54 @@ repository = "https://github.com/quickwit-oss/tantivy"
 readme = "README.md"
 keywords = ["search", "information", "retrieval"]
 edition = "2021"
-rust-version = "1.62"
 
 [dependencies]
-oneshot = "0.1.5"
-base64 = "0.21.0"
+oneshot = "0.1.3"
+base64 = "0.13.0"
+byteorder = "1.4.3"
 crc32fast = "1.3.2"
 once_cell = "1.10.0"
 regex = { version = "1.5.5", default-features = false, features = ["std", "unicode"] }
-aho-corasick = "0.7"
-tantivy-fst = "0.4.0"
+tantivy-fst = "0.3.0"
 memmap2 = { version = "0.5.3", optional = true }
-lz4_flex = { version = "0.10", default-features = false, features = ["checked-decode"], optional = true }
+lz4_flex = { version = "0.9.2", default-features = false, features = ["checked-decode"], optional = true }
 brotli = { version = "3.3.4", optional = true }
-zstd = { version = "0.12", optional = true, default-features = false }
+zstd = { version = "0.11", optional = true }
 snap = { version = "1.0.5", optional = true }
 tempfile = { version = "3.3.0", optional = true }
 log = "0.4.16"
 serde = { version = "1.0.136", features = ["derive"] }
 serde_json = "1.0.79"
 num_cpus = "1.13.1"
-fs2 = { version = "0.4.3", optional = true }
+fs2={ version = "0.4.3", optional = true }
 levenshtein_automata = "0.2.1"
 uuid = { version = "1.0.0", features = ["v4", "serde"] }
 crossbeam-channel = "0.5.4"
+tantivy-query-grammar = { version="0.18.0", path="./query-grammar" }
+tantivy-bitpacker = { version="0.2", path="./bitpacker" }
+common = { version = "0.3", path = "./common/", package = "tantivy-common" }
+fastfield_codecs = { version="0.2", path="./fastfield_codecs", default-features = false }
+ownedbytes = { version="0.3", path="./ownedbytes" }
+stable_deref_trait = "1.2.0"
 rust-stemmers = "1.2.0"
 downcast-rs = "1.2.0"
 bitpacking = { version = "0.8.4", default-features = false, features = ["bitpacker4x"] }
 census = "0.4.0"
-rustc-hash = "1.1.0"
+fnv = "1.0.7"
 thiserror = "1.0.30"
 htmlescape = "0.3.1"
 fail = "0.5.0"
-murmurhash32 = "0.3.0"
-time = { version = "0.3.10", features = ["serde-well-known"] }
+murmurhash32 = "0.2.0"
+time = { version = "0.3.9", features = ["serde-well-known"] }
 smallvec = "1.8.0"
 rayon = "1.5.2"
-lru = "0.9.0"
+lru = "0.7.5"
 fastdivide = "0.4.0"
 itertools = "0.10.3"
 measure_time = "0.8.2"
+pretty_assertions = "1.2.1"
+serde_cbor = { version = "0.11.2", optional = true }
 async-trait = "0.1.53"
-arc-swap = "1.5.0"
-
-columnar = { version="0.1", path="./columnar", package ="tantivy-columnar" }
-sstable = { version="0.1", path="./sstable", package ="tantivy-sstable", optional = true }
-stacker = { version="0.1", path="./stacker", package ="tantivy-stacker" }
-query-grammar = { version= "0.19.0", path="./query-grammar", package = "tantivy-query-grammar" }
-tantivy-bitpacker = { version= "0.3", path="./bitpacker" }
-common = { version= "0.5", path = "./common/", package = "tantivy-common" }
-tokenizer-api = { version="0.1", path="./tokenizer-api", package="tantivy-tokenizer-api" }
 
 [target.'cfg(windows)'.dependencies]
 winapi = "0.3.9"
@@ -69,14 +67,12 @@ winapi = "0.3.9"
 rand = "0.8.5"
 maplit = "1.0.2"
 matches = "0.1.9"
-pretty_assertions = "1.2.1"
 proptest = "1.0.0"
-criterion = "0.4"
+criterion = "0.3.5"
 test-log = "0.2.10"
-env_logger = "0.10.0"
-pprof = { version = "0.11.0", features = ["flamegraph", "criterion"] }
+env_logger = "0.9.0"
+pprof = { version = "0.9.0", features = ["flamegraph", "criterion"] }
 futures = "0.3.21"
-paste = "1.0.11"
 
 [dev-dependencies.fail]
 version = "0.5.0"
@@ -92,9 +88,8 @@ debug-assertions = true
 overflow-checks = true
 
 [features]
-default = ["mmap", "stopwords", "lz4-compression"]
+default = ["mmap", "lz4-compression" ]
 mmap = ["fs2", "tempfile", "memmap2"]
-stopwords = []
 
 brotli-compression = ["brotli"]
 lz4-compression = ["lz4_flex"]
@@ -104,10 +99,10 @@ zstd-compression = ["zstd"]
 failpoints = ["fail/failpoints"]
 unstable = [] # useful for benches.
 
-quickwit = ["sstable"]
+quickwit = ["serde_cbor"]
 
 [workspace]
-members = ["query-grammar", "bitpacker", "common", "ownedbytes", "stacker", "sstable", "tokenizer-api", "columnar"]
+members = ["query-grammar", "bitpacker", "common", "fastfield_codecs", "ownedbytes"]
 
 # Following the "fail" crate best practises, we isolate
 # tests that define specific behavior in fail check points

README.md

@@ -5,6 +5,7 @@
 [![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.
@@ -15,7 +16,7 @@ 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
 
@@ -29,7 +30,7 @@ Your mileage WILL vary depending on the nature of queries and their load.
 # Features
 
 - Full-text search
-- Configurable tokenizer (stemming available for 17 Latin languages) with third party support for Chinese ([tantivy-jieba](https://crates.io/crates/tantivy-jieba) and [cang-jie](https://crates.io/crates/cang-jie)), Japanese ([lindera](https://github.com/lindera-morphology/lindera-tantivy), [Vaporetto](https://crates.io/crates/vaporetto_tantivy), and [tantivy-tokenizer-tiny-segmenter](https://crates.io/crates/tantivy-tokenizer-tiny-segmenter)) and Korean ([lindera](https://github.com/lindera-morphology/lindera-tantivy) + [lindera-ko-dic-builder](https://github.com/lindera-morphology/lindera-ko-dic-builder))
+- Configurable tokenizer (stemming available for 17 Latin languages with third party support for Chinese ([tantivy-jieba](https://crates.io/crates/tantivy-jieba) and [cang-jie](https://crates.io/crates/cang-jie)), Japanese ([lindera](https://github.com/lindera-morphology/lindera-tantivy), [Vaporetto](https://crates.io/crates/vaporetto_tantivy), and [tantivy-tokenizer-tiny-segmenter](https://crates.io/crates/tantivy-tokenizer-tiny-segmenter)) and Korean ([lindera](https://github.com/lindera-morphology/lindera-tantivy) + [lindera-ko-dic-builder](https://github.com/lindera-morphology/lindera-ko-dic-builder))
 - Fast (check out the :racehorse: :sparkles: [benchmark](https://tantivy-search.github.io/bench/) :sparkles: :racehorse:)
 - Tiny startup time (<10ms), perfect for command-line tools
 - BM25 scoring (the same as Lucene)
@@ -41,13 +42,13 @@ Your mileage WILL vary depending on the nature of queries and their load.
 - SIMD integer compression when the platform/CPU includes the SSE2 instruction set
 - Single valued and multivalued u64, i64, and f64 fast fields (equivalent of doc values in Lucene)
 - `&[u8]` fast fields
-- Text, i64, u64, f64, dates, ip, bool, and hierarchical facet fields
-- Compressed document store (LZ4, Zstd, None, Brotli, Snap)
+- Text, i64, u64, f64, dates, and hierarchical facet fields
+- LZ4 compressed document store
 - Range queries
 - Faceted search
 - Configurable indexing (optional term frequency and position indexing)
 - JSON Field
-- Aggregation Collector: histogram, range buckets, average, and stats metrics
+- Aggregation Collector: range buckets, average, and stats metrics
 - LogMergePolicy with deletes
 - Searcher Warmer API
 - Cheesy logo with a horse
@@ -56,9 +57,10 @@ 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 and supports Linux, macOS, and Windows.
+Tantivy works on stable Rust (>= 1.27) and supports Linux, macOS, and Windows.
 
 - [Tantivy's simple search example](https://tantivy-search.github.io/examples/basic_search.html)
 - [tantivy-cli and its tutorial](https://github.com/quickwit-oss/tantivy-cli) - `tantivy-cli` is an actual command-line interface that makes it easy for you to create a search engine,
@@ -80,39 +82,63 @@ There are many ways to support this project.
 # Contributing code
 
 We use the GitHub Pull Request workflow: reference a GitHub ticket and/or include a comprehensive commit message when opening a PR.
-Feel free to update CHANGELOG.md with your contribution.
-
-## Tokenizer
-
-When implementing a tokenizer for tantivy depend on the `tantivy-tokenizer-api` crate.
 
 ## Clone and build locally
 
-Tantivy compiles on stable Rust.
+Tantivy compiles on stable Rust but requires `Rust >= 1.27`.
 To check out and run tests, you can simply run:
 
 ```bash
-git clone https://github.com/quickwit-oss/tantivy.git
-cd tantivy
-cargo test
+    git clone https://github.com/quickwit-oss/tantivy.git
+    cd tantivy
+    cargo build
 ```
 
-# Companies Using Tantivy
+## Run tests
+
+Some tests will not run with just `cargo test` because of `fail-rs`.
+To run the tests exhaustively, run `./run-tests.sh`.
+
+## Debug
+
+You might find it useful to step through the programme with a debugger.
+
+### A failing test
+
+Make sure you haven't run `cargo clean` after the most recent `cargo test` or `cargo build` to guarantee that the `target/` directory exists. Use this bash script to find the name of the most recent debug build of Tantivy and run it under `rust-gdb`:
+
+```bash
+find target/debug/ -maxdepth 1 -executable -type f -name "tantivy*" -printf '%TY-%Tm-%Td %TT %p\n' | sort -r | cut -d " " -f 3 | xargs -I RECENT_DBG_TANTIVY rust-gdb RECENT_DBG_TANTIVY
+```
+
+Now that you are in `rust-gdb`, you can set breakpoints on lines and methods that match your source code and run the debug executable with flags that you normally pass to `cargo test` like this:
+
+```bash
+$gdb run --test-threads 1 --test $NAME_OF_TEST
+```
+
+### An example
+
+By default, `rustc` compiles everything in the `examples/` directory in debug mode. This makes it easy for you to make examples to reproduce bugs:
+
+```bash
+rust-gdb target/debug/examples/$EXAMPLE_NAME
+$ gdb run
+```
+# Companies Using Tantivy 
 
 <p align="left">
-<img align="center" src="doc/assets/images/etsy.png" alt="Etsy" height="25" width="auto" />&nbsp;
 <img align="center" src="doc/assets/images/Nuclia.png#gh-light-mode-only" alt="Nuclia" height="25" width="auto" /> &nbsp;
 <img align="center" src="doc/assets/images/humanfirst.png#gh-light-mode-only" alt="Humanfirst.ai" height="30" width="auto" />
 <img align="center" src="doc/assets/images/element.io.svg#gh-light-mode-only" alt="Element.io" height="25" width="auto" />
 <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)
 
@@ -126,17 +152,4 @@ 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`.
+- According to our [search latency benchmark](https://tantivy-search.github.io/bench/), Tantivy is approximately 2x faster than Lucene.

TODO.txt

@@ -1,18 +0,0 @@
-Make schema_builder API fluent.
-fix doc serialization and prevent compression problems
-
-u64 , etc. shoudl return Resutl<Option> now that we support optional missing a column is really not an error
-remove fastfield codecs
-ditch the first_or_default trick. if it is still useful, improve its implementation.
-rename FastFieldReaders::open to load
-
-
-remove fast field reader
-
-find a way to unify the two DateTime.
-readd type check in the filter wrapper
-
-add unit test on columnar list columns.
-
-make sure sort works
-

benches/index-bench.rs

@@ -34,7 +34,7 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
             let index = Index::create_in_ram(schema.clone());
             let index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
             for _ in 0..NUM_REPEATS {
-                for doc_json in HDFS_LOGS.trim().split('\n') {
+                for doc_json in HDFS_LOGS.trim().split("\n") {
                     let doc = schema.parse_document(doc_json).unwrap();
                     index_writer.add_document(doc).unwrap();
                 }
@@ -46,7 +46,7 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
             let index = Index::create_in_ram(schema.clone());
             let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
             for _ in 0..NUM_REPEATS {
-                for doc_json in HDFS_LOGS.trim().split('\n') {
+                for doc_json in HDFS_LOGS.trim().split("\n") {
                     let doc = schema.parse_document(doc_json).unwrap();
                     index_writer.add_document(doc).unwrap();
                 }
@@ -59,7 +59,7 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
             let index = Index::create_in_ram(schema_with_store.clone());
             let index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
             for _ in 0..NUM_REPEATS {
-                for doc_json in HDFS_LOGS.trim().split('\n') {
+                for doc_json in HDFS_LOGS.trim().split("\n") {
                     let doc = schema.parse_document(doc_json).unwrap();
                     index_writer.add_document(doc).unwrap();
                 }
@@ -71,7 +71,7 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
             let index = Index::create_in_ram(schema_with_store.clone());
             let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
             for _ in 0..NUM_REPEATS {
-                for doc_json in HDFS_LOGS.trim().split('\n') {
+                for doc_json in HDFS_LOGS.trim().split("\n") {
                     let doc = schema.parse_document(doc_json).unwrap();
                     index_writer.add_document(doc).unwrap();
                 }
@@ -85,7 +85,7 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
             let json_field = dynamic_schema.get_field("json").unwrap();
             let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
             for _ in 0..NUM_REPEATS {
-                for doc_json in HDFS_LOGS.trim().split('\n') {
+                for doc_json in HDFS_LOGS.trim().split("\n") {
                     let json_val: serde_json::Map<String, serde_json::Value> =
                         serde_json::from_str(doc_json).unwrap();
                     let doc = tantivy::doc!(json_field=>json_val);
@@ -101,7 +101,7 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
             let json_field = dynamic_schema.get_field("json").unwrap();
             let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
             for _ in 0..NUM_REPEATS {
-                for doc_json in HDFS_LOGS.trim().split('\n') {
+                for doc_json in HDFS_LOGS.trim().split("\n") {
                     let json_val: serde_json::Map<String, serde_json::Value> =
                         serde_json::from_str(doc_json).unwrap();
                     let doc = tantivy::doc!(json_field=>json_val);

bitpacker/Cargo.toml

@@ -1,21 +1,15 @@
 [package]
 name = "tantivy-bitpacker"
-version = "0.3.0"
-edition = "2021"
+version = "0.2.0"
+edition = "2018"
 authors = ["Paul Masurel <paul.masurel@gmail.com>"]
 license = "MIT"
 categories = []
 description = """Tantivy-sub crate: bitpacking"""
 repository = "https://github.com/quickwit-oss/tantivy"
 keywords = []
-documentation = "https://docs.rs/tantivy-bitpacker/latest/tantivy_bitpacker"
-homepage = "https://github.com/quickwit-oss/tantivy"
 
 
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-
-[dev-dependencies]
-rand = "0.8"
-proptest = "1"

bitpacker/benches/bench.rs

@@ -4,39 +4,9 @@ extern crate test;
 
 #[cfg(test)]
 mod tests {
-    use rand::seq::IteratorRandom;
-    use rand::thread_rng;
-    use tantivy_bitpacker::{BitPacker, BitUnpacker, BlockedBitpacker};
+    use tantivy_bitpacker::BlockedBitpacker;
     use test::Bencher;
 
-    #[inline(never)]
-    fn create_bitpacked_data(bit_width: u8, num_els: u32) -> Vec<u8> {
-        let mut bitpacker = BitPacker::new();
-        let mut buffer = Vec::new();
-        for _ in 0..num_els {
-            // the values do not matter.
-            bitpacker.write(0u64, bit_width, &mut buffer).unwrap();
-            bitpacker.flush(&mut buffer).unwrap();
-        }
-        buffer
-    }
-
-    #[bench]
-    fn bench_bitpacking_read(b: &mut Bencher) {
-        let bit_width = 3;
-        let num_els = 1_000_000u32;
-        let bit_unpacker = BitUnpacker::new(bit_width);
-        let data = create_bitpacked_data(bit_width, num_els);
-        let idxs: Vec<u32> = (0..num_els).choose_multiple(&mut thread_rng(), 100_000);
-        b.iter(|| {
-            let mut out = 0u64;
-            for &idx in &idxs {
-                out = out.wrapping_add(bit_unpacker.get(idx, &data[..]));
-            }
-            out
-        });
-    }
-
     #[bench]
     fn bench_blockedbitp_read(b: &mut Bencher) {
         let mut blocked_bitpacker = BlockedBitpacker::new();
@@ -44,9 +14,9 @@ mod tests {
             blocked_bitpacker.add(val * val);
         }
         b.iter(|| {
-            let mut out = 0u64;
+            let mut out = 0;
             for val in 0..=21500 {
-                out = out.wrapping_add(blocked_bitpacker.get(val));
+                out = blocked_bitpacker.get(val);
             }
             out
         });

bitpacker/src/bitpacker.rs

@@ -19,20 +19,21 @@ impl BitPacker {
     }
 
     #[inline]
-    pub fn write<TWrite: io::Write + ?Sized>(
+    pub fn write<TWrite: io::Write>(
         &mut self,
         val: u64,
         num_bits: u8,
         output: &mut TWrite,
     ) -> io::Result<()> {
+        let val_u64 = val as u64;
         let num_bits = num_bits as usize;
         if self.mini_buffer_written + num_bits > 64 {
-            self.mini_buffer |= val.wrapping_shl(self.mini_buffer_written as u32);
+            self.mini_buffer |= val_u64.wrapping_shl(self.mini_buffer_written as u32);
             output.write_all(self.mini_buffer.to_le_bytes().as_ref())?;
-            self.mini_buffer = val.wrapping_shr((64 - self.mini_buffer_written) as u32);
+            self.mini_buffer = val_u64.wrapping_shr((64 - self.mini_buffer_written) as u32);
             self.mini_buffer_written = self.mini_buffer_written + num_bits - 64;
         } else {
-            self.mini_buffer |= val << self.mini_buffer_written;
+            self.mini_buffer |= val_u64 << self.mini_buffer_written;
             self.mini_buffer_written += num_bits;
             if self.mini_buffer_written == 64 {
                 output.write_all(self.mini_buffer.to_le_bytes().as_ref())?;
@@ -43,7 +44,7 @@ impl BitPacker {
         Ok(())
     }
 
-    pub fn flush<TWrite: io::Write + ?Sized>(&mut self, output: &mut TWrite) -> io::Result<()> {
+    pub fn flush<TWrite: io::Write>(&mut self, output: &mut TWrite) -> io::Result<()> {
         if self.mini_buffer_written > 0 {
             let num_bytes = (self.mini_buffer_written + 7) / 8;
             let bytes = self.mini_buffer.to_le_bytes();
@@ -54,69 +55,53 @@ impl BitPacker {
         Ok(())
     }
 
-    pub fn close<TWrite: io::Write + ?Sized>(&mut self, output: &mut TWrite) -> io::Result<()> {
+    pub fn close<TWrite: io::Write>(&mut self, output: &mut TWrite) -> io::Result<()> {
         self.flush(output)?;
+        // Padding the write file to simplify reads.
+        output.write_all(&[0u8; 7])?;
         Ok(())
     }
 }
 
-#[derive(Clone, Debug, Default, Copy)]
+#[derive(Clone, Debug, Default)]
 pub struct BitUnpacker {
-    num_bits: u32,
+    num_bits: u64,
     mask: u64,
 }
 
 impl BitUnpacker {
-    /// Creates a bit unpacker, that assumes the same bitwidth for all values.
-    ///
-    /// The bitunpacker works by doing an unaligned read of 8 bytes.
-    /// For this reason, values of `num_bits` between
-    /// [57..63] are forbidden.
     pub fn new(num_bits: u8) -> BitUnpacker {
-        assert!(num_bits <= 7 * 8 || num_bits == 64);
         let mask: u64 = if num_bits == 64 {
             !0u64
         } else {
             (1u64 << num_bits) - 1u64
         };
         BitUnpacker {
-            num_bits: u32::from(num_bits),
+            num_bits: u64::from(num_bits),
             mask,
         }
     }
 
-    pub fn bit_width(&self) -> u8 {
-        self.num_bits as u8
-    }
-
     #[inline]
-    pub fn get(&self, idx: u32, data: &[u8]) -> u64 {
-        let addr_in_bits = idx * self.num_bits;
-        let addr = (addr_in_bits >> 3) as usize;
-        if addr + 8 > data.len() {
-            if self.num_bits == 0 {
-                return 0;
-            }
-            let bit_shift = addr_in_bits & 7;
-            return self.get_slow_path(addr, bit_shift, data);
+    pub fn get(&self, idx: u64, data: &[u8]) -> u64 {
+        if self.num_bits == 0 {
+            return 0u64;
         }
+        let num_bits = self.num_bits;
+        let mask = self.mask;
+        let addr_in_bits = idx * num_bits;
+        let addr = addr_in_bits >> 3;
         let bit_shift = addr_in_bits & 7;
-        let bytes: [u8; 8] = (&data[addr..addr + 8]).try_into().unwrap();
+        debug_assert!(
+            addr + 8 <= data.len() as u64,
+            "The fast field field should have been padded with 7 bytes."
+        );
+        let bytes: [u8; 8] = (&data[(addr as usize)..(addr as usize) + 8])
+            .try_into()
+            .unwrap();
         let val_unshifted_unmasked: u64 = u64::from_le_bytes(bytes);
-        let val_shifted = val_unshifted_unmasked >> bit_shift;
-        val_shifted & self.mask
-    }
-
-    #[inline(never)]
-    fn get_slow_path(&self, addr: usize, bit_shift: u32, data: &[u8]) -> u64 {
-        let mut bytes: [u8; 8] = [0u8; 8];
-        let available_bytes = data.len() - addr;
-        // This function is meant to only be called if we did not have 8 bytes to load.
-        debug_assert!(available_bytes < 8);
-        bytes[..available_bytes].copy_from_slice(&data[addr..]);
-        let val_unshifted_unmasked: u64 = u64::from_le_bytes(bytes);
-        let val_shifted = val_unshifted_unmasked >> bit_shift;
-        val_shifted & self.mask
+        let val_shifted = (val_unshifted_unmasked >> bit_shift) as u64;
+        val_shifted & mask
     }
 }
 
@@ -124,7 +109,7 @@ impl BitUnpacker {
 mod test {
     use super::{BitPacker, BitUnpacker};
 
-    fn create_bitpacker(len: usize, num_bits: u8) -> (BitUnpacker, Vec<u64>, Vec<u8>) {
+    fn create_fastfield_bitpacker(len: usize, num_bits: u8) -> (BitUnpacker, Vec<u64>, Vec<u8>) {
         let mut data = Vec::new();
         let mut bitpacker = BitPacker::new();
         let max_val: u64 = (1u64 << num_bits as u64) - 1u64;
@@ -135,15 +120,15 @@ mod test {
             bitpacker.write(val, num_bits, &mut data).unwrap();
         }
         bitpacker.close(&mut data).unwrap();
-        assert_eq!(data.len(), ((num_bits as usize) * len + 7) / 8);
+        assert_eq!(data.len(), ((num_bits as usize) * len + 7) / 8 + 7);
         let bitunpacker = BitUnpacker::new(num_bits);
         (bitunpacker, vals, data)
     }
 
     fn test_bitpacker_util(len: usize, num_bits: u8) {
-        let (bitunpacker, vals, data) = create_bitpacker(len, num_bits);
+        let (bitunpacker, vals, data) = create_fastfield_bitpacker(len, num_bits);
         for (i, val) in vals.iter().enumerate() {
-            assert_eq!(bitunpacker.get(i as u32, &data), *val);
+            assert_eq!(bitunpacker.get(i as u64, &data), *val);
         }
     }
 
@@ -155,49 +140,4 @@ mod test {
         test_bitpacker_util(6, 14);
         test_bitpacker_util(1000, 14);
     }
-
-    use proptest::prelude::*;
-
-    fn num_bits_strategy() -> impl Strategy<Value = u8> {
-        prop_oneof!(Just(0), Just(1), 2u8..56u8, Just(56), Just(64),)
-    }
-
-    fn vals_strategy() -> impl Strategy<Value = (u8, Vec<u64>)> {
-        (num_bits_strategy(), 0usize..100usize).prop_flat_map(|(num_bits, len)| {
-            let max_val = if num_bits == 64 {
-                u64::MAX
-            } else {
-                (1u64 << num_bits as u32) - 1
-            };
-            let vals = proptest::collection::vec(0..=max_val, len);
-            vals.prop_map(move |vals| (num_bits, vals))
-        })
-    }
-
-    fn test_bitpacker_aux(num_bits: u8, vals: &[u64]) {
-        let mut buffer: Vec<u8> = Vec::new();
-        let mut bitpacker = BitPacker::new();
-        for &val in vals {
-            bitpacker.write(val, num_bits, &mut buffer).unwrap();
-        }
-        bitpacker.flush(&mut buffer).unwrap();
-        assert_eq!(buffer.len(), (vals.len() * num_bits as usize + 7) / 8);
-        let bitunpacker = BitUnpacker::new(num_bits);
-        let max_val = if num_bits == 64 {
-            u64::MAX
-        } else {
-            (1u64 << num_bits) - 1
-        };
-        for (i, val) in vals.iter().copied().enumerate() {
-            assert!(val <= max_val);
-            assert_eq!(bitunpacker.get(i as u32, &buffer), val);
-        }
-    }
-
-    proptest::proptest! {
-        #[test]
-        fn test_bitpacker_proptest((num_bits, vals) in vals_strategy()) {
-            test_bitpacker_aux(num_bits, &vals);
-        }
-    }
 }

bitpacker/src/blocked_bitpacker.rs

@@ -58,10 +58,6 @@ 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![];
@@ -77,14 +73,16 @@ impl BlockedBitpacker {
     pub fn mem_usage(&self) -> usize {
         std::mem::size_of::<BlockedBitpacker>()
             + self.compressed_blocks.capacity()
-            + mem_usage(&self.offset_and_bits)
-            + mem_usage(&self.buffer)
+            + self.offset_and_bits.capacity()
+                * std::mem::size_of_val(&self.offset_and_bits.get(0).cloned().unwrap_or_default())
+            + self.buffer.capacity()
+                * std::mem::size_of_val(&self.buffer.get(0).cloned().unwrap_or_default())
     }
 
     #[inline]
     pub fn add(&mut self, val: u64) {
         self.buffer.push(val);
-        if self.buffer.len() == BLOCK_SIZE {
+        if self.buffer.len() == BLOCK_SIZE as usize {
             self.flush();
         }
     }
@@ -126,11 +124,11 @@ impl BlockedBitpacker {
     }
     #[inline]
     pub fn get(&self, idx: usize) -> u64 {
-        let metadata_pos = idx / BLOCK_SIZE;
-        let pos_in_block = idx % BLOCK_SIZE;
+        let metadata_pos = idx / BLOCK_SIZE as usize;
+        let pos_in_block = idx % BLOCK_SIZE as usize;
         if let Some(metadata) = self.offset_and_bits.get(metadata_pos) {
             let unpacked = BitUnpacker::new(metadata.num_bits()).get(
-                pos_in_block as u32,
+                pos_in_block as u64,
                 &self.compressed_blocks[metadata.offset() as usize..],
             );
             unpacked + metadata.base_value()

bitpacker/src/lib.rs

@@ -1,8 +1,6 @@
 mod bitpacker;
 mod blocked_bitpacker;
 
-use std::cmp::Ordering;
-
 pub use crate::bitpacker::{BitPacker, BitUnpacker};
 pub use crate::blocked_bitpacker::BlockedBitpacker;
 
@@ -39,104 +37,44 @@ pub fn compute_num_bits(n: u64) -> u8 {
     }
 }
 
-/// Computes the (min, max) of an iterator of `PartialOrd` values.
-///
-/// For values implementing `Ord` (in a way consistent to their `PartialOrd` impl),
-/// this function behaves as expected.
-///
-/// For values with partial ordering, the behavior is non-trivial and may
-/// depends on the order of the values.
-/// For floats however, it simply returns the same results as if NaN were
-/// skipped.
 pub fn minmax<I, T>(mut vals: I) -> Option<(T, T)>
 where
     I: Iterator<Item = T>,
-    T: Copy + PartialOrd,
+    T: Copy + Ord,
 {
-    let first_el = vals.find(|val| {
-        // We use this to make sure we skip all NaN values when
-        // working with a float type.
-        val.partial_cmp(val) == Some(Ordering::Equal)
-    })?;
-    let mut min_so_far: T = first_el;
-    let mut max_so_far: T = first_el;
-    for val in vals {
-        if val.partial_cmp(&min_so_far) == Some(Ordering::Less) {
-            min_so_far = val;
-        }
-        if val.partial_cmp(&max_so_far) == Some(Ordering::Greater) {
-            max_so_far = val;
-        }
+    if let Some(first_el) = vals.next() {
+        return Some(vals.fold((first_el, first_el), |(min_val, max_val), el| {
+            (min_val.min(el), max_val.max(el))
+        }));
     }
-    Some((min_so_far, max_so_far))
+    None
 }
 
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_compute_num_bits() {
-        assert_eq!(compute_num_bits(1), 1u8);
-        assert_eq!(compute_num_bits(0), 0u8);
-        assert_eq!(compute_num_bits(2), 2u8);
-        assert_eq!(compute_num_bits(3), 2u8);
-        assert_eq!(compute_num_bits(4), 3u8);
-        assert_eq!(compute_num_bits(255), 8u8);
-        assert_eq!(compute_num_bits(256), 9u8);
-        assert_eq!(compute_num_bits(5_000_000_000), 33u8);
-    }
-
-    #[test]
-    fn test_minmax_empty() {
-        let vals: Vec<u32> = vec![];
-        assert_eq!(minmax(vals.into_iter()), None);
-    }
-
-    #[test]
-    fn test_minmax_one() {
-        assert_eq!(minmax(vec![1].into_iter()), Some((1, 1)));
-    }
-
-    #[test]
-    fn test_minmax_two() {
-        assert_eq!(minmax(vec![1, 2].into_iter()), Some((1, 2)));
-        assert_eq!(minmax(vec![2, 1].into_iter()), Some((1, 2)));
-    }
-
-    #[test]
-    fn test_minmax_nan() {
-        assert_eq!(
-            minmax(vec![f64::NAN, 1f64, 2f64].into_iter()),
-            Some((1f64, 2f64))
-        );
-        assert_eq!(
-            minmax(vec![2f64, f64::NAN, 1f64].into_iter()),
-            Some((1f64, 2f64))
-        );
-        assert_eq!(
-            minmax(vec![2f64, 1f64, f64::NAN].into_iter()),
-            Some((1f64, 2f64))
-        );
-    }
-
-    #[test]
-    fn test_minmax_inf() {
-        assert_eq!(
-            minmax(vec![f64::INFINITY, 1f64, 2f64].into_iter()),
-            Some((1f64, f64::INFINITY))
-        );
-        assert_eq!(
-            minmax(vec![-f64::INFINITY, 1f64, 2f64].into_iter()),
-            Some((-f64::INFINITY, 2f64))
-        );
-        assert_eq!(
-            minmax(vec![2f64, f64::INFINITY, 1f64].into_iter()),
-            Some((1f64, f64::INFINITY))
-        );
-        assert_eq!(
-            minmax(vec![2f64, 1f64, -f64::INFINITY].into_iter()),
-            Some((-f64::INFINITY, 2f64))
-        );
-    }
+#[test]
+fn test_compute_num_bits() {
+    assert_eq!(compute_num_bits(1), 1u8);
+    assert_eq!(compute_num_bits(0), 0u8);
+    assert_eq!(compute_num_bits(2), 2u8);
+    assert_eq!(compute_num_bits(3), 2u8);
+    assert_eq!(compute_num_bits(4), 3u8);
+    assert_eq!(compute_num_bits(255), 8u8);
+    assert_eq!(compute_num_bits(256), 9u8);
+    assert_eq!(compute_num_bits(5_000_000_000), 33u8);
+}
+
+#[test]
+fn test_minmax_empty() {
+    let vals: Vec<u32> = vec![];
+    assert_eq!(minmax(vals.into_iter()), None);
+}
+
+#[test]
+fn test_minmax_one() {
+    assert_eq!(minmax(vec![1].into_iter()), Some((1, 1)));
+}
+
+#[test]
+fn test_minmax_two() {
+    assert_eq!(minmax(vec![1, 2].into_iter()), Some((1, 2)));
+    assert_eq!(minmax(vec![2, 1].into_iter()), Some((1, 2)));
 }

ci/before_deploy.ps1

@@ -0,0 +1,23 @@
+# This script takes care of packaging the build artifacts that will go in the
+# release zipfile
+
+$SRC_DIR = $PWD.Path
+$STAGE = [System.Guid]::NewGuid().ToString()
+
+Set-Location $ENV:Temp
+New-Item -Type Directory -Name $STAGE
+Set-Location $STAGE
+
+$ZIP = "$SRC_DIR\$($Env:CRATE_NAME)-$($Env:APPVEYOR_REPO_TAG_NAME)-$($Env:TARGET).zip"
+
+# TODO Update this to package the right artifacts
+Copy-Item "$SRC_DIR\target\$($Env:TARGET)\release\hello.exe" '.\'
+
+7z a "$ZIP" *
+
+Push-AppveyorArtifact "$ZIP"
+
+Remove-Item *.* -Force
+Set-Location ..
+Remove-Item $STAGE
+Set-Location $SRC_DIR

ci/before_deploy.sh

@@ -0,0 +1,33 @@
+# This script takes care of building your crate and packaging it for release
+
+set -ex
+
+main() {
+    local src=$(pwd) \
+          stage=
+
+    case $TRAVIS_OS_NAME in
+        linux)
+            stage=$(mktemp -d)
+            ;;
+        osx)
+            stage=$(mktemp -d -t tmp)
+            ;;
+    esac
+
+    test -f Cargo.lock || cargo generate-lockfile
+
+    # TODO Update this to build the artifacts that matter to you
+    cross rustc --bin hello --target $TARGET --release -- -C lto
+
+    # TODO Update this to package the right artifacts
+    cp target/$TARGET/release/hello $stage/
+
+    cd $stage
+    tar czf $src/$CRATE_NAME-$TRAVIS_TAG-$TARGET.tar.gz *
+    cd $src
+
+    rm -rf $stage
+}
+
+main

ci/install.sh

@@ -0,0 +1,47 @@
+set -ex
+
+main() {
+    local target=
+    if [ $TRAVIS_OS_NAME = linux ]; then
+        target=x86_64-unknown-linux-musl
+        sort=sort
+    else
+        target=x86_64-apple-darwin
+        sort=gsort  # for `sort --sort-version`, from brew's coreutils.
+    fi
+
+    # Builds for iOS are done on OSX, but require the specific target to be
+    # installed.
+    case $TARGET in
+        aarch64-apple-ios)
+            rustup target install aarch64-apple-ios
+            ;;
+        armv7-apple-ios)
+            rustup target install armv7-apple-ios
+            ;;
+        armv7s-apple-ios)
+            rustup target install armv7s-apple-ios
+            ;;
+        i386-apple-ios)
+            rustup target install i386-apple-ios
+            ;;
+        x86_64-apple-ios)
+            rustup target install x86_64-apple-ios
+            ;;
+    esac
+
+    # This fetches latest stable release
+    local tag=$(git ls-remote --tags --refs --exit-code https://github.com/japaric/cross \
+                       | cut -d/ -f3 \
+                       | grep -E '^v[0.1.0-9.]+$' \
+                       | $sort --version-sort \
+                       | tail -n1)
+    curl -LSfs https://japaric.github.io/trust/install.sh | \
+        sh -s -- \
+           --force \
+           --git japaric/cross \
+           --tag $tag \
+           --target $target
+}
+
+main

ci/script.sh

@@ -0,0 +1,30 @@
+#!/usr/bin/env bash
+
+# This script takes care of testing your crate
+
+set -ex
+
+main() {
+    if [ ! -z $CODECOV ]; then
+        echo "Codecov"
+        cargo build --verbose && cargo coverage --verbose --all && bash <(curl -s https://codecov.io/bash) -s target/kcov
+    else
+        echo "Build"
+        cross build --target $TARGET
+        if [ ! -z $DISABLE_TESTS ]; then
+            return
+        fi
+        echo "Test"
+        cross test --target $TARGET --no-default-features --features mmap
+        cross test --target $TARGET --no-default-features --features mmap query-grammar
+    fi
+    for example in $(ls examples/*.rs)
+    do
+        cargo run --example  $(basename $example .rs)
+    done
+}
+
+# we don't run the "test phase" when doing deploys
+if [ -z $TRAVIS_TAG ]; then
+    main
+fi

columnar/Cargo.toml

@@ -1,28 +0,0 @@
-[package]
-name = "tantivy-columnar"
-version = "0.1.0"
-edition = "2021"
-license = "MIT"
-
-[dependencies]
-itertools = "0.10.5"
-log = "0.4.17"
-fnv = "1.0.7"
-fastdivide = "0.4.0"
-rand = { version = "0.8.5", optional = true }
-measure_time = { version = "0.8.2", optional = true }
-prettytable-rs = { version = "0.10.0", optional = true }
-
-stacker = { path = "../stacker", package="tantivy-stacker"}
-sstable = { path = "../sstable", package = "tantivy-sstable" }
-common = { path = "../common", package = "tantivy-common" }
-tantivy-bitpacker = { version= "0.3", path = "../bitpacker/" }
-serde = "1.0.152"
-
-[dev-dependencies]
-proptest = "1"
-more-asserts = "0.3.1"
-rand = "0.8.5"
-
-[features]
-unstable = []

columnar/README.md

@@ -1,109 +0,0 @@
-# Columnar format
-
-This crate describes columnar format used in tantivy.
-
-## Goals
-
-This format is special in the following way.
-- it needs to be compact
-- accessing a specific column does not require to load the entire columnar. It can be done in 2 to 3 random access.
-- columns of several types can be associated with the same column name.
-- it needs to support columns with different types `(str, u64, i64, f64)`
-and different cardinality `(required, optional, multivalued)`.
-- columns, once loaded, offer cheap random access.
-- it is designed to allow range queries.
-
-# Coercion rules
-
-Users can create a columnar by inserting rows to a `ColumnarWriter`,
-and serializing it into a `Write` object.
-Nothing prevents a user from recording values with different type to the same `column_name`.
-
-In that case, `tantivy-columnar`'s behavior is as follows:
-- JsonValues are grouped into 3 types (String, Number, bool).
-Values that corresponds to different groups are mapped to different columns. For instance, String values are treated independently
-from Number or boolean values. `tantivy-columnar` will simply emit several columns associated to a given column_name.
-- Only one column for a given json value type is emitted.  If number values with different number types are recorded (e.g. u64, i64, f64),
-`tantivy-columnar` will pick the first type that can represents the set of appended value, with the following prioriy order (`i64`, `u64`, `f64`).
-`i64` is picked over `u64` as it is likely to  yield less change of types. Most use cases strictly requiring `u64` show the
-restriction on 50% of the values (e.g. a 64-bit hash). On the other hand, a lot of use cases can show rare negative value.
-
-# Columnar format
-
-This columnar format may have more than one column (with different types) associated to the same `column_name` (see [Coercion rules](#coercion-rules) above).
-The `(column_name, columne_type)` couple however uniquely identifies a column.
-That couple is serialized as a column `column_key`.  The format of that key is:
-`[column_name][ZERO_BYTE][column_type_header: u8]`
-
-```
-COLUMNAR:=
-    [COLUMNAR_DATA]
-    [COLUMNAR_KEY_TO_DATA_INDEX]
-    [COLUMNAR_FOOTER];
-
-
-# Columns are sorted by their column key.
-COLUMNAR_DATA:=
-    [COLUMN_DATA]+;
-
-COLUMNAR_FOOTER := [RANGE_SSTABLE_BYTES_LEN: 8 bytes little endian]
-
-```
-
-The columnar file starts by the actual column data, concatenated one after the other,
-sorted by column key.
-
-A sstable associates
-`(column name, column_cardinality, column_type) to range of bytes.
-
-Column name may not contain the zero byte `\0`.
-
-Listing all columns associated to `column_name` can therefore
-be done by listing all keys prefixed by
-`[column_name][ZERO_BYTE]`
-
-The associated range of bytes refer to a range of bytes
-
-This crate exposes a columnar format for tantivy.
-This format is described in README.md
-
-
-The crate introduces the following concepts.
-
-`Columnar` is an equivalent of a dataframe.
-It maps `column_key` to `Column`.
-
-A `Column<T>` asssociates a `RowId` (u32) to any
-number of values.
-
-This is made possible by wrapping a `ColumnIndex` and a `ColumnValue` object.
-The `ColumnValue<T>` represents a mapping that associates each `RowId` to
-exactly one single value.
-
-The `ColumnIndex` then maps each RowId to a set of `RowId` in the
-`ColumnValue`.
-
-For optimization, and compression purposes, the `ColumnIndex` has three
-possible representation, each for different cardinalities.
-
-- Full
-
-All RowId have exactly one value. The ColumnIndex is the trivial mapping.
-
-- Optional
-
-All RowIds can have at most one value. The ColumnIndex is the trivial mapping `ColumnRowId -> Option<ColumnValueRowId>`.
-
-- Multivalued
-
-All RowIds can have any number of values.
-The column index is mapping values to a range.
-
-
-All these objects are implemented an unit tested independently
-in their own module:
-
-- columnar
-- column_index
-- column_values
-- column

columnar/benches/bench_u128.rs

@@ -1,124 +0,0 @@
-#![feature(test)]
-
-use std::ops::RangeInclusive;
-use std::sync::Arc;
-
-use common::OwnedBytes;
-use rand::rngs::StdRng;
-use rand::seq::SliceRandom;
-use rand::{random, Rng, SeedableRng};
-use tantivy_columnar::ColumnValues;
-use test::Bencher;
-extern crate test;
-
-// TODO does this make sense for IPv6 ?
-fn generate_random() -> Vec<u64> {
-    let mut permutation: Vec<u64> = (0u64..100_000u64)
-        .map(|el| el + random::<u16>() as u64)
-        .collect();
-    permutation.shuffle(&mut StdRng::from_seed([1u8; 32]));
-    permutation
-}
-
-fn get_u128_column_random() -> Arc<dyn ColumnValues<u128>> {
-    let permutation = generate_random();
-    let permutation = permutation.iter().map(|el| *el as u128).collect::<Vec<_>>();
-    get_u128_column_from_data(&permutation)
-}
-
-fn get_u128_column_from_data(data: &[u128]) -> Arc<dyn ColumnValues<u128>> {
-    let mut out = vec![];
-    tantivy_columnar::column_values::serialize_column_values_u128(&data, &mut out).unwrap();
-    let out = OwnedBytes::new(out);
-    tantivy_columnar::column_values::open_u128_mapped::<u128>(out).unwrap()
-}
-
-const FIFTY_PERCENT_RANGE: RangeInclusive<u64> = 1..=50;
-const SINGLE_ITEM: u64 = 90;
-const SINGLE_ITEM_RANGE: RangeInclusive<u64> = 90..=90;
-
-fn get_data_50percent_item() -> Vec<u128> {
-    let mut rng = StdRng::from_seed([1u8; 32]);
-
-    let mut data = vec![];
-    for _ in 0..300_000 {
-        let val = rng.gen_range(1..=100);
-        data.push(val);
-    }
-    data.push(SINGLE_ITEM);
-    data.shuffle(&mut rng);
-    let data = data.iter().map(|el| *el as u128).collect::<Vec<_>>();
-    data
-}
-
-#[bench]
-fn bench_intfastfield_getrange_u128_50percent_hit(b: &mut Bencher) {
-    let data = get_data_50percent_item();
-    let column = get_u128_column_from_data(&data);
-
-    b.iter(|| {
-        let mut positions = Vec::new();
-        column.get_row_ids_for_value_range(
-            *FIFTY_PERCENT_RANGE.start() as u128..=*FIFTY_PERCENT_RANGE.end() as u128,
-            0..data.len() as u32,
-            &mut positions,
-        );
-        positions
-    });
-}
-
-#[bench]
-fn bench_intfastfield_getrange_u128_single_hit(b: &mut Bencher) {
-    let data = get_data_50percent_item();
-    let column = get_u128_column_from_data(&data);
-
-    b.iter(|| {
-        let mut positions = Vec::new();
-        column.get_row_ids_for_value_range(
-            *SINGLE_ITEM_RANGE.start() as u128..=*SINGLE_ITEM_RANGE.end() as u128,
-            0..data.len() as u32,
-            &mut positions,
-        );
-        positions
-    });
-}
-
-#[bench]
-fn bench_intfastfield_getrange_u128_hit_all(b: &mut Bencher) {
-    let data = get_data_50percent_item();
-    let column = get_u128_column_from_data(&data);
-
-    b.iter(|| {
-        let mut positions = Vec::new();
-        column.get_row_ids_for_value_range(0..=u128::MAX, 0..data.len() as u32, &mut positions);
-        positions
-    });
-}
-// U128 RANGE END
-
-#[bench]
-fn bench_intfastfield_scan_all_fflookup_u128(b: &mut Bencher) {
-    let column = get_u128_column_random();
-
-    b.iter(|| {
-        let mut a = 0u128;
-        for i in 0u64..column.num_vals() as u64 {
-            a += column.get_val(i as u32);
-        }
-        a
-    });
-}
-
-#[bench]
-fn bench_intfastfield_jumpy_stride5_u128(b: &mut Bencher) {
-    let column = get_u128_column_random();
-
-    b.iter(|| {
-        let n = column.num_vals();
-        let mut a = 0u128;
-        for i in (0..n / 5).map(|val| val * 5) {
-            a += column.get_val(i);
-        }
-        a
-    });
-}

columnar/benches/bench_u64.rs

@@ -1,211 +0,0 @@
-#![feature(test)]
-extern crate test;
-
-use std::ops::RangeInclusive;
-use std::sync::Arc;
-
-use rand::prelude::*;
-use tantivy_columnar::column_values::{serialize_and_load_u64_based_column_values, CodecType};
-use tantivy_columnar::*;
-use test::Bencher;
-
-// Warning: this generates the same permutation at each call
-fn generate_permutation() -> Vec<u64> {
-    let mut permutation: Vec<u64> = (0u64..100_000u64).collect();
-    permutation.shuffle(&mut StdRng::from_seed([1u8; 32]));
-    permutation
-}
-
-fn generate_random() -> Vec<u64> {
-    let mut permutation: Vec<u64> = (0u64..100_000u64)
-        .map(|el| el + random::<u16>() as u64)
-        .collect();
-    permutation.shuffle(&mut StdRng::from_seed([1u8; 32]));
-    permutation
-}
-
-// Warning: this generates the same permutation at each call
-fn generate_permutation_gcd() -> Vec<u64> {
-    let mut permutation: Vec<u64> = (1u64..100_000u64).map(|el| el * 1000).collect();
-    permutation.shuffle(&mut StdRng::from_seed([1u8; 32]));
-    permutation
-}
-
-pub fn serialize_and_load(column: &[u64], codec_type: CodecType) -> Arc<dyn ColumnValues<u64>> {
-    serialize_and_load_u64_based_column_values(&column, &[codec_type])
-}
-
-#[bench]
-fn bench_intfastfield_jumpy_veclookup(b: &mut Bencher) {
-    let permutation = generate_permutation();
-    let n = permutation.len();
-    b.iter(|| {
-        let mut a = 0u64;
-        for _ in 0..n {
-            a = permutation[a as usize];
-        }
-        a
-    });
-}
-
-#[bench]
-fn bench_intfastfield_jumpy_fflookup_bitpacked(b: &mut Bencher) {
-    let permutation = generate_permutation();
-    let n = permutation.len();
-    let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&permutation, CodecType::Bitpacked);
-    b.iter(|| {
-        let mut a = 0u64;
-        for _ in 0..n {
-            a = column.get_val(a as u32);
-        }
-        a
-    });
-}
-
-const FIFTY_PERCENT_RANGE: RangeInclusive<u64> = 1..=50;
-const SINGLE_ITEM: u64 = 90;
-const SINGLE_ITEM_RANGE: RangeInclusive<u64> = 90..=90;
-const ONE_PERCENT_ITEM_RANGE: RangeInclusive<u64> = 49..=49;
-fn get_data_50percent_item() -> Vec<u128> {
-    let mut rng = StdRng::from_seed([1u8; 32]);
-
-    let mut data = vec![];
-    for _ in 0..300_000 {
-        let val = rng.gen_range(1..=100);
-        data.push(val);
-    }
-    data.push(SINGLE_ITEM);
-
-    data.shuffle(&mut rng);
-    let data = data.iter().map(|el| *el as u128).collect::<Vec<_>>();
-    data
-}
-
-// U64 RANGE START
-#[bench]
-fn bench_intfastfield_getrange_u64_50percent_hit(b: &mut Bencher) {
-    let data = get_data_50percent_item();
-    let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
-    let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&data, CodecType::Bitpacked);
-    b.iter(|| {
-        let mut positions = Vec::new();
-        column.get_row_ids_for_value_range(
-            FIFTY_PERCENT_RANGE,
-            0..data.len() as u32,
-            &mut positions,
-        );
-        positions
-    });
-}
-
-#[bench]
-fn bench_intfastfield_getrange_u64_1percent_hit(b: &mut Bencher) {
-    let data = get_data_50percent_item();
-    let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
-    let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&data, CodecType::Bitpacked);
-
-    b.iter(|| {
-        let mut positions = Vec::new();
-        column.get_row_ids_for_value_range(
-            ONE_PERCENT_ITEM_RANGE,
-            0..data.len() as u32,
-            &mut positions,
-        );
-        positions
-    });
-}
-
-#[bench]
-fn bench_intfastfield_getrange_u64_single_hit(b: &mut Bencher) {
-    let data = get_data_50percent_item();
-    let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
-    let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&data, CodecType::Bitpacked);
-
-    b.iter(|| {
-        let mut positions = Vec::new();
-        column.get_row_ids_for_value_range(SINGLE_ITEM_RANGE, 0..data.len() as u32, &mut positions);
-        positions
-    });
-}
-
-#[bench]
-fn bench_intfastfield_getrange_u64_hit_all(b: &mut Bencher) {
-    let data = get_data_50percent_item();
-    let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
-    let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&data, CodecType::Bitpacked);
-
-    b.iter(|| {
-        let mut positions = Vec::new();
-        column.get_row_ids_for_value_range(0..=u64::MAX, 0..data.len() as u32, &mut positions);
-        positions
-    });
-}
-// U64 RANGE END
-
-#[bench]
-fn bench_intfastfield_stride7_vec(b: &mut Bencher) {
-    let permutation = generate_permutation();
-    let n = permutation.len();
-    b.iter(|| {
-        let mut a = 0u64;
-        for i in (0..n / 7).map(|val| val * 7) {
-            a += permutation[i as usize];
-        }
-        a
-    });
-}
-
-#[bench]
-fn bench_intfastfield_stride7_fflookup(b: &mut Bencher) {
-    let permutation = generate_permutation();
-    let n = permutation.len();
-    let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&permutation, CodecType::Bitpacked);
-    b.iter(|| {
-        let mut a = 0;
-        for i in (0..n / 7).map(|val| val * 7) {
-            a += column.get_val(i as u32);
-        }
-        a
-    });
-}
-
-#[bench]
-fn bench_intfastfield_scan_all_fflookup(b: &mut Bencher) {
-    let permutation = generate_permutation();
-    let n = permutation.len();
-    let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&permutation, CodecType::Bitpacked);
-    let column_ref = column.as_ref();
-    b.iter(|| {
-        let mut a = 0u64;
-        for i in 0u32..n as u32 {
-            a += column_ref.get_val(i);
-        }
-        a
-    });
-}
-
-#[bench]
-fn bench_intfastfield_scan_all_fflookup_gcd(b: &mut Bencher) {
-    let permutation = generate_permutation_gcd();
-    let n = permutation.len();
-    let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&permutation, CodecType::Bitpacked);
-    b.iter(|| {
-        let mut a = 0u64;
-        for i in 0..n {
-            a += column.get_val(i as u32);
-        }
-        a
-    });
-}
-
-#[bench]
-fn bench_intfastfield_scan_all_vec(b: &mut Bencher) {
-    let permutation = generate_permutation();
-    b.iter(|| {
-        let mut a = 0u64;
-        for i in 0..permutation.len() {
-            a += permutation[i as usize] as u64;
-        }
-        a
-    });
-}

columnar/columnar-cli/Cargo.toml

@@ -1,17 +0,0 @@
-[package]
-name = "tantivy-columnar-cli"
-version = "0.1.0"
-edition = "2021"
-license = "MIT"
-
-[dependencies]
-columnar = {path="../", package="tantivy-columnar"}
-serde_json = "1"
-serde_json_borrow = {git="https://github.com/PSeitz/serde_json_borrow/"}
-serde = "1"
-
-[workspace]
-members = []
-
-[profile.release]
-debug = true

columnar/columnar-cli/src/main.rs

@@ -1,134 +0,0 @@
-use columnar::ColumnarWriter;
-use columnar::NumericalValue;
-use serde_json_borrow;
-use std::fs::File;
-use std::io;
-use std::io::BufRead;
-use std::io::BufReader;
-use std::time::Instant;
-
-#[derive(Default)]
-struct JsonStack {
-    path: String,
-    stack: Vec<usize>,
-}
-
-impl JsonStack {
-    fn push(&mut self, seg: &str) {
-        let len = self.path.len();
-        self.stack.push(len);
-        self.path.push('.');
-        self.path.push_str(seg);
-    }
-
-    fn pop(&mut self) {
-        if let Some(len) = self.stack.pop() {
-            self.path.truncate(len);
-        }
-    }
-
-    fn path(&self) -> &str {
-        &self.path[1..]
-    }
-}
-
-fn append_json_to_columnar(
-    doc: u32,
-    json_value: &serde_json_borrow::Value,
-    columnar: &mut ColumnarWriter,
-    stack: &mut JsonStack,
-) -> usize {
-    let mut count = 0;
-    match json_value {
-        serde_json_borrow::Value::Null => {}
-        serde_json_borrow::Value::Bool(val) => {
-            columnar.record_numerical(
-                doc,
-                stack.path(),
-                NumericalValue::from(if *val { 1u64 } else { 0u64 }),
-            );
-            count += 1;
-        }
-        serde_json_borrow::Value::Number(num) => {
-            let numerical_value: NumericalValue = if let Some(num_i64) = num.as_i64() {
-                num_i64.into()
-            } else if let Some(num_u64) = num.as_u64() {
-                num_u64.into()
-            } else if let Some(num_f64) = num.as_f64() {
-                num_f64.into()
-            } else {
-                panic!();
-            };
-            count += 1;
-            columnar.record_numerical(
-                doc,
-                stack.path(),
-                numerical_value,
-            );
-        }
-        serde_json_borrow::Value::Str(msg) => {
-            columnar.record_str(
-                doc,
-                stack.path(),
-                msg,
-            );
-            count += 1;
-        },
-        serde_json_borrow::Value::Array(vals) => {
-            for val in vals {
-                count += append_json_to_columnar(doc, val, columnar, stack);
-            }
-        },
-        serde_json_borrow::Value::Object(json_map) => {
-            for (child_key, child_val) in json_map {
-                stack.push(child_key);
-                count += append_json_to_columnar(doc, child_val, columnar, stack);
-                stack.pop();
-            }
-        },
-    }
-    count
-}
-
-fn main() -> io::Result<()> {
-    let file = File::open("gh_small.json")?;
-    let mut reader = BufReader::new(file);
-    let mut line = String::with_capacity(100);
-    let mut columnar = columnar::ColumnarWriter::default();
-    let mut doc = 0;
-    let start = Instant::now();
-    let mut stack = JsonStack::default();
-    let mut total_count = 0;
-
-    let start_build = Instant::now();
-    loop {
-        line.clear();
-        let len = reader.read_line(&mut line)?;
-        if len == 0 {
-            break;
-        }
-        let Ok(json_value) = serde_json::from_str::<serde_json_borrow::Value>(&line) else { continue; };
-        total_count += append_json_to_columnar(doc, &json_value, &mut columnar, &mut stack);
-        doc += 1;
-    }
-    println!("Build in {:?}", start_build.elapsed());
-
-    println!("value count {total_count}");
-
-    let mut buffer = Vec::new();
-    let start_serialize = Instant::now();
-    columnar.serialize(doc, None, &mut buffer)?;
-    println!("Serialized in {:?}", start_serialize.elapsed());
-    println!("num docs: {doc}, {:?}", start.elapsed());
-    println!("buffer len {} MB", buffer.len() / 1_000_000);
-    let columnar = columnar::ColumnarReader::open(buffer)?;
-    for (column_name, dynamic_column) in columnar.list_columns()? {
-        let num_bytes = dynamic_column.num_bytes();
-        let typ = dynamic_column.column_type();
-        if num_bytes > 1_000_000 {
-            println!("{column_name} {typ:?}  {} KB", num_bytes / 1_000);
-        }
-    }
-    println!("{} columns", columnar.num_columns());
-    Ok(())
-}

columnar/src/TODO.md

@@ -1,47 +0,0 @@
-# zero to one
-
-* revisit line codec
-* add columns from schema on merge
-* Plugging JSON
-* replug examples
-* move datetime to quickwit common
-* switch to nanos
-* reintroduce the gcd map.
-
-# Perf and Size
-* remove alloc in `ord_to_term`
-+ multivaued range queries restrat frm the beginning all of the time.
-* re-add ZSTD compression for dictionaries
-no systematic monotonic mapping
-consider removing multilinear
-f32?
-adhoc solution for bool?
-add metrics helper for aggregate. sum(row_id)
-review inline absence/presence
-improv perf of select using PDEP
-compare with roaring bitmap/elias fano etc etc.
-SIMD range? (see blog post)
-Add alignment?
-Consider another codec to bridge the gap between few and 5k elements
-
-# Cleanup and rationalization
-in benchmark, unify percent vs ratio, f32 vs f64.
-investigate if should have better errors? io::Error is overused at the moment.
-rename rank/select in unit tests
-Review the public API via cargo doc
-go through TODOs
-remove all  doc_id occurences -> row_id
-use the rank & select naming in unit tests branch.
-multi-linear -> blockwise
-linear codec -> simply a multiplication for the index column
-rename columnar to something more explicit, like column_dictionary or columnar_table
-rename fastfield -> column
-document changes
-rationalization FastFieldValue, HasColumnType
-isolate u128_based and uniform naming
-
-# Other
-fix enhance column-cli
-
-# Santa claus
-autodetect datetime ipaddr, plug customizable tokenizer.

