Closes #896 - Facet reader related

Bugfix. Acquiring a facet reader on a segment that does not contain any
doc with this facet returns `None`.

.gitignore

@@ -1,4 +1,5 @@
 tantivy.iml
+proptest-regressions
 *.swp
 target
 target/debug

CHANGELOG.md

@@ -1,3 +1,72 @@
+Tantivy 0.13.2
+===================
+Bugfix. Acquiring a facet reader on a segment that does not contain any 
+doc with this facet returns `None`. (#896)
+
+Tantivy 0.13.1
+======================
+Made `Query` and `Collector` `Send + Sync`.
+Updated misc dependency versions.
+
+
+Tantivy 0.13.0
+======================
+Tantivy 0.13 introduce a change in the index format that will require
+you to reindex your index (BlockWAND information are added in the skiplist). 
+The index size increase is minor as this information is only added for
+full blocks.
+If you have a massive index for which reindexing is not an option, please contact me
+so that we can discuss possible solutions.
+
+- Bugfix in `FuzzyTermQuery` not matching terms by prefix when it should (@Peachball)
+- Relaxed constraints on the custom/tweak score functions. At the segment level, they can be mut, and they are not required to be Sync + Send.
+- `MMapDirectory::open` does not return a `Result` anymore.
+- Change in the DocSet and Scorer API. (@fulmicoton). 
+A freshly created DocSet point directly to their first doc. A sentinel value called TERMINATED marks the end of a DocSet.
+`.advance()` returns the new DocId. `Scorer::skip(target)` has been replaced by `Scorer::seek(target)` and returns the resulting DocId.
+As a result, iterating through DocSet now looks as follows
+```rust
+let mut doc = docset.doc();
+while doc != TERMINATED {
+   // ...
+   doc = docset.advance();
+}
+```
+The change made it possible to greatly simplify a lot of the docset's code.
+- Misc internal optimization and introduction of the `Scorer::for_each_pruning` function. (@fulmicoton)
+- Added an offset option to the Top(.*)Collectors. (@robyoung)
+- Added Block WAND. Performance on TOP-K on term-unions should be greatly increased. (@fulmicoton, and special thanks 
+to the PISA team for answering all my questions!)
+
+Tantivy 0.12.0
+======================
+- Removing static dispatch in tokenizers for simplicity. (#762)
+- Added backward iteration for `TermDictionary` stream. (@halvorboe)
+- Fixed a performance issue when searching for the posting lists of a missing term (@audunhalland)
+- Added a configurable maximum number of docs (10M by default) for a segment to be considered for merge (@hntd187, landed by @halvorboe #713) 
+- Important Bugfix #777, causing tantivy to retain memory mapping. (diagnosed by @poljar)
+- Added support for field boosting. (#547, @fulmicoton)
+
+## How to update?
+
+Crates relying on custom tokenizer, or registering tokenizer in the manager will require some 
+minor changes. Check https://github.com/tantivy-search/tantivy/blob/master/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
 =====================
 
@@ -9,6 +78,7 @@ Tantivy 0.11.0
 - API change around `Box<BoxableTokenizer>`. See detail in #629
 - Avoid rebuilding Regex automaton whenever a regex query is reused. #639 (@brainlock)
 - Add footer with some metadata to index files. #605 (@fdb-hiroshima)
+- Add a method to check the compatibility of the footer in the index with the running version of tantivy (@petr-tik)
 - TopDocs collector: ensure stable sorting on equal score. #671 (@brainlock)
 - Added handling of pre-tokenized text fields (#642), which will enable users to
   load tokens created outside tantivy. See usage in examples/pre_tokenized_text. (@kkoziara)
@@ -16,10 +86,11 @@ Tantivy 0.11.0
 
 ## How to update?
 
+- The index format is changed. You are required to reindex your data to use tantivy 0.11. 
 - `Box<dyn BoxableTokenizer>` has been replaced by a `BoxedTokenizer` struct.
 - Regex are now compiled when the `RegexQuery` instance is built. As a result, it can now return
 an error and handling the `Result` is required.
-
+- `tantivy::version()` now returns a `Version` object. This object implements `ToString()`
 
 Tantivy 0.10.2
 =====================

Cargo.toml

@@ -1,11 +1,11 @@
 [package]
 name = "tantivy"
-version = "0.11.0"
+version = "0.13.2"
 authors = ["Paul Masurel <paul.masurel@gmail.com>"]
 license = "MIT"
 categories = ["database-implementations", "data-structures"]
 description = """Search engine library"""
-documentation = "https://tantivy-search.github.io/tantivy/tantivy/index.html"
+documentation = "https://docs.rs/tantivy/"
 homepage = "https://github.com/tantivy-search/tantivy"
 repository = "https://github.com/tantivy-search/tantivy"
 readme = "README.md"
@@ -13,46 +13,42 @@ keywords = ["search", "information", "retrieval"]
 edition = "2018"
 
 [dependencies]
-base64 = "0.11.0"
+base64 = "0.12"
-byteorder = "1.0"
+byteorder = "1"
-crc32fast = "1.2.0"
+crc32fast = "1"
-once_cell = "1.0"
+once_cell = "1"
-regex ={version = "1.3.0", default-features = false, features = ["std"]}
+regex ={version = "1", default-features = false, features = ["std"]}
-tantivy-fst = "0.1"
+tantivy-fst = "0.3"
 memmap = {version = "0.7", optional=true}
-lz4 = {version="1.20", optional=true}
+lz4 = {version="1", optional=true}
-snap = {version="0.2"}
+snap = "1"
-atomicwrites = {version="0.2.2", optional=true}
+atomicwrites = {version="0.2", optional=true}
-tempfile = "3.0"
+tempfile = "3"
 log = "0.4"
-serde = "1.0"
+serde = {version="1", features=["derive"]}
-serde_derive = "1.0"
+serde_json = "1"
-serde_json = "1.0"
+num_cpus = "1"
-num_cpus = "1.2"
 fs2={version="0.4", optional=true}
-itertools = "0.8"
+levenshtein_automata = "0.2"
-levenshtein_automata = {version="0.1", features=["fst_automaton"]}
 notify = {version="4", optional=true}
-bit-set = "0.5"
 uuid = { version = "0.8", features = ["v4", "serde"] }
 crossbeam = "0.7"
-futures = "0.1"
+futures = {version = "0.3",  features=["thread-pool"] }
-futures-cpupool = "0.1"
 owning_ref = "0.4"
-stable_deref_trait = "1.0.0"
+stable_deref_trait = "1"
-rust-stemmers = "1.1"
+rust-stemmers = "1"
-downcast-rs = { version="1.0" }
+downcast-rs = "1"
-tantivy-query-grammar = { path="./query-grammar" }
+tantivy-query-grammar = { version="0.13", path="./query-grammar" }
 bitpacking = {version="0.8", default-features = false, features=["bitpacker4x"]}
-census = "0.2"
+census = "0.4"
-fnv = "1.0.6"
+fnv = "1"
 owned-read = "0.4"
 failure = "0.1"
-htmlescape = "0.3.1"
+htmlescape = "0.3"
-fail = "0.3"
+fail = "0.4"
 murmurhash32 = "0.2"
 chrono = "0.4"
-smallvec = "1.0"
+smallvec = "1"
 rayon = "1"
 
 [target.'cfg(windows)'.dependencies]
@@ -62,10 +58,10 @@ winapi = "0.3"
 rand = "0.7"
 maplit = "1"
 matches = "0.1.8"
-time = "0.1.42"
+proptest = "0.10"
 
 [dev-dependencies.fail]
-version = "0.3"
+version = "0.4"
 features = ["failpoints"]
 
 [profile.release]
@@ -84,6 +80,7 @@ lz4-compression = ["lz4"]
 failpoints = ["fail/failpoints"]
 unstable = [] # useful for benches.
 wasm-bindgen = ["uuid/wasm-bindgen"]
+scoref64 = [] # scores are f64 instead of f32. was introduced to debug blockwand.
 
 [workspace]
 members = ["query-grammar"]

README.md

@@ -31,16 +31,20 @@ Tantivy is, in fact, strongly inspired by Lucene's design.
 
 # Benchmark
 
-Tantivy is typically faster than Lucene, but the results depend on 
-the nature of the queries in your workload.
-
 The following [benchmark](https://tantivy-search.github.io/bench/) break downs 
 performance for different type of queries / collection.
 
+
+In general, Tantivy tends to be 
+- slower than Lucene on union with a Top-K due to Block-WAND optimization.
+- faster than Lucene on intersection and phrase queries. 
+
+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)) and [Japanese](https://crates.io/crates/tantivy-tokenizer-tiny-segmenter))
+- Configurable tokenizer (stemming available for 17 Latin languages with third party support for Chinese ([tantivy-jieba](https://crates.io/crates/tantivy-jieba) and [cang-jie](https://crates.io/crates/cang-jie)), Japanese ([lindera](https://github.com/lindera-morphology/lindera-tantivy) and [tantivy-tokenizer-tiny-segmente](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)
@@ -59,18 +63,17 @@ performance for different type of queries / collection.
 - Configurable indexing (optional term frequency and position indexing)
 - Cheesy logo with a horse
 
-# Non-features
+## Non-features
 
 - Distributed search is out of the scope of Tantivy. That being said, Tantivy is a
 library upon which one could build a distributed search. Serializable/mergeable collector state for instance, 
 are within the scope of Tantivy.
 
-# Supported OS and compiler
-
-Tantivy works on stable Rust (>= 1.27) and supports Linux, MacOS, and Windows.
 
 # Getting started
 
+Tantivy 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/tantivy-search/tantivy-cli) - `tantivy-cli` is an actual command line interface that makes it easy for you to create a search engine,
 index documents, and search via the CLI or a small server with a REST API.

appveyor.yml

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

examples/custom_collector.rs

@@ -14,7 +14,7 @@ use tantivy::fastfield::FastFieldReader;
 use tantivy::query::QueryParser;
 use tantivy::schema::Field;
 use tantivy::schema::{Schema, FAST, INDEXED, TEXT};
-use tantivy::{doc, Index, SegmentReader, TantivyError};
+use tantivy::{doc, Index, Score, SegmentReader, TantivyError};
 
 #[derive(Default)]
 struct Stats {
@@ -114,7 +114,7 @@ struct StatsSegmentCollector {
 impl SegmentCollector for StatsSegmentCollector {
     type Fruit = Option<Stats>;
 
-    fn collect(&mut self, doc: u32, _score: f32) {
+    fn collect(&mut self, doc: u32, _score: Score) {
         let value = self.fast_field_reader.get(doc) as f64;
         self.stats.count += 1;
         self.stats.sum += value;

examples/faceted_search.rs

@@ -13,63 +13,100 @@
 // ---
 // Importing tantivy...
 use tantivy::collector::FacetCollector;
-use tantivy::query::AllQuery;
+use tantivy::query::{AllQuery, TermQuery};
 use tantivy::schema::*;
 use tantivy::{doc, Index};
-use tempfile::TempDir;
 
 fn main() -> tantivy::Result<()> {
-    // Let's create a temporary directory for the
+    // Let's create a temporary directory for the sake of this example
-    // sake of this example
-    let index_path = TempDir::new()?;
     let mut schema_builder = Schema::builder();
 
-    schema_builder.add_text_field("name", TEXT | STORED);
+    let name = schema_builder.add_text_field("felin_name", TEXT | STORED);
-
+    // this is our faceted field: its scientific classification
-    // this is our faceted field
+    let classification = schema_builder.add_facet_field("classification");
-    schema_builder.add_facet_field("tags");
 
     let schema = schema_builder.build();
+    let index = Index::create_in_ram(schema);
 
-    let index = Index::create_in_dir(&index_path, schema.clone())?;
+    let mut index_writer = index.writer(30_000_000)?;
-
-    let mut index_writer = index.writer(50_000_000)?;
-
-    let name = schema.get_field("name").unwrap();
-    let tags = schema.get_field("tags").unwrap();
 
     // For convenience, tantivy also comes with a macro to
     // reduce the boilerplate above.
     index_writer.add_document(doc!(
-        name => "the ditch",
+        name => "Cat",
-        tags => Facet::from("/pools/north")
+        classification => Facet::from("/Felidae/Felinae/Felis")
     ));
-
     index_writer.add_document(doc!(
-        name => "little stacey",
+        name => "Canada lynx",
-        tags => Facet::from("/pools/south")
+        classification => Facet::from("/Felidae/Felinae/Lynx")
+    ));
+    index_writer.add_document(doc!(
+        name => "Cheetah",
+        classification => Facet::from("/Felidae/Felinae/Acinonyx")
+    ));
+    index_writer.add_document(doc!(
+        name => "Tiger",
+        classification => Facet::from("/Felidae/Pantherinae/Panthera")
+    ));
+    index_writer.add_document(doc!(
+        name => "Lion",
+        classification => Facet::from("/Felidae/Pantherinae/Panthera")
+    ));
+    index_writer.add_document(doc!(
+        name => "Jaguar",
+        classification => Facet::from("/Felidae/Pantherinae/Panthera")
+    ));
+    index_writer.add_document(doc!(
+        name => "Sunda clouded leopard",
+        classification => Facet::from("/Felidae/Pantherinae/Neofelis")
+    ));
+    index_writer.add_document(doc!(
+        name => "Fossa",
+        classification => Facet::from("/Eupleridae/Cryptoprocta")
     ));
-
     index_writer.commit()?;
 
     let reader = index.reader()?;
-
     let searcher = reader.searcher();
+    {
+        let mut facet_collector = FacetCollector::for_field(classification);
+        facet_collector.add_facet("/Felidae");
+        let facet_counts = searcher.search(&AllQuery, &facet_collector)?;
+        // This lists all of the facet counts, right below "/Felidae".
+        let facets: Vec<(&Facet, u64)> = facet_counts.get("/Felidae").collect();
+        assert_eq!(
+            facets,
+            vec![
+                (&Facet::from("/Felidae/Felinae"), 3),
+                (&Facet::from("/Felidae/Pantherinae"), 4),
+            ]
+        );
+    }
 
-    let mut facet_collector = FacetCollector::for_field(tags);
+    // Facets are also searchable.
-    facet_collector.add_facet("/pools");
+    //
+    // For instance a common UI pattern is to allow the user someone to click on a facet link
+    // (e.g: `Pantherinae`) to drill down and filter the current result set with this subfacet.
+    //
+    // The search would then look as follows.
 
-    let facet_counts = searcher.search(&AllQuery, &facet_collector).unwrap();
+    // Check the reference doc for different ways to create a `Facet` object.
-
+    {
-    // This lists all of the facet counts
+        let facet = Facet::from_text("/Felidae/Pantherinae");
-    let facets: Vec<(&Facet, u64)> = facet_counts.get("/pools").collect();
+        let facet_term = Term::from_facet(classification, &facet);
-    assert_eq!(
+        let facet_term_query = TermQuery::new(facet_term, IndexRecordOption::Basic);
-        facets,
+        let mut facet_collector = FacetCollector::for_field(classification);
-        vec![
+        facet_collector.add_facet("/Felidae/Pantherinae");
-            (&Facet::from("/pools/north"), 1),
+        let facet_counts = searcher.search(&facet_term_query, &facet_collector)?;
-            (&Facet::from("/pools/south"), 1),
+        let facets: Vec<(&Facet, u64)> = facet_counts.get("/Felidae/Pantherinae").collect();
-        ]
+        assert_eq!(
-    );
+            facets,
+            vec![
+                (&Facet::from("/Felidae/Pantherinae/Neofelis"), 1),
+                (&Facet::from("/Felidae/Pantherinae/Panthera"), 3),
+            ]
+        );
+    }
 
     Ok(())
 }

examples/faceted_search_with_tweaked_score.rs

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

examples/iterating_docs_and_positions.rs

@@ -10,7 +10,7 @@
 // ---
 // Importing tantivy...
 use tantivy::schema::*;
-use tantivy::{doc, DocId, DocSet, Index, Postings};
+use tantivy::{doc, DocSet, Index, Postings, TERMINATED};
 
 fn main() -> tantivy::Result<()> {
     // We first create a schema for the sake of the
@@ -62,12 +62,11 @@ fn main() -> tantivy::Result<()> {
         {
             // this buffer will be used to request for positions
             let mut positions: Vec<u32> = Vec::with_capacity(100);
-            while segment_postings.advance() {
+            let mut doc_id = segment_postings.doc();
-                // the number of time the term appears in the document.
+            while doc_id != TERMINATED {
-                let doc_id: DocId = segment_postings.doc(); //< do not try to access this before calling advance once.
-
                 // This MAY contains deleted documents as well.
                 if segment_reader.is_deleted(doc_id) {
+                    doc_id = segment_postings.advance();
                     continue;
                 }
 
@@ -86,6 +85,7 @@ fn main() -> tantivy::Result<()> {
                 // Doc 2: TermFreq 1: [0]
                 // ```
                 println!("Doc {}: TermFreq {}: {:?}", doc_id, term_freq, positions);
+                doc_id = segment_postings.advance();
             }
         }
     }
@@ -117,11 +117,16 @@ fn main() -> tantivy::Result<()> {
         if let Some(mut block_segment_postings) =
             inverted_index.read_block_postings(&term_the, IndexRecordOption::Basic)
         {
-            while block_segment_postings.advance() {
+            loop {
+                let docs = block_segment_postings.docs();
+                if docs.is_empty() {
+                    break;
+                }
                 // Once again these docs MAY contains deleted documents as well.
                 let docs = block_segment_postings.docs();
                 // Prints `Docs [0, 2].`
                 println!("Docs {:?}", docs);
+                block_segment_postings.advance();
             }
         }
     }

examples/pre_tokenized_text.rs

@@ -9,11 +9,10 @@
 // - import tokenized text straight from json,
 // - perform a search on documents with pre-tokenized text
 
-use tantivy::tokenizer::{PreTokenizedString, SimpleTokenizer, Token, TokenStream, Tokenizer};
-
 use tantivy::collector::{Count, TopDocs};
 use tantivy::query::TermQuery;
 use tantivy::schema::*;
+use tantivy::tokenizer::{PreTokenizedString, SimpleTokenizer, Token, Tokenizer};
 use tantivy::{doc, Index, ReloadPolicy};
 use tempfile::TempDir;
 
@@ -65,11 +64,8 @@ fn main() -> tantivy::Result<()> {
         tokens: pre_tokenize_text(body_text),
     };
 
-    // Now lets create a document and add our `PreTokenizedString` using
+    // Now lets create a document and add our `PreTokenizedString`
-    // `add_pre_tokenized_text` method of `Document`
+    let old_man_doc = doc!(title => title_tok, body => body_tok);
-    let mut old_man_doc = Document::default();
-    old_man_doc.add_pre_tokenized_text(title, &title_tok);
-    old_man_doc.add_pre_tokenized_text(body, &body_tok);
 
     // ... now let's just add it to the IndexWriter
     index_writer.add_document(old_man_doc);
@@ -116,6 +112,9 @@ fn main() -> tantivy::Result<()> {
 
     assert_eq!(count, 2);
 
+    // Now let's print out the results.
+    // Note that the tokens are not stored along with the original text
+    // in the document store
     for (_score, doc_address) in top_docs {
         let retrieved_doc = searcher.doc(doc_address)?;
         println!("Document: {}", schema.to_json(&retrieved_doc));

examples/stop_words.rs

@@ -50,7 +50,7 @@ fn main() -> tantivy::Result<()> {
 
     // This tokenizer lowers all of the text (to help with stop word matching)
     // then removes all instances of `the` and `and` from the corpus
-    let tokenizer = SimpleTokenizer
+    let tokenizer = TextAnalyzer::from(SimpleTokenizer)
         .filter(LowerCaser)
         .filter(StopWordFilter::remove(vec![
             "the".to_string(),

query-grammar/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "tantivy-query-grammar"
-version = "0.11.0"
+version = "0.13.0"
 authors = ["Paul Masurel <paul.masurel@gmail.com>"]
 license = "MIT"
 categories = ["database-implementations", "data-structures"]
@@ -13,4 +13,4 @@ keywords = ["search", "information", "retrieval"]
 edition = "2018"
 
 [dependencies]
-combine = ">=3.6.0,<4.0.0"
+combine = {version="4", default-features=false, features=[] }

query-grammar/README.md

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

query-grammar/src/lib.rs

@@ -1,5 +1,3 @@
-#![recursion_limit = "100"]
-
 mod occur;
 mod query_grammar;
 mod user_input_ast;

query-grammar/src/occur.rs

@@ -31,22 +31,12 @@ impl Occur {
 
     /// Compose two occur values.
     pub fn compose(left: Occur, right: Occur) -> Occur {
-        match left {
+        match (left, right) {
-            Occur::Should => right,
+            (Occur::Should, _) => right,
-            Occur::Must => {
+            (Occur::Must, Occur::MustNot) => Occur::MustNot,
-                if right == Occur::MustNot {
+            (Occur::Must, _) => Occur::Must,
-                    Occur::MustNot
+            (Occur::MustNot, Occur::MustNot) => Occur::Must,
-                } else {
+            (Occur::MustNot, _) => Occur::MustNot,
-                    Occur::Must
-                }
-            }
-            Occur::MustNot => {
-                if right == Occur::MustNot {
-                    Occur::Must
-                } else {
-                    Occur::MustNot
-                }
-            }
         }
     }
 }
@@ -56,3 +46,27 @@ impl fmt::Display for Occur {
         f.write_char(self.to_char())
     }
 }
+
+#[cfg(test)]
+mod test {
+    use crate::Occur;
+
+    #[test]
+    fn test_Occur_compose() {
+        assert_eq!(Occur::compose(Occur::Should, Occur::Should), Occur::Should);
+        assert_eq!(Occur::compose(Occur::Should, Occur::Must), Occur::Must);
+        assert_eq!(
+            Occur::compose(Occur::Should, Occur::MustNot),
+            Occur::MustNot
+        );
+        assert_eq!(Occur::compose(Occur::Must, Occur::Should), Occur::Must);
+        assert_eq!(Occur::compose(Occur::Must, Occur::Must), Occur::Must);
+        assert_eq!(Occur::compose(Occur::Must, Occur::MustNot), Occur::MustNot);
+        assert_eq!(
+            Occur::compose(Occur::MustNot, Occur::Should),
+            Occur::MustNot
+        );
+        assert_eq!(Occur::compose(Occur::MustNot, Occur::Must), Occur::MustNot);
+        assert_eq!(Occur::compose(Occur::MustNot, Occur::MustNot), Occur::Must);
+    }
+}

query-grammar/src/query_grammar.rs

@@ -1,171 +1,209 @@
-use super::user_input_ast::*;
+use super::user_input_ast::{UserInputAST, UserInputBound, UserInputLeaf, UserInputLiteral};
 use crate::Occur;
-use combine::char::*;
+use combine::error::StringStreamError;
-use combine::error::StreamError;
+use combine::parser::char::{char, digit, letter, space, spaces, string};
-use combine::stream::StreamErrorFor;
+use combine::parser::Parser;
-use combine::*;
+use combine::{
+    attempt, choice, eof, many, many1, one_of, optional, parser, satisfy, skip_many1, value,
+};
 
-parser! {
+fn field<'a>() -> impl Parser<&'a str, Output = String> {
-    fn field[I]()(I) -> String
+    (
-    where [I: Stream<Item = char>] {
+        letter(),
-        (
+        many(satisfy(|c: char| c.is_alphanumeric() || c == '_')),
-            letter(),
+    )
-            many(satisfy(|c: char| c.is_alphanumeric() || c == '_')),
+        .skip(char(':'))
-        ).skip(char(':')).map(|(s1, s2): (char, String)| format!("{}{}", s1, s2))
-    }
-}
-
-parser! {
-    fn word[I]()(I) -> String
-    where [I: Stream<Item = char>] {
-        (
-            satisfy(|c: char| !c.is_whitespace() && !['-', '`', ':', '{', '}', '"', '[', ']', '(',')'].contains(&c) ),
-            many(satisfy(|c: char| !c.is_whitespace() && ![':', '{', '}', '"', '[', ']', '(',')'].contains(&c)))
-        )
         .map(|(s1, s2): (char, String)| format!("{}{}", s1, s2))
-        .and_then(|s: String|
-           match s.as_str() {
-             "OR" => Err(StreamErrorFor::<I>::unexpected_static_message("OR")),
-             "AND" => Err(StreamErrorFor::<I>::unexpected_static_message("AND")),
-             "NOT" => Err(StreamErrorFor::<I>::unexpected_static_message("NOT")),
-             _ => Ok(s)
-           })
-    }
 }
 
-parser! {
+fn word<'a>() -> impl Parser<&'a str, Output = String> {
-    fn literal[I]()(I) -> UserInputLeaf
+    (
-    where [I: Stream<Item = char>]
+        satisfy(|c: char| {
-    {
+            !c.is_whitespace()
-        let term_val = || {
+                && !['-', '^', '`', ':', '{', '}', '"', '[', ']', '(', ')'].contains(&c)
-            let phrase = char('"').with(many1(satisfy(|c| c != '"'))).skip(char('"'));
+        }),
-            phrase.or(word())
+        many(satisfy(|c: char| {
-        };
+            !c.is_whitespace() && ![':', '^', '{', '}', '"', '[', ']', '(', ')'].contains(&c)
-        let term_val_with_field = negative_number().or(term_val());
+        })),
-        let term_query =
+    )
-            (field(), term_val_with_field)
+        .map(|(s1, s2): (char, String)| format!("{}{}", s1, s2))
-            .map(|(field_name, phrase)| UserInputLiteral {
+        .and_then(|s: String| match s.as_str() {
-                field_name: Some(field_name),
+            "OR" | "AND " | "NOT" => Err(StringStreamError::UnexpectedParse),
-                phrase,
+            _ => Ok(s),
-            });
-        let term_default_field = term_val().map(|phrase| UserInputLiteral {
-            field_name: None,
-            phrase,
-        });
-        attempt(term_query)
-            .or(term_default_field)
-            .map(UserInputLeaf::from)
-    }
-}
-
-parser! {
-    fn negative_number[I]()(I) -> String
-    where [I: Stream<Item = char>]
-    {
-        (char('-'), many1(satisfy(char::is_numeric)),
-         optional((char('.'), many1(satisfy(char::is_numeric)))))
-            .map(|(s1, s2, s3): (char, String, Option<(char, String)>)| {
-                if let Some(('.', s3)) = s3 {
-                    format!("{}{}.{}", s1, s2, s3)
-                } else {
-                    format!("{}{}", s1, s2)
-                }
-            })
-    }
-}
-
-parser! {
-    fn spaces1[I]()(I) -> ()
-    where [I: Stream<Item = char>] {
-        skip_many1(space())
-    }
-}
-
-parser! {
-    /// Function that parses a range out of a Stream
-    /// Supports ranges like:
-    /// [5 TO 10], {5 TO 10}, [* TO 10], [10 TO *], {10 TO *], >5, <=10
-    /// [a TO *], [a TO c], [abc TO bcd}
-    fn range[I]()(I) -> UserInputLeaf
-    where [I: Stream<Item = char>] {
-        let range_term_val = || {
-            word().or(negative_number()).or(char('*').with(value("*".to_string())))
-        };
-
-        // check for unbounded range in the form of <5, <=10, >5, >=5
-        let elastic_unbounded_range = (choice([attempt(string(">=")),
-                                               attempt(string("<=")),
-                                               attempt(string("<")),
-                                               attempt(string(">"))])
-                                       .skip(spaces()),
-                                       range_term_val()).
-            map(|(comparison_sign, bound): (&str, String)|
-                match comparison_sign {
-                    ">=" => (UserInputBound::Inclusive(bound), UserInputBound::Unbounded),
-                    "<=" => (UserInputBound::Unbounded, UserInputBound::Inclusive(bound)),
-                    "<" => (UserInputBound::Unbounded, UserInputBound::Exclusive(bound)),
-                    ">" => (UserInputBound::Exclusive(bound), UserInputBound::Unbounded),
-                    // default case
-                    _ => (UserInputBound::Unbounded, UserInputBound::Unbounded)
-                });
-        let lower_bound = (one_of("{[".chars()), range_term_val())
-            .map(|(boundary_char, lower_bound): (char, String)|
-                 if lower_bound == "*" {
-                     UserInputBound::Unbounded
-                 } else if boundary_char == '{' {
-                         UserInputBound::Exclusive(lower_bound)
-                 } else {
-                     UserInputBound::Inclusive(lower_bound)
-                 });
-        let upper_bound = (range_term_val(), one_of("}]".chars()))
-            .map(|(higher_bound, boundary_char): (String, char)|
-                 if higher_bound == "*" {
-                     UserInputBound::Unbounded
-                 } else if boundary_char == '}' {
-                     UserInputBound::Exclusive(higher_bound)
-                 } else {
-                     UserInputBound::Inclusive(higher_bound)
-                 });
-         // return only lower and upper
-        let lower_to_upper = (lower_bound.
-                                    skip((spaces(),
-                                          string("TO"),
-                                          spaces())),
-                                    upper_bound);
-
-        (optional(field()).skip(spaces()),
-         // try elastic first, if it matches, the range is unbounded
-         attempt(elastic_unbounded_range).or(lower_to_upper))
-            .map(|(field, (lower, upper))|
-                 // Construct the leaf from extracted field (optional)
-                 // and bounds
-                 UserInputLeaf::Range {
-                     field,
-                     lower,
-                     upper
         })
-    }
+}
+
+fn term_val<'a>() -> impl Parser<&'a str, Output = String> {
+    let phrase = char('"').with(many1(satisfy(|c| c != '"'))).skip(char('"'));
+    phrase.or(word())
+}
+
+fn term_query<'a>() -> impl Parser<&'a str, Output = UserInputLiteral> {
+    let term_val_with_field = negative_number().or(term_val());
+    (field(), term_val_with_field).map(|(field_name, phrase)| UserInputLiteral {
+        field_name: Some(field_name),
+        phrase,
+    })
+}
+
+fn literal<'a>() -> impl Parser<&'a str, Output = UserInputLeaf> {
+    let term_default_field = term_val().map(|phrase| UserInputLiteral {
+        field_name: None,
+        phrase,
+    });
+    attempt(term_query())
+        .or(term_default_field)
+        .map(UserInputLeaf::from)
+}
+
+fn negative_number<'a>() -> impl Parser<&'a str, Output = String> {
+    (
+        char('-'),
+        many1(digit()),
+        optional((char('.'), many1(digit()))),
+    )
+        .map(|(s1, s2, s3): (char, String, Option<(char, String)>)| {
+            if let Some(('.', s3)) = s3 {
+                format!("{}{}.{}", s1, s2, s3)
+            } else {
+                format!("{}{}", s1, s2)
+            }
+        })
+}
+
+fn spaces1<'a>() -> impl Parser<&'a str, Output = ()> {
+    skip_many1(space())
+}
+
+/// Function that parses a range out of a Stream
+/// Supports ranges like:
+/// [5 TO 10], {5 TO 10}, [* TO 10], [10 TO *], {10 TO *], >5, <=10
+/// [a TO *], [a TO c], [abc TO bcd}
+fn range<'a>() -> impl Parser<&'a str, Output = UserInputLeaf> {
+    let range_term_val = || {
+        word()
+            .or(negative_number())
+            .or(char('*').with(value("*".to_string())))
+    };
+
+    // check for unbounded range in the form of <5, <=10, >5, >=5
+    let elastic_unbounded_range = (
+        choice([
+            attempt(string(">=")),
+            attempt(string("<=")),
+            attempt(string("<")),
+            attempt(string(">")),
+        ])
+        .skip(spaces()),
+        range_term_val(),
+    )
+        .map(
+            |(comparison_sign, bound): (&str, String)| match comparison_sign {
+                ">=" => (UserInputBound::Inclusive(bound), UserInputBound::Unbounded),
+                "<=" => (UserInputBound::Unbounded, UserInputBound::Inclusive(bound)),
+                "<" => (UserInputBound::Unbounded, UserInputBound::Exclusive(bound)),
+                ">" => (UserInputBound::Exclusive(bound), UserInputBound::Unbounded),
+                // default case
+                _ => (UserInputBound::Unbounded, UserInputBound::Unbounded),
+            },
+        );
+    let lower_bound = (one_of("{[".chars()), range_term_val()).map(
+        |(boundary_char, lower_bound): (char, String)| {
+            if lower_bound == "*" {
+                UserInputBound::Unbounded
+            } else if boundary_char == '{' {
+                UserInputBound::Exclusive(lower_bound)
+            } else {
+                UserInputBound::Inclusive(lower_bound)
+            }
+        },
+    );
+    let upper_bound = (range_term_val(), one_of("}]".chars())).map(
+        |(higher_bound, boundary_char): (String, char)| {
+            if higher_bound == "*" {
+                UserInputBound::Unbounded
+            } else if boundary_char == '}' {
+                UserInputBound::Exclusive(higher_bound)
+            } else {
+                UserInputBound::Inclusive(higher_bound)
+            }
+        },
+    );
+    // return only lower and upper
+    let lower_to_upper = (
+        lower_bound.skip((spaces(), string("TO"), spaces())),
+        upper_bound,
+    );
+
+    (
+        optional(field()).skip(spaces()),
+        // try elastic first, if it matches, the range is unbounded
+        attempt(elastic_unbounded_range).or(lower_to_upper),
+    )
+        .map(|(field, (lower, upper))|
+             // Construct the leaf from extracted field (optional)
+             // and bounds
+             UserInputLeaf::Range {
+                 field,
+                 lower,
+                 upper
+    })
 }
 
 fn negate(expr: UserInputAST) -> UserInputAST {
     expr.unary(Occur::MustNot)
 }
 