columnar/src/column/dictionary_encoded.rs

@@ -1,100 +0,0 @@
-use std::io;
-use std::ops::Deref;
-use std::sync::Arc;
-
-use sstable::{Dictionary, VoidSSTable};
-
-use crate::column::Column;
-use crate::RowId;
-
-/// Dictionary encoded column.
-///
-/// The column simply gives access to a regular u64-column that, in
-/// which the values are term-ordinals.
-///
-/// These ordinals are ids uniquely identify the bytes that are stored in
-/// the column. These ordinals are small, and sorted in the same order
-/// as the term_ord_column.
-#[derive(Clone)]
-pub struct BytesColumn {
-    pub(crate) dictionary: Arc<Dictionary<VoidSSTable>>,
-    pub(crate) term_ord_column: Column<u64>,
-}
-
-impl BytesColumn {
-    /// Fills the given `output` buffer with the term associated to the ordinal `ord`.
-    ///
-    /// Returns `false` if the term does not exist (e.g. `term_ord` is greater or equal to the
-    /// overll number of terms).
-    pub fn ord_to_bytes(&self, ord: u64, output: &mut Vec<u8>) -> io::Result<bool> {
-        self.dictionary.ord_to_term(ord, output)
-    }
-
-    /// Returns the number of rows in the column.
-    pub fn num_rows(&self) -> RowId {
-        self.term_ord_column.num_docs()
-    }
-
-    pub fn term_ords(&self, row_id: RowId) -> impl Iterator<Item = u64> + '_ {
-        self.term_ord_column.values_for_doc(row_id)
-    }
-
-    /// Returns the column of ordinals
-    pub fn ords(&self) -> &Column<u64> {
-        &self.term_ord_column
-    }
-
-    pub fn num_terms(&self) -> usize {
-        self.dictionary.num_terms()
-    }
-
-    pub fn dictionary(&self) -> &Dictionary<VoidSSTable> {
-        self.dictionary.as_ref()
-    }
-}
-
-#[derive(Clone)]
-pub struct StrColumn(BytesColumn);
-
-impl From<StrColumn> for BytesColumn {
-    fn from(str_column: StrColumn) -> BytesColumn {
-        str_column.0
-    }
-}
-
-impl StrColumn {
-    pub(crate) fn wrap(bytes_column: BytesColumn) -> StrColumn {
-        StrColumn(bytes_column)
-    }
-
-    pub fn dictionary(&self) -> &Dictionary<VoidSSTable> {
-        self.0.dictionary.as_ref()
-    }
-
-    /// Fills the buffer
-    pub fn ord_to_str(&self, term_ord: u64, output: &mut String) -> io::Result<bool> {
-        unsafe {
-            let buf = output.as_mut_vec();
-            if !self.0.dictionary.ord_to_term(term_ord, buf)? {
-                return Ok(false);
-            }
-            // TODO consider remove checks if it hurts performance.
-            if std::str::from_utf8(buf.as_slice()).is_err() {
-                buf.clear();
-                return Err(io::Error::new(
-                    io::ErrorKind::InvalidData,
-                    "Not valid utf-8",
-                ));
-            }
-        }
-        Ok(true)
-    }
-}
-
-impl Deref for StrColumn {
-    type Target = BytesColumn;
-
-    fn deref(&self) -> &Self::Target {
-        &self.0
-    }
-}

columnar/src/column/mod.rs

@@ -1,161 +0,0 @@
-mod dictionary_encoded;
-mod serialize;
-
-use std::fmt::Debug;
-use std::io::Write;
-use std::ops::{Deref, Range, RangeInclusive};
-use std::sync::Arc;
-
-use common::BinarySerializable;
-pub use dictionary_encoded::{BytesColumn, StrColumn};
-pub use serialize::{
-    open_column_bytes, open_column_str, open_column_u128, open_column_u64,
-    serialize_column_mappable_to_u128, serialize_column_mappable_to_u64,
-};
-
-use crate::column_index::ColumnIndex;
-use crate::column_values::monotonic_mapping::StrictlyMonotonicMappingToInternal;
-use crate::column_values::{monotonic_map_column, ColumnValues};
-use crate::{Cardinality, MonotonicallyMappableToU64, RowId};
-
-#[derive(Clone)]
-pub struct Column<T = u64> {
-    pub idx: ColumnIndex,
-    pub values: Arc<dyn ColumnValues<T>>,
-}
-
-impl<T: MonotonicallyMappableToU64> Column<T> {
-    pub fn to_u64_monotonic(self) -> Column<u64> {
-        let values = Arc::new(monotonic_map_column(
-            self.values,
-            StrictlyMonotonicMappingToInternal::<T>::new(),
-        ));
-        Column {
-            idx: self.idx,
-            values,
-        }
-    }
-}
-
-impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
-    #[inline]
-    pub fn get_cardinality(&self) -> Cardinality {
-        self.idx.get_cardinality()
-    }
-
-    pub fn num_docs(&self) -> RowId {
-        match &self.idx {
-            ColumnIndex::Empty { num_docs } => *num_docs,
-            ColumnIndex::Full => self.values.num_vals(),
-            ColumnIndex::Optional(optional_index) => optional_index.num_docs(),
-            ColumnIndex::Multivalued(col_index) => {
-                // The multivalued index contains all value start row_id,
-                // and one extra value at the end with the overall number of rows.
-                col_index.num_docs()
-            }
-        }
-    }
-
-    pub fn min_value(&self) -> T {
-        self.values.min_value()
-    }
-
-    pub fn max_value(&self) -> T {
-        self.values.max_value()
-    }
-
-    pub fn first(&self, row_id: RowId) -> Option<T> {
-        self.values_for_doc(row_id).next()
-    }
-
-    pub fn values_for_doc(&self, row_id: RowId) -> impl Iterator<Item = T> + '_ {
-        self.value_row_ids(row_id)
-            .map(|value_row_id: RowId| self.values.get_val(value_row_id))
-    }
-
-    /// Get the docids of values which are in the provided value range.
-    #[inline]
-    pub fn get_docids_for_value_range(
-        &self,
-        value_range: RangeInclusive<T>,
-        selected_docid_range: Range<u32>,
-        doc_ids: &mut Vec<u32>,
-    ) {
-        // convert passed docid range to row id range
-        let rowid_range = self.idx.docid_range_to_rowids(selected_docid_range.clone());
-
-        // Load rows
-        self.values
-            .get_row_ids_for_value_range(value_range, rowid_range, doc_ids);
-        // Convert rows to docids
-        self.idx
-            .select_batch_in_place(selected_docid_range.start, doc_ids);
-    }
-
-    /// Fils the output vector with the (possibly multiple values that are associated_with
-    /// `row_id`.
-    ///
-    /// This method clears the `output` vector.
-    pub fn fill_vals(&self, row_id: RowId, output: &mut Vec<T>) {
-        output.clear();
-        output.extend(self.values_for_doc(row_id));
-    }
-
-    pub fn first_or_default_col(self, default_value: T) -> Arc<dyn ColumnValues<T>> {
-        Arc::new(FirstValueWithDefault {
-            column: self,
-            default_value,
-        })
-    }
-}
-
-impl<T> Deref for Column<T> {
-    type Target = ColumnIndex;
-
-    fn deref(&self) -> &Self::Target {
-        &self.idx
-    }
-}
-
-impl BinarySerializable for Cardinality {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> std::io::Result<()> {
-        self.to_code().serialize(writer)
-    }
-
-    fn deserialize<R: std::io::Read>(reader: &mut R) -> std::io::Result<Self> {
-        let cardinality_code = u8::deserialize(reader)?;
-        let cardinality = Cardinality::try_from_code(cardinality_code)?;
-        Ok(cardinality)
-    }
-}
-
-// TODO simplify or optimize
-struct FirstValueWithDefault<T: Copy> {
-    column: Column<T>,
-    default_value: T,
-}
-
-impl<T: PartialOrd + Debug + Send + Sync + Copy + 'static> ColumnValues<T>
-    for FirstValueWithDefault<T>
-{
-    fn get_val(&self, idx: u32) -> T {
-        self.column.first(idx).unwrap_or(self.default_value)
-    }
-
-    fn min_value(&self) -> T {
-        self.column.values.min_value()
-    }
-
-    fn max_value(&self) -> T {
-        self.column.values.max_value()
-    }
-
-    fn num_vals(&self) -> u32 {
-        match &self.column.idx {
-            ColumnIndex::Empty { .. } => 0u32,
-            ColumnIndex::Full => self.column.values.num_vals(),
-            ColumnIndex::Optional(optional_idx) => optional_idx.num_docs(),
-            ColumnIndex::Multivalued(multivalue_idx) => multivalue_idx.num_docs(),
-        }
-    }
-}

columnar/src/column/serialize.rs

@@ -1,94 +0,0 @@
-use std::io;
-use std::io::Write;
-use std::sync::Arc;
-
-use common::OwnedBytes;
-use sstable::Dictionary;
-
-use crate::column::{BytesColumn, Column};
-use crate::column_index::{serialize_column_index, SerializableColumnIndex};
-use crate::column_values::{
-    load_u64_based_column_values, serialize_column_values_u128, serialize_u64_based_column_values,
-    CodecType, MonotonicallyMappableToU128, MonotonicallyMappableToU64,
-};
-use crate::iterable::Iterable;
-use crate::StrColumn;
-
-pub fn serialize_column_mappable_to_u128<T: MonotonicallyMappableToU128>(
-    column_index: SerializableColumnIndex<'_>,
-    iterable: &dyn Iterable<T>,
-    output: &mut impl Write,
-) -> io::Result<()> {
-    let column_index_num_bytes = serialize_column_index(column_index, output)?;
-    serialize_column_values_u128(iterable, output)?;
-    output.write_all(&column_index_num_bytes.to_le_bytes())?;
-    Ok(())
-}
-
-pub fn serialize_column_mappable_to_u64<T: MonotonicallyMappableToU64>(
-    column_index: SerializableColumnIndex<'_>,
-    column_values: &impl Iterable<T>,
-    output: &mut impl Write,
-) -> io::Result<()> {
-    let column_index_num_bytes = serialize_column_index(column_index, output)?;
-    serialize_u64_based_column_values(
-        column_values,
-        &[CodecType::Bitpacked, CodecType::BlockwiseLinear],
-        output,
-    )?;
-    output.write_all(&column_index_num_bytes.to_le_bytes())?;
-    Ok(())
-}
-
-pub fn open_column_u64<T: MonotonicallyMappableToU64>(bytes: OwnedBytes) -> io::Result<Column<T>> {
-    let (body, column_index_num_bytes_payload) = bytes.rsplit(4);
-    let column_index_num_bytes = u32::from_le_bytes(
-        column_index_num_bytes_payload
-            .as_slice()
-            .try_into()
-            .unwrap(),
-    );
-    let (column_index_data, column_values_data) = body.split(column_index_num_bytes as usize);
-    let column_index = crate::column_index::open_column_index(column_index_data)?;
-    let column_values = load_u64_based_column_values(column_values_data)?;
-    Ok(Column {
-        idx: column_index,
-        values: column_values,
-    })
-}
-
-pub fn open_column_u128<T: MonotonicallyMappableToU128>(
-    bytes: OwnedBytes,
-) -> io::Result<Column<T>> {
-    let (body, column_index_num_bytes_payload) = bytes.rsplit(4);
-    let column_index_num_bytes = u32::from_le_bytes(
-        column_index_num_bytes_payload
-            .as_slice()
-            .try_into()
-            .unwrap(),
-    );
-    let (column_index_data, column_values_data) = body.split(column_index_num_bytes as usize);
-    let column_index = crate::column_index::open_column_index(column_index_data)?;
-    let column_values = crate::column_values::open_u128_mapped(column_values_data)?;
-    Ok(Column {
-        idx: column_index,
-        values: column_values,
-    })
-}
-
-pub fn open_column_bytes(data: OwnedBytes) -> io::Result<BytesColumn> {
-    let (body, dictionary_len_bytes) = data.rsplit(4);
-    let dictionary_len = u32::from_le_bytes(dictionary_len_bytes.as_slice().try_into().unwrap());
-    let (dictionary_bytes, column_bytes) = body.split(dictionary_len as usize);
-    let dictionary = Arc::new(Dictionary::from_bytes(dictionary_bytes)?);
-    let term_ord_column = crate::column::open_column_u64::<u64>(column_bytes)?;
-    Ok(BytesColumn {
-        dictionary,
-        term_ord_column,
-    })
-}
-
-pub fn open_column_str(data: OwnedBytes) -> io::Result<StrColumn> {
-    let bytes_column = open_column_bytes(data)?;
-    Ok(StrColumn::wrap(bytes_column))
-}

columnar/src/column_index/merge/mod.rs

@@ -1,136 +0,0 @@
-mod shuffled;
-mod stacked;
-
-use shuffled::merge_column_index_shuffled;
-use stacked::merge_column_index_stacked;
-
-use crate::column_index::SerializableColumnIndex;
-use crate::{Cardinality, ColumnIndex, MergeRowOrder};
-
-// For simplification, we never have cardinality go down due to deletes.
-fn detect_cardinality(columns: &[Option<ColumnIndex>]) -> Cardinality {
-    columns
-        .iter()
-        .flatten()
-        .map(ColumnIndex::get_cardinality)
-        .max()
-        .unwrap_or(Cardinality::Full)
-}
-
-pub fn merge_column_index<'a>(
-    columns: &'a [Option<ColumnIndex>],
-    merge_row_order: &'a MergeRowOrder,
-) -> SerializableColumnIndex<'a> {
-    // For simplification, we do not try to detect whether the cardinality could be
-    // downgraded thanks to deletes.
-    let cardinality_after_merge = detect_cardinality(columns);
-    match merge_row_order {
-        MergeRowOrder::Stack(stack_merge_order) => {
-            merge_column_index_stacked(columns, cardinality_after_merge, stack_merge_order)
-        }
-        MergeRowOrder::Shuffled(complex_merge_order) => {
-            merge_column_index_shuffled(columns, cardinality_after_merge, complex_merge_order)
-        }
-    }
-}
-
-// TODO actually, the shuffled code path is a bit too general.
-// In practise, we do not really shuffle everything.
-// The merge order restricted to a specific column keeps the original row order.
-//
-// This may offer some optimization that we have not explored yet.
-
-#[cfg(test)]
-mod tests {
-    use crate::column_index::merge::detect_cardinality;
-    use crate::column_index::multivalued_index::MultiValueIndex;
-    use crate::column_index::{merge_column_index, OptionalIndex, SerializableColumnIndex};
-    use crate::{Cardinality, ColumnIndex, MergeRowOrder, RowAddr, RowId, ShuffleMergeOrder};
-
-    #[test]
-    fn test_detect_cardinality() {
-        assert_eq!(detect_cardinality(&[]), Cardinality::Full);
-        let optional_index: ColumnIndex = OptionalIndex::for_test(1, &[]).into();
-        let multivalued_index: ColumnIndex = MultiValueIndex::for_test(&[0, 1]).into();
-        assert_eq!(
-            detect_cardinality(&[Some(optional_index.clone()), None]),
-            Cardinality::Optional
-        );
-        assert_eq!(
-            detect_cardinality(&[Some(optional_index.clone()), Some(ColumnIndex::Full)]),
-            Cardinality::Optional
-        );
-        assert_eq!(
-            detect_cardinality(&[Some(multivalued_index.clone()), None]),
-            Cardinality::Multivalued
-        );
-        assert_eq!(
-            detect_cardinality(&[
-                Some(multivalued_index.clone()),
-                Some(optional_index.clone())
-            ]),
-            Cardinality::Multivalued
-        );
-        assert_eq!(
-            detect_cardinality(&[Some(optional_index), Some(multivalued_index)]),
-            Cardinality::Multivalued
-        );
-    }
-
-    #[test]
-    fn test_merge_index_multivalued_sorted() {
-        let column_indexes: Vec<Option<ColumnIndex>> =
-            vec![Some(MultiValueIndex::for_test(&[0, 2, 5]).into())];
-        let merge_row_order: MergeRowOrder = ShuffleMergeOrder::for_test(
-            &[2],
-            vec![
-                RowAddr {
-                    segment_ord: 0u32,
-                    row_id: 1u32,
-                },
-                RowAddr {
-                    segment_ord: 0u32,
-                    row_id: 0u32,
-                },
-            ],
-        )
-        .into();
-        let merged_column_index = merge_column_index(&column_indexes[..], &merge_row_order);
-        let SerializableColumnIndex::Multivalued(start_index_iterable) = merged_column_index
-        else { panic!("Excpected a multivalued index") };
-        let start_indexes: Vec<RowId> = start_index_iterable.boxed_iter().collect();
-        assert_eq!(&start_indexes, &[0, 3, 5]);
-    }
-
-    #[test]
-    fn test_merge_index_multivalued_sorted_several_segment() {
-        let column_indexes: Vec<Option<ColumnIndex>> = vec![
-            Some(MultiValueIndex::for_test(&[0, 2, 5]).into()),
-            None,
-            Some(MultiValueIndex::for_test(&[0, 1, 4]).into()),
-        ];
-        let merge_row_order: MergeRowOrder = ShuffleMergeOrder::for_test(
-            &[2, 0, 2],
-            vec![
-                RowAddr {
-                    segment_ord: 2u32,
-                    row_id: 1u32,
-                },
-                RowAddr {
-                    segment_ord: 0u32,
-                    row_id: 0u32,
-                },
-                RowAddr {
-                    segment_ord: 2u32,
-                    row_id: 0u32,
-                },
-            ],
-        )
-        .into();
-        let merged_column_index = merge_column_index(&column_indexes[..], &merge_row_order);
-        let SerializableColumnIndex::Multivalued(start_index_iterable) = merged_column_index
-        else { panic!("Excpected a multivalued index") };
-        let start_indexes: Vec<RowId> = start_index_iterable.boxed_iter().collect();
-        assert_eq!(&start_indexes, &[0, 3, 5, 6]);
-    }
-}

columnar/src/column_index/merge/shuffled.rs

@@ -1,168 +0,0 @@
-use std::iter;
-
-use crate::column_index::{SerializableColumnIndex, Set};
-use crate::iterable::Iterable;
-use crate::{Cardinality, ColumnIndex, RowId, ShuffleMergeOrder};
-
-pub fn merge_column_index_shuffled<'a>(
-    column_indexes: &'a [Option<ColumnIndex>],
-    cardinality_after_merge: Cardinality,
-    shuffle_merge_order: &'a ShuffleMergeOrder,
-) -> SerializableColumnIndex<'a> {
-    match cardinality_after_merge {
-        Cardinality::Full => SerializableColumnIndex::Full,
-        Cardinality::Optional => {
-            let non_null_row_ids =
-                merge_column_index_shuffled_optional(column_indexes, shuffle_merge_order);
-            SerializableColumnIndex::Optional {
-                non_null_row_ids,
-                num_rows: shuffle_merge_order.num_rows(),
-            }
-        }
-        Cardinality::Multivalued => {
-            let multivalue_start_index =
-                merge_column_index_shuffled_multivalued(column_indexes, shuffle_merge_order);
-            SerializableColumnIndex::Multivalued(multivalue_start_index)
-        }
-    }
-}
-
-/// Merge several column indexes into one, ordering rows according to the merge_order passed as
-/// argument. While it is true that the `merge_order` may imply deletes and hence could in theory a
-/// multivalued index into an optional one, this is not supported today for simplification.
-///
-/// In other words the column_indexes passed as argument may NOT be multivalued.
-fn merge_column_index_shuffled_optional<'a>(
-    column_indexes: &'a [Option<ColumnIndex>],
-    merge_order: &'a ShuffleMergeOrder,
-) -> Box<dyn Iterable<RowId> + 'a> {
-    Box::new(ShuffledOptionalIndex {
-        column_indexes,
-        merge_order,
-    })
-}
-
-struct ShuffledOptionalIndex<'a> {
-    column_indexes: &'a [Option<ColumnIndex>],
-    merge_order: &'a ShuffleMergeOrder,
-}
-
-impl<'a> Iterable<u32> for ShuffledOptionalIndex<'a> {
-    fn boxed_iter(&self) -> Box<dyn Iterator<Item = u32> + '_> {
-        Box::new(self.merge_order
-        .iter_new_to_old_row_addrs()
-        .enumerate()
-        .filter_map(|(new_row_id, old_row_addr)| {
-            let Some(column_index) = &self.column_indexes[old_row_addr.segment_ord as usize] else {
-                return None;
-            };
-            let row_id = new_row_id as u32;
-            if column_index.has_value(old_row_addr.row_id) {
-                Some(row_id)
-            } else {
-                None
-            }
-        }))
-    }
-}
-
-fn merge_column_index_shuffled_multivalued<'a>(
-    column_indexes: &'a [Option<ColumnIndex>],
-    merge_order: &'a ShuffleMergeOrder,
-) -> Box<dyn Iterable<RowId> + 'a> {
-    Box::new(ShuffledMultivaluedIndex {
-        column_indexes,
-        merge_order,
-    })
-}
-
-struct ShuffledMultivaluedIndex<'a> {
-    column_indexes: &'a [Option<ColumnIndex>],
-    merge_order: &'a ShuffleMergeOrder,
-}
-
-fn iter_num_values<'a>(
-    column_indexes: &'a [Option<ColumnIndex>],
-    merge_order: &'a ShuffleMergeOrder,
-) -> impl Iterator<Item = u32> + 'a {
-    merge_order.iter_new_to_old_row_addrs().map(|row_addr| {
-        let Some(column_index) = &column_indexes[row_addr.segment_ord as usize] else {
-                    // No values in the entire column. It surely means there are 0 values associated to this row.
-                    return 0u32;
-                };
-        match column_index {
-            ColumnIndex::Empty { .. } => 0u32,
-            ColumnIndex::Full => 1,
-            ColumnIndex::Optional(optional_index) => {
-                u32::from(optional_index.contains(row_addr.row_id))
-            }
-            ColumnIndex::Multivalued(multivalued_index) => {
-                multivalued_index.range(row_addr.row_id).len() as u32
-            }
-        }
-    })
-}
-
-/// Transforms an iterator containing the number of vals per row (with `num_rows` elements)
-/// into a `start_offset` iterator starting at 0 and (with `num_rows + 1` element)
-fn integrate_num_vals(num_vals: impl Iterator<Item = u32>) -> impl Iterator<Item = RowId> {
-    iter::once(0u32).chain(num_vals.scan(0, |state, num_vals| {
-        *state += num_vals;
-        Some(*state)
-    }))
-}
-
-impl<'a> Iterable<u32> for ShuffledMultivaluedIndex<'a> {
-    fn boxed_iter(&self) -> Box<dyn Iterator<Item = u32> + '_> {
-        let num_vals_per_row = iter_num_values(self.column_indexes, self.merge_order);
-        Box::new(integrate_num_vals(num_vals_per_row))
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::column_index::OptionalIndex;
-    use crate::RowAddr;
-
-    #[test]
-    fn test_integrate_num_vals_empty() {
-        assert!(integrate_num_vals(iter::empty()).eq(iter::once(0)));
-    }
-
-    #[test]
-    fn test_integrate_num_vals_one_el() {
-        assert!(integrate_num_vals(iter::once(10)).eq([0, 10].into_iter()));
-    }
-
-    #[test]
-    fn test_integrate_num_vals_several() {
-        assert!(integrate_num_vals([3, 0, 10, 20].into_iter()).eq([0, 3, 3, 13, 33].into_iter()));
-    }
-
-    #[test]
-    fn test_merge_column_index_optional_shuffle() {
-        let optional_index: ColumnIndex = OptionalIndex::for_test(2, &[0]).into();
-        let column_indexes = vec![Some(optional_index), Some(ColumnIndex::Full)];
-        let row_addrs = vec![
-            RowAddr {
-                segment_ord: 0u32,
-                row_id: 1u32,
-            },
-            RowAddr {
-                segment_ord: 1u32,
-                row_id: 0u32,
-            },
-        ];
-        let shuffle_merge_order = ShuffleMergeOrder::for_test(&[2, 1], row_addrs);
-        let serializable_index = merge_column_index_shuffled(
-            &column_indexes[..],
-            Cardinality::Optional,
-            &shuffle_merge_order,
-        );
-        let SerializableColumnIndex::Optional { non_null_row_ids, num_rows } = serializable_index else { panic!() };
-        assert_eq!(num_rows, 2);
-        let non_null_rows: Vec<RowId> = non_null_row_ids.boxed_iter().collect();
-        assert_eq!(&non_null_rows, &[1]);
-    }
-}

columnar/src/column_index/merge/stacked.rs

@@ -1,156 +0,0 @@
-use std::iter;
-
-use crate::column_index::{SerializableColumnIndex, Set};
-use crate::iterable::Iterable;
-use crate::{Cardinality, ColumnIndex, RowId, StackMergeOrder};
-
-/// Simple case:
-/// The new mapping just consists in stacking the different column indexes.
-///
-/// There are no sort nor deletes involved.
-pub fn merge_column_index_stacked<'a>(
-    columns: &'a [Option<ColumnIndex>],
-    cardinality_after_merge: Cardinality,
-    stack_merge_order: &'a StackMergeOrder,
-) -> SerializableColumnIndex<'a> {
-    match cardinality_after_merge {
-        Cardinality::Full => SerializableColumnIndex::Full,
-        Cardinality::Optional => SerializableColumnIndex::Optional {
-            non_null_row_ids: Box::new(StackedOptionalIndex {
-                columns,
-                stack_merge_order,
-            }),
-            num_rows: stack_merge_order.num_rows(),
-        },
-        Cardinality::Multivalued => {
-            let stacked_multivalued_index = StackedMultivaluedIndex {
-                columns,
-                stack_merge_order,
-            };
-            SerializableColumnIndex::Multivalued(Box::new(stacked_multivalued_index))
-        }
-    }
-}
-
-struct StackedOptionalIndex<'a> {
-    columns: &'a [Option<ColumnIndex>],
-    stack_merge_order: &'a StackMergeOrder,
-}
-
-impl<'a> Iterable<RowId> for StackedOptionalIndex<'a> {
-    fn boxed_iter(&self) -> Box<dyn Iterator<Item = RowId> + 'a> {
-        Box::new(
-            self.columns
-                .iter()
-                .enumerate()
-                .flat_map(|(columnar_id, column_index_opt)| {
-                    let columnar_row_range = self.stack_merge_order.columnar_range(columnar_id);
-                    let rows_it: Box<dyn Iterator<Item = RowId>> = match column_index_opt {
-                        Some(ColumnIndex::Full) => Box::new(columnar_row_range),
-                        Some(ColumnIndex::Optional(optional_index)) => Box::new(
-                            optional_index
-                                .iter_rows()
-                                .map(move |row_id: RowId| columnar_row_range.start + row_id),
-                        ),
-                        Some(ColumnIndex::Multivalued(_)) => {
-                            panic!("No multivalued index is allowed when stacking column index");
-                        }
-                        None | Some(ColumnIndex::Empty { .. }) => Box::new(std::iter::empty()),
-                    };
-                    rows_it
-                }),
-        )
-    }
-}
-
-#[derive(Clone, Copy)]
-struct StackedMultivaluedIndex<'a> {
-    columns: &'a [Option<ColumnIndex>],
-    stack_merge_order: &'a StackMergeOrder,
-}
-
-fn convert_column_opt_to_multivalued_index<'a>(
-    column_index_opt: Option<&'a ColumnIndex>,
-    num_rows: RowId,
-) -> Box<dyn Iterator<Item = RowId> + 'a> {
-    match column_index_opt {
-        None | Some(ColumnIndex::Empty { .. }) => {
-            Box::new(iter::repeat(0u32).take(num_rows as usize + 1))
-        }
-        Some(ColumnIndex::Full) => Box::new(0..num_rows + 1),
-        Some(ColumnIndex::Optional(optional_index)) => {
-            Box::new(
-                (0..num_rows)
-                    // TODO optimize
-                    .map(|row_id| optional_index.rank(row_id))
-                    .chain(std::iter::once(optional_index.num_non_nulls())),
-            )
-        }
-        Some(ColumnIndex::Multivalued(multivalued_index)) => {
-            multivalued_index.start_index_column.iter()
-        }
-    }
-}
-
-impl<'a> Iterable<RowId> for StackedMultivaluedIndex<'a> {
-    fn boxed_iter(&self) -> Box<dyn Iterator<Item = RowId> + '_> {
-        let multivalued_indexes =
-            self.columns
-                .iter()
-                .map(Option::as_ref)
-                .enumerate()
-                .map(|(columnar_id, column_opt)| {
-                    let num_rows =
-                        self.stack_merge_order.columnar_range(columnar_id).len() as RowId;
-                    convert_column_opt_to_multivalued_index(column_opt, num_rows)
-                });
-        stack_multivalued_indexes(multivalued_indexes)
-    }
-}
-
-// Refactor me
-fn stack_multivalued_indexes<'a>(
-    mut multivalued_indexes: impl Iterator<Item = Box<dyn Iterator<Item = RowId> + 'a>> + 'a,
-) -> Box<dyn Iterator<Item = RowId> + 'a> {
-    let mut offset = 0;
-    let mut last_row_id = 0;
-    let mut current_it = multivalued_indexes.next();
-    Box::new(std::iter::from_fn(move || loop {
-        let Some(multivalued_index) = current_it.as_mut() else {
-            return None;
-        };
-        if let Some(row_id) = multivalued_index.next() {
-            last_row_id = offset + row_id;
-            return Some(last_row_id);
-        }
-        offset = last_row_id;
-        loop {
-            current_it = multivalued_indexes.next();
-            if current_it.as_mut()?.next().is_some() {
-                break;
-            }
-        }
-    }))
-}
-
-#[cfg(test)]
-mod tests {
-    use crate::RowId;
-
-    fn it<'a>(row_ids: &'a [RowId]) -> Box<dyn Iterator<Item = RowId> + 'a> {
-        Box::new(row_ids.iter().copied())
-    }
-
-    #[test]
-    fn test_stack() {
-        let columns = [
-            it(&[0u32, 0u32]),
-            it(&[0u32, 1u32, 1u32, 4u32]),
-            it(&[0u32, 3u32, 5u32]),
-            it(&[0u32, 4u32]),
-        ]
-        .into_iter();
-        let start_offsets: Vec<RowId> = super::stack_multivalued_indexes(columns).collect();
-        assert_eq!(start_offsets, &[0, 0, 1, 1, 4, 7, 9, 13]);
-    }
-}

columnar/src/column_index/mod.rs

@@ -1,115 +0,0 @@
-mod merge;
-mod multivalued_index;
-mod optional_index;
-mod serialize;
-
-use std::ops::Range;
-
-pub use merge::merge_column_index;
-pub use optional_index::{OptionalIndex, Set};
-pub use serialize::{open_column_index, serialize_column_index, SerializableColumnIndex};
-
-use crate::column_index::multivalued_index::MultiValueIndex;
-use crate::{Cardinality, DocId, RowId};
-
-#[derive(Clone)]
-pub enum ColumnIndex {
-    Empty {
-        num_docs: u32,
-    },
-    Full,
-    Optional(OptionalIndex),
-    /// In addition, at index num_rows, an extra value is added
-    /// containing the overal number of values.
-    Multivalued(MultiValueIndex),
-}
-
-impl From<OptionalIndex> for ColumnIndex {
-    fn from(optional_index: OptionalIndex) -> ColumnIndex {
-        ColumnIndex::Optional(optional_index)
-    }
-}
-
-impl From<MultiValueIndex> for ColumnIndex {
-    fn from(multi_value_index: MultiValueIndex) -> ColumnIndex {
-        ColumnIndex::Multivalued(multi_value_index)
-    }
-}
-
-impl ColumnIndex {
-    #[inline]
-    pub fn get_cardinality(&self) -> Cardinality {
-        match self {
-            ColumnIndex::Empty { .. } => Cardinality::Optional,
-            ColumnIndex::Full => Cardinality::Full,
-            ColumnIndex::Optional(_) => Cardinality::Optional,
-            ColumnIndex::Multivalued(_) => Cardinality::Multivalued,
-        }
-    }
-
-    /// Returns true if and only if there are at least one value associated to the row.
-    pub fn has_value(&self, doc_id: DocId) -> bool {
-        match self {
-            ColumnIndex::Empty { .. } => false,
-            ColumnIndex::Full => true,
-            ColumnIndex::Optional(optional_index) => optional_index.contains(doc_id),
-            ColumnIndex::Multivalued(multivalued_index) => {
-                !multivalued_index.range(doc_id).is_empty()
-            }
-        }
-    }
-
-    pub fn value_row_ids(&self, doc_id: DocId) -> Range<RowId> {
-        match self {
-            ColumnIndex::Empty { .. } => 0..0,
-            ColumnIndex::Full => doc_id..doc_id + 1,
-            ColumnIndex::Optional(optional_index) => {
-                if let Some(val) = optional_index.rank_if_exists(doc_id) {
-                    val..val + 1
-                } else {
-                    0..0
-                }
-            }
-            ColumnIndex::Multivalued(multivalued_index) => multivalued_index.range(doc_id),
-        }
-    }
-
-    pub fn docid_range_to_rowids(&self, doc_id: Range<DocId>) -> Range<RowId> {
-        match self {
-            ColumnIndex::Empty { .. } => 0..0,
-            ColumnIndex::Full => doc_id,
-            ColumnIndex::Optional(optional_index) => {
-                let row_start = optional_index.rank(doc_id.start);
-                let row_end = optional_index.rank(doc_id.end);
-                row_start..row_end
-            }
-            ColumnIndex::Multivalued(multivalued_index) => {
-                let end_docid = doc_id.end.min(multivalued_index.num_docs() - 1) + 1;
-                let start_docid = doc_id.start.min(end_docid);
-
-                let row_start = multivalued_index.start_index_column.get_val(start_docid);
-                let row_end = multivalued_index.start_index_column.get_val(end_docid);
-
-                row_start..row_end
-            }
-        }
-    }
-
-    pub fn select_batch_in_place(&self, doc_id_start: DocId, rank_ids: &mut Vec<RowId>) {
-        match self {
-            ColumnIndex::Empty { .. } => {
-                rank_ids.clear();
-            }
-            ColumnIndex::Full => {
-                // No need to do anything:
-                // value_idx and row_idx are the same.
-            }
-            ColumnIndex::Optional(optional_index) => {
-                optional_index.select_batch(&mut rank_ids[..]);
-            }
-            ColumnIndex::Multivalued(multivalued_index) => {
-                multivalued_index.select_batch_in_place(doc_id_start, rank_ids)
-            }
-        }
-    }
-}

columnar/src/column_index/multivalued_index.rs

@@ -1,141 +0,0 @@
-use std::io;
-use std::io::Write;
-use std::ops::Range;
-use std::sync::Arc;
-
-use common::OwnedBytes;
-
-use crate::column_values::{
-    load_u64_based_column_values, serialize_u64_based_column_values, CodecType, ColumnValues,
-};
-use crate::iterable::Iterable;
-use crate::{DocId, RowId};
-
-pub fn serialize_multivalued_index(
-    multivalued_index: &dyn Iterable<RowId>,
-    output: &mut impl Write,
-) -> io::Result<()> {
-    serialize_u64_based_column_values(
-        multivalued_index,
-        &[CodecType::Bitpacked, CodecType::Linear],
-        output,
-    )?;
-    Ok(())
-}
-
-pub fn open_multivalued_index(bytes: OwnedBytes) -> io::Result<MultiValueIndex> {
-    let start_index_column: Arc<dyn ColumnValues<RowId>> = load_u64_based_column_values(bytes)?;
-    Ok(MultiValueIndex { start_index_column })
-}
-
-#[derive(Clone)]
-/// Index to resolve value range for given doc_id.
-/// Starts at 0.
-pub struct MultiValueIndex {
-    pub start_index_column: Arc<dyn crate::ColumnValues<RowId>>,
-}
-
-impl From<Arc<dyn ColumnValues<RowId>>> for MultiValueIndex {
-    fn from(start_index_column: Arc<dyn ColumnValues<RowId>>) -> Self {
-        MultiValueIndex { start_index_column }
-    }
-}
-
-impl MultiValueIndex {
-    pub fn for_test(start_offsets: &[RowId]) -> MultiValueIndex {
-        let mut buffer = Vec::new();
-        serialize_multivalued_index(&start_offsets, &mut buffer).unwrap();
-        let bytes = OwnedBytes::new(buffer);
-        open_multivalued_index(bytes).unwrap()
-    }
-
-    /// Returns `[start, end)`, such that the values associated with
-    /// the given document are `start..end`.
-    #[inline]
-    pub(crate) fn range(&self, doc_id: DocId) -> Range<RowId> {
-        let start = self.start_index_column.get_val(doc_id);
-        let end = self.start_index_column.get_val(doc_id + 1);
-        start..end
-    }
-
-    /// Returns the number of documents in the index.
-    #[inline]
-    pub fn num_docs(&self) -> u32 {
-        self.start_index_column.num_vals() - 1
-    }
-
-    /// Converts a list of ranks (row ids of values) in a 1:n index to the corresponding list of
-    /// docids. Positions are converted inplace to docids.
-    ///
-    /// Since there is no index for value pos -> docid, but docid -> value pos range, we scan the
-    /// index.
-    ///
-    /// Correctness: positions needs to be sorted. idx_reader needs to contain monotonically
-    /// increasing positions.
-    ///
-    /// TODO: Instead of a linear scan we can employ a exponential search into binary search to
-    /// match a docid to its value position.
-    #[allow(clippy::bool_to_int_with_if)]
-    pub(crate) fn select_batch_in_place(&self, docid_start: DocId, ranks: &mut Vec<u32>) {
-        if ranks.is_empty() {
-            return;
-        }
-        let mut cur_doc = docid_start;
-        let mut last_doc = None;
-
-        assert!(self.start_index_column.get_val(docid_start) <= ranks[0]);
-
-        let mut write_doc_pos = 0;
-        for i in 0..ranks.len() {
-            let pos = ranks[i];
-            loop {
-                let end = self.start_index_column.get_val(cur_doc + 1);
-                if end > pos {
-                    ranks[write_doc_pos] = cur_doc;
-                    write_doc_pos += if last_doc == Some(cur_doc) { 0 } else { 1 };
-                    last_doc = Some(cur_doc);
-                    break;
-                }
-                cur_doc += 1;
-            }
-        }
-        ranks.truncate(write_doc_pos);
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use std::ops::Range;
-    use std::sync::Arc;
-
-    use super::MultiValueIndex;
-    use crate::column_values::IterColumn;
-    use crate::{ColumnValues, RowId};
-
-    fn index_to_pos_helper(
-        index: &MultiValueIndex,
-        doc_id_range: Range<u32>,
-        positions: &[u32],
-    ) -> Vec<u32> {
-        let mut positions = positions.to_vec();
-        index.select_batch_in_place(doc_id_range.start, &mut positions);
-        positions
-    }
-
-    #[test]
-    fn test_positions_to_docid() {
-        let offsets: Vec<RowId> = vec![0, 10, 12, 15, 22, 23]; // docid values are [0..10, 10..12, 12..15, etc.]
-        let column: Arc<dyn ColumnValues<RowId>> = Arc::new(IterColumn::from(offsets.into_iter()));
-        let index = MultiValueIndex::from(column);
-        assert_eq!(index.num_docs(), 5);
-        let positions = &[10u32, 11, 15, 20, 21, 22];
-        assert_eq!(index_to_pos_helper(&index, 0..5, positions), vec![1, 3, 4]);
-        assert_eq!(index_to_pos_helper(&index, 1..5, positions), vec![1, 3, 4]);
-        assert_eq!(index_to_pos_helper(&index, 0..5, &[9]), vec![0]);
-        assert_eq!(index_to_pos_helper(&index, 1..5, &[10]), vec![1]);
-        assert_eq!(index_to_pos_helper(&index, 1..5, &[11]), vec![1]);
-        assert_eq!(index_to_pos_helper(&index, 2..5, &[12]), vec![2]);
-        assert_eq!(index_to_pos_helper(&index, 2..5, &[12, 14]), vec![2]);
-        assert_eq!(index_to_pos_helper(&index, 2..5, &[12, 14, 15]), vec![2, 3]);
-    }
-}

columnar/src/column_index/optional_index/mod.rs

@@ -1,515 +0,0 @@
-use std::io::{self, Write};
-use std::sync::Arc;
-
-mod set;
-mod set_block;
-
-use common::{BinarySerializable, OwnedBytes, VInt};
-pub use set::{SelectCursor, Set, SetCodec};
-use set_block::{
-    DenseBlock, DenseBlockCodec, SparseBlock, SparseBlockCodec, DENSE_BLOCK_NUM_BYTES,
-};
-
-use crate::iterable::Iterable;
-use crate::{DocId, InvalidData, RowId};
-
-/// The threshold for for number of elements after which we switch to dense block encoding.
-///
-/// We simply pick the value that minimize the size of the blocks.
-const DENSE_BLOCK_THRESHOLD: u32 =
-    set_block::DENSE_BLOCK_NUM_BYTES / std::mem::size_of::<u16>() as u32; //< 5_120
-
-const ELEMENTS_PER_BLOCK: u32 = u16::MAX as u32 + 1;
-
-const BLOCK_SIZE: RowId = 1 << 16;
-
-#[derive(Copy, Clone, Debug)]
-struct BlockMeta {
-    non_null_rows_before_block: u32,
-    start_byte_offset: u32,
-    block_variant: BlockVariant,
-}
-
-#[derive(Clone, Copy, Debug)]
-enum BlockVariant {
-    Dense,
-    Sparse { num_vals: u16 },
-}
-
-impl BlockVariant {
-    pub fn empty() -> Self {
-        Self::Sparse { num_vals: 0 }
-    }
-    pub fn num_bytes_in_block(&self) -> u32 {
-        match *self {
-            BlockVariant::Dense => set_block::DENSE_BLOCK_NUM_BYTES,
-            BlockVariant::Sparse { num_vals } => num_vals as u32 * 2,
-        }
-    }
-}
-
-/// This codec is inspired by roaring bitmaps.
-/// In the dense blocks, however, in order to accelerate `select`
-/// we interleave an offset over two bytes. (more on this lower)
-///
-/// The lower 16 bits of doc ids are stored as u16 while the upper 16 bits are given by the block
-/// id. Each block contains 1<<16 docids.
-///
-/// # Serialized Data Layout
-/// The data starts with the block data. Each block is either dense or sparse encoded, depending on
-/// the number of values in the block. A block is sparse when it contains less than
-/// DENSE_BLOCK_THRESHOLD (6144) values.
-/// [Sparse data block | dense data block, .. #repeat*; Desc: Either a sparse or dense encoded
-/// block]
-/// ### Sparse block data
-/// [u16 LE, .. #repeat*; Desc: Positions with values in a block]
-/// ### Dense block data
-/// [Dense codec for the whole block; Desc: Similar to a bitvec(0..ELEMENTS_PER_BLOCK) + Metadata
-/// for faster lookups. See dense.rs]
-///
-/// The data is followed by block metadata, to know which area of the raw block data belongs to
-/// which block. Only metadata for blocks with elements is recorded to
-/// keep the overhead low for scenarios with many very sparse columns. The block metadata consists
-/// of the block index and the number of values in the block. Since we don't store empty blocks
-/// num_vals is incremented by 1, e.g. 0 means 1 value.
-///
-/// The last u16 is storing the number of metadata blocks.
-/// [u16 LE, .. #repeat*; Desc: Positions with values in a block][(u16 LE, u16 LE), .. #repeat*;
-/// Desc: (Block Id u16, Num Elements u16)][u16 LE; Desc: num blocks with values u16]
-///
-/// # Opening
-/// When opening the data layout, the data is expanded to `Vec<SparseCodecBlockVariant>`, where the
-/// index is the block index. For each block `byte_start` and `offset` is computed.
-#[derive(Clone)]
-pub struct OptionalIndex {
-    num_rows: RowId,
-    num_non_null_rows: RowId,
-    block_data: OwnedBytes,
-    block_metas: Arc<[BlockMeta]>,
-}
-
-/// Splits a value address into lower and upper 16bits.
-/// The lower 16 bits are the value in the block
-/// The upper 16 bits are the block index
-#[derive(Copy, Debug, Clone)]
-struct RowAddr {
-    block_id: u16,
-    in_block_row_id: u16,
-}
-
-#[inline(always)]
-fn row_addr_from_row_id(row_id: RowId) -> RowAddr {
-    RowAddr {
-        block_id: (row_id / BLOCK_SIZE) as u16,
-        in_block_row_id: (row_id % BLOCK_SIZE) as u16,
-    }
-}
-
-enum BlockSelectCursor<'a> {
-    Dense(<DenseBlock<'a> as Set<u16>>::SelectCursor<'a>),
-    Sparse(<SparseBlock<'a> as Set<u16>>::SelectCursor<'a>),
-}
-
-impl<'a> BlockSelectCursor<'a> {
-    fn select(&mut self, rank: u16) -> u16 {
-        match self {
-            BlockSelectCursor::Dense(dense_select_cursor) => dense_select_cursor.select(rank),
-            BlockSelectCursor::Sparse(sparse_select_cursor) => sparse_select_cursor.select(rank),
-        }
-    }
-}
-pub struct OptionalIndexSelectCursor<'a> {
-    current_block_cursor: BlockSelectCursor<'a>,
-    current_block_id: u16,
-    // The current block is guaranteed to contain ranks < end_rank.
-    current_block_end_rank: RowId,
-    optional_index: &'a OptionalIndex,
-    block_doc_idx_start: RowId,
-    num_null_rows_before_block: RowId,
-}
-
-impl<'a> OptionalIndexSelectCursor<'a> {
-    fn search_and_load_block(&mut self, rank: RowId) {
-        if rank < self.current_block_end_rank {
-            // we are already in the right block
-            return;
-        }
-        self.current_block_id = self.optional_index.find_block(rank, self.current_block_id);
-        self.current_block_end_rank = self
-            .optional_index
-            .block_metas
-            .get(self.current_block_id as usize + 1)
-            .map(|block_meta| block_meta.non_null_rows_before_block)
-            .unwrap_or(u32::MAX);
-        self.block_doc_idx_start = (self.current_block_id as u32) * ELEMENTS_PER_BLOCK;
-        let block_meta = self.optional_index.block_metas[self.current_block_id as usize];
-        self.num_null_rows_before_block = block_meta.non_null_rows_before_block;
-        let block: Block<'_> = self.optional_index.block(block_meta);
-        self.current_block_cursor = match block {
-            Block::Dense(dense_block) => BlockSelectCursor::Dense(dense_block.select_cursor()),
-            Block::Sparse(sparse_block) => BlockSelectCursor::Sparse(sparse_block.select_cursor()),
-        };
-    }
-}
-
-impl<'a> SelectCursor<RowId> for OptionalIndexSelectCursor<'a> {
-    fn select(&mut self, rank: RowId) -> RowId {
-        self.search_and_load_block(rank);
-        let index_in_block = (rank - self.num_null_rows_before_block) as u16;
-        self.current_block_cursor.select(index_in_block) as RowId + self.block_doc_idx_start
-    }
-}
-
-impl Set<RowId> for OptionalIndex {
-    type SelectCursor<'b> = OptionalIndexSelectCursor<'b> where Self: 'b;
-    // Check if value at position is not null.
-    #[inline]
-    fn contains(&self, row_id: RowId) -> bool {
-        let RowAddr {
-            block_id,
-            in_block_row_id,
-        } = row_addr_from_row_id(row_id);
-        let block_meta = self.block_metas[block_id as usize];
-        match self.block(block_meta) {
-            Block::Dense(dense_block) => dense_block.contains(in_block_row_id),
-            Block::Sparse(sparse_block) => sparse_block.contains(in_block_row_id),
-        }
-    }
-
-    #[inline]
-    fn rank(&self, doc_id: DocId) -> RowId {
-        let RowAddr {
-            block_id,
-            in_block_row_id,
-        } = row_addr_from_row_id(doc_id);
-        let block_meta = self.block_metas[block_id as usize];
-        let block = self.block(block_meta);
-        let block_offset_row_id = match block {
-            Block::Dense(dense_block) => dense_block.rank(in_block_row_id),
-            Block::Sparse(sparse_block) => sparse_block.rank(in_block_row_id),
-        } as u32;
-        block_meta.non_null_rows_before_block + block_offset_row_id
-    }
-
-    #[inline]
-    fn rank_if_exists(&self, doc_id: DocId) -> Option<RowId> {
-        let RowAddr {
-            block_id,
-            in_block_row_id,
-        } = row_addr_from_row_id(doc_id);
-        let block_meta = self.block_metas[block_id as usize];
-        let block = self.block(block_meta);
-        let block_offset_row_id = match block {
-            Block::Dense(dense_block) => dense_block.rank_if_exists(in_block_row_id),
-            Block::Sparse(sparse_block) => sparse_block.rank_if_exists(in_block_row_id),
-        }? as u32;
-        Some(block_meta.non_null_rows_before_block + block_offset_row_id)
-    }
-
-    #[inline]
-    fn select(&self, rank: RowId) -> RowId {
-        let block_pos = self.find_block(rank, 0);
-        let block_doc_idx_start = (block_pos as u32) * ELEMENTS_PER_BLOCK;
-        let block_meta = self.block_metas[block_pos as usize];
-        let block: Block<'_> = self.block(block_meta);
-        let index_in_block = (rank - block_meta.non_null_rows_before_block) as u16;
-        let in_block_rank = match block {
-            Block::Dense(dense_block) => dense_block.select(index_in_block),
-            Block::Sparse(sparse_block) => sparse_block.select(index_in_block),
-        };
-        block_doc_idx_start + in_block_rank as u32
-    }
-
-    fn select_cursor(&self) -> OptionalIndexSelectCursor<'_> {
-        OptionalIndexSelectCursor {
-            current_block_cursor: BlockSelectCursor::Sparse(
-                SparseBlockCodec::open(b"").select_cursor(),
-            ),
-            current_block_id: 0u16,
-            current_block_end_rank: 0u32, //< this is sufficient to force the first load
-            optional_index: self,
-            block_doc_idx_start: 0u32,
-            num_null_rows_before_block: 0u32,
-        }
-    }
-}
-
-impl OptionalIndex {
-    pub fn for_test(num_rows: RowId, row_ids: &[RowId]) -> OptionalIndex {
-        assert!(row_ids
-            .last()
-            .copied()
-            .map(|last_row_id| last_row_id < num_rows)
-            .unwrap_or(true));
-        let mut buffer = Vec::new();
-        serialize_optional_index(&row_ids, num_rows, &mut buffer).unwrap();
-        let bytes = OwnedBytes::new(buffer);
-        open_optional_index(bytes).unwrap()
-    }
-
-    pub fn num_docs(&self) -> RowId {
-        self.num_rows
-    }
-
-    pub fn num_non_nulls(&self) -> RowId {
-        self.num_non_null_rows
-    }
-
-    pub fn iter_rows(&self) -> impl Iterator<Item = RowId> + '_ {
-        // TODO optimize
-        let mut select_batch = self.select_cursor();
-        (0..self.num_non_null_rows).map(move |rank| select_batch.select(rank))
-    }
-    pub fn select_batch(&self, ranks: &mut [RowId]) {
-        let mut select_cursor = self.select_cursor();
-        for rank in ranks.iter_mut() {
-            *rank = select_cursor.select(*rank);
-        }
-    }
-
-    #[inline]
-    fn block(&self, block_meta: BlockMeta) -> Block<'_> {
-        let BlockMeta {
-            start_byte_offset,
-            block_variant,
-            ..
-        } = block_meta;
-        let start_byte_offset = start_byte_offset as usize;
-        let bytes = self.block_data.as_slice();
-        match block_variant {
-            BlockVariant::Dense => Block::Dense(DenseBlockCodec::open(
-                &bytes[start_byte_offset..start_byte_offset + DENSE_BLOCK_NUM_BYTES as usize],
-            )),
-            BlockVariant::Sparse { num_vals } => {
-                let end_byte_offset = start_byte_offset + num_vals as usize * 2;
-                let sparse_bytes = &bytes[start_byte_offset..end_byte_offset];
-                Block::Sparse(SparseBlockCodec::open(sparse_bytes))
-            }
-        }
-    }
-
-    #[inline]
-    fn find_block(&self, dense_idx: u32, start_block_pos: u16) -> u16 {
-        for block_pos in start_block_pos..self.block_metas.len() as u16 {
-            let offset = self.block_metas[block_pos as usize].non_null_rows_before_block;
-            if offset > dense_idx {
-                return block_pos - 1u16;
-            }
-        }
-        self.block_metas.len() as u16 - 1u16
-    }
-
-    // TODO Add a good API for the codec_idx to original_idx translation.
-    // The Iterator API is a probably a bad idea
-}
-
-#[derive(Copy, Clone)]
-enum Block<'a> {
-    Dense(DenseBlock<'a>),
-    Sparse(SparseBlock<'a>),
-}
-
-#[derive(Debug, Copy, Clone)]
-enum OptionalIndexCodec {
-    Dense = 0,
-    Sparse = 1,
-}
-
-impl OptionalIndexCodec {
-    fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    fn try_from_code(code: u8) -> Result<Self, InvalidData> {
-        match code {
-            0 => Ok(Self::Dense),
-            1 => Ok(Self::Sparse),
-            _ => Err(InvalidData),
-        }
-    }
-}
-
-impl BinarySerializable for OptionalIndexCodec {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        writer.write_all(&[self.to_code()])
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let optional_codec_code = u8::deserialize(reader)?;
-        let optional_codec = Self::try_from_code(optional_codec_code)?;
-        Ok(optional_codec)
-    }
-}
-
-fn serialize_optional_index_block(block_els: &[u16], out: &mut impl io::Write) -> io::Result<()> {
-    let is_sparse = is_sparse(block_els.len() as u32);
-    if is_sparse {
-        SparseBlockCodec::serialize(block_els.iter().copied(), out)?;
-    } else {
-        DenseBlockCodec::serialize(block_els.iter().copied(), out)?;
-    }
-    Ok(())
-}
-
-pub fn serialize_optional_index<W: io::Write>(
-    non_null_rows: &dyn Iterable<RowId>,
-    num_rows: RowId,
-    output: &mut W,
-) -> io::Result<()> {
-    VInt(num_rows as u64).serialize(output)?;
-
-    let mut rows_it = non_null_rows.boxed_iter();
-    let mut block_metadata: Vec<SerializedBlockMeta> = Vec::new();
-    let mut current_block = Vec::new();
-
-    // This if-statement for the first element ensures that
-    // `block_metadata` is not empty in the loop below.
-    let Some(idx) = rows_it.next() else {
-        output.write_all(&0u16.to_le_bytes())?;
-        return Ok(());
-    };
-
-    let row_addr = row_addr_from_row_id(idx);
-
-    let mut current_block_id = row_addr.block_id;
-    current_block.push(row_addr.in_block_row_id);
-
-    for idx in rows_it {
-        let value_addr = row_addr_from_row_id(idx);
-        if current_block_id != value_addr.block_id {
-            serialize_optional_index_block(&current_block[..], output)?;
-            block_metadata.push(SerializedBlockMeta {
-                block_id: current_block_id,
-                num_non_null_rows: current_block.len() as u32,
-            });
-            current_block.clear();
-            current_block_id = value_addr.block_id;
-        }
-        current_block.push(value_addr.in_block_row_id);
-    }
-
-    // handle last block
-    serialize_optional_index_block(&current_block[..], output)?;
-
-    block_metadata.push(SerializedBlockMeta {
-        block_id: current_block_id,
-        num_non_null_rows: current_block.len() as u32,
-    });
-
-    for block in &block_metadata {
-        output.write_all(&block.to_bytes())?;
-    }
-
-    output.write_all((block_metadata.len() as u16).to_le_bytes().as_ref())?;
-
-    Ok(())
-}
-
-const SERIALIZED_BLOCK_META_NUM_BYTES: usize = 4;
-
-#[derive(Clone, Copy, Debug)]
-struct SerializedBlockMeta {
-    block_id: u16,
-    num_non_null_rows: u32, //< takes values in 1..=u16::MAX
-}
-
-// TODO unit tests
-impl SerializedBlockMeta {
-    #[inline]
-    fn from_bytes(bytes: [u8; SERIALIZED_BLOCK_META_NUM_BYTES]) -> SerializedBlockMeta {
-        let block_id = u16::from_le_bytes(bytes[0..2].try_into().unwrap());
-        let num_non_null_rows: u32 =
-            u16::from_le_bytes(bytes[2..4].try_into().unwrap()) as u32 + 1u32;
-        SerializedBlockMeta {
-            block_id,
-            num_non_null_rows,
-        }
-    }
-
-    #[inline]
-    fn to_bytes(self) -> [u8; SERIALIZED_BLOCK_META_NUM_BYTES] {
-        assert!(self.num_non_null_rows > 0);
-        let mut bytes = [0u8; SERIALIZED_BLOCK_META_NUM_BYTES];
-        bytes[0..2].copy_from_slice(&self.block_id.to_le_bytes());
-        // We don't store empty blocks, therefore we can subtract 1.
-        // This way we will be able to use u16 when the number of elements is 1 << 16 or u16::MAX+1
-        bytes[2..4].copy_from_slice(&((self.num_non_null_rows - 1u32) as u16).to_le_bytes());
-        bytes
-    }
-}
-
-#[inline]
-fn is_sparse(num_rows_in_block: u32) -> bool {
-    num_rows_in_block < DENSE_BLOCK_THRESHOLD
-}
-
-fn deserialize_optional_index_block_metadatas(
-    data: &[u8],
-    num_rows: u32,
-) -> (Box<[BlockMeta]>, u32) {
-    let num_blocks = data.len() / SERIALIZED_BLOCK_META_NUM_BYTES;
-    let mut block_metas = Vec::with_capacity(num_blocks + 1);
-    let mut start_byte_offset = 0;
-    let mut non_null_rows_before_block = 0;
-    for block_meta_bytes in data.chunks_exact(SERIALIZED_BLOCK_META_NUM_BYTES) {
-        let block_meta_bytes: [u8; SERIALIZED_BLOCK_META_NUM_BYTES] =
-            block_meta_bytes.try_into().unwrap();
-        let SerializedBlockMeta {
-            block_id,
-            num_non_null_rows,
-        } = SerializedBlockMeta::from_bytes(block_meta_bytes);
-        block_metas.resize(
-            block_id as usize,
-            BlockMeta {
-                non_null_rows_before_block,
-                start_byte_offset,
-                block_variant: BlockVariant::empty(),
-            },
-        );
-        let block_variant = if is_sparse(num_non_null_rows) {
-            BlockVariant::Sparse {
-                num_vals: num_non_null_rows as u16,
-            }
-        } else {
-            BlockVariant::Dense
-        };
-        block_metas.push(BlockMeta {
-            non_null_rows_before_block,
-            start_byte_offset,
-            block_variant,
-        });
-        start_byte_offset += block_variant.num_bytes_in_block();
-        non_null_rows_before_block += num_non_null_rows;
-    }
-    block_metas.resize(
-        ((num_rows + BLOCK_SIZE - 1) / BLOCK_SIZE) as usize,
-        BlockMeta {
-            non_null_rows_before_block,
-            start_byte_offset,
-            block_variant: BlockVariant::empty(),
-        },
-    );
-    (block_metas.into_boxed_slice(), non_null_rows_before_block)
-}
-
-pub fn open_optional_index(bytes: OwnedBytes) -> io::Result<OptionalIndex> {
-    let (mut bytes, num_non_empty_blocks_bytes) = bytes.rsplit(2);
-    let num_non_empty_block_bytes =
-        u16::from_le_bytes(num_non_empty_blocks_bytes.as_slice().try_into().unwrap());
-    let num_rows = VInt::deserialize_u64(&mut bytes)? as u32;
-    let block_metas_num_bytes =
-        num_non_empty_block_bytes as usize * SERIALIZED_BLOCK_META_NUM_BYTES;
-    let (block_data, block_metas) = bytes.rsplit(block_metas_num_bytes);
-    let (block_metas, num_non_null_rows) =
-        deserialize_optional_index_block_metadatas(block_metas.as_slice(), num_rows);
-    let optional_index = OptionalIndex {
-        num_rows,
-        num_non_null_rows,
-        block_data,
-        block_metas: block_metas.into(),
-    };
-    Ok(optional_index)
-}
-
-#[cfg(test)]
-mod tests;

columnar/src/column_index/optional_index/set.rs

@@ -1,47 +0,0 @@
-use std::io;
-
-/// A codec makes it possible to serialize a set of
-/// elements, and open the resulting Set representation.
-pub trait SetCodec {
-    type Item: Copy + TryFrom<usize> + Eq + std::hash::Hash + std::fmt::Debug;
-    type Reader<'a>: Set<Self::Item>;
-
-    /// Serializes a set of unique sorted u16 elements.
-    ///
-    /// May panic if the elements are not sorted.
-    fn serialize(els: impl Iterator<Item = Self::Item>, wrt: impl io::Write) -> io::Result<()>;
-    fn open(data: &[u8]) -> Self::Reader<'_>;
-}
-
-/// Stateful object that makes it possible to compute several select in a row,
-/// provided the rank passed as argument are increasing.
-pub trait SelectCursor<T> {
-    // May panic if rank is greater than the number of elements in the Set,
-    // or if rank is < than value provided in the previous call.
-    fn select(&mut self, rank: T) -> T;
-}
-
-pub trait Set<T> {
-    type SelectCursor<'b>: SelectCursor<T>
-    where Self: 'b;
-
-    /// Returns true if the elements is contained in the Set
-    fn contains(&self, el: T) -> bool;
-
-    /// Returns the number of rows in the set that are < `el`
-    fn rank(&self, el: T) -> T;
-
-    /// If the set contains `el` returns the element rank.
-    /// If the set does not contain the element, it returns `None`.
-    fn rank_if_exists(&self, el: T) -> Option<T>;
-
-    /// Return the rank-th value stored in this bitmap.
-    ///
-    /// # Panics
-    ///
-    /// May panic if rank is greater than the number of elements in the Set.
-    fn select(&self, rank: T) -> T;
-
-    /// Creates a brand new select cursor.
-    fn select_cursor(&self) -> Self::SelectCursor<'_>;
-}