-fn must(expr: UserInputAST) -> UserInputAST {
+fn leaf<'a>() -> impl Parser<&'a str, Output = UserInputAST> {
-    expr.unary(Occur::Must)
+    parser(|input| {
+        char('(')
+            .with(ast())
+            .skip(char(')'))
+            .or(char('*').map(|_| UserInputAST::from(UserInputLeaf::All)))
+            .or(attempt(
+                string("NOT").skip(spaces1()).with(leaf()).map(negate),
+            ))
+            .or(attempt(range().map(UserInputAST::from)))
+            .or(literal().map(UserInputAST::from))
+            .parse_stream(input)
+            .into_result()
+    })
 }
 
-parser! {
+fn occur_symbol<'a>() -> impl Parser<&'a str, Output = Occur> {
-    fn leaf[I]()(I) -> UserInputAST
+    char('-')
-    where [I: Stream<Item = char>] {
+        .map(|_| Occur::MustNot)
-            char('-').with(leaf()).map(negate)
+        .or(char('+').map(|_| Occur::Must))
-        .or(char('+').with(leaf()).map(must))
+}
-        .or(char('(').with(ast()).skip(char(')')))
+
-        .or(char('*').map(|_| UserInputAST::from(UserInputLeaf::All)))
+fn occur_leaf<'a>() -> impl Parser<&'a str, Output = (Option<Occur>, UserInputAST)> {
-        .or(attempt(string("NOT").skip(spaces1()).with(leaf()).map(negate)))
+    (optional(occur_symbol()), boosted_leaf())
-        .or(attempt(range().map(UserInputAST::from)))
+}
-        .or(literal().map(UserInputAST::from))
+
-    }
+fn positive_float_number<'a>() -> impl Parser<&'a str, Output = f64> {
+    (many1(digit()), optional((char('.'), many1(digit())))).map(
+        |(int_part, decimal_part_opt): (String, Option<(char, String)>)| {
+            let mut float_str = int_part;
+            if let Some((chr, decimal_str)) = decimal_part_opt {
+                float_str.push(chr);
+                float_str.push_str(&decimal_str);
+            }
+            float_str.parse::<f64>().unwrap()
+        },
+    )
+}
+
+fn boost<'a>() -> impl Parser<&'a str, Output = f64> {
+    (char('^'), positive_float_number()).map(|(_, boost)| boost)
+}
+
+fn boosted_leaf<'a>() -> impl Parser<&'a str, Output = UserInputAST> {
+    (leaf(), optional(boost())).map(|(leaf, boost_opt)| match boost_opt {
+        Some(boost) if (boost - 1.0).abs() > std::f64::EPSILON => {
+            UserInputAST::Boost(Box::new(leaf), boost)
+        }
+        _ => leaf,
+    })
 }
 
 #[derive(Clone, Copy)]
@@ -174,13 +212,10 @@ enum BinaryOperand {
     And,
 }
 
-parser! {
+fn binary_operand<'a>() -> impl Parser<&'a str, Output = BinaryOperand> {
-    fn binary_operand[I]()(I) -> BinaryOperand
+    string("AND")
-    where [I: Stream<Item = char>]
+        .with(value(BinaryOperand::And))
-    {
+        .or(string("OR").with(value(BinaryOperand::Or)))
-       string("AND").with(value(BinaryOperand::And))
-       .or(string("OR").with(value(BinaryOperand::Or)))
-    }
 }
 
 fn aggregate_binary_expressions(
@@ -208,37 +243,81 @@ fn aggregate_binary_expressions(
     }
 }
 
-parser! {
+fn operand_leaf<'a>() -> impl Parser<&'a str, Output = (BinaryOperand, UserInputAST)> {
-    pub fn ast[I]()(I) -> UserInputAST
+    (
-    where [I: Stream<Item = char>]
+        binary_operand().skip(spaces()),
-    {
+        boosted_leaf().skip(spaces()),
-        let operand_leaf = (binary_operand().skip(spaces()), leaf().skip(spaces()));
+    )
-        let boolean_expr = (leaf().skip(spaces().silent()), many1(operand_leaf)).map(
-            |(left, right)| aggregate_binary_expressions(left,right));
-        let whitespace_separated_leaves = many1(leaf().skip(spaces().silent()))
-        .map(|subqueries: Vec<UserInputAST>|
-            if subqueries.len() == 1 {
-                subqueries.into_iter().next().unwrap()
-            } else {
-                UserInputAST::Clause(subqueries.into_iter().collect())
-            });
-        let expr = attempt(boolean_expr).or(whitespace_separated_leaves);
-        spaces().with(expr).skip(spaces())
-    }
 }
 
-parser! {
+pub fn ast<'a>() -> impl Parser<&'a str, Output = UserInputAST> {
-    pub fn parse_to_ast[I]()(I) -> UserInputAST
+    let boolean_expr = (boosted_leaf().skip(spaces()), many1(operand_leaf()))
-    where [I: Stream<Item = char>]
+        .map(|(left, right)| aggregate_binary_expressions(left, right));
-    {
+    let whitespace_separated_leaves = many1(occur_leaf().skip(spaces().silent())).map(
-        spaces().with(optional(ast()).skip(eof())).map(|opt_ast| opt_ast.unwrap_or_else(UserInputAST::empty_query))
+        |subqueries: Vec<(Option<Occur>, UserInputAST)>| {
-    }
+            if subqueries.len() == 1 {
+                let (occur_opt, ast) = subqueries.into_iter().next().unwrap();
+                match occur_opt.unwrap_or(Occur::Should) {
+                    Occur::Must | Occur::Should => ast,
+                    Occur::MustNot => UserInputAST::Clause(vec![(Some(Occur::MustNot), ast)]),
+                }
+            } else {
+                UserInputAST::Clause(subqueries.into_iter().collect())
+            }
+        },
+    );
+    let expr = attempt(boolean_expr).or(whitespace_separated_leaves);
+    spaces().with(expr).skip(spaces())
+}
+
+pub fn parse_to_ast<'a>() -> impl Parser<&'a str, Output = UserInputAST> {
+    spaces()
+        .with(optional(ast()).skip(eof()))
+        .map(|opt_ast| opt_ast.unwrap_or_else(UserInputAST::empty_query))
 }
 
 #[cfg(test)]
 mod test {
 
     use super::*;
+    use combine::parser::Parser;
+
+    pub fn nearly_equals(a: f64, b: f64) -> bool {
+        (a - b).abs() < 0.0005 * (a + b).abs()
+    }
+
+    fn assert_nearly_equals(expected: f64, val: f64) {
+        assert!(
+            nearly_equals(val, expected),
+            "Got {}, expected {}.",
+            val,
+            expected
+        );
+    }
+
+    #[test]
+    fn test_occur_symbol() {
+        assert_eq!(super::occur_symbol().parse("-"), Ok((Occur::MustNot, "")));
+        assert_eq!(super::occur_symbol().parse("+"), Ok((Occur::Must, "")));
+    }
+
+    #[test]
+    fn test_positive_float_number() {
+        fn valid_parse(float_str: &str, expected_val: f64, expected_remaining: &str) {
+            let (val, remaining) = positive_float_number().parse(float_str).unwrap();
+            assert_eq!(remaining, expected_remaining);
+            assert_nearly_equals(val, expected_val);
+        }
+        fn error_parse(float_str: &str) {
+            assert!(positive_float_number().parse(float_str).is_err());
+        }
+        valid_parse("1.0", 1.0, "");
+        valid_parse("1", 1.0, "");
+        valid_parse("0.234234 aaa", 0.234234f64, " aaa");
+        error_parse(".3332");
+        error_parse("1.");
+        error_parse("-1.");
+    }
 
     fn test_parse_query_to_ast_helper(query: &str, expected: &str) {
         let query = parse_to_ast().parse(query).unwrap().0;
@@ -269,15 +348,24 @@ mod test {
             "Err(UnexpectedParse)"
         );
         test_parse_query_to_ast_helper("NOTa", "\"NOTa\"");
-        test_parse_query_to_ast_helper("NOT a", "-(\"a\")");
+        test_parse_query_to_ast_helper("NOT a", "(-\"a\")");
+    }
+
+    #[test]
+    fn test_boosting() {
+        assert!(parse_to_ast().parse("a^2^3").is_err());
+        assert!(parse_to_ast().parse("a^2^").is_err());
+        test_parse_query_to_ast_helper("a^3", "(\"a\")^3");
+        test_parse_query_to_ast_helper("a^3 b^2", "(*(\"a\")^3 *(\"b\")^2)");
+        test_parse_query_to_ast_helper("a^1", "\"a\"");
     }
 
     #[test]
     fn test_parse_query_to_ast_binary_op() {
-        test_parse_query_to_ast_helper("a AND b", "(+(\"a\") +(\"b\"))");
+        test_parse_query_to_ast_helper("a AND b", "(+\"a\" +\"b\")");
-        test_parse_query_to_ast_helper("a OR b", "(?(\"a\") ?(\"b\"))");
+        test_parse_query_to_ast_helper("a OR b", "(?\"a\" ?\"b\")");
-        test_parse_query_to_ast_helper("a OR b AND c", "(?(\"a\") ?((+(\"b\") +(\"c\"))))");
+        test_parse_query_to_ast_helper("a OR b AND c", "(?\"a\" ?(+\"b\" +\"c\"))");
-        test_parse_query_to_ast_helper("a AND b         AND c", "(+(\"a\") +(\"b\") +(\"c\"))");
+        test_parse_query_to_ast_helper("a AND b         AND c", "(+\"a\" +\"b\" +\"c\")");
         assert_eq!(
             format!("{:?}", parse_to_ast().parse("a OR b aaa")),
             "Err(UnexpectedParse)"
@@ -315,6 +403,13 @@ mod test {
         test_parse_query_to_ast_helper("weight: <= 70.5", "weight:{\"*\" TO \"70.5\"]");
     }
 
+    #[test]
+    fn test_occur_leaf() {
+        let ((occur, ast), _) = super::occur_leaf().parse("+abc").unwrap();
+        assert_eq!(occur, Some(Occur::Must));
+        assert_eq!(format!("{:?}", ast), "\"abc\"");
+    }
+
     #[test]
     fn test_range_parser() {
         // testing the range() parser separately
@@ -343,32 +438,67 @@ mod test {
     fn test_parse_query_to_triming_spaces() {
         test_parse_query_to_ast_helper("   abc", "\"abc\"");
         test_parse_query_to_ast_helper("abc ", "\"abc\"");
-        test_parse_query_to_ast_helper("(  a OR abc)", "(?(\"a\") ?(\"abc\"))");
+        test_parse_query_to_ast_helper("(  a OR abc)", "(?\"a\" ?\"abc\")");
-        test_parse_query_to_ast_helper("(a  OR abc)", "(?(\"a\") ?(\"abc\"))");
+        test_parse_query_to_ast_helper("(a  OR abc)", "(?\"a\" ?\"abc\")");
-        test_parse_query_to_ast_helper("(a OR  abc)", "(?(\"a\") ?(\"abc\"))");
+        test_parse_query_to_ast_helper("(a OR  abc)", "(?\"a\" ?\"abc\")");
-        test_parse_query_to_ast_helper("a OR abc ", "(?(\"a\") ?(\"abc\"))");
+        test_parse_query_to_ast_helper("a OR abc ", "(?\"a\" ?\"abc\")");
-        test_parse_query_to_ast_helper("(a OR abc )", "(?(\"a\") ?(\"abc\"))");
+        test_parse_query_to_ast_helper("(a OR abc )", "(?\"a\" ?\"abc\")");
-        test_parse_query_to_ast_helper("(a OR  abc) ", "(?(\"a\") ?(\"abc\"))");
+        test_parse_query_to_ast_helper("(a OR  abc) ", "(?\"a\" ?\"abc\")");
     }
 
     #[test]
-    fn test_parse_query_to_ast() {
+    fn test_parse_query_single_term() {
         test_parse_query_to_ast_helper("abc", "\"abc\"");
-        test_parse_query_to_ast_helper("a b", "(\"a\" \"b\")");
+    }
-        test_parse_query_to_ast_helper("+(a b)", "+((\"a\" \"b\"))");
+
-        test_parse_query_to_ast_helper("+d", "+(\"d\")");
+    #[test]
-        test_parse_query_to_ast_helper("+(a b) +d", "(+((\"a\" \"b\")) +(\"d\"))");
+    fn test_parse_query_default_clause() {
-        test_parse_query_to_ast_helper("(+a +b) d", "((+(\"a\") +(\"b\")) \"d\")");
+        test_parse_query_to_ast_helper("a b", "(*\"a\" *\"b\")");
-        test_parse_query_to_ast_helper("(+a)", "+(\"a\")");
+    }
-        test_parse_query_to_ast_helper("(+a +b)", "(+(\"a\") +(\"b\"))");
+
+    #[test]
+    fn test_parse_query_must_default_clause() {
+        test_parse_query_to_ast_helper("+(a b)", "(*\"a\" *\"b\")");
+    }
+
+    #[test]
+    fn test_parse_query_must_single_term() {
+        test_parse_query_to_ast_helper("+d", "\"d\"");
+    }
+
+    #[test]
+    fn test_single_term_with_field() {
         test_parse_query_to_ast_helper("abc:toto", "abc:\"toto\"");
+    }
+
+    #[test]
+    fn test_single_term_with_float() {
         test_parse_query_to_ast_helper("abc:1.1", "abc:\"1.1\"");
-        test_parse_query_to_ast_helper("+abc:toto", "+(abc:\"toto\")");
+    }
-        test_parse_query_to_ast_helper("(+abc:toto -titi)", "(+(abc:\"toto\") -(\"titi\"))");
+
-        test_parse_query_to_ast_helper("-abc:toto", "-(abc:\"toto\")");
+    #[test]
-        test_parse_query_to_ast_helper("abc:a b", "(abc:\"a\" \"b\")");
+    fn test_must_clause() {
+        test_parse_query_to_ast_helper("(+a +b)", "(+\"a\" +\"b\")");
+    }
+
+    #[test]
+    fn test_parse_test_query_plus_a_b_plus_d() {
+        test_parse_query_to_ast_helper("+(a b) +d", "(+(*\"a\" *\"b\") +\"d\")");
+    }
+
+    #[test]
+    fn test_parse_test_query_other() {
+        test_parse_query_to_ast_helper("(+a +b) d", "(*(+\"a\" +\"b\") *\"d\")");
+        test_parse_query_to_ast_helper("+abc:toto", "abc:\"toto\"");
+        test_parse_query_to_ast_helper("(+abc:toto -titi)", "(+abc:\"toto\" -\"titi\")");
+        test_parse_query_to_ast_helper("-abc:toto", "(-abc:\"toto\")");
+        test_parse_query_to_ast_helper("abc:a b", "(*abc:\"a\" *\"b\")");
         test_parse_query_to_ast_helper("abc:\"a b\"", "abc:\"a b\"");
         test_parse_query_to_ast_helper("foo:[1 TO 5]", "foo:[\"1\" TO \"5\"]");
+    }
+
+    #[test]
+    fn test_parse_query_with_range() {
         test_parse_query_to_ast_helper("[1 TO 5]", "[\"1\" TO \"5\"]");
         test_parse_query_to_ast_helper("foo:{a TO z}", "foo:{\"a\" TO \"z\"}");
         test_parse_query_to_ast_helper("foo:[1 TO toto}", "foo:[\"1\" TO \"toto\"}");

query-grammar/src/user_input_ast.rs

@@ -85,14 +85,14 @@ impl UserInputBound {
 }
 
 pub enum UserInputAST {
-    Clause(Vec<UserInputAST>),
+    Clause(Vec<(Option<Occur>, UserInputAST)>),
-    Unary(Occur, Box<UserInputAST>),
     Leaf(Box<UserInputLeaf>),
+    Boost(Box<UserInputAST>, f64),
 }
 
 impl UserInputAST {
     pub fn unary(self, occur: Occur) -> UserInputAST {
-        UserInputAST::Unary(occur, Box::new(self))
+        UserInputAST::Clause(vec![(Some(occur), self)])
     }
 
     fn compose(occur: Occur, asts: Vec<UserInputAST>) -> UserInputAST {
@@ -103,7 +103,7 @@ impl UserInputAST {
         } else {
             UserInputAST::Clause(
                 asts.into_iter()
-                    .map(|ast: UserInputAST| ast.unary(occur))
+                    .map(|ast: UserInputAST| (Some(occur), ast))
                     .collect::<Vec<_>>(),
             )
         }
@@ -134,26 +134,38 @@ impl From<UserInputLeaf> for UserInputAST {
     }
 }
 
+fn print_occur_ast(
+    occur_opt: Option<Occur>,
+    ast: &UserInputAST,
+    formatter: &mut fmt::Formatter,
+) -> fmt::Result {
+    if let Some(occur) = occur_opt {
+        write!(formatter, "{}{:?}", occur, ast)?;
+    } else {
+        write!(formatter, "*{:?}", ast)?;
+    }
+    Ok(())
+}
+
 impl fmt::Debug for UserInputAST {
-    fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> Result<(), fmt::Error> {
+    fn fmt(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
         match *self {
             UserInputAST::Clause(ref subqueries) => {
                 if subqueries.is_empty() {
                     write!(formatter, "<emptyclause>")?;
                 } else {
                     write!(formatter, "(")?;
-                    write!(formatter, "{:?}", &subqueries[0])?;
+                    print_occur_ast(subqueries[0].0, &subqueries[0].1, formatter)?;
                     for subquery in &subqueries[1..] {
-                        write!(formatter, " {:?}", subquery)?;
+                        write!(formatter, " ")?;
+                        print_occur_ast(subquery.0, &subquery.1, formatter)?;
                     }
                     write!(formatter, ")")?;
                 }
                 Ok(())
             }
-            UserInputAST::Unary(ref occur, ref subquery) => {
-                write!(formatter, "{}({:?})", occur, subquery)
-            }
             UserInputAST::Leaf(ref subquery) => write!(formatter, "{:?}", subquery),
+            UserInputAST::Boost(ref leaf, boost) => write!(formatter, "({:?})^{}", leaf, boost),
         }
     }
 }

src/collector/count_collector.rs

@@ -1,7 +1,6 @@
 use super::Collector;
 use crate::collector::SegmentCollector;
 use crate::DocId;
-use crate::Result;
 use crate::Score;
 use crate::SegmentLocalId;
 use crate::SegmentReader;
@@ -13,44 +12,29 @@ use crate::SegmentReader;
 /// use tantivy::collector::Count;
 /// use tantivy::query::QueryParser;
 /// use tantivy::schema::{Schema, TEXT};
-/// use tantivy::{doc, Index, Result};
+/// use tantivy::{doc, Index};
 ///
-/// # fn main() { example().unwrap(); }
+/// let mut schema_builder = Schema::builder();
-/// fn example() -> Result<()> {
+/// let title = schema_builder.add_text_field("title", TEXT);
-///     let mut schema_builder = Schema::builder();
+/// let schema = schema_builder.build();
-///     let title = schema_builder.add_text_field("title", TEXT);
+/// let index = Index::create_in_ram(schema);
-///     let schema = schema_builder.build();
-///     let index = Index::create_in_ram(schema);
-///     {
-///         let mut index_writer = index.writer(3_000_000)?;
-///         index_writer.add_document(doc!(
-///             title => "The Name of the Wind",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of Muadib",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "A Dairy Cow",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of a Young Girl",
-///         ));
-///         index_writer.commit().unwrap();
-///     }
 ///
-///     let reader = index.reader()?;
+/// let mut index_writer = index.writer(3_000_000).unwrap();
-///     let searcher = reader.searcher();
+/// index_writer.add_document(doc!(title => "The Name of the Wind"));
+/// index_writer.add_document(doc!(title => "The Diary of Muadib"));
+/// index_writer.add_document(doc!(title => "A Dairy Cow"));
+/// index_writer.add_document(doc!(title => "The Diary of a Young Girl"));
+/// assert!(index_writer.commit().is_ok());
 ///
-///     {
+/// let reader = index.reader().unwrap();
-///         let query_parser = QueryParser::for_index(&index, vec![title]);
+/// let searcher = reader.searcher();
-///         let query = query_parser.parse_query("diary")?;
-///         let count = searcher.search(&query, &Count).unwrap();
 ///
-///         assert_eq!(count, 2);
+/// // Here comes the important part
-///     }
+/// let query_parser = QueryParser::for_index(&index, vec![title]);
+/// let query = query_parser.parse_query("diary").unwrap();
+/// let count = searcher.search(&query, &Count).unwrap();
 ///
-///     Ok(())
+/// assert_eq!(count, 2);
-/// }
 /// ```
 pub struct Count;
 
@@ -59,7 +43,11 @@ impl Collector for Count {
 
     type Child = SegmentCountCollector;
 
-    fn for_segment(&self, _: SegmentLocalId, _: &SegmentReader) -> Result<SegmentCountCollector> {
+    fn for_segment(
+        &self,
+        _: SegmentLocalId,
+        _: &SegmentReader,
+    ) -> crate::Result<SegmentCountCollector> {
         Ok(SegmentCountCollector::default())
     }
 
@@ -67,7 +55,7 @@ impl Collector for Count {
         false
     }
 
-    fn merge_fruits(&self, segment_counts: Vec<usize>) -> Result<usize> {
+    fn merge_fruits(&self, segment_counts: Vec<usize>) -> crate::Result<usize> {
         Ok(segment_counts.into_iter().sum())
     }
 }
@@ -108,18 +96,18 @@ mod tests {
         }
         {
             let mut count_collector = SegmentCountCollector::default();
-            count_collector.collect(0u32, 1f32);
+            count_collector.collect(0u32, 1.0);
             assert_eq!(count_collector.harvest(), 1);
         }
         {
             let mut count_collector = SegmentCountCollector::default();
-            count_collector.collect(0u32, 1f32);
+            count_collector.collect(0u32, 1.0);
             assert_eq!(count_collector.harvest(), 1);
         }
         {
             let mut count_collector = SegmentCountCollector::default();
-            count_collector.collect(0u32, 1f32);
+            count_collector.collect(0u32, 1.0);
-            count_collector.collect(1u32, 1f32);
+            count_collector.collect(1u32, 1.0);
             assert_eq!(count_collector.harvest(), 2);
         }
     }

src/collector/custom_score_top_collector.rs

@@ -1,6 +1,5 @@
 use crate::collector::top_collector::{TopCollector, TopSegmentCollector};
 use crate::collector::{Collector, SegmentCollector};
-use crate::Result;
 use crate::{DocAddress, DocId, Score, SegmentReader};
 
 pub(crate) struct CustomScoreTopCollector<TCustomScorer, TScore = Score> {
@@ -12,13 +11,13 @@ impl<TCustomScorer, TScore> CustomScoreTopCollector<TCustomScorer, TScore>
 where
     TScore: Clone + PartialOrd,
 {
-    pub fn new(
+    pub(crate) fn new(
         custom_scorer: TCustomScorer,
-        limit: usize,
+        collector: TopCollector<TScore>,
     ) -> CustomScoreTopCollector<TCustomScorer, TScore> {
         CustomScoreTopCollector {
             custom_scorer,
-            collector: TopCollector::with_limit(limit),
+            collector,
         }
     }
 }
@@ -29,7 +28,7 @@ where
 /// It is the segment local version of the [`CustomScorer`](./trait.CustomScorer.html).
 pub trait CustomSegmentScorer<TScore>: 'static {
     /// Computes the score of a specific `doc`.
-    fn score(&self, doc: DocId) -> TScore;
+    fn score(&mut self, doc: DocId) -> TScore;
 }
 
 /// `CustomScorer` makes it possible to define any kind of score.
@@ -42,12 +41,12 @@ pub trait CustomScorer<TScore>: Sync {
     type Child: CustomSegmentScorer<TScore>;
     /// Builds a child scorer for a specific segment. The child scorer is associated to
     /// a specific segment.
-    fn segment_scorer(&self, segment_reader: &SegmentReader) -> Result<Self::Child>;
+    fn segment_scorer(&self, segment_reader: &SegmentReader) -> crate::Result<Self::Child>;
 }
 
 impl<TCustomScorer, TScore> Collector for CustomScoreTopCollector<TCustomScorer, TScore>
 where
-    TCustomScorer: CustomScorer<TScore>,
+    TCustomScorer: CustomScorer<TScore> + Send + Sync,
     TScore: 'static + PartialOrd + Clone + Send + Sync,
 {
     type Fruit = Vec<(TScore, DocAddress)>;
@@ -58,7 +57,7 @@ where
         &self,
         segment_local_id: u32,
         segment_reader: &SegmentReader,
-    ) -> Result<Self::Child> {
+    ) -> crate::Result<Self::Child> {
         let segment_scorer = self.custom_scorer.segment_scorer(segment_reader)?;
         let segment_collector = self
             .collector
@@ -73,7 +72,7 @@ where
         false
     }
 
-    fn merge_fruits(&self, segment_fruits: Vec<Self::Fruit>) -> Result<Self::Fruit> {
+    fn merge_fruits(&self, segment_fruits: Vec<Self::Fruit>) -> crate::Result<Self::Fruit> {
         self.collector.merge_fruits(segment_fruits)
     }
 }
@@ -111,16 +110,16 @@ where
 {
     type Child = T;
 
-    fn segment_scorer(&self, segment_reader: &SegmentReader) -> Result<Self::Child> {
+    fn segment_scorer(&self, segment_reader: &SegmentReader) -> crate::Result<Self::Child> {
         Ok((self)(segment_reader))
     }
 }
 
 impl<F, TScore> CustomSegmentScorer<TScore> for F
 where
-    F: 'static + Sync + Send + Fn(DocId) -> TScore,
+    F: 'static + FnMut(DocId) -> TScore,
 {
-    fn score(&self, doc: DocId) -> TScore {
+    fn score(&mut self, doc: DocId) -> TScore {
         (self)(doc)
     }
 }

src/collector/facet_collector.rs

@@ -1,11 +1,9 @@
 use crate::collector::Collector;
 use crate::collector::SegmentCollector;
-use crate::docset::SkipResult;
 use crate::fastfield::FacetReader;
 use crate::schema::Facet;
 use crate::schema::Field;
 use crate::DocId;
-use crate::Result;
 use crate::Score;
 use crate::SegmentLocalId;
 use crate::SegmentReader;
@@ -84,10 +82,9 @@ fn facet_depth(facet_bytes: &[u8]) -> usize {
 /// use tantivy::collector::FacetCollector;
 /// use tantivy::query::AllQuery;
 /// use tantivy::schema::{Facet, Schema, TEXT};
-/// use tantivy::{doc, Index, Result};
+/// use tantivy::{doc, Index};
 ///
-/// # fn main() { example().unwrap(); }
+/// fn example() -> tantivy::Result<()> {
-/// fn example() -> Result<()> {
 ///     let mut schema_builder = Schema::builder();
 ///
 ///     // Facet have their own specific type.
@@ -127,7 +124,7 @@ fn facet_depth(facet_bytes: &[u8]) -> usize {
 ///     let searcher = reader.searcher();
 ///
 ///     {
-///			let mut facet_collector = FacetCollector::for_field(facet);
+///         let mut facet_collector = FacetCollector::for_field(facet);
 ///         facet_collector.add_facet("/lang");
 ///         facet_collector.add_facet("/category");
 ///         let facet_counts = searcher.search(&AllQuery, &facet_collector)?;
@@ -143,7 +140,7 @@ fn facet_depth(facet_bytes: &[u8]) -> usize {
 ///     }
 ///
 ///     {
-///			let mut facet_collector = FacetCollector::for_field(facet);
+///         let mut facet_collector = FacetCollector::for_field(facet);
 ///         facet_collector.add_facet("/category/fiction");
 ///         let facet_counts = searcher.search(&AllQuery, &facet_collector)?;
 ///
@@ -158,8 +155,8 @@ fn facet_depth(facet_bytes: &[u8]) -> usize {
 ///         ]);
 ///     }
 ///
-///    {
+///     {
-///			let mut facet_collector = FacetCollector::for_field(facet);
+///         let mut facet_collector = FacetCollector::for_field(facet);
 ///         facet_collector.add_facet("/category/fiction");
 ///         let facet_counts = searcher.search(&AllQuery, &facet_collector)?;
 ///
@@ -172,6 +169,7 @@ fn facet_depth(facet_bytes: &[u8]) -> usize {
 ///
 ///     Ok(())
 /// }
+/// # assert!(example().is_ok());
 /// ```
 pub struct FacetCollector {
     field: Field,
@@ -189,6 +187,11 @@ pub struct FacetSegmentCollector {
     collapse_facet_ords: Vec<u64>,
 }
 
+enum SkipResult {
+    Found,
+    NotFound,
+}
+
 fn skip<'a, I: Iterator<Item = &'a Facet>>(
     target: &[u8],
     collapse_it: &mut Peekable<I>,
@@ -198,14 +201,14 @@ fn skip<'a, I: Iterator<Item = &'a Facet>>(
             Some(facet_bytes) => match facet_bytes.encoded_str().as_bytes().cmp(target) {
                 Ordering::Less => {}
                 Ordering::Greater => {
-                    return SkipResult::OverStep;
+                    return SkipResult::NotFound;
                 }
                 Ordering::Equal => {
-                    return SkipResult::Reached;
+                    return SkipResult::Found;
                 }
             },
             None => {
-                return SkipResult::End;
+                return SkipResult::NotFound;
             }
         }
         collapse_it.next();
@@ -262,7 +265,7 @@ impl Collector for FacetCollector {
         &self,
         _: SegmentLocalId,
         reader: &SegmentReader,
-    ) -> Result<FacetSegmentCollector> {
+    ) -> crate::Result<FacetSegmentCollector> {
         let field_name = reader.schema().get_field_name(self.field);
         let facet_reader = reader.facet_reader(self.field).ok_or_else(|| {
             TantivyError::SchemaError(format!("Field {:?} is not a facet field.", field_name))
@@ -282,7 +285,7 @@ impl Collector for FacetCollector {
                     // is positionned on a term that has not been processed yet.
                     let skip_result = skip(facet_streamer.key(), &mut collapse_facet_it);
                     match skip_result {
-                        SkipResult::Reached => {
+                        SkipResult::Found => {
                             // we reach a facet we decided to collapse.
                             let collapse_depth = facet_depth(facet_streamer.key());
                             let mut collapsed_id = 0;
@@ -302,7 +305,7 @@ impl Collector for FacetCollector {
                             }
                             break;
                         }
-                        SkipResult::End | SkipResult::OverStep => {
+                        SkipResult::NotFound => {
                             collapse_mapping.push(0);
                             if !facet_streamer.advance() {
                                 break;
@@ -328,7 +331,7 @@ impl Collector for FacetCollector {
         false
     }
 
-    fn merge_fruits(&self, segments_facet_counts: Vec<FacetCounts>) -> Result<FacetCounts> {
+    fn merge_fruits(&self, segments_facet_counts: Vec<FacetCounts>) -> crate::Result<FacetCounts> {
         let mut facet_counts: BTreeMap<Facet, u64> = BTreeMap::new();
         for segment_facet_counts in segments_facet_counts {
             for (facet, count) in segment_facet_counts.facet_counts {
@@ -452,9 +455,11 @@ impl FacetCounts {
 #[cfg(test)]
 mod tests {
     use super::{FacetCollector, FacetCounts};
+    use crate::collector::Count;
     use crate::core::Index;
-    use crate::query::AllQuery;
+    use crate::query::{AllQuery, QueryParser, TermQuery};
-    use crate::schema::{Document, Facet, Field, Schema};
+    use crate::schema::{Document, Facet, Field, IndexRecordOption, Schema};
+    use crate::Term;
     use rand::distributions::Uniform;
     use rand::prelude::SliceRandom;
     use rand::{thread_rng, Rng};
@@ -544,6 +549,56 @@ mod tests {
         assert_eq!(facets[0].1, 1);
     }
 
+    #[test]
+    fn test_doc_search_by_facet() {
+        let mut schema_builder = Schema::builder();
+        let facet_field = schema_builder.add_facet_field("facet");
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        index_writer.add_document(doc!(
+            facet_field => Facet::from_text(&"/A/A"),
+        ));
+        index_writer.add_document(doc!(
+            facet_field => Facet::from_text(&"/A/B"),
+        ));
+        index_writer.add_document(doc!(
+            facet_field => Facet::from_text(&"/A/C/A"),
+        ));
+        index_writer.add_document(doc!(
+            facet_field => Facet::from_text(&"/D/C/A"),
+        ));
+        index_writer.commit().unwrap();
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+        assert_eq!(searcher.num_docs(), 4);
+
+        let count_facet = |facet_str: &str| {
+            let term = Term::from_facet(facet_field, &Facet::from_text(facet_str));
+            searcher
+                .search(&TermQuery::new(term, IndexRecordOption::Basic), &Count)
+                .unwrap()
+        };
+
+        assert_eq!(count_facet("/"), 4);
+        assert_eq!(count_facet("/A"), 3);
+        assert_eq!(count_facet("/A/B"), 1);
+        assert_eq!(count_facet("/A/C"), 1);
+        assert_eq!(count_facet("/A/C/A"), 1);
+        assert_eq!(count_facet("/C/A"), 0);
+        {
+            let query_parser = QueryParser::for_index(&index, vec![]);
+            {
+                let query = query_parser.parse_query("facet:/A/B").unwrap();
+                assert_eq!(1, searcher.search(&query, &Count).unwrap());
+            }
+            {
+                let query = query_parser.parse_query("facet:/A").unwrap();
+                assert_eq!(3, searcher.search(&query, &Count).unwrap());
+            }
+        }
+    }
+
     #[test]
     fn test_non_used_facet_collector() {
         let mut facet_collector = FacetCollector::for_field(Field::from_field_id(0));

src/collector/mod.rs

@@ -85,7 +85,6 @@ See the `custom_collector` example.
 */
 
 use crate::DocId;
-use crate::Result;
 use crate::Score;
 use crate::SegmentLocalId;
 use crate::SegmentReader;
@@ -110,6 +109,7 @@ pub use self::tweak_score_top_collector::{ScoreSegmentTweaker, ScoreTweaker};
 
 mod facet_collector;
 pub use self::facet_collector::FacetCollector;
+use crate::query::Weight;
 
 /// `Fruit` is the type for the result of our collection.
 /// e.g. `usize` for the `Count` collector.
@@ -133,7 +133,7 @@ impl<T> Fruit for T where T: Send + downcast_rs::Downcast {}
 /// The collection logic itself is in the `SegmentCollector`.
 ///
 /// Segments are not guaranteed to be visited in any specific order.
-pub trait Collector: Sync {
+pub trait Collector: Sync + Send {
     /// `Fruit` is the type for the result of our collection.
     /// e.g. `usize` for the `Count` collector.
     type Fruit: Fruit;
@@ -147,14 +147,37 @@ pub trait Collector: Sync {
         &self,
         segment_local_id: SegmentLocalId,
         segment: &SegmentReader,
-    ) -> Result<Self::Child>;
+    ) -> crate::Result<Self::Child>;
 
     /// Returns true iff the collector requires to compute scores for documents.
     fn requires_scoring(&self) -> bool;
 
     /// Combines the fruit associated to the collection of each segments
     /// into one fruit.
-    fn merge_fruits(&self, segment_fruits: Vec<Self::Fruit>) -> Result<Self::Fruit>;
+    fn merge_fruits(&self, segment_fruits: Vec<Self::Fruit>) -> crate::Result<Self::Fruit>;
+
+    /// Created a segment collector and
+    fn collect_segment(
+        &self,
+        weight: &dyn Weight,
+        segment_ord: u32,
+        reader: &SegmentReader,
+    ) -> crate::Result<<Self::Child as SegmentCollector>::Fruit> {
+        let mut segment_collector = self.for_segment(segment_ord as u32, reader)?;
+
+        if let Some(delete_bitset) = reader.delete_bitset() {
+            weight.for_each(reader, &mut |doc, score| {
+                if delete_bitset.is_alive(doc) {
+                    segment_collector.collect(doc, score);
+                }
+            })?;
+        } else {
+            weight.for_each(reader, &mut |doc, score| {
+                segment_collector.collect(doc, score);
+            })?;
+        }
+        Ok(segment_collector.harvest())
+    }
 }
 
 /// The `SegmentCollector` is the trait in charge of defining the
@@ -185,7 +208,11 @@ where
     type Fruit = (Left::Fruit, Right::Fruit);
     type Child = (Left::Child, Right::Child);
 
-    fn for_segment(&self, segment_local_id: u32, segment: &SegmentReader) -> Result<Self::Child> {
+    fn for_segment(
+        &self,
+        segment_local_id: u32,
+        segment: &SegmentReader,
+    ) -> crate::Result<Self::Child> {
         let left = self.0.for_segment(segment_local_id, segment)?;
         let right = self.1.for_segment(segment_local_id, segment)?;
         Ok((left, right))
@@ -198,7 +225,7 @@ where
     fn merge_fruits(
         &self,
         children: Vec<(Left::Fruit, Right::Fruit)>,
-    ) -> Result<(Left::Fruit, Right::Fruit)> {
+    ) -> crate::Result<(Left::Fruit, Right::Fruit)> {
         let mut left_fruits = vec![];
         let mut right_fruits = vec![];
         for (left_fruit, right_fruit) in children {
@@ -240,7 +267,11 @@ where
     type Fruit = (One::Fruit, Two::Fruit, Three::Fruit);
     type Child = (One::Child, Two::Child, Three::Child);
 
-    fn for_segment(&self, segment_local_id: u32, segment: &SegmentReader) -> Result<Self::Child> {
+    fn for_segment(
+        &self,
+        segment_local_id: u32,
+        segment: &SegmentReader,
+    ) -> crate::Result<Self::Child> {
         let one = self.0.for_segment(segment_local_id, segment)?;
         let two = self.1.for_segment(segment_local_id, segment)?;
         let three = self.2.for_segment(segment_local_id, segment)?;
@@ -251,7 +282,7 @@ where
         self.0.requires_scoring() || self.1.requires_scoring() || self.2.requires_scoring()
     }
 
-    fn merge_fruits(&self, children: Vec<Self::Fruit>) -> Result<Self::Fruit> {
+    fn merge_fruits(&self, children: Vec<Self::Fruit>) -> crate::Result<Self::Fruit> {
         let mut one_fruits = vec![];
         let mut two_fruits = vec![];
         let mut three_fruits = vec![];
@@ -299,7 +330,11 @@ where
     type Fruit = (One::Fruit, Two::Fruit, Three::Fruit, Four::Fruit);
     type Child = (One::Child, Two::Child, Three::Child, Four::Child);
 
-    fn for_segment(&self, segment_local_id: u32, segment: &SegmentReader) -> Result<Self::Child> {
+    fn for_segment(
+        &self,
+        segment_local_id: u32,
+        segment: &SegmentReader,
+    ) -> crate::Result<Self::Child> {
         let one = self.0.for_segment(segment_local_id, segment)?;
         let two = self.1.for_segment(segment_local_id, segment)?;
         let three = self.2.for_segment(segment_local_id, segment)?;
@@ -314,7 +349,7 @@ where
             || self.3.requires_scoring()
     }
 
-    fn merge_fruits(&self, children: Vec<Self::Fruit>) -> Result<Self::Fruit> {
+    fn merge_fruits(&self, children: Vec<Self::Fruit>) -> crate::Result<Self::Fruit> {
         let mut one_fruits = vec![];
         let mut two_fruits = vec![];
         let mut three_fruits = vec![];

src/collector/multi_collector.rs

@@ -2,7 +2,6 @@ use super::Collector;
 use super::SegmentCollector;
 use crate::collector::Fruit;
 use crate::DocId;
-use crate::Result;
 use crate::Score;
 use crate::SegmentLocalId;
 use crate::SegmentReader;
@@ -24,7 +23,7 @@ impl<TCollector: Collector> Collector for CollectorWrapper<TCollector> {
         &self,
         segment_local_id: u32,
         reader: &SegmentReader,
-    ) -> Result<Box<dyn BoxableSegmentCollector>> {
+    ) -> crate::Result<Box<dyn BoxableSegmentCollector>> {
         let child = self.0.for_segment(segment_local_id, reader)?;
         Ok(Box::new(SegmentCollectorWrapper(child)))
     }
@@ -33,7 +32,10 @@ impl<TCollector: Collector> Collector for CollectorWrapper<TCollector> {
         self.0.requires_scoring()
     }
 
-    fn merge_fruits(&self, children: Vec<<Self as Collector>::Fruit>) -> Result<Box<dyn Fruit>> {
+    fn merge_fruits(
+        &self,
+        children: Vec<<Self as Collector>::Fruit>,
+    ) -> crate::Result<Box<dyn Fruit>> {
         let typed_fruit: Vec<TCollector::Fruit> = children
             .into_iter()
             .map(|untyped_fruit| {
@@ -44,7 +46,7 @@ impl<TCollector: Collector> Collector for CollectorWrapper<TCollector> {
                         TantivyError::InvalidArgument("Failed to cast child fruit.".to_string())
                     })
             })
-            .collect::<Result<_>>()?;
+            .collect::<crate::Result<_>>()?;
         let merged_fruit = self.0.merge_fruits(typed_fruit)?;
         Ok(Box::new(merged_fruit))
     }
@@ -53,7 +55,7 @@ impl<TCollector: Collector> Collector for CollectorWrapper<TCollector> {
 impl SegmentCollector for Box<dyn BoxableSegmentCollector> {
     type Fruit = Box<dyn Fruit>;
 
-    fn collect(&mut self, doc: u32, score: f32) {
+    fn collect(&mut self, doc: u32, score: Score) {
         self.as_mut().collect(doc, score);
     }
 
@@ -63,7 +65,7 @@ impl SegmentCollector for Box<dyn BoxableSegmentCollector> {
 }
 
 pub trait BoxableSegmentCollector {
-    fn collect(&mut self, doc: u32, score: f32);
+    fn collect(&mut self, doc: u32, score: Score);
     fn harvest_from_box(self: Box<Self>) -> Box<dyn Fruit>;
 }
 
@@ -72,7 +74,7 @@ pub struct SegmentCollectorWrapper<TSegmentCollector: SegmentCollector>(TSegment
 impl<TSegmentCollector: SegmentCollector> BoxableSegmentCollector
     for SegmentCollectorWrapper<TSegmentCollector>
 {
-    fn collect(&mut self, doc: u32, score: f32) {
+    fn collect(&mut self, doc: u32, score: Score) {
         self.0.collect(doc, score);
     }
 
@@ -108,49 +110,35 @@ impl<TFruit: Fruit> FruitHandle<TFruit> {
 /// use tantivy::collector::{Count, TopDocs, MultiCollector};
 /// use tantivy::query::QueryParser;
 /// use tantivy::schema::{Schema, TEXT};
-/// use tantivy::{doc, Index, Result};
+/// use tantivy::{doc, Index};
 ///
-/// # fn main() { example().unwrap(); }
+/// let mut schema_builder = Schema::builder();
-/// fn example() -> Result<()> {
+/// let title = schema_builder.add_text_field("title", TEXT);
-///     let mut schema_builder = Schema::builder();
+/// let schema = schema_builder.build();
-///     let title = schema_builder.add_text_field("title", TEXT);
+/// let index = Index::create_in_ram(schema);
-///     let schema = schema_builder.build();
-///     let index = Index::create_in_ram(schema);
-///     {
-///         let mut index_writer = index.writer(3_000_000)?;
-///         index_writer.add_document(doc!(
-///             title => "The Name of the Wind",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of Muadib",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "A Dairy Cow",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of a Young Girl",
-///         ));
-///         index_writer.commit().unwrap();
-///     }
 ///
-///     let reader = index.reader()?;
+/// let mut index_writer = index.writer(3_000_000).unwrap();
-///     let searcher = reader.searcher();
+/// index_writer.add_document(doc!(title => "The Name of the Wind"));
+/// index_writer.add_document(doc!(title => "The Diary of Muadib"));
+/// index_writer.add_document(doc!(title => "A Dairy Cow"));
+/// index_writer.add_document(doc!(title => "The Diary of a Young Girl"));
+/// assert!(index_writer.commit().is_ok());
 ///
-///     let mut collectors = MultiCollector::new();
+/// let reader = index.reader().unwrap();
-///     let top_docs_handle = collectors.add_collector(TopDocs::with_limit(2));
+/// let searcher = reader.searcher();
-///     let count_handle = collectors.add_collector(Count);
-///     let query_parser = QueryParser::for_index(&index, vec![title]);
-///     let query = query_parser.parse_query("diary")?;
-///     let mut multi_fruit = searcher.search(&query, &collectors)?;
 ///
-///     let count = count_handle.extract(&mut multi_fruit);
+/// let mut collectors = MultiCollector::new();
-///     let top_docs = top_docs_handle.extract(&mut multi_fruit);
+/// let top_docs_handle = collectors.add_collector(TopDocs::with_limit(2));
+/// let count_handle = collectors.add_collector(Count);
+/// let query_parser = QueryParser::for_index(&index, vec![title]);
+/// let query = query_parser.parse_query("diary").unwrap();
+/// let mut multi_fruit = searcher.search(&query, &collectors).unwrap();
 ///
-///     # assert_eq!(count, 2);
+/// let count = count_handle.extract(&mut multi_fruit);
-///     # assert_eq!(top_docs.len(), 2);
+/// let top_docs = top_docs_handle.extract(&mut multi_fruit);
 ///
-///     Ok(())
+/// assert_eq!(count, 2);
-/// }
+/// assert_eq!(top_docs.len(), 2);
 /// ```
 #[allow(clippy::type_complexity)]
 #[derive(Default)]
@@ -189,12 +177,12 @@ impl<'a> Collector for MultiCollector<'a> {
         &self,
         segment_local_id: SegmentLocalId,
         segment: &SegmentReader,
-    ) -> Result<MultiCollectorChild> {
+    ) -> crate::Result<MultiCollectorChild> {
         let children = self
             .collector_wrappers
             .iter()
             .map(|collector_wrapper| collector_wrapper.for_segment(segment_local_id, segment))
-            .collect::<Result<Vec<_>>>()?;
+            .collect::<crate::Result<Vec<_>>>()?;
         Ok(MultiCollectorChild { children })
     }
 
@@ -205,7 +193,7 @@ impl<'a> Collector for MultiCollector<'a> {
             .any(Collector::requires_scoring)
     }
 
-    fn merge_fruits(&self, segments_multifruits: Vec<MultiFruit>) -> Result<MultiFruit> {
+    fn merge_fruits(&self, segments_multifruits: Vec<MultiFruit>) -> crate::Result<MultiFruit> {
         let mut segment_fruits_list: Vec<Vec<Box<dyn Fruit>>> = (0..self.collector_wrappers.len())
             .map(|_| Vec::with_capacity(segments_multifruits.len()))
             .collect::<Vec<_>>();
@@ -223,7 +211,7 @@ impl<'a> Collector for MultiCollector<'a> {
             .map(|(child_collector, segment_fruits)| {
                 Ok(Some(child_collector.merge_fruits(segment_fruits)?))
             })
-            .collect::<Result<_>>()?;
+            .collect::<crate::Result<_>>()?;
         Ok(MultiFruit { sub_fruits })
     }
 }

src/collector/tests.rs

@@ -55,7 +55,7 @@ impl Collector for TestCollector {
         &self,
         segment_id: SegmentLocalId,
         _reader: &SegmentReader,
-    ) -> Result<TestSegmentCollector> {
+    ) -> crate::Result<TestSegmentCollector> {
         Ok(TestSegmentCollector {
             segment_id,
             fruit: TestFruit::default(),
@@ -66,7 +66,7 @@ impl Collector for TestCollector {
         self.compute_score
     }
 
-    fn merge_fruits(&self, mut children: Vec<TestFruit>) -> Result<TestFruit> {
+    fn merge_fruits(&self, mut children: Vec<TestFruit>) -> crate::Result<TestFruit> {
         children.sort_by_key(|fruit| {
             if fruit.docs().is_empty() {
                 0
@@ -124,7 +124,7 @@ impl Collector for FastFieldTestCollector {
         &self,
         _: SegmentLocalId,
         segment_reader: &SegmentReader,
-    ) -> Result<FastFieldSegmentCollector> {
+    ) -> crate::Result<FastFieldSegmentCollector> {
         let reader = segment_reader
             .fast_fields()
             .u64(self.field)
@@ -139,7 +139,7 @@ impl Collector for FastFieldTestCollector {
         false
     }
 
-    fn merge_fruits(&self, children: Vec<Vec<u64>>) -> Result<Vec<u64>> {
+    fn merge_fruits(&self, children: Vec<Vec<u64>>) -> crate::Result<Vec<u64>> {
         Ok(children.into_iter().flat_map(|v| v.into_iter()).collect())
     }
 }
@@ -184,7 +184,7 @@ impl Collector for BytesFastFieldTestCollector {
         &self,
         _segment_local_id: u32,
         segment_reader: &SegmentReader,
-    ) -> Result<BytesFastFieldSegmentCollector> {
+    ) -> crate::Result<BytesFastFieldSegmentCollector> {
         Ok(BytesFastFieldSegmentCollector {
             vals: Vec::new(),
             reader: segment_reader
@@ -198,7 +198,7 @@ impl Collector for BytesFastFieldTestCollector {
         false
     }
 
-    fn merge_fruits(&self, children: Vec<Vec<u8>>) -> Result<Vec<u8>> {
+    fn merge_fruits(&self, children: Vec<Vec<u8>>) -> crate::Result<Vec<u8>> {
         Ok(children.into_iter().flat_map(|c| c.into_iter()).collect())
     }
 }
@@ -206,7 +206,7 @@ impl Collector for BytesFastFieldTestCollector {
 impl SegmentCollector for BytesFastFieldSegmentCollector {
     type Fruit = Vec<u8>;
 
-    fn collect(&mut self, doc: u32, _score: f32) {
+    fn collect(&mut self, doc: u32, _score: Score) {
         let data = self.reader.get_bytes(doc);
         self.vals.extend(data);
     }

src/collector/top_collector.rs

@@ -1,6 +1,5 @@
 use crate::DocAddress;
 use crate::DocId;
-use crate::Result;
 use crate::SegmentLocalId;
 use crate::SegmentReader;
 use serde::export::PhantomData;
@@ -19,9 +18,9 @@ use std::collections::BinaryHeap;
 /// Two elements are equal if their feature is equal, and regardless of whether `doc`
 /// is equal. This should be perfectly fine for this usage, but let's make sure this
 /// struct is never public.
-struct ComparableDoc<T, D> {
+pub(crate) struct ComparableDoc<T, D> {
-    feature: T,
+    pub feature: T,
-    doc: D,
+    pub doc: D,
 }
 
 impl<T: PartialOrd, D: PartialOrd> PartialOrd for ComparableDoc<T, D> {
@@ -57,7 +56,8 @@ impl<T: PartialOrd, D: PartialOrd> PartialEq for ComparableDoc<T, D> {
 impl<T: PartialOrd, D: PartialOrd> Eq for ComparableDoc<T, D> {}
 
 pub(crate) struct TopCollector<T> {
-    limit: usize,
+    pub limit: usize,
+    pub offset: usize,
     _marker: PhantomData<T>,
 }
 
@@ -73,27 +73,33 @@ where
         if limit < 1 {
             panic!("Limit must be strictly greater than 0.");
         }
-        TopCollector {
+        Self {
             limit,
+            offset: 0,
             _marker: PhantomData,
         }
     }
 
-    pub fn limit(&self) -> usize {
+    /// Skip the first "offset" documents when collecting.
-        self.limit
+    ///
+    /// This is equivalent to `OFFSET` in MySQL or PostgreSQL and `start` in
+    /// Lucene's TopDocsCollector.
+    pub fn and_offset(mut self, offset: usize) -> TopCollector<T> {
+        self.offset = offset;
+        self
     }
 
     pub fn merge_fruits(
         &self,
         children: Vec<Vec<(T, DocAddress)>>,
-    ) -> Result<Vec<(T, DocAddress)>> {
+    ) -> crate::Result<Vec<(T, DocAddress)>> {
         if self.limit == 0 {
             return Ok(Vec::new());
         }
         let mut top_collector = BinaryHeap::new();
         for child_fruit in children {
             for (feature, doc) in child_fruit {
-                if top_collector.len() < self.limit {
+                if top_collector.len() < (self.limit + self.offset) {
                     top_collector.push(ComparableDoc { feature, doc });
                 } else if let Some(mut head) = top_collector.peek_mut() {
                     if head.feature < feature {
@@ -105,6 +111,7 @@ where
         Ok(top_collector
             .into_sorted_vec()
             .into_iter()
+            .skip(self.offset)
             .map(|cdoc| (cdoc.feature, cdoc.doc))
             .collect())
     }
@@ -113,8 +120,24 @@ where
         &self,
         segment_id: SegmentLocalId,
         _: &SegmentReader,
-    ) -> Result<TopSegmentCollector<F>> {
+    ) -> crate::Result<TopSegmentCollector<F>> {
-        Ok(TopSegmentCollector::new(segment_id, self.limit))
+        Ok(TopSegmentCollector::new(
+            segment_id,
+            self.limit + self.offset,
+        ))
+    }
+
+    /// Create a new TopCollector with the same limit and offset.
+    ///
+    /// Ideally we would use Into but the blanket implementation seems to cause the Scorer traits
+    /// to fail.
+    #[doc(hidden)]
+    pub(crate) fn into_tscore<TScore: PartialOrd + Clone>(self) -> TopCollector<TScore> {
+        TopCollector {
+            limit: self.limit,
+            offset: self.offset,
+            _marker: PhantomData,
+        }
     }
 }
 
@@ -188,7 +211,7 @@ impl<T: PartialOrd + Clone> TopSegmentCollector<T> {
 
 #[cfg(test)]
 mod tests {
-    use super::TopSegmentCollector;
+    use super::{TopCollector, TopSegmentCollector};
     use crate::DocAddress;
 
     #[test]
@@ -249,6 +272,48 @@ mod tests {
             top_collector_limit_3.harvest()[..2].to_vec(),
         );
     }
+
+    #[test]
+    fn test_top_collector_with_limit_and_offset() {
+        let collector = TopCollector::with_limit(2).and_offset(1);
+
+        let results = collector
+            .merge_fruits(vec![vec![
+                (0.9, DocAddress(0, 1)),
+                (0.8, DocAddress(0, 2)),
+                (0.7, DocAddress(0, 3)),
+                (0.6, DocAddress(0, 4)),
+                (0.5, DocAddress(0, 5)),
+            ]])
+            .unwrap();
+
+        assert_eq!(
+            results,
+            vec![(0.8, DocAddress(0, 2)), (0.7, DocAddress(0, 3)),]
+        );
+    }
+
+    #[test]
+    fn test_top_collector_with_limit_larger_than_set_and_offset() {
+        let collector = TopCollector::with_limit(2).and_offset(1);
+
+        let results = collector
+            .merge_fruits(vec![vec![(0.9, DocAddress(0, 1)), (0.8, DocAddress(0, 2))]])
+            .unwrap();
+
+        assert_eq!(results, vec![(0.8, DocAddress(0, 2)),]);
+    }
+
+    #[test]
+    fn test_top_collector_with_limit_and_offset_larger_than_set() {
+        let collector = TopCollector::with_limit(2).and_offset(20);
+
+        let results = collector
+            .merge_fruits(vec![vec![(0.9, DocAddress(0, 1)), (0.8, DocAddress(0, 2))]])
+            .unwrap();
+
+        assert_eq!(results, vec![]);
+    }
 }
 
 #[cfg(all(test, feature = "unstable"))]

src/collector/top_score_collector.rs

@@ -1,18 +1,20 @@
 use super::Collector;
 use crate::collector::custom_score_top_collector::CustomScoreTopCollector;
-use crate::collector::top_collector::TopCollector;
 use crate::collector::top_collector::TopSegmentCollector;
+use crate::collector::top_collector::{ComparableDoc, TopCollector};
 use crate::collector::tweak_score_top_collector::TweakedScoreTopCollector;
 use crate::collector::{
     CustomScorer, CustomSegmentScorer, ScoreSegmentTweaker, ScoreTweaker, SegmentCollector,
 };
+use crate::fastfield::FastFieldReader;
+use crate::query::Weight;
 use crate::schema::Field;
 use crate::DocAddress;
 use crate::DocId;
-use crate::Result;
 use crate::Score;
 use crate::SegmentLocalId;
 use crate::SegmentReader;
+use std::collections::BinaryHeap;
 use std::fmt;
 
 /// The `TopDocs` collector keeps track of the top `K` documents
@@ -29,49 +31,70 @@ use std::fmt;
 /// use tantivy::collector::TopDocs;
 /// use tantivy::query::QueryParser;
 /// use tantivy::schema::{Schema, TEXT};
-/// use tantivy::{doc, DocAddress, Index, Result};
+/// use tantivy::{doc, DocAddress, Index};
 ///
-/// # fn main() { example().unwrap(); }
+/// let mut schema_builder = Schema::builder();
-/// fn example() -> Result<()> {
+/// let title = schema_builder.add_text_field("title", TEXT);
-///     let mut schema_builder = Schema::builder();
+/// let schema = schema_builder.build();
-///     let title = schema_builder.add_text_field("title", TEXT);
+/// let index = Index::create_in_ram(schema);
-///     let schema = schema_builder.build();
-///     let index = Index::create_in_ram(schema);
-///     {
-///         let mut index_writer = index.writer_with_num_threads(1, 3_000_000)?;
-///         index_writer.add_document(doc!(
-///             title => "The Name of the Wind",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of Muadib",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "A Dairy Cow",
-///         ));
-///         index_writer.add_document(doc!(
-///             title => "The Diary of a Young Girl",
-///         ));
-///         index_writer.commit().unwrap();
-///     }
 ///
-///     let reader = index.reader()?;
+/// let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-///     let searcher = reader.searcher();
+/// index_writer.add_document(doc!(title => "The Name of the Wind"));
+/// index_writer.add_document(doc!(title => "The Diary of Muadib"));
+/// index_writer.add_document(doc!(title => "A Dairy Cow"));
+/// index_writer.add_document(doc!(title => "The Diary of a Young Girl"));
+/// assert!(index_writer.commit().is_ok());
 ///
-///     let query_parser = QueryParser::for_index(&index, vec![title]);
+/// let reader = index.reader().unwrap();
-///     let query = query_parser.parse_query("diary")?;
+/// let searcher = reader.searcher();
-///     let top_docs = searcher.search(&query, &TopDocs::with_limit(2))?;
 ///
-///     assert_eq!(&top_docs[0], &(0.7261542, DocAddress(0, 1)));
+/// let query_parser = QueryParser::for_index(&index, vec![title]);
-///     assert_eq!(&top_docs[1], &(0.6099695, DocAddress(0, 3)));
+/// let query = query_parser.parse_query("diary").unwrap();
+/// let top_docs = searcher.search(&query, &TopDocs::with_limit(2)).unwrap();
 ///
-///     Ok(())
+/// assert_eq!(top_docs[0].1, DocAddress(0, 1));
-/// }
+/// assert_eq!(top_docs[1].1, DocAddress(0, 3));
 /// ```
 pub struct TopDocs(TopCollector<Score>);
 
 impl fmt::Debug for TopDocs {
     fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        write!(f, "TopDocs({})", self.0.limit())
+        write!(
+            f,
+            "TopDocs(limit={}, offset={})",
+            self.0.limit, self.0.offset
+        )
+    }
+}
+
+struct ScorerByFastFieldReader {
+    ff_reader: FastFieldReader<u64>,
+}
+
+impl CustomSegmentScorer<u64> for ScorerByFastFieldReader {
+    fn score(&mut self, doc: DocId) -> u64 {
+        self.ff_reader.get_u64(u64::from(doc))
+    }
+}
+
+struct ScorerByField {
+    field: Field,
+}
+
+impl CustomScorer<u64> for ScorerByField {
+    type Child = ScorerByFastFieldReader;
+
+    fn segment_scorer(&self, segment_reader: &SegmentReader) -> crate::Result<Self::Child> {
+        let ff_reader = segment_reader
+            .fast_fields()
+            .u64(self.field)
+            .ok_or_else(|| {
+                crate::TantivyError::SchemaError(format!(
+                    "Field requested ({:?}) is not a i64/u64 fast field.",
+                    self.field
+                ))
+            })?;
+        Ok(ScorerByFastFieldReader { ff_reader })
     }
 }
 
@@ -84,11 +107,50 @@ impl TopDocs {
         TopDocs(TopCollector::with_limit(limit))
     }
 
+    /// Skip the first "offset" documents when collecting.
+    ///
+    /// This is equivalent to `OFFSET` in MySQL or PostgreSQL and `start` in
+    /// Lucene's TopDocsCollector.
+    ///
+    /// ```rust
+    /// use tantivy::collector::TopDocs;
+    /// use tantivy::query::QueryParser;
+    /// use tantivy::schema::{Schema, TEXT};
+    /// use tantivy::{doc, DocAddress, Index};
+    ///
+    /// let mut schema_builder = Schema::builder();
+    /// let title = schema_builder.add_text_field("title", TEXT);
+    /// let schema = schema_builder.build();
+    /// let index = Index::create_in_ram(schema);
+    ///
+    /// let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+    /// index_writer.add_document(doc!(title => "The Name of the Wind"));
+    /// index_writer.add_document(doc!(title => "The Diary of Muadib"));
+    /// index_writer.add_document(doc!(title => "A Dairy Cow"));
+    /// index_writer.add_document(doc!(title => "The Diary of a Young Girl"));
+    /// index_writer.add_document(doc!(title => "The Diary of Lena Mukhina"));
+    /// assert!(index_writer.commit().is_ok());
+    ///
+    /// let reader = index.reader().unwrap();
+    /// let searcher = reader.searcher();
+    ///
+    /// let query_parser = QueryParser::for_index(&index, vec![title]);
+    /// let query = query_parser.parse_query("diary").unwrap();
+    /// let top_docs = searcher.search(&query, &TopDocs::with_limit(2).and_offset(1)).unwrap();
+    ///
+    /// assert_eq!(top_docs.len(), 2);
+    /// assert_eq!(top_docs[0].1, DocAddress(0, 4));
+    /// assert_eq!(top_docs[1].1, DocAddress(0, 3));
+    /// ```
+    pub fn and_offset(self, offset: usize) -> TopDocs {
+        TopDocs(self.0.and_offset(offset))
+    }
+
     /// Set top-K to rank documents by a given fast field.
     ///
     /// ```rust
     /// # use tantivy::schema::{Schema, FAST, TEXT};
-    /// # use tantivy::{doc, Index, Result, DocAddress};
+    /// # use tantivy::{doc, Index, DocAddress};
     /// # use tantivy::query::{Query, QueryParser};
     /// use tantivy::Searcher;
     /// use tantivy::collector::TopDocs;
@@ -102,15 +164,12 @@ impl TopDocs {
     /// #  
     /// #   let index = Index::create_in_ram(schema);
     /// #   let mut index_writer = index.writer_with_num_threads(1, 3_000_000)?;
-    /// #   index_writer.add_document(doc!(
+    /// #   index_writer.add_document(doc!(title => "The Name of the Wind", rating => 92u64));
-    /// #       title => "The Name of the Wind",
-    /// #       rating => 92u64,
-    /// #   ));
     /// #   index_writer.add_document(doc!(title => "The Diary of Muadib", rating => 97u64));
     /// #   index_writer.add_document(doc!(title => "A Dairy Cow", rating => 63u64));
     /// #   index_writer.add_document(doc!(title => "The Diary of a Young Girl", rating => 80u64));
-    /// #   index_writer.commit()?;
+    /// #   assert!(index_writer.commit().is_ok());
-    /// #   let reader = index.reader()?;
+    /// #   let reader = index.reader().unwrap();
     /// #   let query = QueryParser::for_index(&index, vec![title]).parse_query("diary")?;
     /// #   let top_docs = docs_sorted_by_rating(&reader.searcher(), &query, rating)?;
     /// #   assert_eq!(top_docs,
@@ -128,7 +187,7 @@ impl TopDocs {
     /// fn docs_sorted_by_rating(searcher: &Searcher,
     ///                          query: &dyn Query,
     ///                          sort_by_field: Field)
-    ///     -> Result<Vec<(u64, DocAddress)>> {
+    ///     -> tantivy::Result<Vec<(u64, DocAddress)>> {
     ///
     ///     // This is where we build our topdocs collector
     ///     //
@@ -160,14 +219,7 @@ impl TopDocs {
         self,
         field: Field,
     ) -> impl Collector<Fruit = Vec<(u64, DocAddress)>> {
-        self.custom_score(move |segment_reader: &SegmentReader| {
+        self.custom_score(ScorerByField { field })
-            let ff_reader = segment_reader
-                .fast_fields()
-                .u64(field)
-                .expect("Field requested is not a i64/u64 fast field.");
-            //TODO error message missmatch actual behavior for i64
-            move |doc: DocId| ff_reader.get(doc)
-        })
     }
 