columnar/src/column_index/optional_index/set_block/dense.rs

@@ -1,278 +0,0 @@
-use std::convert::TryInto;
-use std::io::{self, Write};
-
-use common::BinarySerializable;
-
-use crate::column_index::optional_index::{SelectCursor, Set, SetCodec, ELEMENTS_PER_BLOCK};
-
-#[inline(always)]
-fn get_bit_at(input: u64, n: u16) -> bool {
-    input & (1 << n) != 0
-}
-
-#[inline]
-fn set_bit_at(input: &mut u64, n: u16) {
-    *input |= 1 << n;
-}
-
-/// For the `DenseCodec`, `data` which contains the encoded blocks.
-/// Each block consists of [u8; 12]. The first 8 bytes is a bitvec for 64 elements.
-/// The last 4 bytes are the offset, the number of set bits so far.
-///
-/// When translating the original index to a dense index, the correct block can be computed
-/// directly `orig_idx/64`. Inside the block the position is `orig_idx%64`.
-///
-/// When translating a dense index to the original index, we can use the offset to find the correct
-/// block. Direct computation is not possible, but we can employ a linear or binary search.
-
-const ELEMENTS_PER_MINI_BLOCK: u16 = 64;
-const MINI_BLOCK_BITVEC_NUM_BYTES: usize = 8;
-const MINI_BLOCK_OFFSET_NUM_BYTES: usize = 2;
-pub const MINI_BLOCK_NUM_BYTES: usize = MINI_BLOCK_BITVEC_NUM_BYTES + MINI_BLOCK_OFFSET_NUM_BYTES;
-
-/// Number of bytes in a dense block.
-pub const DENSE_BLOCK_NUM_BYTES: u32 =
-    (ELEMENTS_PER_BLOCK / ELEMENTS_PER_MINI_BLOCK as u32) * MINI_BLOCK_NUM_BYTES as u32;
-
-pub struct DenseBlockCodec;
-
-impl SetCodec for DenseBlockCodec {
-    type Item = u16;
-    type Reader<'a> = DenseBlock<'a>;
-
-    fn serialize(els: impl Iterator<Item = u16>, wrt: impl io::Write) -> io::Result<()> {
-        serialize_dense_codec(els, wrt)
-    }
-
-    #[inline]
-    fn open(data: &[u8]) -> Self::Reader<'_> {
-        assert_eq!(data.len(), DENSE_BLOCK_NUM_BYTES as usize);
-        DenseBlock(data)
-    }
-}
-
-/// Interpreting the bitvec as a set of integer within 0..=63
-/// and given an element, returns the number of elements in the
-/// set lesser than the element.
-///
-/// # Panics
-///
-/// May panic or return a wrong result if el <= 64.
-#[inline(always)]
-fn rank_u64(bitvec: u64, el: u16) -> u16 {
-    debug_assert!(el < 64);
-    let mask = (1u64 << el) - 1;
-    let masked_bitvec = bitvec & mask;
-    masked_bitvec.count_ones() as u16
-}
-
-#[inline(always)]
-fn select_u64(mut bitvec: u64, rank: u16) -> u16 {
-    for _ in 0..rank {
-        bitvec &= bitvec - 1;
-    }
-    bitvec.trailing_zeros() as u16
-}
-
-// TODO test the following solution on Intel... on Ryzen Zen <3 it is a catastrophy.
-// #[target_feature(enable = "bmi2")]
-// unsafe fn select_bitvec_unsafe(bitvec: u64, rank: u16) -> u16 {
-//     let pdep = _pdep_u64(1u64 << rank, bitvec);
-//     pdep.trailing_zeros() as u16
-// }
-
-#[derive(Clone, Copy, Debug)]
-struct DenseMiniBlock {
-    bitvec: u64,
-    rank: u16,
-}
-
-impl DenseMiniBlock {
-    fn from_bytes(data: [u8; MINI_BLOCK_NUM_BYTES]) -> Self {
-        let bitvec = u64::from_le_bytes(data[..MINI_BLOCK_BITVEC_NUM_BYTES].try_into().unwrap());
-        let rank = u16::from_le_bytes(data[MINI_BLOCK_BITVEC_NUM_BYTES..].try_into().unwrap());
-        Self { bitvec, rank }
-    }
-
-    fn to_bytes(self) -> [u8; MINI_BLOCK_NUM_BYTES] {
-        let mut bytes = [0u8; MINI_BLOCK_NUM_BYTES];
-        bytes[..MINI_BLOCK_BITVEC_NUM_BYTES].copy_from_slice(&self.bitvec.to_le_bytes());
-        bytes[MINI_BLOCK_BITVEC_NUM_BYTES..].copy_from_slice(&self.rank.to_le_bytes());
-        bytes
-    }
-}
-
-#[derive(Copy, Clone)]
-pub struct DenseBlock<'a>(&'a [u8]);
-
-pub struct DenseBlockSelectCursor<'a> {
-    block_id: u16,
-    dense_block: DenseBlock<'a>,
-}
-
-impl<'a> SelectCursor<u16> for DenseBlockSelectCursor<'a> {
-    #[inline]
-    fn select(&mut self, rank: u16) -> u16 {
-        self.block_id = self
-            .dense_block
-            .find_miniblock_containing_rank(rank, self.block_id)
-            .unwrap();
-        let index_block = self.dense_block.mini_block(self.block_id);
-        let in_block_rank = rank - index_block.rank;
-        self.block_id * ELEMENTS_PER_MINI_BLOCK + select_u64(index_block.bitvec, in_block_rank)
-    }
-}
-
-impl<'a> Set<u16> for DenseBlock<'a> {
-    type SelectCursor<'b> = DenseBlockSelectCursor<'a> where Self: 'b;
-
-    #[inline(always)]
-    fn contains(&self, el: u16) -> bool {
-        let mini_block_id = el / ELEMENTS_PER_MINI_BLOCK;
-        let bitvec = self.mini_block(mini_block_id).bitvec;
-        let pos_in_bitvec = el % ELEMENTS_PER_MINI_BLOCK;
-        get_bit_at(bitvec, pos_in_bitvec)
-    }
-
-    #[inline(always)]
-    fn rank_if_exists(&self, el: u16) -> Option<u16> {
-        let block_pos = el / ELEMENTS_PER_MINI_BLOCK;
-        let index_block = self.mini_block(block_pos);
-        let pos_in_block_bit_vec = el % ELEMENTS_PER_MINI_BLOCK;
-        let ones_in_block = rank_u64(index_block.bitvec, pos_in_block_bit_vec);
-        let rank = index_block.rank + ones_in_block;
-        if get_bit_at(index_block.bitvec, pos_in_block_bit_vec) {
-            Some(rank)
-        } else {
-            None
-        }
-    }
-
-    #[inline(always)]
-    fn rank(&self, el: u16) -> u16 {
-        let block_pos = el / ELEMENTS_PER_MINI_BLOCK;
-        let index_block = self.mini_block(block_pos);
-        let pos_in_block_bit_vec = el % ELEMENTS_PER_MINI_BLOCK;
-        let ones_in_block = rank_u64(index_block.bitvec, pos_in_block_bit_vec);
-        index_block.rank + ones_in_block
-    }
-
-    #[inline(always)]
-    fn select(&self, rank: u16) -> u16 {
-        let block_id = self.find_miniblock_containing_rank(rank, 0).unwrap();
-        let index_block = self.mini_block(block_id);
-        let in_block_rank = rank - index_block.rank;
-        block_id * ELEMENTS_PER_MINI_BLOCK + select_u64(index_block.bitvec, in_block_rank)
-    }
-
-    #[inline(always)]
-    fn select_cursor(&self) -> Self::SelectCursor<'_> {
-        DenseBlockSelectCursor {
-            block_id: 0,
-            dense_block: *self,
-        }
-    }
-}
-
-impl<'a> DenseBlock<'a> {
-    #[inline]
-    fn mini_block(&self, mini_block_id: u16) -> DenseMiniBlock {
-        let data_start_pos = mini_block_id as usize * MINI_BLOCK_NUM_BYTES;
-        DenseMiniBlock::from_bytes(
-            self.0[data_start_pos..data_start_pos + MINI_BLOCK_NUM_BYTES]
-                .try_into()
-                .unwrap(),
-        )
-    }
-
-    #[inline]
-    fn iter_miniblocks(
-        &self,
-        from_block_id: u16,
-    ) -> impl Iterator<Item = (u16, DenseMiniBlock)> + '_ {
-        self.0
-            .chunks_exact(MINI_BLOCK_NUM_BYTES)
-            .enumerate()
-            .skip(from_block_id as usize)
-            .map(|(block_id, bytes)| {
-                let mini_block = DenseMiniBlock::from_bytes(bytes.try_into().unwrap());
-                (block_id as u16, mini_block)
-            })
-    }
-
-    /// Finds the block position containing the dense_idx.
-    ///
-    /// # Correctness
-    /// dense_idx needs to be smaller than the number of values in the index
-    ///
-    /// The last offset number is equal to the number of values in the index.
-    #[inline]
-    fn find_miniblock_containing_rank(&self, rank: u16, from_block_id: u16) -> Option<u16> {
-        self.iter_miniblocks(from_block_id)
-            .take_while(|(_, block)| block.rank <= rank)
-            .map(|(block_id, _)| block_id)
-            .last()
-    }
-}
-
-/// Iterator over all values, true if set, otherwise false
-pub fn serialize_dense_codec(
-    els: impl Iterator<Item = u16>,
-    mut output: impl Write,
-) -> io::Result<()> {
-    let mut non_null_rows_before: u16 = 0u16;
-    let mut block = 0u64;
-    let mut current_block_id = 0u16;
-    for el in els {
-        let block_id = el / ELEMENTS_PER_MINI_BLOCK;
-        let in_offset = el % ELEMENTS_PER_MINI_BLOCK;
-        while block_id > current_block_id {
-            let dense_mini_block = DenseMiniBlock {
-                bitvec: block,
-                rank: non_null_rows_before,
-            };
-            output.write_all(&dense_mini_block.to_bytes())?;
-            non_null_rows_before += block.count_ones() as u16;
-            block = 0u64;
-            current_block_id += 1u16;
-        }
-        set_bit_at(&mut block, in_offset);
-    }
-    while current_block_id <= u16::MAX / ELEMENTS_PER_MINI_BLOCK {
-        block.serialize(&mut output)?;
-        non_null_rows_before.serialize(&mut output)?;
-        // This will overflow to 0 exactly if all bits are set.
-        // This is however not problem as we won't use this last value.
-        non_null_rows_before = non_null_rows_before.wrapping_add(block.count_ones() as u16);
-        block = 0u64;
-        current_block_id += 1u16;
-    }
-    Ok(())
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_select_bitvec() {
-        assert_eq!(select_u64(1u64, 0), 0);
-        assert_eq!(select_u64(2u64, 0), 1);
-        assert_eq!(select_u64(4u64, 0), 2);
-        assert_eq!(select_u64(8u64, 0), 3);
-        assert_eq!(select_u64(1 | 8u64, 0), 0);
-        assert_eq!(select_u64(1 | 8u64, 1), 3);
-    }
-
-    #[test]
-    fn test_count_ones() {
-        for i in 0..=63 {
-            assert_eq!(rank_u64(u64::MAX, i), i);
-        }
-    }
-
-    #[test]
-    fn test_dense() {
-        assert_eq!(DENSE_BLOCK_NUM_BYTES, 10_240);
-    }
-}

columnar/src/column_index/optional_index/set_block/mod.rs

@@ -1,8 +0,0 @@
-mod dense;
-mod sparse;
-
-pub use dense::{DenseBlock, DenseBlockCodec, DENSE_BLOCK_NUM_BYTES};
-pub use sparse::{SparseBlock, SparseBlockCodec};
-
-#[cfg(test)]
-mod tests;

columnar/src/column_index/optional_index/set_block/sparse.rs

@@ -1,111 +0,0 @@
-use crate::column_index::optional_index::{SelectCursor, Set, SetCodec};
-
-pub struct SparseBlockCodec;
-
-impl SetCodec for SparseBlockCodec {
-    type Item = u16;
-    type Reader<'a> = SparseBlock<'a>;
-
-    fn serialize(
-        els: impl Iterator<Item = u16>,
-        mut wrt: impl std::io::Write,
-    ) -> std::io::Result<()> {
-        for el in els {
-            wrt.write_all(&el.to_le_bytes())?;
-        }
-        Ok(())
-    }
-
-    fn open(data: &[u8]) -> Self::Reader<'_> {
-        SparseBlock(data)
-    }
-}
-
-#[derive(Copy, Clone)]
-pub struct SparseBlock<'a>(&'a [u8]);
-
-impl<'a> SelectCursor<u16> for SparseBlock<'a> {
-    #[inline]
-    fn select(&mut self, rank: u16) -> u16 {
-        <SparseBlock<'a> as Set<u16>>::select(self, rank)
-    }
-}
-
-impl<'a> Set<u16> for SparseBlock<'a> {
-    type SelectCursor<'b> = Self where Self: 'b;
-
-    #[inline(always)]
-    fn contains(&self, el: u16) -> bool {
-        self.binary_search(el).is_ok()
-    }
-
-    #[inline(always)]
-    fn rank_if_exists(&self, el: u16) -> Option<u16> {
-        self.binary_search(el).ok()
-    }
-
-    #[inline(always)]
-    fn rank(&self, el: u16) -> u16 {
-        self.binary_search(el).unwrap_or_else(|el| el)
-    }
-
-    #[inline(always)]
-    fn select(&self, rank: u16) -> u16 {
-        let offset = rank as usize * 2;
-        u16::from_le_bytes(self.0[offset..offset + 2].try_into().unwrap())
-    }
-
-    #[inline(always)]
-    fn select_cursor(&self) -> Self::SelectCursor<'_> {
-        *self
-    }
-}
-
-#[inline(always)]
-fn get_u16(data: &[u8], byte_position: usize) -> u16 {
-    let bytes: [u8; 2] = data[byte_position..byte_position + 2].try_into().unwrap();
-    u16::from_le_bytes(bytes)
-}
-
-impl<'a> SparseBlock<'a> {
-    #[inline(always)]
-    fn value_at_idx(&self, data: &[u8], idx: u16) -> u16 {
-        let start_offset: usize = idx as usize * 2;
-        get_u16(data, start_offset)
-    }
-
-    #[inline]
-    fn num_vals(&self) -> u16 {
-        (self.0.len() / 2) as u16
-    }
-
-    #[inline]
-    #[allow(clippy::comparison_chain)]
-    // Looks for the element in the block. Returns the positions if found.
-    fn binary_search(&self, target: u16) -> Result<u16, u16> {
-        let data = &self.0;
-        let mut size = self.num_vals();
-        let mut left = 0;
-        let mut right = size;
-        // TODO try different implem.
-        //  e.g. exponential search into binary search
-        while left < right {
-            let mid = left + size / 2;
-
-            // TODO do boundary check only once, and then use an
-            // unsafe `value_at_idx`
-            let mid_val = self.value_at_idx(data, mid);
-
-            if target > mid_val {
-                left = mid + 1;
-            } else if target < mid_val {
-                right = mid;
-            } else {
-                return Ok(mid);
-            }
-
-            size = right - left;
-        }
-        Err(left)
-    }
-}

columnar/src/column_index/optional_index/set_block/tests.rs

@@ -1,109 +0,0 @@
-use std::collections::HashMap;
-
-use crate::column_index::optional_index::set_block::dense::DENSE_BLOCK_NUM_BYTES;
-use crate::column_index::optional_index::set_block::{DenseBlockCodec, SparseBlockCodec};
-use crate::column_index::optional_index::{SelectCursor, Set, SetCodec};
-
-fn test_set_helper<C: SetCodec<Item = u16>>(vals: &[u16]) -> usize {
-    let mut buffer = Vec::new();
-    C::serialize(vals.iter().copied(), &mut buffer).unwrap();
-    let tested_set = C::open(buffer.as_slice());
-    let hash_set: HashMap<C::Item, C::Item> = vals
-        .iter()
-        .copied()
-        .enumerate()
-        .map(|(ord, val)| (val, C::Item::try_from(ord).ok().unwrap()))
-        .collect();
-    for val in 0u16..=u16::MAX {
-        assert_eq!(tested_set.contains(val), hash_set.contains_key(&val));
-        assert_eq!(tested_set.rank_if_exists(val), hash_set.get(&val).copied());
-        assert_eq!(
-            tested_set.rank(val),
-            vals.iter().cloned().take_while(|v| *v < val).count() as u16
-        );
-    }
-    for rank in 0..vals.len() {
-        assert_eq!(tested_set.select(rank as u16), vals[rank]);
-    }
-    buffer.len()
-}
-
-#[test]
-fn test_dense_block_set_u16_empty() {
-    let buffer_len = test_set_helper::<DenseBlockCodec>(&[]);
-    assert_eq!(buffer_len, DENSE_BLOCK_NUM_BYTES as usize);
-}
-
-#[test]
-fn test_dense_block_set_u16_max() {
-    let buffer_len = test_set_helper::<DenseBlockCodec>(&[u16::MAX]);
-    assert_eq!(buffer_len, DENSE_BLOCK_NUM_BYTES as usize);
-}
-
-#[test]
-fn test_sparse_block_set_u16_empty() {
-    let buffer_len = test_set_helper::<SparseBlockCodec>(&[]);
-    assert_eq!(buffer_len, 0);
-}
-
-#[test]
-fn test_sparse_block_set_u16_max() {
-    let buffer_len = test_set_helper::<SparseBlockCodec>(&[u16::MAX]);
-    assert_eq!(buffer_len, 2);
-}
-
-use proptest::prelude::*;
-
-proptest! {
-    #![proptest_config(ProptestConfig::with_cases(1))]
-    #[test]
-    fn test_prop_test_dense(els in proptest::collection::btree_set(0..=u16::MAX, 0..=u16::MAX as usize)) {
-        let vals: Vec<u16> = els.into_iter().collect();
-        let buffer_len = test_set_helper::<DenseBlockCodec>(&vals);
-        assert_eq!(buffer_len, DENSE_BLOCK_NUM_BYTES as usize);
-    }
-
-    #[test]
-    fn test_prop_test_sparse(els in proptest::collection::btree_set(0..=u16::MAX, 0..=u16::MAX as usize)) {
-        let vals: Vec<u16> = els.into_iter().collect();
-        let buffer_len = test_set_helper::<SparseBlockCodec>(&vals);
-        assert_eq!(buffer_len, vals.len() * 2);
-    }
-}
-
-#[test]
-fn test_simple_translate_codec_codec_idx_to_original_idx_dense() {
-    let mut buffer = Vec::new();
-    DenseBlockCodec::serialize([1, 3, 17, 32, 30_000, 30_001].iter().copied(), &mut buffer)
-        .unwrap();
-    let tested_set = DenseBlockCodec::open(buffer.as_slice());
-    assert!(tested_set.contains(1));
-    let mut select_cursor = tested_set.select_cursor();
-    assert_eq!(select_cursor.select(0), 1);
-    assert_eq!(select_cursor.select(1), 3);
-    assert_eq!(select_cursor.select(2), 17);
-}
-
-#[test]
-fn test_simple_translate_codec_idx_to_original_idx_sparse() {
-    let mut buffer = Vec::new();
-    SparseBlockCodec::serialize([1, 3, 17].iter().copied(), &mut buffer).unwrap();
-    let tested_set = SparseBlockCodec::open(buffer.as_slice());
-    assert!(tested_set.contains(1));
-    let mut select_cursor = tested_set.select_cursor();
-    assert_eq!(SelectCursor::select(&mut select_cursor, 0), 1);
-    assert_eq!(SelectCursor::select(&mut select_cursor, 1), 3);
-    assert_eq!(SelectCursor::select(&mut select_cursor, 2), 17);
-}
-
-#[test]
-fn test_simple_translate_codec_idx_to_original_idx_dense() {
-    let mut buffer = Vec::new();
-    DenseBlockCodec::serialize(0u16..150u16, &mut buffer).unwrap();
-    let tested_set = DenseBlockCodec::open(buffer.as_slice());
-    assert!(tested_set.contains(1));
-    let mut select_cursor = tested_set.select_cursor();
-    for i in 0..150 {
-        assert_eq!(i, select_cursor.select(i));
-    }
-}

columnar/src/column_index/optional_index/tests.rs

@@ -1,371 +0,0 @@
-use proptest::prelude::{any, prop, *};
-use proptest::strategy::Strategy;
-use proptest::{prop_oneof, proptest};
-
-use super::*;
-
-#[test]
-fn test_dense_block_threshold() {
-    assert_eq!(super::DENSE_BLOCK_THRESHOLD, 5_120);
-}
-
-fn random_bitvec() -> BoxedStrategy<Vec<bool>> {
-    prop_oneof![
-        1 => prop::collection::vec(proptest::bool::weighted(1.0), 0..100),
-        1 => prop::collection::vec(proptest::bool::weighted(0.00), 0..(ELEMENTS_PER_BLOCK as usize * 3)), // empty blocks
-        1 => prop::collection::vec(proptest::bool::weighted(1.00), 0..(ELEMENTS_PER_BLOCK as usize + 10)), // full block
-        1 => prop::collection::vec(proptest::bool::weighted(0.01), 0..100),
-        1 => prop::collection::vec(proptest::bool::weighted(0.01), 0..u16::MAX as usize),
-        8 => vec![any::<bool>()],
-    ]
-    .boxed()
-}
-
-proptest! {
-    #![proptest_config(ProptestConfig::with_cases(50))]
-    #[test]
-    fn test_with_random_bitvecs(bitvec1 in random_bitvec(), bitvec2 in random_bitvec(), bitvec3 in random_bitvec()) {
-        let mut bitvec = Vec::new();
-        bitvec.extend_from_slice(&bitvec1);
-        bitvec.extend_from_slice(&bitvec2);
-        bitvec.extend_from_slice(&bitvec3);
-        test_null_index(&bitvec[..]);
-    }
-}
-
-#[test]
-fn test_with_random_sets_simple() {
-    let vals = 10..BLOCK_SIZE * 2;
-    let mut out: Vec<u8> = Vec::new();
-    serialize_optional_index(&vals, 100, &mut out).unwrap();
-    let null_index = open_optional_index(OwnedBytes::new(out)).unwrap();
-    let ranks: Vec<u32> = (65_472u32..65_473u32).collect();
-    let els: Vec<u32> = ranks.iter().copied().map(|rank| rank + 10).collect();
-    let mut select_cursor = null_index.select_cursor();
-    for (rank, el) in ranks.iter().copied().zip(els.iter().copied()) {
-        assert_eq!(select_cursor.select(rank), el);
-    }
-}
-
-#[test]
-fn test_optional_index_trailing_empty_blocks() {
-    test_null_index(&[false]);
-}
-
-#[test]
-fn test_optional_index_one_block_false() {
-    let mut iter = vec![false; ELEMENTS_PER_BLOCK as usize];
-    iter.push(true);
-    test_null_index(&iter[..]);
-}
-
-#[test]
-fn test_optional_index_one_block_true() {
-    let mut iter = vec![true; ELEMENTS_PER_BLOCK as usize];
-    iter.push(true);
-    test_null_index(&iter[..]);
-}
-
-impl<'a> Iterable<RowId> for &'a [bool] {
-    fn boxed_iter(&self) -> Box<dyn Iterator<Item = RowId> + 'a> {
-        Box::new(
-            self.iter()
-                .cloned()
-                .enumerate()
-                .filter(|(_pos, val)| *val)
-                .map(|(pos, _val)| pos as u32),
-        )
-    }
-}
-
-fn test_null_index(data: &[bool]) {
-    let mut out: Vec<u8> = Vec::new();
-    serialize_optional_index(&data, data.len() as RowId, &mut out).unwrap();
-    let null_index = open_optional_index(OwnedBytes::new(out)).unwrap();
-    let orig_idx_with_value: Vec<u32> = data
-        .iter()
-        .enumerate()
-        .filter(|(_pos, val)| **val)
-        .map(|(pos, _val)| pos as u32)
-        .collect();
-    let mut select_iter = null_index.select_cursor();
-    for i in 0..orig_idx_with_value.len() {
-        assert_eq!(select_iter.select(i as u32), orig_idx_with_value[i]);
-    }
-
-    let step_size = (orig_idx_with_value.len() / 100).max(1);
-    for (dense_idx, orig_idx) in orig_idx_with_value.iter().enumerate().step_by(step_size) {
-        assert_eq!(null_index.rank_if_exists(*orig_idx), Some(dense_idx as u32));
-    }
-
-    // 100 samples
-    let step_size = (data.len() / 100).max(1);
-    for (pos, value) in data.iter().enumerate().step_by(step_size) {
-        assert_eq!(null_index.contains(pos as u32), *value);
-    }
-}
-
-#[test]
-fn test_optional_index_test_translation() {
-    let optional_index = OptionalIndex::for_test(4, &[0, 2]);
-    let mut select_cursor = optional_index.select_cursor();
-    assert_eq!(select_cursor.select(0), 0);
-    assert_eq!(select_cursor.select(1), 2);
-}
-
-#[test]
-fn test_optional_index_translate() {
-    let optional_index = OptionalIndex::for_test(4, &[0, 2]);
-    assert_eq!(optional_index.rank_if_exists(0), Some(0));
-    assert_eq!(optional_index.rank_if_exists(2), Some(1));
-}
-
-#[test]
-fn test_optional_index_small() {
-    let optional_index = OptionalIndex::for_test(4, &[0, 2]);
-    assert!(optional_index.contains(0));
-    assert!(!optional_index.contains(1));
-    assert!(optional_index.contains(2));
-    assert!(!optional_index.contains(3));
-}
-
-#[test]
-fn test_optional_index_large() {
-    let row_ids = &[ELEMENTS_PER_BLOCK, ELEMENTS_PER_BLOCK + 1];
-    let optional_index = OptionalIndex::for_test(ELEMENTS_PER_BLOCK + 2, row_ids);
-    assert!(!optional_index.contains(0));
-    assert!(!optional_index.contains(100));
-    assert!(!optional_index.contains(ELEMENTS_PER_BLOCK - 1));
-    assert!(optional_index.contains(ELEMENTS_PER_BLOCK));
-    assert!(optional_index.contains(ELEMENTS_PER_BLOCK + 1));
-}
-
-fn test_optional_index_iter_aux(row_ids: &[RowId], num_rows: RowId) {
-    let optional_index = OptionalIndex::for_test(num_rows, row_ids);
-    assert_eq!(optional_index.num_docs(), num_rows);
-    assert!(optional_index.iter_rows().eq(row_ids.iter().copied()));
-}
-
-#[test]
-fn test_optional_index_iter_empty() {
-    test_optional_index_iter_aux(&[], 0u32);
-}
-
-fn test_optional_index_rank_aux(row_ids: &[RowId]) {
-    let num_rows = row_ids.last().copied().unwrap_or(0u32) + 1;
-    let null_index = OptionalIndex::for_test(num_rows, row_ids);
-    assert_eq!(null_index.num_docs(), num_rows);
-    for (row_id, row_val) in row_ids.iter().copied().enumerate() {
-        assert_eq!(null_index.rank(row_val), row_id as u32);
-        assert_eq!(null_index.rank_if_exists(row_val), Some(row_id as u32));
-        if row_val > 0 && !null_index.contains(&row_val - 1) {
-            assert_eq!(null_index.rank(row_val - 1), row_id as u32);
-        }
-        assert_eq!(null_index.rank(row_val + 1), row_id as u32 + 1);
-    }
-}
-
-#[test]
-fn test_optional_index_rank() {
-    test_optional_index_rank_aux(&[1u32]);
-    test_optional_index_rank_aux(&[0u32, 1u32]);
-    let mut block = Vec::new();
-    block.push(3u32);
-    block.extend((0..BLOCK_SIZE).map(|i| i + BLOCK_SIZE + 1));
-    test_optional_index_rank_aux(&block);
-}
-
-#[test]
-fn test_optional_index_iter_empty_one() {
-    test_optional_index_iter_aux(&[1], 2u32);
-    test_optional_index_iter_aux(&[100_000], 200_000u32);
-}
-
-#[test]
-fn test_optional_index_iter_dense_block() {
-    let mut block = Vec::new();
-    block.push(3u32);
-    block.extend((0..BLOCK_SIZE).map(|i| i + BLOCK_SIZE + 1));
-    test_optional_index_iter_aux(&block, 3 * BLOCK_SIZE);
-}
-
-#[test]
-fn test_optional_index_for_tests() {
-    let optional_index = OptionalIndex::for_test(4, &[1, 2]);
-    assert!(!optional_index.contains(0));
-    assert!(optional_index.contains(1));
-    assert!(optional_index.contains(2));
-    assert!(!optional_index.contains(3));
-    assert_eq!(optional_index.num_docs(), 4);
-}
-
-#[cfg(all(test, feature = "unstable"))]
-mod bench {
-
-    use rand::rngs::StdRng;
-    use rand::{Rng, SeedableRng};
-    use test::Bencher;
-
-    use super::*;
-
-    const TOTAL_NUM_VALUES: u32 = 1_000_000;
-    fn gen_bools(fill_ratio: f64) -> OptionalIndex {
-        let mut out = Vec::new();
-        let mut rng: StdRng = StdRng::from_seed([1u8; 32]);
-        let vals: Vec<RowId> = (0..TOTAL_NUM_VALUES)
-            .map(|_| rng.gen_bool(fill_ratio))
-            .enumerate()
-            .filter(|(pos, val)| *val)
-            .map(|(pos, _)| pos as RowId)
-            .collect();
-        serialize_optional_index(&&vals[..], TOTAL_NUM_VALUES, &mut out).unwrap();
-        let codec = open_optional_index(OwnedBytes::new(out)).unwrap();
-        codec
-    }
-
-    fn random_range_iterator(
-        start: u32,
-        end: u32,
-        avg_step_size: u32,
-        avg_deviation: u32,
-    ) -> impl Iterator<Item = u32> {
-        let mut rng: StdRng = StdRng::from_seed([1u8; 32]);
-        let mut current = start;
-        std::iter::from_fn(move || {
-            current += rng.gen_range(avg_step_size - avg_deviation..=avg_step_size + avg_deviation);
-            if current >= end {
-                None
-            } else {
-                Some(current)
-            }
-        })
-    }
-
-    fn n_percent_step_iterator(percent: f32, num_values: u32) -> impl Iterator<Item = u32> {
-        let ratio = percent as f32 / 100.0;
-        let step_size = (1f32 / ratio) as u32;
-        let deviation = step_size - 1;
-        random_range_iterator(0, num_values, step_size, deviation)
-    }
-
-    fn walk_over_data(codec: &OptionalIndex, avg_step_size: u32) -> Option<u32> {
-        walk_over_data_from_positions(
-            codec,
-            random_range_iterator(0, TOTAL_NUM_VALUES, avg_step_size, 0),
-        )
-    }
-
-    fn walk_over_data_from_positions(
-        codec: &OptionalIndex,
-        positions: impl Iterator<Item = u32>,
-    ) -> Option<u32> {
-        let mut dense_idx: Option<u32> = None;
-        for idx in positions {
-            dense_idx = dense_idx.or(codec.rank_if_exists(idx));
-        }
-        dense_idx
-    }
-
-    #[bench]
-    fn bench_translate_orig_to_codec_1percent_filled_10percent_hit(bench: &mut Bencher) {
-        let codec = gen_bools(0.01f64);
-        bench.iter(|| walk_over_data(&codec, 100));
-    }
-
-    #[bench]
-    fn bench_translate_orig_to_codec_5percent_filled_10percent_hit(bench: &mut Bencher) {
-        let codec = gen_bools(0.05f64);
-        bench.iter(|| walk_over_data(&codec, 100));
-    }
-
-    #[bench]
-    fn bench_translate_orig_to_codec_5percent_filled_1percent_hit(bench: &mut Bencher) {
-        let codec = gen_bools(0.05f64);
-        bench.iter(|| walk_over_data(&codec, 1000));
-    }
-
-    #[bench]
-    fn bench_translate_orig_to_codec_full_scan_1percent_filled(bench: &mut Bencher) {
-        let codec = gen_bools(0.01f64);
-        bench.iter(|| walk_over_data_from_positions(&codec, 0..TOTAL_NUM_VALUES));
-    }
-
-    #[bench]
-    fn bench_translate_orig_to_codec_full_scan_10percent_filled(bench: &mut Bencher) {
-        let codec = gen_bools(0.1f64);
-        bench.iter(|| walk_over_data_from_positions(&codec, 0..TOTAL_NUM_VALUES));
-    }
-
-    #[bench]
-    fn bench_translate_orig_to_codec_full_scan_90percent_filled(bench: &mut Bencher) {
-        let codec = gen_bools(0.9f64);
-        bench.iter(|| walk_over_data_from_positions(&codec, 0..TOTAL_NUM_VALUES));
-    }
-
-    #[bench]
-    fn bench_translate_orig_to_codec_10percent_filled_1percent_hit(bench: &mut Bencher) {
-        let codec = gen_bools(0.1f64);
-        bench.iter(|| walk_over_data(&codec, 100));
-    }
-
-    #[bench]
-    fn bench_translate_orig_to_codec_50percent_filled_1percent_hit(bench: &mut Bencher) {
-        let codec = gen_bools(0.5f64);
-        bench.iter(|| walk_over_data(&codec, 100));
-    }
-
-    #[bench]
-    fn bench_translate_orig_to_codec_90percent_filled_1percent_hit(bench: &mut Bencher) {
-        let codec = gen_bools(0.9f64);
-        bench.iter(|| walk_over_data(&codec, 100));
-    }
-
-    #[bench]
-    fn bench_translate_codec_to_orig_1percent_filled_0comma005percent_hit(bench: &mut Bencher) {
-        bench_translate_codec_to_orig_util(0.01f64, 0.005f32, bench);
-    }
-
-    #[bench]
-    fn bench_translate_codec_to_orig_10percent_filled_0comma005percent_hit(bench: &mut Bencher) {
-        bench_translate_codec_to_orig_util(0.1f64, 0.005f32, bench);
-    }
-
-    #[bench]
-    fn bench_translate_codec_to_orig_1percent_filled_10percent_hit(bench: &mut Bencher) {
-        bench_translate_codec_to_orig_util(0.01f64, 10f32, bench);
-    }
-
-    #[bench]
-    fn bench_translate_codec_to_orig_1percent_filled_full_scan(bench: &mut Bencher) {
-        bench_translate_codec_to_orig_util(0.01f64, 100f32, bench);
-    }
-
-    fn bench_translate_codec_to_orig_util(
-        percent_filled: f64,
-        percent_hit: f32,
-        bench: &mut Bencher,
-    ) {
-        let codec = gen_bools(percent_filled);
-        let num_non_nulls = codec.num_non_nulls();
-        let idxs: Vec<u32> = if percent_hit == 100.0f32 {
-            (0..num_non_nulls).collect()
-        } else {
-            n_percent_step_iterator(percent_hit, num_non_nulls).collect()
-        };
-        let mut output = vec![0u32; idxs.len()];
-        bench.iter(|| {
-            output.copy_from_slice(&idxs[..]);
-            codec.select_batch(&mut output);
-        });
-    }
-
-    #[bench]
-    fn bench_translate_codec_to_orig_90percent_filled_0comma005percent_hit(bench: &mut Bencher) {
-        bench_translate_codec_to_orig_util(0.9f64, 0.005, bench);
-    }
-
-    #[bench]
-    fn bench_translate_codec_to_orig_90percent_filled_full_scan(bench: &mut Bencher) {
-        bench_translate_codec_to_orig_util(0.9f64, 100.0f32, bench);
-    }
-}

columnar/src/column_index/serialize.rs

@@ -1,77 +0,0 @@
-use std::io;
-use std::io::Write;
-
-use common::{CountingWriter, OwnedBytes};
-
-use crate::column_index::multivalued_index::serialize_multivalued_index;
-use crate::column_index::optional_index::serialize_optional_index;
-use crate::column_index::ColumnIndex;
-use crate::iterable::Iterable;
-use crate::{Cardinality, RowId};
-
-pub enum SerializableColumnIndex<'a> {
-    Full,
-    Optional {
-        non_null_row_ids: Box<dyn Iterable<RowId> + 'a>,
-        num_rows: RowId,
-    },
-    // TODO remove the Arc<dyn> apart from serialization this is not
-    // dynamic at all.
-    Multivalued(Box<dyn Iterable<RowId> + 'a>),
-}
-
-impl<'a> SerializableColumnIndex<'a> {
-    pub fn get_cardinality(&self) -> Cardinality {
-        match self {
-            SerializableColumnIndex::Full => Cardinality::Full,
-            SerializableColumnIndex::Optional { .. } => Cardinality::Optional,
-            SerializableColumnIndex::Multivalued(_) => Cardinality::Multivalued,
-        }
-    }
-}
-
-pub fn serialize_column_index(
-    column_index: SerializableColumnIndex,
-    output: &mut impl Write,
-) -> io::Result<u32> {
-    let mut output = CountingWriter::wrap(output);
-    let cardinality = column_index.get_cardinality().to_code();
-    output.write_all(&[cardinality])?;
-    match column_index {
-        SerializableColumnIndex::Full => {}
-        SerializableColumnIndex::Optional {
-            non_null_row_ids,
-            num_rows,
-        } => serialize_optional_index(non_null_row_ids.as_ref(), num_rows, &mut output)?,
-        SerializableColumnIndex::Multivalued(multivalued_index) => {
-            serialize_multivalued_index(&*multivalued_index, &mut output)?
-        }
-    }
-    let column_index_num_bytes = output.written_bytes() as u32;
-    Ok(column_index_num_bytes)
-}
-
-pub fn open_column_index(mut bytes: OwnedBytes) -> io::Result<ColumnIndex> {
-    if bytes.is_empty() {
-        return Err(io::Error::new(
-            io::ErrorKind::UnexpectedEof,
-            "Failed to deserialize column index. Empty buffer.",
-        ));
-    }
-    let cardinality_code = bytes[0];
-    let cardinality = Cardinality::try_from_code(cardinality_code)?;
-    bytes.advance(1);
-    match cardinality {
-        Cardinality::Full => Ok(ColumnIndex::Full),
-        Cardinality::Optional => {
-            let optional_index = super::optional_index::open_optional_index(bytes)?;
-            Ok(ColumnIndex::Optional(optional_index))
-        }
-        Cardinality::Multivalued => {
-            let multivalue_index = super::multivalued_index::open_multivalued_index(bytes)?;
-            Ok(ColumnIndex::Multivalued(multivalue_index))
-        }
-    }
-}
-
-// TODO unit tests

columnar/src/column_values/bench.rs

@@ -1,135 +0,0 @@
-use std::sync::Arc;
-
-use common::OwnedBytes;
-use rand::rngs::StdRng;
-use rand::{Rng, SeedableRng};
-use test::{self, Bencher};
-
-use super::*;
-use crate::column_values::u64_based::*;
-
-fn get_data() -> Vec<u64> {
-    let mut rng = StdRng::seed_from_u64(2u64);
-    let mut data: Vec<_> = (100..55000_u64)
-        .map(|num| num + rng.gen::<u8>() as u64)
-        .collect();
-    data.push(99_000);
-    data.insert(1000, 2000);
-    data.insert(2000, 100);
-    data.insert(3000, 4100);
-    data.insert(4000, 100);
-    data.insert(5000, 800);
-    data
-}
-
-fn compute_stats(vals: impl Iterator<Item = u64>) -> ColumnStats {
-    let mut stats_collector = StatsCollector::default();
-    for val in vals {
-        stats_collector.collect(val);
-    }
-    stats_collector.stats()
-}
-
-#[inline(never)]
-fn value_iter() -> impl Iterator<Item = u64> {
-    0..20_000
-}
-fn get_reader_for_bench<Codec: ColumnCodec>(data: &[u64]) -> Codec::ColumnValues {
-    let mut bytes = Vec::new();
-    let stats = compute_stats(data.iter().cloned());
-    let mut codec_serializer = Codec::estimator();
-    for val in data {
-        codec_serializer.collect(*val);
-    }
-    codec_serializer.serialize(&stats, Box::new(data.iter().copied()).as_mut(), &mut bytes);
-
-    Codec::load(OwnedBytes::new(bytes)).unwrap()
-}
-fn bench_get<Codec: ColumnCodec>(b: &mut Bencher, data: &[u64]) {
-    let col = get_reader_for_bench::<Codec>(data);
-    b.iter(|| {
-        let mut sum = 0u64;
-        for pos in value_iter() {
-            let val = col.get_val(pos as u32);
-            sum = sum.wrapping_add(val);
-        }
-        sum
-    });
-}
-
-#[inline(never)]
-fn bench_get_dynamic_helper(b: &mut Bencher, col: Arc<dyn ColumnValues>) {
-    b.iter(|| {
-        let mut sum = 0u64;
-        for pos in value_iter() {
-            let val = col.get_val(pos as u32);
-            sum = sum.wrapping_add(val);
-        }
-        sum
-    });
-}
-
-fn bench_get_dynamic<Codec: ColumnCodec>(b: &mut Bencher, data: &[u64]) {
-    let col = Arc::new(get_reader_for_bench::<Codec>(data));
-    bench_get_dynamic_helper(b, col);
-}
-fn bench_create<Codec: ColumnCodec>(b: &mut Bencher, data: &[u64]) {
-    let stats = compute_stats(data.iter().cloned());
-
-    let mut bytes = Vec::new();
-    b.iter(|| {
-        bytes.clear();
-        let mut codec_serializer = Codec::estimator();
-        for val in data.iter().take(1024) {
-            codec_serializer.collect(*val);
-        }
-
-        codec_serializer.serialize(&stats, Box::new(data.iter().copied()).as_mut(), &mut bytes)
-    });
-}
-
-#[bench]
-fn bench_fastfield_bitpack_create(b: &mut Bencher) {
-    let data: Vec<_> = get_data();
-    bench_create::<BitpackedCodec>(b, &data);
-}
-#[bench]
-fn bench_fastfield_linearinterpol_create(b: &mut Bencher) {
-    let data: Vec<_> = get_data();
-    bench_create::<LinearCodec>(b, &data);
-}
-#[bench]
-fn bench_fastfield_multilinearinterpol_create(b: &mut Bencher) {
-    let data: Vec<_> = get_data();
-    bench_create::<BlockwiseLinearCodec>(b, &data);
-}
-#[bench]
-fn bench_fastfield_bitpack_get(b: &mut Bencher) {
-    let data: Vec<_> = get_data();
-    bench_get::<BitpackedCodec>(b, &data);
-}
-#[bench]
-fn bench_fastfield_bitpack_get_dynamic(b: &mut Bencher) {
-    let data: Vec<_> = get_data();
-    bench_get_dynamic::<BitpackedCodec>(b, &data);
-}
-#[bench]
-fn bench_fastfield_linearinterpol_get(b: &mut Bencher) {
-    let data: Vec<_> = get_data();
-    bench_get::<LinearCodec>(b, &data);
-}
-#[bench]
-fn bench_fastfield_linearinterpol_get_dynamic(b: &mut Bencher) {
-    let data: Vec<_> = get_data();
-    bench_get_dynamic::<LinearCodec>(b, &data);
-}
-#[bench]
-fn bench_fastfield_multilinearinterpol_get(b: &mut Bencher) {
-    let data: Vec<_> = get_data();
-    bench_get::<BlockwiseLinearCodec>(b, &data);
-}
-#[bench]
-fn bench_fastfield_multilinearinterpol_get_dynamic(b: &mut Bencher) {
-    let data: Vec<_> = get_data();
-    bench_get_dynamic::<BlockwiseLinearCodec>(b, &data);
-}

columnar/src/column_values/merge.rs

@@ -1,41 +0,0 @@
-use std::fmt::Debug;
-use std::sync::Arc;
-
-use crate::iterable::Iterable;
-use crate::{ColumnIndex, ColumnValues, MergeRowOrder};
-
-pub(crate) struct MergedColumnValues<'a, T> {
-    pub(crate) column_indexes: &'a [Option<ColumnIndex>],
-    pub(crate) column_values: &'a [Option<Arc<dyn ColumnValues<T>>>],
-    pub(crate) merge_row_order: &'a MergeRowOrder,
-}
-
-impl<'a, T: Copy + PartialOrd + Debug> Iterable<T> for MergedColumnValues<'a, T> {
-    fn boxed_iter(&self) -> Box<dyn Iterator<Item = T> + '_> {
-        match self.merge_row_order {
-            MergeRowOrder::Stack(_) => Box::new(
-                self.column_values
-                    .iter()
-                    .flatten()
-                    .flat_map(|column_value| column_value.iter()),
-            ),
-            MergeRowOrder::Shuffled(shuffle_merge_order) => Box::new(
-                shuffle_merge_order
-                    .iter_new_to_old_row_addrs()
-                    .flat_map(|row_addr| {
-                        let column_index =
-                            self.column_indexes[row_addr.segment_ord as usize].as_ref()?;
-                        let column_values =
-                            self.column_values[row_addr.segment_ord as usize].as_ref()?;
-                        let value_range = column_index.value_row_ids(row_addr.row_id);
-                        Some((value_range, column_values))
-                    })
-                    .flat_map(|(value_range, column_values)| {
-                        value_range
-                            .into_iter()
-                            .map(|val| column_values.get_val(val))
-                    }),
-            ),
-        }
-    }
-}

columnar/src/column_values/mod.rs

@@ -1,220 +0,0 @@
-#![warn(missing_docs)]
-
-//! # `fastfield_codecs`
-//!
-//! - Columnar storage of data for tantivy [`Column`].
-//! - Encode data in different codecs.
-//! - Monotonically map values to u64/u128
-
-use std::fmt::Debug;
-use std::ops::{Range, RangeInclusive};
-use std::sync::Arc;
-
-pub use monotonic_mapping::{MonotonicallyMappableToU64, StrictlyMonotonicFn};
-pub use monotonic_mapping_u128::MonotonicallyMappableToU128;
-
-mod merge;
-pub(crate) mod monotonic_mapping;
-pub(crate) mod monotonic_mapping_u128;
-mod stats;
-mod u128_based;
-mod u64_based;
-mod vec_column;
-
-mod monotonic_column;
-
-pub(crate) use merge::MergedColumnValues;
-pub use stats::ColumnStats;
-pub use u128_based::{open_u128_mapped, serialize_column_values_u128};
-pub use u64_based::{
-    load_u64_based_column_values, serialize_and_load_u64_based_column_values,
-    serialize_u64_based_column_values, CodecType, ALL_U64_CODEC_TYPES,
-};
-pub use vec_column::VecColumn;
-
-pub use self::monotonic_column::monotonic_map_column;
-use crate::RowId;
-
-/// `ColumnValues` provides access to a dense field column.
-///
-/// `Column` are just a wrapper over `ColumnValues` and a `ColumnIndex`.
-///
-/// Any methods with a default and specialized implementation need to be called in the
-/// wrappers that implement the trait: Arc and MonotonicMappingColumn
-pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
-    /// Return the value associated with 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: u32) -> T;
-
-    /// Allows to push down multiple fetch calls, to avoid dynamic dispatch overhead.
-    ///
-    /// idx and output should have the same length
-    ///
-    /// # Panics
-    ///
-    /// May panic if `idx` is greater than the column length.
-    fn get_vals(&self, idx: &[u32], output: &mut [T]) {
-        assert!(idx.len() == output.len());
-        for (out, idx) in output.iter_mut().zip(idx.iter()) {
-            *out = self.get_val(*idx as u32);
-        }
-    }
-
-    /// Fills an output buffer with the fast field values
-    /// associated with the `DocId` going from
-    /// `start` to `start + output.len()`.
-    ///
-    /// # Panics
-    ///
-    /// Must panic if `start + output.len()` is greater than
-    /// the segment's `maxdoc`.
-    #[inline(always)]
-    fn get_range(&self, start: u64, output: &mut [T]) {
-        for (out, idx) in output.iter_mut().zip(start..) {
-            *out = self.get_val(idx as u32);
-        }
-    }
-
-    /// Get the row ids of values which are in the provided value range.
-    ///
-    /// Note that position == docid for single value fast fields
-    #[inline(always)]
-    fn get_row_ids_for_value_range(
-        &self,
-        value_range: RangeInclusive<T>,
-        row_id_range: Range<RowId>,
-        row_id_hits: &mut Vec<RowId>,
-    ) {
-        let row_id_range = row_id_range.start..row_id_range.end.min(self.num_vals());
-        for idx in row_id_range.start..row_id_range.end {
-            let val = self.get_val(idx);
-            if value_range.contains(&val) {
-                row_id_hits.push(idx);
-            }
-        }
-    }
-
-    /// Returns the minimum value for this fast field.
-    ///
-    /// This min_value may not be exact.
-    /// For instance, the min value does not take in account of possible
-    /// deleted document. All values are however guaranteed to be higher than
-    /// `.min_value()`.
-    fn min_value(&self) -> T;
-
-    /// Returns the maximum value for this fast field.
-    ///
-    /// This max_value may not be exact.
-    /// For instance, the max value does not take in account of possible
-    /// deleted document. All values are however guaranteed to be higher than
-    /// `.max_value()`.
-    fn max_value(&self) -> T;
-
-    /// The number of values in the column.
-    fn num_vals(&self) -> u32;
-
-    /// 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)))
-    }
-}
-
-impl<T: Copy + PartialOrd + Debug> ColumnValues<T> for Arc<dyn ColumnValues<T>> {
-    #[inline(always)]
-    fn get_val(&self, idx: u32) -> T {
-        self.as_ref().get_val(idx)
-    }
-
-    #[inline(always)]
-    fn min_value(&self) -> T {
-        self.as_ref().min_value()
-    }
-
-    #[inline(always)]
-    fn max_value(&self) -> T {
-        self.as_ref().max_value()
-    }
-
-    #[inline(always)]
-    fn num_vals(&self) -> u32 {
-        self.as_ref().num_vals()
-    }
-
-    #[inline(always)]
-    fn iter<'b>(&'b self) -> Box<dyn Iterator<Item = T> + 'b> {
-        self.as_ref().iter()
-    }
-
-    #[inline(always)]
-    fn get_range(&self, start: u64, output: &mut [T]) {
-        self.as_ref().get_range(start, output)
-    }
-
-    #[inline(always)]
-    fn get_row_ids_for_value_range(
-        &self,
-        range: RangeInclusive<T>,
-        doc_id_range: Range<u32>,
-        positions: &mut Vec<u32>,
-    ) {
-        self.as_ref()
-            .get_row_ids_for_value_range(range, doc_id_range, positions)
-    }
-}
-
-/// Wraps an cloneable iterator into a `Column`.
-pub struct IterColumn<T>(T);
-
-impl<T> From<T> for IterColumn<T>
-where T: Iterator + Clone + ExactSizeIterator
-{
-    fn from(iter: T) -> Self {
-        IterColumn(iter)
-    }
-}
-
-impl<T> ColumnValues<T::Item> for IterColumn<T>
-where
-    T: Iterator + Clone + ExactSizeIterator + Send + Sync,
-    T::Item: PartialOrd + Debug,
-{
-    fn get_val(&self, idx: u32) -> T::Item {
-        self.0.clone().nth(idx as usize).unwrap()
-    }
-
-    fn min_value(&self) -> T::Item {
-        self.0.clone().next().unwrap()
-    }
-
-    fn max_value(&self) -> T::Item {
-        self.0.clone().last().unwrap()
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.0.len() as u32
-    }
-
-    fn iter(&self) -> Box<dyn Iterator<Item = T::Item> + '_> {
-        Box::new(self.0.clone())
-    }
-}
-
-#[cfg(all(test, feature = "unstable"))]
-mod bench;
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_range_as_col() {
-        let col = IterColumn::from(10..100);
-        assert_eq!(col.num_vals(), 90);
-        assert_eq!(col.max_value(), 99);
-    }
-}

columnar/src/column_values/monotonic_column.rs

@@ -1,120 +0,0 @@
-use std::fmt::Debug;
-use std::marker::PhantomData;
-use std::ops::{Range, RangeInclusive};
-
-use crate::column_values::monotonic_mapping::StrictlyMonotonicFn;
-use crate::ColumnValues;
-
-struct MonotonicMappingColumn<C, T, Input> {
-    from_column: C,
-    monotonic_mapping: T,
-    _phantom: PhantomData<Input>,
-}
-
-/// Creates a view of a column transformed by a strictly monotonic mapping. See
-/// [`StrictlyMonotonicFn`].
-///
-/// E.g. apply a gcd monotonic_mapping([100, 200, 300]) == [1, 2, 3]
-/// monotonic_mapping.mapping() is expected to be injective, and we should always have
-/// monotonic_mapping.inverse(monotonic_mapping.mapping(el)) == el
-///
-/// The inverse of the mapping is required for:
-/// `fn get_positions_for_value_range(&self, range: RangeInclusive<T>) -> Vec<u64> `
-/// The user provides the original value range and we need to monotonic map them in the same way the
-/// serialization does before calling the underlying column.
-///
-/// Note that when opening a codec, the monotonic_mapping should be the inverse of the mapping
-/// during serialization. And therefore the monotonic_mapping_inv when opening is the same as
-/// monotonic_mapping during serialization.
-pub fn monotonic_map_column<C, T, Input, Output>(
-    from_column: C,
-    monotonic_mapping: T,
-) -> impl ColumnValues<Output>
-where
-    C: ColumnValues<Input>,
-    T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
-    Input: PartialOrd + Debug + Send + Sync + Clone,
-    Output: PartialOrd + Debug + Send + Sync + Clone,
-{
-    MonotonicMappingColumn {
-        from_column,
-        monotonic_mapping,
-        _phantom: PhantomData,
-    }
-}
-
-impl<C, T, Input, Output> ColumnValues<Output> for MonotonicMappingColumn<C, T, Input>
-where
-    C: ColumnValues<Input>,
-    T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
-    Input: PartialOrd + Send + Debug + Sync + Clone,
-    Output: PartialOrd + Send + Debug + Sync + Clone,
-{
-    #[inline]
-    fn get_val(&self, idx: u32) -> Output {
-        let from_val = self.from_column.get_val(idx);
-        self.monotonic_mapping.mapping(from_val)
-    }
-
-    fn min_value(&self) -> Output {
-        let from_min_value = self.from_column.min_value();
-        self.monotonic_mapping.mapping(from_min_value)
-    }
-
-    fn max_value(&self) -> Output {
-        let from_max_value = self.from_column.max_value();
-        self.monotonic_mapping.mapping(from_max_value)
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.from_column.num_vals()
-    }
-
-    fn iter(&self) -> Box<dyn Iterator<Item = Output> + '_> {
-        Box::new(
-            self.from_column
-                .iter()
-                .map(|el| self.monotonic_mapping.mapping(el)),
-        )
-    }
-
-    fn get_row_ids_for_value_range(
-        &self,
-        range: RangeInclusive<Output>,
-        doc_id_range: Range<u32>,
-        positions: &mut Vec<u32>,
-    ) {
-        self.from_column.get_row_ids_for_value_range(
-            self.monotonic_mapping.inverse(range.start().clone())
-                ..=self.monotonic_mapping.inverse(range.end().clone()),
-            doc_id_range,
-            positions,
-        )
-    }
-
-    // We voluntarily do not implement get_range as it yields a regression,
-    // and we do not have any specialized implementation anyway.
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::column_values::monotonic_mapping::{
-        StrictlyMonotonicMappingInverter, StrictlyMonotonicMappingToInternal,
-    };
-    use crate::column_values::VecColumn;
-
-    #[test]
-    fn test_monotonic_mapping_iter() {
-        let vals: Vec<u64> = (0..100u64).map(|el| el * 10).collect();
-        let col = VecColumn::from(&vals);
-        let mapped = monotonic_map_column(
-            col,
-            StrictlyMonotonicMappingInverter::from(StrictlyMonotonicMappingToInternal::<i64>::new()),
-        );
-        let val_i64s: Vec<u64> = mapped.iter().collect();
-        for i in 0..100 {
-            assert_eq!(val_i64s[i as usize], mapped.get_val(i));
-        }
-    }
-}

columnar/src/column_values/monotonic_mapping.rs

@@ -1,211 +0,0 @@
-use std::fmt::Debug;
-use std::marker::PhantomData;
-
-use common::DateTime;
-
-use super::MonotonicallyMappableToU128;
-use crate::RowId;
-
-/// Monotonic maps a value to u64 value space.
-/// Monotonic mapping enables `PartialOrd` on u64 space without conversion to original space.
-pub trait MonotonicallyMappableToU64: 'static + PartialOrd + Debug + Copy + Send + Sync {
-    /// Converts a value to u64.
-    ///
-    /// Internally all fast field values are encoded as u64.
-    fn to_u64(self) -> u64;
-
-    /// Converts a value from u64
-    ///
-    /// Internally all fast field values are encoded as u64.
-    /// **Note: To be used for converting encoded Term, Posting values.**
-    fn from_u64(val: u64) -> Self;
-}
-
-/// Values need to be strictly monotonic mapped to a `Internal` value (u64 or u128) that can be
-/// used in fast field codecs.
-///
-/// The monotonic mapping is required so that `PartialOrd` can be used on `Internal` without
-/// converting to `External`.
-///
-/// All strictly monotonic functions are invertible because they are guaranteed to have a one-to-one
-/// mapping from their range to their domain. The `inverse` method is required when opening a codec,
-/// so a value can be converted back to its original domain (e.g. ip address or f64) from its
-/// internal representation.
-pub trait StrictlyMonotonicFn<External, Internal> {
-    /// Strictly monotonically maps the value from External to Internal.
-    fn mapping(&self, inp: External) -> Internal;
-    /// Inverse of `mapping`. Maps the value from Internal to External.
-    fn inverse(&self, out: Internal) -> External;
-}
-
-/// Inverts a strictly monotonic mapping from `StrictlyMonotonicFn<A, B>` to
-/// `StrictlyMonotonicFn<B, A>`.
-///
-/// # Warning
-///
-/// This type comes with a footgun. A type being strictly monotonic does not impose that the inverse
-/// mapping is strictly monotonic over the entire space External. e.g. a -> a * 2. Use at your own
-/// risks.
-pub(crate) struct StrictlyMonotonicMappingInverter<T> {
-    orig_mapping: T,
-}
-impl<T> From<T> for StrictlyMonotonicMappingInverter<T> {
-    fn from(orig_mapping: T) -> Self {
-        Self { orig_mapping }
-    }
-}
-
-impl<From, To, T> StrictlyMonotonicFn<To, From> for StrictlyMonotonicMappingInverter<T>
-where T: StrictlyMonotonicFn<From, To>
-{
-    #[inline(always)]
-    fn mapping(&self, val: To) -> From {
-        self.orig_mapping.inverse(val)
-    }
-
-    #[inline(always)]
-    fn inverse(&self, val: From) -> To {
-        self.orig_mapping.mapping(val)
-    }
-}
-
-/// Applies the strictly monotonic mapping from `T` without any additional changes.
-pub(crate) struct StrictlyMonotonicMappingToInternal<T> {
-    _phantom: PhantomData<T>,
-}
-
-impl<T> StrictlyMonotonicMappingToInternal<T> {
-    pub(crate) fn new() -> StrictlyMonotonicMappingToInternal<T> {
-        Self {
-            _phantom: PhantomData,
-        }
-    }
-}
-
-impl<External: MonotonicallyMappableToU128, T: MonotonicallyMappableToU128>
-    StrictlyMonotonicFn<External, u128> for StrictlyMonotonicMappingToInternal<T>
-where T: MonotonicallyMappableToU128
-{
-    #[inline(always)]
-    fn mapping(&self, inp: External) -> u128 {
-        External::to_u128(inp)
-    }
-
-    #[inline(always)]
-    fn inverse(&self, out: u128) -> External {
-        External::from_u128(out)
-    }
-}
-
-impl<External: MonotonicallyMappableToU64, T: MonotonicallyMappableToU64>
-    StrictlyMonotonicFn<External, u64> for StrictlyMonotonicMappingToInternal<T>
-where T: MonotonicallyMappableToU64
-{
-    #[inline(always)]
-    fn mapping(&self, inp: External) -> u64 {
-        External::to_u64(inp)
-    }
-
-    #[inline(always)]
-    fn inverse(&self, out: u64) -> External {
-        External::from_u64(out)
-    }
-}
-
-impl MonotonicallyMappableToU64 for u64 {
-    #[inline(always)]
-    fn to_u64(self) -> u64 {
-        self
-    }
-
-    #[inline(always)]
-    fn from_u64(val: u64) -> Self {
-        val
-    }
-}
-
-impl MonotonicallyMappableToU64 for i64 {
-    #[inline(always)]
-    fn to_u64(self) -> u64 {
-        common::i64_to_u64(self)
-    }
-
-    #[inline(always)]
-    fn from_u64(val: u64) -> Self {
-        common::u64_to_i64(val)
-    }
-}
-
-impl MonotonicallyMappableToU64 for DateTime {
-    #[inline(always)]
-    fn to_u64(self) -> u64 {
-        common::i64_to_u64(self.into_timestamp_micros())
-    }
-
-    #[inline(always)]
-    fn from_u64(val: u64) -> Self {
-        DateTime::from_timestamp_micros(common::u64_to_i64(val))
-    }
-}
-
-impl MonotonicallyMappableToU64 for bool {
-    #[inline(always)]
-    fn to_u64(self) -> u64 {
-        u64::from(self)
-    }
-
-    #[inline(always)]
-    fn from_u64(val: u64) -> Self {
-        val > 0
-    }
-}
-
-impl MonotonicallyMappableToU64 for RowId {
-    #[inline(always)]
-    fn to_u64(self) -> u64 {
-        u64::from(self)
-    }
-
-    #[inline(always)]
-    fn from_u64(val: u64) -> RowId {
-        val as RowId
-    }
-}
-
-// TODO remove me.
-// Tantivy should refuse NaN values and work with NotNaN internally.
-impl MonotonicallyMappableToU64 for f64 {
-    #[inline(always)]
-    fn to_u64(self) -> u64 {
-        common::f64_to_u64(self)
-    }
-
-    #[inline(always)]
-    fn from_u64(val: u64) -> Self {
-        common::u64_to_f64(val)
-    }
-}
-
-#[cfg(test)]
-mod tests {
-
-    use super::*;
-
-    #[test]
-    fn strictly_monotonic_test() {
-        // identity mapping
-        test_round_trip(&StrictlyMonotonicMappingToInternal::<u64>::new(), 100u64);
-        // round trip to i64
-        test_round_trip(&StrictlyMonotonicMappingToInternal::<i64>::new(), 100u64);
-        // TODO
-        // identity mapping
-        // test_round_trip(&StrictlyMonotonicMappingToInternal::<u128>::new(), 100u128);
-    }
-
-    fn test_round_trip<T: StrictlyMonotonicFn<K, L>, K: std::fmt::Debug + Eq + Copy, L>(
-        mapping: &T,
-        test_val: K,
-    ) {
-        assert_eq!(mapping.inverse(mapping.mapping(test_val)), test_val);
-    }
-}

columnar/src/column_values/monotonic_mapping_u128.rs

@@ -1,41 +0,0 @@
-use std::fmt::Debug;
-use std::net::Ipv6Addr;
-
-/// Montonic maps a value to u128 value space
-/// Monotonic mapping enables `PartialOrd` on u128 space without conversion to original space.
-pub trait MonotonicallyMappableToU128: 'static + PartialOrd + Copy + Debug + Send + Sync {
-    /// Converts a value to u128.
-    ///
-    /// Internally all fast field values are encoded as u64.
-    fn to_u128(self) -> u128;
-
-    /// Converts a value from u128
-    ///
-    /// Internally all fast field values are encoded as u64.
-    /// **Note: To be used for converting encoded Term, Posting values.**
-    fn from_u128(val: u128) -> Self;
-}
-
-impl MonotonicallyMappableToU128 for u128 {
-    fn to_u128(self) -> u128 {
-        self
-    }
-
-    fn from_u128(val: u128) -> Self {
-        val
-    }
-}
-
-impl MonotonicallyMappableToU128 for Ipv6Addr {
-    fn to_u128(self) -> u128 {
-        ip_to_u128(self)
-    }
-
-    fn from_u128(val: u128) -> Self {
-        Ipv6Addr::from(val.to_be_bytes())
-    }
-}
-
-fn ip_to_u128(ip_addr: Ipv6Addr) -> u128 {
-    u128::from_be_bytes(ip_addr.octets())
-}

columnar/src/column_values/stats.rs

@@ -1,103 +0,0 @@
-use std::io;
-use std::io::Write;
-use std::num::NonZeroU64;
-
-use common::{BinarySerializable, VInt};
-
-use crate::RowId;
-
-/// Column statistics.
-#[derive(Debug, Clone, Eq, PartialEq)]
-pub struct ColumnStats {
-    /// GCD of the elements `el - min(column)`.
-    pub gcd: NonZeroU64,
-    /// Minimum value of the column.
-    pub min_value: u64,
-    /// Maximum value of the column.
-    pub max_value: u64,
-    /// Number of rows in the column.
-    pub num_rows: RowId,
-}
-
-impl ColumnStats {
-    /// Amplitude of value.
-    /// Difference between the maximum and the minimum value.
-    pub fn amplitude(&self) -> u64 {
-        self.max_value - self.min_value
-    }
-}
-
-impl BinarySerializable for ColumnStats {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        VInt(self.min_value).serialize(writer)?;
-        VInt(self.gcd.get()).serialize(writer)?;
-        VInt(self.amplitude() / self.gcd).serialize(writer)?;
-        VInt(self.num_rows as u64).serialize(writer)?;
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let min_value = VInt::deserialize(reader)?.0;
-        let gcd = VInt::deserialize(reader)?.0;
-        let gcd = NonZeroU64::new(gcd)
-            .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidData, "GCD of 0 is forbidden"))?;
-        let amplitude = VInt::deserialize(reader)?.0 * gcd.get();
-        let max_value = min_value + amplitude;
-        let num_rows = VInt::deserialize(reader)?.0 as RowId;
-        Ok(ColumnStats {
-            min_value,
-            max_value,
-            num_rows,
-            gcd,
-        })
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use std::num::NonZeroU64;
-
-    use common::BinarySerializable;
-
-    use crate::column_values::ColumnStats;
-
-    #[track_caller]
-    fn test_stats_ser_deser_aux(stats: &ColumnStats, num_bytes: usize) {
-        let mut buffer: Vec<u8> = Vec::new();
-        stats.serialize(&mut buffer).unwrap();
-        assert_eq!(buffer.len(), num_bytes);
-        let deser_stats = ColumnStats::deserialize(&mut &buffer[..]).unwrap();
-        assert_eq!(stats, &deser_stats);
-    }
-
-    #[test]
-    fn test_stats_serialization() {
-        test_stats_ser_deser_aux(
-            &(ColumnStats {
-                gcd: NonZeroU64::new(3).unwrap(),
-                min_value: 1,
-                max_value: 3001,
-                num_rows: 10,
-            }),
-            5,
-        );
-        test_stats_ser_deser_aux(
-            &(ColumnStats {
-                gcd: NonZeroU64::new(1_000).unwrap(),
-                min_value: 1,
-                max_value: 3001,
-                num_rows: 10,
-            }),
-            5,
-        );
-        test_stats_ser_deser_aux(
-            &(ColumnStats {
-                gcd: NonZeroU64::new(1).unwrap(),
-                min_value: 0,
-                max_value: 0,
-                num_rows: 0,
-            }),
-            4,
-        );
-    }
-}