     /// Ranks the documents using a custom score.
@@ -202,27 +254,33 @@ impl TopDocs {
     /// use tantivy::collector::TopDocs;
     /// use tantivy::schema::Field;
     ///
-    /// # fn create_schema() -> Schema {
+    /// fn create_schema() -> Schema {
-    /// #    let mut schema_builder = Schema::builder();
+    ///    let mut schema_builder = Schema::builder();
-    /// #    schema_builder.add_text_field("product_name", TEXT);
+    ///    schema_builder.add_text_field("product_name", TEXT);
-    /// #    schema_builder.add_u64_field("popularity", FAST);
+    ///    schema_builder.add_u64_field("popularity", FAST);
-    /// #    schema_builder.build()
+    ///    schema_builder.build()
-    /// # }
+    /// }
-    /// #
+    ///
-    /// # fn main() -> tantivy::Result<()> {
+    /// fn create_index() -> tantivy::Result<Index> {
-    /// #   let schema = create_schema();
+    ///   let schema = create_schema();
-    /// #   let index = Index::create_in_ram(schema);
+    ///   let index = Index::create_in_ram(schema);
-    /// #   let mut index_writer = index.writer_with_num_threads(1, 3_000_000)?;
+    ///   let mut index_writer = index.writer_with_num_threads(1, 3_000_000)?;
-    /// #   let product_name = index.schema().get_field("product_name").unwrap();
+    ///   let product_name = index.schema().get_field("product_name").unwrap();
-    /// #   
+    ///   let popularity: Field = index.schema().get_field("popularity").unwrap();
+    ///   index_writer.add_document(doc!(product_name => "The Diary of Muadib", popularity => 1u64));
+    ///   index_writer.add_document(doc!(product_name => "A Dairy Cow", popularity => 10u64));
+    ///   index_writer.add_document(doc!(product_name => "The Diary of a Young Girl", popularity => 15u64));
+    ///   index_writer.commit()?;
+    ///   Ok(index)
+    /// }
+    ///
+    /// let index = create_index().unwrap();
+    /// let product_name = index.schema().get_field("product_name").unwrap();
     /// let popularity: Field = index.schema().get_field("popularity").unwrap();
-    /// #   index_writer.add_document(doc!(product_name => "The Diary of Muadib", popularity => 1u64));
+    ///
-    /// #   index_writer.add_document(doc!(product_name => "A Dairy Cow", popularity => 10u64));
+    /// let user_query_str = "diary";
-    /// #   index_writer.add_document(doc!(product_name => "The Diary of a Young Girl", popularity => 15u64));
+    /// let query_parser = QueryParser::for_index(&index, vec![product_name]);
-    /// #   index_writer.commit()?;
+    /// let query = query_parser.parse_query(user_query_str).unwrap();
-    /// // ...
-    /// # let user_query = "diary";
-    /// # let query = QueryParser::for_index(&index, vec![product_name]).parse_query(user_query)?;
     ///
     /// // This is where we build our collector with our custom score.
     /// let top_docs_by_custom_score = TopDocs
@@ -245,19 +303,16 @@ impl TopDocs {
     ///                 let popularity: u64 = popularity_reader.get(doc);
     ///                 // Well.. For the sake of the example we use a simple logarithm
     ///                 // function.
-    ///                 let popularity_boost_score = ((2u64 + popularity) as f32).log2();
+    ///                 let popularity_boost_score = ((2u64 + popularity) as Score).log2();
     ///                 popularity_boost_score * original_score
     ///             }
     ///           });
-    /// # let reader = index.reader()?;
+    /// let reader = index.reader().unwrap();
-    /// # let searcher = reader.searcher();
+    /// let searcher = reader.searcher();
     /// // ... and here are our documents. Note this is a simple vec.
     /// // The `Score` in the pair is our tweaked score.
     /// let resulting_docs: Vec<(Score, DocAddress)> =
-    ///      searcher.search(&*query, &top_docs_by_custom_score)?;
+    ///      searcher.search(&query, &top_docs_by_custom_score).unwrap();
-    ///
-    /// # Ok(())
-    /// # }
     /// ```
     ///
     /// # See also
@@ -269,9 +324,9 @@ impl TopDocs {
     where
         TScore: 'static + Send + Sync + Clone + PartialOrd,
         TScoreSegmentTweaker: ScoreSegmentTweaker<TScore> + 'static,
-        TScoreTweaker: ScoreTweaker<TScore, Child = TScoreSegmentTweaker>,
+        TScoreTweaker: ScoreTweaker<TScore, Child = TScoreSegmentTweaker> + Send + Sync,
     {
-        TweakedScoreTopCollector::new(score_tweaker, self.0.limit())
+        TweakedScoreTopCollector::new(score_tweaker, self.0.into_tscore())
     }
 
     /// Ranks the documents using a custom score.
@@ -383,9 +438,9 @@ impl TopDocs {
     where
         TScore: 'static + Send + Sync + Clone + PartialOrd,
         TCustomSegmentScorer: CustomSegmentScorer<TScore> + 'static,
-        TCustomScorer: CustomScorer<TScore, Child = TCustomSegmentScorer>,
+        TCustomScorer: CustomScorer<TScore, Child = TCustomSegmentScorer> + Send + Sync,
     {
-        CustomScoreTopCollector::new(custom_score, self.0.limit())
+        CustomScoreTopCollector::new(custom_score, self.0.into_tscore())
     }
 }
 
@@ -398,7 +453,7 @@ impl Collector for TopDocs {
         &self,
         segment_local_id: SegmentLocalId,
         reader: &SegmentReader,
-    ) -> Result<Self::Child> {
+    ) -> crate::Result<Self::Child> {
         let collector = self.0.for_segment(segment_local_id, reader)?;
         Ok(TopScoreSegmentCollector(collector))
     }
@@ -407,9 +462,70 @@ impl Collector for TopDocs {
         true
     }
 
-    fn merge_fruits(&self, child_fruits: Vec<Vec<(Score, DocAddress)>>) -> Result<Self::Fruit> {
+    fn merge_fruits(
+        &self,
+        child_fruits: Vec<Vec<(Score, DocAddress)>>,
+    ) -> crate::Result<Self::Fruit> {
         self.0.merge_fruits(child_fruits)
     }
+
+    fn collect_segment(
+        &self,
+        weight: &dyn Weight,
+        segment_ord: u32,
+        reader: &SegmentReader,
+    ) -> crate::Result<<Self::Child as SegmentCollector>::Fruit> {
+        let heap_len = self.0.limit + self.0.offset;
+        let mut heap: BinaryHeap<ComparableDoc<Score, DocId>> = BinaryHeap::with_capacity(heap_len);
+
+        if let Some(delete_bitset) = reader.delete_bitset() {
+            let mut threshold = Score::MIN;
+            weight.for_each_pruning(threshold, reader, &mut |doc, score| {
+                if delete_bitset.is_deleted(doc) {
+                    return threshold;
+                }
+                let heap_item = ComparableDoc {
+                    feature: score,
+                    doc,
+                };
+                if heap.len() < heap_len {
+                    heap.push(heap_item);
+                    if heap.len() == heap_len {
+                        threshold = heap.peek().map(|el| el.feature).unwrap_or(Score::MIN);
+                    }
+                    return threshold;
+                }
+                *heap.peek_mut().unwrap() = heap_item;
+                threshold = heap.peek().map(|el| el.feature).unwrap_or(Score::MIN);
+                threshold
+            })?;
+        } else {
+            weight.for_each_pruning(Score::MIN, reader, &mut |doc, score| {
+                let heap_item = ComparableDoc {
+                    feature: score,
+                    doc,
+                };
+                if heap.len() < heap_len {
+                    heap.push(heap_item);
+                    // TODO the threshold is suboptimal for heap.len == heap_len
+                    if heap.len() == heap_len {
+                        return heap.peek().map(|el| el.feature).unwrap_or(Score::MIN);
+                    } else {
+                        return Score::MIN;
+                    }
+                }
+                *heap.peek_mut().unwrap() = heap_item;
+                heap.peek().map(|el| el.feature).unwrap_or(Score::MIN)
+            })?;
+        }
+
+        let fruit = heap
+            .into_sorted_vec()
+            .into_iter()
+            .map(|cid| (cid.feature, DocAddress(segment_ord, cid.doc)))
+            .collect();
+        Ok(fruit)
+    }
 }
 
 /// Segment Collector associated to `TopDocs`.
@@ -419,7 +535,7 @@ impl SegmentCollector for TopScoreSegmentCollector {
     type Fruit = Vec<(Score, DocAddress)>;
 
     fn collect(&mut self, doc: DocId, score: Score) {
-        self.0.collect(doc, score)
+        self.0.collect(doc, score);
     }
 
     fn harvest(self) -> Vec<(Score, DocAddress)> {
@@ -433,11 +549,10 @@ mod tests {
     use crate::collector::Collector;
     use crate::query::{AllQuery, Query, QueryParser};
     use crate::schema::{Field, Schema, FAST, STORED, TEXT};
-    use crate::DocAddress;
     use crate::Index;
     use crate::IndexWriter;
     use crate::Score;
-    use itertools::Itertools;
+    use crate::{DocAddress, DocId, SegmentReader};
 
     fn make_index() -> Index {
         let mut schema_builder = Schema::builder();
@@ -455,6 +570,13 @@ mod tests {
         index
     }
 
+    fn assert_results_equals(results: &[(Score, DocAddress)], expected: &[(Score, DocAddress)]) {
+        for (result, expected) in results.iter().zip(expected.iter()) {
+            assert_eq!(result.1, expected.1);
+            crate::assert_nearly_equals!(result.0, expected.0);
+        }
+    }
+
     #[test]
     fn test_top_collector_not_at_capacity() {
         let index = make_index();
@@ -467,16 +589,31 @@ mod tests {
             .searcher()
             .search(&text_query, &TopDocs::with_limit(4))
             .unwrap();
-        assert_eq!(
+        assert_results_equals(
-            score_docs,
+            &score_docs,
-            vec![
+            &[
                 (0.81221175, DocAddress(0u32, 1)),
                 (0.5376842, DocAddress(0u32, 2)),
-                (0.48527452, DocAddress(0, 0))
+                (0.48527452, DocAddress(0, 0)),
-            ]
+            ],
         );
     }
 
+    #[test]
+    fn test_top_collector_not_at_capacity_with_offset() {
+        let index = make_index();
+        let field = index.schema().get_field("text").unwrap();
+        let query_parser = QueryParser::for_index(&index, vec![field]);
+        let text_query = query_parser.parse_query("droopy tax").unwrap();
+        let score_docs: Vec<(Score, DocAddress)> = index
+            .reader()
+            .unwrap()
+            .searcher()
+            .search(&text_query, &TopDocs::with_limit(4).and_offset(2))
+            .unwrap();
+        assert_results_equals(&score_docs[..], &[(0.48527452, DocAddress(0, 0))]);
+    }
+
     #[test]
     fn test_top_collector_at_capacity() {
         let index = make_index();
@@ -489,12 +626,33 @@ mod tests {
             .searcher()
             .search(&text_query, &TopDocs::with_limit(2))
             .unwrap();
-        assert_eq!(
+        assert_results_equals(
-            score_docs,
+            &score_docs,
-            vec![
+            &[
                 (0.81221175, DocAddress(0u32, 1)),
                 (0.5376842, DocAddress(0u32, 2)),
-            ]
+            ],
+        );
+    }
+
+    #[test]
+    fn test_top_collector_at_capacity_with_offset() {
+        let index = make_index();
+        let field = index.schema().get_field("text").unwrap();
+        let query_parser = QueryParser::for_index(&index, vec![field]);
+        let text_query = query_parser.parse_query("droopy tax").unwrap();
+        let score_docs: Vec<(Score, DocAddress)> = index
+            .reader()
+            .unwrap()
+            .searcher()
+            .search(&text_query, &TopDocs::with_limit(2).and_offset(1))
+            .unwrap();
+        assert_results_equals(
+            &score_docs[..],
+            &[
+                (0.5376842, DocAddress(0u32, 2)),
+                (0.48527452, DocAddress(0, 0)),
+            ],
         );
     }
 
@@ -511,8 +669,8 @@ mod tests {
 
         // precondition for the test to be meaningful: we did get documents
         // with the same score
-        assert!(page_1.iter().map(|result| result.0).all_equal());
+        assert!(page_1.iter().all(|result| result.0 == page_1[0].0));
-        assert!(page_2.iter().map(|result| result.0).all_equal());
+        assert!(page_2.iter().all(|result| result.0 == page_2[0].0));
 
         // sanity check since we're relying on make_index()
         assert_eq!(page_1.len(), 2);
@@ -555,8 +713,8 @@ mod tests {
         let top_collector = TopDocs::with_limit(4).order_by_u64_field(size);
         let top_docs: Vec<(u64, DocAddress)> = searcher.search(&query, &top_collector).unwrap();
         assert_eq!(
-            top_docs,
+            &top_docs[..],
-            vec![
+            &[
                 (64, DocAddress(0, 1)),
                 (16, DocAddress(0, 2)),
                 (12, DocAddress(0, 0))
@@ -586,7 +744,6 @@ mod tests {
     }
 
     #[test]
-    #[should_panic(expected = "Field requested is not a i64/u64 fast field")]
     fn test_field_not_fast_field() {
         let mut schema_builder = Schema::builder();
         let title = schema_builder.add_text_field(TITLE, TEXT);
@@ -601,7 +758,59 @@ mod tests {
         let searcher = index.reader().unwrap().searcher();
         let segment = searcher.segment_reader(0);
         let top_collector = TopDocs::with_limit(4).order_by_u64_field(size);
-        assert!(top_collector.for_segment(0, segment).is_ok());
+        let err = top_collector.for_segment(0, segment);
+        if let Err(crate::TantivyError::SchemaError(msg)) = err {
+            assert_eq!(
+                msg,
+                "Field requested (Field(1)) is not a i64/u64 fast field."
+            );
+        } else {
+            assert!(false);
+        }
+    }
+
+    #[test]
+    fn test_tweak_score_top_collector_with_offset() {
+        let index = make_index();
+        let field = index.schema().get_field("text").unwrap();
+        let query_parser = QueryParser::for_index(&index, vec![field]);
+        let text_query = query_parser.parse_query("droopy tax").unwrap();
+        let collector = TopDocs::with_limit(2).and_offset(1).tweak_score(
+            move |_segment_reader: &SegmentReader| move |doc: DocId, _original_score: Score| doc,
+        );
+        let score_docs: Vec<(u32, DocAddress)> = index
+            .reader()
+            .unwrap()
+            .searcher()
+            .search(&text_query, &collector)
+            .unwrap();
+
+        assert_eq!(
+            score_docs,
+            vec![(1, DocAddress(0, 1)), (0, DocAddress(0, 0)),]
+        );
+    }
+
+    #[test]
+    fn test_custom_score_top_collector_with_offset() {
+        let index = make_index();
+        let field = index.schema().get_field("text").unwrap();
+        let query_parser = QueryParser::for_index(&index, vec![field]);
+        let text_query = query_parser.parse_query("droopy tax").unwrap();
+        let collector = TopDocs::with_limit(2)
+            .and_offset(1)
+            .custom_score(move |_segment_reader: &SegmentReader| move |doc: DocId| doc);
+        let score_docs: Vec<(u32, DocAddress)> = index
+            .reader()
+            .unwrap()
+            .searcher()
+            .search(&text_query, &collector)
+            .unwrap();
+
+        assert_eq!(
+            score_docs,
+            vec![(1, DocAddress(0, 1)), (0, DocAddress(0, 0)),]
+        );
     }
 
     fn index(

src/collector/tweak_score_top_collector.rs

@@ -14,11 +14,11 @@ where
 {
     pub fn new(
         score_tweaker: TScoreTweaker,
-        limit: usize,
+        collector: TopCollector<TScore>,
     ) -> TweakedScoreTopCollector<TScoreTweaker, TScore> {
         TweakedScoreTopCollector {
             score_tweaker,
-            collector: TopCollector::with_limit(limit),
+            collector,
         }
     }
 }
@@ -29,7 +29,7 @@ where
 /// It is the segment local version of the [`ScoreTweaker`](./trait.ScoreTweaker.html).
 pub trait ScoreSegmentTweaker<TScore>: 'static {
     /// Tweak the given `score` for the document `doc`.
-    fn score(&self, doc: DocId, score: Score) -> TScore;
+    fn score(&mut self, doc: DocId, score: Score) -> TScore;
 }
 
 /// `ScoreTweaker` makes it possible to tweak the score
@@ -49,7 +49,7 @@ pub trait ScoreTweaker<TScore>: Sync {
 
 impl<TScoreTweaker, TScore> Collector for TweakedScoreTopCollector<TScoreTweaker, TScore>
 where
-    TScoreTweaker: ScoreTweaker<TScore>,
+    TScoreTweaker: ScoreTweaker<TScore> + Send + Sync,
     TScore: 'static + PartialOrd + Clone + Send + Sync,
 {
     type Fruit = Vec<(TScore, DocAddress)>;
@@ -121,9 +121,9 @@ where
 
 impl<F, TScore> ScoreSegmentTweaker<TScore> for F
 where
-    F: 'static + Sync + Send + Fn(DocId, Score) -> TScore,
+    F: 'static + FnMut(DocId, Score) -> TScore,
 {
-    fn score(&self, doc: DocId, score: Score) -> TScore {
+    fn score(&mut self, doc: DocId, score: Score) -> TScore {
         (self)(doc, score)
     }
 }

src/common/bitset.rs

@@ -33,6 +33,10 @@ impl TinySet {
         TinySet(0u64)
     }
 
+    pub fn clear(&mut self) {
+        self.0 = 0u64;
+    }
+
     /// Returns the complement of the set in `[0, 64[`.
     fn complement(self) -> TinySet {
         TinySet(!self.0)
@@ -43,6 +47,11 @@ impl TinySet {
         !self.intersect(TinySet::singleton(el)).is_empty()
     }
 
+    /// Returns the number of elements in the TinySet.
+    pub fn len(self) -> u32 {
+        self.0.count_ones()
+    }
+
     /// Returns the intersection of `self` and `other`
     pub fn intersect(self, other: TinySet) -> TinySet {
         TinySet(self.0 & other.0)
@@ -109,22 +118,12 @@ impl TinySet {
     pub fn range_greater_or_equal(from_included: u32) -> TinySet {
         TinySet::range_lower(from_included).complement()
     }
-
-    pub fn clear(&mut self) {
-        self.0 = 0u64;
-    }
-
-    pub fn len(self) -> u32 {
-        self.0.count_ones()
-    }
 }
 
 #[derive(Clone)]
 pub struct BitSet {
     tinysets: Box<[TinySet]>,
-    len: usize, //< Technically it should be u32, but we
+    len: usize,
-    // count multiple inserts.
-    // `usize` guards us from overflow.
     max_value: u32,
 }
 
@@ -204,7 +203,7 @@ mod tests {
 
     use super::BitSet;
     use super::TinySet;
-    use crate::docset::DocSet;
+    use crate::docset::{DocSet, TERMINATED};
     use crate::query::BitSetDocSet;
     use crate::tests;
     use crate::tests::generate_nonunique_unsorted;
@@ -278,11 +277,13 @@ mod tests {
         }
         assert_eq!(btreeset.len(), bitset.len());
         let mut bitset_docset = BitSetDocSet::from(bitset);
+        let mut remaining = true;
         for el in btreeset.into_iter() {
-            bitset_docset.advance();
+            assert!(remaining);
             assert_eq!(bitset_docset.doc(), el);
+            remaining = bitset_docset.advance() != TERMINATED;
         }
-        assert!(!bitset_docset.advance());
+        assert!(!remaining);
     }
 
     #[test]

src/common/mod.rs

@@ -10,7 +10,9 @@ pub(crate) use self::bitset::TinySet;
 pub(crate) use self::composite_file::{CompositeFile, CompositeWrite};
 pub use self::counting_writer::CountingWriter;
 pub use self::serialize::{BinarySerializable, FixedSize};
-pub use self::vint::{read_u32_vint, serialize_vint_u32, write_u32_vint, VInt};
+pub use self::vint::{
+    read_u32_vint, read_u32_vint_no_advance, serialize_vint_u32, write_u32_vint, VInt,
+};
 pub use byteorder::LittleEndian as Endianness;
 
 /// Segment's max doc must be `< MAX_DOC_LIMIT`.
@@ -18,6 +20,19 @@ pub use byteorder::LittleEndian as Endianness;
 /// We do not allow segments with more than
 pub const MAX_DOC_LIMIT: u32 = 1 << 31;
 
+pub fn minmax<I, T>(mut vals: I) -> Option<(T, T)>
+where
+    I: Iterator<Item = T>,
+    T: Copy + Ord,
+{
+    if let Some(first_el) = vals.next() {
+        return Some(vals.fold((first_el, first_el), |(min_val, max_val), el| {
+            (min_val.min(el), max_val.max(el))
+        }));
+    }
+    None
+}
+
 /// Computes the number of bits that will be used for bitpacking.
 ///
 /// In general the target is the minimum number of bits
@@ -134,6 +149,7 @@ pub fn u64_to_f64(val: u64) -> f64 {
 #[cfg(test)]
 pub(crate) mod test {
 
+    pub use super::minmax;
     pub use super::serialize::test::fixed_size_test;
     use super::{compute_num_bits, f64_to_u64, i64_to_u64, u64_to_f64, u64_to_i64};
     use std::f64;
@@ -199,4 +215,21 @@ pub(crate) mod test {
         assert!(((super::MAX_DOC_LIMIT - 1) as i32) >= 0);
         assert!((super::MAX_DOC_LIMIT as i32) < 0);
     }
+
+    #[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)));
+    }
 }

src/common/serialize.rs

@@ -89,6 +89,19 @@ impl FixedSize for u64 {
     const SIZE_IN_BYTES: usize = 8;
 }
 
+impl BinarySerializable for f32 {
+    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
+        writer.write_f32::<Endianness>(*self)
+    }
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
+        reader.read_f32::<Endianness>()
+    }
+}
+
+impl FixedSize for f32 {
+    const SIZE_IN_BYTES: usize = 4;
+}
+
 impl BinarySerializable for i64 {
     fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
         writer.write_i64::<Endianness>(*self)

src/common/vint.rs

@@ -5,12 +5,12 @@ use std::io::Read;
 use std::io::Write;
 
 ///   Wrapper over a `u64` that serializes as a variable int.
-#[derive(Debug, Eq, PartialEq)]
+#[derive(Clone, Copy, Debug, Eq, PartialEq)]
 pub struct VInt(pub u64);
 
 const STOP_BIT: u8 = 128;
 
-pub fn serialize_vint_u32(val: u32) -> (u64, usize) {
+pub fn serialize_vint_u32(val: u32, buf: &mut [u8; 8]) -> &[u8] {
     const START_2: u64 = 1 << 7;
     const START_3: u64 = 1 << 14;
     const START_4: u64 = 1 << 21;
@@ -29,7 +29,7 @@ pub fn serialize_vint_u32(val: u32) -> (u64, usize) {
 
     let val = u64::from(val);
     const STOP_BIT: u64 = 128u64;
-    match val {
+    let (res, num_bytes) = match val {
         0..=STOP_1 => (val | STOP_BIT, 1),
         START_2..=STOP_2 => (
             (val & MASK_1) | ((val & MASK_2) << 1) | (STOP_BIT << (8)),
@@ -56,7 +56,9 @@ pub fn serialize_vint_u32(val: u32) -> (u64, usize) {
                 | (STOP_BIT << (8 * 4)),
             5,
         ),
-    }
+    };
+    LittleEndian::write_u64(&mut buf[..], res);
+    &buf[0..num_bytes]
 }
 
 /// Returns the number of bytes covered by a
@@ -85,23 +87,26 @@ 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 vlen = vint_len(*data);
+    let (result, vlen) = read_u32_vint_no_advance(*data);
+    *data = &data[vlen..];
+    result
+}
+
+pub fn read_u32_vint_no_advance(data: &[u8]) -> (u32, usize) {
+    let vlen = vint_len(data);
     let mut result = 0u32;
     let mut shift = 0u64;
     for &b in &data[..vlen] {
         result |= u32::from(b & 127u8) << shift;
         shift += 7;
     }
-    *data = &data[vlen..];
+    (result, vlen)
-    result
 }
-
 /// Write a `u32` as a vint payload.
 pub fn write_u32_vint<W: io::Write>(val: u32, writer: &mut W) -> io::Result<()> {
-    let (val, num_bytes) = serialize_vint_u32(val);
+    let mut buf = [0u8; 8];
-    let mut buffer = [0u8; 8];
+    let data = serialize_vint_u32(val, &mut buf);
-    LittleEndian::write_u64(&mut buffer, val);
+    writer.write_all(&data)
-    writer.write_all(&buffer[..num_bytes])
 }
 
 impl VInt {
@@ -172,7 +177,6 @@ mod tests {
     use super::serialize_vint_u32;
     use super::VInt;
     use crate::common::BinarySerializable;
-    use byteorder::{ByteOrder, LittleEndian};
 
     fn aux_test_vint(val: u64) {
         let mut v = [14u8; 10];
@@ -208,12 +212,10 @@ mod tests {
 
     fn aux_test_serialize_vint_u32(val: u32) {
         let mut buffer = [0u8; 10];
-        let mut buffer2 = [0u8; 10];
+        let mut buffer2 = [0u8; 8];
         let len_vint = VInt(val as u64).serialize_into(&mut buffer);
-        let (vint, len) = serialize_vint_u32(val);
+        let res2 = serialize_vint_u32(val, &mut buffer2);
-        assert_eq!(len, len_vint, "len wrong for val {}", val);
+        assert_eq!(&buffer[..len_vint], res2, "array wrong for {}", val);
-        LittleEndian::write_u64(&mut buffer2, vint);
-        assert_eq!(&buffer[..len], &buffer2[..len], "array wrong for {}", val);
     }
 
     #[test]

src/core/executor.rs

@@ -1,4 +1,3 @@
-use crate::Result;
 use crossbeam::channel;
 use rayon::{ThreadPool, ThreadPoolBuilder};
 
@@ -10,7 +9,9 @@ use rayon::{ThreadPool, ThreadPoolBuilder};
 /// API of a dependency, knowing it might conflict with a different version
 /// used by the client. Second, we may stop using rayon in the future.
 pub enum Executor {
+    /// Single thread variant of an Executor
     SingleThread,
+    /// Thread pool variant of an Executor
     ThreadPool(ThreadPool),
 }
 
@@ -20,8 +21,8 @@ impl Executor {
         Executor::SingleThread
     }
 
-    // Creates an Executor that dispatches the tasks in a thread pool.
+    /// Creates an Executor that dispatches the tasks in a thread pool.
-    pub fn multi_thread(num_threads: usize, prefix: &'static str) -> Result<Executor> {
+    pub fn multi_thread(num_threads: usize, prefix: &'static str) -> crate::Result<Executor> {
         let pool = ThreadPoolBuilder::new()
             .num_threads(num_threads)
             .thread_name(move |num| format!("{}{}", prefix, num))
@@ -29,22 +30,22 @@ impl Executor {
         Ok(Executor::ThreadPool(pool))
     }
 
-    // Perform a map in the thread pool.
+    /// Perform a map in the thread pool.
-    //
+    ///
-    // Regardless of the executor (`SingleThread` or `ThreadPool`), panics in the task
+    /// Regardless of the executor (`SingleThread` or `ThreadPool`), panics in the task
-    // will propagate to the caller.
+    /// will propagate to the caller.
     pub fn map<
         A: Send,
         R: Send,
         AIterator: Iterator<Item = A>,
-        F: Sized + Sync + Fn(A) -> Result<R>,
+        F: Sized + Sync + Fn(A) -> crate::Result<R>,
     >(
         &self,
         f: F,
         args: AIterator,
-    ) -> Result<Vec<R>> {
+    ) -> crate::Result<Vec<R>> {
         match self {
-            Executor::SingleThread => args.map(f).collect::<Result<_>>(),
+            Executor::SingleThread => args.map(f).collect::<crate::Result<_>>(),
             Executor::ThreadPool(pool) => {
                 let args_with_indices: Vec<(usize, A)> = args.enumerate().collect();
                 let num_fruits = args_with_indices.len();

src/core/index.rs

@@ -1,4 +1,3 @@
-use super::segment::create_segment;
 use super::segment::Segment;
 use crate::core::Executor;
 use crate::core::IndexMeta;
@@ -20,19 +19,21 @@ use crate::reader::IndexReaderBuilder;
 use crate::schema::Field;
 use crate::schema::FieldType;
 use crate::schema::Schema;
-use crate::tokenizer::BoxedTokenizer;
+use crate::tokenizer::{TextAnalyzer, TokenizerManager};
-use crate::tokenizer::TokenizerManager;
 use crate::IndexWriter;
-use crate::Result;
-use num_cpus;
 use std::borrow::BorrowMut;
 use std::collections::HashSet;
 use std::fmt;
+
 #[cfg(feature = "mmap")]
-use std::path::{Path, PathBuf};
+use std::path::Path;
+use std::path::PathBuf;
 use std::sync::Arc;
 
-fn load_metas(directory: &dyn Directory, inventory: &SegmentMetaInventory) -> Result<IndexMeta> {
+fn load_metas(
+    directory: &dyn Directory,
+    inventory: &SegmentMetaInventory,
+) -> crate::Result<IndexMeta> {
     let meta_data = directory.atomic_read(&META_FILEPATH)?;
     let meta_string = String::from_utf8_lossy(&meta_data);
     IndexMeta::deserialize(&meta_string, &inventory)
@@ -73,14 +74,14 @@ impl Index {
 
     /// Replace the default single thread search executor pool
     /// by a thread pool with a given number of threads.
-    pub fn set_multithread_executor(&mut self, num_threads: usize) -> Result<()> {
+    pub fn set_multithread_executor(&mut self, num_threads: usize) -> crate::Result<()> {
         self.executor = Arc::new(Executor::multi_thread(num_threads, "thrd-tantivy-search-")?);
         Ok(())
     }
 
     /// Replace the default single thread search executor pool
     /// by a thread pool with a given number of threads.
-    pub fn set_default_multithread_executor(&mut self) -> Result<()> {
+    pub fn set_default_multithread_executor(&mut self) -> crate::Result<()> {
         let default_num_threads = num_cpus::get();
         self.set_multithread_executor(default_num_threads)
     }
@@ -99,28 +100,29 @@ impl Index {
     ///
     /// If a previous index was in this directory, then its meta file will be destroyed.
     #[cfg(feature = "mmap")]
-    pub fn create_in_dir<P: AsRef<Path>>(directory_path: P, schema: Schema) -> Result<Index> {
+    pub fn create_in_dir<P: AsRef<Path>>(
+        directory_path: P,
+        schema: Schema,
+    ) -> crate::Result<Index> {
         let mmap_directory = MmapDirectory::open(directory_path)?;
         if Index::exists(&mmap_directory) {
             return Err(TantivyError::IndexAlreadyExists);
         }
-
         Index::create(mmap_directory, schema)
     }
 
     /// Opens or creates a new index in the provided directory
-    pub fn open_or_create<Dir: Directory>(dir: Dir, schema: Schema) -> Result<Index> {
+    pub fn open_or_create<Dir: Directory>(dir: Dir, schema: Schema) -> crate::Result<Index> {
-        if Index::exists(&dir) {
+        if !Index::exists(&dir) {
-            let index = Index::open(dir)?;
+            return Index::create(dir, schema);
-            if index.schema() == schema {
+        }
-                Ok(index)
+        let index = Index::open(dir)?;
-            } else {
+        if index.schema() == schema {
-                Err(TantivyError::SchemaError(
+            Ok(index)
-                    "An index exists but the schema does not match.".to_string(),
-                ))
-            }
         } else {
-            Index::create(dir, schema)
+            Err(TantivyError::SchemaError(
+                "An index exists but the schema does not match.".to_string(),
+            ))
         }
     }
 
@@ -133,13 +135,13 @@ impl Index {
     /// The temp directory is only used for testing the `MmapDirectory`.
     /// For other unit tests, prefer the `RAMDirectory`, see: `create_in_ram`.
     #[cfg(feature = "mmap")]
-    pub fn create_from_tempdir(schema: Schema) -> Result<Index> {
+    pub fn create_from_tempdir(schema: Schema) -> crate::Result<Index> {
         let mmap_directory = MmapDirectory::create_from_tempdir()?;
         Index::create(mmap_directory, schema)
     }
 
     /// Creates a new index given an implementation of the trait `Directory`
-    pub fn create<Dir: Directory>(dir: Dir, schema: Schema) -> Result<Index> {
+    pub fn create<Dir: Directory>(dir: Dir, schema: Schema) -> crate::Result<Index> {
         let directory = ManagedDirectory::wrap(dir)?;
         Index::from_directory(directory, schema)
     }
@@ -147,7 +149,7 @@ impl Index {
     /// Create a new index from a directory.
     ///
     /// This will overwrite existing meta.json
-    fn from_directory(mut directory: ManagedDirectory, schema: Schema) -> Result<Index> {
+    fn from_directory(mut directory: ManagedDirectory, schema: Schema) -> crate::Result<Index> {
         save_new_metas(schema.clone(), directory.borrow_mut())?;
         let metas = IndexMeta::with_schema(schema);
         Index::create_from_metas(directory, &metas, SegmentMetaInventory::default())
@@ -158,7 +160,7 @@ impl Index {
         directory: ManagedDirectory,
         metas: &IndexMeta,
         inventory: SegmentMetaInventory,
-    ) -> Result<Index> {
+    ) -> crate::Result<Index> {
         let schema = metas.schema.clone();
         Ok(Index {
             directory,
@@ -175,11 +177,11 @@ impl Index {
     }
 
     /// Helper to access the tokenizer associated to a specific field.
-    pub fn tokenizer_for_field(&self, field: Field) -> Result<BoxedTokenizer> {
+    pub fn tokenizer_for_field(&self, field: Field) -> crate::Result<TextAnalyzer> {
         let field_entry = self.schema.get_field_entry(field);
         let field_type = field_entry.field_type();
         let tokenizer_manager: &TokenizerManager = self.tokenizers();
-        let tokenizer_name_opt: Option<BoxedTokenizer> = match field_type {
+        let tokenizer_name_opt: Option<TextAnalyzer> = match field_type {
             FieldType::Str(text_options) => text_options
                 .get_indexing_options()
                 .map(|text_indexing_options| text_indexing_options.tokenizer().to_string())
@@ -198,7 +200,7 @@ impl Index {
     /// Create a default `IndexReader` for the given index.
     ///
     /// See [`Index.reader_builder()`](#method.reader_builder).
-    pub fn reader(&self) -> Result<IndexReader> {
+    pub fn reader(&self) -> crate::Result<IndexReader> {
         self.reader_builder().try_into()
     }
 
@@ -213,7 +215,7 @@ impl Index {
 
     /// Opens a new directory from an index path.
     #[cfg(feature = "mmap")]
-    pub fn open_in_dir<P: AsRef<Path>>(directory_path: P) -> Result<Index> {
+    pub fn open_in_dir<P: AsRef<Path>>(directory_path: P) -> crate::Result<Index> {
         let mmap_directory = MmapDirectory::open(directory_path)?;
         Index::open(mmap_directory)
     }
@@ -237,7 +239,7 @@ impl Index {
     }
 
     /// Open the index using the provided directory
-    pub fn open<D: Directory>(directory: D) -> Result<Index> {
+    pub fn open<D: Directory>(directory: D) -> crate::Result<Index> {
         let directory = ManagedDirectory::wrap(directory)?;
         let inventory = SegmentMetaInventory::default();
         let metas = load_metas(&directory, &inventory)?;
@@ -245,7 +247,7 @@ impl Index {
     }
 
     /// Reads the index meta file from the directory.
-    pub fn load_metas(&self) -> Result<IndexMeta> {
+    pub fn load_metas(&self) -> crate::Result<IndexMeta> {
         load_metas(self.directory(), &self.inventory)
     }
 
@@ -273,7 +275,7 @@ impl Index {
         &self,
         num_threads: usize,
         overall_heap_size_in_bytes: usize,
-    ) -> Result<IndexWriter> {
+    ) -> crate::Result<IndexWriter> {
         let directory_lock = self
             .directory
             .acquire_lock(&INDEX_WRITER_LOCK)
@@ -281,7 +283,7 @@ impl Index {
                 TantivyError::LockFailure(
                     err,
                     Some(
-                        "Failed to acquire index lock. If you are using\
+                        "Failed to acquire index lock. If you are using \
                          a regular directory, this means there is already an \
                          `IndexWriter` working on this `Directory`, in this process \
                          or in a different process."
@@ -308,7 +310,7 @@ impl Index {
     /// If the lockfile already exists, returns `Error::FileAlreadyExists`.
     /// # Panics
     /// If the heap size per thread is too small, panics.
-    pub fn writer(&self, overall_heap_size_in_bytes: usize) -> Result<IndexWriter> {
+    pub fn writer(&self, overall_heap_size_in_bytes: usize) -> crate::Result<IndexWriter> {
         let mut num_threads = num_cpus::get();
         let heap_size_in_bytes_per_thread = overall_heap_size_in_bytes / num_threads;
         if heap_size_in_bytes_per_thread < HEAP_SIZE_MIN {
@@ -325,7 +327,7 @@ impl Index {
     }
 
     /// Returns the list of segments that are searchable
-    pub fn searchable_segments(&self) -> Result<Vec<Segment>> {
+    pub fn searchable_segments(&self) -> crate::Result<Vec<Segment>> {
         Ok(self
             .searchable_segment_metas()?
             .into_iter()
@@ -335,7 +337,7 @@ impl Index {
 
     #[doc(hidden)]
     pub fn segment(&self, segment_meta: SegmentMeta) -> Segment {
-        create_segment(self.clone(), segment_meta)
+        Segment::for_index(self.clone(), segment_meta)
     }
 
     /// Creates a new segment.
@@ -358,12 +360,12 @@ impl Index {
 
     /// Reads the meta.json and returns the list of
     /// `SegmentMeta` from the last commit.
-    pub fn searchable_segment_metas(&self) -> Result<Vec<SegmentMeta>> {
+    pub fn searchable_segment_metas(&self) -> crate::Result<Vec<SegmentMeta>> {
         Ok(self.load_metas()?.segments)
     }
 
     /// Returns the list of segment ids that are searchable.
-    pub fn searchable_segment_ids(&self) -> Result<Vec<SegmentId>> {
+    pub fn searchable_segment_ids(&self) -> crate::Result<Vec<SegmentId>> {
         Ok(self
             .searchable_segment_metas()?
             .iter()
@@ -372,7 +374,7 @@ impl Index {
     }
 
     /// Returns the set of corrupted files
-    pub fn validate_checksum(&self) -> Result<HashSet<PathBuf>> {
+    pub fn validate_checksum(&self) -> crate::Result<HashSet<PathBuf>> {
         self.directory.list_damaged().map_err(Into::into)
     }
 }
@@ -388,12 +390,9 @@ mod tests {
     use crate::directory::RAMDirectory;
     use crate::schema::Field;
     use crate::schema::{Schema, INDEXED, TEXT};
-    use crate::Index;
     use crate::IndexReader;
-    use crate::IndexWriter;
     use crate::ReloadPolicy;
-    use std::thread;
+    use crate::{Directory, Index};
-    use std::time::Duration;
 
     #[test]
     fn test_indexer_for_field() {
@@ -471,14 +470,14 @@ mod tests {
             .try_into()
             .unwrap();
         assert_eq!(reader.searcher().num_docs(), 0);
-        let mut writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        test_index_on_commit_reload_policy_aux(field, &index, &reader);
-        test_index_on_commit_reload_policy_aux(field, &mut writer, &reader);
     }
 
     #[cfg(feature = "mmap")]
     mod mmap_specific {
 
         use super::*;
+        use crate::Directory;
         use std::path::PathBuf;
         use tempfile::TempDir;
 
@@ -489,22 +488,20 @@ mod tests {
             let tempdir = TempDir::new().unwrap();
             let tempdir_path = PathBuf::from(tempdir.path());
             let index = Index::create_in_dir(&tempdir_path, schema).unwrap();
-            let mut writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-            writer.commit().unwrap();
             let reader = index
                 .reader_builder()
                 .reload_policy(ReloadPolicy::OnCommit)
                 .try_into()
                 .unwrap();
             assert_eq!(reader.searcher().num_docs(), 0);
-            test_index_on_commit_reload_policy_aux(field, &mut writer, &reader);
+            test_index_on_commit_reload_policy_aux(field, &index, &reader);
         }
 
         #[test]
         fn test_index_manual_policy_mmap() {
             let schema = throw_away_schema();
             let field = schema.get_field("num_likes").unwrap();
-            let index = Index::create_from_tempdir(schema).unwrap();
+            let mut index = Index::create_from_tempdir(schema).unwrap();
             let mut writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
             writer.commit().unwrap();
             let reader = index
@@ -514,8 +511,12 @@ mod tests {
                 .unwrap();
             assert_eq!(reader.searcher().num_docs(), 0);
             writer.add_document(doc!(field=>1u64));
+            let (sender, receiver) = crossbeam::channel::unbounded();
+            let _handle = index.directory_mut().watch(Box::new(move || {
+                let _ = sender.send(());
+            }));
             writer.commit().unwrap();
-            thread::sleep(Duration::from_millis(500));
+            assert!(receiver.recv().is_ok());
             assert_eq!(reader.searcher().num_docs(), 0);
             reader.reload().unwrap();
             assert_eq!(reader.searcher().num_docs(), 1);
@@ -535,39 +536,26 @@ mod tests {
                 .try_into()
                 .unwrap();
             assert_eq!(reader.searcher().num_docs(), 0);
-            let mut writer = write_index.writer_with_num_threads(1, 3_000_000).unwrap();
+            test_index_on_commit_reload_policy_aux(field, &write_index, &reader);
-            test_index_on_commit_reload_policy_aux(field, &mut writer, &reader);
         }
     }
 
-    fn test_index_on_commit_reload_policy_aux(
+    fn test_index_on_commit_reload_policy_aux(field: Field, index: &Index, reader: &IndexReader) {
-        field: Field,
+        let mut reader_index = reader.index();
-        writer: &mut IndexWriter,
+        let (sender, receiver) = crossbeam::channel::unbounded();
-        reader: &IndexReader,
+        let _watch_handle = reader_index.directory_mut().watch(Box::new(move || {
-    ) {
+            let _ = sender.send(());
+        }));
+        let mut writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
         assert_eq!(reader.searcher().num_docs(), 0);
         writer.add_document(doc!(field=>1u64));
         writer.commit().unwrap();
-        let mut count = 0;
+        assert!(receiver.recv().is_ok());
-        for _ in 0..100 {
+        assert_eq!(reader.searcher().num_docs(), 1);
-            count = reader.searcher().num_docs();
-            if count > 0 {
-                break;
-            }
-            thread::sleep(Duration::from_millis(100));
-        }
-        assert_eq!(count, 1);
         writer.add_document(doc!(field=>2u64));
         writer.commit().unwrap();
-        let mut count = 0;
+        assert!(receiver.recv().is_ok());
-        for _ in 0..10 {
+        assert_eq!(reader.searcher().num_docs(), 2);
-            count = reader.searcher().num_docs();
-            if count > 1 {
-                break;
-            }
-            thread::sleep(Duration::from_millis(100));
-        }
-        assert_eq!(count, 2);
     }
 
     // This test will not pass on windows, because windows
@@ -584,9 +572,13 @@ mod tests {
         for i in 0u64..8_000u64 {
             writer.add_document(doc!(field => i));
         }
+        let (sender, receiver) = crossbeam::channel::unbounded();
+        let _handle = directory.watch(Box::new(move || {
+            let _ = sender.send(());
+        }));
         writer.commit().unwrap();
         let mem_right_after_commit = directory.total_mem_usage();
-        thread::sleep(Duration::from_millis(1_000));
+        assert!(receiver.recv().is_ok());
         let reader = index
             .reader_builder()
             .reload_policy(ReloadPolicy::Manual)
@@ -600,6 +592,11 @@ mod tests {
         reader.reload().unwrap();
         let searcher = reader.searcher();
         assert_eq!(searcher.num_docs(), 8_000);
-        assert!(mem_right_after_merge_finished < mem_right_after_commit);
+        assert!(
+            mem_right_after_merge_finished < mem_right_after_commit,
+            "(mem after merge){} is expected < (mem before merge){}",
+            mem_right_after_merge_finished,
+            mem_right_after_commit
+        );
     }
 }

src/core/index_meta.rs

@@ -3,8 +3,7 @@ use crate::core::SegmentId;
 use crate::schema::Schema;
 use crate::Opstamp;
 use census::{Inventory, TrackedObject};
-use serde;
+use serde::{Deserialize, Serialize};
-use serde_json;
 use std::collections::HashSet;
 use std::fmt;
 use std::path::PathBuf;
@@ -214,7 +213,7 @@ pub struct IndexMeta {
     #[serde(skip_serializing_if = "Option::is_none")]
     /// Payload associated to the last commit.
     ///
-    /// Upon commit, clients can optionally add a small `Striing` payload to their commit
+    /// Upon commit, clients can optionally add a small `String` payload to their commit
     /// to help identify this commit.
     /// This payload is entirely unused by tantivy.
     pub payload: Option<String>,
@@ -300,6 +299,9 @@ mod tests {
             payload: None,
         };
         let json = serde_json::ser::to_string(&index_metas).expect("serialization failed");
-        assert_eq!(json, r#"{"segments":[],"schema":[{"name":"text","type":"text","options":{"indexing":{"record":"position","tokenizer":"default"},"stored":false}}],"opstamp":0}"#);
+        assert_eq!(
+            json,
+            r#"{"segments":[],"schema":[{"name":"text","type":"text","options":{"indexing":{"record":"position","tokenizer":"default"},"stored":false}}],"opstamp":0}"#
+        );
     }
 }

src/core/inverted_index_reader.rs

@@ -7,7 +7,6 @@ use crate::schema::FieldType;
 use crate::schema::IndexRecordOption;
 use crate::schema::Term;
 use crate::termdict::TermDictionary;
-use owned_read::OwnedRead;
 
 /// The inverted index reader is in charge of accessing
 /// the inverted index associated to a specific field.
@@ -60,7 +59,7 @@ impl InvertedIndexReader {
             .get_index_record_option()
             .unwrap_or(IndexRecordOption::Basic);
         InvertedIndexReader {
-            termdict: TermDictionary::empty(&field_type),
+            termdict: TermDictionary::empty(),
             postings_source: ReadOnlySource::empty(),
             positions_source: ReadOnlySource::empty(),
             positions_idx_source: ReadOnlySource::empty(),
@@ -97,8 +96,7 @@ impl InvertedIndexReader {
         let offset = term_info.postings_offset as usize;
         let end_source = self.postings_source.len();
         let postings_slice = self.postings_source.slice(offset, end_source);
-        let postings_reader = OwnedRead::new(postings_slice);
+        block_postings.reset(term_info.doc_freq, postings_slice);
-        block_postings.reset(term_info.doc_freq, postings_reader);
     }
 
     /// Returns a block postings given a `Term`.
@@ -127,7 +125,7 @@ impl InvertedIndexReader {
         let postings_data = self.postings_source.slice_from(offset);
         BlockSegmentPostings::from_data(
             term_info.doc_freq,
-            OwnedRead::new(postings_data),
+            postings_data,
             self.record_option,
             requested_option,
         )

src/core/searcher.rs

@@ -1,11 +1,8 @@
 use crate::collector::Collector;
-use crate::collector::SegmentCollector;
 use crate::core::Executor;
 use crate::core::InvertedIndexReader;
 use crate::core::SegmentReader;
 use crate::query::Query;
-use crate::query::Scorer;
-use crate::query::Weight;
 use crate::schema::Document;
 use crate::schema::Schema;
 use crate::schema::{Field, Term};
@@ -14,30 +11,9 @@ use crate::store::StoreReader;
 use crate::termdict::TermMerger;
 use crate::DocAddress;
 use crate::Index;
-use crate::Result;
 use std::fmt;
 use std::sync::Arc;
 
-fn collect_segment<C: Collector>(
-    collector: &C,
-    weight: &dyn Weight,
-    segment_ord: u32,
-    segment_reader: &SegmentReader,
-) -> Result<C::Fruit> {
-    let mut scorer = weight.scorer(segment_reader)?;
-    let mut segment_collector = collector.for_segment(segment_ord as u32, segment_reader)?;
-    if let Some(delete_bitset) = segment_reader.delete_bitset() {
-        scorer.for_each(&mut |doc, score| {
-            if delete_bitset.is_alive(doc) {
-                segment_collector.collect(doc, score);
-            }
-        });
-    } else {
-        scorer.for_each(&mut |doc, score| segment_collector.collect(doc, score));
-    }
-    Ok(segment_collector.harvest())
-}
-
 /// Holds a list of `SegmentReader`s ready for search.
 ///
 /// It guarantees that the `Segment` will not be removed before
@@ -78,7 +54,7 @@ impl Searcher {
     ///
     /// The searcher uses the segment ordinal to route the
     /// the request to the right `Segment`.
-    pub fn doc(&self, doc_address: DocAddress) -> Result<Document> {
+    pub fn doc(&self, doc_address: DocAddress) -> crate::Result<Document> {
         let DocAddress(segment_local_id, doc_id) = doc_address;
         let store_reader = &self.store_readers[segment_local_id as usize];
         store_reader.get(doc_id)
@@ -132,7 +108,11 @@ impl Searcher {
     ///
     ///  Finally, the Collector merges each of the child collectors into itself for result usability
     ///  by the caller.
-    pub fn search<C: Collector>(&self, query: &dyn Query, collector: &C) -> Result<C::Fruit> {
+    pub fn search<C: Collector>(
+        &self,
+        query: &dyn Query,
+        collector: &C,
+    ) -> crate::Result<C::Fruit> {
         let executor = self.index.search_executor();
         self.search_with_executor(query, collector, executor)
     }
@@ -154,18 +134,13 @@ impl Searcher {
         query: &dyn Query,
         collector: &C,
         executor: &Executor,
-    ) -> Result<C::Fruit> {
+    ) -> crate::Result<C::Fruit> {
         let scoring_enabled = collector.requires_scoring();
         let weight = query.weight(self, scoring_enabled)?;
         let segment_readers = self.segment_readers();
         let fruits = executor.map(
             |(segment_ord, segment_reader)| {
-                collect_segment(
+                collector.collect_segment(weight.as_ref(), segment_ord as u32, segment_reader)
-                    collector,
-                    weight.as_ref(),
-                    segment_ord as u32,
-                    segment_reader,
-                )
             },
             segment_readers.iter().enumerate(),
         )?;

src/core/segment.rs

@@ -8,10 +8,8 @@ use crate::directory::{ReadOnlySource, WritePtr};
 use crate::indexer::segment_serializer::SegmentSerializer;
 use crate::schema::Schema;
 use crate::Opstamp;
-use crate::Result;
 use std::fmt;
 use std::path::PathBuf;
-use std::result;
 
 /// A segment is a piece of the index.
 #[derive(Clone)]
@@ -26,15 +24,12 @@ impl fmt::Debug for Segment {
     }
 }
 
-/// Creates a new segment given an `Index` and a `SegmentId`
-///
-/// The function is here to make it private outside `tantivy`.
-/// #[doc(hidden)]
-pub fn create_segment(index: Index, meta: SegmentMeta) -> Segment {
-    Segment { index, meta }
-}
-
 impl Segment {
+    /// Creates a new segment given an `Index` and a `SegmentId`
+    pub(crate) fn for_index(index: Index, meta: SegmentMeta) -> Segment {
+        Segment { index, meta }
+    }
+
     /// Returns the index the segment belongs to.
     pub fn index(&self) -> &Index {
         &self.index
@@ -83,20 +78,14 @@ impl Segment {
     }
 
     /// Open one of the component file for a *regular* read.
-    pub fn open_read(
+    pub fn open_read(&self, component: SegmentComponent) -> Result<ReadOnlySource, OpenReadError> {
-        &self,
-        component: SegmentComponent,
-    ) -> result::Result<ReadOnlySource, OpenReadError> {
         let path = self.relative_path(component);
         let source = self.index.directory().open_read(&path)?;
         Ok(source)
     }
 
     /// Open one of the component file for *regular* write.
-    pub fn open_write(
+    pub fn open_write(&mut self, component: SegmentComponent) -> Result<WritePtr, OpenWriteError> {
-        &mut self,
-        component: SegmentComponent,
-    ) -> result::Result<WritePtr, OpenWriteError> {
         let path = self.relative_path(component);
         let write = self.index.directory_mut().open_write(&path)?;
         Ok(write)
@@ -109,5 +98,5 @@ pub trait SerializableSegment {
     ///
     /// # Returns
     /// The number of documents in the segment.
-    fn write(&self, serializer: SegmentSerializer) -> Result<u32>;
+    fn write(&self, serializer: SegmentSerializer) -> crate::Result<u32>;
 }

src/core/segment_id.rs

@@ -4,6 +4,7 @@ use uuid::Uuid;
 
 #[cfg(test)]
 use once_cell::sync::Lazy;
+use serde::{Deserialize, Serialize};
 use std::error::Error;
 use std::str::FromStr;
 #[cfg(test)]

src/core/segment_reader.rs

@@ -8,7 +8,7 @@ use crate::directory::ReadOnlySource;
 use crate::fastfield::DeleteBitSet;
 use crate::fastfield::FacetReader;
 use crate::fastfield::FastFieldReaders;
-use crate::fieldnorm::FieldNormReader;
+use crate::fieldnorm::{FieldNormReader, FieldNormReaders};
 use crate::schema::Field;
 use crate::schema::FieldType;
 use crate::schema::Schema;
@@ -16,7 +16,6 @@ use crate::space_usage::SegmentSpaceUsage;
 use crate::store::StoreReader;
 use crate::termdict::TermDictionary;
 use crate::DocId;
-use crate::Result;
 use fail::fail_point;
 use std::collections::HashMap;
 use std::fmt;
@@ -49,7 +48,7 @@ pub struct SegmentReader {
     positions_composite: CompositeFile,
     positions_idx_composite: CompositeFile,
     fast_fields_readers: Arc<FastFieldReaders>,
-    fieldnorms_composite: CompositeFile,
+    fieldnorm_readers: FieldNormReaders,
 
     store_source: ReadOnlySource,
     delete_bitset_opt: Option<DeleteBitSet>,
@@ -113,8 +112,10 @@ impl SegmentReader {
             return None;
         }
         let term_ords_reader = self.fast_fields().u64s(field)?;
-        let termdict_source = self.termdict_composite.open_read(field)?;
+        let termdict = self.termdict_composite
-        let termdict = TermDictionary::from_source(&termdict_source);
+            .open_read(field)
+            .map(|source| TermDictionary::from_source(&source))
+            .unwrap_or_else(TermDictionary::empty);
         let facet_reader = FacetReader::new(term_ords_reader, termdict);
         Some(facet_reader)
     }
@@ -127,8 +128,8 @@ impl SegmentReader {
     /// They are simply stored as a fast field, serialized in
     /// the `.fieldnorm` file of the segment.
     pub fn get_fieldnorms_reader(&self, field: Field) -> FieldNormReader {
-        if let Some(fieldnorm_source) = self.fieldnorms_composite.open_read(field) {
+        if let Some(fieldnorm_reader) = self.fieldnorm_readers.get_field(field) {
-            FieldNormReader::open(fieldnorm_source)
+            fieldnorm_reader
         } else {
             let field_name = self.schema.get_field_name(field);
             let err_msg = format!(
@@ -145,7 +146,7 @@ impl SegmentReader {
     }
 
     /// Open a new segment for reading.
-    pub fn open(segment: &Segment) -> Result<SegmentReader> {
+    pub fn open(segment: &Segment) -> crate::Result<SegmentReader> {
         let termdict_source = segment.open_read(SegmentComponent::TERMS)?;
         let termdict_composite = CompositeFile::open(&termdict_source)?;
 
@@ -179,8 +180,8 @@ impl SegmentReader {
         let fast_field_readers =
             Arc::new(FastFieldReaders::load_all(&schema, &fast_fields_composite)?);
 
-        let fieldnorms_data = segment.open_read(SegmentComponent::FIELDNORMS)?;
+        let fieldnorm_data = segment.open_read(SegmentComponent::FIELDNORMS)?;
-        let fieldnorms_composite = CompositeFile::open(&fieldnorms_data)?;
+        let fieldnorm_readers = FieldNormReaders::open(fieldnorm_data)?;
 
         let delete_bitset_opt = if segment.meta().has_deletes() {
             let delete_data = segment.open_read(SegmentComponent::DELETE)?;
@@ -196,7 +197,7 @@ impl SegmentReader {
             termdict_composite,
             postings_composite,
             fast_fields_readers: fast_field_readers,
-            fieldnorms_composite,
+            fieldnorm_readers,
             segment_id: segment.id(),
             store_source,
             delete_bitset_opt,
@@ -296,8 +297,8 @@ impl SegmentReader {
     }
 
     /// Returns an iterator that will iterate over the alive document ids
-    pub fn doc_ids_alive(&self) -> SegmentReaderAliveDocsIterator<'_> {
+    pub fn doc_ids_alive<'a>(&'a self) -> impl Iterator<Item = DocId> + 'a {
-        SegmentReaderAliveDocsIterator::new(&self)
+        (0u32..self.max_doc).filter(move |doc| !self.is_deleted(*doc))
     }
 
     /// Summarize total space usage of this segment.
@@ -309,7 +310,7 @@ impl SegmentReader {
             self.positions_composite.space_usage(),
             self.positions_idx_composite.space_usage(),
             self.fast_fields_readers.space_usage(),
-            self.fieldnorms_composite.space_usage(),
+            self.fieldnorm_readers.space_usage(),
             self.get_store_reader().space_usage(),
             self.delete_bitset_opt
                 .as_ref()
@@ -325,52 +326,6 @@ impl fmt::Debug for SegmentReader {
     }
 }
 
-/// Implements the iterator trait to allow easy iteration
-/// over non-deleted ("alive") DocIds in a SegmentReader
-pub struct SegmentReaderAliveDocsIterator<'a> {
-    reader: &'a SegmentReader,
-    max_doc: DocId,
-    current: DocId,
-}
-
-impl<'a> SegmentReaderAliveDocsIterator<'a> {
-    pub fn new(reader: &'a SegmentReader) -> SegmentReaderAliveDocsIterator<'a> {
-        SegmentReaderAliveDocsIterator {
-            reader,
-            max_doc: reader.max_doc(),
-            current: 0,
-        }
-    }
-}
-
-impl<'a> Iterator for SegmentReaderAliveDocsIterator<'a> {
-    type Item = DocId;
-
-    fn next(&mut self) -> Option<Self::Item> {
-        // TODO: Use TinySet (like in BitSetDocSet) to speed this process up
-        if self.current >= self.max_doc {
-            return None;
-        }
-
-        // find the next alive doc id
-        while self.reader.is_deleted(self.current) {
-            self.current += 1;
-
-            if self.current >= self.max_doc {
-                return None;
-            }
-        }
-
-        // capture the current alive DocId
-        let result = Some(self.current);
-
-        // move down the chain
-        self.current += 1;
-
-        result
-    }
-}
-
 #[cfg(test)]
 mod test {
     use crate::core::Index;

src/directory/directory.rs

@@ -119,7 +119,7 @@ pub trait Directory: DirectoryClone + fmt::Debug + Send + Sync + 'static {
     /// Specifically, subsequent writes or flushes should
     /// have no effect on the returned `ReadOnlySource` object.
     ///
-    /// You should only use this to read files create with [`open_write`]
+    /// You should only use this to read files create with [Directory::open_write].
     fn open_read(&self, path: &Path) -> result::Result<ReadOnlySource, OpenReadError>;
 
     /// Removes a file
@@ -160,7 +160,7 @@ pub trait Directory: DirectoryClone + fmt::Debug + Send + Sync + 'static {
     ///
     /// This should only be used for small files.
     ///
-    /// You should only use this to read files create with [`atomic_write`]
+    /// You should only use this to read files create with [Directory::atomic_write].
     fn atomic_read(&self, path: &Path) -> Result<Vec<u8>, OpenReadError>;
 
     /// Atomically replace the content of a file with data.
@@ -197,7 +197,7 @@ pub trait Directory: DirectoryClone + fmt::Debug + Send + Sync + 'static {
     /// Registers a callback that will be called whenever a change on the `meta.json`
     /// using the `atomic_write` API is detected.
     ///
-    /// The behavior when using `.watch()` on a file using `.open_write(...)` is, on the other
+    /// The behavior when using `.watch()` on a file using [Directory::open_write] is, on the other
     /// hand, undefined.
     ///
     /// The file will be watched for the lifetime of the returned `WatchHandle`. The caller is

src/directory/error.rs

@@ -1,3 +1,4 @@
+use crate::Version;
 use std::error::Error as StdError;
 use std::fmt;
 use std::io;
@@ -156,6 +157,65 @@ impl StdError for OpenWriteError {
     }
 }
 
+/// Type of index incompatibility between the library and the index found on disk
+/// Used to catch and provide a hint to solve this incompatibility issue
+pub enum Incompatibility {
+    /// This library cannot decompress the index found on disk
+    CompressionMismatch {
+        /// Compression algorithm used by the current version of tantivy
+        library_compression_format: String,
+        /// Compression algorithm that was used to serialise the index
+        index_compression_format: String,
+    },
+    /// The index format found on disk isn't supported by this version of the library
+    IndexMismatch {
+        /// Version used by the library
+        library_version: Version,
+        /// Version the index was built with
+        index_version: Version,
+    },
+}
+
+impl fmt::Debug for Incompatibility {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> Result<(), fmt::Error> {
+        match self {
+            Incompatibility::CompressionMismatch {
+                library_compression_format,
+                index_compression_format,
+            } => {
+                let err = format!(
+                    "Library was compiled with {:?} compression, index was compressed with {:?}",
+                    library_compression_format, index_compression_format
+                );
+                let advice = format!(
+                    "Change the feature flag to {:?} and rebuild the library",
+                    index_compression_format
+                );
+                write!(f, "{}. {}", err, advice)?;
+            }
+            Incompatibility::IndexMismatch {
+                library_version,
+                index_version,
+            } => {
+                let err = format!(
+                    "Library version: {}, index version: {}",
+                    library_version.index_format_version, index_version.index_format_version
+                );
+                // TODO make a more useful error message
+                // include the version range that supports this index_format_version
+                let advice = format!(
+                    "Change tantivy to a version compatible with index format {} (e.g. {}.{}.x) \
+                     and rebuild your project.",
+                    index_version.index_format_version, index_version.major, index_version.minor
+                );
+                write!(f, "{}. {}", err, advice)?;
+            }
+        }
+
+        Ok(())
+    }
+}
+
 /// Error that may occur when accessing a file read
 #[derive(Debug)]
 pub enum OpenReadError {
@@ -164,6 +224,8 @@ pub enum OpenReadError {
     /// Any kind of IO error that happens when
     /// interacting with the underlying IO device.
     IOError(IOError),
+    /// This library doesn't support the index version found on disk
+    IncompatibleIndex(Incompatibility),
 }
 
 impl From<IOError> for OpenReadError {
@@ -183,19 +245,9 @@ impl fmt::Display for OpenReadError {
                 "an io error occurred while opening a file for reading: '{}'",
                 err
             ),
-        }
+            OpenReadError::IncompatibleIndex(ref footer) => {
-    }
+                write!(f, "Incompatible index format: {:?}", footer)
-}
+            }
-
-impl StdError for OpenReadError {
-    fn description(&self) -> &str {
-        "error occurred while opening a file for reading"
-    }
-
-    fn cause(&self) -> Option<&dyn StdError> {
-        match *self {
-            OpenReadError::FileDoesNotExist(_) => None,
-            OpenReadError::IOError(ref err) => Some(err),
         }
     }
 }
@@ -216,6 +268,12 @@ impl From<IOError> for DeleteError {
     }
 }
 
+impl From<Incompatibility> for OpenReadError {
+    fn from(incompatibility: Incompatibility) -> Self {
+        OpenReadError::IncompatibleIndex(incompatibility)
+    }
+}
+
 impl fmt::Display for DeleteError {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         match *self {

src/directory/footer.rs

@@ -1,159 +1,213 @@
+use crate::common::{BinarySerializable, CountingWriter, FixedSize, VInt};
+use crate::directory::error::Incompatibility;
 use crate::directory::read_only_source::ReadOnlySource;
 use crate::directory::{AntiCallToken, TerminatingWrite};
-use byteorder::{ByteOrder, LittleEndian};
+use crate::Version;
+use byteorder::{ByteOrder, LittleEndian, WriteBytesExt};
 use crc32fast::Hasher;
 use std::io;
 use std::io::Write;
 
-const COMMON_FOOTER_SIZE: usize = 4 * 5;
+const FOOTER_MAX_LEN: usize = 10_000;
+
+type CrcHashU32 = u32;
 
 #[derive(Debug, Clone, PartialEq)]
 pub struct Footer {
-    pub tantivy_version: (u32, u32, u32),
+    pub version: Version,
     pub meta: String,
     pub versioned_footer: VersionedFooter,
 }
 