columnar/src/column_values/u128_based/compact_space/blank_range.rs

@@ -1,43 +0,0 @@
-use std::ops::RangeInclusive;
-
-/// The range of a blank in value space.
-///
-/// A blank is an unoccupied space in the data.
-/// Use try_into() to construct.
-/// A range has to have at least length of 3. Invalid ranges will be rejected.
-///
-/// Ordered by range length.
-#[derive(Debug, Eq, PartialEq, Clone)]
-pub(crate) struct BlankRange {
-    blank_range: RangeInclusive<u128>,
-}
-impl TryFrom<RangeInclusive<u128>> for BlankRange {
-    type Error = &'static str;
-    fn try_from(range: RangeInclusive<u128>) -> Result<Self, Self::Error> {
-        let blank_size = range.end().saturating_sub(*range.start());
-        if blank_size < 2 {
-            Err("invalid range")
-        } else {
-            Ok(BlankRange { blank_range: range })
-        }
-    }
-}
-impl BlankRange {
-    pub(crate) fn blank_size(&self) -> u128 {
-        self.blank_range.end() - self.blank_range.start() + 1
-    }
-    pub(crate) fn blank_range(&self) -> RangeInclusive<u128> {
-        self.blank_range.clone()
-    }
-}
-
-impl Ord for BlankRange {
-    fn cmp(&self, other: &Self) -> std::cmp::Ordering {
-        self.blank_size().cmp(&other.blank_size())
-    }
-}
-impl PartialOrd for BlankRange {
-    fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
-        Some(self.blank_size().cmp(&other.blank_size()))
-    }
-}

columnar/src/column_values/u128_based/compact_space/build_compact_space.rs

@@ -1,231 +0,0 @@
-use std::collections::{BTreeSet, BinaryHeap};
-use std::iter;
-use std::ops::RangeInclusive;
-
-use itertools::Itertools;
-
-use super::blank_range::BlankRange;
-use super::{CompactSpace, RangeMapping};
-
-/// Put the blanks for the sorted values into a binary heap
-fn get_blanks(values_sorted: &BTreeSet<u128>) -> BinaryHeap<BlankRange> {
-    let mut blanks: BinaryHeap<BlankRange> = BinaryHeap::new();
-    for (first, second) in values_sorted.iter().tuple_windows() {
-        // Correctness Overflow: the values are deduped and sorted (BTreeSet property), that means
-        // there's always space between two values.
-        let blank_range = first + 1..=second - 1;
-        let blank_range: Result<BlankRange, _> = blank_range.try_into();
-        if let Ok(blank_range) = blank_range {
-            blanks.push(blank_range);
-        }
-    }
-
-    blanks
-}
-
-struct BlankCollector {
-    blanks: Vec<BlankRange>,
-    staged_blanks_sum: u128,
-}
-impl BlankCollector {
-    fn new() -> Self {
-        Self {
-            blanks: vec![],
-            staged_blanks_sum: 0,
-        }
-    }
-    fn stage_blank(&mut self, blank: BlankRange) {
-        self.staged_blanks_sum += blank.blank_size();
-        self.blanks.push(blank);
-    }
-    fn drain(&mut self) -> impl Iterator<Item = BlankRange> + '_ {
-        self.staged_blanks_sum = 0;
-        self.blanks.drain(..)
-    }
-    fn staged_blanks_sum(&self) -> u128 {
-        self.staged_blanks_sum
-    }
-    fn num_staged_blanks(&self) -> usize {
-        self.blanks.len()
-    }
-}
-fn num_bits(val: u128) -> u8 {
-    (128u32 - val.leading_zeros()) as u8
-}
-
-/// Will collect blanks and add them to compact space if more bits are saved than cost from
-/// metadata.
-pub fn get_compact_space(
-    values_deduped_sorted: &BTreeSet<u128>,
-    total_num_values: u32,
-    cost_per_blank: usize,
-) -> CompactSpace {
-    let mut compact_space_builder = CompactSpaceBuilder::new();
-    if values_deduped_sorted.is_empty() {
-        return compact_space_builder.finish();
-    }
-
-    let mut blanks: BinaryHeap<BlankRange> = get_blanks(values_deduped_sorted);
-    // Replace after stabilization of https://github.com/rust-lang/rust/issues/62924
-
-    // We start by space that's limited to min_value..=max_value
-    let min_value = *values_deduped_sorted.iter().next().unwrap_or(&0);
-    let max_value = *values_deduped_sorted.iter().last().unwrap_or(&0);
-
-    // +1 for null, in case min and max covers the whole space, we are off by one.
-    let mut amplitude_compact_space = (max_value - min_value).saturating_add(1);
-    if min_value != 0 {
-        compact_space_builder.add_blanks(iter::once(0..=min_value - 1));
-    }
-    if max_value != u128::MAX {
-        compact_space_builder.add_blanks(iter::once(max_value + 1..=u128::MAX));
-    }
-
-    let mut amplitude_bits: u8 = num_bits(amplitude_compact_space);
-
-    let mut blank_collector = BlankCollector::new();
-    // We will stage blanks until they reduce the compact space by at least 1 bit and then flush
-    // them if the metadata cost is lower than the total number of saved bits.
-    // Binary heap to process the gaps by their size
-    while let Some(blank_range) = blanks.pop() {
-        blank_collector.stage_blank(blank_range);
-
-        let staged_spaces_sum: u128 = blank_collector.staged_blanks_sum();
-        let amplitude_new_compact_space = amplitude_compact_space - staged_spaces_sum;
-        let amplitude_new_bits = num_bits(amplitude_new_compact_space);
-        if amplitude_bits == amplitude_new_bits {
-            continue;
-        }
-        let saved_bits = (amplitude_bits - amplitude_new_bits) as usize * total_num_values as usize;
-        // TODO: Maybe calculate exact cost of blanks and run this more expensive computation only,
-        // when amplitude_new_bits changes
-        let cost = blank_collector.num_staged_blanks() * cost_per_blank;
-        if cost >= saved_bits {
-            // Continue here, since although we walk over the blanks by size,
-            // we can potentially save a lot at the last bits, which are smaller blanks
-            //
-            // E.g. if the first range reduces the compact space by 1000 from 2000 to 1000, which
-            // saves 11-10=1 bit and the next range reduces the compact space by 950 to
-            // 50, which saves 10-6=4 bit
-            continue;
-        }
-
-        amplitude_compact_space = amplitude_new_compact_space;
-        amplitude_bits = amplitude_new_bits;
-        compact_space_builder.add_blanks(blank_collector.drain().map(|blank| blank.blank_range()));
-    }
-
-    // special case, when we don't collected any blanks because:
-    // * the data is empty (early exit)
-    // * the algorithm did decide it's not worth the cost, which can be the case for single values
-    //
-    // We drain one collected blank unconditionally, so the empty case is reserved for empty
-    // data, and therefore empty compact_space means the data is empty and no data is covered
-    // (conversely to all data) and we can assign null to it.
-    if compact_space_builder.is_empty() {
-        compact_space_builder.add_blanks(
-            blank_collector
-                .drain()
-                .map(|blank| blank.blank_range())
-                .take(1),
-        );
-    }
-
-    let compact_space = compact_space_builder.finish();
-    if max_value - min_value != u128::MAX {
-        debug_assert_eq!(
-            compact_space.amplitude_compact_space(),
-            amplitude_compact_space
-        );
-    }
-    compact_space
-}
-
-#[derive(Debug, Clone, Eq, PartialEq)]
-struct CompactSpaceBuilder {
-    blanks: Vec<RangeInclusive<u128>>,
-}
-
-impl CompactSpaceBuilder {
-    /// Creates a new compact space builder which will initially cover the whole space.
-    fn new() -> Self {
-        Self { blanks: Vec::new() }
-    }
-
-    /// Assumes that repeated add_blank calls don't overlap and are not adjacent,
-    /// e.g. [3..=5, 5..=10] is not allowed
-    ///
-    /// Both of those assumptions are true when blanks are produced from sorted values.
-    fn add_blanks(&mut self, blank: impl Iterator<Item = RangeInclusive<u128>>) {
-        self.blanks.extend(blank);
-    }
-
-    fn is_empty(&self) -> bool {
-        self.blanks.is_empty()
-    }
-
-    /// Convert blanks to covered space and assign null value
-    fn finish(mut self) -> CompactSpace {
-        // sort by start. ranges are not allowed to overlap
-        self.blanks.sort_unstable_by_key(|blank| *blank.start());
-
-        let mut covered_space = Vec::with_capacity(self.blanks.len());
-
-        // begining of the blanks
-        if let Some(first_blank_start) = self.blanks.first().map(RangeInclusive::start) {
-            if *first_blank_start != 0 {
-                covered_space.push(0..=first_blank_start - 1);
-            }
-        }
-
-        // Between the blanks
-        let between_blanks = self.blanks.iter().tuple_windows().map(|(left, right)| {
-            assert!(
-                left.end() < right.start(),
-                "overlapping or adjacent ranges detected"
-            );
-            *left.end() + 1..=*right.start() - 1
-        });
-        covered_space.extend(between_blanks);
-
-        // end of the blanks
-        if let Some(last_blank_end) = self.blanks.last().map(RangeInclusive::end) {
-            if *last_blank_end != u128::MAX {
-                covered_space.push(last_blank_end + 1..=u128::MAX);
-            }
-        }
-
-        if covered_space.is_empty() {
-            covered_space.push(0..=0); // empty data case
-        };
-
-        let mut compact_start: u64 = 1; // 0 is reserved for `null`
-        let mut ranges_mapping: Vec<RangeMapping> = Vec::with_capacity(covered_space.len());
-        for cov in covered_space {
-            let range_mapping = super::RangeMapping {
-                value_range: cov,
-                compact_start,
-            };
-            let covered_range_len = range_mapping.range_length();
-            ranges_mapping.push(range_mapping);
-            compact_start += covered_range_len;
-        }
-        // println!("num ranges {}", ranges_mapping.len());
-        CompactSpace { ranges_mapping }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_binary_heap_pop_order() {
-        let mut blanks: BinaryHeap<BlankRange> = BinaryHeap::new();
-        blanks.push((0..=10).try_into().unwrap());
-        blanks.push((100..=200).try_into().unwrap());
-        blanks.push((100..=110).try_into().unwrap());
-        assert_eq!(blanks.pop().unwrap().blank_size(), 101);
-        assert_eq!(blanks.pop().unwrap().blank_size(), 11);
-    }
-}

columnar/src/column_values/u128_based/compact_space/mod.rs

@@ -1,809 +0,0 @@
-/// This codec takes a large number space (u128) and reduces it to a compact number space.
-///
-/// It will find spaces in the number range. For example:
-///
-/// 100, 101, 102, 103, 104, 50000, 50001
-/// could be mapped to
-/// 100..104 -> 0..4
-/// 50000..50001 -> 5..6
-///
-/// Compact space 0..=6 requires much less bits than 100..=50001
-///
-/// The codec is created to compress ip addresses, but may be employed in other use cases.
-use std::{
-    cmp::Ordering,
-    collections::BTreeSet,
-    io::{self, Write},
-    ops::{Range, RangeInclusive},
-};
-
-mod blank_range;
-mod build_compact_space;
-
-use build_compact_space::get_compact_space;
-use common::{BinarySerializable, CountingWriter, OwnedBytes, VInt, VIntU128};
-use tantivy_bitpacker::{self, BitPacker, BitUnpacker};
-
-use crate::column_values::ColumnValues;
-use crate::RowId;
-
-/// The cost per blank is quite hard actually, since blanks are delta encoded, the actual cost of
-/// blanks depends on the number of blanks.
-///
-/// The number is taken by looking at a real dataset. It is optimized for larger datasets.
-const COST_PER_BLANK_IN_BITS: usize = 36;
-
-#[derive(Debug, Clone, Eq, PartialEq)]
-pub struct CompactSpace {
-    ranges_mapping: Vec<RangeMapping>,
-}
-
-/// Maps the range from the original space to compact_start + range.len()
-#[derive(Debug, Clone, Eq, PartialEq)]
-struct RangeMapping {
-    value_range: RangeInclusive<u128>,
-    compact_start: u64,
-}
-impl RangeMapping {
-    fn range_length(&self) -> u64 {
-        (self.value_range.end() - self.value_range.start()) as u64 + 1
-    }
-
-    // The last value of the compact space in this range
-    fn compact_end(&self) -> u64 {
-        self.compact_start + self.range_length() - 1
-    }
-}
-
-impl BinarySerializable for CompactSpace {
-    fn serialize<W: io::Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        VInt(self.ranges_mapping.len() as u64).serialize(writer)?;
-
-        let mut prev_value = 0;
-        for value_range in self
-            .ranges_mapping
-            .iter()
-            .map(|range_mapping| &range_mapping.value_range)
-        {
-            let blank_delta_start = value_range.start() - prev_value;
-            VIntU128(blank_delta_start).serialize(writer)?;
-            prev_value = *value_range.start();
-
-            let blank_delta_end = value_range.end() - prev_value;
-            VIntU128(blank_delta_end).serialize(writer)?;
-            prev_value = *value_range.end();
-        }
-
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let num_ranges = VInt::deserialize(reader)?.0;
-        let mut ranges_mapping: Vec<RangeMapping> = vec![];
-        let mut value = 0u128;
-        let mut compact_start = 1u64; // 0 is reserved for `null`
-        for _ in 0..num_ranges {
-            let blank_delta_start = VIntU128::deserialize(reader)?.0;
-            value += blank_delta_start;
-            let blank_start = value;
-
-            let blank_delta_end = VIntU128::deserialize(reader)?.0;
-            value += blank_delta_end;
-            let blank_end = value;
-
-            let range_mapping = RangeMapping {
-                value_range: blank_start..=blank_end,
-                compact_start,
-            };
-            let range_length = range_mapping.range_length();
-            ranges_mapping.push(range_mapping);
-            compact_start += range_length;
-        }
-
-        Ok(Self { ranges_mapping })
-    }
-}
-
-impl CompactSpace {
-    /// Amplitude is the value range of the compact space including the sentinel value used to
-    /// identify null values. The compact space is 0..=amplitude .
-    ///
-    /// It's only used to verify we don't exceed u64 number space, which would indicate a bug.
-    fn amplitude_compact_space(&self) -> u128 {
-        self.ranges_mapping
-            .last()
-            .map(|last_range| last_range.compact_end() as u128)
-            .unwrap_or(1) // compact space starts at 1, 0 == null
-    }
-
-    fn get_range_mapping(&self, pos: usize) -> &RangeMapping {
-        &self.ranges_mapping[pos]
-    }
-
-    /// Returns either Ok(the value in the compact space) or if it is outside the compact space the
-    /// Err(position where it would be inserted)
-    fn u128_to_compact(&self, value: u128) -> Result<u64, usize> {
-        self.ranges_mapping
-            .binary_search_by(|probe| {
-                let value_range = &probe.value_range;
-                if value < *value_range.start() {
-                    Ordering::Greater
-                } else if value > *value_range.end() {
-                    Ordering::Less
-                } else {
-                    Ordering::Equal
-                }
-            })
-            .map(|pos| {
-                let range_mapping = &self.ranges_mapping[pos];
-                let pos_in_range = (value - range_mapping.value_range.start()) as u64;
-                range_mapping.compact_start + pos_in_range
-            })
-    }
-
-    /// Unpacks a value from compact space u64 to u128 space
-    fn compact_to_u128(&self, compact: u64) -> u128 {
-        let pos = self
-            .ranges_mapping
-            .binary_search_by_key(&compact, |range_mapping| range_mapping.compact_start)
-            // Correctness: Overflow. The first range starts at compact space 0, the error from
-            // binary search can never be 0
-            .map_or_else(|e| e - 1, |v| v);
-
-        let range_mapping = &self.ranges_mapping[pos];
-        let diff = compact - range_mapping.compact_start;
-        range_mapping.value_range.start() + diff as u128
-    }
-}
-
-pub struct CompactSpaceCompressor {
-    params: IPCodecParams,
-}
-
-#[derive(Debug, Clone)]
-pub struct IPCodecParams {
-    compact_space: CompactSpace,
-    bit_unpacker: BitUnpacker,
-    min_value: u128,
-    max_value: u128,
-    num_vals: RowId,
-    num_bits: u8,
-}
-
-impl CompactSpaceCompressor {
-    pub fn num_vals(&self) -> RowId {
-        self.params.num_vals
-    }
-
-    /// Taking the vals as Vec may cost a lot of memory. It is used to sort the vals.
-    pub fn train_from(iter: impl Iterator<Item = u128>) -> Self {
-        let mut values_sorted = BTreeSet::new();
-        let mut total_num_values = 0u32;
-        for val in iter {
-            total_num_values += 1u32;
-            values_sorted.insert(val);
-        }
-        let compact_space =
-            get_compact_space(&values_sorted, total_num_values, COST_PER_BLANK_IN_BITS);
-        let amplitude_compact_space = compact_space.amplitude_compact_space();
-
-        assert!(
-            amplitude_compact_space <= u64::MAX as u128,
-            "case unsupported."
-        );
-
-        let num_bits = tantivy_bitpacker::compute_num_bits(amplitude_compact_space as u64);
-        let min_value = *values_sorted.iter().next().unwrap_or(&0);
-        let max_value = *values_sorted.iter().last().unwrap_or(&0);
-        assert_eq!(
-            compact_space
-                .u128_to_compact(max_value)
-                .expect("could not convert max value to compact space"),
-            amplitude_compact_space as u64
-        );
-        CompactSpaceCompressor {
-            params: IPCodecParams {
-                compact_space,
-                bit_unpacker: BitUnpacker::new(num_bits),
-                min_value,
-                max_value,
-                num_vals: total_num_values,
-                num_bits,
-            },
-        }
-    }
-
-    fn write_footer(self, writer: &mut impl Write) -> io::Result<()> {
-        let writer = &mut CountingWriter::wrap(writer);
-        self.params.serialize(writer)?;
-
-        let footer_len = writer.written_bytes() as u32;
-        footer_len.serialize(writer)?;
-
-        Ok(())
-    }
-
-    pub fn compress_into(
-        self,
-        vals: impl Iterator<Item = u128>,
-        write: &mut impl Write,
-    ) -> io::Result<()> {
-        let mut bitpacker = BitPacker::default();
-        for val in vals {
-            let compact = self
-                .params
-                .compact_space
-                .u128_to_compact(val)
-                .map_err(|_| {
-                    io::Error::new(
-                        io::ErrorKind::InvalidData,
-                        "Could not convert value to compact_space. This is a bug.",
-                    )
-                })?;
-            bitpacker.write(compact, self.params.num_bits, write)?;
-        }
-        bitpacker.close(write)?;
-        self.write_footer(write)?;
-        Ok(())
-    }
-}
-
-#[derive(Debug, Clone)]
-pub struct CompactSpaceDecompressor {
-    data: OwnedBytes,
-    params: IPCodecParams,
-}
-
-impl BinarySerializable for IPCodecParams {
-    fn serialize<W: io::Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        // header flags for future optional dictionary encoding
-        let footer_flags = 0u64;
-        footer_flags.serialize(writer)?;
-
-        VIntU128(self.min_value).serialize(writer)?;
-        VIntU128(self.max_value).serialize(writer)?;
-        VIntU128(self.num_vals as u128).serialize(writer)?;
-        self.num_bits.serialize(writer)?;
-
-        self.compact_space.serialize(writer)?;
-
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let _header_flags = u64::deserialize(reader)?;
-        let min_value = VIntU128::deserialize(reader)?.0;
-        let max_value = VIntU128::deserialize(reader)?.0;
-        let num_vals = VIntU128::deserialize(reader)?.0 as u32;
-        let num_bits = u8::deserialize(reader)?;
-        let compact_space = CompactSpace::deserialize(reader)?;
-
-        Ok(Self {
-            compact_space,
-            bit_unpacker: BitUnpacker::new(num_bits),
-            min_value,
-            max_value,
-            num_vals,
-            num_bits,
-        })
-    }
-}
-
-impl ColumnValues<u128> for CompactSpaceDecompressor {
-    #[inline]
-    fn get_val(&self, doc: u32) -> u128 {
-        self.get(doc)
-    }
-
-    fn min_value(&self) -> u128 {
-        self.min_value()
-    }
-
-    fn max_value(&self) -> u128 {
-        self.max_value()
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.params.num_vals
-    }
-
-    #[inline]
-    fn iter(&self) -> Box<dyn Iterator<Item = u128> + '_> {
-        Box::new(self.iter())
-    }
-
-    #[inline]
-    fn get_row_ids_for_value_range(
-        &self,
-        value_range: RangeInclusive<u128>,
-        positions_range: Range<u32>,
-        positions: &mut Vec<u32>,
-    ) {
-        self.get_positions_for_value_range(value_range, positions_range, positions)
-    }
-}
-
-impl CompactSpaceDecompressor {
-    pub fn open(data: OwnedBytes) -> io::Result<CompactSpaceDecompressor> {
-        let (data_slice, footer_len_bytes) = data.split_at(data.len() - 4);
-        let footer_len = u32::deserialize(&mut &footer_len_bytes[..])?;
-
-        let data_footer = &data_slice[data_slice.len() - footer_len as usize..];
-        let params = IPCodecParams::deserialize(&mut &data_footer[..])?;
-        let decompressor = CompactSpaceDecompressor { data, params };
-
-        Ok(decompressor)
-    }
-
-    /// Converting to compact space for the decompressor is more complex, since we may get values
-    /// which are outside the compact space. e.g. if we map
-    /// 1000 => 5
-    /// 2000 => 6
-    ///
-    /// and we want a mapping for 1005, there is no equivalent compact space. We instead return an
-    /// error with the index of the next range.
-    fn u128_to_compact(&self, value: u128) -> Result<u64, usize> {
-        self.params.compact_space.u128_to_compact(value)
-    }
-
-    fn compact_to_u128(&self, compact: u64) -> u128 {
-        self.params.compact_space.compact_to_u128(compact)
-    }
-
-    /// Comparing on compact space: Random dataset 0,24 (50% random hit) - 1.05 GElements/s
-    /// Comparing on compact space: Real dataset 1.08 GElements/s
-    ///
-    /// Comparing on original space: Real dataset .06 GElements/s (not completely optimized)
-    #[inline]
-    pub fn get_positions_for_value_range(
-        &self,
-        value_range: RangeInclusive<u128>,
-        position_range: Range<u32>,
-        positions: &mut Vec<u32>,
-    ) {
-        if value_range.start() > value_range.end() {
-            return;
-        }
-        let position_range = position_range.start..position_range.end.min(self.num_vals());
-        let from_value = *value_range.start();
-        let to_value = *value_range.end();
-        assert!(to_value >= from_value);
-        let compact_from = self.u128_to_compact(from_value);
-        let compact_to = self.u128_to_compact(to_value);
-
-        // Quick return, if both ranges fall into the same non-mapped space, the range can't cover
-        // any values, so we can early exit
-        match (compact_to, compact_from) {
-            (Err(pos1), Err(pos2)) if pos1 == pos2 => return,
-            _ => {}
-        }
-
-        let compact_from = compact_from.unwrap_or_else(|pos| {
-            // Correctness: Out of bounds, if this value is Err(last_index + 1), we early exit,
-            // since the to_value also mapps into the same non-mapped space
-            let range_mapping = self.params.compact_space.get_range_mapping(pos);
-            range_mapping.compact_start
-        });
-        // If there is no compact space, we go to the closest upperbound compact space
-        let compact_to = compact_to.unwrap_or_else(|pos| {
-            // Correctness: Overflow, if this value is Err(0), we early exit,
-            // since the from_value also mapps into the same non-mapped space
-
-            // Get end of previous range
-            let pos = pos - 1;
-            let range_mapping = self.params.compact_space.get_range_mapping(pos);
-            range_mapping.compact_end()
-        });
-
-        let range = compact_from..=compact_to;
-
-        let scan_num_docs = position_range.end - position_range.start;
-
-        let step_size = 4;
-        let cutoff = position_range.start + scan_num_docs - scan_num_docs % step_size;
-
-        let mut push_if_in_range = |idx, val| {
-            if range.contains(&val) {
-                positions.push(idx);
-            }
-        };
-        let get_val = |idx| self.params.bit_unpacker.get(idx, &self.data);
-        // unrolled loop
-        for idx in (position_range.start..cutoff).step_by(step_size as usize) {
-            let idx1 = idx;
-            let idx2 = idx + 1;
-            let idx3 = idx + 2;
-            let idx4 = idx + 3;
-            let val1 = get_val(idx1);
-            let val2 = get_val(idx2);
-            let val3 = get_val(idx3);
-            let val4 = get_val(idx4);
-            push_if_in_range(idx1, val1);
-            push_if_in_range(idx2, val2);
-            push_if_in_range(idx3, val3);
-            push_if_in_range(idx4, val4);
-        }
-
-        // handle rest
-        for idx in cutoff..position_range.end {
-            push_if_in_range(idx, get_val(idx));
-        }
-    }
-
-    #[inline]
-    fn iter_compact(&self) -> impl Iterator<Item = u64> + '_ {
-        (0..self.params.num_vals).map(move |idx| self.params.bit_unpacker.get(idx, &self.data))
-    }
-
-    #[inline]
-    fn iter(&self) -> impl Iterator<Item = u128> + '_ {
-        // TODO: Performance. It would be better to iterate on the ranges and check existence via
-        // the bit_unpacker.
-        self.iter_compact()
-            .map(|compact| self.compact_to_u128(compact))
-    }
-
-    #[inline]
-    pub fn get(&self, idx: u32) -> u128 {
-        let compact = self.params.bit_unpacker.get(idx, &self.data);
-        self.compact_to_u128(compact)
-    }
-
-    pub fn min_value(&self) -> u128 {
-        self.params.min_value
-    }
-
-    pub fn max_value(&self) -> u128 {
-        self.params.max_value
-    }
-}
-
-#[cfg(test)]
-mod tests {
-
-    use itertools::Itertools;
-
-    use super::*;
-    use crate::column_values::u128_based::U128Header;
-    use crate::column_values::{open_u128_mapped, serialize_column_values_u128};
-
-    #[test]
-    fn compact_space_test() {
-        let ips = &[
-            2u128, 4u128, 1000, 1001, 1002, 1003, 1004, 1005, 1008, 1010, 1012, 1260,
-        ]
-        .into_iter()
-        .collect();
-        let compact_space = get_compact_space(ips, ips.len() as u32, 11);
-        let amplitude = compact_space.amplitude_compact_space();
-        assert_eq!(amplitude, 17);
-        assert_eq!(1, compact_space.u128_to_compact(2).unwrap());
-        assert_eq!(2, compact_space.u128_to_compact(3).unwrap());
-        assert_eq!(compact_space.u128_to_compact(100).unwrap_err(), 1);
-
-        for (num1, num2) in (0..3).tuple_windows() {
-            assert_eq!(
-                compact_space.get_range_mapping(num1).compact_end() + 1,
-                compact_space.get_range_mapping(num2).compact_start
-            );
-        }
-
-        let mut output: Vec<u8> = Vec::new();
-        compact_space.serialize(&mut output).unwrap();
-
-        assert_eq!(
-            compact_space,
-            CompactSpace::deserialize(&mut &output[..]).unwrap()
-        );
-
-        for ip in ips {
-            let compact = compact_space.u128_to_compact(*ip).unwrap();
-            assert_eq!(compact_space.compact_to_u128(compact), *ip);
-        }
-    }
-
-    #[test]
-    fn compact_space_amplitude_test() {
-        let ips = &[100000u128, 1000000].into_iter().collect();
-        let compact_space = get_compact_space(ips, ips.len() as u32, 1);
-        let amplitude = compact_space.amplitude_compact_space();
-        assert_eq!(amplitude, 2);
-    }
-
-    fn test_all(mut data: OwnedBytes, expected: &[u128]) {
-        let _header = U128Header::deserialize(&mut data);
-        let decompressor = CompactSpaceDecompressor::open(data).unwrap();
-        for (idx, expected_val) in expected.iter().cloned().enumerate() {
-            let val = decompressor.get(idx as u32);
-            assert_eq!(val, expected_val);
-
-            let test_range = |range: RangeInclusive<u128>| {
-                let expected_positions = expected
-                    .iter()
-                    .positions(|val| range.contains(val))
-                    .map(|pos| pos as u32)
-                    .collect::<Vec<_>>();
-                let mut positions = Vec::new();
-                decompressor.get_positions_for_value_range(
-                    range,
-                    0..decompressor.num_vals(),
-                    &mut positions,
-                );
-                assert_eq!(positions, expected_positions);
-            };
-
-            test_range(expected_val.saturating_sub(1)..=expected_val);
-            test_range(expected_val..=expected_val);
-            test_range(expected_val..=expected_val.saturating_add(1));
-            test_range(expected_val.saturating_sub(1)..=expected_val.saturating_add(1));
-        }
-    }
-
-    fn test_aux_vals(u128_vals: &[u128]) -> OwnedBytes {
-        let mut out = Vec::new();
-        serialize_column_values_u128(&u128_vals, &mut out).unwrap();
-        let data = OwnedBytes::new(out);
-        test_all(data.clone(), u128_vals);
-        data
-    }
-
-    #[test]
-    fn test_range_1() {
-        let vals = &[
-            1u128,
-            100u128,
-            3u128,
-            99999u128,
-            100000u128,
-            100001u128,
-            4_000_211_221u128,
-            4_000_211_222u128,
-            333u128,
-        ];
-        let mut data = test_aux_vals(vals);
-
-        let _header = U128Header::deserialize(&mut data);
-        let decomp = CompactSpaceDecompressor::open(data).unwrap();
-        let complete_range = 0..vals.len() as u32;
-        for (pos, val) in vals.iter().enumerate() {
-            let val = *val;
-            let pos = pos as u32;
-            let mut positions = Vec::new();
-            decomp.get_positions_for_value_range(val..=val, pos..pos + 1, &mut positions);
-            assert_eq!(positions, vec![pos]);
-        }
-
-        // handle docid range out of bounds
-        let positions: Vec<u32> = get_positions_for_value_range_helper(&decomp, 0..=1, 1..u32::MAX);
-        assert!(positions.is_empty());
-
-        let positions =
-            get_positions_for_value_range_helper(&decomp, 0..=1, complete_range.clone());
-        assert_eq!(positions, vec![0]);
-        let positions =
-            get_positions_for_value_range_helper(&decomp, 0..=2, complete_range.clone());
-        assert_eq!(positions, vec![0]);
-        let positions =
-            get_positions_for_value_range_helper(&decomp, 0..=3, complete_range.clone());
-        assert_eq!(positions, vec![0, 2]);
-        assert_eq!(
-            get_positions_for_value_range_helper(
-                &decomp,
-                99999u128..=99999u128,
-                complete_range.clone()
-            ),
-            vec![3]
-        );
-        assert_eq!(
-            get_positions_for_value_range_helper(
-                &decomp,
-                99999u128..=100000u128,
-                complete_range.clone()
-            ),
-            vec![3, 4]
-        );
-        assert_eq!(
-            get_positions_for_value_range_helper(
-                &decomp,
-                99998u128..=100000u128,
-                complete_range.clone()
-            ),
-            vec![3, 4]
-        );
-        assert_eq!(
-            &get_positions_for_value_range_helper(
-                &decomp,
-                99998u128..=99999u128,
-                complete_range.clone()
-            ),
-            &[3]
-        );
-        assert!(get_positions_for_value_range_helper(
-            &decomp,
-            99998u128..=99998u128,
-            complete_range.clone()
-        )
-        .is_empty());
-        assert_eq!(
-            &get_positions_for_value_range_helper(
-                &decomp,
-                333u128..=333u128,
-                complete_range.clone()
-            ),
-            &[8]
-        );
-        assert_eq!(
-            &get_positions_for_value_range_helper(
-                &decomp,
-                332u128..=333u128,
-                complete_range.clone()
-            ),
-            &[8]
-        );
-        assert_eq!(
-            &get_positions_for_value_range_helper(
-                &decomp,
-                332u128..=334u128,
-                complete_range.clone()
-            ),
-            &[8]
-        );
-        assert_eq!(
-            &get_positions_for_value_range_helper(
-                &decomp,
-                333u128..=334u128,
-                complete_range.clone()
-            ),
-            &[8]
-        );
-
-        assert_eq!(
-            &get_positions_for_value_range_helper(
-                &decomp,
-                4_000_211_221u128..=5_000_000_000u128,
-                complete_range
-            ),
-            &[6, 7]
-        );
-    }
-
-    #[test]
-    fn test_empty() {
-        let vals = &[];
-        let data = test_aux_vals(vals);
-        let _decomp = CompactSpaceDecompressor::open(data).unwrap();
-    }
-
-    #[test]
-    fn test_range_2() {
-        let vals = &[
-            100u128,
-            99999u128,
-            100000u128,
-            100001u128,
-            4_000_211_221u128,
-            4_000_211_222u128,
-            333u128,
-        ];
-        let mut data = test_aux_vals(vals);
-        let _header = U128Header::deserialize(&mut data);
-        let decomp = CompactSpaceDecompressor::open(data).unwrap();
-        let complete_range = 0..vals.len() as u32;
-        assert!(
-            &get_positions_for_value_range_helper(&decomp, 0..=5, complete_range.clone())
-                .is_empty(),
-        );
-        assert_eq!(
-            &get_positions_for_value_range_helper(&decomp, 0..=100, complete_range.clone()),
-            &[0]
-        );
-        assert_eq!(
-            &get_positions_for_value_range_helper(&decomp, 0..=105, complete_range),
-            &[0]
-        );
-    }
-
-    fn get_positions_for_value_range_helper<C: ColumnValues<T> + ?Sized, T: PartialOrd>(
-        column: &C,
-        value_range: RangeInclusive<T>,
-        doc_id_range: Range<u32>,
-    ) -> Vec<u32> {
-        let mut positions = Vec::new();
-        column.get_row_ids_for_value_range(value_range, doc_id_range, &mut positions);
-        positions
-    }
-
-    #[test]
-    fn test_range_3() {
-        let vals = &[
-            200u128,
-            201,
-            202,
-            203,
-            204,
-            204,
-            206,
-            207,
-            208,
-            209,
-            210,
-            1_000_000,
-            5_000_000_000,
-        ];
-        let mut out = Vec::new();
-        serialize_column_values_u128(&&vals[..], &mut out).unwrap();
-        let decomp = open_u128_mapped(OwnedBytes::new(out)).unwrap();
-        let complete_range = 0..vals.len() as u32;
-
-        assert_eq!(
-            get_positions_for_value_range_helper(&*decomp, 199..=200, complete_range.clone()),
-            vec![0]
-        );
-
-        assert_eq!(
-            get_positions_for_value_range_helper(&*decomp, 199..=201, complete_range.clone()),
-            vec![0, 1]
-        );
-
-        assert_eq!(
-            get_positions_for_value_range_helper(&*decomp, 200..=200, complete_range.clone()),
-            vec![0]
-        );
-
-        assert_eq!(
-            get_positions_for_value_range_helper(&*decomp, 1_000_000..=1_000_000, complete_range),
-            vec![11]
-        );
-    }
-
-    #[test]
-    fn test_bug1() {
-        let vals = &[9223372036854775806];
-        let _data = test_aux_vals(vals);
-    }
-
-    #[test]
-    fn test_bug2() {
-        let vals = &[340282366920938463463374607431768211455u128];
-        let _data = test_aux_vals(vals);
-    }
-
-    #[test]
-    fn test_bug3() {
-        let vals = &[340282366920938463463374607431768211454];
-        let _data = test_aux_vals(vals);
-    }
-
-    #[test]
-    fn test_bug4() {
-        let vals = &[340282366920938463463374607431768211455, 0];
-        let _data = test_aux_vals(vals);
-    }
-
-    #[test]
-    fn test_first_large_gaps() {
-        let vals = &[1_000_000_000u128; 100];
-        let _data = test_aux_vals(vals);
-    }
-
-    use proptest::prelude::*;
-
-    fn num_strategy() -> impl Strategy<Value = u128> {
-        prop_oneof![
-            1 => prop::num::u128::ANY.prop_map(|num| u128::MAX - (num % 10) ),
-            1 => prop::num::u128::ANY.prop_map(|num| i64::MAX as u128 + 5 - (num % 10) ),
-            1 => prop::num::u128::ANY.prop_map(|num| i128::MAX as u128 + 5 - (num % 10) ),
-            1 => prop::num::u128::ANY.prop_map(|num| num % 10 ),
-            20 => prop::num::u128::ANY,
-        ]
-    }
-
-    proptest! {
-        #![proptest_config(ProptestConfig::with_cases(10))]
-
-        #[test]
-        fn compress_decompress_random(vals in proptest::collection::vec(num_strategy() , 1..1000)) {
-            let _data = test_aux_vals(&vals);
-        }
-    }
-}

columnar/src/column_values/u128_based/mod.rs

@@ -1,178 +0,0 @@
-use std::fmt::Debug;
-use std::io;
-use std::io::Write;
-use std::sync::Arc;
-
-mod compact_space;
-
-use common::{BinarySerializable, OwnedBytes, VInt};
-use compact_space::{CompactSpaceCompressor, CompactSpaceDecompressor};
-
-use crate::column_values::monotonic_map_column;
-use crate::column_values::monotonic_mapping::{
-    StrictlyMonotonicMappingInverter, StrictlyMonotonicMappingToInternal,
-};
-use crate::iterable::Iterable;
-use crate::{ColumnValues, MonotonicallyMappableToU128};
-
-#[derive(Debug, Copy, Clone, PartialEq, Eq)]
-pub(crate) struct U128Header {
-    pub num_vals: u32,
-    pub codec_type: U128FastFieldCodecType,
-}
-
-impl BinarySerializable for U128Header {
-    fn serialize<W: io::Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        VInt(self.num_vals as u64).serialize(writer)?;
-        self.codec_type.serialize(writer)?;
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let num_vals = VInt::deserialize(reader)?.0 as u32;
-        let codec_type = U128FastFieldCodecType::deserialize(reader)?;
-        Ok(U128Header {
-            num_vals,
-            codec_type,
-        })
-    }
-}
-
-/// Serializes u128 values with the compact space codec.
-pub fn serialize_column_values_u128<T: MonotonicallyMappableToU128>(
-    iterable: &dyn Iterable<T>,
-    output: &mut impl io::Write,
-) -> io::Result<()> {
-    let compressor = CompactSpaceCompressor::train_from(
-        iterable
-            .boxed_iter()
-            .map(MonotonicallyMappableToU128::to_u128),
-    );
-    let header = U128Header {
-        num_vals: compressor.num_vals(),
-        codec_type: U128FastFieldCodecType::CompactSpace,
-    };
-    header.serialize(output)?;
-    compressor.compress_into(
-        iterable
-            .boxed_iter()
-            .map(MonotonicallyMappableToU128::to_u128),
-        output,
-    )?;
-    Ok(())
-}
-
-#[derive(PartialEq, Eq, PartialOrd, Ord, Debug, Clone, Copy)]
-#[repr(u8)]
-/// Available codecs to use to encode the u128 (via [`MonotonicallyMappableToU128`]) converted data.
-pub(crate) enum U128FastFieldCodecType {
-    /// This codec takes a large number space (u128) and reduces it to a compact number space, by
-    /// removing the holes.
-    CompactSpace = 1,
-}
-
-impl BinarySerializable for U128FastFieldCodecType {
-    fn serialize<W: Write + ?Sized>(&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 U128FastFieldCodecType {
-    pub(crate) fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    pub(crate) fn from_code(code: u8) -> Option<Self> {
-        match code {
-            1 => Some(Self::CompactSpace),
-            _ => None,
-        }
-    }
-}
-
-/// Returns the correct codec reader wrapped in the `Arc` for the data.
-pub fn open_u128_mapped<T: MonotonicallyMappableToU128 + Debug>(
-    mut bytes: OwnedBytes,
-) -> io::Result<Arc<dyn ColumnValues<T>>> {
-    let header = U128Header::deserialize(&mut bytes)?;
-    assert_eq!(header.codec_type, U128FastFieldCodecType::CompactSpace);
-    let reader = CompactSpaceDecompressor::open(bytes)?;
-    let inverted: StrictlyMonotonicMappingInverter<StrictlyMonotonicMappingToInternal<T>> =
-        StrictlyMonotonicMappingToInternal::<T>::new().into();
-    Ok(Arc::new(monotonic_map_column(reader, inverted)))
-}
-#[cfg(test)]
-pub mod tests {
-    use super::*;
-    use crate::column_values::u64_based::{
-        serialize_and_load_u64_based_column_values, serialize_u64_based_column_values,
-        ALL_U64_CODEC_TYPES,
-    };
-    use crate::column_values::CodecType;
-
-    #[test]
-    fn test_serialize_deserialize_u128_header() {
-        let original = U128Header {
-            num_vals: 11,
-            codec_type: U128FastFieldCodecType::CompactSpace,
-        };
-        let mut out = Vec::new();
-        original.serialize(&mut out).unwrap();
-        let restored = U128Header::deserialize(&mut &out[..]).unwrap();
-        assert_eq!(restored, original);
-    }
-
-    #[test]
-    fn test_serialize_deserialize() {
-        let original = [1u64, 5u64, 10u64];
-        let restored: Vec<u64> =
-            serialize_and_load_u64_based_column_values(&&original[..], &ALL_U64_CODEC_TYPES)
-                .iter()
-                .collect();
-        assert_eq!(&restored, &original[..]);
-    }
-
-    #[test]
-    fn test_fastfield_bool_size_bitwidth_1() {
-        let mut buffer = Vec::new();
-        serialize_u64_based_column_values::<bool>(
-            &&[false, true][..],
-            &ALL_U64_CODEC_TYPES,
-            &mut buffer,
-        )
-        .unwrap();
-        // TODO put the header as a footer so that it serves as a padding.
-        // 5 bytes of header, 1 byte of value, 7 bytes of padding.
-        assert_eq!(buffer.len(), 5 + 1);
-    }
-
-    #[test]
-    fn test_fastfield_bool_bit_size_bitwidth_0() {
-        let mut buffer = Vec::new();
-        serialize_u64_based_column_values::<bool>(
-            &&[false, true][..],
-            &ALL_U64_CODEC_TYPES,
-            &mut buffer,
-        )
-        .unwrap();
-        // 6 bytes of header, 0 bytes of value, 7 bytes of padding.
-        assert_eq!(buffer.len(), 6);
-    }
-
-    #[test]
-    fn test_fastfield_gcd() {
-        let mut buffer = Vec::new();
-        let vals: Vec<u64> = (0..80).map(|val| (val % 7) * 1_000u64).collect();
-        serialize_u64_based_column_values(&&vals[..], &[CodecType::Bitpacked], &mut buffer)
-            .unwrap();
-        // Values are stored over 3 bits.
-        assert_eq!(buffer.len(), 6 + (3 * 80 / 8));
-    }
-}

columnar/src/column_values/u64_based/bitpacked.rs

@@ -1,127 +0,0 @@
-use std::io::{self, Write};
-
-use common::{BinarySerializable, OwnedBytes};
-use fastdivide::DividerU64;
-use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
-
-use crate::column_values::u64_based::{ColumnCodec, ColumnCodecEstimator, ColumnStats};
-use crate::{ColumnValues, RowId};
-
-/// Depending on the field type, a different
-/// fast field is required.
-#[derive(Clone)]
-pub struct BitpackedReader {
-    data: OwnedBytes,
-    bit_unpacker: BitUnpacker,
-    stats: ColumnStats,
-}
-
-impl ColumnValues for BitpackedReader {
-    #[inline(always)]
-    fn get_val(&self, doc: u32) -> u64 {
-        self.stats.min_value + self.stats.gcd.get() * self.bit_unpacker.get(doc, &self.data)
-    }
-
-    #[inline]
-    fn min_value(&self) -> u64 {
-        self.stats.min_value
-    }
-    #[inline]
-    fn max_value(&self) -> u64 {
-        self.stats.max_value
-    }
-    #[inline]
-    fn num_vals(&self) -> RowId {
-        self.stats.num_rows
-    }
-}
-
-fn num_bits(stats: &ColumnStats) -> u8 {
-    compute_num_bits(stats.amplitude() / stats.gcd)
-}
-
-#[derive(Default)]
-pub struct BitpackedCodecEstimator;
-
-impl ColumnCodecEstimator for BitpackedCodecEstimator {
-    fn collect(&mut self, _value: u64) {}
-
-    fn estimate(&self, stats: &ColumnStats) -> Option<u64> {
-        let num_bits_per_value = num_bits(stats);
-        Some(stats.num_bytes() + (stats.num_rows as u64 * (num_bits_per_value as u64) + 7) / 8)
-    }
-
-    fn serialize(
-        &self,
-        stats: &ColumnStats,
-        vals: &mut dyn Iterator<Item = u64>,
-        wrt: &mut dyn Write,
-    ) -> io::Result<()> {
-        stats.serialize(wrt)?;
-        let num_bits = num_bits(stats);
-        let mut bit_packer = BitPacker::new();
-        let divider = DividerU64::divide_by(stats.gcd.get());
-        for val in vals {
-            bit_packer.write(divider.divide(val - stats.min_value), num_bits, wrt)?;
-        }
-        bit_packer.close(wrt)?;
-        Ok(())
-    }
-}
-
-pub struct BitpackedCodec;
-
-impl ColumnCodec for BitpackedCodec {
-    type ColumnValues = BitpackedReader;
-    type Estimator = BitpackedCodecEstimator;
-
-    /// Opens a fast field given a file.
-    fn load(mut data: OwnedBytes) -> io::Result<Self::ColumnValues> {
-        let stats = ColumnStats::deserialize(&mut data)?;
-        let num_bits = num_bits(&stats);
-        let bit_unpacker = BitUnpacker::new(num_bits);
-        Ok(BitpackedReader {
-            data,
-            bit_unpacker,
-            stats,
-        })
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::column_values::u64_based::tests::create_and_validate;
-
-    #[test]
-    fn test_with_codec_data_sets_simple() {
-        create_and_validate::<BitpackedCodec>(&[4, 3, 12], "name");
-    }
-
-    #[test]
-    fn test_with_codec_data_sets_simple_gcd() {
-        create_and_validate::<BitpackedCodec>(&[1000, 2000, 3000], "name");
-    }
-
-    #[test]
-    fn test_with_codec_data_sets() {
-        let data_sets = crate::column_values::u64_based::tests::get_codec_test_datasets();
-        for (mut data, name) in data_sets {
-            create_and_validate::<BitpackedCodec>(&data, name);
-            data.reverse();
-            create_and_validate::<BitpackedCodec>(&data, name);
-        }
-    }
-
-    #[test]
-    fn bitpacked_fast_field_rand() {
-        for _ in 0..500 {
-            let mut data = (0..1 + rand::random::<u8>() as usize)
-                .map(|_| rand::random::<i64>() as u64 / 2)
-                .collect::<Vec<_>>();
-            create_and_validate::<BitpackedCodec>(&data, "rand");
-            data.reverse();
-            create_and_validate::<BitpackedCodec>(&data, "rand");
-        }
-    }
-}

columnar/src/column_values/u64_based/blockwise_linear.rs

@@ -1,281 +0,0 @@
-use std::io::Write;
-use std::sync::Arc;
-use std::{io, iter};
-
-use common::{BinarySerializable, CountingWriter, DeserializeFrom, OwnedBytes};
-use fastdivide::DividerU64;
-use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
-
-use crate::column_values::u64_based::line::Line;
-use crate::column_values::u64_based::{ColumnCodec, ColumnCodecEstimator, ColumnStats};
-use crate::column_values::{ColumnValues, VecColumn};
-use crate::MonotonicallyMappableToU64;
-
-const BLOCK_SIZE: u32 = 512u32;
-
-#[derive(Debug, Default)]
-struct Block {
-    line: Line,
-    bit_unpacker: BitUnpacker,
-    data_start_offset: usize,
-}
-
-impl BinarySerializable for Block {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        self.line.serialize(writer)?;
-        self.bit_unpacker.bit_width().serialize(writer)?;
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let line = Line::deserialize(reader)?;
-        let bit_width = u8::deserialize(reader)?;
-        Ok(Block {
-            line,
-            bit_unpacker: BitUnpacker::new(bit_width),
-            data_start_offset: 0,
-        })
-    }
-}
-
-fn compute_num_blocks(num_vals: u32) -> u32 {
-    (num_vals + BLOCK_SIZE - 1) / BLOCK_SIZE
-}
-
-pub struct BlockwiseLinearEstimator {
-    block: Vec<u64>,
-    values_num_bytes: u64,
-    meta_num_bytes: u64,
-}
-
-impl Default for BlockwiseLinearEstimator {
-    fn default() -> Self {
-        Self {
-            block: Vec::with_capacity(BLOCK_SIZE as usize),
-            values_num_bytes: 0u64,
-            meta_num_bytes: 0u64,
-        }
-    }
-}
-
-impl BlockwiseLinearEstimator {
-    fn flush_block_estimate(&mut self) {
-        if self.block.is_empty() {
-            return;
-        }
-        let line = Line::train(&VecColumn::from(&self.block));
-        let mut max_value = 0u64;
-        for (i, buffer_val) in self.block.iter().enumerate() {
-            let interpolated_val = line.eval(i as u32);
-            let val = buffer_val.wrapping_sub(interpolated_val);
-            max_value = val.max(max_value);
-        }
-        let bit_width = compute_num_bits(max_value) as usize;
-        self.values_num_bytes += (bit_width * self.block.len() + 7) as u64 / 8;
-        self.meta_num_bytes += 1 + line.num_bytes();
-    }
-}
-
-impl ColumnCodecEstimator for BlockwiseLinearEstimator {
-    fn collect(&mut self, value: u64) {
-        self.block.push(value);
-        if self.block.len() == BLOCK_SIZE as usize {
-            self.flush_block_estimate();
-            self.block.clear();
-        }
-    }
-    fn estimate(&self, stats: &ColumnStats) -> Option<u64> {
-        let mut estimate = 4 + stats.num_bytes() + self.meta_num_bytes + self.values_num_bytes;
-        if stats.gcd.get() > 1 {
-            let estimate_gain_from_gcd =
-                (stats.gcd.get() as f32).log2().floor() * stats.num_rows as f32 / 8.0f32;
-            estimate = estimate.saturating_sub(estimate_gain_from_gcd as u64);
-        }
-        Some(estimate)
-    }
-
-    fn finalize(&mut self) {
-        self.flush_block_estimate();
-    }
-
-    fn serialize(
-        &self,
-        stats: &ColumnStats,
-        mut vals: &mut dyn Iterator<Item = u64>,
-        wrt: &mut dyn Write,
-    ) -> io::Result<()> {
-        stats.serialize(wrt)?;
-        let mut buffer = Vec::with_capacity(BLOCK_SIZE as usize);
-        let num_blocks = compute_num_blocks(stats.num_rows) as usize;
-        let mut blocks = Vec::with_capacity(num_blocks);
-
-        let mut bit_packer = BitPacker::new();
-
-        let gcd_divider = DividerU64::divide_by(stats.gcd.get());
-
-        for _ in 0..num_blocks {
-            buffer.clear();
-            buffer.extend(
-                (&mut vals)
-                    .map(MonotonicallyMappableToU64::to_u64)
-                    .take(BLOCK_SIZE as usize),
-            );
-
-            for buffer_val in buffer.iter_mut() {
-                *buffer_val = gcd_divider.divide(*buffer_val - stats.min_value);
-            }
-
-            let line = Line::train(&VecColumn::from(&buffer));
-
-            assert!(!buffer.is_empty());
-
-            for (i, buffer_val) in buffer.iter_mut().enumerate() {
-                let interpolated_val = line.eval(i as u32);
-                *buffer_val = buffer_val.wrapping_sub(interpolated_val);
-            }
-
-            let bit_width = buffer.iter().copied().map(compute_num_bits).max().unwrap();
-
-            for &buffer_val in &buffer {
-                bit_packer.write(buffer_val, bit_width, wrt)?;
-            }
-
-            blocks.push(Block {
-                line,
-                bit_unpacker: BitUnpacker::new(bit_width),
-                data_start_offset: 0,
-            });
-        }
-
-        bit_packer.close(wrt)?;
-
-        assert_eq!(blocks.len(), num_blocks);
-
-        let mut counting_wrt = CountingWriter::wrap(wrt);
-        for block in &blocks {
-            block.serialize(&mut counting_wrt)?;
-        }
-        let footer_len = counting_wrt.written_bytes();
-        (footer_len as u32).serialize(&mut counting_wrt)?;
-
-        Ok(())
-    }
-}
-
-pub struct BlockwiseLinearCodec;
-
-impl ColumnCodec<u64> for BlockwiseLinearCodec {
-    type ColumnValues = BlockwiseLinearReader;
-
-    type Estimator = BlockwiseLinearEstimator;
-
-    fn load(mut bytes: OwnedBytes) -> io::Result<Self::ColumnValues> {
-        let stats = ColumnStats::deserialize(&mut bytes)?;
-        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 num_blocks = compute_num_blocks(stats.num_rows);
-        let mut blocks: Vec<Block> = iter::repeat_with(|| Block::deserialize(&mut footer))
-            .take(num_blocks as usize)
-            .collect::<io::Result<_>>()?;
-        let mut start_offset = 0;
-        for block in &mut blocks {
-            block.data_start_offset = start_offset;
-            start_offset += (block.bit_unpacker.bit_width() as usize) * BLOCK_SIZE as usize / 8;
-        }
-        Ok(BlockwiseLinearReader {
-            blocks: blocks.into_boxed_slice().into(),
-            data,
-            stats,
-        })
-    }
-}
-
-#[derive(Clone)]
-pub struct BlockwiseLinearReader {
-    blocks: Arc<[Block]>,
-    data: OwnedBytes,
-    stats: ColumnStats,
-}
-
-impl ColumnValues for BlockwiseLinearReader {
-    #[inline(always)]
-    fn get_val(&self, idx: u32) -> u64 {
-        let block_id = (idx / BLOCK_SIZE) as usize;
-        let idx_within_block = idx % BLOCK_SIZE;
-        let block = &self.blocks[block_id];
-        let interpoled_val: u64 = block.line.eval(idx_within_block);
-        let block_bytes = &self.data[block.data_start_offset..];
-        let bitpacked_diff = block.bit_unpacker.get(idx_within_block, block_bytes);
-        // TODO optimize me! the line parameters could be tweaked to include the multiplication and
-        // remove the dependency.
-        self.stats.min_value
-            + self
-                .stats
-                .gcd
-                .get()
-                .wrapping_mul(interpoled_val.wrapping_add(bitpacked_diff))
-    }
-
-    #[inline(always)]
-    fn min_value(&self) -> u64 {
-        self.stats.min_value
-    }
-
-    #[inline(always)]
-    fn max_value(&self) -> u64 {
-        self.stats.max_value
-    }
-
-    #[inline(always)]
-    fn num_vals(&self) -> u32 {
-        self.stats.num_rows
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::column_values::u64_based::tests::create_and_validate;
-
-    #[test]
-    fn test_with_codec_data_sets_simple() {
-        create_and_validate::<BlockwiseLinearCodec>(
-            &[11, 20, 40, 20, 10, 10, 10, 10, 10, 10],
-            "simple test",
-        )
-        .unwrap();
-    }
-
-    #[test]
-    fn test_with_codec_data_sets_simple_gcd() {
-        let (_, actual_compression_rate) = create_and_validate::<BlockwiseLinearCodec>(
-            &[10, 20, 40, 20, 10, 10, 10, 10, 10, 10],
-            "name",
-        )
-        .unwrap();
-        assert_eq!(actual_compression_rate, 0.175);
-    }
-
-    #[test]
-    fn test_with_codec_data_sets() {
-        let data_sets = crate::column_values::u64_based::tests::get_codec_test_datasets();
-        for (mut data, name) in data_sets {
-            create_and_validate::<BlockwiseLinearCodec>(&data, name);
-            data.reverse();
-            create_and_validate::<BlockwiseLinearCodec>(&data, name);
-        }
-    }
-
-    #[test]
-    fn test_blockwise_linear_fast_field_rand() {
-        for _ in 0..500 {
-            let mut data = (0..1 + rand::random::<u8>() as usize)
-                .map(|_| rand::random::<i64>() as u64 / 2)
-                .collect::<Vec<_>>();
-            create_and_validate::<BlockwiseLinearCodec>(&data, "rand");
-            data.reverse();
-            create_and_validate::<BlockwiseLinearCodec>(&data, "rand");
-        }
-    }
-}

columnar/src/column_values/u64_based/line.rs

@@ -1,210 +0,0 @@
-use std::io;
-use std::num::NonZeroU32;
-
-use common::{BinarySerializable, VInt};
-
-use crate::column_values::ColumnValues;
-
-const MID_POINT: u64 = (1u64 << 32) - 1u64;
-
-/// `Line` describes a line function `y: ax + b` using integer
-/// arithmetics.
-///
-/// The slope is in fact a decimal split into a 32 bit integer value,
-/// and a 32-bit decimal value.
-///
-/// The multiplication then becomes.
-/// `y = m * x >> 32 + b`
-#[derive(Debug, Clone, Copy, Default)]
-pub struct Line {
-    pub(crate) slope: u64,
-    pub(crate) intercept: u64,
-}
-
-/// Compute the line slope.
-///
-/// This function has the nice property of being
-/// invariant by translation.
-/// `
-///   compute_slope(y0, y1)
-/// = compute_slope(y0 + X % 2^64, y1 + X % 2^64)
-/// `
-fn compute_slope(y0: u64, y1: u64, num_vals: NonZeroU32) -> u64 {
-    let dy = y1.wrapping_sub(y0);
-    let sign = dy <= (1 << 63);
-    let abs_dy = if sign {
-        y1.wrapping_sub(y0)
-    } else {
-        y0.wrapping_sub(y1)
-    };
-    if abs_dy >= 1 << 32 {
-        // This is outside of realm we handle.
-        // Let's just bail.
-        return 0u64;
-    }
-
-    let abs_slope = (abs_dy << 32) / num_vals.get() as u64;
-    if sign {
-        abs_slope
-    } else {
-        // The complement does indeed create the
-        // opposite decreasing slope...
-        //
-        // Intuitively (without the bitshifts and % u64::MAX)
-        // ```
-        //    (x + shift)*(u64::MAX - abs_slope)
-        // -  (x * (u64::MAX - abs_slope))
-        // = - shift * abs_slope
-        // ```
-        u64::MAX - abs_slope
-    }
-}
-
-impl Line {
-    #[inline(always)]
-    pub fn eval(&self, x: u32) -> u64 {
-        let linear_part = ((x as u64).wrapping_mul(self.slope) >> 32) as i32 as u64;
-        self.intercept.wrapping_add(linear_part)
-    }
-
-    // Intercept is only computed from provided positions
-    pub fn train_from(
-        first_val: u64,
-        last_val: u64,
-        num_vals: u32,
-        positions_and_values: impl Iterator<Item = (u64, u64)>,
-    ) -> Self {
-        // TODO replace with let else
-        let idx_last_val = if let Some(idx_last_val) = NonZeroU32::new(num_vals - 1) {
-            idx_last_val
-        } else {
-            return Line::default();
-        };
-
-        let y0 = first_val;
-        let y1 = last_val;
-
-        // We first independently pick our slope.
-        let slope = compute_slope(y0, y1, idx_last_val);
-
-        // We picked our slope. Note that it does not have to be perfect.
-        // Now we need to compute the best intercept.
-        //
-        // Intuitively, the best intercept is such that line passes through one of the
-        // `(i, ys[])`.
-        //
-        // The best intercept therefore has the form
-        // `y[i] - line.eval(i)` (using wrapping arithmetics).
-        // In other words, the best intercept is one of the `y - Line::eval(ys[i])`
-        // and our task is just to pick the one that minimizes our error.
-        //
-        // Without sorting our values, this is a difficult problem.
-        // We however rely on the following trick...
-        //
-        // We only focus on the case where the interpolation is half decent.
-        // If the line interpolation is doing its job on a dataset suited for it,
-        // we can hope that the maximum error won't be larger than `u64::MAX / 2`.
-        //
-        // In other words, even without the intercept the values `y - Line::eval(ys[i])` will all be
-        // within an interval that takes less than half of the modulo space of `u64`.
-        //
-        // Our task is therefore to identify this interval.
-        // Here we simply translate all of our values by `y0 - 2^63` and pick the min.
-        let mut line = Line {
-            slope,
-            intercept: 0,
-        };
-        let heuristic_shift = y0.wrapping_sub(MID_POINT);
-        line.intercept = positions_and_values
-            .map(|(pos, y)| y.wrapping_sub(line.eval(pos as u32)))
-            .min_by_key(|&val| val.wrapping_sub(heuristic_shift))
-            .unwrap_or(0u64); //< Never happens.
-        line
-    }
-
-    /// Returns a line that attemps to approximate a function
-    /// f: i in 0..[ys.num_vals()) -> ys[i].
-    ///
-    /// - The approximation is always lower than the actual value.
-    /// Or more rigorously, formally `f(i).wrapping_sub(ys[i])` is small
-    /// for any i in [0..ys.len()).
-    /// - It computes without panicking for any value of it.
-    ///
-    /// This function is only invariable by translation if all of the
-    /// `ys` are packaged into half of the space. (See heuristic below)
-    /// TODO USE array
-    pub fn train(ys: &dyn ColumnValues) -> Self {
-        let first_val = ys.iter().next().unwrap();
-        let last_val = ys.iter().nth(ys.num_vals() as usize - 1).unwrap();
-        Self::train_from(
-            first_val,
-            last_val,
-            ys.num_vals(),
-            ys.iter().enumerate().map(|(pos, val)| (pos as u64, val)),
-        )
-    }
-}
-
-impl BinarySerializable for Line {
-    fn serialize<W: io::Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        VInt(self.slope).serialize(writer)?;
-        VInt(self.intercept).serialize(writer)?;
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let slope = VInt::deserialize(reader)?.0;
-        let intercept = VInt::deserialize(reader)?.0;
-        Ok(Line { slope, intercept })
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::column_values::VecColumn;
-
-    /// Test training a line and ensuring that the maximum difference between
-    /// the data points and the line is `expected`.
-    ///
-    /// This function operates translation over the data for better coverage.
-    #[track_caller]
-    fn test_line_interpol_with_translation(ys: &[u64], expected: Option<u64>) {
-        let mut translations = vec![0, 100, u64::MAX / 2, u64::MAX, u64::MAX - 1];
-        translations.extend_from_slice(ys);
-        for translation in translations {
-            let translated_ys: Vec<u64> = ys
-                .iter()
-                .copied()
-                .map(|y| y.wrapping_add(translation))
-                .collect();
-            let largest_err = test_eval_max_err(&translated_ys);
-            assert_eq!(largest_err, expected);
-        }
-    }
-
-    fn test_eval_max_err(ys: &[u64]) -> Option<u64> {
-        let line = Line::train(&VecColumn::from(&ys));
-        ys.iter()
-            .enumerate()
-            .map(|(x, y)| y.wrapping_sub(line.eval(x as u32)))
-            .max()
-    }
-
-    #[test]
-    fn test_train() {
-        test_line_interpol_with_translation(&[11, 11, 11, 12, 12, 13], Some(1));
-        test_line_interpol_with_translation(&[13, 12, 12, 11, 11, 11], Some(1));
-        test_line_interpol_with_translation(&[13, 13, 12, 11, 11, 11], Some(1));
-        test_line_interpol_with_translation(&[13, 13, 12, 11, 11, 11], Some(1));
-        test_line_interpol_with_translation(&[u64::MAX - 1, 0, 0, 1], Some(1));
-        test_line_interpol_with_translation(&[u64::MAX - 1, u64::MAX, 0, 1], Some(0));
-        test_line_interpol_with_translation(&[0, 1, 2, 3, 5], Some(0));
-        test_line_interpol_with_translation(&[1, 2, 3, 4], Some(0));
-
-        let data: Vec<u64> = (0..255).collect();
-        test_line_interpol_with_translation(&data, Some(0));
-        let data: Vec<u64> = (0..255).map(|el| el * 2).collect();
-        test_line_interpol_with_translation(&data, Some(0));
-    }
-}