+/// Serialises the footer to a byte-array
+/// - versioned_footer_len : 4 bytes
+///-  versioned_footer: variable bytes
+/// - meta_len: 4 bytes
+/// - meta: variable bytes
+/// - version_len: 4 bytes
+/// - version json: variable bytes
+impl BinarySerializable for Footer {
+    fn serialize<W: io::Write>(&self, writer: &mut W) -> io::Result<()> {
+        BinarySerializable::serialize(&self.versioned_footer, writer)?;
+        BinarySerializable::serialize(&self.meta, writer)?;
+        let version_string =
+            serde_json::to_string(&self.version).map_err(|_err| io::ErrorKind::InvalidInput)?;
+        BinarySerializable::serialize(&version_string, writer)?;
+        Ok(())
+    }
+
+    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
+        let versioned_footer = VersionedFooter::deserialize(reader)?;
+        let meta = String::deserialize(reader)?;
+        let version_json = String::deserialize(reader)?;
+        let version = serde_json::from_str(&version_json)?;
+        Ok(Footer {
+            version,
+            meta,
+            versioned_footer,
+        })
+    }
+}
+
 impl Footer {
     pub fn new(versioned_footer: VersionedFooter) -> Self {
-        let tantivy_version = (
+        let version = crate::VERSION.clone();
-            env!("CARGO_PKG_VERSION_MAJOR").parse().unwrap(),
+        let meta = version.to_string();
-            env!("CARGO_PKG_VERSION_MINOR").parse().unwrap(),
-            env!("CARGO_PKG_VERSION_PATCH").parse().unwrap(),
-        );
         Footer {
-            tantivy_version,
+            version,
-            meta: format!(
+            meta,
-                "tantivy {}.{}.{}, index v{}",
-                tantivy_version.0,
-                tantivy_version.1,
-                tantivy_version.2,
-                versioned_footer.version()
-            ),
             versioned_footer,
         }
     }
 
-    pub fn to_bytes(&self) -> Vec<u8> {
+    pub fn append_footer<W: io::Write>(&self, mut write: &mut W) -> io::Result<()> {
-        let mut res = self.versioned_footer.to_bytes();
+        let mut counting_write = CountingWriter::wrap(&mut write);
-        res.extend_from_slice(self.meta.as_bytes());
+        self.serialize(&mut counting_write)?;
-        let len = res.len();
+        let written_len = counting_write.written_bytes();
-        res.resize(len + COMMON_FOOTER_SIZE, 0);
+        write.write_u32::<LittleEndian>(written_len as u32)?;
-        let mut common_footer = &mut res[len..];
+        Ok(())
-        LittleEndian::write_u32(&mut common_footer, self.meta.len() as u32);
-        LittleEndian::write_u32(&mut common_footer[4..], self.tantivy_version.0);
-        LittleEndian::write_u32(&mut common_footer[8..], self.tantivy_version.1);
-        LittleEndian::write_u32(&mut common_footer[12..], self.tantivy_version.2);
-        LittleEndian::write_u32(&mut common_footer[16..], (len + COMMON_FOOTER_SIZE) as u32);
-        res
-    }
-
-    pub fn from_bytes(data: &[u8]) -> Result<Self, io::Error> {
-        let len = data.len();
-        if len < COMMON_FOOTER_SIZE + 4 {
-            // 4 bytes for index version, stored in versioned footer
-            return Err(io::Error::new(
-                    io::ErrorKind::UnexpectedEof,
-                    format!("File corrupted. The footer len must be over 24, while the entire file len is {}", len)
-                    )
-            );
-        }
-
-        let size = LittleEndian::read_u32(&data[len - 4..]) as usize;
-        if len < size as usize {
-            return Err(io::Error::new(
-                io::ErrorKind::UnexpectedEof,
-                format!(
-                    "File corrupted. The footer len is {}, while the entire file len is {}",
-                    size, len
-                ),
-            ));
-        }
-        let footer = &data[len - size as usize..];
-        let meta_len = LittleEndian::read_u32(&footer[size - 20..]) as usize;
-        let tantivy_major = LittleEndian::read_u32(&footer[size - 16..]);
-        let tantivy_minor = LittleEndian::read_u32(&footer[size - 12..]);
-        let tantivy_patch = LittleEndian::read_u32(&footer[size - 8..]);
-        Ok(Footer {
-            tantivy_version: (tantivy_major, tantivy_minor, tantivy_patch),
-            meta: String::from_utf8_lossy(&footer[size - meta_len - 20..size - 20]).into_owned(),
-            versioned_footer: VersionedFooter::from_bytes(&footer[..size - meta_len - 20])?,
-        })
     }
 
     pub fn extract_footer(source: ReadOnlySource) -> Result<(Footer, ReadOnlySource), io::Error> {
-        let footer = Footer::from_bytes(source.as_slice())?;
+        if source.len() < 4 {
-        let reader = source.slice_to(source.as_slice().len() - footer.size());
+            return Err(io::Error::new(
-        Ok((footer, reader))
+                io::ErrorKind::UnexpectedEof,
-    }
+                format!(
-
+                    "File corrupted. The file is smaller than 4 bytes (len={}).",
-    pub fn size(&self) -> usize {
+                    source.len()
-        self.versioned_footer.size() as usize + self.meta.len() + 20
+                ),
-    }
+            ));
-}
-
-#[derive(Debug, Clone, PartialEq)]
-pub enum VersionedFooter {
-    UnknownVersion { version: u32, size: u32 },
-    V0(u32), // crc
-}
-
-impl VersionedFooter {
-    pub fn to_bytes(&self) -> Vec<u8> {
-        match self {
-            VersionedFooter::V0(crc) => {
-                let mut res = vec![0; 8];
-                LittleEndian::write_u32(&mut res, 0);
-                LittleEndian::write_u32(&mut res[4..], *crc);
-                res
-            }
-            VersionedFooter::UnknownVersion { .. } => {
-                panic!("Unsupported index should never get serialized");
-            }
         }
+        let (body_footer, footer_len_bytes) = source.split_from_end(u32::SIZE_IN_BYTES);
+        let footer_len = LittleEndian::read_u32(footer_len_bytes.as_slice()) as usize;
+        let body_len = body_footer.len() - footer_len;
+        let (body, footer_data) = body_footer.split(body_len);
+        let mut cursor = footer_data.as_slice();
+        let footer = Footer::deserialize(&mut cursor)?;
+        Ok((footer, body))
     }
 
-    pub fn from_bytes(footer: &[u8]) -> Result<Self, io::Error> {
+    /// Confirms that the index will be read correctly by this version of tantivy
-        assert!(footer.len() >= 4);
+    /// Has to be called after `extract_footer` to make sure it's not accessing uninitialised memory
-        let version = LittleEndian::read_u32(footer);
+    pub fn is_compatible(&self) -> Result<(), Incompatibility> {
-        match version {
+        let library_version = crate::version();
-            0 => {
+        match &self.versioned_footer {
-                if footer.len() == 8 {
+            VersionedFooter::V1 {
-                    Ok(VersionedFooter::V0(LittleEndian::read_u32(&footer[4..])))
+                crc32: _crc,
-                } else {
+                store_compression,
-                    Err(io::Error::new(
+            } => {
-                        io::ErrorKind::UnexpectedEof,
+                if &library_version.store_compression != store_compression {
-                        format!(
+                    return Err(Incompatibility::CompressionMismatch {
-                            "File corrupted. The versioned footer len is {}, while it should be 8",
+                        library_compression_format: library_version.store_compression.to_string(),
-                            footer.len()
+                        index_compression_format: store_compression.to_string(),
-                        ),
+                    });
-                    ))
                 }
+                Ok(())
             }
-            version => Ok(VersionedFooter::UnknownVersion {
+            VersionedFooter::V2 {
-                version,
+                crc32: _crc,
-                size: footer.len() as u32,
+                store_compression,
+            } => {
+                if &library_version.store_compression != store_compression {
+                    return Err(Incompatibility::CompressionMismatch {
+                        library_compression_format: library_version.store_compression.to_string(),
+                        index_compression_format: store_compression.to_string(),
+                    });
+                }
+                Ok(())
+            }
+            VersionedFooter::UnknownVersion => Err(Incompatibility::IndexMismatch {
+                library_version: library_version.clone(),
+                index_version: self.version.clone(),
             }),
         }
     }
+}
 
-    pub fn size(&self) -> u32 {
+/// Footer that includes a crc32 hash that enables us to checksum files in the index
+#[derive(Debug, Clone, PartialEq)]
+pub enum VersionedFooter {
+    UnknownVersion,
+    V1 {
+        crc32: CrcHashU32,
+        store_compression: String,
+    },
+    // Introduction of the Block WAND information.
+    V2 {
+        crc32: CrcHashU32,
+        store_compression: String,
+    },
+}
+
+impl BinarySerializable for VersionedFooter {
+    fn serialize<W: io::Write>(&self, writer: &mut W) -> io::Result<()> {
+        let mut buf = Vec::new();
         match self {
-            VersionedFooter::V0(_) => 8,
+            VersionedFooter::V2 {
-            VersionedFooter::UnknownVersion { size, .. } => *size,
+                crc32,
+                store_compression: compression,
+            } => {
+                // Serializes a valid `VersionedFooter` or panics if the version is unknown
+                // [   version    |   crc_hash  | compression_mode ]
+                // [    0..4      |     4..8    |     variable     ]
+                BinarySerializable::serialize(&2u32, &mut buf)?;
+                BinarySerializable::serialize(crc32, &mut buf)?;
+                BinarySerializable::serialize(compression, &mut buf)?;
+            }
+            VersionedFooter::V1 { .. } | VersionedFooter::UnknownVersion => {
+                return Err(io::Error::new(
+                    io::ErrorKind::InvalidInput,
+                    "Cannot serialize an unknown versioned footer ",
+                ));
+            }
         }
+        BinarySerializable::serialize(&VInt(buf.len() as u64), writer)?;
+        assert!(buf.len() <= FOOTER_MAX_LEN);
+        writer.write_all(&buf[..])?;
+        Ok(())
     }
 
-    pub fn version(&self) -> u32 {
+    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        match self {
+        let len = VInt::deserialize(reader)?.0 as usize;
-            VersionedFooter::V0(_) => 0,
+        if len > FOOTER_MAX_LEN {
-            VersionedFooter::UnknownVersion { version, .. } => *version,
+            return Err(io::Error::new(
+                io::ErrorKind::InvalidData,
+                format!(
+                    "Footer seems invalid as it suggests a footer len of {}. File is corrupted, \
+            or the index was created with a different & old version of tantivy.",
+                    len
+                ),
+            ));
         }
+        let mut buf = vec![0u8; len];
+        reader.read_exact(&mut buf[..])?;
+        let mut cursor = &buf[..];
+        let version = u32::deserialize(&mut cursor)?;
+        if version != 1 && version != 2 {
+            return Ok(VersionedFooter::UnknownVersion);
+        }
+        let crc32 = u32::deserialize(&mut cursor)?;
+        let store_compression = String::deserialize(&mut cursor)?;
+        Ok(if version == 1 {
+            VersionedFooter::V1 {
+                crc32,
+                store_compression,
+            }
+        } else {
+            assert_eq!(version, 2);
+            VersionedFooter::V2 {
+                crc32,
+                store_compression,
+            }
+        })
     }
+}
 
-    pub fn crc(&self) -> Option<u32> {
+impl VersionedFooter {
+    pub fn crc(&self) -> Option<CrcHashU32> {
         match self {
-            VersionedFooter::V0(crc) => Some(*crc),
+            VersionedFooter::V2 { crc32, .. } => Some(*crc32),
+            VersionedFooter::V1 { crc32, .. } => Some(*crc32),
             VersionedFooter::UnknownVersion { .. } => None,
         }
     }
@@ -189,25 +243,148 @@ impl<W: TerminatingWrite> Write for FooterProxy<W> {
 
 impl<W: TerminatingWrite> TerminatingWrite for FooterProxy<W> {
     fn terminate_ref(&mut self, _: AntiCallToken) -> io::Result<()> {
-        let crc = self.hasher.take().unwrap().finalize();
+        let crc32 = self.hasher.take().unwrap().finalize();
-
+        let footer = Footer::new(VersionedFooter::V2 {
-        let footer = Footer::new(VersionedFooter::V0(crc)).to_bytes();
+            crc32,
+            store_compression: crate::store::COMPRESSION.to_string(),
+        });
         let mut writer = self.writer.take().unwrap();
-        writer.write_all(&footer)?;
+        footer.append_footer(&mut writer)?;
         writer.terminate()
     }
 }
 
 #[cfg(test)]
 mod tests {
+
+    use super::CrcHashU32;
+    use super::FooterProxy;
+    use crate::common::{BinarySerializable, VInt};
     use crate::directory::footer::{Footer, VersionedFooter};
+    use crate::directory::TerminatingWrite;
+    use byteorder::{ByteOrder, LittleEndian};
+    use regex::Regex;
+    use std::io;
+
+    #[test]
+    fn test_versioned_footer() {
+        let mut vec = Vec::new();
+        let footer_proxy = FooterProxy::new(&mut vec);
+        assert!(footer_proxy.terminate().is_ok());
+        assert_eq!(vec.len(), 167);
+        let footer = Footer::deserialize(&mut &vec[..]).unwrap();
+        assert!(matches!(
+           footer.versioned_footer,
+           VersionedFooter::V2 { store_compression, .. }
+           if store_compression == crate::store::COMPRESSION
+        ));
+        assert_eq!(&footer.version, crate::version());
+    }
 
     #[test]
     fn test_serialize_deserialize_footer() {
-        let crc = 123456;
+        let mut buffer = Vec::new();
-        let footer = Footer::new(VersionedFooter::V0(crc));
+        let crc32 = 123456u32;
-        let footer_bytes = footer.to_bytes();
+        let footer: Footer = Footer::new(VersionedFooter::V2 {
+            crc32,
+            store_compression: "lz4".to_string(),
+        });
+        footer.serialize(&mut buffer).unwrap();
+        let footer_deser = Footer::deserialize(&mut &buffer[..]).unwrap();
+        assert_eq!(footer_deser, footer);
+    }
 
-        assert_eq!(Footer::from_bytes(&footer_bytes).unwrap(), footer);
+    #[test]
+    fn footer_length() {
+        let crc32 = 1111111u32;
+        let versioned_footer = VersionedFooter::V2 {
+            crc32,
+            store_compression: "lz4".to_string(),
+        };
+        let mut buf = Vec::new();
+        versioned_footer.serialize(&mut buf).unwrap();
+        assert_eq!(buf.len(), 13);
+        let footer = Footer::new(versioned_footer);
+        let regex_ptn = Regex::new(
+            "tantivy v[0-9]{1,3}\\.[0-9]{1,3}\\.[0-9]{1,3}\\.{0,10}, index_format v[0-9]{1,5}",
+        )
+        .unwrap();
+        assert!(regex_ptn.is_match(&footer.meta));
+    }
+
+    #[test]
+    fn versioned_footer_from_bytes() {
+        let v_footer_bytes = vec![
+            // versionned footer length
+            12 | 128,
+            // index format version
+            2,
+            0,
+            0,
+            0,
+            // crc 32
+            12,
+            35,
+            89,
+            18,
+            // compression format
+            3 | 128,
+            b'l',
+            b'z',
+            b'4',
+        ];
+        let mut cursor = &v_footer_bytes[..];
+        let versioned_footer = VersionedFooter::deserialize(&mut cursor).unwrap();
+        assert!(cursor.is_empty());
+        let expected_crc: u32 = LittleEndian::read_u32(&v_footer_bytes[5..9]) as CrcHashU32;
+        let expected_versioned_footer: VersionedFooter = VersionedFooter::V2 {
+            crc32: expected_crc,
+            store_compression: "lz4".to_string(),
+        };
+        assert_eq!(versioned_footer, expected_versioned_footer);
+        let mut buffer = Vec::new();
+        assert!(versioned_footer.serialize(&mut buffer).is_ok());
+        assert_eq!(&v_footer_bytes[..], &buffer[..]);
+    }
+
+    #[test]
+    fn versioned_footer_panic() {
+        let v_footer_bytes = vec![6u8 | 128u8, 3u8, 0u8, 0u8, 1u8, 0u8, 0u8];
+        let mut b = &v_footer_bytes[..];
+        let versioned_footer = VersionedFooter::deserialize(&mut b).unwrap();
+        assert!(b.is_empty());
+        let expected_versioned_footer = VersionedFooter::UnknownVersion;
+        assert_eq!(versioned_footer, expected_versioned_footer);
+        let mut buf = Vec::new();
+        assert!(versioned_footer.serialize(&mut buf).is_err());
+    }
+
+    #[test]
+    #[cfg(not(feature = "lz4"))]
+    fn compression_mismatch() {
+        let crc32 = 1111111u32;
+        let versioned_footer = VersionedFooter::V1 {
+            crc32,
+            store_compression: "lz4".to_string(),
+        };
+        let footer = Footer::new(versioned_footer);
+        let res = footer.is_compatible();
+        assert!(res.is_err());
+    }
+
+    #[test]
+    fn test_deserialize_too_large_footer() {
+        let mut buf = vec![];
+        assert!(FooterProxy::new(&mut buf).terminate().is_ok());
+        let mut long_len_buf = [0u8; 10];
+        let num_bytes = VInt(super::FOOTER_MAX_LEN as u64 + 1u64).serialize_into(&mut long_len_buf);
+        buf[0..num_bytes].copy_from_slice(&long_len_buf[..num_bytes]);
+        let err = Footer::deserialize(&mut &buf[..]).unwrap_err();
+        assert_eq!(err.kind(), io::ErrorKind::InvalidData);
+        assert_eq!(
+            err.to_string(),
+            "Footer seems invalid as it suggests a footer len of 10001. File is corrupted, \
+            or the index was created with a different & old version of tantivy."
+        );
     }
 }

src/directory/managed_directory.rs

@@ -2,15 +2,15 @@ use crate::core::MANAGED_FILEPATH;
 use crate::directory::error::{DeleteError, IOError, LockError, OpenReadError, OpenWriteError};
 use crate::directory::footer::{Footer, FooterProxy};
 use crate::directory::DirectoryLock;
+use crate::directory::GarbageCollectionResult;
 use crate::directory::Lock;
 use crate::directory::META_LOCK;
 use crate::directory::{ReadOnlySource, WritePtr};
 use crate::directory::{WatchCallback, WatchHandle};
 use crate::error::DataCorruption;
 use crate::Directory;
-use crate::Result;
+
 use crc32fast::Hasher;
-use serde_json;
 use std::collections::HashSet;
 use std::io;
 use std::io::Write;
@@ -64,7 +64,7 @@ fn save_managed_paths(
 
 impl ManagedDirectory {
     /// Wraps a directory as managed directory.
-    pub fn wrap<Dir: Directory>(directory: Dir) -> Result<ManagedDirectory> {
+    pub fn wrap<Dir: Directory>(directory: Dir) -> crate::Result<ManagedDirectory> {
         match directory.atomic_read(&MANAGED_FILEPATH) {
             Ok(data) => {
                 let managed_files_json = String::from_utf8_lossy(&data);
@@ -87,6 +87,11 @@ impl ManagedDirectory {
                 meta_informations: Arc::default(),
             }),
             Err(OpenReadError::IOError(e)) => Err(From::from(e)),
+            Err(OpenReadError::IncompatibleIndex(incompatibility)) => {
+                // For the moment, this should never happen  `meta.json`
+                // do not have any footer and cannot detect incompatibility.
+                Err(crate::TantivyError::IncompatibleIndex(incompatibility))
+            }
         }
     }
 
@@ -104,7 +109,10 @@ impl ManagedDirectory {
     /// If a file cannot be deleted (for permission reasons for instance)
     /// an error is simply logged, and the file remains in the list of managed
     /// files.
-    pub fn garbage_collect<L: FnOnce() -> HashSet<PathBuf>>(&mut self, get_living_files: L) {
+    pub fn garbage_collect<L: FnOnce() -> HashSet<PathBuf>>(
+        &mut self,
+        get_living_files: L,
+    ) -> crate::Result<GarbageCollectionResult> {
         info!("Garbage collect");
         let mut files_to_delete = vec![];
 
@@ -130,19 +138,25 @@ impl ManagedDirectory {
             // 2) writer change meta.json (for instance after a merge or a commit)
             // 3) gc kicks in.
             // 4) gc removes a file that was useful for process B, before process B opened it.
-            if let Ok(_meta_lock) = self.acquire_lock(&META_LOCK) {
+            match self.acquire_lock(&META_LOCK) {
-                let living_files = get_living_files();
+                Ok(_meta_lock) => {
-                for managed_path in &meta_informations_rlock.managed_paths {
+                    let living_files = get_living_files();
-                    if !living_files.contains(managed_path) {
+                    for managed_path in &meta_informations_rlock.managed_paths {
-                        files_to_delete.push(managed_path.clone());
+                        if !living_files.contains(managed_path) {
+                            files_to_delete.push(managed_path.clone());
+                        }
                     }
                 }
-            } else {
+                Err(err) => {
-                error!("Failed to acquire lock for GC");
+                    error!("Failed to acquire lock for GC");
+                    return Err(crate::TantivyError::from(err));
+                }
             }
         }
 
+        let mut failed_to_delete_files = vec![];
         let mut deleted_files = vec![];
+
         for file_to_delete in files_to_delete {
             match self.delete(&file_to_delete) {
                 Ok(_) => {
@@ -152,9 +166,10 @@ impl ManagedDirectory {
                 Err(file_error) => {
                     match file_error {
                         DeleteError::FileDoesNotExist(_) => {
-                            deleted_files.push(file_to_delete);
+                            deleted_files.push(file_to_delete.clone());
                         }
                         DeleteError::IOError(_) => {
+                            failed_to_delete_files.push(file_to_delete.clone());
                             if !cfg!(target_os = "windows") {
                                 // On windows, delete is expected to fail if the file
                                 // is mmapped.
@@ -177,10 +192,13 @@ impl ManagedDirectory {
             for delete_file in &deleted_files {
                 managed_paths_write.remove(delete_file);
             }
-            if save_managed_paths(self.directory.as_mut(), &meta_informations_wlock).is_err() {
+            save_managed_paths(self.directory.as_mut(), &meta_informations_wlock)?;
-                error!("Failed to save the list of managed files.");
-            }
         }
+
+        Ok(GarbageCollectionResult {
+            deleted_files,
+            failed_to_delete_files,
+        })
     }
 
     /// Registers a file as managed
@@ -247,8 +265,9 @@ impl ManagedDirectory {
 impl Directory for ManagedDirectory {
     fn open_read(&self, path: &Path) -> result::Result<ReadOnlySource, OpenReadError> {
         let read_only_source = self.directory.open_read(path)?;
-        let (_footer, reader) = Footer::extract_footer(read_only_source)
+        let (footer, reader) = Footer::extract_footer(read_only_source)
             .map_err(|err| IOError::with_path(path.to_path_buf(), err))?;
+        footer.is_compatible()?;
         Ok(reader)
     }
 
@@ -328,7 +347,7 @@ mod tests_mmap_specific {
             assert!(managed_directory.exists(test_path1));
             assert!(managed_directory.exists(test_path2));
             let living_files: HashSet<PathBuf> = [test_path1.to_owned()].iter().cloned().collect();
-            managed_directory.garbage_collect(|| living_files);
+            assert!(managed_directory.garbage_collect(|| living_files).is_ok());
             assert!(managed_directory.exists(test_path1));
             assert!(!managed_directory.exists(test_path2));
         }
@@ -338,7 +357,7 @@ mod tests_mmap_specific {
             assert!(managed_directory.exists(test_path1));
             assert!(!managed_directory.exists(test_path2));
             let living_files: HashSet<PathBuf> = HashSet::new();
-            managed_directory.garbage_collect(|| living_files);
+            assert!(managed_directory.garbage_collect(|| living_files).is_ok());
             assert!(!managed_directory.exists(test_path1));
             assert!(!managed_directory.exists(test_path2));
         }
@@ -360,7 +379,9 @@ mod tests_mmap_specific {
         assert!(managed_directory.exists(test_path1));
 
         let _mmap_read = managed_directory.open_read(test_path1).unwrap();
-        managed_directory.garbage_collect(|| living_files.clone());
+        assert!(managed_directory
+            .garbage_collect(|| living_files.clone())
+            .is_ok());
         if cfg!(target_os = "windows") {
             // On Windows, gc should try and fail the file as it is mmapped.
             assert!(managed_directory.exists(test_path1));
@@ -368,7 +389,7 @@ mod tests_mmap_specific {
             drop(_mmap_read);
             // The file should still be in the list of managed file and
             // eventually be deleted once mmap is released.
-            managed_directory.garbage_collect(|| living_files);
+            assert!(managed_directory.garbage_collect(|| living_files).is_ok());
             assert!(!managed_directory.exists(test_path1));
         } else {
             assert!(!managed_directory.exists(test_path1));
@@ -393,6 +414,8 @@ mod tests_mmap_specific {
         write.write_all(&[3u8, 4u8, 5u8]).unwrap();
         write.terminate().unwrap();
 
+        let read_source = managed_directory.open_read(test_path2).unwrap();
+        assert_eq!(read_source.as_slice(), &[3u8, 4u8, 5u8]);
         assert!(managed_directory.list_damaged().unwrap().is_empty());
 
         let mut corrupted_path = tempdir_path.clone();

src/directory/mmap_directory.rs

@@ -1,10 +1,3 @@
-use fs2;
-use notify;
-
-use self::fs2::FileExt;
-use self::notify::RawEvent;
-use self::notify::RecursiveMode;
-use self::notify::Watcher;
 use crate::core::META_FILEPATH;
 use crate::directory::error::LockError;
 use crate::directory::error::{
@@ -20,8 +13,12 @@ use crate::directory::WatchCallback;
 use crate::directory::WatchCallbackList;
 use crate::directory::WatchHandle;
 use crate::directory::{TerminatingWrite, WritePtr};
-use atomicwrites;
+use fs2::FileExt;
 use memmap::Mmap;
+use notify::RawEvent;
+use notify::RecursiveMode;
+use notify::Watcher;
+use serde::{Deserialize, Serialize};
 use std::collections::HashMap;
 use std::convert::From;
 use std::fmt;
@@ -131,14 +128,13 @@ impl MmapCache {
         }
         self.cache.remove(full_path);
         self.counters.miss += 1;
-        Ok(if let Some(mmap) = open_mmap(full_path)? {
+        let mmap_opt = open_mmap(full_path)?;
+        Ok(mmap_opt.map(|mmap| {
             let mmap_arc: Arc<BoxedData> = Arc::new(Box::new(mmap));
             let mmap_weak = Arc::downgrade(&mmap_arc);
             self.cache.insert(full_path.to_owned(), mmap_weak);
-            Some(mmap_arc)
+            mmap_arc
-        } else {
+        }))
-            None
-        })
     }
 }
 
@@ -174,7 +170,7 @@ impl WatcherWrapper {
                             // We might want to be more accurate than this at one point.
                             if let Some(filename) = changed_path.file_name() {
                                 if filename == *META_FILEPATH {
-                                    watcher_router_clone.broadcast();
+                                    let _ = watcher_router_clone.broadcast();
                                 }
                             }
                         }
@@ -224,17 +220,13 @@ struct MmapDirectoryInner {
 }
 
 impl MmapDirectoryInner {
-    fn new(
+    fn new(root_path: PathBuf, temp_directory: Option<TempDir>) -> MmapDirectoryInner {
-        root_path: PathBuf,
+        MmapDirectoryInner {
-        temp_directory: Option<TempDir>,
-    ) -> Result<MmapDirectoryInner, OpenDirectoryError> {
-        let mmap_directory_inner = MmapDirectoryInner {
             root_path,
             mmap_cache: Default::default(),
             _temp_directory: temp_directory,
             watcher: RwLock::new(None),
-        };
+        }
-        Ok(mmap_directory_inner)
     }
 
     fn watch(&self, watch_callback: WatchCallback) -> crate::Result<WatchHandle> {
@@ -268,14 +260,11 @@ impl fmt::Debug for MmapDirectory {
 }
 
 impl MmapDirectory {
-    fn new(
+    fn new(root_path: PathBuf, temp_directory: Option<TempDir>) -> MmapDirectory {
-        root_path: PathBuf,
+        let inner = MmapDirectoryInner::new(root_path, temp_directory);
-        temp_directory: Option<TempDir>,
+        MmapDirectory {
-    ) -> Result<MmapDirectory, OpenDirectoryError> {
-        let inner = MmapDirectoryInner::new(root_path, temp_directory)?;
-        Ok(MmapDirectory {
             inner: Arc::new(inner),
-        })
+        }
     }
 
     /// Creates a new MmapDirectory in a temporary directory.
@@ -285,7 +274,7 @@ impl MmapDirectory {
     pub fn create_from_tempdir() -> Result<MmapDirectory, OpenDirectoryError> {
         let tempdir = TempDir::new().map_err(OpenDirectoryError::IoError)?;
         let tempdir_path = PathBuf::from(tempdir.path());
-        MmapDirectory::new(tempdir_path, Some(tempdir))
+        Ok(MmapDirectory::new(tempdir_path, Some(tempdir)))
     }
 
     /// Opens a MmapDirectory in a directory.
@@ -303,7 +292,7 @@ impl MmapDirectory {
                 directory_path,
             )))
         } else {
-            Ok(MmapDirectory::new(PathBuf::from(directory_path), None)?)
+            Ok(MmapDirectory::new(PathBuf::from(directory_path), None))
         }
     }
 
@@ -538,16 +527,15 @@ mod tests {
     // The following tests are specific to the MmapDirectory
 
     use super::*;
+    use crate::indexer::LogMergePolicy;
     use crate::schema::{Schema, SchemaBuilder, TEXT};
     use crate::Index;
     use crate::ReloadPolicy;
     use std::fs;
     use std::sync::atomic::{AtomicUsize, Ordering};
-    use std::thread;
-    use std::time::Duration;
 
     #[test]
-    fn test_open_non_existant_path() {
+    fn test_open_non_existent_path() {
         assert!(MmapDirectory::open(PathBuf::from("./nowhere")).is_err());
     }
 
@@ -640,13 +628,18 @@ mod tests {
         let tmp_dir = tempfile::TempDir::new().unwrap();
         let tmp_dirpath = tmp_dir.path().to_owned();
         let mut watch_wrapper = WatcherWrapper::new(&tmp_dirpath).unwrap();
-        let tmp_file = tmp_dirpath.join("coucou");
+        let tmp_file = tmp_dirpath.join(*META_FILEPATH);
         let _handle = watch_wrapper.watch(Box::new(move || {
             counter_clone.fetch_add(1, Ordering::SeqCst);
         }));
+        let (sender, receiver) = crossbeam::channel::unbounded();
+        let _handle2 = watch_wrapper.watch(Box::new(move || {
+            let _ = sender.send(());
+        }));
         assert_eq!(counter.load(Ordering::SeqCst), 0);
         fs::write(&tmp_file, b"whateverwilldo").unwrap();
-        thread::sleep(Duration::new(0, 1_000u32));
+        assert!(receiver.recv().is_ok());
+        assert!(counter.load(Ordering::SeqCst) >= 1);
     }
 
     #[test]
@@ -655,34 +648,42 @@ mod tests {
         let mut schema_builder: SchemaBuilder = Schema::builder();
         let text_field = schema_builder.add_text_field("text", TEXT);
         let schema = schema_builder.build();
+
         {
             let index = Index::create(mmap_directory.clone(), schema).unwrap();
+
             let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-            for _num_commits in 0..16 {
+            let mut log_merge_policy = LogMergePolicy::default();
+            log_merge_policy.set_min_merge_size(3);
+            index_writer.set_merge_policy(Box::new(log_merge_policy));
+            for _num_commits in 0..10 {
                 for _ in 0..10 {
                     index_writer.add_document(doc!(text_field=>"abc"));
                 }
                 index_writer.commit().unwrap();
             }
+
             let reader = index
                 .reader_builder()
                 .reload_policy(ReloadPolicy::Manual)
                 .try_into()
                 .unwrap();
-            for _ in 0..30 {
+
+            for _ in 0..4 {
                 index_writer.add_document(doc!(text_field=>"abc"));
                 index_writer.commit().unwrap();
                 reader.reload().unwrap();
             }
             index_writer.wait_merging_threads().unwrap();
+
             reader.reload().unwrap();
             let num_segments = reader.searcher().segment_readers().len();
-            assert_eq!(num_segments, 4);
+            assert!(num_segments <= 4);
             assert_eq!(
                 num_segments * 7,
                 mmap_directory.get_cache_info().mmapped.len()
             );
         }
-        assert_eq!(mmap_directory.get_cache_info().mmapped.len(), 0);
+        assert!(mmap_directory.get_cache_info().mmapped.is_empty());
     }
 }

src/directory/mod.rs

@@ -23,9 +23,22 @@ pub use self::directory::{Directory, DirectoryClone};
 pub use self::directory_lock::{Lock, INDEX_WRITER_LOCK, META_LOCK};
 pub use self::ram_directory::RAMDirectory;
 pub use self::read_only_source::ReadOnlySource;
-pub(crate) use self::watch_event_router::WatchCallbackList;
+pub use self::watch_event_router::{WatchCallback, WatchCallbackList, WatchHandle};
-pub use self::watch_event_router::{WatchCallback, WatchHandle};
 use std::io::{self, BufWriter, Write};
+use std::path::PathBuf;
+/// Outcome of the Garbage collection
+pub struct GarbageCollectionResult {
+    /// List of files that were deleted in this cycle
+    pub deleted_files: Vec<PathBuf>,
+    /// List of files that were schedule to be deleted in this cycle,
+    /// but deletion did not work. This typically happens on windows,
+    /// as deleting a memory mapped file is forbidden.
+    ///
+    /// If a searcher is still held, a file cannot be deleted.
+    /// This is not considered a bug, the file will simply be deleted
+    /// in the next GC.
+    pub failed_to_delete_files: Vec<PathBuf>,
+}
 