columnar/src/column_values/u64_based/linear.rs

@@ -1,277 +0,0 @@
-use std::io;
-
-use common::{BinarySerializable, OwnedBytes};
-use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
-
-use super::line::Line;
-use super::ColumnValues;
-use crate::column_values::u64_based::{ColumnCodec, ColumnCodecEstimator, ColumnStats};
-use crate::column_values::VecColumn;
-use crate::RowId;
-
-const HALF_SPACE: u64 = u64::MAX / 2;
-const LINE_ESTIMATION_BLOCK_LEN: usize = 512;
-
-/// Depending on the field type, a different
-/// fast field is required.
-#[derive(Clone)]
-pub struct LinearReader {
-    data: OwnedBytes,
-    linear_params: LinearParams,
-    stats: ColumnStats,
-}
-
-impl ColumnValues for LinearReader {
-    #[inline]
-    fn get_val(&self, doc: u32) -> u64 {
-        let interpoled_val: u64 = self.linear_params.line.eval(doc);
-        let bitpacked_diff = self.linear_params.bit_unpacker.get(doc, &self.data);
-        interpoled_val.wrapping_add(bitpacked_diff)
-    }
-
-    #[inline(always)]
-    fn min_value(&self) -> u64 {
-        self.stats.min_value
-    }
-
-    #[inline(always)]
-    fn max_value(&self) -> u64 {
-        self.stats.max_value
-    }
-
-    #[inline]
-    fn num_vals(&self) -> u32 {
-        self.stats.num_rows
-    }
-}
-
-/// Fastfield serializer, which tries to guess values by linear interpolation
-/// and stores the difference bitpacked.
-pub struct LinearCodec;
-
-#[derive(Debug, Clone)]
-struct LinearParams {
-    line: Line,
-    bit_unpacker: BitUnpacker,
-}
-
-impl BinarySerializable for LinearParams {
-    fn serialize<W: io::Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        self.line.serialize(writer)?;
-        self.bit_unpacker.bit_width().serialize(writer)?;
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let line = Line::deserialize(reader)?;
-        let bit_width = u8::deserialize(reader)?;
-        Ok(Self {
-            line,
-            bit_unpacker: BitUnpacker::new(bit_width),
-        })
-    }
-}
-
-pub struct LinearCodecEstimator {
-    block: Vec<u64>,
-    line: Option<Line>,
-    row_id: RowId,
-    min_deviation: u64,
-    max_deviation: u64,
-    first_val: u64,
-    last_val: u64,
-}
-
-impl Default for LinearCodecEstimator {
-    fn default() -> LinearCodecEstimator {
-        LinearCodecEstimator {
-            block: Vec::with_capacity(LINE_ESTIMATION_BLOCK_LEN),
-            line: None,
-            row_id: 0,
-            min_deviation: u64::MAX,
-            max_deviation: u64::MIN,
-            first_val: 0u64,
-            last_val: 0u64,
-        }
-    }
-}
-
-impl ColumnCodecEstimator for LinearCodecEstimator {
-    fn finalize(&mut self) {
-        if let Some(line) = self.line.as_mut() {
-            line.intercept = line
-                .intercept
-                .wrapping_add(self.min_deviation)
-                .wrapping_sub(HALF_SPACE);
-        }
-    }
-
-    fn estimate(&self, stats: &ColumnStats) -> Option<u64> {
-        let line = self.line?;
-        let amplitude = self.max_deviation - self.min_deviation;
-        let num_bits = compute_num_bits(amplitude);
-        let linear_params = LinearParams {
-            line,
-            bit_unpacker: BitUnpacker::new(num_bits),
-        };
-        Some(
-            stats.num_bytes()
-                + linear_params.num_bytes()
-                + (num_bits as u64 * stats.num_rows as u64 + 7) / 8,
-        )
-    }
-
-    fn serialize(
-        &self,
-        stats: &ColumnStats,
-        vals: &mut dyn Iterator<Item = u64>,
-        wrt: &mut dyn io::Write,
-    ) -> io::Result<()> {
-        stats.serialize(wrt)?;
-        let line = self.line.unwrap();
-        let amplitude = self.max_deviation - self.min_deviation;
-        let num_bits = compute_num_bits(amplitude);
-        let linear_params = LinearParams {
-            line,
-            bit_unpacker: BitUnpacker::new(num_bits),
-        };
-        linear_params.serialize(wrt)?;
-        let mut bit_packer = BitPacker::new();
-        for (pos, value) in vals.enumerate() {
-            let calculated_value = line.eval(pos as u32);
-            let offset = value.wrapping_sub(calculated_value);
-            bit_packer.write(offset, num_bits, wrt)?;
-        }
-        bit_packer.close(wrt)?;
-        Ok(())
-    }
-
-    fn collect(&mut self, value: u64) {
-        if let Some(line) = self.line {
-            self.collect_after_line_estimation(&line, value);
-        } else {
-            self.collect_before_line_estimation(value);
-        }
-    }
-}
-
-impl LinearCodecEstimator {
-    #[inline]
-    fn collect_after_line_estimation(&mut self, line: &Line, value: u64) {
-        let interpoled_val: u64 = line.eval(self.row_id);
-        let deviation = value.wrapping_add(HALF_SPACE).wrapping_sub(interpoled_val);
-        self.min_deviation = self.min_deviation.min(deviation);
-        self.max_deviation = self.max_deviation.max(deviation);
-        if self.row_id == 0 {
-            self.first_val = value;
-        }
-        self.last_val = value;
-        self.row_id += 1u32;
-    }
-
-    #[inline]
-    fn collect_before_line_estimation(&mut self, value: u64) {
-        self.block.push(value);
-        if self.block.len() == LINE_ESTIMATION_BLOCK_LEN {
-            let line = Line::train(&VecColumn::from(&self.block));
-            let block = std::mem::take(&mut self.block);
-            for val in block {
-                self.collect_after_line_estimation(&line, val);
-            }
-            self.line = Some(line);
-        }
-    }
-}
-
-impl ColumnCodec for LinearCodec {
-    type ColumnValues = LinearReader;
-
-    type Estimator = LinearCodecEstimator;
-
-    fn load(mut data: OwnedBytes) -> io::Result<Self::ColumnValues> {
-        let stats = ColumnStats::deserialize(&mut data)?;
-        let linear_params = LinearParams::deserialize(&mut data)?;
-        Ok(LinearReader {
-            stats,
-            linear_params,
-            data,
-        })
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use rand::RngCore;
-
-    use super::*;
-    use crate::column_values::u64_based::tests::{create_and_validate, get_codec_test_datasets};
-
-    #[test]
-    fn test_compression_simple() {
-        let vals = (100u64..)
-            .take(super::LINE_ESTIMATION_BLOCK_LEN)
-            .collect::<Vec<_>>();
-        create_and_validate::<LinearCodec>(&vals, "simple monotonically large").unwrap();
-    }
-
-    #[test]
-    fn test_compression() {
-        let data = (10..=6_000_u64).collect::<Vec<_>>();
-        let (estimate, actual_compression) =
-            create_and_validate::<LinearCodec>(&data, "simple monotonically large").unwrap();
-        assert_le!(actual_compression, 0.001);
-        assert_le!(estimate, 0.02);
-    }
-
-    #[test]
-    fn test_with_codec_datasets() {
-        let data_sets = get_codec_test_datasets();
-        for (mut data, name) in data_sets {
-            create_and_validate::<LinearCodec>(&data, name);
-            data.reverse();
-            create_and_validate::<LinearCodec>(&data, name);
-        }
-    }
-    #[test]
-    fn linear_interpol_fast_field_test_large_amplitude() {
-        let data = vec![
-            i64::MAX as u64 / 2,
-            i64::MAX as u64 / 3,
-            i64::MAX as u64 / 2,
-        ];
-        create_and_validate::<LinearCodec>(&data, "large amplitude");
-    }
-
-    #[test]
-    fn overflow_error_test() {
-        let data = vec![1572656989877777, 1170935903116329, 720575940379279, 0];
-        create_and_validate::<LinearCodec>(&data, "overflow test");
-    }
-
-    #[test]
-    fn linear_interpol_fast_concave_data() {
-        let data = vec![0, 1, 2, 5, 8, 10, 20, 50];
-        create_and_validate::<LinearCodec>(&data, "concave data");
-    }
-    #[test]
-    fn linear_interpol_fast_convex_data() {
-        let data = vec![0, 40, 60, 70, 75, 77];
-        create_and_validate::<LinearCodec>(&data, "convex data");
-    }
-    #[test]
-    fn linear_interpol_fast_field_test_simple() {
-        let data = (10..=20_u64).collect::<Vec<_>>();
-        create_and_validate::<LinearCodec>(&data, "simple monotonically");
-    }
-
-    #[test]
-    fn linear_interpol_fast_field_rand() {
-        let mut rng = rand::thread_rng();
-        for _ in 0..50 {
-            let mut data = (0..10_000).map(|_| rng.next_u64()).collect::<Vec<_>>();
-            create_and_validate::<LinearCodec>(&data, "random");
-            data.reverse();
-            create_and_validate::<LinearCodec>(&data, "random");
-        }
-    }
-}

columnar/src/column_values/u64_based/mod.rs

@@ -1,214 +0,0 @@
-mod bitpacked;
-mod blockwise_linear;
-mod line;
-mod linear;
-mod stats_collector;
-
-use std::io;
-use std::io::Write;
-use std::sync::Arc;
-
-use common::{BinarySerializable, OwnedBytes};
-
-use crate::column_values::monotonic_mapping::{
-    StrictlyMonotonicMappingInverter, StrictlyMonotonicMappingToInternal,
-};
-pub use crate::column_values::u64_based::bitpacked::BitpackedCodec;
-pub use crate::column_values::u64_based::blockwise_linear::BlockwiseLinearCodec;
-pub use crate::column_values::u64_based::linear::LinearCodec;
-pub use crate::column_values::u64_based::stats_collector::StatsCollector;
-use crate::column_values::{monotonic_map_column, ColumnStats};
-use crate::iterable::Iterable;
-use crate::{ColumnValues, MonotonicallyMappableToU64};
-
-/// A `ColumnCodecEstimator` is in charge of gathering all
-/// data required to serialize a column.
-///
-/// This happens during a first pass on data of the column elements.
-/// During that pass, all column estimators receive a call to their
-/// `.collect(el)`.
-///
-/// After this first pass, finalize is called.
-/// `.estimate(..)` then should return an accurate estimation of the
-/// size of the serialized column (were we to pick this codec.).
-/// `.serialize(..)` then serializes the column using this codec.
-pub trait ColumnCodecEstimator<T = u64>: 'static {
-    /// Records a new value for estimation.
-    /// This method will be called for each element of the column during
-    /// `estimation`.
-    fn collect(&mut self, value: u64);
-    /// Finalizes the first pass phase.
-    fn finalize(&mut self) {}
-    /// Returns an accurate estimation of the number of bytes that will
-    /// be used to represent this column.
-    fn estimate(&self, stats: &ColumnStats) -> Option<u64>;
-    /// Serializes the column using the given codec.
-    /// This constitutes a second pass over the columns values.
-    fn serialize(
-        &self,
-        stats: &ColumnStats,
-        vals: &mut dyn Iterator<Item = T>,
-        wrt: &mut dyn io::Write,
-    ) -> io::Result<()>;
-}
-
-/// A column codec describes a colunm serialization format.
-pub trait ColumnCodec<T: PartialOrd = u64> {
-    /// Specialized `ColumnValues` type.
-    type ColumnValues: ColumnValues<T> + 'static;
-    /// `Estimator` for the given codec.
-    type Estimator: ColumnCodecEstimator + Default;
-
-    /// Loads a column that has been serialized using this codec.
-    fn load(bytes: OwnedBytes) -> io::Result<Self::ColumnValues>;
-
-    /// Returns an estimator.
-    fn estimator() -> Self::Estimator {
-        Self::Estimator::default()
-    }
-
-    /// Returns a boxed estimator.
-    fn boxed_estimator() -> Box<dyn ColumnCodecEstimator> {
-        Box::new(Self::estimator())
-    }
-}
-
-/// Available codecs to use to encode the u64 (via [`MonotonicallyMappableToU64`]) converted data.
-#[derive(PartialEq, Eq, PartialOrd, Ord, Debug, Clone, Copy)]
-#[repr(u8)]
-pub enum CodecType {
-    /// Bitpack all values in the value range. The number of bits is defined by the amplitude
-    /// `column.max_value() - column.min_value()`
-    Bitpacked = 0u8,
-    /// Linear interpolation puts a line between the first and last value and then bitpacks the
-    /// values by the offset from the line. The number of bits is defined by the max deviation from
-    /// the line.
-    Linear = 1u8,
-    /// Same as [`CodecType::Linear`], but encodes in blocks of 512 elements.
-    BlockwiseLinear = 2u8,
-}
-
-/// List of all available u64-base codecs.
-pub const ALL_U64_CODEC_TYPES: [CodecType; 3] = [
-    CodecType::Bitpacked,
-    CodecType::Linear,
-    CodecType::BlockwiseLinear,
-];
-
-impl CodecType {
-    fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    fn try_from_code(code: u8) -> Option<CodecType> {
-        match code {
-            0u8 => Some(CodecType::Bitpacked),
-            1u8 => Some(CodecType::Linear),
-            2u8 => Some(CodecType::BlockwiseLinear),
-            _ => None,
-        }
-    }
-
-    fn load<T: MonotonicallyMappableToU64>(
-        &self,
-        bytes: OwnedBytes,
-    ) -> io::Result<Arc<dyn ColumnValues<T>>> {
-        match self {
-            CodecType::Bitpacked => load_specific_codec::<BitpackedCodec, T>(bytes),
-            CodecType::Linear => load_specific_codec::<LinearCodec, T>(bytes),
-            CodecType::BlockwiseLinear => load_specific_codec::<BlockwiseLinearCodec, T>(bytes),
-        }
-    }
-}
-
-fn load_specific_codec<C: ColumnCodec, T: MonotonicallyMappableToU64>(
-    bytes: OwnedBytes,
-) -> io::Result<Arc<dyn ColumnValues<T>>> {
-    let reader = C::load(bytes)?;
-    let reader_typed = monotonic_map_column(
-        reader,
-        StrictlyMonotonicMappingInverter::from(StrictlyMonotonicMappingToInternal::<T>::new()),
-    );
-    Ok(Arc::new(reader_typed))
-}
-
-impl CodecType {
-    /// Returns a boxed codec estimator associated to a given `CodecType`.
-    pub fn estimator(&self) -> Box<dyn ColumnCodecEstimator> {
-        match self {
-            CodecType::Bitpacked => BitpackedCodec::boxed_estimator(),
-            CodecType::Linear => LinearCodec::boxed_estimator(),
-            CodecType::BlockwiseLinear => BlockwiseLinearCodec::boxed_estimator(),
-        }
-    }
-}
-
-/// Serializes a given column of u64-mapped values.
-pub fn serialize_u64_based_column_values<T: MonotonicallyMappableToU64>(
-    vals: &dyn Iterable<T>,
-    codec_types: &[CodecType],
-    wrt: &mut dyn Write,
-) -> io::Result<()> {
-    let mut stats_collector = StatsCollector::default();
-    let mut estimators: Vec<(CodecType, Box<dyn ColumnCodecEstimator>)> =
-        Vec::with_capacity(codec_types.len());
-    for &codec_type in codec_types {
-        estimators.push((codec_type, codec_type.estimator()));
-    }
-    for val in vals.boxed_iter() {
-        let val_u64 = val.to_u64();
-        stats_collector.collect(val_u64);
-        for (_, estimator) in &mut estimators {
-            estimator.collect(val_u64);
-        }
-    }
-    for (_, estimator) in &mut estimators {
-        estimator.finalize();
-    }
-    let stats = stats_collector.stats();
-    let (_, best_codec, best_codec_estimator) = estimators
-        .into_iter()
-        .flat_map(|(codec_type, estimator)| {
-            let num_bytes = estimator.estimate(&stats)?;
-            Some((num_bytes, codec_type, estimator))
-        })
-        .min_by_key(|(num_bytes, _, _)| *num_bytes)
-        .ok_or_else(|| {
-            io::Error::new(io::ErrorKind::InvalidData, "No available applicable codec.")
-        })?;
-    best_codec.to_code().serialize(wrt)?;
-    best_codec_estimator.serialize(
-        &stats,
-        &mut vals.boxed_iter().map(MonotonicallyMappableToU64::to_u64),
-        wrt,
-    )?;
-    Ok(())
-}
-
-/// Load u64-based column values.
-///
-/// This method first identifies the codec off the first byte.
-pub fn load_u64_based_column_values<T: MonotonicallyMappableToU64>(
-    mut bytes: OwnedBytes,
-) -> io::Result<Arc<dyn ColumnValues<T>>> {
-    let codec_type: CodecType = bytes
-        .first()
-        .copied()
-        .and_then(CodecType::try_from_code)
-        .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidData, "Failed to read codec type"))?;
-    bytes.advance(1);
-    codec_type.load(bytes)
-}
-
-/// Helper function to serialize a column (autodetect from all codecs) and then open it
-pub fn serialize_and_load_u64_based_column_values<T: MonotonicallyMappableToU64>(
-    vals: &dyn Iterable,
-    codec_types: &[CodecType],
-) -> Arc<dyn ColumnValues<T>> {
-    let mut buffer = Vec::new();
-    serialize_u64_based_column_values(vals, codec_types, &mut buffer).unwrap();
-    load_u64_based_column_values::<T>(OwnedBytes::new(buffer)).unwrap()
-}
-
-#[cfg(test)]
-mod tests;

columnar/src/column_values/u64_based/stats_collector.rs

@@ -1,200 +0,0 @@
-use std::num::NonZeroU64;
-
-use fastdivide::DividerU64;
-
-use crate::column_values::ColumnStats;
-use crate::RowId;
-
-/// 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;
-        }
-    }
-}
-
-#[derive(Default)]
-pub struct StatsCollector {
-    min_max_opt: Option<(u64, u64)>,
-    num_rows: RowId,
-    // We measure the GCD of the difference between the values and the minimal value.
-    // This is the same as computing the difference between the values and the first value.
-    //
-    // This way, we can compress i64-converted-to-u64 (e.g. timestamp that were supplied in
-    // seconds, only to be converted in microseconds).
-    increment_gcd_opt: Option<(NonZeroU64, DividerU64)>,
-    first_value_opt: Option<u64>,
-}
-
-impl StatsCollector {
-    pub fn stats(&self) -> ColumnStats {
-        let (min_value, max_value) = self.min_max_opt.unwrap_or((0u64, 0u64));
-        let increment_gcd = if let Some((increment_gcd, _)) = self.increment_gcd_opt {
-            increment_gcd
-        } else {
-            NonZeroU64::new(1u64).unwrap()
-        };
-        ColumnStats {
-            min_value,
-            max_value,
-            num_rows: self.num_rows,
-            gcd: increment_gcd,
-        }
-    }
-
-    #[inline]
-    fn update_increment_gcd(&mut self, value: u64) {
-        let Some(first_value) = self.first_value_opt else {
-            // We set the first value and just quit.
-            self.first_value_opt = Some(value);
-            return;
-        };
-        let Some(non_zero_value) = NonZeroU64::new(value.abs_diff(first_value)) else {
-            // We can simply skip 0 values.
-            return;
-        };
-        let Some((gcd, gcd_divider)) = self.increment_gcd_opt else {
-            self.set_increment_gcd(non_zero_value);
-            return;
-        };
-        if gcd.get() == 1 {
-            // It won't see any update now.
-            return;
-        }
-        let remainder =
-            non_zero_value.get() - (gcd_divider.divide(non_zero_value.get())) * gcd.get();
-        if remainder == 0 {
-            return;
-        }
-        let new_gcd = compute_gcd(non_zero_value, gcd);
-        self.set_increment_gcd(new_gcd);
-    }
-
-    fn set_increment_gcd(&mut self, gcd: NonZeroU64) {
-        let new_divider = DividerU64::divide_by(gcd.get());
-        self.increment_gcd_opt = Some((gcd, new_divider));
-    }
-
-    pub fn collect(&mut self, value: u64) {
-        self.min_max_opt = Some(if let Some((min, max)) = self.min_max_opt {
-            (min.min(value), max.max(value))
-        } else {
-            (value, value)
-        });
-        self.num_rows += 1;
-        self.update_increment_gcd(value);
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use std::num::NonZeroU64;
-
-    use crate::column_values::u64_based::stats_collector::{compute_gcd, StatsCollector};
-    use crate::column_values::u64_based::ColumnStats;
-
-    fn compute_stats(vals: impl Iterator<Item = u64>) -> ColumnStats {
-        let mut stats_collector = StatsCollector::default();
-        for val in vals {
-            stats_collector.collect(val);
-        }
-        stats_collector.stats()
-    }
-
-    fn find_gcd(vals: impl Iterator<Item = u64>) -> u64 {
-        compute_stats(vals).gcd.get()
-    }
-
-    #[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 test_gcd() {
-        assert_eq!(find_gcd([0].into_iter()), 1);
-        assert_eq!(find_gcd([0, 10].into_iter()), 10);
-        assert_eq!(find_gcd([10, 0].into_iter()), 10);
-        assert_eq!(find_gcd([].into_iter()), 1);
-        assert_eq!(find_gcd([15, 30, 5, 10].into_iter()), 5);
-        assert_eq!(find_gcd([15, 16, 10].into_iter()), 1);
-        assert_eq!(find_gcd([0, 5, 5, 5].into_iter()), 5);
-        assert_eq!(find_gcd([0, 0].into_iter()), 1);
-        assert_eq!(find_gcd([1, 10, 4, 1, 7, 10].into_iter()), 3);
-        assert_eq!(find_gcd([1, 10, 0, 4, 1, 7, 10].into_iter()), 1);
-    }
-
-    #[test]
-    fn test_stats() {
-        assert_eq!(
-            compute_stats([].into_iter()),
-            ColumnStats {
-                gcd: NonZeroU64::new(1).unwrap(),
-                min_value: 0,
-                max_value: 0,
-                num_rows: 0
-            }
-        );
-        assert_eq!(
-            compute_stats([0, 1].into_iter()),
-            ColumnStats {
-                gcd: NonZeroU64::new(1).unwrap(),
-                min_value: 0,
-                max_value: 1,
-                num_rows: 2
-            }
-        );
-        assert_eq!(
-            compute_stats([0, 1].into_iter()),
-            ColumnStats {
-                gcd: NonZeroU64::new(1).unwrap(),
-                min_value: 0,
-                max_value: 1,
-                num_rows: 2
-            }
-        );
-        assert_eq!(
-            compute_stats([10, 20, 30].into_iter()),
-            ColumnStats {
-                gcd: NonZeroU64::new(10).unwrap(),
-                min_value: 10,
-                max_value: 30,
-                num_rows: 3
-            }
-        );
-        assert_eq!(
-            compute_stats([10, 50, 10, 30].into_iter()),
-            ColumnStats {
-                gcd: NonZeroU64::new(20).unwrap(),
-                min_value: 10,
-                max_value: 50,
-                num_rows: 4
-            }
-        );
-        assert_eq!(
-            compute_stats([10, 0, 30].into_iter()),
-            ColumnStats {
-                gcd: NonZeroU64::new(10).unwrap(),
-                min_value: 0,
-                max_value: 30,
-                num_rows: 3
-            }
-        );
-    }
-}

columnar/src/column_values/u64_based/tests.rs

@@ -1,401 +0,0 @@
-use proptest::prelude::*;
-use proptest::strategy::Strategy;
-use proptest::{num, prop_oneof, proptest};
-
-#[test]
-fn test_serialize_and_load_simple() {
-    let mut buffer = Vec::new();
-    let vals = &[1u64, 2u64, 5u64];
-    serialize_u64_based_column_values(
-        &&vals[..],
-        &[CodecType::Bitpacked, CodecType::BlockwiseLinear],
-        &mut buffer,
-    )
-    .unwrap();
-    assert_eq!(buffer.len(), 7);
-    let col = load_u64_based_column_values::<u64>(OwnedBytes::new(buffer)).unwrap();
-    assert_eq!(col.num_vals(), 3);
-    assert_eq!(col.get_val(0), 1);
-    assert_eq!(col.get_val(1), 2);
-    assert_eq!(col.get_val(2), 5);
-}
-
-#[test]
-fn test_empty_column_i64() {
-    let vals: [i64; 0] = [];
-    let mut num_acceptable_codecs = 0;
-    for codec in ALL_U64_CODEC_TYPES {
-        let mut buffer = Vec::new();
-        if serialize_u64_based_column_values(&&vals[..], &[codec], &mut buffer).is_err() {
-            continue;
-        }
-        num_acceptable_codecs += 1;
-        let col = load_u64_based_column_values::<i64>(OwnedBytes::new(buffer)).unwrap();
-        assert_eq!(col.num_vals(), 0);
-        assert_eq!(col.min_value(), i64::MIN);
-        assert_eq!(col.max_value(), i64::MIN);
-    }
-    assert!(num_acceptable_codecs > 0);
-}
-
-#[test]
-fn test_empty_column_u64() {
-    let vals: [u64; 0] = [];
-    let mut num_acceptable_codecs = 0;
-    for codec in ALL_U64_CODEC_TYPES {
-        let mut buffer = Vec::new();
-        if serialize_u64_based_column_values(&&vals[..], &[codec], &mut buffer).is_err() {
-            continue;
-        }
-        num_acceptable_codecs += 1;
-        let col = load_u64_based_column_values::<u64>(OwnedBytes::new(buffer)).unwrap();
-        assert_eq!(col.num_vals(), 0);
-        assert_eq!(col.min_value(), u64::MIN);
-        assert_eq!(col.max_value(), u64::MIN);
-    }
-    assert!(num_acceptable_codecs > 0);
-}
-
-#[test]
-fn test_empty_column_f64() {
-    let vals: [f64; 0] = [];
-    let mut num_acceptable_codecs = 0;
-    for codec in ALL_U64_CODEC_TYPES {
-        let mut buffer = Vec::new();
-        if serialize_u64_based_column_values(&&vals[..], &[codec], &mut buffer).is_err() {
-            continue;
-        }
-        num_acceptable_codecs += 1;
-        let col = load_u64_based_column_values::<f64>(OwnedBytes::new(buffer)).unwrap();
-        assert_eq!(col.num_vals(), 0);
-        // FIXME. f64::MIN would be better!
-        assert!(col.min_value().is_nan());
-        assert!(col.max_value().is_nan());
-    }
-    assert!(num_acceptable_codecs > 0);
-}
-
-pub(crate) fn create_and_validate<TColumnCodec: ColumnCodec>(
-    vals: &[u64],
-    name: &str,
-) -> Option<(f32, f32)> {
-    let mut stats_collector = StatsCollector::default();
-    let mut codec_estimator: TColumnCodec::Estimator = Default::default();
-
-    for val in vals.boxed_iter() {
-        stats_collector.collect(val);
-        codec_estimator.collect(val);
-    }
-    codec_estimator.finalize();
-    let stats = stats_collector.stats();
-    let estimation = codec_estimator.estimate(&stats)?;
-
-    let mut buffer = Vec::new();
-    codec_estimator
-        .serialize(&stats, vals.boxed_iter().as_mut(), &mut buffer)
-        .unwrap();
-
-    let actual_compression = buffer.len() as u64;
-
-    let reader = TColumnCodec::load(OwnedBytes::new(buffer)).unwrap();
-    assert_eq!(reader.num_vals(), vals.len() as u32);
-    for (doc, orig_val) in vals.iter().copied().enumerate() {
-        let val = reader.get_val(doc as u32);
-        assert_eq!(
-            val, orig_val,
-            "val `{val}` does not match orig_val {orig_val:?}, in data set {name}, data `{vals:?}`",
-        );
-    }
-
-    if !vals.is_empty() {
-        let test_rand_idx = rand::thread_rng().gen_range(0..=vals.len() - 1);
-        let expected_positions: Vec<u32> = vals
-            .iter()
-            .enumerate()
-            .filter(|(_, el)| **el == vals[test_rand_idx])
-            .map(|(pos, _)| pos as u32)
-            .collect();
-        let mut positions = Vec::new();
-        reader.get_row_ids_for_value_range(
-            vals[test_rand_idx]..=vals[test_rand_idx],
-            0..vals.len() as u32,
-            &mut positions,
-        );
-        assert_eq!(expected_positions, positions);
-    }
-    if actual_compression > 1000 {
-        assert!(relative_difference(estimation, actual_compression) < 0.10f32);
-    }
-    Some((
-        compression_rate(estimation, stats.num_rows),
-        compression_rate(actual_compression, stats.num_rows),
-    ))
-}
-
-fn compression_rate(num_bytes: u64, num_values: u32) -> f32 {
-    num_bytes as f32 / (num_values as f32 * 8.0)
-}
-
-fn relative_difference(left: u64, right: u64) -> f32 {
-    let left = left as f32;
-    let right = right as f32;
-    2.0f32 * (left - right).abs() / (left + right)
-}
-
-proptest! {
-    #![proptest_config(ProptestConfig::with_cases(100))]
-
-    #[test]
-    fn test_proptest_small_bitpacked(data in proptest::collection::vec(num_strategy(), 1..10)) {
-        create_and_validate::<BitpackedCodec>(&data, "proptest bitpacked");
-    }
-
-    #[test]
-    fn test_proptest_small_linear(data in proptest::collection::vec(num_strategy(), 1..10)) {
-        create_and_validate::<LinearCodec>(&data, "proptest linearinterpol");
-    }
-
-
-    #[test]
-    fn test_proptest_small_blockwise_linear(data in proptest::collection::vec(num_strategy(), 1..10)) {
-        create_and_validate::<BlockwiseLinearCodec>(&data, "proptest multilinearinterpol");
-    }
-}
-
-#[test]
-fn test_small_blockwise_linear_example() {
-    create_and_validate::<BlockwiseLinearCodec>(
-        &[9223372036854775808, 9223370937344622593],
-        "proptest multilinearinterpol",
-    );
-}
-
-proptest! {
-    #![proptest_config(ProptestConfig::with_cases(10))]
-
-    #[test]
-    fn test_proptest_large_bitpacked(data in proptest::collection::vec(num_strategy(), 1..6000)) {
-        create_and_validate::<BitpackedCodec>(&data, "proptest bitpacked");
-    }
-
-    #[test]
-    fn test_proptest_large_linear(data in proptest::collection::vec(num_strategy(), 1..6000)) {
-        create_and_validate::<LinearCodec>(&data, "proptest linearinterpol");
-    }
-
-    #[test]
-    fn test_proptest_large_blockwise_linear(data in proptest::collection::vec(num_strategy(), 1..6000)) {
-        create_and_validate::<BlockwiseLinearCodec>(&data, "proptest multilinearinterpol");
-    }
-}
-
-fn num_strategy() -> impl Strategy<Value = u64> {
-    prop_oneof![
-        1 => prop::num::u64::ANY.prop_map(|num| u64::MAX - (num % 10) ),
-        1 => prop::num::u64::ANY.prop_map(|num| num % 10 ),
-        20 => prop::num::u64::ANY,
-    ]
-}
-
-pub fn get_codec_test_datasets() -> Vec<(Vec<u64>, &'static str)> {
-    let mut data_and_names = vec![];
-
-    let data = (10..=10_000_u64).collect::<Vec<_>>();
-    data_and_names.push((data, "simple monotonically increasing"));
-
-    data_and_names.push((
-        vec![5, 6, 7, 8, 9, 10, 99, 100],
-        "offset in linear interpol",
-    ));
-    data_and_names.push((vec![5, 50, 3, 13, 1, 1000, 35], "rand small"));
-    data_and_names.push((vec![10], "single value"));
-
-    data_and_names.push((
-        vec![1572656989877777, 1170935903116329, 720575940379279, 0],
-        "overflow error",
-    ));
-
-    data_and_names
-}
-
-fn test_codec<C: ColumnCodec>() {
-    let codec_name = std::any::type_name::<C>();
-    for (data, dataset_name) in get_codec_test_datasets() {
-        let estimate_actual_opt: Option<(f32, f32)> =
-            tests::create_and_validate::<C>(&data, dataset_name);
-        let result = if let Some((estimate, actual)) = estimate_actual_opt {
-            format!("Estimate `{estimate}` Actual `{actual}`")
-        } else {
-            "Disabled".to_string()
-        };
-        println!("Codec {codec_name}, DataSet {dataset_name}, {result}");
-    }
-}
-#[test]
-fn test_codec_bitpacking() {
-    test_codec::<BitpackedCodec>();
-}
-#[test]
-fn test_codec_interpolation() {
-    test_codec::<LinearCodec>();
-}
-#[test]
-fn test_codec_multi_interpolation() {
-    test_codec::<BlockwiseLinearCodec>();
-}
-
-use super::*;
-
-fn estimate<C: ColumnCodec>(vals: &[u64]) -> Option<f32> {
-    let mut stats_collector = StatsCollector::default();
-    let mut estimator = C::Estimator::default();
-    for &val in vals {
-        stats_collector.collect(val);
-        estimator.collect(val);
-    }
-    estimator.finalize();
-    let stats = stats_collector.stats();
-    let num_bytes = estimator.estimate(&stats)?;
-    if stats.num_rows == 0 {
-        return None;
-    }
-    Some(num_bytes as f32 / (8.0 * stats.num_rows as f32))
-}
-
-#[test]
-fn estimation_good_interpolation_case() {
-    let data = (10..=20000_u64).collect::<Vec<_>>();
-
-    let linear_interpol_estimation = estimate::<LinearCodec>(&data).unwrap();
-    assert_le!(linear_interpol_estimation, 0.01);
-
-    let multi_linear_interpol_estimation = estimate::<BlockwiseLinearCodec>(&data).unwrap();
-    assert_le!(multi_linear_interpol_estimation, 0.2);
-    assert_lt!(linear_interpol_estimation, multi_linear_interpol_estimation);
-
-    let bitpacked_estimation = estimate::<BitpackedCodec>(&data).unwrap();
-    assert_lt!(linear_interpol_estimation, bitpacked_estimation);
-}
-
-#[test]
-fn estimation_test_bad_interpolation_case_monotonically_increasing() {
-    let mut data: Vec<u64> = (201..=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 = estimate::<LinearCodec>(&data[..]).unwrap();
-    assert_le!(linear_interpol_estimation, 0.35);
-
-    let bitpacked_estimation = estimate::<BitpackedCodec>(&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) = CodecType::try_from_code(code) {
-            assert_eq!(codec_type.to_code(), code);
-            count_codec += 1;
-        }
-    }
-    assert_eq!(count_codec, 3);
-}
-
-fn test_fastfield_gcd_i64_with_codec(codec_type: CodecType, num_vals: usize) -> io::Result<()> {
-    let mut vals: Vec<i64> = (-4..=(num_vals as i64) - 5).map(|val| val * 1000).collect();
-    let mut buffer: Vec<u8> = Vec::new();
-    crate::column_values::serialize_u64_based_column_values(
-        &&vals[..],
-        &[codec_type],
-        &mut buffer,
-    )?;
-    let buffer = OwnedBytes::new(buffer);
-    let column = crate::column_values::load_u64_based_column_values::<i64>(buffer.clone())?;
-    assert_eq!(column.get_val(0), -4000i64);
-    assert_eq!(column.get_val(1), -3000i64);
-    assert_eq!(column.get_val(2), -2000i64);
-    assert_eq!(column.max_value(), (num_vals as i64 - 5) * 1000);
-    assert_eq!(column.min_value(), -4000i64);
-
-    // Can't apply gcd
-    let mut buffer_without_gcd = Vec::new();
-    vals.pop();
-    vals.push(1001i64);
-    crate::column_values::serialize_u64_based_column_values(
-        &&vals[..],
-        &[codec_type],
-        &mut buffer_without_gcd,
-    )?;
-    let buffer_without_gcd = OwnedBytes::new(buffer_without_gcd);
-    assert!(buffer_without_gcd.len() > buffer.len());
-
-    Ok(())
-}
-
-#[test]
-fn test_fastfield_gcd_i64() -> io::Result<()> {
-    for &codec_type in &[
-        CodecType::Bitpacked,
-        CodecType::BlockwiseLinear,
-        CodecType::Linear,
-    ] {
-        test_fastfield_gcd_i64_with_codec(codec_type, 5500)?;
-    }
-    Ok(())
-}
-
-fn test_fastfield_gcd_u64_with_codec(codec_type: CodecType, num_vals: usize) -> io::Result<()> {
-    let mut vals: Vec<u64> = (1..=num_vals).map(|i| i as u64 * 1000u64).collect();
-    let mut buffer: Vec<u8> = Vec::new();
-    crate::column_values::serialize_u64_based_column_values(
-        &&vals[..],
-        &[codec_type],
-        &mut buffer,
-    )?;
-    let buffer = OwnedBytes::new(buffer);
-    let column = crate::column_values::load_u64_based_column_values::<u64>(buffer.clone())?;
-    assert_eq!(column.get_val(0), 1000u64);
-    assert_eq!(column.get_val(1), 2000u64);
-    assert_eq!(column.get_val(2), 3000u64);
-    assert_eq!(column.max_value(), num_vals as u64 * 1000);
-    assert_eq!(column.min_value(), 1000u64);
-
-    // Can't apply gcd
-    let mut buffer_without_gcd = Vec::new();
-    vals.pop();
-    vals.push(1001u64);
-    crate::column_values::serialize_u64_based_column_values(
-        &&vals[..],
-        &[codec_type],
-        &mut buffer_without_gcd,
-    )?;
-    let buffer_without_gcd = OwnedBytes::new(buffer_without_gcd);
-    assert!(buffer_without_gcd.len() > buffer.len());
-    Ok(())
-}
-
-#[test]
-fn test_fastfield_gcd_u64() -> io::Result<()> {
-    for &codec_type in &[
-        CodecType::Bitpacked,
-        CodecType::BlockwiseLinear,
-        CodecType::Linear,
-    ] {
-        test_fastfield_gcd_u64_with_codec(codec_type, 5500)?;
-    }
-    Ok(())
-}
-
-#[test]
-pub fn test_fastfield2() {
-    let test_fastfield = crate::column_values::serialize_and_load_u64_based_column_values::<u64>(
-        &&[100u64, 200u64, 300u64][..],
-        &ALL_U64_CODEC_TYPES,
-    );
-    assert_eq!(test_fastfield.get_val(0), 100);
-    assert_eq!(test_fastfield.get_val(1), 200);
-    assert_eq!(test_fastfield.get_val(2), 300);
-}

columnar/src/column_values/vec_column.rs

@@ -1,52 +0,0 @@
-use std::fmt::Debug;
-
-use tantivy_bitpacker::minmax;
-
-use crate::ColumnValues;
-
-/// VecColumn provides `Column` over a slice.
-pub struct VecColumn<'a, T = u64> {
-    pub(crate) values: &'a [T],
-    pub(crate) min_value: T,
-    pub(crate) max_value: T,
-}
-
-impl<'a, T: Copy + PartialOrd + Send + Sync + Debug> ColumnValues<T> for VecColumn<'a, T> {
-    fn get_val(&self, position: u32) -> T {
-        self.values[position as usize]
-    }
-
-    fn iter(&self) -> Box<dyn Iterator<Item = T> + '_> {
-        Box::new(self.values.iter().copied())
-    }
-
-    fn min_value(&self) -> T {
-        self.min_value
-    }
-
-    fn max_value(&self) -> T {
-        self.max_value
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.values.len() as u32
-    }
-
-    fn get_range(&self, start: u64, output: &mut [T]) {
-        output.copy_from_slice(&self.values[start as usize..][..output.len()])
-    }
-}
-
-impl<'a, T: Copy + PartialOrd + Default, V> From<&'a V> for VecColumn<'a, T>
-where V: AsRef<[T]> + ?Sized
-{
-    fn from(values: &'a V) -> Self {
-        let values = values.as_ref();
-        let (min_value, max_value) = minmax(values.iter().copied()).unwrap_or_default();
-        Self {
-            values,
-            min_value,
-            max_value,
-        }
-    }
-}

columnar/src/columnar/column_type.rs

@@ -1,163 +0,0 @@
-use std::fmt::Debug;
-use std::net::Ipv6Addr;
-
-use serde::{Deserialize, Serialize};
-
-use crate::value::NumericalType;
-use crate::InvalidData;
-
-/// The column type represents the column type.
-/// Any changes need to be propagated to `COLUMN_TYPES`.
-#[derive(Hash, Eq, PartialEq, Debug, Clone, Copy, Ord, PartialOrd, Serialize, Deserialize)]
-#[repr(u8)]
-pub enum ColumnType {
-    I64 = 0u8,
-    U64 = 1u8,
-    F64 = 2u8,
-    Bytes = 3u8,
-    Str = 4u8,
-    Bool = 5u8,
-    IpAddr = 6u8,
-    DateTime = 7u8,
-}
-
-// The order needs to match _exactly_ the order in the enum
-const COLUMN_TYPES: [ColumnType; 8] = [
-    ColumnType::I64,
-    ColumnType::U64,
-    ColumnType::F64,
-    ColumnType::Bytes,
-    ColumnType::Str,
-    ColumnType::Bool,
-    ColumnType::IpAddr,
-    ColumnType::DateTime,
-];
-
-impl ColumnType {
-    pub fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    pub(crate) fn try_from_code(code: u8) -> Result<ColumnType, InvalidData> {
-        COLUMN_TYPES.get(code as usize).copied().ok_or(InvalidData)
-    }
-}
-
-impl From<NumericalType> for ColumnType {
-    fn from(numerical_type: NumericalType) -> Self {
-        match numerical_type {
-            NumericalType::I64 => ColumnType::I64,
-            NumericalType::U64 => ColumnType::U64,
-            NumericalType::F64 => ColumnType::F64,
-        }
-    }
-}
-
-impl ColumnType {
-    pub fn numerical_type(&self) -> Option<NumericalType> {
-        match self {
-            ColumnType::I64 => Some(NumericalType::I64),
-            ColumnType::U64 => Some(NumericalType::U64),
-            ColumnType::F64 => Some(NumericalType::F64),
-            ColumnType::Bytes
-            | ColumnType::Str
-            | ColumnType::Bool
-            | ColumnType::IpAddr
-            | ColumnType::DateTime => None,
-        }
-    }
-}
-
-// TODO remove if possible
-pub trait HasAssociatedColumnType: 'static + Debug + Send + Sync + Copy + PartialOrd {
-    fn column_type() -> ColumnType;
-    fn default_value() -> Self;
-}
-
-impl HasAssociatedColumnType for u64 {
-    fn column_type() -> ColumnType {
-        ColumnType::U64
-    }
-
-    fn default_value() -> Self {
-        0u64
-    }
-}
-
-impl HasAssociatedColumnType for i64 {
-    fn column_type() -> ColumnType {
-        ColumnType::I64
-    }
-
-    fn default_value() -> Self {
-        0i64
-    }
-}
-
-impl HasAssociatedColumnType for f64 {
-    fn column_type() -> ColumnType {
-        ColumnType::F64
-    }
-
-    fn default_value() -> Self {
-        Default::default()
-    }
-}
-
-impl HasAssociatedColumnType for bool {
-    fn column_type() -> ColumnType {
-        ColumnType::Bool
-    }
-    fn default_value() -> Self {
-        Default::default()
-    }
-}
-
-impl HasAssociatedColumnType for common::DateTime {
-    fn column_type() -> ColumnType {
-        ColumnType::DateTime
-    }
-    fn default_value() -> Self {
-        Default::default()
-    }
-}
-
-impl HasAssociatedColumnType for Ipv6Addr {
-    fn column_type() -> ColumnType {
-        ColumnType::IpAddr
-    }
-
-    fn default_value() -> Self {
-        Ipv6Addr::from([0u8; 16])
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::Cardinality;
-
-    #[test]
-    fn test_column_type_to_code() {
-        for (code, expected_column_type) in super::COLUMN_TYPES.iter().copied().enumerate() {
-            if let Ok(column_type) = ColumnType::try_from_code(code as u8) {
-                assert_eq!(column_type, expected_column_type);
-            }
-        }
-        for code in COLUMN_TYPES.len() as u8..=u8::MAX {
-            assert!(ColumnType::try_from_code(code).is_err());
-        }
-    }
-
-    #[test]
-    fn test_cardinality_to_code() {
-        let mut num_cardinality = 0;
-        for code in u8::MIN..=u8::MAX {
-            if let Ok(cardinality) = Cardinality::try_from_code(code) {
-                assert_eq!(cardinality.to_code(), code);
-                num_cardinality += 1;
-            }
-        }
-        assert_eq!(num_cardinality, 3);
-    }
-}

columnar/src/columnar/format_version.rs

@@ -1,73 +0,0 @@
-use crate::InvalidData;
-
-pub const VERSION_FOOTER_NUM_BYTES: usize = MAGIC_BYTES.len() + std::mem::size_of::<u32>();
-
-/// We end the file by these 4 bytes just to somewhat identify that
-/// this is indeed a columnar file.
-const MAGIC_BYTES: [u8; 4] = [2, 113, 119, 66];
-
-pub fn footer() -> [u8; VERSION_FOOTER_NUM_BYTES] {
-    let mut footer_bytes = [0u8; VERSION_FOOTER_NUM_BYTES];
-    footer_bytes[0..4].copy_from_slice(&Version::V1.to_bytes());
-    footer_bytes[4..8].copy_from_slice(&MAGIC_BYTES[..]);
-    footer_bytes
-}
-
-pub fn parse_footer(footer_bytes: [u8; VERSION_FOOTER_NUM_BYTES]) -> Result<Version, InvalidData> {
-    if footer_bytes[4..8] != MAGIC_BYTES {
-        return Err(InvalidData);
-    }
-    Version::try_from_bytes(footer_bytes[0..4].try_into().unwrap())
-}
-
-#[derive(Debug, Copy, Clone, Eq, PartialEq)]
-#[repr(u32)]
-pub enum Version {
-    V1 = 1u32,
-}
-
-impl Version {
-    fn to_bytes(self) -> [u8; 4] {
-        (self as u32).to_le_bytes()
-    }
-
-    fn try_from_bytes(bytes: [u8; 4]) -> Result<Version, InvalidData> {
-        let code = u32::from_le_bytes(bytes);
-        match code {
-            1u32 => Ok(Version::V1),
-            _ => Err(InvalidData),
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use std::collections::HashSet;
-
-    use super::*;
-
-    #[test]
-    fn test_footer_dserialization() {
-        let parsed_version: Version = parse_footer(footer()).unwrap();
-        assert_eq!(Version::V1, parsed_version);
-    }
-
-    #[test]
-    fn test_version_serialization() {
-        let version_to_tests: Vec<u32> = [0, 1 << 8, 1 << 16, 1 << 24]
-            .iter()
-            .copied()
-            .flat_map(|offset| (0..255).map(move |el| el + offset))
-            .collect();
-        let mut valid_versions: HashSet<u32> = HashSet::default();
-        for &i in &version_to_tests {
-            let version_res = Version::try_from_bytes(i.to_le_bytes());
-            if let Ok(version) = version_res {
-                assert_eq!(version, Version::V1);
-                assert_eq!(version.to_bytes(), i.to_le_bytes());
-                valid_versions.insert(i);
-            }
-        }
-        assert_eq!(valid_versions.len(), 1);
-    }
-}

columnar/src/columnar/merge/merge_dict_column.rs

@@ -1,204 +0,0 @@
-use std::io::{self, Write};
-
-use common::{BitSet, CountingWriter, ReadOnlyBitSet};
-use sstable::{SSTable, TermOrdinal};
-
-use super::term_merger::TermMerger;
-use crate::column::serialize_column_mappable_to_u64;
-use crate::column_index::SerializableColumnIndex;
-use crate::iterable::Iterable;
-use crate::{BytesColumn, MergeRowOrder, ShuffleMergeOrder};
-
-// Serialize [Dictionary, Column, dictionary num bytes U32::LE]
-// Column: [Column Index, Column Values, column index num bytes U32::LE]
-pub fn merge_bytes_or_str_column(
-    column_index: SerializableColumnIndex<'_>,
-    bytes_columns: &[Option<BytesColumn>],
-    merge_row_order: &MergeRowOrder,
-    output: &mut impl Write,
-) -> io::Result<()> {
-    // Serialize dict and generate mapping for values
-    let mut output = CountingWriter::wrap(output);
-    // TODO !!! Remove useless terms.
-    let term_ord_mapping = serialize_merged_dict(bytes_columns, merge_row_order, &mut output)?;
-    let dictionary_num_bytes: u32 = output.written_bytes() as u32;
-    let output = output.finish();
-    let remapped_term_ordinals_values = RemappedTermOrdinalsValues {
-        bytes_columns,
-        term_ord_mapping: &term_ord_mapping,
-        merge_row_order,
-    };
-    serialize_column_mappable_to_u64(column_index, &remapped_term_ordinals_values, output)?;
-    output.write_all(&dictionary_num_bytes.to_le_bytes())?;
-    Ok(())
-}
-
-struct RemappedTermOrdinalsValues<'a> {
-    bytes_columns: &'a [Option<BytesColumn>],
-    term_ord_mapping: &'a TermOrdinalMapping,
-    merge_row_order: &'a MergeRowOrder,
-}
-
-impl<'a> Iterable for RemappedTermOrdinalsValues<'a> {
-    fn boxed_iter(&self) -> Box<dyn Iterator<Item = u64> + '_> {
-        match self.merge_row_order {
-            MergeRowOrder::Stack(_) => self.boxed_iter_stacked(),
-            MergeRowOrder::Shuffled(shuffle_merge_order) => {
-                self.boxed_iter_shuffled(shuffle_merge_order)
-            }
-        }
-    }
-}
-
-impl<'a> RemappedTermOrdinalsValues<'a> {
-    fn boxed_iter_stacked(&self) -> Box<dyn Iterator<Item = u64> + '_> {
-        let iter = self
-            .bytes_columns
-            .iter()
-            .enumerate()
-            .flat_map(|(segment_ord, byte_column)| {
-                let segment_ord = self.term_ord_mapping.get_segment(segment_ord as u32);
-                byte_column.iter().flat_map(move |bytes_column| {
-                    bytes_column
-                        .ords()
-                        .values
-                        .iter()
-                        .map(move |term_ord| segment_ord[term_ord as usize])
-                })
-            });
-        // TODO see if we can better decompose the mapping / and the stacking
-        Box::new(iter)
-    }
-
-    fn boxed_iter_shuffled<'b>(
-        &'b self,
-        shuffle_merge_order: &'b ShuffleMergeOrder,
-    ) -> Box<dyn Iterator<Item = u64> + 'b> {
-        Box::new(
-            shuffle_merge_order
-                .iter_new_to_old_row_addrs()
-                .flat_map(move |old_addr| {
-                    let segment_ord = self.term_ord_mapping.get_segment(old_addr.segment_ord);
-                    self.bytes_columns[old_addr.segment_ord as usize]
-                        .as_ref()
-                        .into_iter()
-                        .flat_map(move |bytes_column| {
-                            bytes_column
-                                .term_ords(old_addr.row_id)
-                                .map(|old_term_ord: u64| segment_ord[old_term_ord as usize])
-                        })
-                }),
-        )
-    }
-}
-
-fn compute_term_bitset(column: &BytesColumn, row_bitset: &ReadOnlyBitSet) -> BitSet {
-    let num_terms = column.dictionary().num_terms();
-    let mut term_bitset = BitSet::with_max_value(num_terms as u32);
-    for row_id in row_bitset.iter() {
-        for term_ord in column.term_ord_column.values_for_doc(row_id) {
-            term_bitset.insert(term_ord as u32);
-        }
-    }
-    term_bitset
-}
-
-fn is_term_present(bitsets: &[Option<BitSet>], term_merger: &TermMerger) -> bool {
-    for (segment_ord, from_term_ord) in term_merger.matching_segments() {
-        if let Some(bitset) = bitsets[segment_ord].as_ref() {
-            if bitset.contains(from_term_ord as u32) {
-                return true;
-            }
-        } else {
-            return true;
-        }
-    }
-    false
-}
-
-fn serialize_merged_dict(
-    bytes_columns: &[Option<BytesColumn>],
-    merge_row_order: &MergeRowOrder,
-    output: &mut impl Write,
-) -> io::Result<TermOrdinalMapping> {
-    let mut term_ord_mapping = TermOrdinalMapping::default();
-
-    let mut field_term_streams = Vec::new();
-    for column in bytes_columns.iter().flatten() {
-        term_ord_mapping.add_segment(column.dictionary.num_terms());
-        let terms = column.dictionary.stream()?;
-        field_term_streams.push(terms);
-    }
-
-    let mut merged_terms = TermMerger::new(field_term_streams);
-    let mut sstable_builder = sstable::VoidSSTable::writer(output);
-
-    // TODO support complex `merge_row_order`.
-    match merge_row_order {
-        MergeRowOrder::Stack(_) => {
-            let mut current_term_ord = 0;
-            while merged_terms.advance() {
-                let term_bytes: &[u8] = merged_terms.key();
-                sstable_builder.insert(term_bytes, &())?;
-                for (segment_ord, from_term_ord) in merged_terms.matching_segments() {
-                    term_ord_mapping.register_from_to(segment_ord, from_term_ord, current_term_ord);
-                }
-                current_term_ord += 1;
-            }
-            sstable_builder.finish()?;
-        }
-        MergeRowOrder::Shuffled(shuffle_merge_order) => {
-            assert_eq!(shuffle_merge_order.alive_bitsets.len(), bytes_columns.len());
-            let mut term_bitsets: Vec<Option<BitSet>> = Vec::with_capacity(bytes_columns.len());
-            for (alive_bitset_opt, bytes_column_opt) in shuffle_merge_order
-                .alive_bitsets
-                .iter()
-                .zip(bytes_columns.iter())
-            {
-                match (alive_bitset_opt, bytes_column_opt) {
-                    (Some(alive_bitset), Some(bytes_column)) => {
-                        let term_bitset = compute_term_bitset(bytes_column, alive_bitset);
-                        term_bitsets.push(Some(term_bitset));
-                    }
-                    _ => {
-                        term_bitsets.push(None);
-                    }
-                }
-            }
-            let mut current_term_ord = 0;
-            while merged_terms.advance() {
-                let term_bytes: &[u8] = merged_terms.key();
-                if !is_term_present(&term_bitsets[..], &merged_terms) {
-                    continue;
-                }
-                sstable_builder.insert(term_bytes, &())?;
-                for (segment_ord, from_term_ord) in merged_terms.matching_segments() {
-                    term_ord_mapping.register_from_to(segment_ord, from_term_ord, current_term_ord);
-                }
-                current_term_ord += 1;
-            }
-            sstable_builder.finish()?;
-        }
-    }
-    Ok(term_ord_mapping)
-}
-
-#[derive(Default, Debug)]
-struct TermOrdinalMapping {
-    per_segment_new_term_ordinals: Vec<Vec<TermOrdinal>>,
-}
-
-impl TermOrdinalMapping {
-    fn add_segment(&mut self, max_term_ord: usize) {
-        self.per_segment_new_term_ordinals
-            .push(vec![TermOrdinal::default(); max_term_ord]);
-    }
-
-    fn register_from_to(&mut self, segment_ord: usize, from_ord: TermOrdinal, to_ord: TermOrdinal) {
-        self.per_segment_new_term_ordinals[segment_ord][from_ord as usize] = to_ord;
-    }
-
-    fn get_segment(&self, segment_ord: u32) -> &[TermOrdinal] {
-        &(self.per_segment_new_term_ordinals[segment_ord as usize])[..]
-    }
-}

columnar/src/columnar/merge/merge_mapping.rs

@@ -1,118 +0,0 @@
-use std::ops::Range;
-
-use common::{BitSet, OwnedBytes, ReadOnlyBitSet};
-
-use crate::{ColumnarReader, RowAddr, RowId};
-
-pub struct StackMergeOrder {
-    // This does not start at 0. The first row is the number of
-    // rows in the first columnar.
-    cumulated_row_ids: Vec<RowId>,
-}
-
-impl StackMergeOrder {
-    pub fn stack(columnars: &[&ColumnarReader]) -> StackMergeOrder {
-        let mut cumulated_row_ids: Vec<RowId> = Vec::with_capacity(columnars.len());
-        let mut cumulated_row_id = 0;
-        for columnar in columnars {
-            cumulated_row_id += columnar.num_rows();
-            cumulated_row_ids.push(cumulated_row_id);
-        }
-        StackMergeOrder { cumulated_row_ids }
-    }
-
-    pub fn num_rows(&self) -> RowId {
-        self.cumulated_row_ids.last().copied().unwrap_or(0)
-    }
-
-    pub fn offset(&self, columnar_id: usize) -> RowId {
-        if columnar_id == 0 {
-            return 0;
-        }
-        self.cumulated_row_ids[columnar_id - 1]
-    }
-
-    pub fn columnar_range(&self, columnar_id: usize) -> Range<RowId> {
-        self.offset(columnar_id)..self.offset(columnar_id + 1)
-    }
-}
-
-pub enum MergeRowOrder {
-    /// Columnar tables are simply stacked one above the other.
-    /// If the i-th columnar_readers has n_rows_i rows, then
-    /// in the resulting columnar,
-    /// rows [r0..n_row_0) contains the row of columnar_readers[0], in ordder
-    /// rows [n_row_0..n_row_0 + n_row_1 contains the row of columnar_readers[1], in order.
-    /// ..
-    /// No documents is deleted.
-    Stack(StackMergeOrder),
-    /// Some more complex mapping, that may interleaves rows from the different readers and
-    /// drop rows, or do both.
-    Shuffled(ShuffleMergeOrder),
-}
-
-impl From<StackMergeOrder> for MergeRowOrder {
-    fn from(stack_merge_order: StackMergeOrder) -> MergeRowOrder {
-        MergeRowOrder::Stack(stack_merge_order)
-    }
-}
-
-impl From<ShuffleMergeOrder> for MergeRowOrder {
-    fn from(shuffle_merge_order: ShuffleMergeOrder) -> MergeRowOrder {
-        MergeRowOrder::Shuffled(shuffle_merge_order)
-    }
-}
-
-impl MergeRowOrder {
-    pub fn num_rows(&self) -> RowId {
-        match self {
-            MergeRowOrder::Stack(stack_row_order) => stack_row_order.num_rows(),
-            MergeRowOrder::Shuffled(complex_mapping) => complex_mapping.num_rows(),
-        }
-    }
-}
-
-pub struct ShuffleMergeOrder {
-    pub new_row_id_to_old_row_id: Vec<RowAddr>,
-    pub alive_bitsets: Vec<Option<ReadOnlyBitSet>>,
-}
-
-impl ShuffleMergeOrder {
-    pub fn for_test(
-        segment_num_rows: &[RowId],
-        new_row_id_to_old_row_id: Vec<RowAddr>,
-    ) -> ShuffleMergeOrder {
-        let mut alive_bitsets: Vec<BitSet> = segment_num_rows
-            .iter()
-            .map(|&num_rows| BitSet::with_max_value(num_rows))
-            .collect();
-        for &RowAddr {
-            segment_ord,
-            row_id,
-        } in &new_row_id_to_old_row_id
-        {
-            alive_bitsets[segment_ord as usize].insert(row_id);
-        }
-        let alive_bitsets: Vec<Option<ReadOnlyBitSet>> = alive_bitsets
-            .into_iter()
-            .map(|alive_bitset| {
-                let mut buffer = Vec::new();
-                alive_bitset.serialize(&mut buffer).unwrap();
-                let data = OwnedBytes::new(buffer);
-                Some(ReadOnlyBitSet::open(data))
-            })
-            .collect();
-        ShuffleMergeOrder {
-            new_row_id_to_old_row_id,
-            alive_bitsets,
-        }
-    }
-
-    pub fn num_rows(&self) -> RowId {
-        self.new_row_id_to_old_row_id.len() as RowId
-    }
-
-    pub fn iter_new_to_old_row_addrs(&self) -> impl Iterator<Item = RowAddr> + '_ {
-        self.new_row_id_to_old_row_id.iter().copied()
-    }
-}

columnar/src/columnar/merge/mod.rs

@@ -1,375 +0,0 @@
-mod merge_dict_column;
-mod merge_mapping;
-mod term_merger;
-
-use std::collections::{BTreeMap, HashMap, HashSet};
-use std::io;
-use std::net::Ipv6Addr;
-use std::sync::Arc;
-
-pub use merge_mapping::{MergeRowOrder, ShuffleMergeOrder, StackMergeOrder};
-
-use super::writer::ColumnarSerializer;
-use crate::column::{serialize_column_mappable_to_u128, serialize_column_mappable_to_u64};
-use crate::column_values::MergedColumnValues;
-use crate::columnar::merge::merge_dict_column::merge_bytes_or_str_column;
-use crate::columnar::writer::CompatibleNumericalTypes;
-use crate::columnar::ColumnarReader;
-use crate::dynamic_column::DynamicColumn;
-use crate::{
-    BytesColumn, Column, ColumnIndex, ColumnType, ColumnValues, NumericalType, NumericalValue,
-};
-
-/// Column types are grouped into different categories.
-/// After merge, all columns belonging to the same category are coerced to
-/// the same column type.
-///
-/// In practise, today, only Numerical colummns are coerced into one type today.
-///
-/// See also [README.md].
-#[derive(Copy, Clone, Eq, PartialEq, Hash, Debug)]
-enum ColumnTypeCategory {
-    Bool,
-    Str,
-    Numerical,
-    DateTime,
-    Bytes,
-    IpAddr,
-}
-
-impl From<ColumnType> for ColumnTypeCategory {
-    fn from(column_type: ColumnType) -> Self {
-        match column_type {
-            ColumnType::I64 => ColumnTypeCategory::Numerical,
-            ColumnType::U64 => ColumnTypeCategory::Numerical,
-            ColumnType::F64 => ColumnTypeCategory::Numerical,
-            ColumnType::Bytes => ColumnTypeCategory::Bytes,
-            ColumnType::Str => ColumnTypeCategory::Str,
-            ColumnType::Bool => ColumnTypeCategory::Bool,
-            ColumnType::IpAddr => ColumnTypeCategory::IpAddr,
-            ColumnType::DateTime => ColumnTypeCategory::DateTime,
-        }
-    }
-}
-
-/// Merge several columnar table together.
-///
-/// If several columns with the same name are conflicting with the numerical types in the
-/// input columnars, the first type compatible out of i64, u64, f64 in that order will be used.
-///
-/// `require_columns` makes it possible to ensure that some columns will be present in the
-/// resulting columnar. When a required column is a numerical column type, one of two things can
-/// happen:
-/// - If the required column type is compatible with all of the input columnar, the resulsting
-///   merged
-/// columnar will simply coerce the input column and use the required column type.
-/// - If the required column type is incompatible with one of the input columnar, the merged
-/// will fail with an InvalidData error.
-///
-/// `merge_row_order` makes it possible to remove or reorder row in the resulting
-/// `Columnar` table.
-///
-/// Reminder: a string and a numerical column may bare the same column name. This is not
-/// considered a conflict.
-pub fn merge_columnar(
-    columnar_readers: &[&ColumnarReader],
-    required_columns: &[(String, ColumnType)],
-    merge_row_order: MergeRowOrder,
-    output: &mut impl io::Write,
-) -> io::Result<()> {
-    let mut serializer = ColumnarSerializer::new(output);
-
-    let columns_to_merge = group_columns_for_merge(columnar_readers, required_columns)?;
-    for ((column_name, column_type), columns) in columns_to_merge {
-        let mut column_serializer =
-            serializer.serialize_column(column_name.as_bytes(), column_type);
-        merge_column(
-            column_type,
-            columns,
-            &merge_row_order,
-            &mut column_serializer,
-        )?;
-    }
-    serializer.finalize(merge_row_order.num_rows())?;
-
-    Ok(())
-}
-
-fn dynamic_column_to_u64_monotonic(dynamic_column: DynamicColumn) -> Option<Column<u64>> {
-    match dynamic_column {
-        DynamicColumn::Bool(column) => Some(column.to_u64_monotonic()),
-        DynamicColumn::I64(column) => Some(column.to_u64_monotonic()),
-        DynamicColumn::U64(column) => Some(column.to_u64_monotonic()),
-        DynamicColumn::F64(column) => Some(column.to_u64_monotonic()),
-        DynamicColumn::DateTime(column) => Some(column.to_u64_monotonic()),
-        DynamicColumn::IpAddr(_) | DynamicColumn::Bytes(_) | DynamicColumn::Str(_) => None,
-    }
-}
-
-fn merge_column(
-    column_type: ColumnType,
-    columns: Vec<Option<DynamicColumn>>,
-    merge_row_order: &MergeRowOrder,
-    wrt: &mut impl io::Write,
-) -> io::Result<()> {
-    match column_type {
-        ColumnType::I64
-        | ColumnType::U64
-        | ColumnType::F64
-        | ColumnType::DateTime
-        | ColumnType::Bool => {
-            let mut column_indexes: Vec<Option<ColumnIndex>> = Vec::with_capacity(columns.len());
-            let mut column_values: Vec<Option<Arc<dyn ColumnValues>>> =
-                Vec::with_capacity(columns.len());
-            for dynamic_column_opt in columns {
-                if let Some(Column { idx, values }) =
-                    dynamic_column_opt.and_then(dynamic_column_to_u64_monotonic)
-                {
-                    column_indexes.push(Some(idx));
-                    column_values.push(Some(values));
-                } else {
-                    column_indexes.push(None);
-                    column_values.push(None);
-                }
-            }
-            let merged_column_index =
-                crate::column_index::merge_column_index(&column_indexes[..], merge_row_order);
-            let merge_column_values = MergedColumnValues {
-                column_indexes: &column_indexes[..],
-                column_values: &column_values[..],
-                merge_row_order,
-            };
-            serialize_column_mappable_to_u64(merged_column_index, &merge_column_values, wrt)?;
-        }
-        ColumnType::IpAddr => {
-            let mut column_indexes: Vec<Option<ColumnIndex>> = Vec::with_capacity(columns.len());
-            let mut column_values: Vec<Option<Arc<dyn ColumnValues<Ipv6Addr>>>> =
-                Vec::with_capacity(columns.len());
-            for dynamic_column_opt in columns {
-                if let Some(DynamicColumn::IpAddr(Column { idx, values })) = dynamic_column_opt {
-                    column_indexes.push(Some(idx));
-                    column_values.push(Some(values));
-                } else {
-                    column_indexes.push(None);
-                    column_values.push(None);
-                }
-            }
-
-            let merged_column_index =
-                crate::column_index::merge_column_index(&column_indexes[..], merge_row_order);
-            let merge_column_values = MergedColumnValues {
-                column_indexes: &column_indexes[..],
-                column_values: &column_values,
-                merge_row_order,
-            };
-
-            serialize_column_mappable_to_u128(merged_column_index, &merge_column_values, wrt)?;
-        }
-        ColumnType::Bytes | ColumnType::Str => {
-            let mut column_indexes: Vec<Option<ColumnIndex>> = Vec::with_capacity(columns.len());
-            let mut bytes_columns: Vec<Option<BytesColumn>> = Vec::with_capacity(columns.len());
-            for dynamic_column_opt in columns {
-                match dynamic_column_opt {
-                    Some(DynamicColumn::Str(str_column)) => {
-                        column_indexes.push(Some(str_column.term_ord_column.idx.clone()));
-                        bytes_columns.push(Some(str_column.into()));
-                    }
-                    Some(DynamicColumn::Bytes(bytes_column)) => {
-                        column_indexes.push(Some(bytes_column.term_ord_column.idx.clone()));
-                        bytes_columns.push(Some(bytes_column));
-                    }
-                    _ => {
-                        column_indexes.push(None);
-                        bytes_columns.push(None);
-                    }
-                }
-            }
-            let merged_column_index =
-                crate::column_index::merge_column_index(&column_indexes[..], merge_row_order);
-            merge_bytes_or_str_column(merged_column_index, &bytes_columns, merge_row_order, wrt)?;
-        }
-    }
-    Ok(())
-}
-
-struct GroupedColumns {
-    required_column_type: Option<ColumnType>,
-    columns: Vec<Option<DynamicColumn>>,
-    column_category: ColumnTypeCategory,
-}
-
-impl GroupedColumns {
-    fn for_category(column_category: ColumnTypeCategory, num_columnars: usize) -> Self {
-        GroupedColumns {
-            required_column_type: None,
-            columns: vec![None; num_columnars],
-            column_category,
-        }
-    }
-
-    /// Set the dynamic column for a given columnar.
-    fn set_column(&mut self, columnar_id: usize, column: DynamicColumn) {
-        self.columns[columnar_id] = Some(column);
-    }
-
-    /// Force the existence of a column, as well as its type.
-    fn require_type(&mut self, required_type: ColumnType) -> io::Result<()> {
-        if let Some(existing_required_type) = self.required_column_type {
-            if existing_required_type == required_type {
-                // This was just a duplicate in the `required_columns`.
-                // Nothing to do.
-                return Ok(());
-            } else {
-                return Err(io::Error::new(
-                    io::ErrorKind::InvalidInput,
-                    "Required column conflicts with another required column of the same type \
-                     category.",
-                ));
-            }
-        }
-        self.required_column_type = Some(required_type);
-        Ok(())
-    }
-
-    /// Returns the column type after merge.
-    ///
-    /// This method does not check if the column types can actually be coerced to
-    /// this type.
-    fn column_type_after_merge(&self) -> ColumnType {
-        if let Some(required_type) = self.required_column_type {
-            return required_type;
-        }
-        let column_type: HashSet<ColumnType> = self
-            .columns
-            .iter()
-            .flatten()
-            .map(|column| column.column_type())
-            .collect();
-        if column_type.len() == 1 {
-            return column_type.into_iter().next().unwrap();
-        }
-        // At the moment, only the numerical categorical column type has more than one possible
-        // column type.
-        assert_eq!(self.column_category, ColumnTypeCategory::Numerical);
-        merged_numerical_columns_type(self.columns.iter().flatten()).into()
-    }
-}
-
-/// Returns the type of the merged numerical column.
-///
-/// This function picks the first numerical type out of i64, u64, f64 (order matters
-/// here), that is compatible with all the `columns`.
-///
-/// # Panics
-/// Panics if one of the column is not numerical.
-fn merged_numerical_columns_type<'a>(
-    columns: impl Iterator<Item = &'a DynamicColumn>,
-) -> NumericalType {
-    let mut compatible_numerical_types = CompatibleNumericalTypes::default();
-    for column in columns {
-        let (min_value, max_value) =
-            min_max_if_numerical(column).expect("All columns re required to be numerical");
-        compatible_numerical_types.accept_value(min_value);
-        compatible_numerical_types.accept_value(max_value);
-    }
-    compatible_numerical_types.to_numerical_type()
-}
-
-#[allow(clippy::type_complexity)]
-fn group_columns_for_merge(
-    columnar_readers: &[&ColumnarReader],
-    required_columns: &[(String, ColumnType)],
-) -> io::Result<BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>>> {
-    // Each column name may have multiple types of column associated.
-    // For merging we are interested in the same column type category since they can be merged.
-    let mut columns_grouped: HashMap<(String, ColumnTypeCategory), GroupedColumns> = HashMap::new();
-
-    for &(ref column_name, column_type) in required_columns {
-        columns_grouped
-            .entry((column_name.clone(), column_type.into()))
-            .or_insert_with(|| {
-                GroupedColumns::for_category(column_type.into(), columnar_readers.len())
-            })
-            .require_type(column_type)?;
-    }
-
-    for (columnar_id, columnar_reader) in columnar_readers.iter().enumerate() {
-        let column_name_and_handle = columnar_reader.list_columns()?;
-        for (column_name, handle) in column_name_and_handle {
-            let column_category: ColumnTypeCategory = handle.column_type().into();
-            let column = handle.open()?;
-            columns_grouped
-                .entry((column_name, column_category))
-                .or_insert_with(|| {
-                    GroupedColumns::for_category(column_category, columnar_readers.len())
-                })
-                .set_column(columnar_id, column);
-        }
-    }
-
-    let mut merge_columns: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
-        Default::default();
-
-    for ((column_name, _), mut grouped_columns) in columns_grouped {
-        let column_type = grouped_columns.column_type_after_merge();
-        coerce_columns(column_type, &mut grouped_columns.columns)?;
-        merge_columns.insert((column_name, column_type), grouped_columns.columns);
-    }
-
-    Ok(merge_columns)
-}
-
-fn coerce_columns(
-    column_type: ColumnType,
-    columns: &mut [Option<DynamicColumn>],
-) -> io::Result<()> {
-    for column_opt in columns.iter_mut() {
-        if let Some(column) = column_opt.take() {
-            *column_opt = Some(coerce_column(column_type, column)?);
-        }
-    }
-    Ok(())
-}
-
-fn coerce_column(column_type: ColumnType, column: DynamicColumn) -> io::Result<DynamicColumn> {
-    if let Some(numerical_type) = column_type.numerical_type() {
-        column
-            .coerce_numerical(numerical_type)
-            .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidInput, ""))
-    } else {
-        if column.column_type() != column_type {
-            return Err(io::Error::new(
-                io::ErrorKind::InvalidInput,
-                format!(
-                    "Cannot coerce column of type `{:?}` to `{column_type:?}`",
-                    column.column_type()
-                ),
-            ));
-        }
-        Ok(column)
-    }
-}
-
-/// Returns the (min, max) of a column provided it is numerical (i64, u64. f64).
-///
-/// The min and the max are simply the numerical value as defined by `ColumnValue::min_value()`,
-/// and `ColumnValue::max_value()`.
-///
-/// It is important to note that these values are only guaranteed to be lower/upper bound
-/// (as opposed to min/max value).
-/// If a column is empty, the min and max values are currently set to 0.
-fn min_max_if_numerical(column: &DynamicColumn) -> Option<(NumericalValue, NumericalValue)> {
-    match column {
-        DynamicColumn::I64(column) => Some((column.min_value().into(), column.max_value().into())),
-        DynamicColumn::U64(column) => Some((column.min_value().into(), column.min_value().into())),
-        DynamicColumn::F64(column) => Some((column.min_value().into(), column.min_value().into())),
-        DynamicColumn::Bool(_)
-        | DynamicColumn::IpAddr(_)
-        | DynamicColumn::DateTime(_)
-        | DynamicColumn::Bytes(_)
-        | DynamicColumn::Str(_) => None,
-    }
-}
-
-#[cfg(test)]
-mod tests;

columnar/src/columnar/merge/term_merger.rs

@@ -1,107 +0,0 @@
-use std::cmp::Ordering;
-use std::collections::BinaryHeap;
-
-use sstable::TermOrdinal;
-
-use crate::Streamer;
-
-pub struct HeapItem<'a> {
-    pub streamer: Streamer<'a>,
-    pub segment_ord: usize,
-}
-
-impl<'a> PartialEq for HeapItem<'a> {
-    fn eq(&self, other: &Self) -> bool {
-        self.segment_ord == other.segment_ord
-    }
-}
-
-impl<'a> Eq for HeapItem<'a> {}
-
-impl<'a> PartialOrd for HeapItem<'a> {
-    fn partial_cmp(&self, other: &HeapItem<'a>) -> Option<Ordering> {
-        Some(self.cmp(other))
-    }
-}
-
-impl<'a> Ord for HeapItem<'a> {
-    fn cmp(&self, other: &HeapItem<'a>) -> Ordering {
-        (&other.streamer.key(), &other.segment_ord).cmp(&(&self.streamer.key(), &self.segment_ord))
-    }
-}
-
-/// Given a list of sorted term streams,
-/// returns an iterator over sorted unique terms.
-///
-/// The item yield is actually a pair with
-/// - the term
-/// - a slice with the ordinal of the segments containing
-/// the terms.
-pub struct TermMerger<'a> {
-    heap: BinaryHeap<HeapItem<'a>>,
-    current_streamers: Vec<HeapItem<'a>>,
-}
-
-impl<'a> TermMerger<'a> {
-    /// Stream of merged term dictionary
-    pub fn new(streams: Vec<Streamer<'a>>) -> TermMerger<'a> {
-        TermMerger {
-            heap: BinaryHeap::new(),
-            current_streamers: streams
-                .into_iter()
-                .enumerate()
-                .map(|(ord, streamer)| HeapItem {
-                    streamer,
-                    segment_ord: ord,
-                })
-                .collect(),
-        }
-    }
-
-    pub(crate) fn matching_segments<'b: 'a>(
-        &'b self,
-    ) -> impl 'b + Iterator<Item = (usize, TermOrdinal)> {
-        self.current_streamers
-            .iter()
-            .map(|heap_item| (heap_item.segment_ord, heap_item.streamer.term_ord()))
-    }
-
-    fn advance_segments(&mut self) {
-        let streamers = &mut self.current_streamers;
-        let heap = &mut self.heap;
-        for mut heap_item in streamers.drain(..) {
-            if heap_item.streamer.advance() {
-                heap.push(heap_item);
-            }
-        }
-    }
-
-    /// Advance the term iterator to the next term.
-    /// Returns true if there is indeed another term
-    /// False if there is none.
-    pub fn advance(&mut self) -> bool {
-        self.advance_segments();
-        if let Some(head) = self.heap.pop() {
-            self.current_streamers.push(head);
-            while let Some(next_streamer) = self.heap.peek() {
-                if self.current_streamers[0].streamer.key() != next_streamer.streamer.key() {
-                    break;
-                }
-                let next_heap_it = self.heap.pop().unwrap(); // safe : we peeked beforehand
-                self.current_streamers.push(next_heap_it);
-            }
-            true
-        } else {
-            false
-        }
-    }
-
-    /// Returns the current term.
-    ///
-    /// This method may be called
-    /// if and only if advance() has been called before
-    /// and "true" was returned.
-    pub fn key(&self) -> &[u8] {
-        self.current_streamers[0].streamer.key()
-    }
-}

columnar/src/columnar/merge/tests.rs

@@ -1,318 +0,0 @@
-use super::*;
-use crate::{Cardinality, ColumnarWriter, HasAssociatedColumnType, RowId};
-
-fn make_columnar<T: Into<NumericalValue> + HasAssociatedColumnType + Copy>(
-    column_name: &str,
-    vals: &[T],
-) -> ColumnarReader {
-    let mut dataframe_writer = ColumnarWriter::default();
-    dataframe_writer.record_column_type(column_name, T::column_type(), false);
-    for (row_id, val) in vals.iter().copied().enumerate() {
-        dataframe_writer.record_numerical(row_id as RowId, column_name, val.into());
-    }
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer
-        .serialize(vals.len() as RowId, None, &mut buffer)
-        .unwrap();
-    ColumnarReader::open(buffer).unwrap()
-}
-
-#[test]
-fn test_column_coercion_to_u64() {
-    // i64 type
-    let columnar1 = make_columnar("numbers", &[1i64]);
-    // u64 type
-    let columnar2 = make_columnar("numbers", &[u64::MAX]);
-    let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
-        group_columns_for_merge(&[&columnar1, &columnar2], &[]).unwrap();
-    assert_eq!(column_map.len(), 1);
-    assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
-}
-
-#[test]
-fn test_column_no_coercion_if_all_the_same() {
-    let columnar1 = make_columnar("numbers", &[1u64]);
-    let columnar2 = make_columnar("numbers", &[2u64]);
-    let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
-        group_columns_for_merge(&[&columnar1, &columnar2], &[]).unwrap();
-    assert_eq!(column_map.len(), 1);
-    assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
-}
-
-#[test]
-fn test_column_coercion_to_i64() {
-    let columnar1 = make_columnar("numbers", &[-1i64]);
-    let columnar2 = make_columnar("numbers", &[2u64]);
-    let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
-        group_columns_for_merge(&[&columnar1, &columnar2], &[]).unwrap();
-    assert_eq!(column_map.len(), 1);
-    assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::I64)));
-}
-
-#[test]
-fn test_impossible_coercion_returns_an_error() {
-    let columnar1 = make_columnar("numbers", &[u64::MAX]);
-    let group_error =
-        group_columns_for_merge(&[&columnar1], &[("numbers".to_string(), ColumnType::I64)])
-            .map(|_| ())
-            .unwrap_err();
-    assert_eq!(group_error.kind(), io::ErrorKind::InvalidInput);
-}
-
-#[test]
-fn test_group_columns_with_required_column() {
-    let columnar1 = make_columnar("numbers", &[1i64]);
-    let columnar2 = make_columnar("numbers", &[2u64]);
-    let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
-        group_columns_for_merge(
-            &[&columnar1, &columnar2],
-            &[("numbers".to_string(), ColumnType::U64)],
-        )
-        .unwrap();
-    assert_eq!(column_map.len(), 1);
-    assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
-}
-
-#[test]
-fn test_group_columns_required_column_with_no_existing_columns() {
-    let columnar1 = make_columnar("numbers", &[2u64]);
-    let columnar2 = make_columnar("numbers", &[2u64]);
-    let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
-        group_columns_for_merge(
-            &[&columnar1, &columnar2],
-            &[("required_col".to_string(), ColumnType::Str)],
-        )
-        .unwrap();
-    assert_eq!(column_map.len(), 2);
-    let columns = column_map
-        .get(&("required_col".to_string(), ColumnType::Str))
-        .unwrap();
-    assert_eq!(columns.len(), 2);
-    assert!(columns[0].is_none());
-    assert!(columns[1].is_none());
-}
-
-#[test]
-fn test_group_columns_required_column_is_above_all_columns_have_the_same_type_rule() {
-    let columnar1 = make_columnar("numbers", &[2i64]);
-    let columnar2 = make_columnar("numbers", &[2i64]);
-    let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
-        group_columns_for_merge(
-            &[&columnar1, &columnar2],
-            &[("numbers".to_string(), ColumnType::U64)],
-        )
-        .unwrap();
-    assert_eq!(column_map.len(), 1);
-    assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
-}
-
-#[test]
-fn test_missing_column() {
-    let columnar1 = make_columnar("numbers", &[-1i64]);
-    let columnar2 = make_columnar("numbers2", &[2u64]);
-    let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
-        group_columns_for_merge(&[&columnar1, &columnar2], &[]).unwrap();
-    assert_eq!(column_map.len(), 2);
-    assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::I64)));
-    {
-        let columns = column_map
-            .get(&("numbers".to_string(), ColumnType::I64))
-            .unwrap();
-        assert!(columns[0].is_some());
-        assert!(columns[1].is_none());
-    }
-    {
-        let columns = column_map
-            .get(&("numbers2".to_string(), ColumnType::U64))
-            .unwrap();
-        assert!(columns[0].is_none());
-        assert!(columns[1].is_some());
-    }
-}
-
-fn make_numerical_columnar_multiple_columns(
-    columns: &[(&str, &[&[NumericalValue]])],
-) -> ColumnarReader {
-    let mut dataframe_writer = ColumnarWriter::default();
-    for (column_name, column_values) in columns {
-        for (row_id, vals) in column_values.iter().enumerate() {
-            for val in vals.iter() {
-                dataframe_writer.record_numerical(row_id as u32, column_name, *val);
-            }
-        }
-    }
-    let num_rows = columns
-        .iter()
-        .map(|(_, val_rows)| val_rows.len() as RowId)
-        .max()
-        .unwrap_or(0u32);
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer
-        .serialize(num_rows, None, &mut buffer)
-        .unwrap();
-    ColumnarReader::open(buffer).unwrap()
-}
-
-fn make_byte_columnar_multiple_columns(columns: &[(&str, &[&[&[u8]]])]) -> ColumnarReader {
-    let mut dataframe_writer = ColumnarWriter::default();
-    for (column_name, column_values) in columns {
-        for (row_id, vals) in column_values.iter().enumerate() {
-            for val in vals.iter() {
-                dataframe_writer.record_bytes(row_id as u32, column_name, val);
-            }
-        }
-    }
-    let num_rows = columns
-        .iter()
-        .map(|(_, val_rows)| val_rows.len() as RowId)
-        .max()
-        .unwrap_or(0u32);
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer
-        .serialize(num_rows, None, &mut buffer)
-        .unwrap();
-    ColumnarReader::open(buffer).unwrap()
-}
-
-fn make_text_columnar_multiple_columns(columns: &[(&str, &[&[&str]])]) -> ColumnarReader {
-    let mut dataframe_writer = ColumnarWriter::default();
-    for (column_name, column_values) in columns {
-        for (row_id, vals) in column_values.iter().enumerate() {
-            for val in vals.iter() {
-                dataframe_writer.record_str(row_id as u32, column_name, val);
-            }
-        }
-    }
-    let num_rows = columns
-        .iter()
-        .map(|(_, val_rows)| val_rows.len() as RowId)
-        .max()
-        .unwrap_or(0u32);
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer
-        .serialize(num_rows, None, &mut buffer)
-        .unwrap();
-    ColumnarReader::open(buffer).unwrap()
-}
-
-#[test]
-fn test_merge_columnar_numbers() {
-    let columnar1 =
-        make_numerical_columnar_multiple_columns(&[("numbers", &[&[NumericalValue::from(-1f64)]])]);
-    let columnar2 = make_numerical_columnar_multiple_columns(&[(
-        "numbers",
-        &[&[], &[NumericalValue::from(-3f64)]],
-    )]);
-    let mut buffer = Vec::new();
-    let columnars = &[&columnar1, &columnar2];
-    let stack_merge_order = StackMergeOrder::stack(columnars);
-    crate::columnar::merge_columnar(
-        columnars,
-        &[],
-        MergeRowOrder::Stack(stack_merge_order),
-        &mut buffer,
-    )
-    .unwrap();
-    let columnar_reader = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar_reader.num_rows(), 3);
-    assert_eq!(columnar_reader.num_columns(), 1);
-    let cols = columnar_reader.read_columns("numbers").unwrap();
-    let dynamic_column = cols[0].open().unwrap();
-    let DynamicColumn::F64(vals) = dynamic_column else { panic!() };
-    assert_eq!(vals.get_cardinality(), Cardinality::Optional);
-    assert_eq!(vals.first(0u32), Some(-1f64));
-    assert_eq!(vals.first(1u32), None);
-    assert_eq!(vals.first(2u32), Some(-3f64));
-}
-
-#[test]
-fn test_merge_columnar_texts() {
-    let columnar1 = make_text_columnar_multiple_columns(&[("texts", &[&["a"]])]);
-    let columnar2 = make_text_columnar_multiple_columns(&[("texts", &[&[], &["b"]])]);
-    let mut buffer = Vec::new();
-    let columnars = &[&columnar1, &columnar2];
-    let stack_merge_order = StackMergeOrder::stack(columnars);
-    crate::columnar::merge_columnar(
-        columnars,
-        &[],
-        MergeRowOrder::Stack(stack_merge_order),
-        &mut buffer,
-    )
-    .unwrap();
-    let columnar_reader = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar_reader.num_rows(), 3);
-    assert_eq!(columnar_reader.num_columns(), 1);
-    let cols = columnar_reader.read_columns("texts").unwrap();
-    let dynamic_column = cols[0].open().unwrap();
-    let DynamicColumn::Str(vals) = dynamic_column else { panic!() };
-    let get_str_for_ord = |ord| {
-        let mut out = String::new();
-        vals.ord_to_str(ord, &mut out).unwrap();
-        out
-    };
-
-    assert_eq!(vals.dictionary.num_terms(), 2);
-    assert_eq!(get_str_for_ord(0), "a");
-    assert_eq!(get_str_for_ord(1), "b");
-
-    let get_str_for_row = |row_id| {
-        let term_ords: Vec<u64> = vals.term_ords(row_id).collect();
-        assert!(term_ords.len() <= 1);
-        let mut out = String::new();
-        if term_ords.len() == 1 {
-            vals.ord_to_str(term_ords[0], &mut out).unwrap();
-        }
-        out
-    };
-
-    assert_eq!(get_str_for_row(0), "a");
-    assert_eq!(get_str_for_row(1), "");
-    assert_eq!(get_str_for_row(2), "b");
-}
-
-#[test]
-fn test_merge_columnar_byte() {
-    let columnar1 = make_byte_columnar_multiple_columns(&[("bytes", &[&[b"bbbb"], &[b"baaa"]])]);
-    let columnar2 = make_byte_columnar_multiple_columns(&[("bytes", &[&[], &[b"a"]])]);
-    let mut buffer = Vec::new();
-    let columnars = &[&columnar1, &columnar2];
-    let stack_merge_order = StackMergeOrder::stack(columnars);
-    crate::columnar::merge_columnar(
-        columnars,
-        &[],
-        MergeRowOrder::Stack(stack_merge_order),
-        &mut buffer,
-    )
-    .unwrap();
-    let columnar_reader = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar_reader.num_rows(), 4);
-    assert_eq!(columnar_reader.num_columns(), 1);
-    let cols = columnar_reader.read_columns("bytes").unwrap();
-    let dynamic_column = cols[0].open().unwrap();
-    let DynamicColumn::Bytes(vals) = dynamic_column else { panic!() };
-    let get_bytes_for_ord = |ord| {
-        let mut out = Vec::new();
-        vals.ord_to_bytes(ord, &mut out).unwrap();
-        out
-    };
-
-    assert_eq!(vals.dictionary.num_terms(), 3);
-    assert_eq!(get_bytes_for_ord(0), b"a");
-    assert_eq!(get_bytes_for_ord(1), b"baaa");
-    assert_eq!(get_bytes_for_ord(2), b"bbbb");
-
-    let get_bytes_for_row = |row_id| {
-        let term_ords: Vec<u64> = vals.term_ords(row_id).collect();
-        assert!(term_ords.len() <= 1);
-        let mut out = Vec::new();
-        if term_ords.len() == 1 {
-            vals.ord_to_bytes(term_ords[0], &mut out).unwrap();
-        }
-        out
-    };
-
-    assert_eq!(get_bytes_for_row(0), b"bbbb");
-    assert_eq!(get_bytes_for_row(1), b"baaa");
-    assert_eq!(get_bytes_for_row(2), b"");
-    assert_eq!(get_bytes_for_row(3), b"a");
-}

columnar/src/columnar/mod.rs

@@ -1,10 +0,0 @@
-mod column_type;
-mod format_version;
-mod merge;
-mod reader;
-mod writer;
-
-pub use column_type::{ColumnType, HasAssociatedColumnType};
-pub use merge::{merge_columnar, MergeRowOrder, ShuffleMergeOrder, StackMergeOrder};
-pub use reader::ColumnarReader;
-pub use writer::ColumnarWriter;

columnar/src/columnar/reader/mod.rs

@@ -1,192 +0,0 @@
-use std::{io, mem};
-
-use common::file_slice::FileSlice;
-use common::BinarySerializable;
-use sstable::{Dictionary, RangeSSTable};
-
-use crate::columnar::{format_version, ColumnType};
-use crate::dynamic_column::DynamicColumnHandle;
-use crate::RowId;
-
-fn io_invalid_data(msg: String) -> io::Error {
-    io::Error::new(io::ErrorKind::InvalidData, msg)
-}
-
-/// The ColumnarReader makes it possible to access a set of columns
-/// associated to field names.
-#[derive(Clone)]
-pub struct ColumnarReader {
-    column_dictionary: Dictionary<RangeSSTable>,
-    column_data: FileSlice,
-    num_rows: RowId,
-}
-
-/// Functions by both the async/sync code listing columns.
-/// It takes a stream from the column sstable and return the list of
-/// `DynamicColumn` available in it.
-fn read_all_columns_in_stream(
-    mut stream: sstable::Streamer<'_, RangeSSTable>,
-    column_data: &FileSlice,
-) -> io::Result<Vec<DynamicColumnHandle>> {
-    let mut results = Vec::new();
-    while stream.advance() {
-        let key_bytes: &[u8] = stream.key();
-        let Some(column_code) = key_bytes.last().copied() else {
-            return Err(io_invalid_data("Empty column name.".to_string()));
-        };
-        let column_type = ColumnType::try_from_code(column_code)
-            .map_err(|_| io_invalid_data(format!("Unknown column code `{column_code}`")))?;
-        let range = stream.value();
-        let file_slice = column_data.slice(range.start as usize..range.end as usize);
-        let dynamic_column_handle = DynamicColumnHandle {
-            file_slice,
-            column_type,
-        };
-        results.push(dynamic_column_handle);
-    }
-    Ok(results)
-}
-
-impl ColumnarReader {
-    /// Opens a new Columnar file.
-    pub fn open<F>(file_slice: F) -> io::Result<ColumnarReader>
-    where FileSlice: From<F> {
-        Self::open_inner(file_slice.into())
-    }
-
-    fn open_inner(file_slice: FileSlice) -> io::Result<ColumnarReader> {
-        let (file_slice_without_sstable_len, footer_slice) = file_slice
-            .split_from_end(mem::size_of::<u64>() + 4 + format_version::VERSION_FOOTER_NUM_BYTES);
-        let footer_bytes = footer_slice.read_bytes()?;
-        let sstable_len = u64::deserialize(&mut &footer_bytes[0..8])?;
-        let num_rows = u32::deserialize(&mut &footer_bytes[8..12])?;
-        let version_footer_bytes: [u8; format_version::VERSION_FOOTER_NUM_BYTES] =
-            footer_bytes[12..].try_into().unwrap();
-        let _version = format_version::parse_footer(version_footer_bytes)?;
-        let (column_data, sstable) =
-            file_slice_without_sstable_len.split_from_end(sstable_len as usize);
-        let column_dictionary = Dictionary::open(sstable)?;
-        Ok(ColumnarReader {
-            column_dictionary,
-            column_data,
-            num_rows,
-        })
-    }
-
-    pub fn num_rows(&self) -> RowId {
-        self.num_rows
-    }
-
-    // TODO Add unit tests
-    pub fn list_columns(&self) -> io::Result<Vec<(String, DynamicColumnHandle)>> {
-        let mut stream = self.column_dictionary.stream()?;
-        let mut results = Vec::new();
-        while stream.advance() {
-            let key_bytes: &[u8] = stream.key();
-            let column_code: u8 = key_bytes.last().cloned().unwrap();
-            let column_type: ColumnType = ColumnType::try_from_code(column_code)
-                .map_err(|_| io_invalid_data(format!("Unknown column code `{column_code}`")))?;
-            let range = stream.value().clone();
-            let column_name =
-                // The last two bytes are respectively the 0u8 separator and the column_type.
-                String::from_utf8_lossy(&key_bytes[..key_bytes.len() - 2]).to_string();
-            let file_slice = self
-                .column_data
-                .slice(range.start as usize..range.end as usize);
-            let column_handle = DynamicColumnHandle {
-                file_slice,
-                column_type,
-            };
-            results.push((column_name, column_handle));
-        }
-        Ok(results)
-    }
-
-    fn stream_for_column_range(&self, column_name: &str) -> sstable::StreamerBuilder<RangeSSTable> {
-        // Each column is a associated to a given `column_key`,
-        // that starts by `column_name\0column_header`.
-        //
-        // Listing the columns associated to the given column name is therefore equivalent to
-        // listing `column_key` with the prefix `column_name\0`.
-        //
-        // This is in turn equivalent to searching for the range
-        // `[column_name,\0`..column_name\1)`.
-        // TODO can we get some more generic `prefix(..)` logic in the dictionary.
-        let mut start_key = column_name.to_string();
-        start_key.push('\0');
-        let mut end_key = column_name.to_string();
-        end_key.push(1u8 as char);
-        self.column_dictionary
-            .range()
-            .ge(start_key.as_bytes())
-            .lt(end_key.as_bytes())
-    }
-
-    pub async fn read_columns_async(
-        &self,
-        column_name: &str,
-    ) -> io::Result<Vec<DynamicColumnHandle>> {
-        let stream = self
-            .stream_for_column_range(column_name)
-            .into_stream_async()
-            .await?;
-        read_all_columns_in_stream(stream, &self.column_data)
-    }
-
-    /// Get all columns for the given column name.
-    ///
-    /// There can be more than one column associated to a given column name, provided they have
-    /// different types.
-    pub fn read_columns(&self, column_name: &str) -> io::Result<Vec<DynamicColumnHandle>> {
-        let stream = self.stream_for_column_range(column_name).into_stream()?;
-        read_all_columns_in_stream(stream, &self.column_data)
-    }
-
-    /// Return the number of columns in the columnar.
-    pub fn num_columns(&self) -> usize {
-        self.column_dictionary.num_terms()
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use crate::{ColumnType, ColumnarReader, ColumnarWriter};
-
-    #[test]
-    fn test_list_columns() {
-        let mut columnar_writer = ColumnarWriter::default();
-        columnar_writer.record_column_type("col1", ColumnType::Str, false);
-        columnar_writer.record_column_type("col2", ColumnType::U64, false);
-        let mut buffer = Vec::new();
-        columnar_writer.serialize(1, None, &mut buffer).unwrap();
-        let columnar = ColumnarReader::open(buffer).unwrap();
-        let columns = columnar.list_columns().unwrap();
-        assert_eq!(columns.len(), 2);
-        assert_eq!(&columns[0].0, "col1");
-        assert_eq!(columns[0].1.column_type(), ColumnType::Str);
-        assert_eq!(&columns[1].0, "col2");
-        assert_eq!(columns[1].1.column_type(), ColumnType::U64);
-    }
-
-    #[test]
-    fn test_list_columns_strict_typing_prevents_coercion() {
-        let mut columnar_writer = ColumnarWriter::default();
-        columnar_writer.record_column_type("count", ColumnType::U64, false);
-        columnar_writer.record_numerical(1, "count", 1u64);
-        let mut buffer = Vec::new();
-        columnar_writer.serialize(2, None, &mut buffer).unwrap();
-        let columnar = ColumnarReader::open(buffer).unwrap();
-        let columns = columnar.list_columns().unwrap();
-        assert_eq!(columns.len(), 1);
-        assert_eq!(&columns[0].0, "count");
-        assert_eq!(columns[0].1.column_type(), ColumnType::U64);
-    }
-
-    #[test]
-    #[should_panic(expected = "Input type forbidden")]
-    fn test_list_columns_strict_typing_panics_on_wrong_types() {
-        let mut columnar_writer = ColumnarWriter::default();
-        columnar_writer.record_column_type("count", ColumnType::U64, false);
-        columnar_writer.record_numerical(1, "count", 1i64);
-    }
-}

columnar/src/columnar/writer/column_operation.rs

@@ -1,360 +0,0 @@
-use std::net::Ipv6Addr;
-
-use crate::dictionary::UnorderedId;
-use crate::utils::{place_bits, pop_first_byte, select_bits};
-use crate::value::NumericalValue;
-use crate::{InvalidData, NumericalType, RowId};
-
-/// When we build a columnar dataframe, we first just group
-/// all mutations per column, and appends them in append-only buffer
-/// in the stacker.
-///
-/// These ColumnOperation<T> are therefore serialize/deserialized
-/// in memory.
-///
-/// We represents all of these operations as `ColumnOperation`.
-#[derive(Eq, PartialEq, Debug, Clone, Copy)]
-pub(super) enum ColumnOperation<T> {
-    NewDoc(RowId),
-    Value(T),
-}
-
-#[derive(Copy, Clone, Eq, PartialEq, Debug)]
-struct ColumnOperationMetadata {
-    op_type: ColumnOperationType,
-    len: u8,
-}
-
-impl ColumnOperationMetadata {
-    fn to_code(self) -> u8 {
-        place_bits::<0, 6>(self.len) | place_bits::<6, 8>(self.op_type.to_code())
-    }
-
-    fn try_from_code(code: u8) -> Result<Self, InvalidData> {
-        let len = select_bits::<0, 6>(code);
-        let typ_code = select_bits::<6, 8>(code);
-        let column_type = ColumnOperationType::try_from_code(typ_code)?;
-        Ok(ColumnOperationMetadata {
-            op_type: column_type,
-            len,
-        })
-    }
-}
-
-#[derive(Copy, Clone, Eq, PartialEq, Debug)]
-#[repr(u8)]
-enum ColumnOperationType {
-    NewDoc = 0u8,
-    AddValue = 1u8,
-}
-
-impl ColumnOperationType {
-    pub fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    pub fn try_from_code(code: u8) -> Result<Self, InvalidData> {
-        match code {
-            0 => Ok(Self::NewDoc),
-            1 => Ok(Self::AddValue),
-            _ => Err(InvalidData),
-        }
-    }
-}
-
-impl<V: SymbolValue> ColumnOperation<V> {
-    pub(super) fn serialize(self) -> impl AsRef<[u8]> {
-        let mut minibuf = MiniBuffer::default();
-        let column_op_metadata = match self {
-            ColumnOperation::NewDoc(new_doc) => {
-                let symbol_len = new_doc.serialize(&mut minibuf.bytes[1..]);
-                ColumnOperationMetadata {
-                    op_type: ColumnOperationType::NewDoc,
-                    len: symbol_len,
-                }
-            }
-            ColumnOperation::Value(val) => {
-                let symbol_len = val.serialize(&mut minibuf.bytes[1..]);
-                ColumnOperationMetadata {
-                    op_type: ColumnOperationType::AddValue,
-                    len: symbol_len,
-                }
-            }
-        };
-        minibuf.bytes[0] = column_op_metadata.to_code();
-        // +1 for the metadata
-        minibuf.len = 1 + column_op_metadata.len;
-        minibuf
-    }
-
-    /// Deserialize a colummn operation.
-    /// Returns None if the buffer is empty.
-    ///
-    /// Panics if the payload is invalid:
-    /// this deserialize method is meant to target in memory.
-    pub(super) fn deserialize(bytes: &mut &[u8]) -> Option<Self> {
-        let column_op_metadata_byte = pop_first_byte(bytes)?;
-        let column_op_metadata = ColumnOperationMetadata::try_from_code(column_op_metadata_byte)
-            .expect("Invalid op metadata byte");
-        let symbol_bytes: &[u8];
-        (symbol_bytes, *bytes) = bytes.split_at(column_op_metadata.len as usize);
-        match column_op_metadata.op_type {
-            ColumnOperationType::NewDoc => {
-                let new_doc = u32::deserialize(symbol_bytes);
-                Some(ColumnOperation::NewDoc(new_doc))
-            }
-            ColumnOperationType::AddValue => {
-                let value = V::deserialize(symbol_bytes);
-                Some(ColumnOperation::Value(value))
-            }
-        }
-    }
-}
-
-impl<T> From<T> for ColumnOperation<T> {
-    fn from(value: T) -> Self {
-        ColumnOperation::Value(value)
-    }
-}
-
-// Serialization trait very local to the writer.
-// As we write fast fields, we accumulate them in "in memory".
-// In order to limit memory usage, and in order
-// to benefit from the stacker, we do this by serialization our data
-// as "Symbols".
-#[allow(clippy::from_over_into)]
-pub(super) trait SymbolValue: Clone + Copy {
-    // Serializes the symbol into the given buffer.
-    // Returns the number of bytes written into the buffer.
-    /// # Panics
-    /// May not exceed 9bytes
-    fn serialize(self, buffer: &mut [u8]) -> u8;
-    // Panics if invalid
-    fn deserialize(bytes: &[u8]) -> Self;
-}
-
-impl SymbolValue for bool {
-    fn serialize(self, buffer: &mut [u8]) -> u8 {
-        buffer[0] = u8::from(self);
-        1u8
-    }
-
-    fn deserialize(bytes: &[u8]) -> Self {
-        bytes[0] == 1u8
-    }
-}
-
-impl SymbolValue for Ipv6Addr {
-    fn serialize(self, buffer: &mut [u8]) -> u8 {
-        buffer[0..16].copy_from_slice(&self.octets());
-        16
-    }
-
-    fn deserialize(bytes: &[u8]) -> Self {
-        let octets: [u8; 16] = bytes[0..16].try_into().unwrap();
-        Ipv6Addr::from(octets)
-    }
-}
-
-#[derive(Default)]
-struct MiniBuffer {
-    pub bytes: [u8; 17],
-    pub len: u8,
-}
-
-impl AsRef<[u8]> for MiniBuffer {
-    fn as_ref(&self) -> &[u8] {
-        &self.bytes[..self.len as usize]
-    }
-}
-
-impl SymbolValue for NumericalValue {
-    fn deserialize(mut bytes: &[u8]) -> Self {
-        let type_code = pop_first_byte(&mut bytes).unwrap();
-        let symbol_type = NumericalType::try_from_code(type_code).unwrap();
-        let mut octet: [u8; 8] = [0u8; 8];
-        octet[..bytes.len()].copy_from_slice(bytes);
-        match symbol_type {
-            NumericalType::U64 => {
-                let val: u64 = u64::from_le_bytes(octet);
-                NumericalValue::U64(val)
-            }
-            NumericalType::I64 => {
-                let encoded: u64 = u64::from_le_bytes(octet);
-                let val: i64 = decode_zig_zag(encoded);
-                NumericalValue::I64(val)
-            }
-            NumericalType::F64 => {
-                debug_assert_eq!(bytes.len(), 8);
-                let val: f64 = f64::from_le_bytes(octet);
-                NumericalValue::F64(val)
-            }
-        }
-    }
-
-    /// F64: Serialize with a fixed size of 9 bytes
-    /// U64: Serialize without leading zeroes
-    /// I64: ZigZag encoded and serialize without leading zeroes
-    fn serialize(self, output: &mut [u8]) -> u8 {
-        match self {
-            NumericalValue::F64(val) => {
-                output[0] = NumericalType::F64 as u8;
-                output[1..9].copy_from_slice(&val.to_le_bytes());
-                9u8
-            }
-            NumericalValue::U64(val) => {
-                let len = compute_num_bytes_for_u64(val) as u8;
-                output[0] = NumericalType::U64 as u8;
-                output[1..9].copy_from_slice(&val.to_le_bytes());
-                len + 1u8
-            }
-            NumericalValue::I64(val) => {
-                let zig_zag_encoded = encode_zig_zag(val);
-                let len = compute_num_bytes_for_u64(zig_zag_encoded) as u8;
-                output[0] = NumericalType::I64 as u8;
-                output[1..9].copy_from_slice(&zig_zag_encoded.to_le_bytes());
-                len + 1u8
-            }
-        }
-    }
-}
-
-impl SymbolValue for u32 {
-    fn serialize(self, output: &mut [u8]) -> u8 {
-        let len = compute_num_bytes_for_u64(self as u64);
-        output[0..4].copy_from_slice(&self.to_le_bytes());
-        len as u8
-    }
-
-    fn deserialize(bytes: &[u8]) -> Self {
-        let mut quartet: [u8; 4] = [0u8; 4];
-        quartet[..bytes.len()].copy_from_slice(bytes);
-        u32::from_le_bytes(quartet)
-    }
-}
-
-impl SymbolValue for UnorderedId {
-    fn serialize(self, output: &mut [u8]) -> u8 {
-        self.0.serialize(output)
-    }
-
-    fn deserialize(bytes: &[u8]) -> Self {
-        UnorderedId(u32::deserialize(bytes))
-    }
-}
-
-fn compute_num_bytes_for_u64(val: u64) -> usize {
-    let msb = (64u32 - val.leading_zeros()) as usize;
-    (msb + 7) / 8
-}
-
-fn encode_zig_zag(n: i64) -> u64 {
-    ((n << 1) ^ (n >> 63)) as u64
-}
-
-fn decode_zig_zag(n: u64) -> i64 {
-    ((n >> 1) as i64) ^ (-((n & 1) as i64))
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[track_caller]
-    fn test_zig_zag_aux(val: i64) {
-        let encoded = super::encode_zig_zag(val);
-        assert_eq!(decode_zig_zag(encoded), val);
-        if let Some(abs_val) = val.checked_abs() {
-            let abs_val = abs_val as u64;
-            assert!(encoded <= abs_val * 2);
-        }
-    }
-
-    #[test]
-    fn test_zig_zag() {
-        assert_eq!(encode_zig_zag(0i64), 0u64);
-        assert_eq!(encode_zig_zag(-1i64), 1u64);
-        assert_eq!(encode_zig_zag(1i64), 2u64);
-        test_zig_zag_aux(0i64);
-        test_zig_zag_aux(i64::MIN);
-        test_zig_zag_aux(i64::MAX);
-    }
-
-    use proptest::prelude::any;
-    use proptest::proptest;
-
-    proptest! {
-        #[test]
-        fn test_proptest_zig_zag(val in any::<i64>()) {
-            test_zig_zag_aux(val);
-        }
-    }
-
-    #[test]
-    fn test_column_op_metadata_byte_serialization() {
-        for len in 0..=15 {
-            for op_type in [ColumnOperationType::AddValue, ColumnOperationType::NewDoc] {
-                let column_op_metadata = ColumnOperationMetadata { op_type, len };
-                let column_op_metadata_code = column_op_metadata.to_code();
-                let serdeser_metadata =
-                    ColumnOperationMetadata::try_from_code(column_op_metadata_code).unwrap();
-                assert_eq!(column_op_metadata, serdeser_metadata);
-            }
-        }
-    }
-
-    #[track_caller]
-    fn ser_deser_symbol(column_op: ColumnOperation<NumericalValue>) {
-        let buf = column_op.serialize();
-        let mut buffer = buf.as_ref().to_vec();
-        buffer.extend_from_slice(b"234234");
-        let mut bytes = &buffer[..];
-        let serdeser_symbol = ColumnOperation::deserialize(&mut bytes).unwrap();
-        assert_eq!(bytes.len() + buf.as_ref().len(), buffer.len());
-        assert_eq!(column_op, serdeser_symbol);
-    }
-
-    #[test]
-    fn test_compute_num_bytes_for_u64() {
-        assert_eq!(compute_num_bytes_for_u64(0), 0);
-        assert_eq!(compute_num_bytes_for_u64(1), 1);
-        assert_eq!(compute_num_bytes_for_u64(255), 1);
-        assert_eq!(compute_num_bytes_for_u64(256), 2);
-        assert_eq!(compute_num_bytes_for_u64((1 << 16) - 1), 2);
-        assert_eq!(compute_num_bytes_for_u64(1 << 16), 3);
-    }
-
-    #[test]
-    fn test_symbol_serialization() {
-        ser_deser_symbol(ColumnOperation::NewDoc(0));
-        ser_deser_symbol(ColumnOperation::NewDoc(3));
-        ser_deser_symbol(ColumnOperation::Value(NumericalValue::I64(0i64)));
-        ser_deser_symbol(ColumnOperation::Value(NumericalValue::I64(1i64)));
-        ser_deser_symbol(ColumnOperation::Value(NumericalValue::U64(257u64)));
-        ser_deser_symbol(ColumnOperation::Value(NumericalValue::I64(-257i64)));
-        ser_deser_symbol(ColumnOperation::Value(NumericalValue::I64(i64::MIN)));
-        ser_deser_symbol(ColumnOperation::Value(NumericalValue::U64(0u64)));
-        ser_deser_symbol(ColumnOperation::Value(NumericalValue::U64(u64::MIN)));
-        ser_deser_symbol(ColumnOperation::Value(NumericalValue::U64(u64::MAX)));
-    }
-
-    fn test_column_operation_unordered_aux(val: u32, expected_len: usize) {
-        let column_op = ColumnOperation::Value(UnorderedId(val));
-        let minibuf = column_op.serialize();
-        assert_eq!({ minibuf.as_ref().len() }, expected_len);
-        let mut buf = minibuf.as_ref().to_vec();
-        buf.extend_from_slice(&[2, 2, 2, 2, 2, 2]);
-        let mut cursor = &buf[..];
-        let column_op_serdeser: ColumnOperation<UnorderedId> =
-            ColumnOperation::deserialize(&mut cursor).unwrap();
-        assert_eq!(column_op_serdeser, ColumnOperation::Value(UnorderedId(val)));
-        assert_eq!(cursor.len() + expected_len, buf.len());
-    }
-
-    #[test]
-    fn test_column_operation_unordered() {
-        test_column_operation_unordered_aux(300u32, 3);
-        test_column_operation_unordered_aux(1u32, 2);
-        test_column_operation_unordered_aux(0u32, 1);
-    }
-}

columnar/src/columnar/writer/column_writers.rs

@@ -1,363 +0,0 @@
-use std::cmp::Ordering;
-
-use stacker::{ExpUnrolledLinkedList, MemoryArena};
-
-use crate::columnar::writer::column_operation::{ColumnOperation, SymbolValue};
-use crate::dictionary::{DictionaryBuilder, UnorderedId};
-use crate::{Cardinality, NumericalType, NumericalValue, RowId};
-
-#[derive(Copy, Clone, Debug, Eq, PartialEq)]
-#[repr(u8)]
-enum DocumentStep {
-    Same = 0,
-    Next = 1,
-    Skipped = 2,
-}
-
-#[inline(always)]
-fn delta_with_last_doc(last_doc_opt: Option<u32>, doc: u32) -> DocumentStep {
-    let expected_next_doc = last_doc_opt.map(|last_doc| last_doc + 1).unwrap_or(0u32);
-    match doc.cmp(&expected_next_doc) {
-        Ordering::Less => DocumentStep::Same,
-        Ordering::Equal => DocumentStep::Next,
-        Ordering::Greater => DocumentStep::Skipped,
-    }
-}
-
-#[derive(Copy, Clone, Default)]
-pub struct ColumnWriter {
-    // Detected cardinality of the column so far.
-    cardinality: Cardinality,
-    // Last document inserted.
-    // None if no doc has been added yet.
-    last_doc_opt: Option<u32>,
-    // Buffer containing the serialized values.
-    values: ExpUnrolledLinkedList,
-}
-
-impl ColumnWriter {
-    /// Returns an iterator over the Symbol that have been recorded
-    /// for the given column.
-    pub(super) fn operation_iterator<'a, V: SymbolValue>(
-        &self,
-        arena: &MemoryArena,
-        old_to_new_ids_opt: Option<&[RowId]>,
-        buffer: &'a mut Vec<u8>,
-    ) -> impl Iterator<Item = ColumnOperation<V>> + 'a {
-        buffer.clear();
-        self.values.read_to_end(arena, buffer);
-        if let Some(old_to_new_ids) = old_to_new_ids_opt {
-            // TODO avoid the extra deserialization / serialization.
-            let mut sorted_ops: Vec<(RowId, ColumnOperation<V>)> = Vec::new();
-            let mut new_doc = 0u32;
-            let mut cursor = &buffer[..];
-            for op in std::iter::from_fn(|| ColumnOperation::<V>::deserialize(&mut cursor)) {
-                if let ColumnOperation::NewDoc(doc) = &op {
-                    new_doc = old_to_new_ids[*doc as usize];
-                    sorted_ops.push((new_doc, ColumnOperation::NewDoc(new_doc)));
-                } else {
-                    sorted_ops.push((new_doc, op));
-                }
-            }
-            // stable sort is crucial here.
-            sorted_ops.sort_by_key(|(new_doc_id, _)| *new_doc_id);
-            buffer.clear();
-            for (_, op) in sorted_ops {
-                buffer.extend_from_slice(op.serialize().as_ref());
-            }
-        }
-        let mut cursor: &[u8] = &buffer[..];
-        std::iter::from_fn(move || ColumnOperation::deserialize(&mut cursor))
-    }
-
-    /// Records a change of the document being recorded.
-    ///
-    /// This function will also update the cardinality of the column
-    /// if necessary.
-    pub(super) fn record<S: SymbolValue>(&mut self, doc: RowId, value: S, arena: &mut MemoryArena) {
-        // Difference between `doc` and the last doc.
-        match delta_with_last_doc(self.last_doc_opt, doc) {
-            DocumentStep::Same => {
-                // This is the last encounterred document.
-                self.cardinality = Cardinality::Multivalued;
-            }
-            DocumentStep::Next => {
-                self.last_doc_opt = Some(doc);
-                self.write_symbol::<S>(ColumnOperation::NewDoc(doc), arena);
-            }
-            DocumentStep::Skipped => {
-                self.cardinality = self.cardinality.max(Cardinality::Optional);
-                self.last_doc_opt = Some(doc);
-                self.write_symbol::<S>(ColumnOperation::NewDoc(doc), arena);
-            }
-        }
-        self.write_symbol(ColumnOperation::Value(value), arena);
-    }
-
-    // Get the cardinality.
-    // The overall number of docs in the column is necessary to
-    // deal with the case where the all docs contain 1 value, except some documents
-    // at the end of the column.
-    pub(crate) fn get_cardinality(&self, num_docs: RowId) -> Cardinality {
-        match delta_with_last_doc(self.last_doc_opt, num_docs) {
-            DocumentStep::Same | DocumentStep::Next => self.cardinality,
-            DocumentStep::Skipped => self.cardinality.max(Cardinality::Optional),
-        }
-    }
-
-    /// Appends a new symbol to the `ColumnWriter`.
-    fn write_symbol<V: SymbolValue>(
-        &mut self,
-        column_operation: ColumnOperation<V>,
-        arena: &mut MemoryArena,
-    ) {
-        self.values
-            .writer(arena)
-            .extend_from_slice(column_operation.serialize().as_ref());
-    }
-}
-
-#[derive(Clone, Copy, Default)]
-pub(crate) struct NumericalColumnWriter {
-    compatible_numerical_types: CompatibleNumericalTypes,
-    column_writer: ColumnWriter,
-}
-
-impl NumericalColumnWriter {
-    pub fn force_numerical_type(&mut self, numerical_type: NumericalType) {
-        assert!(self
-            .compatible_numerical_types
-            .is_type_accepted(numerical_type));
-        self.compatible_numerical_types = CompatibleNumericalTypes::StaticType(numerical_type);
-    }
-}
-
-/// State used to store what types are still acceptable
-/// after having seen a set of numerical values.
-#[derive(Clone, Copy)]
-pub(crate) enum CompatibleNumericalTypes {
-    Dynamic {
-        all_values_within_i64_range: bool,
-        all_values_within_u64_range: bool,
-    },
-    StaticType(NumericalType),
-}
-
-impl Default for CompatibleNumericalTypes {
-    fn default() -> CompatibleNumericalTypes {
-        CompatibleNumericalTypes::Dynamic {
-            all_values_within_i64_range: true,
-            all_values_within_u64_range: true,
-        }
-    }
-}
-
-impl CompatibleNumericalTypes {
-    pub fn is_type_accepted(&self, numerical_type: NumericalType) -> bool {
-        match self {
-            CompatibleNumericalTypes::Dynamic {
-                all_values_within_i64_range,
-                all_values_within_u64_range,
-            } => match numerical_type {
-                NumericalType::I64 => *all_values_within_i64_range,
-                NumericalType::U64 => *all_values_within_u64_range,
-                NumericalType::F64 => true,
-            },
-            CompatibleNumericalTypes::StaticType(static_numerical_type) => {
-                *static_numerical_type == numerical_type
-            }
-        }
-    }
-
-    pub fn accept_value(&mut self, numerical_value: NumericalValue) {
-        match self {
-            CompatibleNumericalTypes::Dynamic {
-                all_values_within_i64_range,
-                all_values_within_u64_range,
-            } => match numerical_value {
-                NumericalValue::I64(val_i64) => {
-                    let value_within_u64_range = val_i64 >= 0i64;
-                    *all_values_within_u64_range &= value_within_u64_range;
-                }
-                NumericalValue::U64(val_u64) => {
-                    let value_within_i64_range = val_u64 < i64::MAX as u64;
-                    *all_values_within_i64_range &= value_within_i64_range;
-                }
-                NumericalValue::F64(_) => {
-                    *all_values_within_i64_range = false;
-                    *all_values_within_u64_range = false;
-                }
-            },
-            CompatibleNumericalTypes::StaticType(typ) => {
-                assert_eq!(
-                    numerical_value.numerical_type(),
-                    *typ,
-                    "Input type forbidden. This column has been forced to type {typ:?}, received \
-                     {numerical_value:?}"
-                );
-            }
-        }
-    }
-
-    pub fn to_numerical_type(self) -> NumericalType {
-        for numerical_type in [NumericalType::I64, NumericalType::U64] {
-            if self.is_type_accepted(numerical_type) {
-                return numerical_type;
-            }
-        }
-        NumericalType::F64
-    }
-}
-
-impl NumericalColumnWriter {
-    pub fn numerical_type(&self) -> NumericalType {
-        self.compatible_numerical_types.to_numerical_type()
-    }
-
-    pub fn cardinality(&self, num_docs: RowId) -> Cardinality {
-        self.column_writer.get_cardinality(num_docs)
-    }
-
-    pub fn record_numerical_value(
-        &mut self,
-        doc: RowId,
-        value: NumericalValue,
-        arena: &mut MemoryArena,
-    ) {
-        self.compatible_numerical_types.accept_value(value);
-        self.column_writer.record(doc, value, arena);
-    }
-
-    pub(super) fn operation_iterator<'a>(
-        self,
-        arena: &MemoryArena,
-        old_to_new_ids: Option<&[RowId]>,
-        buffer: &'a mut Vec<u8>,
-    ) -> impl Iterator<Item = ColumnOperation<NumericalValue>> + 'a {
-        self.column_writer
-            .operation_iterator(arena, old_to_new_ids, buffer)
-    }
-}
-
-#[derive(Copy, Clone)]
-pub(crate) struct StrOrBytesColumnWriter {
-    pub(crate) dictionary_id: u32,
-    pub(crate) column_writer: ColumnWriter,
-    // If true, when facing a multivalued cardinality,
-    // values associated to a given document will be sorted.
-    //
-    // This is useful for facets.
-    //
-    // If false, the order of appearance in the document will be
-    // observed.
-    pub(crate) sort_values_within_row: bool,
-}
-
-impl StrOrBytesColumnWriter {
-    pub(crate) fn with_dictionary_id(dictionary_id: u32) -> StrOrBytesColumnWriter {
-        StrOrBytesColumnWriter {
-            dictionary_id,
-            column_writer: Default::default(),
-            sort_values_within_row: false,
-        }
-    }
-
-    pub(crate) fn record_bytes(
-        &mut self,
-        doc: RowId,
-        bytes: &[u8],
-        dictionaries: &mut [DictionaryBuilder],
-        arena: &mut MemoryArena,
-    ) {
-        let unordered_id = dictionaries[self.dictionary_id as usize].get_or_allocate_id(bytes);
-        self.column_writer.record(doc, unordered_id, arena);
-    }
-
-    pub(super) fn operation_iterator<'a>(
-        &self,
-        arena: &MemoryArena,
-        old_to_new_ids: Option<&[RowId]>,
-        byte_buffer: &'a mut Vec<u8>,
-    ) -> impl Iterator<Item = ColumnOperation<UnorderedId>> + 'a {
-        self.column_writer
-            .operation_iterator(arena, old_to_new_ids, byte_buffer)
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_delta_with_last_doc() {
-        assert_eq!(delta_with_last_doc(None, 0u32), DocumentStep::Next);
-        assert_eq!(delta_with_last_doc(None, 1u32), DocumentStep::Skipped);
-        assert_eq!(delta_with_last_doc(None, 2u32), DocumentStep::Skipped);
-        assert_eq!(delta_with_last_doc(Some(0u32), 0u32), DocumentStep::Same);
-        assert_eq!(delta_with_last_doc(Some(1u32), 1u32), DocumentStep::Same);
-        assert_eq!(delta_with_last_doc(Some(1u32), 2u32), DocumentStep::Next);
-        assert_eq!(delta_with_last_doc(Some(1u32), 3u32), DocumentStep::Skipped);
-        assert_eq!(delta_with_last_doc(Some(1u32), 4u32), DocumentStep::Skipped);
-    }
-
-    #[track_caller]
-    fn test_column_writer_coercion_iter_aux(
-        values: impl Iterator<Item = NumericalValue>,
-        expected_numerical_type: NumericalType,
-    ) {
-        let mut compatible_numerical_types = CompatibleNumericalTypes::default();
-        for value in values {
-            compatible_numerical_types.accept_value(value);
-        }
-        assert_eq!(
-            compatible_numerical_types.to_numerical_type(),
-            expected_numerical_type
-        );
-    }
-
-    #[track_caller]
-    fn test_column_writer_coercion_aux(
-        values: &[NumericalValue],
-        expected_numerical_type: NumericalType,
-    ) {
-        test_column_writer_coercion_iter_aux(values.iter().copied(), expected_numerical_type);
-        test_column_writer_coercion_iter_aux(values.iter().rev().copied(), expected_numerical_type);
-    }
-
-    #[test]
-    fn test_column_writer_coercion() {
-        test_column_writer_coercion_aux(&[], NumericalType::I64);
-        test_column_writer_coercion_aux(&[1i64.into()], NumericalType::I64);
-        test_column_writer_coercion_aux(&[1u64.into()], NumericalType::I64);
-        // We don't detect exact integer at the moment. We could!
-        test_column_writer_coercion_aux(&[1f64.into()], NumericalType::F64);
-        test_column_writer_coercion_aux(&[u64::MAX.into()], NumericalType::U64);
-        test_column_writer_coercion_aux(&[(i64::MAX as u64).into()], NumericalType::U64);
-        test_column_writer_coercion_aux(&[(1u64 << 63).into()], NumericalType::U64);
-        test_column_writer_coercion_aux(&[1i64.into(), 1u64.into()], NumericalType::I64);
-        test_column_writer_coercion_aux(&[u64::MAX.into(), (-1i64).into()], NumericalType::F64);
-    }
-
-    #[test]
-    #[should_panic]
-    fn test_compatible_numerical_types_static_incompatible_type() {
-        let mut compatible_numerical_types =
-            CompatibleNumericalTypes::StaticType(NumericalType::U64);
-        compatible_numerical_types.accept_value(NumericalValue::I64(1i64));
-    }
-
-    #[test]
-    fn test_compatible_numerical_types_static_different_type_forbidden() {
-        let mut compatible_numerical_types =
-            CompatibleNumericalTypes::StaticType(NumericalType::U64);
-        compatible_numerical_types.accept_value(NumericalValue::U64(u64::MAX));
-    }
-
-    #[test]
-    fn test_compatible_numerical_types_static() {
-        for typ in [NumericalType::I64, NumericalType::I64, NumericalType::F64] {
-            let compatible_numerical_types = CompatibleNumericalTypes::StaticType(typ);
-            assert_eq!(compatible_numerical_types.to_numerical_type(), typ);
-        }
-    }
-}