 #[cfg(feature = "mmap")]
 pub use self::mmap_directory::MmapDirectory;
@@ -33,6 +46,9 @@ pub use self::mmap_directory::MmapDirectory;
 pub use self::managed_directory::ManagedDirectory;
 
 /// Struct used to prevent from calling [`terminate_ref`](trait.TerminatingWrite#method.terminate_ref) directly
+///
+/// The point is that while the type is public, it cannot be built by anyone
+/// outside of this module.
 pub struct AntiCallToken(());
 
 /// Trait used to indicate when no more write need to be done on a writer
@@ -63,6 +79,13 @@ impl<W: TerminatingWrite> TerminatingWrite for BufWriter<W> {
     }
 }
 
+#[cfg(test)]
+impl<'a> TerminatingWrite for &'a mut Vec<u8> {
+    fn terminate_ref(&mut self, _a: AntiCallToken) -> io::Result<()> {
+        self.flush()
+    }
+}
+
 /// Write object for Directory.
 ///
 /// `WritePtr` are required to implement both Write

src/directory/ram_directory.rs

@@ -144,6 +144,22 @@ impl RAMDirectory {
     pub fn total_mem_usage(&self) -> usize {
         self.fs.read().unwrap().total_mem_usage()
     }
+
+    /// Write a copy of all of the files saved in the RAMDirectory in the target `Directory`.
+    ///
+    /// Files are all written using the `Directory::write` meaning, even if they were
+    /// written using the `atomic_write` api.
+    ///
+    /// If an error is encounterred, files may be persisted partially.
+    pub fn persist(&self, dest: &mut dyn Directory) -> crate::Result<()> {
+        let wlock = self.fs.write().unwrap();
+        for (path, source) in wlock.fs.iter() {
+            let mut dest_wrt = dest.open_write(path)?;
+            dest_wrt.write_all(source.as_slice())?;
+            dest_wrt.terminate()?;
+        }
+        Ok(())
+    }
 }
 
 impl Directory for RAMDirectory {
@@ -191,11 +207,11 @@ impl Directory for RAMDirectory {
         // Reserve the path to prevent calls to .write() to succeed.
         self.fs.write().unwrap().write(path_buf.clone(), &[]);
 
-        let mut vec_writer = VecWriter::new(path_buf.clone(), self.clone());
+        let mut vec_writer = VecWriter::new(path_buf, self.clone());
         vec_writer.write_all(data)?;
         vec_writer.flush()?;
         if path == Path::new(&*META_FILEPATH) {
-            self.fs.write().unwrap().watch_router.broadcast();
+            let _ = self.fs.write().unwrap().watch_router.broadcast();
         }
         Ok(())
     }
@@ -204,3 +220,28 @@ impl Directory for RAMDirectory {
         Ok(self.fs.write().unwrap().watch(watch_callback))
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::RAMDirectory;
+    use crate::Directory;
+    use std::io::Write;
+    use std::path::Path;
+
+    #[test]
+    fn test_persist() {
+        let msg_atomic: &'static [u8] = b"atomic is the way";
+        let msg_seq: &'static [u8] = b"sequential is the way";
+        let path_atomic: &'static Path = Path::new("atomic");
+        let path_seq: &'static Path = Path::new("seq");
+        let mut directory = RAMDirectory::create();
+        assert!(directory.atomic_write(path_atomic, msg_atomic).is_ok());
+        let mut wrt = directory.open_write(path_seq).unwrap();
+        assert!(wrt.write_all(msg_seq).is_ok());
+        assert!(wrt.flush().is_ok());
+        let mut directory_copy = RAMDirectory::create();
+        assert!(directory.persist(&mut directory_copy).is_ok());
+        assert_eq!(directory_copy.atomic_read(path_atomic).unwrap(), msg_atomic);
+        assert_eq!(directory_copy.atomic_read(path_seq).unwrap(), msg_seq);
+    }
+}

src/directory/read_only_source.rs

@@ -70,6 +70,12 @@ impl ReadOnlySource {
         (left, right)
     }
 
+    /// Splits into 2 `ReadOnlySource`, at the offset `end - right_len`.
+    pub fn split_from_end(self, right_len: usize) -> (ReadOnlySource, ReadOnlySource) {
+        let left_len = self.len() - right_len;
+        self.split(left_len)
+    }
+
     /// Creates a ReadOnlySource that is just a
     /// view over a slice of the data.
     ///

src/directory/tests.rs

@@ -1,25 +1,117 @@
 use super::*;
+use futures::channel::oneshot;
+use futures::executor::block_on;
 use std::io::Write;
 use std::mem;
 use std::path::{Path, PathBuf};
-use std::sync::atomic::AtomicUsize;
+use std::sync::atomic::Ordering::SeqCst;
-use std::sync::atomic::Ordering;
+use std::sync::atomic::{AtomicBool, AtomicUsize};
 use std::sync::Arc;
-use std::thread;
-use std::time;
 use std::time::Duration;
 
-#[test]
+#[cfg(feature = "mmap")]
-fn test_ram_directory() {
+mod mmap_directory_tests {
-    let mut ram_directory = RAMDirectory::create();
+    use crate::directory::MmapDirectory;
-    test_directory(&mut ram_directory);
+
+    type DirectoryImpl = MmapDirectory;
+
+    fn make_directory() -> DirectoryImpl {
+        MmapDirectory::create_from_tempdir().unwrap()
+    }
+
+    #[test]
+    fn test_simple() {
+        let mut directory = make_directory();
+        super::test_simple(&mut directory);
+    }
+
+    #[test]
+    fn test_write_create_the_file() {
+        let mut directory = make_directory();
+        super::test_write_create_the_file(&mut directory);
+    }
+
+    #[test]
+    fn test_rewrite_forbidden() {
+        let mut directory = make_directory();
+        super::test_rewrite_forbidden(&mut directory);
+    }
+
+    #[test]
+    fn test_directory_delete() {
+        let mut directory = make_directory();
+        super::test_directory_delete(&mut directory);
+    }
+
+    #[test]
+    fn test_lock_non_blocking() {
+        let mut directory = make_directory();
+        super::test_lock_non_blocking(&mut directory);
+    }
+
+    #[test]
+    fn test_lock_blocking() {
+        let mut directory = make_directory();
+        super::test_lock_blocking(&mut directory);
+    }
+
+    #[test]
+    fn test_watch() {
+        let mut directory = make_directory();
+        super::test_watch(&mut directory);
+    }
 }
 
-#[test]
+mod ram_directory_tests {
-#[cfg(feature = "mmap")]
+    use crate::directory::RAMDirectory;
-fn test_mmap_directory() {
+
-    let mut mmap_directory = MmapDirectory::create_from_tempdir().unwrap();
+    type DirectoryImpl = RAMDirectory;
-    test_directory(&mut mmap_directory);
+
+    fn make_directory() -> DirectoryImpl {
+        RAMDirectory::default()
+    }
+
+    #[test]
+    fn test_simple() {
+        let mut directory = make_directory();
+        super::test_simple(&mut directory);
+    }
+
+    #[test]
+    fn test_write_create_the_file() {
+        let mut directory = make_directory();
+        super::test_write_create_the_file(&mut directory);
+    }
+
+    #[test]
+    fn test_rewrite_forbidden() {
+        let mut directory = make_directory();
+        super::test_rewrite_forbidden(&mut directory);
+    }
+
+    #[test]
+    fn test_directory_delete() {
+        let mut directory = make_directory();
+        super::test_directory_delete(&mut directory);
+    }
+
+    #[test]
+    fn test_lock_non_blocking() {
+        let mut directory = make_directory();
+        super::test_lock_non_blocking(&mut directory);
+    }
+
+    #[test]
+    fn test_lock_blocking() {
+        let mut directory = make_directory();
+        super::test_lock_blocking(&mut directory);
+    }
+
+    #[test]
+    fn test_watch() {
+        let mut directory = make_directory();
+        super::test_watch(&mut directory);
+    }
 }
 
 #[test]
@@ -99,48 +191,39 @@ fn test_directory_delete(directory: &mut dyn Directory) {
     assert!(directory.delete(&test_path).is_err());
 }
 
-fn test_directory(directory: &mut dyn Directory) {
-    test_simple(directory);
-    test_rewrite_forbidden(directory);
-    test_write_create_the_file(directory);
-    test_directory_delete(directory);
-    test_lock_non_blocking(directory);
-    test_lock_blocking(directory);
-    test_watch(directory);
-}
-
 fn test_watch(directory: &mut dyn Directory) {
+    let num_progress: Arc<AtomicUsize> = Default::default();
     let counter: Arc<AtomicUsize> = Default::default();
     let counter_clone = counter.clone();
+    let (sender, receiver) = crossbeam::channel::unbounded();
     let watch_callback = Box::new(move || {
-        counter_clone.fetch_add(1, Ordering::SeqCst);
+        counter_clone.fetch_add(1, SeqCst);
     });
-    assert!(directory
+    // This callback is used to synchronize watching in our unit test.
-        .atomic_write(Path::new("meta.json"), b"random_test_data")
+    // We bind it to a variable because the callback is removed when that
-        .is_ok());
+    // handle is dropped.
-    thread::sleep(Duration::new(0, 10_000));
-    assert_eq!(0, counter.load(Ordering::SeqCst));
-
     let watch_handle = directory.watch(watch_callback).unwrap();
+    let _progress_listener = directory
+        .watch(Box::new(move || {
+            let val = num_progress.fetch_add(1, SeqCst);
+            let _ = sender.send(val);
+        }))
+        .unwrap();
+
     for i in 0..10 {
-        assert_eq!(i, counter.load(Ordering::SeqCst));
+        assert_eq!(i, counter.load(SeqCst));
         assert!(directory
             .atomic_write(Path::new("meta.json"), b"random_test_data_2")
             .is_ok());
-        for _ in 0..1_000 {
+        assert_eq!(receiver.recv_timeout(Duration::from_millis(500)), Ok(i));
-            if counter.load(Ordering::SeqCst) > i {
+        assert_eq!(i + 1, counter.load(SeqCst));
-                break;
-            }
-            thread::sleep(Duration::from_millis(10));
-        }
-        assert_eq!(i + 1, counter.load(Ordering::SeqCst));
     }
     mem::drop(watch_handle);
     assert!(directory
         .atomic_write(Path::new("meta.json"), b"random_test_data")
         .is_ok());
-    thread::sleep(Duration::from_millis(200));
+    assert!(receiver.recv_timeout(Duration::from_millis(500)).is_ok());
-    assert_eq!(10, counter.load(Ordering::SeqCst));
+    assert_eq!(10, counter.load(SeqCst));
 }
 
 fn test_lock_non_blocking(directory: &mut dyn Directory) {
@@ -174,9 +257,13 @@ fn test_lock_blocking(directory: &mut dyn Directory) {
         is_blocking: true,
     });
     assert!(lock_a_res.is_ok());
+    let in_thread = Arc::new(AtomicBool::default());
+    let in_thread_clone = in_thread.clone();
+    let (sender, receiver) = oneshot::channel();
     std::thread::spawn(move || {
         //< lock_a_res is sent to the thread.
-        std::thread::sleep(time::Duration::from_millis(10));
+        in_thread_clone.store(true, SeqCst);
+        let _just_sync = block_on(receiver);
         // explicitely droping lock_a_res. It would have been sufficient to just force it
         // to be part of the move, but the intent seems clearer that way.
         drop(lock_a_res);
@@ -189,14 +276,18 @@ fn test_lock_blocking(directory: &mut dyn Directory) {
         });
         assert!(lock_a_res.is_err());
     }
-    {
+    let directory_clone = directory.box_clone();
-        // the blocking call should wait for at least 10ms.
+    let (sender2, receiver2) = oneshot::channel();
-        let start = time::Instant::now();
+    let join_handle = std::thread::spawn(move || {
-        let lock_a_res = directory.acquire_lock(&Lock {
+        assert!(sender2.send(()).is_ok());
+        let lock_a_res = directory_clone.acquire_lock(&Lock {
             filepath: PathBuf::from("a.lock"),
             is_blocking: true,
         });
+        assert!(in_thread.load(SeqCst));
         assert!(lock_a_res.is_ok());
-        assert!(start.elapsed().subsec_millis() >= 10);
+    });
-    }
+    assert!(block_on(receiver2).is_ok());
+    assert!(sender.send(()).is_ok());
+    assert!(join_handle.join().is_ok());
 }

src/directory/watch_event_router.rs

@@ -1,9 +1,11 @@
+use futures::channel::oneshot;
+use futures::{Future, TryFutureExt};
 use std::sync::Arc;
 use std::sync::RwLock;
 use std::sync::Weak;
 
 /// Type alias for callbacks registered when watching files of a `Directory`.
-pub type WatchCallback = Box<dyn Fn() -> () + Sync + Send>;
+pub type WatchCallback = Box<dyn Fn() + Sync + Send>;
 
 /// Helper struct to implement the watch method in `Directory` implementations.
 ///
@@ -22,13 +24,20 @@ pub struct WatchCallbackList {
 #[derive(Clone)]
 pub struct WatchHandle(Arc<WatchCallback>);
 
+impl WatchHandle {
+    /// Create a WatchHandle handle.
+    pub fn new(watch_callback: Arc<WatchCallback>) -> WatchHandle {
+        WatchHandle(watch_callback)
+    }
+}
+
 impl WatchCallbackList {
-    /// Suscribes a new callback and returns a handle that controls the lifetime of the callback.
+    /// Subscribes a new callback and returns a handle that controls the lifetime of the callback.
     pub fn subscribe(&self, watch_callback: WatchCallback) -> WatchHandle {
         let watch_callback_arc = Arc::new(watch_callback);
         let watch_callback_weak = Arc::downgrade(&watch_callback_arc);
         self.router.write().unwrap().push(watch_callback_weak);
-        WatchHandle(watch_callback_arc)
+        WatchHandle::new(watch_callback_arc)
     }
 
     fn list_callback(&self) -> Vec<Arc<WatchCallback>> {
@@ -47,14 +56,21 @@ impl WatchCallbackList {
     }
 
     /// Triggers all callbacks
-    pub fn broadcast(&self) {
+    pub fn broadcast(&self) -> impl Future<Output = ()> {
         let callbacks = self.list_callback();
+        let (sender, receiver) = oneshot::channel();
+        let result = receiver.unwrap_or_else(|_| ());
+        if callbacks.is_empty() {
+            let _ = sender.send(());
+            return result;
+        }
         let spawn_res = std::thread::Builder::new()
             .name("watch-callbacks".to_string())
             .spawn(move || {
                 for callback in callbacks {
                     callback();
                 }
+                let _ = sender.send(());
             });
         if let Err(err) = spawn_res {
             error!(
@@ -62,19 +78,17 @@ impl WatchCallbackList {
                 err
             );
         }
+        result
     }
 }
 
 #[cfg(test)]
 mod tests {
     use crate::directory::WatchCallbackList;
+    use futures::executor::block_on;
     use std::mem;
     use std::sync::atomic::{AtomicUsize, Ordering};
     use std::sync::Arc;
-    use std::thread;
-    use std::time::Duration;
-
-    const WAIT_TIME: u64 = 20;
 
     #[test]
     fn test_watch_event_router_simple() {
@@ -84,22 +98,22 @@ mod tests {
         let inc_callback = Box::new(move || {
             counter_clone.fetch_add(1, Ordering::SeqCst);
         });
-        watch_event_router.broadcast();
+        block_on(watch_event_router.broadcast());
         assert_eq!(0, counter.load(Ordering::SeqCst));
         let handle_a = watch_event_router.subscribe(inc_callback);
-        thread::sleep(Duration::from_millis(WAIT_TIME));
         assert_eq!(0, counter.load(Ordering::SeqCst));
-        watch_event_router.broadcast();
+        block_on(watch_event_router.broadcast());
-        thread::sleep(Duration::from_millis(WAIT_TIME));
         assert_eq!(1, counter.load(Ordering::SeqCst));
-        watch_event_router.broadcast();
+        block_on(async {
-        watch_event_router.broadcast();
+            (
-        watch_event_router.broadcast();
+                watch_event_router.broadcast().await,
-        thread::sleep(Duration::from_millis(WAIT_TIME));
+                watch_event_router.broadcast().await,
+                watch_event_router.broadcast().await,
+            )
+        });
         assert_eq!(4, counter.load(Ordering::SeqCst));
         mem::drop(handle_a);
-        watch_event_router.broadcast();
+        block_on(watch_event_router.broadcast());
-        thread::sleep(Duration::from_millis(WAIT_TIME));
         assert_eq!(4, counter.load(Ordering::SeqCst));
     }
 
@@ -115,20 +129,20 @@ mod tests {
         };
         let handle_a = watch_event_router.subscribe(inc_callback(1));
         let handle_a2 = watch_event_router.subscribe(inc_callback(10));
-        thread::sleep(Duration::from_millis(WAIT_TIME));
         assert_eq!(0, counter.load(Ordering::SeqCst));
-        watch_event_router.broadcast();
+        block_on(async {
-        watch_event_router.broadcast();
+            futures::join!(
-        thread::sleep(Duration::from_millis(WAIT_TIME));
+                watch_event_router.broadcast(),
+                watch_event_router.broadcast()
+            )
+        });
         assert_eq!(22, counter.load(Ordering::SeqCst));
         mem::drop(handle_a);
-        watch_event_router.broadcast();
+        block_on(watch_event_router.broadcast());
-        thread::sleep(Duration::from_millis(WAIT_TIME));
         assert_eq!(32, counter.load(Ordering::SeqCst));
         mem::drop(handle_a2);
-        watch_event_router.broadcast();
+        block_on(watch_event_router.broadcast());
-        watch_event_router.broadcast();
+        block_on(watch_event_router.broadcast());
-        thread::sleep(Duration::from_millis(WAIT_TIME));
         assert_eq!(32, counter.load(Ordering::SeqCst));
     }
 
@@ -142,14 +156,15 @@ mod tests {
         });
         let handle_a = watch_event_router.subscribe(inc_callback);
         assert_eq!(0, counter.load(Ordering::SeqCst));
-        watch_event_router.broadcast();
+        block_on(async {
-        watch_event_router.broadcast();
+            let future1 = watch_event_router.broadcast();
-        thread::sleep(Duration::from_millis(WAIT_TIME));
+            let future2 = watch_event_router.broadcast();
+            futures::join!(future1, future2)
+        });
         assert_eq!(2, counter.load(Ordering::SeqCst));
-        thread::sleep(Duration::from_millis(WAIT_TIME));
         mem::drop(handle_a);
-        watch_event_router.broadcast();
+        let _ = watch_event_router.broadcast();
-        thread::sleep(Duration::from_millis(WAIT_TIME));
+        block_on(watch_event_router.broadcast());
         assert_eq!(2, counter.load(Ordering::SeqCst));
     }
 }

src/docset.rs

@@ -1,58 +1,48 @@
-use crate::common::BitSet;
 use crate::fastfield::DeleteBitSet;
 use crate::DocId;
 use std::borrow::Borrow;
 use std::borrow::BorrowMut;
-use std::cmp::Ordering;
 
-/// Expresses the outcome of a call to `DocSet`'s `.skip_next(...)`.
+/// Sentinel value returned when a DocSet has been entirely consumed.
-#[derive(PartialEq, Eq, Debug)]
+///
-pub enum SkipResult {
+/// This is not u32::MAX as one would have expected, due to the lack of SSE2 instructions
-    /// target was in the docset
+/// to compare [u32; 4].
-    Reached,
+pub const TERMINATED: DocId = std::i32::MAX as u32;
-    /// target was not in the docset, skipping stopped as a greater element was found
-    OverStep,
-    /// the docset was entirely consumed without finding the target, nor any
-    /// element greater than the target.
-    End,
-}
 
 /// Represents an iterable set of sorted doc ids.
 pub trait DocSet {
     /// Goes to the next element.
-    /// `.advance(...)` needs to be called a first time to point to the correct
+    ///
-    /// element.
+    /// The DocId of the next element is returned.
-    fn advance(&mut self) -> bool;
+    /// In other words we should always have :
+    /// ```ignore
+    /// let doc = docset.advance();
+    /// assert_eq!(doc, docset.doc());
+    /// ```
+    ///
+    /// If we reached the end of the DocSet, TERMINATED should be returned.
+    ///
+    /// Calling `.advance()` on a terminated DocSet should be supported, and TERMINATED should
+    /// be returned.
+    /// TODO Test existing docsets.
+    fn advance(&mut self) -> DocId;
 
-    /// After skipping, position the iterator in such a way that `.doc()`
+    /// Advances the DocSet forward until reaching the target, or going to the
-    /// will return a value greater than or equal to target.
+    /// lowest DocId greater than the target.
     ///
-    /// SkipResult expresses whether the `target value` was reached, overstepped,
+    /// If the end of the DocSet is reached, TERMINATED is returned.
-    /// or if the `DocSet` was entirely consumed without finding any value
-    /// greater or equal to the `target`.
     ///
-    /// WARNING: Calling skip always advances the docset.
+    /// Calling `.seek(target)` on a terminated DocSet is legal. Implementation
-    /// More specifically, if the docset is already positionned on the target
+    /// of DocSet should support it.
-    /// skipping will advance to the next position and return SkipResult::Overstep.
     ///
-    /// If `.skip_next()` oversteps, then the docset must be positionned correctly
+    /// Calling `seek(TERMINATED)` is also legal and is the normal way to consume a DocSet.
-    /// on an existing document. In other words, `.doc()` should return the first document
+    fn seek(&mut self, target: DocId) -> DocId {
-    /// greater than `DocId`.
+        let mut doc = self.doc();
-    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        debug_assert!(doc <= target);
-        if !self.advance() {
+        while doc < target {
-            return SkipResult::End;
+            doc = self.advance();
-        }
-        loop {
-            match self.doc().cmp(&target) {
-                Ordering::Less => {
-                    if !self.advance() {
-                        return SkipResult::End;
-                    }
-                }
-                Ordering::Equal => return SkipResult::Reached,
-                Ordering::Greater => return SkipResult::OverStep,
-            }
         }
+        doc
     }
 
     /// Fills a given mutable buffer with the next doc ids from the
@@ -71,38 +61,38 @@ pub trait DocSet {
     /// use case where batching. The normal way to
     /// go through the `DocId`'s is to call `.advance()`.
     fn fill_buffer(&mut self, buffer: &mut [DocId]) -> usize {
+        if self.doc() == TERMINATED {
+            return 0;
+        }
         for (i, buffer_val) in buffer.iter_mut().enumerate() {
-            if self.advance() {
+            *buffer_val = self.doc();
-                *buffer_val = self.doc();
+            if self.advance() == TERMINATED {
-            } else {
+                return i + 1;
-                return i;
             }
         }
         buffer.len()
     }
 
     /// Returns the current document
+    /// Right after creating a new DocSet, the docset points to the first document.
+    ///
+    /// If the DocSet is empty, .doc() should return `TERMINATED`.
     fn doc(&self) -> DocId;
 
     /// Returns a best-effort hint of the
     /// length of the docset.
     fn size_hint(&self) -> u32;
 
-    /// Appends all docs to a `bitset`.
-    fn append_to_bitset(&mut self, bitset: &mut BitSet) {
-        while self.advance() {
-            bitset.insert(self.doc());
-        }
-    }
-
     /// Returns the number documents matching.
     /// Calling this method consumes the `DocSet`.
     fn count(&mut self, delete_bitset: &DeleteBitSet) -> u32 {
         let mut count = 0u32;
-        while self.advance() {
+        let mut doc = self.doc();
-            if !delete_bitset.is_deleted(self.doc()) {
+        while doc != TERMINATED {
+            if !delete_bitset.is_deleted(doc) {
                 count += 1u32;
             }
+            doc = self.advance();
         }
         count
     }
@@ -114,22 +104,42 @@ pub trait DocSet {
     /// given by `count()`.
     fn count_including_deleted(&mut self) -> u32 {
         let mut count = 0u32;
-        while self.advance() {
+        let mut doc = self.doc();
+        while doc != TERMINATED {
             count += 1u32;
+            doc = self.advance();
         }
         count
     }
 }
 
+impl<'a> DocSet for &'a mut dyn DocSet {
+    fn advance(&mut self) -> u32 {
+        (**self).advance()
+    }
+
+    fn seek(&mut self, target: DocId) -> DocId {
+        (**self).seek(target)
+    }
+
+    fn doc(&self) -> u32 {
+        (**self).doc()
+    }
+
+    fn size_hint(&self) -> u32 {
+        (**self).size_hint()
+    }
+}
+
 impl<TDocSet: DocSet + ?Sized> DocSet for Box<TDocSet> {
-    fn advance(&mut self) -> bool {
+    fn advance(&mut self) -> DocId {
         let unboxed: &mut TDocSet = self.borrow_mut();
         unboxed.advance()
     }
 
-    fn skip_next(&mut self, target: DocId) -> SkipResult {
+    fn seek(&mut self, target: DocId) -> DocId {
         let unboxed: &mut TDocSet = self.borrow_mut();
-        unboxed.skip_next(target)
+        unboxed.seek(target)
     }
 
     fn doc(&self) -> DocId {
@@ -151,9 +161,4 @@ impl<TDocSet: DocSet + ?Sized> DocSet for Box<TDocSet> {
         let unboxed: &mut TDocSet = self.borrow_mut();
         unboxed.count_including_deleted()
     }
-
-    fn append_to_bitset(&mut self, bitset: &mut BitSet) {
-        let unboxed: &mut TDocSet = self.borrow_mut();
-        unboxed.append_to_bitset(bitset);
-    }
 }

src/error.rs

@@ -2,12 +2,11 @@
 
 use std::io;
 
-use crate::directory::error::LockError;
 use crate::directory::error::{IOError, OpenDirectoryError, OpenReadError, OpenWriteError};
+use crate::directory::error::{Incompatibility, LockError};
 use crate::fastfield::FastFieldNotAvailableError;
 use crate::query;
 use crate::schema;
-use serde_json;
 use std::fmt;
 use std::path::PathBuf;
 use std::sync::PoisonError;
@@ -80,6 +79,9 @@ pub enum TantivyError {
     /// System error. (e.g.: We failed spawning a new thread)
     #[fail(display = "System error.'{}'", _0)]
     SystemError(String),
+    /// Index incompatible with current version of tantivy
+    #[fail(display = "{:?}", _0)]
+    IncompatibleIndex(Incompatibility),
 }
 
 impl From<DataCorruption> for TantivyError {
@@ -129,6 +131,9 @@ impl From<OpenReadError> for TantivyError {
         match error {
             OpenReadError::FileDoesNotExist(filepath) => TantivyError::PathDoesNotExist(filepath),
             OpenReadError::IOError(io_error) => TantivyError::IOError(io_error),
+            OpenReadError::IncompatibleIndex(incompatibility) => {
+                TantivyError::IncompatibleIndex(incompatibility)
+            }
         }
     }
 }

src/fastfield/delete.rs

@@ -1,15 +1,16 @@
-use crate::common::HasLen;
+use crate::common::{BitSet, HasLen};
 use crate::directory::ReadOnlySource;
 use crate::directory::WritePtr;
 use crate::space_usage::ByteCount;
 use crate::DocId;
-use bit_set::BitSet;
 use std::io;
 use std::io::Write;
 
 /// Write a delete `BitSet`
 ///
 /// where `delete_bitset` is the set of deleted `DocId`.
+/// Warning: this function does not call terminate. The caller is in charge of
+/// closing the writer properly.
 pub fn write_delete_bitset(
     delete_bitset: &BitSet,
     max_doc: u32,
@@ -17,7 +18,7 @@ pub fn write_delete_bitset(
 ) -> io::Result<()> {
     let mut byte = 0u8;
     let mut shift = 0u8;
-    for doc in 0..(max_doc as usize) {
+    for doc in 0..max_doc {
         if delete_bitset.contains(doc) {
             byte |= 1 << shift;
         }
@@ -32,7 +33,7 @@ pub fn write_delete_bitset(
     if max_doc % 8 > 0 {
         writer.write_all(&[byte])?;
     }
-    writer.flush()
+    Ok(())
 }
 
 /// Set of deleted `DocId`s.
@@ -43,6 +44,24 @@ pub struct DeleteBitSet {
 }
 
 impl DeleteBitSet {
+    #[cfg(test)]
+    pub(crate) fn for_test(docs: &[DocId], max_doc: u32) -> DeleteBitSet {
+        use crate::directory::{Directory, RAMDirectory, TerminatingWrite};
+        use std::path::Path;
+        assert!(docs.iter().all(|&doc| doc < max_doc));
+        let mut bitset = BitSet::with_max_value(max_doc);
+        for &doc in docs {
+            bitset.insert(doc);
+        }
+        let mut directory = RAMDirectory::create();
+        let path = Path::new("dummydeletebitset");
+        let mut wrt = directory.open_write(path).unwrap();
+        write_delete_bitset(&bitset, max_doc, &mut wrt).unwrap();
+        wrt.terminate().unwrap();
+        let source = directory.open_read(path).unwrap();
+        Self::open(source)
+    }
+
     /// Opens a delete bitset given its data source.
     pub fn open(data: ReadOnlySource) -> DeleteBitSet {
         let num_deleted: usize = data
@@ -84,44 +103,35 @@ impl HasLen for DeleteBitSet {
 
 #[cfg(test)]
 mod tests {
-    use super::*;
+    use super::DeleteBitSet;
-    use crate::directory::*;
+    use crate::common::HasLen;
-    use bit_set::BitSet;
-    use std::path::PathBuf;
 
-    fn test_delete_bitset_helper(bitset: &BitSet, max_doc: u32) {
+    #[test]
-        let test_path = PathBuf::from("test");
+    fn test_delete_bitset_empty() {
-        let mut directory = RAMDirectory::create();
+        let delete_bitset = DeleteBitSet::for_test(&[], 10);
-        {
+        for doc in 0..10 {
-            let mut writer = directory.open_write(&*test_path).unwrap();
+            assert_eq!(delete_bitset.is_deleted(doc), !delete_bitset.is_alive(doc));
-            write_delete_bitset(bitset, max_doc, &mut writer).unwrap();
-        }
-        {
-            let source = directory.open_read(&test_path).unwrap();
-            let delete_bitset = DeleteBitSet::open(source);
-            for doc in 0..max_doc as usize {
-                assert_eq!(bitset.contains(doc), delete_bitset.is_deleted(doc as DocId));
-            }
-            assert_eq!(delete_bitset.len(), bitset.len());
         }
+        assert_eq!(delete_bitset.len(), 0);
     }
 