columnar/src/columnar/writer/mod.rs

@@ -1,848 +0,0 @@
-mod column_operation;
-mod column_writers;
-mod serializer;
-mod value_index;
-
-use std::io;
-use std::net::Ipv6Addr;
-
-use column_operation::ColumnOperation;
-pub(crate) use column_writers::CompatibleNumericalTypes;
-use common::CountingWriter;
-pub(crate) use serializer::ColumnarSerializer;
-use stacker::{Addr, ArenaHashMap, MemoryArena};
-
-use crate::column_index::SerializableColumnIndex;
-use crate::column_values::{
-    ColumnValues, MonotonicallyMappableToU128, MonotonicallyMappableToU64, VecColumn,
-};
-use crate::columnar::column_type::ColumnType;
-use crate::columnar::writer::column_writers::{
-    ColumnWriter, NumericalColumnWriter, StrOrBytesColumnWriter,
-};
-use crate::columnar::writer::value_index::{IndexBuilder, PreallocatedIndexBuilders};
-use crate::dictionary::{DictionaryBuilder, TermIdMapping, UnorderedId};
-use crate::value::{Coerce, NumericalType, NumericalValue};
-use crate::{Cardinality, RowId};
-
-/// This is a set of buffers that are used to temporarily write the values into before passing them
-/// to the fast field codecs.
-#[derive(Default)]
-struct SpareBuffers {
-    value_index_builders: PreallocatedIndexBuilders,
-    u64_values: Vec<u64>,
-    ip_addr_values: Vec<Ipv6Addr>,
-}
-
-/// Makes it possible to create a new columnar.
-///
-/// ```rust
-/// use tantivy_columnar::ColumnarWriter;
-///
-/// let mut columnar_writer = ColumnarWriter::default();
-/// columnar_writer.record_str(0u32 /* doc id */, "product_name", "Red backpack");
-/// columnar_writer.record_numerical(0u32 /* doc id */, "price", 10u64);
-/// columnar_writer.record_str(1u32 /* doc id */, "product_name", "Apple");
-/// columnar_writer.record_numerical(0u32 /* doc id */, "price", 10.5f64); //< uh oh we ended up mixing integer and floats.
-/// let mut wrt: Vec<u8> =  Vec::new();
-/// columnar_writer.serialize(2u32, None, &mut wrt).unwrap();
-/// ```
-#[derive(Default)]
-pub struct ColumnarWriter {
-    numerical_field_hash_map: ArenaHashMap,
-    datetime_field_hash_map: ArenaHashMap,
-    bool_field_hash_map: ArenaHashMap,
-    ip_addr_field_hash_map: ArenaHashMap,
-    bytes_field_hash_map: ArenaHashMap,
-    str_field_hash_map: ArenaHashMap,
-    arena: MemoryArena,
-    // Dictionaries used to store dictionary-encoded values.
-    dictionaries: Vec<DictionaryBuilder>,
-    buffers: SpareBuffers,
-}
-
-#[inline]
-fn mutate_or_create_column<V, TMutator>(
-    arena_hash_map: &mut ArenaHashMap,
-    column_name: &str,
-    updater: TMutator,
-) where
-    V: Copy + 'static,
-    TMutator: FnMut(Option<V>) -> V,
-{
-    assert!(
-        !column_name.as_bytes().contains(&0u8),
-        "key may not contain the 0 byte"
-    );
-    arena_hash_map.mutate_or_create(column_name.as_bytes(), updater);
-}
-
-impl ColumnarWriter {
-    pub fn mem_usage(&self) -> usize {
-        // TODO add dictionary builders.
-        self.arena.mem_usage()
-            + self.numerical_field_hash_map.mem_usage()
-            + self.bool_field_hash_map.mem_usage()
-            + self.bytes_field_hash_map.mem_usage()
-            + self.str_field_hash_map.mem_usage()
-            + self.ip_addr_field_hash_map.mem_usage()
-            + self.datetime_field_hash_map.mem_usage()
-    }
-
-    /// Returns the list of doc ids from 0..num_docs sorted by the `sort_field`
-    /// column.
-    ///
-    /// If the column is multivalued, use the first value for scoring.
-    /// If no value is associated to a specific row, the document is assigned
-    /// the lowest possible score.
-    ///
-    /// The sort applied is stable.
-    pub fn sort_order(&self, sort_field: &str, num_docs: RowId, reversed: bool) -> Vec<u32> {
-        let Some(numerical_col_writer) =
-            self.numerical_field_hash_map.get::<NumericalColumnWriter>(sort_field.as_bytes()) else {
-                return Vec::new();
-        };
-        let mut symbols_buffer = Vec::new();
-        let mut values = Vec::new();
-        let mut last_doc_opt: Option<RowId> = None;
-        for op in numerical_col_writer.operation_iterator(&self.arena, None, &mut symbols_buffer) {
-            match op {
-                ColumnOperation::NewDoc(doc) => {
-                    last_doc_opt = Some(doc);
-                }
-                ColumnOperation::Value(numerical_value) => {
-                    if let Some(last_doc) = last_doc_opt {
-                        let score: f32 = f64::coerce(numerical_value) as f32;
-                        values.push((score, last_doc));
-                    }
-                }
-            }
-        }
-        for doc in values.len() as u32..num_docs {
-            values.push((0.0f32, doc));
-        }
-        values.sort_by(|(left_score, _), (right_score, _)| {
-            if reversed {
-                right_score.partial_cmp(left_score).unwrap()
-            } else {
-                left_score.partial_cmp(right_score).unwrap()
-            }
-        });
-        values.into_iter().map(|(_score, doc)| doc).collect()
-    }
-
-    /// Records a column type. This is useful to bypass the coercion process,
-    /// makes sure the empty is present in the resulting columnar, or set
-    /// the `sort_values_within_row`.
-    ///
-    /// `sort_values_within_row` is only allowed for `Bytes` or `Str` columns.
-    pub fn record_column_type(
-        &mut self,
-        column_name: &str,
-        column_type: ColumnType,
-        sort_values_within_row: bool,
-    ) {
-        if sort_values_within_row {
-            assert!(
-                column_type == ColumnType::Bytes || column_type == ColumnType::Str,
-                "sort_values_within_row is only allowed for Bytes and Str columns",
-            );
-        }
-        match column_type {
-            ColumnType::Str | ColumnType::Bytes => {
-                let (hash_map, dictionaries) = (
-                    if column_type == ColumnType::Str {
-                        &mut self.str_field_hash_map
-                    } else {
-                        &mut self.bytes_field_hash_map
-                    },
-                    &mut self.dictionaries,
-                );
-                mutate_or_create_column(
-                    hash_map,
-                    column_name,
-                    |column_opt: Option<StrOrBytesColumnWriter>| {
-                        let mut column_writer = if let Some(column_writer) = column_opt {
-                            column_writer
-                        } else {
-                            let dictionary_id = dictionaries.len() as u32;
-                            dictionaries.push(DictionaryBuilder::default());
-                            StrOrBytesColumnWriter::with_dictionary_id(dictionary_id)
-                        };
-                        column_writer.sort_values_within_row = sort_values_within_row;
-                        column_writer
-                    },
-                );
-            }
-            ColumnType::Bool => {
-                mutate_or_create_column(
-                    &mut self.bool_field_hash_map,
-                    column_name,
-                    |column_opt: Option<ColumnWriter>| column_opt.unwrap_or_default(),
-                );
-            }
-            ColumnType::DateTime => {
-                mutate_or_create_column(
-                    &mut self.datetime_field_hash_map,
-                    column_name,
-                    |column_opt: Option<ColumnWriter>| column_opt.unwrap_or_default(),
-                );
-            }
-            ColumnType::I64 | ColumnType::F64 | ColumnType::U64 => {
-                let numerical_type = column_type.numerical_type().unwrap();
-                mutate_or_create_column(
-                    &mut self.numerical_field_hash_map,
-                    column_name,
-                    |column_opt: Option<NumericalColumnWriter>| {
-                        let mut column: NumericalColumnWriter = column_opt.unwrap_or_default();
-                        column.force_numerical_type(numerical_type);
-                        column
-                    },
-                );
-            }
-            ColumnType::IpAddr => mutate_or_create_column(
-                &mut self.ip_addr_field_hash_map,
-                column_name,
-                |column_opt: Option<ColumnWriter>| column_opt.unwrap_or_default(),
-            ),
-        }
-    }
-
-    pub fn record_numerical<T: Into<NumericalValue> + Copy>(
-        &mut self,
-        doc: RowId,
-        column_name: &str,
-        numerical_value: T,
-    ) {
-        let (hash_map, arena) = (&mut self.numerical_field_hash_map, &mut self.arena);
-        mutate_or_create_column(
-            hash_map,
-            column_name,
-            |column_opt: Option<NumericalColumnWriter>| {
-                let mut column: NumericalColumnWriter = column_opt.unwrap_or_default();
-                column.record_numerical_value(doc, numerical_value.into(), arena);
-                column
-            },
-        );
-    }
-
-    pub fn record_ip_addr(&mut self, doc: RowId, column_name: &str, ip_addr: Ipv6Addr) {
-        assert!(
-            !column_name.as_bytes().contains(&0u8),
-            "key may not contain the 0 byte"
-        );
-        let (hash_map, arena) = (&mut self.ip_addr_field_hash_map, &mut self.arena);
-        hash_map.mutate_or_create(
-            column_name.as_bytes(),
-            |column_opt: Option<ColumnWriter>| {
-                let mut column: ColumnWriter = column_opt.unwrap_or_default();
-                column.record(doc, ip_addr, arena);
-                column
-            },
-        );
-    }
-
-    pub fn record_bool(&mut self, doc: RowId, column_name: &str, val: bool) {
-        let (hash_map, arena) = (&mut self.bool_field_hash_map, &mut self.arena);
-        mutate_or_create_column(hash_map, column_name, |column_opt: Option<ColumnWriter>| {
-            let mut column: ColumnWriter = column_opt.unwrap_or_default();
-            column.record(doc, val, arena);
-            column
-        });
-    }
-
-    pub fn record_datetime(&mut self, doc: RowId, column_name: &str, datetime: common::DateTime) {
-        let (hash_map, arena) = (&mut self.datetime_field_hash_map, &mut self.arena);
-        mutate_or_create_column(hash_map, column_name, |column_opt: Option<ColumnWriter>| {
-            let mut column: ColumnWriter = column_opt.unwrap_or_default();
-            column.record(
-                doc,
-                NumericalValue::I64(datetime.into_timestamp_micros()),
-                arena,
-            );
-            column
-        });
-    }
-
-    pub fn record_str(&mut self, doc: RowId, column_name: &str, value: &str) {
-        let (hash_map, arena, dictionaries) = (
-            &mut self.str_field_hash_map,
-            &mut self.arena,
-            &mut self.dictionaries,
-        );
-        hash_map.mutate_or_create(
-            column_name.as_bytes(),
-            |column_opt: Option<StrOrBytesColumnWriter>| {
-                let mut column: StrOrBytesColumnWriter = column_opt.unwrap_or_else(|| {
-                    // Each column has its own dictionary
-                    let dictionary_id = dictionaries.len() as u32;
-                    dictionaries.push(DictionaryBuilder::default());
-                    StrOrBytesColumnWriter::with_dictionary_id(dictionary_id)
-                });
-                column.record_bytes(doc, value.as_bytes(), dictionaries, arena);
-                column
-            },
-        );
-    }
-
-    pub fn record_bytes(&mut self, doc: RowId, column_name: &str, value: &[u8]) {
-        assert!(
-            !column_name.as_bytes().contains(&0u8),
-            "key may not contain the 0 byte"
-        );
-        let (hash_map, arena, dictionaries) = (
-            &mut self.bytes_field_hash_map,
-            &mut self.arena,
-            &mut self.dictionaries,
-        );
-        hash_map.mutate_or_create(
-            column_name.as_bytes(),
-            |column_opt: Option<StrOrBytesColumnWriter>| {
-                let mut column: StrOrBytesColumnWriter = column_opt.unwrap_or_else(|| {
-                    // Each column has its own dictionary
-                    let dictionary_id = dictionaries.len() as u32;
-                    dictionaries.push(DictionaryBuilder::default());
-                    StrOrBytesColumnWriter::with_dictionary_id(dictionary_id)
-                });
-                column.record_bytes(doc, value, dictionaries, arena);
-                column
-            },
-        );
-    }
-    pub fn serialize(
-        &mut self,
-        num_docs: RowId,
-        old_to_new_row_ids: Option<&[RowId]>,
-        wrt: &mut dyn io::Write,
-    ) -> io::Result<()> {
-        let mut serializer = ColumnarSerializer::new(wrt);
-        let mut columns: Vec<(&[u8], ColumnType, Addr)> = self
-            .numerical_field_hash_map
-            .iter()
-            .map(|(column_name, addr, _)| {
-                let numerical_column_writer: NumericalColumnWriter =
-                    self.numerical_field_hash_map.read(addr);
-                let column_type = numerical_column_writer.numerical_type().into();
-                (column_name, column_type, addr)
-            })
-            .collect();
-        columns.extend(
-            self.bytes_field_hash_map
-                .iter()
-                .map(|(term, addr, _)| (term, ColumnType::Bytes, addr)),
-        );
-        columns.extend(
-            self.str_field_hash_map
-                .iter()
-                .map(|(column_name, addr, _)| (column_name, ColumnType::Str, addr)),
-        );
-        columns.extend(
-            self.bool_field_hash_map
-                .iter()
-                .map(|(column_name, addr, _)| (column_name, ColumnType::Bool, addr)),
-        );
-        columns.extend(
-            self.ip_addr_field_hash_map
-                .iter()
-                .map(|(column_name, addr, _)| (column_name, ColumnType::IpAddr, addr)),
-        );
-        columns.extend(
-            self.datetime_field_hash_map
-                .iter()
-                .map(|(column_name, addr, _)| (column_name, ColumnType::DateTime, addr)),
-        );
-        columns.sort_unstable_by_key(|(column_name, col_type, _)| (*column_name, *col_type));
-
-        let (arena, buffers, dictionaries) = (&self.arena, &mut self.buffers, &self.dictionaries);
-        let mut symbol_byte_buffer: Vec<u8> = Vec::new();
-        for (column_name, column_type, addr) in columns {
-            match column_type {
-                ColumnType::Bool => {
-                    let column_writer: ColumnWriter = self.bool_field_hash_map.read(addr);
-                    let cardinality = column_writer.get_cardinality(num_docs);
-                    let mut column_serializer =
-                        serializer.serialize_column(column_name, column_type);
-                    serialize_bool_column(
-                        cardinality,
-                        num_docs,
-                        column_writer.operation_iterator(
-                            arena,
-                            old_to_new_row_ids,
-                            &mut symbol_byte_buffer,
-                        ),
-                        buffers,
-                        &mut column_serializer,
-                    )?;
-                }
-                ColumnType::IpAddr => {
-                    let column_writer: ColumnWriter = self.ip_addr_field_hash_map.read(addr);
-                    let cardinality = column_writer.get_cardinality(num_docs);
-                    let mut column_serializer =
-                        serializer.serialize_column(column_name, ColumnType::IpAddr);
-                    serialize_ip_addr_column(
-                        cardinality,
-                        num_docs,
-                        column_writer.operation_iterator(
-                            arena,
-                            old_to_new_row_ids,
-                            &mut symbol_byte_buffer,
-                        ),
-                        buffers,
-                        &mut column_serializer,
-                    )?;
-                }
-                ColumnType::Bytes | ColumnType::Str => {
-                    let str_or_bytes_column_writer: StrOrBytesColumnWriter =
-                        if column_type == ColumnType::Bytes {
-                            self.bytes_field_hash_map.read(addr)
-                        } else {
-                            self.str_field_hash_map.read(addr)
-                        };
-                    let dictionary_builder =
-                        &dictionaries[str_or_bytes_column_writer.dictionary_id as usize];
-                    let cardinality = str_or_bytes_column_writer
-                        .column_writer
-                        .get_cardinality(num_docs);
-                    let mut column_serializer =
-                        serializer.serialize_column(column_name, column_type);
-                    serialize_bytes_or_str_column(
-                        cardinality,
-                        num_docs,
-                        str_or_bytes_column_writer.sort_values_within_row,
-                        dictionary_builder,
-                        str_or_bytes_column_writer.operation_iterator(
-                            arena,
-                            old_to_new_row_ids,
-                            &mut symbol_byte_buffer,
-                        ),
-                        buffers,
-                        &mut column_serializer,
-                    )?;
-                }
-                ColumnType::F64 | ColumnType::I64 | ColumnType::U64 => {
-                    let numerical_column_writer: NumericalColumnWriter =
-                        self.numerical_field_hash_map.read(addr);
-                    let cardinality = numerical_column_writer.cardinality(num_docs);
-                    let mut column_serializer =
-                        serializer.serialize_column(column_name, column_type);
-                    let numerical_type = column_type.numerical_type().unwrap();
-                    serialize_numerical_column(
-                        cardinality,
-                        num_docs,
-                        numerical_type,
-                        numerical_column_writer.operation_iterator(
-                            arena,
-                            old_to_new_row_ids,
-                            &mut symbol_byte_buffer,
-                        ),
-                        buffers,
-                        &mut column_serializer,
-                    )?;
-                }
-                ColumnType::DateTime => {
-                    let column_writer: ColumnWriter = self.datetime_field_hash_map.read(addr);
-                    let cardinality = column_writer.get_cardinality(num_docs);
-                    let mut column_serializer =
-                        serializer.serialize_column(column_name, ColumnType::DateTime);
-                    serialize_numerical_column(
-                        cardinality,
-                        num_docs,
-                        NumericalType::I64,
-                        column_writer.operation_iterator(
-                            arena,
-                            old_to_new_row_ids,
-                            &mut symbol_byte_buffer,
-                        ),
-                        buffers,
-                        &mut column_serializer,
-                    )?;
-                }
-            };
-        }
-        serializer.finalize(num_docs)?;
-        Ok(())
-    }
-}
-
-// Serialize [Dictionary, Column, dictionary num bytes U32::LE]
-// Column: [Column Index, Column Values, column index num bytes U32::LE]
-fn serialize_bytes_or_str_column(
-    cardinality: Cardinality,
-    num_docs: RowId,
-    sort_values_within_row: bool,
-    dictionary_builder: &DictionaryBuilder,
-    operation_it: impl Iterator<Item = ColumnOperation<UnorderedId>>,
-    buffers: &mut SpareBuffers,
-    wrt: impl io::Write,
-) -> io::Result<()> {
-    let SpareBuffers {
-        value_index_builders,
-        u64_values,
-        ..
-    } = buffers;
-    let mut counting_writer = CountingWriter::wrap(wrt);
-    let term_id_mapping: TermIdMapping = dictionary_builder.serialize(&mut counting_writer)?;
-    let dictionary_num_bytes: u32 = counting_writer.written_bytes() as u32;
-    let mut wrt = counting_writer.finish();
-    let operation_iterator = operation_it.map(|symbol: ColumnOperation<UnorderedId>| {
-        // We map unordered ids to ordered ids.
-        match symbol {
-            ColumnOperation::Value(unordered_id) => {
-                let ordered_id = term_id_mapping.to_ord(unordered_id);
-                ColumnOperation::Value(ordered_id.0 as u64)
-            }
-            ColumnOperation::NewDoc(doc) => ColumnOperation::NewDoc(doc),
-        }
-    });
-    send_to_serialize_column_mappable_to_u64(
-        operation_iterator,
-        cardinality,
-        num_docs,
-        sort_values_within_row,
-        value_index_builders,
-        u64_values,
-        &mut wrt,
-    )?;
-    wrt.write_all(&dictionary_num_bytes.to_le_bytes()[..])?;
-    Ok(())
-}
-
-fn serialize_numerical_column(
-    cardinality: Cardinality,
-    num_docs: RowId,
-    numerical_type: NumericalType,
-    op_iterator: impl Iterator<Item = ColumnOperation<NumericalValue>>,
-    buffers: &mut SpareBuffers,
-    wrt: &mut impl io::Write,
-) -> io::Result<()> {
-    let SpareBuffers {
-        value_index_builders,
-        u64_values,
-        ..
-    } = buffers;
-    match numerical_type {
-        NumericalType::I64 => {
-            send_to_serialize_column_mappable_to_u64(
-                coerce_numerical_symbol::<i64>(op_iterator),
-                cardinality,
-                num_docs,
-                false,
-                value_index_builders,
-                u64_values,
-                wrt,
-            )?;
-        }
-        NumericalType::U64 => {
-            send_to_serialize_column_mappable_to_u64(
-                coerce_numerical_symbol::<u64>(op_iterator),
-                cardinality,
-                num_docs,
-                false,
-                value_index_builders,
-                u64_values,
-                wrt,
-            )?;
-        }
-        NumericalType::F64 => {
-            send_to_serialize_column_mappable_to_u64(
-                coerce_numerical_symbol::<f64>(op_iterator),
-                cardinality,
-                num_docs,
-                false,
-                value_index_builders,
-                u64_values,
-                wrt,
-            )?;
-        }
-    };
-    Ok(())
-}
-
-fn serialize_bool_column(
-    cardinality: Cardinality,
-    num_docs: RowId,
-    column_operations_it: impl Iterator<Item = ColumnOperation<bool>>,
-    buffers: &mut SpareBuffers,
-    wrt: &mut impl io::Write,
-) -> io::Result<()> {
-    let SpareBuffers {
-        value_index_builders,
-        u64_values,
-        ..
-    } = buffers;
-    send_to_serialize_column_mappable_to_u64(
-        column_operations_it.map(|bool_column_operation| match bool_column_operation {
-            ColumnOperation::NewDoc(doc) => ColumnOperation::NewDoc(doc),
-            ColumnOperation::Value(bool_val) => ColumnOperation::Value(bool_val.to_u64()),
-        }),
-        cardinality,
-        num_docs,
-        false,
-        value_index_builders,
-        u64_values,
-        wrt,
-    )?;
-    Ok(())
-}
-
-fn serialize_ip_addr_column(
-    cardinality: Cardinality,
-    num_docs: RowId,
-    column_operations_it: impl Iterator<Item = ColumnOperation<Ipv6Addr>>,
-    buffers: &mut SpareBuffers,
-    wrt: &mut impl io::Write,
-) -> io::Result<()> {
-    let SpareBuffers {
-        value_index_builders,
-        ip_addr_values,
-        ..
-    } = buffers;
-    send_to_serialize_column_mappable_to_u128(
-        column_operations_it,
-        cardinality,
-        num_docs,
-        value_index_builders,
-        ip_addr_values,
-        wrt,
-    )?;
-    Ok(())
-}
-
-fn send_to_serialize_column_mappable_to_u128<
-    T: Copy + Ord + std::fmt::Debug + Send + Sync + MonotonicallyMappableToU128 + PartialOrd,
->(
-    op_iterator: impl Iterator<Item = ColumnOperation<T>>,
-    cardinality: Cardinality,
-    num_rows: RowId,
-    value_index_builders: &mut PreallocatedIndexBuilders,
-    values: &mut Vec<T>,
-    mut wrt: impl io::Write,
-) -> io::Result<()>
-where
-    for<'a> VecColumn<'a, T>: ColumnValues<T>,
-{
-    values.clear();
-    // TODO: split index and values
-    let serializable_column_index = match cardinality {
-        Cardinality::Full => {
-            consume_operation_iterator(
-                op_iterator,
-                value_index_builders.borrow_required_index_builder(),
-                values,
-            );
-            SerializableColumnIndex::Full
-        }
-        Cardinality::Optional => {
-            let optional_index_builder = value_index_builders.borrow_optional_index_builder();
-            consume_operation_iterator(op_iterator, optional_index_builder, values);
-            let optional_index = optional_index_builder.finish(num_rows);
-            SerializableColumnIndex::Optional {
-                num_rows,
-                non_null_row_ids: Box::new(optional_index),
-            }
-        }
-        Cardinality::Multivalued => {
-            let multivalued_index_builder = value_index_builders.borrow_multivalued_index_builder();
-            consume_operation_iterator(op_iterator, multivalued_index_builder, values);
-            let multivalued_index = multivalued_index_builder.finish(num_rows);
-            SerializableColumnIndex::Multivalued(Box::new(multivalued_index))
-        }
-    };
-    crate::column::serialize_column_mappable_to_u128(
-        serializable_column_index,
-        &&values[..],
-        &mut wrt,
-    )?;
-    Ok(())
-}
-
-fn sort_values_within_row_in_place(multivalued_index: &[RowId], values: &mut [u64]) {
-    let mut start_index: usize = 0;
-    for end_index in multivalued_index.iter().copied() {
-        let end_index = end_index as usize;
-        values[start_index..end_index].sort_unstable();
-        start_index = end_index;
-    }
-}
-
-fn send_to_serialize_column_mappable_to_u64(
-    op_iterator: impl Iterator<Item = ColumnOperation<u64>>,
-    cardinality: Cardinality,
-    num_rows: RowId,
-    sort_values_within_row: bool,
-    value_index_builders: &mut PreallocatedIndexBuilders,
-    values: &mut Vec<u64>,
-    mut wrt: impl io::Write,
-) -> io::Result<()>
-where
-    for<'a> VecColumn<'a, u64>: ColumnValues<u64>,
-{
-    values.clear();
-    let serializable_column_index = match cardinality {
-        Cardinality::Full => {
-            consume_operation_iterator(
-                op_iterator,
-                value_index_builders.borrow_required_index_builder(),
-                values,
-            );
-            SerializableColumnIndex::Full
-        }
-        Cardinality::Optional => {
-            let optional_index_builder = value_index_builders.borrow_optional_index_builder();
-            consume_operation_iterator(op_iterator, optional_index_builder, values);
-            let optional_index = optional_index_builder.finish(num_rows);
-            SerializableColumnIndex::Optional {
-                non_null_row_ids: Box::new(optional_index),
-                num_rows,
-            }
-        }
-        Cardinality::Multivalued => {
-            let multivalued_index_builder = value_index_builders.borrow_multivalued_index_builder();
-            consume_operation_iterator(op_iterator, multivalued_index_builder, values);
-            let multivalued_index = multivalued_index_builder.finish(num_rows);
-            if sort_values_within_row {
-                sort_values_within_row_in_place(multivalued_index, values);
-            }
-            SerializableColumnIndex::Multivalued(Box::new(multivalued_index))
-        }
-    };
-    crate::column::serialize_column_mappable_to_u64(
-        serializable_column_index,
-        &&values[..],
-        &mut wrt,
-    )?;
-    Ok(())
-}
-
-fn coerce_numerical_symbol<T>(
-    operation_iterator: impl Iterator<Item = ColumnOperation<NumericalValue>>,
-) -> impl Iterator<Item = ColumnOperation<u64>>
-where T: Coerce + MonotonicallyMappableToU64 {
-    operation_iterator.map(|symbol| match symbol {
-        ColumnOperation::NewDoc(doc) => ColumnOperation::NewDoc(doc),
-        ColumnOperation::Value(numerical_value) => {
-            ColumnOperation::Value(T::coerce(numerical_value).to_u64())
-        }
-    })
-}
-
-fn consume_operation_iterator<T: Ord, TIndexBuilder: IndexBuilder>(
-    operation_iterator: impl Iterator<Item = ColumnOperation<T>>,
-    index_builder: &mut TIndexBuilder,
-    values: &mut Vec<T>,
-) {
-    for symbol in operation_iterator {
-        match symbol {
-            ColumnOperation::NewDoc(doc) => {
-                index_builder.record_row(doc);
-            }
-            ColumnOperation::Value(value) => {
-                index_builder.record_value();
-                values.push(value);
-            }
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use stacker::MemoryArena;
-
-    use crate::columnar::writer::column_operation::ColumnOperation;
-    use crate::{Cardinality, NumericalValue};
-
-    #[test]
-    fn test_column_writer_required_simple() {
-        let mut arena = MemoryArena::default();
-        let mut column_writer = super::ColumnWriter::default();
-        column_writer.record(0u32, NumericalValue::from(14i64), &mut arena);
-        column_writer.record(1u32, NumericalValue::from(15i64), &mut arena);
-        column_writer.record(2u32, NumericalValue::from(-16i64), &mut arena);
-        assert_eq!(column_writer.get_cardinality(3), Cardinality::Full);
-        let mut buffer = Vec::new();
-        let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
-            .operation_iterator(&arena, None, &mut buffer)
-            .collect();
-        assert_eq!(symbols.len(), 6);
-        assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));
-        assert!(matches!(
-            symbols[1],
-            ColumnOperation::Value(NumericalValue::I64(14i64))
-        ));
-        assert!(matches!(symbols[2], ColumnOperation::NewDoc(1u32)));
-        assert!(matches!(
-            symbols[3],
-            ColumnOperation::Value(NumericalValue::I64(15i64))
-        ));
-        assert!(matches!(symbols[4], ColumnOperation::NewDoc(2u32)));
-        assert!(matches!(
-            symbols[5],
-            ColumnOperation::Value(NumericalValue::I64(-16i64))
-        ));
-    }
-
-    #[test]
-    fn test_column_writer_optional_cardinality_missing_first() {
-        let mut arena = MemoryArena::default();
-        let mut column_writer = super::ColumnWriter::default();
-        column_writer.record(1u32, NumericalValue::from(15i64), &mut arena);
-        column_writer.record(2u32, NumericalValue::from(-16i64), &mut arena);
-        assert_eq!(column_writer.get_cardinality(3), Cardinality::Optional);
-        let mut buffer = Vec::new();
-        let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
-            .operation_iterator(&arena, None, &mut buffer)
-            .collect();
-        assert_eq!(symbols.len(), 4);
-        assert!(matches!(symbols[0], ColumnOperation::NewDoc(1u32)));
-        assert!(matches!(
-            symbols[1],
-            ColumnOperation::Value(NumericalValue::I64(15i64))
-        ));
-        assert!(matches!(symbols[2], ColumnOperation::NewDoc(2u32)));
-        assert!(matches!(
-            symbols[3],
-            ColumnOperation::Value(NumericalValue::I64(-16i64))
-        ));
-    }
-
-    #[test]
-    fn test_column_writer_optional_cardinality_missing_last() {
-        let mut arena = MemoryArena::default();
-        let mut column_writer = super::ColumnWriter::default();
-        column_writer.record(0u32, NumericalValue::from(15i64), &mut arena);
-        assert_eq!(column_writer.get_cardinality(2), Cardinality::Optional);
-        let mut buffer = Vec::new();
-        let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
-            .operation_iterator(&arena, None, &mut buffer)
-            .collect();
-        assert_eq!(symbols.len(), 2);
-        assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));
-        assert!(matches!(
-            symbols[1],
-            ColumnOperation::Value(NumericalValue::I64(15i64))
-        ));
-    }
-
-    #[test]
-    fn test_column_writer_multivalued() {
-        let mut arena = MemoryArena::default();
-        let mut column_writer = super::ColumnWriter::default();
-        column_writer.record(0u32, NumericalValue::from(16i64), &mut arena);
-        column_writer.record(0u32, NumericalValue::from(17i64), &mut arena);
-        assert_eq!(column_writer.get_cardinality(1), Cardinality::Multivalued);
-        let mut buffer = Vec::new();
-        let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
-            .operation_iterator(&arena, None, &mut buffer)
-            .collect();
-        assert_eq!(symbols.len(), 3);
-        assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));
-        assert!(matches!(
-            symbols[1],
-            ColumnOperation::Value(NumericalValue::I64(16i64))
-        ));
-        assert!(matches!(
-            symbols[2],
-            ColumnOperation::Value(NumericalValue::I64(17i64))
-        ));
-    }
-}

columnar/src/columnar/writer/serializer.rs

@@ -1,108 +0,0 @@
-use std::io;
-use std::io::Write;
-
-use common::{BinarySerializable, CountingWriter};
-use sstable::value::RangeValueWriter;
-use sstable::RangeSSTable;
-
-use crate::columnar::ColumnType;
-use crate::RowId;
-
-pub struct ColumnarSerializer<W: io::Write> {
-    wrt: CountingWriter<W>,
-    sstable_range: sstable::Writer<Vec<u8>, RangeValueWriter>,
-    prepare_key_buffer: Vec<u8>,
-}
-
-/// Returns a key consisting of the concatenation of the key and the column_type_and_cardinality
-/// code.
-fn prepare_key(key: &[u8], column_type: ColumnType, buffer: &mut Vec<u8>) {
-    buffer.clear();
-    buffer.extend_from_slice(key);
-    buffer.push(0u8);
-    buffer.push(column_type.to_code());
-}
-
-impl<W: io::Write> ColumnarSerializer<W> {
-    pub(crate) fn new(wrt: W) -> ColumnarSerializer<W> {
-        let sstable_range: sstable::Writer<Vec<u8>, RangeValueWriter> =
-            sstable::Dictionary::<RangeSSTable>::builder(Vec::with_capacity(100_000)).unwrap();
-        ColumnarSerializer {
-            wrt: CountingWriter::wrap(wrt),
-            sstable_range,
-            prepare_key_buffer: Vec::new(),
-        }
-    }
-
-    pub fn serialize_column<'a>(
-        &'a mut self,
-        column_name: &[u8],
-        column_type: ColumnType,
-    ) -> impl io::Write + 'a {
-        let start_offset = self.wrt.written_bytes();
-        prepare_key(column_name, column_type, &mut self.prepare_key_buffer);
-        ColumnSerializer {
-            columnar_serializer: self,
-            start_offset,
-        }
-    }
-
-    pub(crate) fn finalize(mut self, num_rows: RowId) -> io::Result<()> {
-        let sstable_bytes: Vec<u8> = self.sstable_range.finish()?;
-        let sstable_num_bytes: u64 = sstable_bytes.len() as u64;
-        self.wrt.write_all(&sstable_bytes)?;
-        self.wrt.write_all(&sstable_num_bytes.to_le_bytes()[..])?;
-        num_rows.serialize(&mut self.wrt)?;
-        self.wrt
-            .write_all(&super::super::format_version::footer())?;
-        self.wrt.flush()?;
-        Ok(())
-    }
-}
-
-struct ColumnSerializer<'a, W: io::Write> {
-    columnar_serializer: &'a mut ColumnarSerializer<W>,
-    start_offset: u64,
-}
-
-impl<'a, W: io::Write> Drop for ColumnSerializer<'a, W> {
-    fn drop(&mut self) {
-        let end_offset: u64 = self.columnar_serializer.wrt.written_bytes();
-        let byte_range = self.start_offset..end_offset;
-        self.columnar_serializer.sstable_range.insert_cannot_fail(
-            &self.columnar_serializer.prepare_key_buffer[..],
-            &byte_range,
-        );
-        self.columnar_serializer.prepare_key_buffer.clear();
-    }
-}
-
-impl<'a, W: io::Write> io::Write for ColumnSerializer<'a, W> {
-    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
-        self.columnar_serializer.wrt.write(buf)
-    }
-
-    fn flush(&mut self) -> io::Result<()> {
-        self.columnar_serializer.wrt.flush()
-    }
-
-    fn write_all(&mut self, buf: &[u8]) -> io::Result<()> {
-        self.columnar_serializer.wrt.write_all(buf)
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::columnar::column_type::ColumnType;
-
-    #[test]
-    fn test_prepare_key_bytes() {
-        let mut buffer: Vec<u8> = b"somegarbage".to_vec();
-        prepare_key(b"root\0child", ColumnType::Str, &mut buffer);
-        assert_eq!(buffer.len(), 12);
-        assert_eq!(&buffer[..10], b"root\0child");
-        assert_eq!(buffer[10], 0u8);
-        assert_eq!(buffer[11], ColumnType::Str.to_code());
-    }
-}

columnar/src/columnar/writer/value_index.rs

@@ -1,165 +0,0 @@
-use crate::iterable::Iterable;
-use crate::RowId;
-
-/// The `IndexBuilder` interprets a sequence of
-/// calls of the form:
-/// (record_doc,record_value+)*
-/// and can then serialize the results into an index to associate docids with their value[s].
-///
-/// It has different implementation depending on whether the
-/// cardinality is required, optional, or multivalued.
-pub(crate) trait IndexBuilder {
-    fn record_row(&mut self, doc: RowId);
-    #[inline]
-    fn record_value(&mut self) {}
-}
-
-/// The FullIndexBuilder does nothing.
-#[derive(Default)]
-pub struct FullIndexBuilder;
-
-impl IndexBuilder for FullIndexBuilder {
-    #[inline(always)]
-    fn record_row(&mut self, _doc: RowId) {}
-}
-
-#[derive(Default)]
-pub struct OptionalIndexBuilder {
-    docs: Vec<RowId>,
-}
-
-impl OptionalIndexBuilder {
-    pub fn finish(&mut self, num_rows: RowId) -> impl Iterable<RowId> + '_ {
-        debug_assert!(self
-            .docs
-            .last()
-            .copied()
-            .map(|last_doc| last_doc < num_rows)
-            .unwrap_or(true));
-        &self.docs[..]
-    }
-
-    fn reset(&mut self) {
-        self.docs.clear();
-    }
-}
-
-impl IndexBuilder for OptionalIndexBuilder {
-    #[inline(always)]
-    fn record_row(&mut self, doc: RowId) {
-        debug_assert!(self
-            .docs
-            .last()
-            .copied()
-            .map(|prev_doc| doc > prev_doc)
-            .unwrap_or(true));
-        self.docs.push(doc);
-    }
-}
-
-#[derive(Default)]
-pub struct MultivaluedIndexBuilder {
-    start_offsets: Vec<RowId>,
-    total_num_vals_seen: u32,
-}
-
-impl MultivaluedIndexBuilder {
-    pub fn finish(&mut self, num_docs: RowId) -> &[u32] {
-        self.start_offsets
-            .resize(num_docs as usize + 1, self.total_num_vals_seen);
-        &self.start_offsets[..]
-    }
-
-    fn reset(&mut self) {
-        self.start_offsets.clear();
-        self.start_offsets.push(0u32);
-        self.total_num_vals_seen = 0;
-    }
-}
-
-impl IndexBuilder for MultivaluedIndexBuilder {
-    fn record_row(&mut self, row_id: RowId) {
-        self.start_offsets
-            .resize(row_id as usize + 1, self.total_num_vals_seen);
-    }
-
-    fn record_value(&mut self) {
-        self.total_num_vals_seen += 1;
-    }
-}
-
-/// The `SpareIndexBuilders` is there to avoid allocating a
-/// new index builder for every single column.
-#[derive(Default)]
-pub struct PreallocatedIndexBuilders {
-    required_index_builder: FullIndexBuilder,
-    optional_index_builder: OptionalIndexBuilder,
-    multivalued_index_builder: MultivaluedIndexBuilder,
-}
-
-impl PreallocatedIndexBuilders {
-    pub fn borrow_required_index_builder(&mut self) -> &mut FullIndexBuilder {
-        &mut self.required_index_builder
-    }
-
-    pub fn borrow_optional_index_builder(&mut self) -> &mut OptionalIndexBuilder {
-        self.optional_index_builder.reset();
-        &mut self.optional_index_builder
-    }
-
-    pub fn borrow_multivalued_index_builder(&mut self) -> &mut MultivaluedIndexBuilder {
-        self.multivalued_index_builder.reset();
-        &mut self.multivalued_index_builder
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_optional_value_index_builder() {
-        let mut opt_value_index_builder = OptionalIndexBuilder::default();
-        opt_value_index_builder.record_row(0u32);
-        opt_value_index_builder.record_value();
-        assert_eq!(
-            &opt_value_index_builder
-                .finish(1u32)
-                .boxed_iter()
-                .collect::<Vec<u32>>(),
-            &[0]
-        );
-        opt_value_index_builder.reset();
-        opt_value_index_builder.record_row(1u32);
-        opt_value_index_builder.record_value();
-        assert_eq!(
-            &opt_value_index_builder
-                .finish(2u32)
-                .boxed_iter()
-                .collect::<Vec<u32>>(),
-            &[1]
-        );
-    }
-
-    #[test]
-    fn test_multivalued_value_index_builder() {
-        let mut multivalued_value_index_builder = MultivaluedIndexBuilder::default();
-        multivalued_value_index_builder.record_row(1u32);
-        multivalued_value_index_builder.record_value();
-        multivalued_value_index_builder.record_value();
-        multivalued_value_index_builder.record_row(2u32);
-        multivalued_value_index_builder.record_value();
-        assert_eq!(
-            multivalued_value_index_builder.finish(4u32).to_vec(),
-            vec![0, 0, 2, 3, 3]
-        );
-        multivalued_value_index_builder.reset();
-        multivalued_value_index_builder.record_row(2u32);
-        multivalued_value_index_builder.record_value();
-        multivalued_value_index_builder.record_value();
-        assert_eq!(
-            multivalued_value_index_builder.finish(4u32).to_vec(),
-            vec![0, 0, 0, 2, 2]
-        );
-    }
-}

columnar/src/dictionary.rs

@@ -1,84 +0,0 @@
-use std::io;
-
-use fnv::FnvHashMap;
-use sstable::SSTable;
-
-pub(crate) struct TermIdMapping {
-    unordered_to_ord: Vec<OrderedId>,
-}
-
-impl TermIdMapping {
-    pub fn to_ord(&self, unordered: UnorderedId) -> OrderedId {
-        self.unordered_to_ord[unordered.0 as usize]
-    }
-}
-
-/// When we add values, we cannot know their ordered id yet.
-/// For this reason, we temporarily assign them a `UnorderedId`
-/// that will be mapped to an `OrderedId` upon serialization.
-#[derive(Clone, Copy, Debug, Hash, PartialEq, Eq)]
-pub struct UnorderedId(pub u32);
-
-#[derive(Clone, Copy, Hash, PartialEq, Eq, Debug)]
-pub struct OrderedId(pub u32);
-
-/// `DictionaryBuilder` for dictionary encoding.
-///
-/// It stores the different terms encounterred and assigns them a temporary value
-/// we call unordered id.
-///
-/// Upon serialization, we will sort the ids and hence build a `UnorderedId -> Term ordinal`
-/// mapping.
-#[derive(Default)]
-pub(crate) struct DictionaryBuilder {
-    dict: FnvHashMap<Vec<u8>, UnorderedId>,
-}
-
-impl DictionaryBuilder {
-    /// Get or allocate an unordered id.
-    /// (This ID is simply an auto-incremented id.)
-    pub fn get_or_allocate_id(&mut self, term: &[u8]) -> UnorderedId {
-        if let Some(term_id) = self.dict.get(term) {
-            return *term_id;
-        }
-        let new_id = UnorderedId(self.dict.len() as u32);
-        self.dict.insert(term.to_vec(), new_id);
-        new_id
-    }
-
-    /// Serialize the dictionary into an fst, and returns the
-    /// `UnorderedId -> TermOrdinal` map.
-    pub fn serialize<'a, W: io::Write + 'a>(&self, wrt: &mut W) -> io::Result<TermIdMapping> {
-        let mut terms: Vec<(&[u8], UnorderedId)> =
-            self.dict.iter().map(|(k, v)| (k.as_slice(), *v)).collect();
-        terms.sort_unstable_by_key(|(key, _)| *key);
-        // TODO Remove the allocation.
-        let mut unordered_to_ord: Vec<OrderedId> = vec![OrderedId(0u32); terms.len()];
-        let mut sstable_builder = sstable::VoidSSTable::writer(wrt);
-        for (ord, (key, unordered_id)) in terms.into_iter().enumerate() {
-            let ordered_id = OrderedId(ord as u32);
-            sstable_builder.insert(key, &())?;
-            unordered_to_ord[unordered_id.0 as usize] = ordered_id;
-        }
-        sstable_builder.finish()?;
-        Ok(TermIdMapping { unordered_to_ord })
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_dictionary_builder() {
-        let mut dictionary_builder = DictionaryBuilder::default();
-        let hello_uid = dictionary_builder.get_or_allocate_id(b"hello");
-        let happy_uid = dictionary_builder.get_or_allocate_id(b"happy");
-        let tax_uid = dictionary_builder.get_or_allocate_id(b"tax");
-        let mut buffer = Vec::new();
-        let id_mapping = dictionary_builder.serialize(&mut buffer).unwrap();
-        assert_eq!(id_mapping.to_ord(hello_uid), OrderedId(1));
-        assert_eq!(id_mapping.to_ord(happy_uid), OrderedId(0));
-        assert_eq!(id_mapping.to_ord(tax_uid), OrderedId(2));
-    }
-}

columnar/src/dynamic_column.rs

@@ -1,258 +0,0 @@
-use std::io;
-use std::net::Ipv6Addr;
-use std::sync::Arc;
-
-use common::file_slice::FileSlice;
-use common::{DateTime, HasLen, OwnedBytes};
-
-use crate::column::{BytesColumn, Column, StrColumn};
-use crate::column_values::{monotonic_map_column, StrictlyMonotonicFn};
-use crate::columnar::ColumnType;
-use crate::{Cardinality, NumericalType};
-
-#[derive(Clone)]
-pub enum DynamicColumn {
-    Bool(Column<bool>),
-    I64(Column<i64>),
-    U64(Column<u64>),
-    F64(Column<f64>),
-    IpAddr(Column<Ipv6Addr>),
-    DateTime(Column<DateTime>),
-    Bytes(BytesColumn),
-    Str(StrColumn),
-}
-
-impl DynamicColumn {
-    pub fn get_cardinality(&self) -> Cardinality {
-        match self {
-            DynamicColumn::Bool(c) => c.get_cardinality(),
-            DynamicColumn::I64(c) => c.get_cardinality(),
-            DynamicColumn::U64(c) => c.get_cardinality(),
-            DynamicColumn::F64(c) => c.get_cardinality(),
-            DynamicColumn::IpAddr(c) => c.get_cardinality(),
-            DynamicColumn::DateTime(c) => c.get_cardinality(),
-            DynamicColumn::Bytes(c) => c.ords().get_cardinality(),
-            DynamicColumn::Str(c) => c.ords().get_cardinality(),
-        }
-    }
-    pub fn column_type(&self) -> ColumnType {
-        match self {
-            DynamicColumn::Bool(_) => ColumnType::Bool,
-            DynamicColumn::I64(_) => ColumnType::I64,
-            DynamicColumn::U64(_) => ColumnType::U64,
-            DynamicColumn::F64(_) => ColumnType::F64,
-            DynamicColumn::IpAddr(_) => ColumnType::IpAddr,
-            DynamicColumn::DateTime(_) => ColumnType::DateTime,
-            DynamicColumn::Bytes(_) => ColumnType::Bytes,
-            DynamicColumn::Str(_) => ColumnType::Str,
-        }
-    }
-
-    pub fn coerce_numerical(self, target_numerical_type: NumericalType) -> Option<Self> {
-        match target_numerical_type {
-            NumericalType::I64 => self.coerce_to_i64(),
-            NumericalType::U64 => self.coerce_to_u64(),
-            NumericalType::F64 => self.coerce_to_f64(),
-        }
-    }
-
-    pub fn is_numerical(&self) -> bool {
-        self.column_type().numerical_type().is_some()
-    }
-
-    pub fn is_f64(&self) -> bool {
-        self.column_type().numerical_type() == Some(NumericalType::F64)
-    }
-    pub fn is_i64(&self) -> bool {
-        self.column_type().numerical_type() == Some(NumericalType::I64)
-    }
-    pub fn is_u64(&self) -> bool {
-        self.column_type().numerical_type() == Some(NumericalType::U64)
-    }
-
-    fn coerce_to_f64(self) -> Option<DynamicColumn> {
-        match self {
-            DynamicColumn::I64(column) => Some(DynamicColumn::F64(Column {
-                idx: column.idx,
-                values: Arc::new(monotonic_map_column(column.values, MapI64ToF64)),
-            })),
-            DynamicColumn::U64(column) => Some(DynamicColumn::F64(Column {
-                idx: column.idx,
-                values: Arc::new(monotonic_map_column(column.values, MapU64ToF64)),
-            })),
-            DynamicColumn::F64(_) => Some(self),
-            _ => None,
-        }
-    }
-    fn coerce_to_i64(self) -> Option<DynamicColumn> {
-        match self {
-            DynamicColumn::U64(column) => {
-                if column.max_value() > i64::MAX as u64 {
-                    return None;
-                }
-                Some(DynamicColumn::I64(Column {
-                    idx: column.idx,
-                    values: Arc::new(monotonic_map_column(column.values, MapU64ToI64)),
-                }))
-            }
-            DynamicColumn::I64(_) => Some(self),
-            _ => None,
-        }
-    }
-    fn coerce_to_u64(self) -> Option<DynamicColumn> {
-        match self {
-            DynamicColumn::I64(column) => {
-                if column.min_value() < 0 {
-                    return None;
-                }
-                Some(DynamicColumn::U64(Column {
-                    idx: column.idx,
-                    values: Arc::new(monotonic_map_column(column.values, MapI64ToU64)),
-                }))
-            }
-            DynamicColumn::U64(_) => Some(self),
-            _ => None,
-        }
-    }
-}
-
-struct MapI64ToF64;
-impl StrictlyMonotonicFn<i64, f64> for MapI64ToF64 {
-    #[inline(always)]
-    fn mapping(&self, inp: i64) -> f64 {
-        inp as f64
-    }
-    #[inline(always)]
-    fn inverse(&self, out: f64) -> i64 {
-        out as i64
-    }
-}
-
-struct MapU64ToF64;
-impl StrictlyMonotonicFn<u64, f64> for MapU64ToF64 {
-    #[inline(always)]
-    fn mapping(&self, inp: u64) -> f64 {
-        inp as f64
-    }
-    #[inline(always)]
-    fn inverse(&self, out: f64) -> u64 {
-        out as u64
-    }
-}
-
-struct MapU64ToI64;
-impl StrictlyMonotonicFn<u64, i64> for MapU64ToI64 {
-    #[inline(always)]
-    fn mapping(&self, inp: u64) -> i64 {
-        inp as i64
-    }
-    #[inline(always)]
-    fn inverse(&self, out: i64) -> u64 {
-        out as u64
-    }
-}
-
-struct MapI64ToU64;
-impl StrictlyMonotonicFn<i64, u64> for MapI64ToU64 {
-    #[inline(always)]
-    fn mapping(&self, inp: i64) -> u64 {
-        inp as u64
-    }
-    #[inline(always)]
-    fn inverse(&self, out: u64) -> i64 {
-        out as i64
-    }
-}
-
-macro_rules! static_dynamic_conversions {
-    ($typ:ty, $enum_name:ident) => {
-        impl From<DynamicColumn> for Option<$typ> {
-            fn from(dynamic_column: DynamicColumn) -> Option<$typ> {
-                if let DynamicColumn::$enum_name(col) = dynamic_column {
-                    Some(col)
-                } else {
-                    None
-                }
-            }
-        }
-
-        impl From<$typ> for DynamicColumn {
-            fn from(typed_column: $typ) -> Self {
-                DynamicColumn::$enum_name(typed_column)
-            }
-        }
-    };
-}
-
-static_dynamic_conversions!(Column<bool>, Bool);
-static_dynamic_conversions!(Column<u64>, U64);
-static_dynamic_conversions!(Column<i64>, I64);
-static_dynamic_conversions!(Column<f64>, F64);
-static_dynamic_conversions!(Column<DateTime>, DateTime);
-static_dynamic_conversions!(StrColumn, Str);
-static_dynamic_conversions!(BytesColumn, Bytes);
-static_dynamic_conversions!(Column<Ipv6Addr>, IpAddr);
-
-#[derive(Clone)]
-pub struct DynamicColumnHandle {
-    pub(crate) file_slice: FileSlice,
-    pub(crate) column_type: ColumnType,
-}
-
-impl DynamicColumnHandle {
-    // TODO rename load
-    pub fn open(&self) -> io::Result<DynamicColumn> {
-        let column_bytes: OwnedBytes = self.file_slice.read_bytes()?;
-        self.open_internal(column_bytes)
-    }
-
-    #[doc(hidden)]
-    pub fn file_slice(&self) -> &FileSlice {
-        &self.file_slice
-    }
-
-    /// Returns the `u64` fast field reader reader associated with `fields` of types
-    /// Str, u64, i64, f64, or datetime.
-    ///
-    /// If not, the fastfield reader will returns the u64-value associated with the original
-    /// FastValue.
-    pub fn open_u64_lenient(&self) -> io::Result<Option<Column<u64>>> {
-        let column_bytes = self.file_slice.read_bytes()?;
-        match self.column_type {
-            ColumnType::Str | ColumnType::Bytes => {
-                let column: BytesColumn = crate::column::open_column_bytes(column_bytes)?;
-                Ok(Some(column.term_ord_column))
-            }
-            ColumnType::Bool => Ok(None),
-            ColumnType::IpAddr => Ok(None),
-            ColumnType::I64 | ColumnType::U64 | ColumnType::F64 | ColumnType::DateTime => {
-                let column = crate::column::open_column_u64::<u64>(column_bytes)?;
-                Ok(Some(column))
-            }
-        }
-    }
-
-    fn open_internal(&self, column_bytes: OwnedBytes) -> io::Result<DynamicColumn> {
-        let dynamic_column: DynamicColumn = match self.column_type {
-            ColumnType::Bytes => crate::column::open_column_bytes(column_bytes)?.into(),
-            ColumnType::Str => crate::column::open_column_str(column_bytes)?.into(),
-            ColumnType::I64 => crate::column::open_column_u64::<i64>(column_bytes)?.into(),
-            ColumnType::U64 => crate::column::open_column_u64::<u64>(column_bytes)?.into(),
-            ColumnType::F64 => crate::column::open_column_u64::<f64>(column_bytes)?.into(),
-            ColumnType::Bool => crate::column::open_column_u64::<bool>(column_bytes)?.into(),
-            ColumnType::IpAddr => crate::column::open_column_u128::<Ipv6Addr>(column_bytes)?.into(),
-            ColumnType::DateTime => {
-                crate::column::open_column_u64::<DateTime>(column_bytes)?.into()
-            }
-        };
-        Ok(dynamic_column)
-    }
-
-    pub fn num_bytes(&self) -> usize {
-        self.file_slice.len()
-    }
-
-    pub fn column_type(&self) -> ColumnType {
-        self.column_type
-    }
-}

columnar/src/iterable.rs

@@ -1,19 +0,0 @@
-use std::ops::Range;
-
-pub trait Iterable<T = u64> {
-    fn boxed_iter(&self) -> Box<dyn Iterator<Item = T> + '_>;
-}
-
-impl<'a, T: Copy> Iterable<T> for &'a [T] {
-    fn boxed_iter(&self) -> Box<dyn Iterator<Item = T> + '_> {
-        Box::new(self.iter().copied())
-    }
-}
-
-impl<T: Copy> Iterable<T> for Range<T>
-where Range<T>: Iterator<Item = T>
-{
-    fn boxed_iter(&self) -> Box<dyn Iterator<Item = T> + '_> {
-        Box::new(self.clone())
-    }
-}

columnar/src/lib.rs

@@ -1,96 +0,0 @@
-#![cfg_attr(all(feature = "unstable", test), feature(test))]
-
-#[cfg(test)]
-#[macro_use]
-extern crate more_asserts;
-
-#[cfg(all(test, feature = "unstable"))]
-extern crate test;
-
-use std::io;
-
-mod column;
-mod column_index;
-pub mod column_values;
-mod columnar;
-mod dictionary;
-mod dynamic_column;
-mod iterable;
-pub(crate) mod utils;
-mod value;
-
-pub use column::{BytesColumn, Column, StrColumn};
-pub use column_index::ColumnIndex;
-pub use column_values::{ColumnValues, MonotonicallyMappableToU128, MonotonicallyMappableToU64};
-pub use columnar::{
-    merge_columnar, ColumnType, ColumnarReader, ColumnarWriter, HasAssociatedColumnType,
-    MergeRowOrder, ShuffleMergeOrder, StackMergeOrder,
-};
-use sstable::VoidSSTable;
-pub use value::{NumericalType, NumericalValue};
-
-pub use self::dynamic_column::{DynamicColumn, DynamicColumnHandle};
-
-pub type RowId = u32;
-pub type DocId = u32;
-
-#[derive(Clone, Copy)]
-pub struct RowAddr {
-    pub segment_ord: u32,
-    pub row_id: RowId,
-}
-
-pub use sstable::Dictionary;
-pub type Streamer<'a> = sstable::Streamer<'a, VoidSSTable>;
-
-pub use common::DateTime;
-
-#[derive(Copy, Clone, Debug)]
-pub struct InvalidData;
-
-impl From<InvalidData> for io::Error {
-    fn from(_: InvalidData) -> Self {
-        io::Error::new(io::ErrorKind::InvalidData, "Invalid data")
-    }
-}
-
-/// Enum describing the number of values that can exist per document
-/// (or per row if you will).
-///
-/// The cardinality must fit on 2 bits.
-#[derive(Clone, Copy, Hash, Default, Debug, PartialEq, Eq, PartialOrd, Ord)]
-#[repr(u8)]
-pub enum Cardinality {
-    /// All documents contain exactly one value.
-    /// `Full` is the default for auto-detecting the Cardinality, since it is the most strict.
-    #[default]
-    Full = 0,
-    /// All documents contain at most one value.
-    Optional = 1,
-    /// All documents may contain any number of values.
-    Multivalued = 2,
-}
-
-impl Cardinality {
-    pub fn is_optional(&self) -> bool {
-        matches!(self, Cardinality::Optional)
-    }
-    pub fn is_multivalue(&self) -> bool {
-        matches!(self, Cardinality::Multivalued)
-    }
-    pub(crate) fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    pub(crate) fn try_from_code(code: u8) -> Result<Cardinality, InvalidData> {
-        match code {
-            0 => Ok(Cardinality::Full),
-            1 => Ok(Cardinality::Optional),
-            2 => Ok(Cardinality::Multivalued),
-            _ => Err(InvalidData),
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests;

columnar/src/tests.rs

@@ -1,212 +0,0 @@
-use std::net::Ipv6Addr;
-
-use crate::column_values::MonotonicallyMappableToU128;
-use crate::columnar::ColumnType;
-use crate::dynamic_column::{DynamicColumn, DynamicColumnHandle};
-use crate::value::NumericalValue;
-use crate::{Cardinality, ColumnarReader, ColumnarWriter};
-
-#[test]
-fn test_dataframe_writer_str() {
-    let mut dataframe_writer = ColumnarWriter::default();
-    dataframe_writer.record_str(1u32, "my_string", "hello");
-    dataframe_writer.record_str(3u32, "my_string", "helloeee");
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer.serialize(5, None, &mut buffer).unwrap();
-    let columnar = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar.num_columns(), 1);
-    let cols: Vec<DynamicColumnHandle> = columnar.read_columns("my_string").unwrap();
-    assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 158);
-}
-
-#[test]
-fn test_dataframe_writer_bytes() {
-    let mut dataframe_writer = ColumnarWriter::default();
-    dataframe_writer.record_bytes(1u32, "my_string", b"hello");
-    dataframe_writer.record_bytes(3u32, "my_string", b"helloeee");
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer.serialize(5, None, &mut buffer).unwrap();
-    let columnar = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar.num_columns(), 1);
-    let cols: Vec<DynamicColumnHandle> = columnar.read_columns("my_string").unwrap();
-    assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 158);
-}
-
-#[test]
-fn test_dataframe_writer_bool() {
-    let mut dataframe_writer = ColumnarWriter::default();
-    dataframe_writer.record_bool(1u32, "bool.value", false);
-    dataframe_writer.record_bool(3u32, "bool.value", true);
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer.serialize(5, None, &mut buffer).unwrap();
-    let columnar = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar.num_columns(), 1);
-    let cols: Vec<DynamicColumnHandle> = columnar.read_columns("bool.value").unwrap();
-    assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 22);
-    assert_eq!(cols[0].column_type(), ColumnType::Bool);
-    let dyn_bool_col = cols[0].open().unwrap();
-    let DynamicColumn::Bool(bool_col) = dyn_bool_col else { panic!(); };
-    let vals: Vec<Option<bool>> = (0..5).map(|row_id| bool_col.first(row_id)).collect();
-    assert_eq!(&vals, &[None, Some(false), None, Some(true), None,]);
-}
-
-#[test]
-fn test_dataframe_writer_u64_multivalued() {
-    let mut dataframe_writer = ColumnarWriter::default();
-    dataframe_writer.record_numerical(2u32, "divisor", 2u64);
-    dataframe_writer.record_numerical(3u32, "divisor", 3u64);
-    dataframe_writer.record_numerical(4u32, "divisor", 2u64);
-    dataframe_writer.record_numerical(5u32, "divisor", 5u64);
-    dataframe_writer.record_numerical(6u32, "divisor", 2u64);
-    dataframe_writer.record_numerical(6u32, "divisor", 3u64);
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer.serialize(7, None, &mut buffer).unwrap();
-    let columnar = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar.num_columns(), 1);
-    let cols: Vec<DynamicColumnHandle> = columnar.read_columns("divisor").unwrap();
-    assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 29);
-    let dyn_i64_col = cols[0].open().unwrap();
-    let DynamicColumn::I64(divisor_col) = dyn_i64_col else { panic!(); };
-    assert_eq!(
-        divisor_col.get_cardinality(),
-        crate::Cardinality::Multivalued
-    );
-    assert_eq!(divisor_col.num_docs(), 7);
-}
-
-#[test]
-fn test_dataframe_writer_ip_addr() {
-    let mut dataframe_writer = ColumnarWriter::default();
-    dataframe_writer.record_ip_addr(1, "ip_addr", Ipv6Addr::from_u128(1001));
-    dataframe_writer.record_ip_addr(3, "ip_addr", Ipv6Addr::from_u128(1050));
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer.serialize(5, None, &mut buffer).unwrap();
-    let columnar = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar.num_columns(), 1);
-    let cols: Vec<DynamicColumnHandle> = columnar.read_columns("ip_addr").unwrap();
-    assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 42);
-    assert_eq!(cols[0].column_type(), ColumnType::IpAddr);
-    let dyn_bool_col = cols[0].open().unwrap();
-    let DynamicColumn::IpAddr(ip_col) = dyn_bool_col else { panic!(); };
-    let vals: Vec<Option<Ipv6Addr>> = (0..5).map(|row_id| ip_col.first(row_id)).collect();
-    assert_eq!(
-        &vals,
-        &[
-            None,
-            Some(Ipv6Addr::from_u128(1001)),
-            None,
-            Some(Ipv6Addr::from_u128(1050)),
-            None,
-        ]
-    );
-}
-
-#[test]
-fn test_dataframe_writer_numerical() {
-    let mut dataframe_writer = ColumnarWriter::default();
-    dataframe_writer.record_numerical(1u32, "srical.value", NumericalValue::U64(12u64));
-    dataframe_writer.record_numerical(2u32, "srical.value", NumericalValue::U64(13u64));
-    dataframe_writer.record_numerical(4u32, "srical.value", NumericalValue::U64(15u64));
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer.serialize(6, None, &mut buffer).unwrap();
-    let columnar = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar.num_columns(), 1);
-    let cols: Vec<DynamicColumnHandle> = columnar.read_columns("srical.value").unwrap();
-    assert_eq!(cols.len(), 1);
-    // Right now this 31 bytes are spent as follows
-    //
-    // - header 14 bytes
-    // - vals  8 //< due to padding? could have been 1byte?.
-    // - null footer 6 bytes
-    assert_eq!(cols[0].num_bytes(), 33);
-    let column = cols[0].open().unwrap();
-    let DynamicColumn::I64(column_i64) = column else { panic!(); };
-    assert_eq!(column_i64.idx.get_cardinality(), Cardinality::Optional);
-    assert_eq!(column_i64.first(0), None);
-    assert_eq!(column_i64.first(1), Some(12i64));
-    assert_eq!(column_i64.first(2), Some(13i64));
-    assert_eq!(column_i64.first(3), None);
-    assert_eq!(column_i64.first(4), Some(15i64));
-    assert_eq!(column_i64.first(5), None);
-    assert_eq!(column_i64.first(6), None); //< we can change the spec for that one.
-}
-
-#[test]
-fn test_dictionary_encoded_str() {
-    let mut buffer = Vec::new();
-    let mut columnar_writer = ColumnarWriter::default();
-    columnar_writer.record_str(1, "my.column", "a");
-    columnar_writer.record_str(3, "my.column", "c");
-    columnar_writer.record_str(3, "my.column2", "different_column!");
-    columnar_writer.record_str(4, "my.column", "b");
-    columnar_writer.serialize(5, None, &mut buffer).unwrap();
-    let columnar_reader = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar_reader.num_columns(), 2);
-    let col_handles = columnar_reader.read_columns("my.column").unwrap();
-    assert_eq!(col_handles.len(), 1);
-    let DynamicColumn::Str(str_col) = col_handles[0].open().unwrap() else  { panic!(); };
-    let index: Vec<Option<u64>> = (0..5).map(|row_id| str_col.ords().first(row_id)).collect();
-    assert_eq!(index, &[None, Some(0), None, Some(2), Some(1)]);
-    assert_eq!(str_col.num_rows(), 5);
-    let mut term_buffer = String::new();
-    let term_ords = str_col.ords();
-    assert_eq!(term_ords.first(0), None);
-    assert_eq!(term_ords.first(1), Some(0));
-    str_col.ord_to_str(0u64, &mut term_buffer).unwrap();
-    assert_eq!(term_buffer, "a");
-    assert_eq!(term_ords.first(2), None);
-    assert_eq!(term_ords.first(3), Some(2));
-    str_col.ord_to_str(2u64, &mut term_buffer).unwrap();
-    assert_eq!(term_buffer, "c");
-    assert_eq!(term_ords.first(4), Some(1));
-    str_col.ord_to_str(1u64, &mut term_buffer).unwrap();
-    assert_eq!(term_buffer, "b");
-}
-
-#[test]
-fn test_dictionary_encoded_bytes() {
-    let mut buffer = Vec::new();
-    let mut columnar_writer = ColumnarWriter::default();
-    columnar_writer.record_bytes(1, "my.column", b"a");
-    columnar_writer.record_bytes(3, "my.column", b"c");
-    columnar_writer.record_bytes(3, "my.column2", b"different_column!");
-    columnar_writer.record_bytes(4, "my.column", b"b");
-    columnar_writer.serialize(5, None, &mut buffer).unwrap();
-    let columnar_reader = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar_reader.num_columns(), 2);
-    let col_handles = columnar_reader.read_columns("my.column").unwrap();
-    assert_eq!(col_handles.len(), 1);
-    let DynamicColumn::Bytes(bytes_col) = col_handles[0].open().unwrap() else  { panic!(); };
-    let index: Vec<Option<u64>> = (0..5)
-        .map(|row_id| bytes_col.ords().first(row_id))
-        .collect();
-    assert_eq!(index, &[None, Some(0), None, Some(2), Some(1)]);
-    assert_eq!(bytes_col.num_rows(), 5);
-    let mut term_buffer = Vec::new();
-    let term_ords = bytes_col.ords();
-    assert_eq!(term_ords.first(0), None);
-    assert_eq!(term_ords.first(1), Some(0));
-    bytes_col
-        .dictionary
-        .ord_to_term(0u64, &mut term_buffer)
-        .unwrap();
-    assert_eq!(term_buffer, b"a");
-    assert_eq!(term_ords.first(2), None);
-    assert_eq!(term_ords.first(3), Some(2));
-    bytes_col
-        .dictionary
-        .ord_to_term(2u64, &mut term_buffer)
-        .unwrap();
-    assert_eq!(term_buffer, b"c");
-    assert_eq!(term_ords.first(4), Some(1));
-    bytes_col
-        .dictionary
-        .ord_to_term(1u64, &mut term_buffer)
-        .unwrap();
-    assert_eq!(term_buffer, b"b");
-}

columnar/src/utils.rs

@@ -1,76 +0,0 @@
-const fn compute_mask(num_bits: u8) -> u8 {
-    if num_bits == 8 {
-        u8::MAX
-    } else {
-        (1u8 << num_bits) - 1
-    }
-}
-
-#[inline(always)]
-#[must_use]
-pub(crate) fn select_bits<const START: u8, const END: u8>(code: u8) -> u8 {
-    assert!(START <= END);
-    assert!(END <= 8);
-    let num_bits: u8 = END - START;
-    let mask: u8 = compute_mask(num_bits);
-    (code >> START) & mask
-}
-
-#[inline(always)]
-#[must_use]
-pub(crate) fn place_bits<const START: u8, const END: u8>(code: u8) -> u8 {
-    assert!(START <= END);
-    assert!(END <= 8);
-    let num_bits: u8 = END - START;
-    let mask: u8 = compute_mask(num_bits);
-    assert!(code <= mask);
-    code << START
-}
-
-/// Pop-front one bytes from a slice of bytes.
-#[inline(always)]
-pub fn pop_first_byte(bytes: &mut &[u8]) -> Option<u8> {
-    if bytes.is_empty() {
-        return None;
-    }
-    let first_byte = bytes[0];
-    *bytes = &bytes[1..];
-    Some(first_byte)
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_select_bits() {
-        assert_eq!(255u8, select_bits::<0, 8>(255u8));
-        assert_eq!(0u8, select_bits::<0, 0>(255u8));
-        assert_eq!(8u8, select_bits::<0, 4>(8u8));
-        assert_eq!(4u8, select_bits::<1, 4>(8u8));
-        assert_eq!(0u8, select_bits::<1, 3>(8u8));
-    }
-
-    #[test]
-    fn test_place_bits() {
-        assert_eq!(255u8, place_bits::<0, 8>(255u8));
-        assert_eq!(4u8, place_bits::<2, 3>(1u8));
-        assert_eq!(0u8, place_bits::<2, 2>(0u8));
-    }
-
-    #[test]
-    #[should_panic]
-    fn test_place_bits_overflows() {
-        let _ = place_bits::<1, 4>(8u8);
-    }
-
-    #[test]
-    fn test_pop_first_byte() {
-        let mut cursor: &[u8] = &b"abcd"[..];
-        assert_eq!(pop_first_byte(&mut cursor), Some(b'a'));
-        assert_eq!(pop_first_byte(&mut cursor), Some(b'b'));
-        assert_eq!(pop_first_byte(&mut cursor), Some(b'c'));
-        assert_eq!(pop_first_byte(&mut cursor), Some(b'd'));
-        assert_eq!(pop_first_byte(&mut cursor), None);
-    }
-}

columnar/src/value.rs

@@ -1,131 +0,0 @@
-use common::DateTime;
-
-use crate::InvalidData;
-
-#[derive(Copy, Clone, PartialEq, Debug)]
-pub enum NumericalValue {
-    I64(i64),
-    U64(u64),
-    F64(f64),
-}
-
-impl NumericalValue {
-    pub fn numerical_type(&self) -> NumericalType {
-        match self {
-            NumericalValue::I64(_) => NumericalType::I64,
-            NumericalValue::U64(_) => NumericalType::U64,
-            NumericalValue::F64(_) => NumericalType::F64,
-        }
-    }
-}
-
-impl From<u64> for NumericalValue {
-    fn from(val: u64) -> NumericalValue {
-        NumericalValue::U64(val)
-    }
-}
-
-impl From<i64> for NumericalValue {
-    fn from(val: i64) -> Self {
-        NumericalValue::I64(val)
-    }
-}
-
-impl From<f64> for NumericalValue {
-    fn from(val: f64) -> Self {
-        NumericalValue::F64(val)
-    }
-}
-
-#[derive(Clone, Copy, Debug, Default, Hash, Eq, PartialEq)]
-#[repr(u8)]
-pub enum NumericalType {
-    #[default]
-    I64 = 0,
-    U64 = 1,
-    F64 = 2,
-}
-
-impl NumericalType {
-    pub fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    pub fn try_from_code(code: u8) -> Result<NumericalType, InvalidData> {
-        match code {
-            0 => Ok(NumericalType::I64),
-            1 => Ok(NumericalType::U64),
-            2 => Ok(NumericalType::F64),
-            _ => Err(InvalidData),
-        }
-    }
-}
-
-/// We voluntarily avoid using `Into` here to keep this
-/// implementation quirk as private as possible.
-///
-/// # Panics
-/// This coercion trait actually panics if it is used
-/// to convert a loose types to a stricter type.
-///
-/// The level is strictness is somewhat arbitrary.
-/// - i64
-/// - u64
-/// - f64.
-pub(crate) trait Coerce {
-    fn coerce(numerical_value: NumericalValue) -> Self;
-}
-
-impl Coerce for i64 {
-    fn coerce(value: NumericalValue) -> Self {
-        match value {
-            NumericalValue::I64(val) => val,
-            NumericalValue::U64(val) => val as i64,
-            NumericalValue::F64(_) => unreachable!(),
-        }
-    }
-}
-
-impl Coerce for u64 {
-    fn coerce(value: NumericalValue) -> Self {
-        match value {
-            NumericalValue::I64(val) => val as u64,
-            NumericalValue::U64(val) => val,
-            NumericalValue::F64(_) => unreachable!(),
-        }
-    }
-}
-
-impl Coerce for f64 {
-    fn coerce(value: NumericalValue) -> Self {
-        match value {
-            NumericalValue::I64(val) => val as f64,
-            NumericalValue::U64(val) => val as f64,
-            NumericalValue::F64(val) => val,
-        }
-    }
-}
-
-impl Coerce for DateTime {
-    fn coerce(value: NumericalValue) -> Self {
-        let timestamp_micros = i64::coerce(value);
-        DateTime::from_timestamp_micros(timestamp_micros)
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::NumericalType;
-
-    #[test]
-    fn test_numerical_type_code() {
-        let mut num_numerical_type = 0;
-        for code in u8::MIN..=u8::MAX {
-            if let Ok(numerical_type) = NumericalType::try_from_code(code) {
-                assert_eq!(numerical_type.to_code(), code);
-                num_numerical_type += 1;
-            }
-        }
-        assert_eq!(num_numerical_type, 3);
-    }
-}

common/Cargo.toml

@@ -1,22 +1,16 @@
 [package]
 name = "tantivy-common"
-version = "0.5.0"
+version = "0.3.0"
 authors = ["Paul Masurel <paul@quickwit.io>", "Pascal Seitz <pascal@quickwit.io>"]
 license = "MIT"
-edition = "2021"
+edition = "2018"
 description = "common traits and utility functions used by multiple tantivy subcrates"
-documentation = "https://docs.rs/tantivy_common/"
-homepage = "https://github.com/quickwit-oss/tantivy"
-repository = "https://github.com/quickwit-oss/tantivy"
-
 
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-ownedbytes = { version= "0.5", path="../ownedbytes" }
-async-trait = "0.1"
-time = { version = "0.3.10", features = ["serde-well-known"] }
-serde = { version = "1.0.136", features = ["derive"] }
+byteorder = "1.4.3"
+ownedbytes = { version="0.3", path="../ownedbytes" }
 
 [dev-dependencies]
 proptest = "1.0.0"

common/src/bitset.rs

@@ -151,7 +151,7 @@ impl TinySet {
         if self.is_empty() {
             None
         } else {
-            let lowest = self.0.trailing_zeros();
+            let lowest = self.0.trailing_zeros() as u32;
             self.0 ^= TinySet::singleton(lowest).0;
             Some(lowest)
         }
@@ -259,7 +259,11 @@ impl BitSet {
         // we do not check saturated els.
         let higher = el / 64u32;
         let lower = el % 64u32;
-        self.len += u64::from(self.tinysets[higher as usize].insert_mut(lower));
+        self.len += if self.tinysets[higher as usize].insert_mut(lower) {
+            1
+        } else {
+            0
+        };
     }
 
     /// Inserts an element in the `BitSet`
@@ -268,7 +272,11 @@ impl BitSet {
         // we do not check saturated els.
         let higher = el / 64u32;
         let lower = el % 64u32;
-        self.len -= u64::from(self.tinysets[higher as usize].remove_mut(lower));
+        self.len -= if self.tinysets[higher as usize].remove_mut(lower) {
+            1
+        } else {
+            0
+        };
     }
 
     /// Returns true iff the elements is in the `BitSet`.
@@ -277,7 +285,7 @@ impl BitSet {
         self.tinyset(el / 64u32).contains(el % 64)
     }
 
-    /// Returns the first non-empty `TinySet` associated with a bucket lower
+    /// Returns the first non-empty `TinySet` associated to a bucket lower
     /// or greater than bucket.
     ///
     /// Reminder: the tiny set with the bucket `bucket`, represents the
@@ -421,7 +429,7 @@ mod tests {
             bitset.serialize(&mut out).unwrap();
 
             let bitset = ReadOnlyBitSet::open(OwnedBytes::new(out));
-            assert_eq!(bitset.len(), i as usize);
+            assert_eq!(bitset.len() as usize, i as usize);
         }
     }
 
@@ -432,7 +440,7 @@ mod tests {
         bitset.serialize(&mut out).unwrap();
 
         let bitset = ReadOnlyBitSet::open(OwnedBytes::new(out));
-        assert_eq!(bitset.len(), 64);
+        assert_eq!(bitset.len() as usize, 64);
     }
 
     #[test]

common/src/datetime.rs

@@ -1,136 +0,0 @@
-use std::fmt;
-
-use serde::{Deserialize, Serialize};
-use time::format_description::well_known::Rfc3339;
-use time::{OffsetDateTime, PrimitiveDateTime, UtcOffset};
-
-/// DateTime Precision
-#[derive(
-    Clone, Copy, Debug, Hash, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize, Default,
-)]
-#[serde(rename_all = "lowercase")]
-pub enum DatePrecision {
-    /// Seconds precision
-    #[default]
-    Seconds,
-    /// Milli-seconds precision.
-    Milliseconds,
-    /// Micro-seconds precision.
-    Microseconds,
-}
-
-/// A date/time value with microsecond precision.
-///
-/// This timestamp does not carry any explicit time zone information.
-/// Users are responsible for applying the provided conversion
-/// functions consistently. Internally the time zone is assumed
-/// to be UTC, which is also used implicitly for JSON serialization.
-///
-/// All constructors and conversions are provided as explicit
-/// functions and not by implementing any `From`/`Into` traits
-/// to prevent unintended usage.
-#[derive(Clone, Default, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
-pub struct DateTime {
-    // Timestamp in microseconds.
-    pub(crate) timestamp_micros: i64,
-}
-
-impl DateTime {
-    /// Create new from UNIX timestamp in seconds
-    pub const fn from_timestamp_secs(seconds: i64) -> Self {
-        Self {
-            timestamp_micros: seconds * 1_000_000,
-        }
-    }
-
-    /// Create new from UNIX timestamp in milliseconds
-    pub const fn from_timestamp_millis(milliseconds: i64) -> Self {
-        Self {
-            timestamp_micros: milliseconds * 1_000,
-        }
-    }
-
-    /// Create new from UNIX timestamp in microseconds.
-    pub const fn from_timestamp_micros(microseconds: i64) -> Self {
-        Self {
-            timestamp_micros: microseconds,
-        }
-    }
-
-    /// Create new from `OffsetDateTime`
-    ///
-    /// The given date/time is converted to UTC and the actual
-    /// time zone is discarded.
-    pub const fn from_utc(dt: OffsetDateTime) -> Self {
-        let timestamp_micros = dt.unix_timestamp() * 1_000_000 + dt.microsecond() as i64;
-        Self { timestamp_micros }
-    }
-
-    /// Create new from `PrimitiveDateTime`
-    ///
-    /// Implicitly assumes that the given date/time is in UTC!
-    /// Otherwise the original value must only be reobtained with
-    /// [`Self::into_primitive()`].
-    pub fn from_primitive(dt: PrimitiveDateTime) -> Self {
-        Self::from_utc(dt.assume_utc())
-    }
-
-    /// Convert to UNIX timestamp in seconds.
-    pub const fn into_timestamp_secs(self) -> i64 {
-        self.timestamp_micros / 1_000_000
-    }
-
-    /// Convert to UNIX timestamp in milliseconds.
-    pub const fn into_timestamp_millis(self) -> i64 {
-        self.timestamp_micros / 1_000
-    }
-
-    /// Convert to UNIX timestamp in microseconds.
-    pub const fn into_timestamp_micros(self) -> i64 {
-        self.timestamp_micros
-    }
-
-    /// Convert to UTC `OffsetDateTime`
-    pub fn into_utc(self) -> OffsetDateTime {
-        let timestamp_nanos = self.timestamp_micros as i128 * 1000;
-        let utc_datetime = OffsetDateTime::from_unix_timestamp_nanos(timestamp_nanos)
-            .expect("valid UNIX timestamp");
-        debug_assert_eq!(UtcOffset::UTC, utc_datetime.offset());
-        utc_datetime
-    }
-
-    /// Convert to `OffsetDateTime` with the given time zone
-    pub fn into_offset(self, offset: UtcOffset) -> OffsetDateTime {
-        self.into_utc().to_offset(offset)
-    }
-
-    /// Convert to `PrimitiveDateTime` without any time zone
-    ///
-    /// The value should have been constructed with [`Self::from_primitive()`].
-    /// Otherwise the time zone is implicitly assumed to be UTC.
-    pub fn into_primitive(self) -> PrimitiveDateTime {
-        let utc_datetime = self.into_utc();
-        // Discard the UTC time zone offset
-        debug_assert_eq!(UtcOffset::UTC, utc_datetime.offset());
-        PrimitiveDateTime::new(utc_datetime.date(), utc_datetime.time())
-    }
-
-    /// Truncates the microseconds value to the corresponding precision.
-    pub fn truncate(self, precision: DatePrecision) -> Self {
-        let truncated_timestamp_micros = match precision {
-            DatePrecision::Seconds => (self.timestamp_micros / 1_000_000) * 1_000_000,
-            DatePrecision::Milliseconds => (self.timestamp_micros / 1_000) * 1_000,
-            DatePrecision::Microseconds => self.timestamp_micros,
-        };
-        Self {
-            timestamp_micros: truncated_timestamp_micros,
-        }
-    }
-}
-
-impl fmt::Debug for DateTime {
-    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        let utc_rfc3339 = self.into_utc().format(&Rfc3339).map_err(|_| fmt::Error)?;
-        f.write_str(&utc_rfc3339)
-    }
-}

common/src/group_by.rs

@@ -1,166 +0,0 @@
-use std::cell::RefCell;
-use std::iter::Peekable;
-use std::rc::Rc;
-
-pub trait GroupByIteratorExtended: Iterator {
-    /// Return an `Iterator` that groups iterator elements. Consecutive elements that map to the
-    /// same key are assigned to the same group.
-    ///
-    /// The returned Iterator item is `(K, impl Iterator)`, where Iterator are the items of the
-    /// group.
-    ///
-    /// ```
-    /// use tantivy_common::GroupByIteratorExtended;
-    ///
-    /// // group data into blocks of larger than zero or not.
-    /// let data: Vec<i32> = vec![1, 3, -2, -2, 1, 0, 1, 2];
-    /// // groups:               |---->|------>|--------->|
-    ///
-    /// let mut data_grouped = Vec::new();
-    /// // Note: group is an iterator
-    /// for (key, group) in data.into_iter().group_by(|val| *val >= 0) {
-    ///     data_grouped.push((key, group.collect()));
-    /// }
-    /// assert_eq!(data_grouped, vec![(true, vec![1, 3]), (false, vec![-2, -2]), (true, vec![1, 0, 1, 2])]);
-    /// ```
-    fn group_by<K, F>(self, key: F) -> GroupByIterator<Self, F, K>
-    where
-        Self: Sized,
-        F: FnMut(&Self::Item) -> K,
-        K: PartialEq + Copy,
-        Self::Item: Copy,
-    {
-        GroupByIterator::new(self, key)
-    }
-}
-impl<I: Iterator> GroupByIteratorExtended for I {}
-
-pub struct GroupByIterator<I, F, K: Copy>
-where
-    I: Iterator,
-    F: FnMut(&I::Item) -> K,
-{
-    // I really would like to avoid the Rc<RefCell>, but the Iterator is shared between
-    // `GroupByIterator` and `GroupIter`. In practice they are used consecutive and
-    // `GroupByIter` is finished before calling next on `GroupByIterator`. I'm not sure there
-    // is a solution with lifetimes for that, because we would need to enforce it in the usage
-    // somehow.
-    //
-    // One potential solution would be to replace the iterator approach with something similar.
-    inner: Rc<RefCell<GroupByShared<I, F, K>>>,
-}
-
-struct GroupByShared<I, F, K: Copy>
-where
-    I: Iterator,
-    F: FnMut(&I::Item) -> K,
-{
-    iter: Peekable<I>,
-    group_by_fn: F,
-}
-
-impl<I, F, K> GroupByIterator<I, F, K>
-where
-    I: Iterator,
-    F: FnMut(&I::Item) -> K,
-    K: Copy,
-{
-    fn new(inner: I, group_by_fn: F) -> Self {
-        let inner = GroupByShared {
-            iter: inner.peekable(),
-            group_by_fn,
-        };
-
-        Self {
-            inner: Rc::new(RefCell::new(inner)),
-        }
-    }
-}
-
-impl<I, F, K> Iterator for GroupByIterator<I, F, K>
-where
-    I: Iterator,
-    I::Item: Copy,
-    F: FnMut(&I::Item) -> K,
-    K: Copy,
-{
-    type Item = (K, GroupIterator<I, F, K>);
-
-    fn next(&mut self) -> Option<Self::Item> {
-        let mut inner = self.inner.borrow_mut();
-        let value = *inner.iter.peek()?;
-        let key = (inner.group_by_fn)(&value);
-
-        let inner = self.inner.clone();
-
-        let group_iter = GroupIterator {
-            inner,
-            group_key: key,
-        };
-        Some((key, group_iter))
-    }
-}
-
-pub struct GroupIterator<I, F, K: Copy>
-where
-    I: Iterator,
-    F: FnMut(&I::Item) -> K,
-{
-    inner: Rc<RefCell<GroupByShared<I, F, K>>>,
-    group_key: K,
-}
-
-impl<I, F, K: PartialEq + Copy> Iterator for GroupIterator<I, F, K>
-where
-    I: Iterator,
-    I::Item: Copy,
-    F: FnMut(&I::Item) -> K,
-{
-    type Item = I::Item;
-
-    fn next(&mut self) -> Option<Self::Item> {
-        let mut inner = self.inner.borrow_mut();
-        // peek if next value is in group
-        let peek_val = *inner.iter.peek()?;
-        if (inner.group_by_fn)(&peek_val) == self.group_key {
-            inner.iter.next()
-        } else {
-            None
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    fn group_by_collect<I: Iterator<Item = u32>>(iter: I) -> Vec<(I::Item, Vec<I::Item>)> {
-        iter.group_by(|val| val / 10)
-            .map(|(el, iter)| (el, iter.collect::<Vec<_>>()))
-            .collect::<Vec<_>>()
-    }
-
-    #[test]
-    fn group_by_two_groups() {
-        let vals = vec![1u32, 4, 15];
-        let grouped_vals = group_by_collect(vals.into_iter());
-        assert_eq!(grouped_vals, vec![(0, vec![1, 4]), (1, vec![15])]);
-    }
-
-    #[test]
-    fn group_by_test_empty() {
-        let vals = vec![];
-        let grouped_vals = group_by_collect(vals.into_iter());
-        assert_eq!(grouped_vals, vec![]);
-    }
-
-    #[test]
-    fn group_by_three_groups() {
-        let vals = vec![1u32, 4, 15, 1];
-        let grouped_vals = group_by_collect(vals.into_iter());
-        assert_eq!(
-            grouped_vals,
-            vec![(0, vec![1, 4]), (1, vec![15]), (0, vec![1])]
-        );
-    }
-}

common/src/lib.rs

@@ -2,22 +2,16 @@
 
 use std::ops::Deref;
 
+pub use byteorder::LittleEndian as Endianness;
+
 mod bitset;
-mod datetime;
-pub mod file_slice;
-mod group_by;
 mod serialize;
 mod vint;
 mod writer;
+
 pub use bitset::*;
-pub use datetime::{DatePrecision, DateTime};
-pub use group_by::GroupByIteratorExtended;
-pub use ownedbytes::{OwnedBytes, StableDeref};
 pub use serialize::{BinarySerializable, DeserializeFrom, FixedSize};
-pub use vint::{
-    deserialize_vint_u128, read_u32_vint, read_u32_vint_no_advance, serialize_vint_u128,
-    serialize_vint_u32, write_u32_vint, VInt, VIntU128,
-};
+pub use vint::{read_u32_vint, read_u32_vint_no_advance, serialize_vint_u32, write_u32_vint, VInt};
 pub use writer::{AntiCallToken, CountingWriter, TerminatingWrite};
 
 /// Has length trait
@@ -58,13 +52,13 @@ const HIGHEST_BIT: u64 = 1 << 63;
 /// to values over 2^63, and all values end up requiring 64 bits.
 ///
 /// # See also
-/// The reverse mapping is [`u64_to_i64()`].
+/// The [reverse mapping is `u64_to_i64`](./fn.u64_to_i64.html).
 #[inline]
 pub fn i64_to_u64(val: i64) -> u64 {
     (val as u64) ^ HIGHEST_BIT
 }
 
-/// Reverse the mapping given by [`i64_to_u64()`].
+/// Reverse the mapping given by [`i64_to_u64`](./fn.i64_to_u64.html).
 #[inline]
 pub fn u64_to_i64(val: u64) -> i64 {
     (val ^ HIGHEST_BIT) as i64
@@ -86,7 +80,7 @@ pub fn u64_to_i64(val: u64) -> i64 {
 /// explains the mapping in a clear manner.
 ///
 /// # See also
-/// The reverse mapping is [`u64_to_f64()`].
+/// The [reverse mapping is `u64_to_f64`](./fn.u64_to_f64.html).
 #[inline]
 pub fn f64_to_u64(val: f64) -> u64 {
     let bits = val.to_bits();
@@ -97,7 +91,7 @@ pub fn f64_to_u64(val: f64) -> u64 {
     }
 }
 
-/// Reverse the mapping given by [`f64_to_u64()`].
+/// Reverse the mapping given by [`i64_to_u64`](./fn.i64_to_u64.html).
 #[inline]
 pub fn u64_to_f64(val: u64) -> f64 {
     f64::from_bits(if val & HIGHEST_BIT != 0 {
@@ -107,24 +101,11 @@ pub fn u64_to_f64(val: u64) -> f64 {
     })
 }
 
-/// Replaces a given byte in the `bytes` slice of bytes.
-///
-/// This function assumes that the needle is rarely contained in the bytes string
-/// and offers a fast path if the needle is not present.
-pub fn replace_in_place(needle: u8, replacement: u8, bytes: &mut [u8]) {
-    if !bytes.contains(&needle) {
-        return;
-    }
-    for b in bytes {
-        if *b == needle {
-            *b = replacement;
-        }
-    }
-}
-
 #[cfg(test)]
 pub mod test {
 
+    use std::f64;
+
     use proptest::prelude::*;
 
     use super::{f64_to_u64, i64_to_u64, u64_to_f64, u64_to_i64, BinarySerializable, FixedSize};
@@ -154,11 +135,11 @@ pub mod test {
 
     #[test]
     fn test_i64_converter() {
-        assert_eq!(i64_to_u64(i64::MIN), u64::MIN);
-        assert_eq!(i64_to_u64(i64::MAX), u64::MAX);
+        assert_eq!(i64_to_u64(i64::min_value()), u64::min_value());
+        assert_eq!(i64_to_u64(i64::max_value()), u64::max_value());
         test_i64_converter_helper(0i64);
-        test_i64_converter_helper(i64::MIN);
-        test_i64_converter_helper(i64::MAX);
+        test_i64_converter_helper(i64::min_value());
+        test_i64_converter_helper(i64::max_value());
         for i in -1000i64..1000i64 {
             test_i64_converter_helper(i);
         }
@@ -186,20 +167,4 @@ pub mod test {
         assert!(f64_to_u64(-2.0) < f64_to_u64(1.0));
         assert!(f64_to_u64(-2.0) < f64_to_u64(-1.5));
     }
-
-    #[test]
-    fn test_replace_in_place() {
-        let test_aux = |before_replacement: &[u8], expected: &[u8]| {
-            let mut bytes: Vec<u8> = before_replacement.to_vec();
-            super::replace_in_place(b'b', b'c', &mut bytes);
-            assert_eq!(&bytes[..], expected);
-        };
-        test_aux(b"", b"");
-        test_aux(b"b", b"c");
-        test_aux(b"baaa", b"caaa");
-        test_aux(b"aaab", b"aaac");
-        test_aux(b"aaabaa", b"aaacaa");
-        test_aux(b"aaaaaa", b"aaaaaa");
-        test_aux(b"bbbb", b"cccc");
-    }
 }

common/src/serialize.rs

@@ -1,39 +1,16 @@
 use std::io::{Read, Write};
 use std::{fmt, io};
 
-use crate::VInt;
+use byteorder::{ReadBytesExt, WriteBytesExt};
 
-#[derive(Default)]
-struct Counter(u64);
-
-impl io::Write for Counter {
-    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
-        self.0 += buf.len() as u64;
-        Ok(buf.len())
-    }
-
-    fn write_all(&mut self, buf: &[u8]) -> io::Result<()> {
-        self.0 += buf.len() as u64;
-        Ok(())
-    }
-
-    fn flush(&mut self) -> io::Result<()> {
-        Ok(())
-    }
-}
+use crate::{Endianness, VInt};
 
 /// Trait for a simple binary serialization.
 pub trait BinarySerializable: fmt::Debug + Sized {
     /// Serialize
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()>;
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()>;
     /// Deserialize
     fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self>;
-
-    fn num_bytes(&self) -> u64 {
-        let mut counter = Counter::default();
-        self.serialize(&mut counter).unwrap();
-        counter.0
-    }
 }
 
 pub trait DeserializeFrom<T: BinarySerializable> {
@@ -42,7 +19,7 @@ pub trait DeserializeFrom<T: BinarySerializable> {
 
 /// Implement deserialize from &[u8] for all types which implement BinarySerializable.
 ///
-/// TryFrom would actually be preferable, but not possible because of the orphan
+/// TryFrom would actually be preferrable, but not possible because of the orphan
 /// rules (not completely sure if this could be resolved)
 impl<T: BinarySerializable> DeserializeFrom<T> for &[u8] {
     fn deserialize(&mut self) -> io::Result<T> {
@@ -57,7 +34,7 @@ pub trait FixedSize: BinarySerializable {
 }
 
 impl BinarySerializable for () {
-    fn serialize<W: Write + ?Sized>(&self, _: &mut W) -> io::Result<()> {
+    fn serialize<W: Write>(&self, _: &mut W) -> io::Result<()> {
         Ok(())
     }
     fn deserialize<R: Read>(_: &mut R) -> io::Result<Self> {
@@ -70,7 +47,7 @@ impl FixedSize for () {
 }
 
 impl<T: BinarySerializable> BinarySerializable for Vec<T> {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
         VInt(self.len() as u64).serialize(writer)?;
         for it in self {
             it.serialize(writer)?;
@@ -89,7 +66,7 @@ impl<T: BinarySerializable> BinarySerializable for Vec<T> {
 }
 
 impl<Left: BinarySerializable, Right: BinarySerializable> BinarySerializable for (Left, Right) {
-    fn serialize<W: Write + ?Sized>(&self, write: &mut W) -> io::Result<()> {
+    fn serialize<W: Write>(&self, write: &mut W) -> io::Result<()> {
         self.0.serialize(write)?;
         self.1.serialize(write)
     }
@@ -104,14 +81,12 @@ impl<Left: BinarySerializable + FixedSize, Right: BinarySerializable + FixedSize
 }
 
 impl BinarySerializable for u32 {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        writer.write_all(&self.to_le_bytes())
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
+        writer.write_u32::<Endianness>(*self)
     }
 
     fn deserialize<R: Read>(reader: &mut R) -> io::Result<u32> {
-        let mut buf = [0u8; 4];
-        reader.read_exact(&mut buf)?;
-        Ok(u32::from_le_bytes(buf))
+        reader.read_u32::<Endianness>()
     }
 }
 
@@ -119,30 +94,12 @@ impl FixedSize for u32 {
     const SIZE_IN_BYTES: usize = 4;
 }
 
-impl BinarySerializable for u16 {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        writer.write_all(&self.to_le_bytes())
-    }
-
-    fn deserialize<R: Read>(reader: &mut R) -> io::Result<u16> {
-        let mut buf = [0u8; 2];
-        reader.read_exact(&mut buf)?;
-        Ok(Self::from_le_bytes(buf))
-    }
-}
-
-impl FixedSize for u16 {
-    const SIZE_IN_BYTES: usize = 2;
-}
-
 impl BinarySerializable for u64 {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        writer.write_all(&self.to_le_bytes())
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
+        writer.write_u64::<Endianness>(*self)
     }
     fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
-        let mut buf = [0u8; 8];
-        reader.read_exact(&mut buf)?;
-        Ok(Self::from_le_bytes(buf))
+        reader.read_u64::<Endianness>()
     }
 }
 
@@ -150,29 +107,12 @@ impl FixedSize for u64 {
     const SIZE_IN_BYTES: usize = 8;
 }
 
-impl BinarySerializable for u128 {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        writer.write_all(&self.to_le_bytes())
-    }
-    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
-        let mut buf = [0u8; 16];
-        reader.read_exact(&mut buf)?;
-        Ok(Self::from_le_bytes(buf))
-    }
-}
-
-impl FixedSize for u128 {
-    const SIZE_IN_BYTES: usize = 16;
-}
-
 impl BinarySerializable for f32 {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        writer.write_all(&self.to_le_bytes())
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
+        writer.write_f32::<Endianness>(*self)
     }
     fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
-        let mut buf = [0u8; 4];
-        reader.read_exact(&mut buf)?;
-        Ok(Self::from_le_bytes(buf))
+        reader.read_f32::<Endianness>()
     }
 }
 
@@ -181,13 +121,11 @@ impl FixedSize for f32 {
 }
 
 impl BinarySerializable for i64 {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        writer.write_all(&self.to_le_bytes())
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
+        writer.write_i64::<Endianness>(*self)
     }
     fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
-        let mut buf = [0u8; Self::SIZE_IN_BYTES];
-        reader.read_exact(&mut buf)?;
-        Ok(Self::from_le_bytes(buf))
+        reader.read_i64::<Endianness>()
     }
 }
 
@@ -196,13 +134,11 @@ impl FixedSize for i64 {
 }
 
 impl BinarySerializable for f64 {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        writer.write_all(&self.to_le_bytes())
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
+        writer.write_f64::<Endianness>(*self)
     }
     fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
-        let mut buf = [0u8; Self::SIZE_IN_BYTES];
-        reader.read_exact(&mut buf)?;
-        Ok(Self::from_le_bytes(buf))
+        reader.read_f64::<Endianness>()
     }
 }
 
@@ -211,13 +147,11 @@ impl FixedSize for f64 {
 }
 
 impl BinarySerializable for u8 {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        writer.write_all(&self.to_le_bytes())
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
+        writer.write_u8(*self)
     }
-    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
-        let mut buf = [0u8; Self::SIZE_IN_BYTES];
-        reader.read_exact(&mut buf)?;
-        Ok(Self::from_le_bytes(buf))
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<u8> {
+        reader.read_u8()
     }
 }
 
@@ -226,11 +160,12 @@ impl FixedSize for u8 {
 }
 
 impl BinarySerializable for bool {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        (*self as u8).serialize(writer)
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
+        let val = if *self { 1 } else { 0 };
+        writer.write_u8(val)
     }
     fn deserialize<R: Read>(reader: &mut R) -> io::Result<bool> {
-        let val = u8::deserialize(reader)?;
+        let val = reader.read_u8()?;
         match val {
             0 => Ok(false),
             1 => Ok(true),
@@ -247,7 +182,7 @@ impl FixedSize for bool {
 }
 
 impl BinarySerializable for String {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
         let data: &[u8] = self.as_bytes();
         VInt(data.len() as u64).serialize(writer)?;
         writer.write_all(data)
@@ -294,7 +229,7 @@ pub mod test {
         fixed_size_test::<u32>();
         assert_eq!(4, serialize_test(3u32));
         assert_eq!(4, serialize_test(5u32));
-        assert_eq!(4, serialize_test(u32::MAX));
+        assert_eq!(4, serialize_test(u32::max_value()));
     }
 
     #[test]
@@ -312,11 +247,6 @@ pub mod test {
         fixed_size_test::<u64>();
     }
 
-    #[test]
-    fn test_serialize_bool() {
-        fixed_size_test::<bool>();
-    }
-
     #[test]
     fn test_serialize_string() {
         assert_eq!(serialize_test(String::from("")), 1);
@@ -342,6 +272,6 @@ pub mod test {
         assert_eq!(serialize_test(VInt(1234u64)), 2);
         assert_eq!(serialize_test(VInt(16_383u64)), 2);
         assert_eq!(serialize_test(VInt(16_384u64)), 3);
-        assert_eq!(serialize_test(VInt(u64::MAX)), 10);
+        assert_eq!(serialize_test(VInt(u64::max_value())), 10);
     }
 }

common/src/vint.rs

@@ -1,77 +1,10 @@
 use std::io;
 use std::io::{Read, Write};
 
+use byteorder::{ByteOrder, LittleEndian};
+
 use super::BinarySerializable;
 
-/// Variable int serializes a u128 number
-pub fn serialize_vint_u128(mut val: u128, output: &mut Vec<u8>) {
-    loop {
-        let next_byte: u8 = (val % 128u128) as u8;
-        val /= 128u128;
-        if val == 0 {
-            output.push(next_byte | STOP_BIT);
-            return;
-        } else {
-            output.push(next_byte);
-        }
-    }
-}
-
-/// Deserializes a u128 number
-///
-/// Returns the number and the slice after the vint
-pub fn deserialize_vint_u128(data: &[u8]) -> io::Result<(u128, &[u8])> {
-    let mut result = 0u128;
-    let mut shift = 0u64;
-    for i in 0..19 {
-        let b = data[i];
-        result |= u128::from(b % 128u8) << shift;
-        if b >= STOP_BIT {
-            return Ok((result, &data[i + 1..]));
-        }
-        shift += 7;
-    }
-    Err(io::Error::new(
-        io::ErrorKind::InvalidData,
-        "Failed to deserialize u128 vint",
-    ))
-}
-
-///   Wrapper over a `u128` that serializes as a variable int.
-#[derive(Clone, Copy, Debug, Eq, PartialEq)]
-pub struct VIntU128(pub u128);
-
-impl BinarySerializable for VIntU128 {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        let mut buffer = vec![];
-        serialize_vint_u128(self.0, &mut buffer);
-        writer.write_all(&buffer)
-    }
-
-    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
-        let mut bytes = reader.bytes();
-        let mut result = 0u128;
-        let mut shift = 0u64;
-        loop {
-            match bytes.next() {
-                Some(Ok(b)) => {
-                    result |= u128::from(b % 128u8) << shift;
-                    if b >= STOP_BIT {
-                        return Ok(VIntU128(result));
-                    }
-                    shift += 7;
-                }
-                _ => {
-                    return Err(io::Error::new(
-                        io::ErrorKind::InvalidData,
-                        "Reach end of buffer while reading VInt",
-                    ));
-                }
-            }
-        }
-    }
-}
-
 ///   Wrapper over a `u64` that serializes as a variable int.
 #[derive(Clone, Copy, Debug, Eq, PartialEq)]
 pub struct VInt(pub u64);
@@ -125,7 +58,7 @@ pub fn serialize_vint_u32(val: u32, buf: &mut [u8; 8]) -> &[u8] {
             5,
         ),
     };
-    buf.copy_from_slice(&res.to_le_bytes());
+    LittleEndian::write_u64(&mut buf[..], res);
     &buf[0..num_bytes]
 }
 
@@ -155,7 +88,7 @@ fn vint_len(data: &[u8]) -> usize {
 /// If the buffer does not start by a valid
 /// vint payload
 pub fn read_u32_vint(data: &mut &[u8]) -> u32 {
-    let (result, vlen) = read_u32_vint_no_advance(data);
+    let (result, vlen) = read_u32_vint_no_advance(*data);
     *data = &data[vlen..];
     result
 }
@@ -209,7 +142,7 @@ impl VInt {
 }
 
 impl BinarySerializable for VInt {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
         let mut buffer = [0u8; 10];
         let num_bytes = self.serialize_into(&mut buffer);
         writer.write_all(&buffer[0..num_bytes])
@@ -243,7 +176,6 @@ impl BinarySerializable for VInt {
 mod tests {
 
     use super::{serialize_vint_u32, BinarySerializable, VInt};
-    use crate::vint::{deserialize_vint_u128, serialize_vint_u128, VIntU128};
 
     fn aux_test_vint(val: u64) {
         let mut v = [14u8; 10];
@@ -267,7 +199,7 @@ mod tests {
         aux_test_vint(0);
         aux_test_vint(1);
         aux_test_vint(5);
-        aux_test_vint(u64::MAX);
+        aux_test_vint(u64::max_value());
         for i in 1..9 {
             let power_of_128 = 1u64 << (7 * i);
             aux_test_vint(power_of_128 - 1u64);
@@ -285,26 +217,6 @@ mod tests {
         assert_eq!(&buffer[..len_vint], res2, "array wrong for {}", val);
     }
 
-    fn aux_test_vint_u128(val: u128) {
-        let mut data = vec![];
-        serialize_vint_u128(val, &mut data);
-        let (deser_val, _data) = deserialize_vint_u128(&data).unwrap();
-        assert_eq!(val, deser_val);
-
-        let mut out = vec![];
-        VIntU128(val).serialize(&mut out).unwrap();
-        let deser_val = VIntU128::deserialize(&mut &out[..]).unwrap();
-        assert_eq!(val, deser_val.0);
-    }
-
-    #[test]
-    fn test_vint_u128() {
-        aux_test_vint_u128(0);
-        aux_test_vint_u128(1);
-        aux_test_vint_u128(u128::MAX / 3);
-        aux_test_vint_u128(u128::MAX);
-    }
-
     #[test]
     fn test_vint_u32() {
         aux_test_serialize_vint_u32(0);
@@ -316,6 +228,6 @@ mod tests {
             aux_test_serialize_vint_u32(power_of_128);
             aux_test_serialize_vint_u32(power_of_128 + 1u32);
         }
-        aux_test_serialize_vint_u32(u32::MAX);
+        aux_test_serialize_vint_u32(u32::max_value());
     }
 }

common/src/writer.rs

@@ -55,14 +55,14 @@ impl<W: TerminatingWrite> TerminatingWrite for CountingWriter<W> {
 }
 
 /// Struct used to prevent from calling
-/// [`terminate_ref`](TerminatingWrite::terminate_ref) directly
+/// [`terminate_ref`](trait.TerminatingWrite.html#tymethod.terminate_ref) directly
 ///
 /// The point is that while the type is public, it cannot be built by anyone
 /// outside of this module.
 pub struct AntiCallToken(());
 
 /// Trait used to indicate when no more write need to be done on a writer
-pub trait TerminatingWrite: Write + Send + Sync {
+pub trait TerminatingWrite: Write {
     /// Indicate that the writer will no longer be used. Internally call terminate_ref.
     fn terminate(mut self) -> io::Result<()>
     where Self: Sized {

doc/src/SUMMARY.md

@@ -1,5 +1,7 @@
 # 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 targeted use cases.
+they both have the same scope and targetted 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 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.
+`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.
 
 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.
 
@@ -22,6 +22,7 @@ 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.
@@ -50,9 +51,13 @@ to get tantivy to fit your use case:
 
 *Example 1* You could for instance use hadoop to build a very large search index in a timely manner, copy all of the resulting segment files in the same directory and edit the `meta.json` to get a functional index.[^2]
 
-*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 with segment `D-7`.
+*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
@@ -61,7 +66,11 @@ 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
 
@@ -15,34 +15,32 @@ 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.
+Compression ratio is mainly affected on the fast field of the sorted property, every thing else is likely unaffected. 
+###### Top-N Optimization
 
-### 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.
+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 tantvy 0.16 only fast fields are allowed to be used.
 
-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(),
@@ -60,3 +58,4 @@ 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

@@ -21,17 +21,16 @@ 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`.
@@ -42,16 +41,16 @@ 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.
@@ -71,25 +70,23 @@ 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
-
-```rust
-(my_path.my_segment, String, 233) or (my_path.my_segment, u64, 233)
-```
+`(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:
-`text:hello` could be reasonably interpreted as targeting the text field or as targeting the json field called `json_dynamic` with the json_path "text".
+`text:hello` could be reasonably interpreted as targetting the text field or as targetting the json field called `json_dynamic` with the json_path "text".
 
 If there is such an ambiguity, we decide to only search in the "text" field: `text:hello`.
 
@@ -99,11 +96,11 @@ 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.
@@ -123,8 +120,9 @@ 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/aggregation.rs

@@ -1,319 +1,129 @@
 // # Aggregation example
 //
 // This example shows how you can use built-in aggregations.
-// We will use nested aggregations with buckets and metrics:
-// - Range buckets and compute the average in each bucket.
-// - Term aggregation and compute the min price in each bucket
-// ---
+// We will use range buckets and compute the average in each bucket.
+//
 
-use serde_json::{Deserializer, Value};
+use serde_json::Value;
 use tantivy::aggregation::agg_req::{
     Aggregation, Aggregations, BucketAggregation, BucketAggregationType, MetricAggregation,
     RangeAggregation,
 };
 use tantivy::aggregation::agg_result::AggregationResults;
-use tantivy::aggregation::bucket::RangeAggregationRange;
 use tantivy::aggregation::metric::AverageAggregation;
 use tantivy::aggregation::AggregationCollector;
-use tantivy::query::AllQuery;
-use tantivy::schema::{self, IndexRecordOption, Schema, TextFieldIndexing, FAST};
-use tantivy::Index;
+use tantivy::query::TermQuery;
+use tantivy::schema::{self, Cardinality, IndexRecordOption, Schema, TextFieldIndexing};
+use tantivy::{doc, Index, Term};
 
 fn main() -> tantivy::Result<()> {
-    // # Create Schema
-    //
-    // Lets create a schema for a footwear shop, with 4 fields: name, category, stock and price.
-    // category, stock and price will be fast fields as that's the requirement
-    // for aggregation queries.
-    //
-
     let mut schema_builder = Schema::builder();
-    // In preparation of the `TermsAggregation`, the category field is configured with:
-    // - `set_fast`
-    // - `raw` tokenizer
-    //
-    // The tokenizer is set to "raw", because the fast field uses the same dictionary as the
-    // inverted index. (This behaviour will change in tantivy 0.20, where the fast field will
-    // always be raw tokenized independent from the regular tokenizing)
-    //
     let text_fieldtype = schema::TextOptions::default()
         .set_indexing_options(
-            TextFieldIndexing::default()
-                .set_index_option(IndexRecordOption::WithFreqs)
-                .set_tokenizer("raw"),
+            TextFieldIndexing::default().set_index_option(IndexRecordOption::WithFreqs),
         )
-        .set_fast()
         .set_stored();
-    schema_builder.add_text_field("category", text_fieldtype);
-    schema_builder.add_f64_field("stock", FAST);
-    schema_builder.add_f64_field("price", FAST);
+    let text_field = schema_builder.add_text_field("text", text_fieldtype);
+    let score_fieldtype =
+        crate::schema::NumericOptions::default().set_fast(Cardinality::SingleValue);
+    let highscore_field = schema_builder.add_f64_field("highscore", score_fieldtype.clone());
+    let price_field = schema_builder.add_f64_field("price", score_fieldtype.clone());
 
     let schema = schema_builder.build();
 
     // # Indexing documents
     //
     // Lets index a bunch of documents for this example.
-    let index = Index::create_in_ram(schema.clone());
-
-    let data = r#"{
-        "name": "Almond Toe Court Shoes, Patent Black",
-        "category": "Womens Footwear",
-        "price": 99.00,
-        "stock": 5
-    }
-    {
-        "name": "Suede Shoes, Blue",
-        "category": "Womens Footwear",
-        "price": 42.00,
-        "stock": 4
-    }
-    {
-        "name": "Leather Driver Saddle Loafers, Tan",
-        "category": "Mens Footwear",
-        "price": 34.00,
-        "stock": 12
-    }
-    {
-        "name": "Flip Flops, Red",
-        "category": "Mens Footwear",
-        "price": 19.00,
-        "stock": 6
-    }
-    {
-        "name": "Flip Flops, Blue",
-        "category": "Mens Footwear",
-        "price": 19.00,
-        "stock": 0
-    }
-    {
-        "name": "Gold Button Cardigan, Black",
-        "category": "Womens Casualwear",
-        "price": 167.00,
-        "stock": 6
-    }
-    {
-        "name": "Cotton Shorts, Medium Red",
-        "category": "Womens Casualwear",
-        "price": 30.00,
-        "stock": 5
-    }
-    {
-        "name": "Fine Stripe Short SleeveShirt, Grey",
-        "category": "Mens Casualwear",
-        "price": 49.99,
-        "stock": 9
-    }
-    {
-        "name": "Fine Stripe Short SleeveShirt, Green",
-        "category": "Mens Casualwear",
-        "price": 49.99,
-        "offer": 39.99,
-        "stock": 9
-    }
-    {
-        "name": "Sharkskin Waistcoat, Charcoal",
-        "category": "Mens Formalwear",
-        "price": 75.00,
-        "stock": 2
-    }
-    {
-        "name": "Lightweight Patch PocketBlazer, Deer",
-        "category": "Mens Formalwear",
-        "price": 175.50,
-        "stock": 1
-    }
-    {
-        "name": "Bird Print Dress, Black",
-        "category": "Womens Formalwear",
-        "price": 270.00,
-        "stock": 10
-    }
-    {
-        "name": "Mid Twist Cut-Out Dress, Pink",
-        "category": "Womens Formalwear",
-        "price": 540.00,
-        "stock": 5
-    }"#;
-
-    let stream = Deserializer::from_str(data).into_iter::<Value>();
+    let index = Index::create_in_ram(schema);
 
     let mut index_writer = index.writer(50_000_000)?;
-    let mut num_indexed = 0;
-    for value in stream {
-        let doc = schema.parse_document(&serde_json::to_string(&value.unwrap())?)?;
-        index_writer.add_document(doc)?;
-        num_indexed += 1;
-        if num_indexed > 4 {
-            // Writing the first segment
-            index_writer.commit()?;
-        }
-    }
+    // writing the segment
+    index_writer.add_document(doc!(
+        text_field => "cool",
+        highscore_field => 1f64,
+        price_field => 0f64,
+    ))?;
+    index_writer.add_document(doc!(
+        text_field => "cool",
+        highscore_field => 3f64,
+        price_field => 1f64,
+    ))?;
+    index_writer.add_document(doc!(
+        text_field => "cool",
+        highscore_field => 5f64,
+        price_field => 1f64,
+    ))?;
+    index_writer.add_document(doc!(
+        text_field => "nohit",
+        highscore_field => 6f64,
+        price_field => 2f64,
+    ))?;
+    index_writer.add_document(doc!(
+        text_field => "cool",
+        highscore_field => 7f64,
+        price_field => 2f64,
+    ))?;
+    index_writer.commit()?;
+    index_writer.add_document(doc!(
+        text_field => "cool",
+        highscore_field => 11f64,
+        price_field => 10f64,
+    ))?;
+    index_writer.add_document(doc!(
+        text_field => "cool",
+        highscore_field => 14f64,
+        price_field => 15f64,
+    ))?;
+
+    index_writer.add_document(doc!(
+        text_field => "cool",
+        highscore_field => 15f64,
+        price_field => 20f64,
+    ))?;
 
-    // Writing the second segment
     index_writer.commit()?;
 
-    // We have two segments now. The `AggregationCollector` will run the aggregation on each
-    // segment and then merge the results into an `IntermediateAggregationResult`.
-
     let reader = index.reader()?;
-    let searcher = reader.searcher();
-    // ---
-    // # Aggregation Query
-    //
-    //
-    // We can construct the query by building the request structure or by deserializing from JSON.
-    // The JSON API is more stable and therefore recommended.
-    //
-    // ## Request 1
+    let text_field = reader.searcher().schema().get_field("text").unwrap();
 
-    let agg_req_str = r#"
-    {
-      "group_by_stock": {
-        "aggs": {
-          "average_price": { "avg": { "field": "price" } }
-        },
-        "range": {
-          "field": "stock",
-          "ranges": [
-            { "key": "few", "to": 1.0 },
-            { "key": "some", "from": 1.0, "to": 10.0 },
-            { "key": "many", "from": 10.0 }
-          ]
-        }
-      }
-    } "#;
+    let term_query = TermQuery::new(
+        Term::from_field_text(text_field, "cool"),
+        IndexRecordOption::Basic,
+    );
 
-    // In this Aggregation we want to get the average price for different groups, depending on how
-    // many items are in stock. We define custom ranges `few`, `some`, `many` via the
-    // range aggregation.
-    // For every bucket we want the average price, so we create a nested metric aggregation on the
-    // range bucket aggregation. Only buckets support nested aggregations.
-    // ### Request JSON API
-    //
+    let sub_agg_req_1: Aggregations = vec![(
+        "average_price".to_string(),
+        Aggregation::Metric(MetricAggregation::Average(
+            AverageAggregation::from_field_name("price".to_string()),
+        )),
+    )]
+    .into_iter()
+    .collect();
 
-    let agg_req: Aggregations = serde_json::from_str(agg_req_str)?;
-    let collector = AggregationCollector::from_aggs(agg_req, None);
-
-    let agg_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
-    let res2: Value = serde_json::to_value(agg_res)?;
-
-    // ### Request Rust API
-    //
-    // This is exactly the same request as above, but via the rust structures.
-    //
-
-    let agg_req: Aggregations = vec![(
-        "group_by_stock".to_string(),
+    let agg_req_1: Aggregations = vec![(
+        "score_ranges".to_string(),
         Aggregation::Bucket(BucketAggregation {
             bucket_agg: BucketAggregationType::Range(RangeAggregation {
-                field: "stock".to_string(),
+                field: "highscore".to_string(),
                 ranges: vec![
-                    RangeAggregationRange {
-                        key: Some("few".into()),
-                        from: None,
-                        to: Some(1f64),
-                    },
-                    RangeAggregationRange {
-                        key: Some("some".into()),
-                        from: Some(1f64),
-                        to: Some(10f64),
-                    },
-                    RangeAggregationRange {
-                        key: Some("many".into()),
-                        from: Some(10f64),
-                        to: None,
-                    },
+                    (-1f64..9f64).into(),
+                    (9f64..14f64).into(),
+                    (14f64..20f64).into(),
                 ],
-                ..Default::default()
             }),
-            sub_aggregation: vec![(
-                "average_price".to_string(),
-                Aggregation::Metric(MetricAggregation::Average(
-                    AverageAggregation::from_field_name("price".to_string()),
-                )),
-            )]
-            .into_iter()
-            .collect(),
+            sub_aggregation: sub_agg_req_1.clone(),
         }),
     )]
     .into_iter()
     .collect();
 
-    let collector = AggregationCollector::from_aggs(agg_req, None);
-    // We use the `AllQuery` which will pass all documents to the AggregationCollector.
-    let agg_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
+    let collector = AggregationCollector::from_aggs(agg_req_1);
 
-    let res1: Value = serde_json::to_value(agg_res)?;
+    let searcher = reader.searcher();
+    let agg_res: AggregationResults = searcher.search(&term_query, &collector).unwrap();
 
-    // ### Aggregation Result
-    //
-    // The resulting structure deserializes in the same JSON format as elastic search.
-    //
-    let expected_res = r#"
-    {
-        "group_by_stock":{
-            "buckets":[
-                {"average_price":{"value":19.0},"doc_count":1,"key":"few","to":1.0},
-                {"average_price":{"value":124.748},"doc_count":10,"from":1.0,"key":"some","to":10.0},
-                {"average_price":{"value":152.0},"doc_count":2,"from":10.0,"key":"many"}
-            ]
-        }
-    }
-    "#;
-    let expected_json: Value = serde_json::from_str(expected_res)?;
-    assert_eq!(expected_json, res1);
-    assert_eq!(expected_json, res2);
-
-    // ### Request 2
-    //
-    // Now we are interested in the minimum price per category, so we create a bucket per
-    // category via `TermsAggregation`. We are interested in the highest minimum prices, and set the
-    // order of the buckets `"order": { "min_price": "desc" }` to be sorted by the the metric of
-    // the sub aggregation. (awesome)
-    //
-    let agg_req_str = r#"
-    {
-      "min_price_per_category": {
-        "aggs": {
-          "min_price": { "min": { "field": "price" } }
-        },
-        "terms": {
-          "field": "category",
-          "min_doc_count": 1,
-          "order": { "min_price": "desc" }
-        }
-      }
-    } "#;
-
-    let agg_req: Aggregations = serde_json::from_str(agg_req_str)?;
-
-    let collector = AggregationCollector::from_aggs(agg_req, None);
-
-    let agg_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
-    let res: Value = serde_json::to_value(agg_res)?;
-
-    // Minimum price per category, sorted by minimum price descending
-    //
-    // As you can see, the starting prices for `Formalwear` are higher than `Casualwear`.
-    //
-    let expected_res = r#"
-    {
-      "min_price_per_category": {
-        "buckets": [
-          { "doc_count": 2, "key": "Womens Formalwear", "min_price": { "value": 270.0 } },
-          { "doc_count": 2, "key": "Mens Formalwear", "min_price": { "value": 75.0 } },
-          { "doc_count": 2, "key": "Mens Casualwear", "min_price": { "value": 49.99 } },
-          { "doc_count": 2, "key": "Womens Footwear", "min_price": { "value": 42.0 } },
-          { "doc_count": 2, "key": "Womens Casualwear", "min_price": { "value": 30.0 } },
-          { "doc_count": 3, "key": "Mens Footwear", "min_price": { "value": 19.0 } }
-        ],
-        "sum_other_doc_count": 0
-      }
-    }
-    "#;
-    let expected_json: Value = serde_json::from_str(expected_res)?;
-
-    assert_eq!(expected_json, res);
+    let res: Value = serde_json::to_value(&agg_res)?;
+    println!("{}", serde_json::to_string_pretty(&res)?);
 
     Ok(())
 }

examples/custom_collector.rs

@@ -7,12 +7,12 @@
 // Of course, you can have a look at the tantivy's built-in collectors
 // such as the `CountCollector` for more examples.
 
-use columnar::Column;
 // ---
 // Importing tantivy...
 use tantivy::collector::{Collector, SegmentCollector};
+use tantivy::fastfield::{DynamicFastFieldReader, FastFieldReader};
 use tantivy::query::QueryParser;
-use tantivy::schema::{Schema, FAST, INDEXED, TEXT};
+use tantivy::schema::{Field, Schema, FAST, INDEXED, TEXT};
 use tantivy::{doc, Index, Score, SegmentReader};
 
 #[derive(Default)]
@@ -50,11 +50,11 @@ impl Stats {
 }
 
 struct StatsCollector {
-    field: String,
+    field: Field,
 }
 
 impl StatsCollector {
-    fn with_field(field: String) -> StatsCollector {
+    fn with_field(field: Field) -> StatsCollector {
         StatsCollector { field }
     }
 }
@@ -71,7 +71,7 @@ impl Collector for StatsCollector {
         _segment_local_id: u32,
         segment_reader: &SegmentReader,
     ) -> tantivy::Result<StatsSegmentCollector> {
-        let fast_field_reader = segment_reader.fast_fields().u64(&self.field)?;
+        let fast_field_reader = segment_reader.fast_fields().u64(self.field)?;
         Ok(StatsSegmentCollector {
             fast_field_reader,
             stats: Stats::default(),
@@ -95,7 +95,7 @@ impl Collector for StatsCollector {
 }
 
 struct StatsSegmentCollector {
-    fast_field_reader: Column,
+    fast_field_reader: DynamicFastFieldReader<u64>,
     stats: Stats,
 }
 
@@ -103,14 +103,10 @@ impl SegmentCollector for StatsSegmentCollector {
     type Fruit = Option<Stats>;
 
     fn collect(&mut self, doc: u32, _score: Score) {
-        // Since we know the values are single value, we could call `first_or_default_col` on the
-        // column and fetch single values.
-        for value in self.fast_field_reader.values_for_doc(doc) {
-            let value = value as f64;
-            self.stats.count += 1;
-            self.stats.sum += value;
-            self.stats.squared_sum += value * value;
-        }
+        let value = self.fast_field_reader.get(doc) as f64;
+        self.stats.count += 1;
+        self.stats.sum += value;
+        self.stats.squared_sum += value * value;
     }
 
     fn harvest(self) -> <Self as SegmentCollector>::Fruit {
@@ -171,11 +167,9 @@ fn main() -> tantivy::Result<()> {
     let searcher = reader.searcher();
     let query_parser = QueryParser::for_index(&index, vec![product_name, product_description]);
 
-    // here we want to search for `broom` and use `StatsCollector` on the hits.
+    // here we want to get a hit on the 'ken' in Frankenstein
     let query = query_parser.parse_query("broom")?;
-    if let Some(stats) =
-        searcher.search(&query, &StatsCollector::with_field("price".to_string()))?
-    {
+    if let Some(stats) = searcher.search(&query, &StatsCollector::with_field(price))? {
         println!("count: {}", stats.count());
         println!("mean: {}", stats.mean());
         println!("standard deviation: {}", stats.standard_deviation());