     #[test]
     fn test_delete_bitset() {
-        {
+        let delete_bitset = DeleteBitSet::for_test(&[1, 9], 10);
-            let mut bitset = BitSet::with_capacity(10);
+        assert!(delete_bitset.is_alive(0));
-            bitset.insert(1);
+        assert!(delete_bitset.is_deleted(1));
-            bitset.insert(9);
+        assert!(delete_bitset.is_alive(2));
-            test_delete_bitset_helper(&bitset, 10);
+        assert!(delete_bitset.is_alive(3));
-        }
+        assert!(delete_bitset.is_alive(4));
-        {
+        assert!(delete_bitset.is_alive(5));
-            let mut bitset = BitSet::with_capacity(8);
+        assert!(delete_bitset.is_alive(6));
-            bitset.insert(1);
+        assert!(delete_bitset.is_alive(6));
-            bitset.insert(2);
+        assert!(delete_bitset.is_alive(7));
-            bitset.insert(3);
+        assert!(delete_bitset.is_alive(8));
-            bitset.insert(5);
+        assert!(delete_bitset.is_deleted(9));
-            bitset.insert(7);
+        for doc in 0..10 {
-            test_delete_bitset_helper(&bitset, 8);
+            assert_eq!(delete_bitset.is_deleted(doc), !delete_bitset.is_alive(doc));
         }
+        assert_eq!(delete_bitset.len(), 2);
     }
 }

src/fastfield/facet_reader.rs

@@ -73,7 +73,52 @@ impl FacetReader {
     }
 
     /// Return the list of facet ordinals associated to a document.
-    pub fn facet_ords(&mut self, doc: DocId, output: &mut Vec<u64>) {
+    pub fn facet_ords(&self, doc: DocId, output: &mut Vec<u64>) {
         self.term_ords.get_vals(doc, output);
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use crate::{Document, schema::{Facet, SchemaBuilder}};
+    use crate::Index;
+
+    #[test]
+    fn test_facet_not_populated_for_all_docs() -> crate::Result<()> {
+        let mut schema_builder = SchemaBuilder::default();
+        let facet_field = schema_builder.add_facet_field("facet");
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_with_num_threads(1, 10_000_000)?;
+        index_writer.add_document(doc!(facet_field=>Facet::from_text("/a/b")));
+        index_writer.add_document(Document::default());
+        index_writer.commit()?;
+        let searcher = index.reader()?.searcher();
+        let facet_reader = searcher.segment_reader(0u32).facet_reader(facet_field).unwrap();
+        let mut facet_ords = Vec::new();
+        facet_reader.facet_ords(0u32, &mut facet_ords);
+        assert_eq!(&facet_ords, &[2u64]);
+        facet_reader.facet_ords(1u32, &mut facet_ords);
+        assert!(facet_ords.is_empty());
+        Ok(())
+    }
+    #[test]
+    fn test_facet_not_populated_for_any_docs() -> crate::Result<()> {
+        let mut schema_builder = SchemaBuilder::default();
+        let facet_field = schema_builder.add_facet_field("facet");
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_with_num_threads(1, 10_000_000)?;
+        index_writer.add_document(Document::default());
+        index_writer.add_document(Document::default());
+        index_writer.commit()?;
+        let searcher = index.reader()?.searcher();
+        let facet_reader = searcher.segment_reader(0u32).facet_reader(facet_field).unwrap();
+        let mut facet_ords = Vec::new();
+        facet_reader.facet_ords(0u32, &mut facet_ords);
+        assert!(facet_ords.is_empty());
+        facet_reader.facet_ords(1u32, &mut facet_ords);
+        assert!(facet_ords.is_empty());
+        Ok(())
+    }
+}

src/fastfield/mod.rs

@@ -33,6 +33,7 @@ pub use self::reader::FastFieldReader;
 pub use self::readers::FastFieldReaders;
 pub use self::serializer::FastFieldSerializer;
 pub use self::writer::{FastFieldsWriter, IntFastFieldWriter};
+use crate::chrono::{NaiveDateTime, Utc};
 use crate::common;
 use crate::schema::Cardinality;
 use crate::schema::FieldType;
@@ -49,7 +50,7 @@ mod serializer;
 mod writer;
 
 /// Trait for types that are allowed for fast fields: (u64, i64 and f64).
-pub trait FastValue: Default + Clone + Copy + Send + Sync + PartialOrd {
+pub trait FastValue: Clone + Copy + Send + Sync + PartialOrd {
     /// Converts a value from u64
     ///
     /// Internally all fast field values are encoded as u64.
@@ -69,6 +70,12 @@ pub trait FastValue: Default + Clone + Copy + Send + Sync + PartialOrd {
     /// Cast value to `u64`.
     /// The value is just reinterpreted in memory.
     fn as_u64(&self) -> u64;
+
+    /// Build a default value. This default value is never used, so the value does not
+    /// really matter.
+    fn make_zero() -> Self {
+        Self::from_u64(0i64.to_u64())
+    }
 }
 
 impl FastValue for u64 {
@@ -135,11 +142,34 @@ impl FastValue for f64 {
     }
 }
 
+impl FastValue for crate::DateTime {
+    fn from_u64(timestamp_u64: u64) -> Self {
+        let timestamp_i64 = i64::from_u64(timestamp_u64);
+        crate::DateTime::from_utc(NaiveDateTime::from_timestamp(timestamp_i64, 0), Utc)
+    }
+
+    fn to_u64(&self) -> u64 {
+        self.timestamp().to_u64()
+    }
+
+    fn fast_field_cardinality(field_type: &FieldType) -> Option<Cardinality> {
+        match *field_type {
+            FieldType::Date(ref integer_options) => integer_options.get_fastfield_cardinality(),
+            _ => None,
+        }
+    }
+
+    fn as_u64(&self) -> u64 {
+        self.timestamp().as_u64()
+    }
+}
+
 fn value_to_u64(value: &Value) -> u64 {
     match *value {
         Value::U64(ref val) => *val,
         Value::I64(ref val) => common::i64_to_u64(*val),
         Value::F64(ref val) => common::f64_to_u64(*val),
+        Value::Date(ref datetime) => common::i64_to_u64(datetime.timestamp()),
         _ => panic!("Expected a u64/i64/f64 field, got {:?} ", value),
     }
 }
@@ -151,10 +181,12 @@ mod tests {
     use crate::common::CompositeFile;
     use crate::directory::{Directory, RAMDirectory, WritePtr};
     use crate::fastfield::FastFieldReader;
-    use crate::schema::Document;
+    use crate::merge_policy::NoMergePolicy;
     use crate::schema::Field;
     use crate::schema::Schema;
     use crate::schema::FAST;
+    use crate::schema::{Document, IntOptions};
+    use crate::{Index, SegmentId, SegmentReader};
     use once_cell::sync::Lazy;
     use rand::prelude::SliceRandom;
     use rand::rngs::StdRng;
@@ -178,6 +210,12 @@ mod tests {
         assert_eq!(test_fastfield.get(2), 300);
     }
 
+    #[test]
+    pub fn test_fastfield_i64_u64() {
+        let datetime = crate::DateTime::from_utc(NaiveDateTime::from_timestamp(0i64, 0), Utc);
+        assert_eq!(i64::from_u64(datetime.to_u64()), 0i64);
+    }
+
     #[test]
     fn test_intfastfield_small() {
         let path = Path::new("test");
@@ -429,6 +467,93 @@ mod tests {
             }
         }
     }
+
+    #[test]
+    fn test_merge_missing_date_fast_field() {
+        let mut schema_builder = Schema::builder();
+        let date_field = schema_builder.add_date_field("date", FAST);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        index_writer.set_merge_policy(Box::new(NoMergePolicy));
+        index_writer.add_document(doc!(date_field =>crate::chrono::prelude::Utc::now()));
+        index_writer.commit().unwrap();
+        index_writer.add_document(doc!());
+        index_writer.commit().unwrap();
+        let reader = index.reader().unwrap();
+        let segment_ids: Vec<SegmentId> = reader
+            .searcher()
+            .segment_readers()
+            .iter()
+            .map(SegmentReader::segment_id)
+            .collect();
+        assert_eq!(segment_ids.len(), 2);
+        let merge_future = index_writer.merge(&segment_ids[..]);
+        let merge_res = futures::executor::block_on(merge_future);
+        assert!(merge_res.is_ok());
+        assert!(reader.reload().is_ok());
+        assert_eq!(reader.searcher().segment_readers().len(), 1);
+    }
+
+    #[test]
+    fn test_default_datetime() {
+        assert_eq!(crate::DateTime::make_zero().timestamp(), 0i64);
+    }
+
+    #[test]
+    fn test_datefastfield() {
+        use crate::fastfield::FastValue;
+        let mut schema_builder = Schema::builder();
+        let date_field = schema_builder.add_date_field("date", FAST);
+        let multi_date_field = schema_builder.add_date_field(
+            "multi_date",
+            IntOptions::default().set_fast(Cardinality::MultiValues),
+        );
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        index_writer.set_merge_policy(Box::new(NoMergePolicy));
+        index_writer.add_document(doc!(
+            date_field => crate::DateTime::from_u64(1i64.to_u64()),
+            multi_date_field => crate::DateTime::from_u64(2i64.to_u64()),
+            multi_date_field => crate::DateTime::from_u64(3i64.to_u64())
+        ));
+        index_writer.add_document(doc!(
+            date_field => crate::DateTime::from_u64(4i64.to_u64())
+        ));
+        index_writer.add_document(doc!(
+            multi_date_field => crate::DateTime::from_u64(5i64.to_u64()),
+            multi_date_field => crate::DateTime::from_u64(6i64.to_u64())
+        ));
+        index_writer.commit().unwrap();
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+        assert_eq!(searcher.segment_readers().len(), 1);
+        let segment_reader = searcher.segment_reader(0);
+        let fast_fields = segment_reader.fast_fields();
+        let date_fast_field = fast_fields.date(date_field).unwrap();
+        let dates_fast_field = fast_fields.dates(multi_date_field).unwrap();
+        let mut dates = vec![];
+        {
+            assert_eq!(date_fast_field.get(0u32).timestamp(), 1i64);
+            dates_fast_field.get_vals(0u32, &mut dates);
+            assert_eq!(dates.len(), 2);
+            assert_eq!(dates[0].timestamp(), 2i64);
+            assert_eq!(dates[1].timestamp(), 3i64);
+        }
+        {
+            assert_eq!(date_fast_field.get(1u32).timestamp(), 4i64);
+            dates_fast_field.get_vals(1u32, &mut dates);
+            assert!(dates.is_empty());
+        }
+        {
+            assert_eq!(date_fast_field.get(2u32).timestamp(), 0i64);
+            dates_fast_field.get_vals(2u32, &mut dates);
+            assert_eq!(dates.len(), 2);
+            assert_eq!(dates[0].timestamp(), 5i64);
+            assert_eq!(dates[1].timestamp(), 6i64);
+        }
+    }
 }
 
 #[cfg(all(test, feature = "unstable"))]

src/fastfield/multivalued/mod.rs

@@ -7,9 +7,6 @@ pub use self::writer::MultiValueIntFastFieldWriter;
 #[cfg(test)]
 mod tests {
 
-    use time;
-
-    use self::time::Duration;
     use crate::collector::TopDocs;
     use crate::query::QueryParser;
     use crate::schema::Cardinality;
@@ -17,6 +14,7 @@ mod tests {
     use crate::schema::IntOptions;
     use crate::schema::Schema;
     use crate::Index;
+    use chrono::Duration;
 
     #[test]
     fn test_multivalued_u64() {

src/fastfield/multivalued/reader.rs

@@ -45,7 +45,7 @@ impl<Item: FastValue> MultiValueIntFastFieldReader<Item> {
     pub fn get_vals(&self, doc: DocId, vals: &mut Vec<Item>) {
         let (start, stop) = self.range(doc);
         let len = (stop - start) as usize;
-        vals.resize(len, Item::default());
+        vals.resize(len, Item::make_zero());
         self.vals_reader.get_range_u64(start, &mut vals[..]);
     }
 

src/fastfield/multivalued/writer.rs

@@ -6,7 +6,6 @@ use crate::schema::{Document, Field};
 use crate::termdict::TermOrdinal;
 use crate::DocId;
 use fnv::FnvHashMap;
-use itertools::Itertools;
 use std::io;
 
 /// Writer for multi-valued (as in, more than one value per document)
@@ -151,8 +150,8 @@ impl MultiValueIntFastFieldWriter {
                     }
                 }
                 None => {
-                    let val_min_max = self.vals.iter().cloned().minmax();
+                    let val_min_max = crate::common::minmax(self.vals.iter().cloned());
-                    let (val_min, val_max) = val_min_max.into_option().unwrap_or((0u64, 0u64));
+                    let (val_min, val_max) = val_min_max.unwrap_or((0u64, 0u64));
                     value_serializer =
                         serializer.new_u64_fast_field_with_idx(self.field, val_min, val_max, 1)?;
                     for &val in &self.vals {

src/fastfield/readers.rs

@@ -4,7 +4,6 @@ use crate::fastfield::MultiValueIntFastFieldReader;
 use crate::fastfield::{FastFieldNotAvailableError, FastFieldReader};
 use crate::schema::{Cardinality, Field, FieldType, Schema};
 use crate::space_usage::PerFieldSpaceUsage;
-use crate::Result;
 use std::collections::HashMap;
 
 /// Provides access to all of the FastFieldReader.
@@ -15,9 +14,11 @@ pub struct FastFieldReaders {
     fast_field_i64: HashMap<Field, FastFieldReader<i64>>,
     fast_field_u64: HashMap<Field, FastFieldReader<u64>>,
     fast_field_f64: HashMap<Field, FastFieldReader<f64>>,
+    fast_field_date: HashMap<Field, FastFieldReader<crate::DateTime>>,
     fast_field_i64s: HashMap<Field, MultiValueIntFastFieldReader<i64>>,
     fast_field_u64s: HashMap<Field, MultiValueIntFastFieldReader<u64>>,
     fast_field_f64s: HashMap<Field, MultiValueIntFastFieldReader<f64>>,
+    fast_field_dates: HashMap<Field, MultiValueIntFastFieldReader<crate::DateTime>>,
     fast_bytes: HashMap<Field, BytesFastFieldReader>,
     fast_fields_composite: CompositeFile,
 }
@@ -26,6 +27,7 @@ enum FastType {
     I64,
     U64,
     F64,
+    Date,
 }
 
 fn type_and_cardinality(field_type: &FieldType) -> Option<(FastType, Cardinality)> {
@@ -39,6 +41,9 @@ fn type_and_cardinality(field_type: &FieldType) -> Option<(FastType, Cardinality
         FieldType::F64(options) => options
             .get_fastfield_cardinality()
             .map(|cardinality| (FastType::F64, cardinality)),
+        FieldType::Date(options) => options
+            .get_fastfield_cardinality()
+            .map(|cardinality| (FastType::Date, cardinality)),
         FieldType::HierarchicalFacet => Some((FastType::U64, Cardinality::MultiValues)),
         _ => None,
     }
@@ -48,14 +53,16 @@ impl FastFieldReaders {
     pub(crate) fn load_all(
         schema: &Schema,
         fast_fields_composite: &CompositeFile,
-    ) -> Result<FastFieldReaders> {
+    ) -> crate::Result<FastFieldReaders> {
         let mut fast_field_readers = FastFieldReaders {
             fast_field_i64: Default::default(),
             fast_field_u64: Default::default(),
             fast_field_f64: Default::default(),
+            fast_field_date: Default::default(),
             fast_field_i64s: Default::default(),
             fast_field_u64s: Default::default(),
             fast_field_f64s: Default::default(),
+            fast_field_dates: Default::default(),
             fast_bytes: Default::default(),
             fast_fields_composite: fast_fields_composite.clone(),
         };
@@ -95,6 +102,12 @@ impl FastFieldReaders {
                                         FastFieldReader::open(fast_field_data.clone()),
                                     );
                                 }
+                                FastType::Date => {
+                                    fast_field_readers.fast_field_date.insert(
+                                        field,
+                                        FastFieldReader::open(fast_field_data.clone()),
+                                    );
+                                }
                             }
                         } else {
                             return Err(From::from(FastFieldNotAvailableError::new(field_entry)));
@@ -130,6 +143,14 @@ impl FastFieldReaders {
                                         .fast_field_f64s
                                         .insert(field, multivalued_int_fast_field);
                                 }
+                                FastType::Date => {
+                                    let vals_reader = FastFieldReader::open(fast_field_data);
+                                    let multivalued_int_fast_field =
+                                        MultiValueIntFastFieldReader::open(idx_reader, vals_reader);
+                                    fast_field_readers
+                                        .fast_field_dates
+                                        .insert(field, multivalued_int_fast_field);
+                                }
                             }
                         } else {
                             return Err(From::from(FastFieldNotAvailableError::new(field_entry)));
@@ -156,8 +177,6 @@ impl FastFieldReaders {
     /// If the field is a i64-fast field, return the associated u64 reader. Values are
     /// mapped from i64 to u64 using a (well the, it is unique) monotonic mapping.    ///
     ///
-    ///TODO should it also be lenient with f64?
-    ///
     /// This method is useful when merging segment reader.
     pub(crate) fn u64_lenient(&self, field: Field) -> Option<FastFieldReader<u64>> {
         if let Some(u64_ff_reader) = self.u64(field) {
@@ -166,6 +185,12 @@ impl FastFieldReaders {
         if let Some(i64_ff_reader) = self.i64(field) {
             return Some(i64_ff_reader.into_u64_reader());
         }
+        if let Some(f64_ff_reader) = self.f64(field) {
+            return Some(f64_ff_reader.into_u64_reader());
+        }
+        if let Some(date_ff_reader) = self.date(field) {
+            return Some(date_ff_reader.into_u64_reader());
+        }
         None
     }
 
@@ -176,6 +201,13 @@ impl FastFieldReaders {
         self.fast_field_i64.get(&field).cloned()
     }
 
+    /// Returns the `i64` fast field reader reader associated to `field`.
+    ///
+    /// If `field` is not a i64 fast field, this method returns `None`.
+    pub fn date(&self, field: Field) -> Option<FastFieldReader<crate::DateTime>> {
+        self.fast_field_date.get(&field).cloned()
+    }
+
     /// Returns the `f64` fast field reader reader associated to `field`.
     ///
     /// If `field` is not a f64 fast field, this method returns `None`.
@@ -202,6 +234,9 @@ impl FastFieldReaders {
         if let Some(i64s_ff_reader) = self.i64s(field) {
             return Some(i64s_ff_reader.into_u64s_reader());
         }
+        if let Some(f64s_ff_reader) = self.f64s(field) {
+            return Some(f64s_ff_reader.into_u64s_reader());
+        }
         None
     }
 
@@ -219,6 +254,13 @@ impl FastFieldReaders {
         self.fast_field_f64s.get(&field).cloned()
     }
 
+    /// Returns a `crate::DateTime` multi-valued fast field reader reader associated to `field`.
+    ///
+    /// If `field` is not a `crate::DateTime` multi-valued fast field, this method returns `None`.
+    pub fn dates(&self, field: Field) -> Option<MultiValueIntFastFieldReader<crate::DateTime>> {
+        self.fast_field_dates.get(&field).cloned()
+    }
+
     /// Returns the `bytes` fast field reader associated to `field`.
     ///
     /// If `field` is not a bytes fast field, returns `None`.

src/fastfield/writer.rs

@@ -4,7 +4,7 @@ use crate::common::BinarySerializable;
 use crate::common::VInt;
 use crate::fastfield::{BytesFastFieldWriter, FastFieldSerializer};
 use crate::postings::UnorderedTermId;
-use crate::schema::{Cardinality, Document, Field, FieldType, Schema};
+use crate::schema::{Cardinality, Document, Field, FieldEntry, FieldType, Schema};
 use crate::termdict::TermOrdinal;
 use fnv::FnvHashMap;
 use std::collections::HashMap;
@@ -17,6 +17,14 @@ pub struct FastFieldsWriter {
     bytes_value_writers: Vec<BytesFastFieldWriter>,
 }
 
+fn fast_field_default_value(field_entry: &FieldEntry) -> u64 {
+    match *field_entry.field_type() {
+        FieldType::I64(_) | FieldType::Date(_) => common::i64_to_u64(0i64),
+        FieldType::F64(_) => common::f64_to_u64(0.0f64),
+        _ => 0u64,
+    }
+}
+
 impl FastFieldsWriter {
     /// Create all `FastFieldWriter` required by the schema.
     pub fn from_schema(schema: &Schema) -> FastFieldsWriter {
@@ -25,18 +33,15 @@ impl FastFieldsWriter {
         let mut bytes_value_writers = Vec::new();
 
         for (field, field_entry) in schema.fields() {
-            let default_value = match *field_entry.field_type() {
-                FieldType::I64(_) => common::i64_to_u64(0i64),
-                FieldType::F64(_) => common::f64_to_u64(0.0f64),
-                _ => 0u64,
-            };
             match *field_entry.field_type() {
                 FieldType::I64(ref int_options)
                 | FieldType::U64(ref int_options)
-                | FieldType::F64(ref int_options) => {
+                | FieldType::F64(ref int_options)
+                | FieldType::Date(ref int_options) => {
                     match int_options.get_fastfield_cardinality() {
                         Some(Cardinality::SingleValue) => {
                             let mut fast_field_writer = IntFastFieldWriter::new(field);
+                            let default_value = fast_field_default_value(field_entry);
                             fast_field_writer.set_val_if_missing(default_value);
                             single_value_writers.push(fast_field_writer);
                         }
@@ -121,6 +126,7 @@ impl FastFieldsWriter {
         for field_writer in &self.single_value_writers {
             field_writer.serialize(serializer)?;
         }
+        
         for field_writer in &self.multi_values_writers {
             let field = field_writer.field();
             field_writer.serialize(serializer, mapping.get(&field))?;

src/fieldnorm/mod.rs

@@ -21,7 +21,7 @@ mod reader;
 mod serializer;
 mod writer;
 
-pub use self::reader::FieldNormReader;
+pub use self::reader::{FieldNormReader, FieldNormReaders};
 pub use self::serializer::FieldNormsSerializer;
 pub use self::writer::FieldNormsWriter;
 

src/fieldnorm/reader.rs

@@ -1,6 +1,41 @@
 use super::{fieldnorm_to_id, id_to_fieldnorm};
+use crate::common::CompositeFile;
 use crate::directory::ReadOnlySource;
+use crate::schema::Field;
+use crate::space_usage::PerFieldSpaceUsage;
 use crate::DocId;
+use std::sync::Arc;
+
+/// Reader for the fieldnorm (for each document, the number of tokens indexed in the
+/// field) of all indexed fields in the index.
+///
+/// Each fieldnorm is approximately compressed over one byte. We refer to this byte as
+/// `fieldnorm_id`.
+/// The mapping from `fieldnorm` to `fieldnorm_id` is given by monotonic.
+#[derive(Clone)]
+pub struct FieldNormReaders {
+    data: Arc<CompositeFile>,
+}
+
+impl FieldNormReaders {
+    /// Creates a field norm reader.
+    pub fn open(source: ReadOnlySource) -> crate::Result<FieldNormReaders> {
+        let data = CompositeFile::open(&source)?;
+        Ok(FieldNormReaders {
+            data: Arc::new(data),
+        })
+    }
+
+    /// Returns the FieldNormReader for a specific field.
+    pub fn get_field(&self, field: Field) -> Option<FieldNormReader> {
+        self.data.open_read(field).map(FieldNormReader::open)
+    }
+
+    /// Return a break down of the space usage per field.
+    pub fn space_usage(&self) -> PerFieldSpaceUsage {
+        self.data.space_usage()
+    }
+}
 
 /// Reads the fieldnorm associated to a document.
 /// The fieldnorm represents the length associated to
@@ -19,6 +54,7 @@ use crate::DocId;
 /// Apart from compression, this scale also makes it possible to
 /// precompute computationally expensive functions of the fieldnorm
 /// in a very short array.
+#[derive(Clone)]
 pub struct FieldNormReader {
     data: ReadOnlySource,
 }
@@ -29,6 +65,11 @@ impl FieldNormReader {
         FieldNormReader { data }
     }
 
+    /// Returns the number of documents in this segment.
+    pub fn num_docs(&self) -> u32 {
+        self.data.len() as u32
+    }
+
     /// Returns the `fieldnorm` associated to a doc id.
     /// The fieldnorm is a value approximating the number
     /// of tokens in a given field of the `doc_id`.
@@ -62,13 +103,12 @@ impl FieldNormReader {
     pub fn fieldnorm_to_id(fieldnorm: u32) -> u8 {
         fieldnorm_to_id(fieldnorm)
     }
-}
 
-#[cfg(test)]
+    #[cfg(test)]
-impl From<Vec<u32>> for FieldNormReader {
+    pub fn for_test(field_norms: &[u32]) -> FieldNormReader {
-    fn from(field_norms: Vec<u32>) -> FieldNormReader {
         let field_norms_id = field_norms
-            .into_iter()
+            .iter()
+            .cloned()
             .map(FieldNormReader::fieldnorm_to_id)
             .collect::<Vec<u8>>();
         let field_norms_data = ReadOnlySource::from(field_norms_id);
@@ -77,3 +117,20 @@ impl From<Vec<u32>> for FieldNormReader {
         }
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use crate::fieldnorm::FieldNormReader;
+
+    #[test]
+    fn test_from_fieldnorms_array() {
+        let fieldnorms = &[1, 2, 3, 4, 1_000_000];
+        let fieldnorm_reader = FieldNormReader::for_test(fieldnorms);
+        assert_eq!(fieldnorm_reader.num_docs(), 5);
+        assert_eq!(fieldnorm_reader.fieldnorm(0), 1);
+        assert_eq!(fieldnorm_reader.fieldnorm(1), 2);
+        assert_eq!(fieldnorm_reader.fieldnorm(2), 3);
+        assert_eq!(fieldnorm_reader.fieldnorm(3), 4);
+        assert_eq!(fieldnorm_reader.fieldnorm(4), 983_064);
+    }
+}

src/fieldnorm/writer.rs

@@ -78,11 +78,12 @@ impl FieldNormsWriter {
     }
 
     /// Serialize the seen fieldnorm values to the serializer for all fields.
-    pub fn serialize(&self, fieldnorms_serializer: &mut FieldNormsSerializer) -> io::Result<()> {
+    pub fn serialize(&self, mut fieldnorms_serializer: FieldNormsSerializer) -> io::Result<()> {
         for &field in self.fields.iter() {
             let fieldnorm_values: &[u8] = &self.fieldnorms_buffer[field.field_id() as usize][..];
             fieldnorms_serializer.serialize_field(field, fieldnorm_values)?;
         }
+        fieldnorms_serializer.close()?;
         Ok(())
     }
 }

src/indexer/delete_queue.rs

@@ -2,7 +2,7 @@ use super::operation::DeleteOperation;
 use crate::Opstamp;
 use std::mem;
 use std::ops::DerefMut;
-use std::sync::{Arc, RwLock};
+use std::sync::{Arc, RwLock, Weak};
 
 // The DeleteQueue is similar in conceptually to a multiple
 // consumer single producer broadcast channel.
@@ -14,14 +14,15 @@ use std::sync::{Arc, RwLock};
 //
 // New consumer can be created in two ways
 // - calling `delete_queue.cursor()` returns a cursor, that
-//   will include all future delete operation (and no past operations).
+//   will include all future delete operation (and some or none
+//   of the past operations... The client is in charge of checking the opstamps.).
 // - cloning an existing cursor returns a new cursor, that
 //   is at the exact same position, and can now advance independently
 //   from the original cursor.
 #[derive(Default)]
 struct InnerDeleteQueue {
     writer: Vec<DeleteOperation>,
-    last_block: Option<Arc<Block>>,
+    last_block: Weak<Block>,
 }
 
 #[derive(Clone)]
@@ -32,21 +33,31 @@ pub struct DeleteQueue {
 impl DeleteQueue {
     // Creates a new delete queue.
     pub fn new() -> DeleteQueue {
-        let delete_queue = DeleteQueue {
+        DeleteQueue {
             inner: Arc::default(),
-        };
-
-        let next_block = NextBlock::from(delete_queue.clone());
-
-        {
-            let mut delete_queue_wlock = delete_queue.inner.write().unwrap();
-            delete_queue_wlock.last_block = Some(Arc::new(Block {
-                operations: Arc::default(),
-                next: next_block,
-            }));
         }
+    }
 
-        delete_queue
+    fn get_last_block(&self) -> Arc<Block> {
+        {
+            // try get the last block with simply acquiring the read lock.
+            let rlock = self.inner.read().unwrap();
+            if let Some(block) = rlock.last_block.upgrade() {
+                return block;
+            }
+        }
+        // It failed. Let's double check after acquiring the write, as someone could have called
+        // `get_last_block` right after we released the rlock.
+        let mut wlock = self.inner.write().unwrap();
+        if let Some(block) = wlock.last_block.upgrade() {
+            return block;
+        }
+        let block = Arc::new(Block {
+            operations: Arc::default(),
+            next: NextBlock::from(self.clone()),
+        });
+        wlock.last_block = Arc::downgrade(&block);
+        block
     }
 
     // Creates a new cursor that makes it possible to
@@ -54,17 +65,7 @@ impl DeleteQueue {
     //
     // Past delete operations are not accessible.
     pub fn cursor(&self) -> DeleteCursor {
-        let last_block = self
+        let last_block = self.get_last_block();
-            .inner
-            .read()
-            .expect("Read lock poisoned when opening delete queue cursor")
-            .last_block
-            .clone()
-            .expect(
-                "Failed to unwrap last_block. This should never happen
-                as the Option<> is only here to make
-                initialization possible",
-            );
         let operations_len = last_block.operations.len();
         DeleteCursor {
             block: last_block,
@@ -100,23 +101,19 @@ impl DeleteQueue {
             .write()
             .expect("Failed to acquire write lock on delete queue writer");
 
-        let delete_operations;
+        if self_wlock.writer.is_empty() {
-        {
+            return None;
-            let writer: &mut Vec<DeleteOperation> = &mut self_wlock.writer;
-            if writer.is_empty() {
-                return None;
-            }
-            delete_operations = mem::replace(writer, vec![]);
         }
 
-        let next_block = NextBlock::from(self.clone());
+        let delete_operations = mem::replace(&mut self_wlock.writer, vec![]);
-        {
+
-            self_wlock.last_block = Some(Arc::new(Block {
+        let new_block = Arc::new(Block {
-                operations: Arc::new(delete_operations),
+            operations: Arc::new(delete_operations.into_boxed_slice()),
-                next: next_block,
+            next: NextBlock::from(self.clone()),
-            }));
+        });
-        }
+
-        self_wlock.last_block.clone()
+        self_wlock.last_block = Arc::downgrade(&new_block);
+        Some(new_block)
     }
 }
 
@@ -170,7 +167,7 @@ impl NextBlock {
 }
 
 struct Block {
-    operations: Arc<Vec<DeleteOperation>>,
+    operations: Arc<Box<[DeleteOperation]>>,
     next: NextBlock,
 }
 

src/indexer/index_writer.rs

@@ -1,15 +1,16 @@
 use super::operation::{AddOperation, UserOperation};
 use super::segment_updater::SegmentUpdater;
 use super::PreparedCommit;
+use crate::common::BitSet;
 use crate::core::Index;
 use crate::core::Segment;
 use crate::core::SegmentComponent;
 use crate::core::SegmentId;
 use crate::core::SegmentMeta;
 use crate::core::SegmentReader;
-use crate::directory::DirectoryLock;
 use crate::directory::TerminatingWrite;
-use crate::docset::DocSet;
+use crate::directory::{DirectoryLock, GarbageCollectionResult};
+use crate::docset::{DocSet, TERMINATED};
 use crate::error::TantivyError;
 use crate::fastfield::write_delete_bitset;
 use crate::indexer::delete_queue::{DeleteCursor, DeleteQueue};
@@ -23,10 +24,9 @@ use crate::schema::Document;
 use crate::schema::IndexRecordOption;
 use crate::schema::Term;
 use crate::Opstamp;
-use crate::Result;
-use bit_set::BitSet;
 use crossbeam::channel;
-use futures::{Canceled, Future};
+use futures::executor::block_on;
+use futures::future::Future;
 use smallvec::smallvec;
 use smallvec::SmallVec;
 use std::mem;
@@ -72,7 +72,7 @@ pub struct IndexWriter {
 
     heap_size_in_bytes_per_thread: usize,
 
-    workers_join_handle: Vec<JoinHandle<Result<()>>>,
+    workers_join_handle: Vec<JoinHandle<crate::Result<()>>>,
 
     operation_receiver: OperationReceiver,
     operation_sender: OperationSender,
@@ -95,7 +95,7 @@ fn compute_deleted_bitset(
     delete_cursor: &mut DeleteCursor,
     doc_opstamps: &DocToOpstampMapping,
     target_opstamp: Opstamp,
-) -> Result<bool> {
+) -> crate::Result<bool> {
     let mut might_have_changed = false;
     while let Some(delete_op) = delete_cursor.get() {
         if delete_op.opstamp > target_opstamp {
@@ -112,65 +112,79 @@ fn compute_deleted_bitset(
         if let Some(mut docset) =
             inverted_index.read_postings(&delete_op.term, IndexRecordOption::Basic)
         {
-            while docset.advance() {
+            let mut deleted_doc = docset.doc();
-                let deleted_doc = docset.doc();
+            while deleted_doc != TERMINATED {
                 if deleted_doc < limit_doc {
-                    delete_bitset.insert(deleted_doc as usize);
+                    delete_bitset.insert(deleted_doc);
                     might_have_changed = true;
                 }
+                deleted_doc = docset.advance();
             }
         }
-
         delete_cursor.advance();
     }
     Ok(might_have_changed)
 }
 
-/// Advance delete for the given segment up
+/// Advance delete for the given segment up to the target opstamp.
-/// to the target opstamp.
+///
+/// Note that there are no guarantee that the resulting `segment_entry` delete_opstamp
+/// is `==` target_opstamp.
+/// For instance, there was no delete operation between the state of the `segment_entry` and
+/// the `target_opstamp`, `segment_entry` is not updated.
 pub(crate) fn advance_deletes(
     mut segment: Segment,
     segment_entry: &mut SegmentEntry,
     target_opstamp: Opstamp,
-) -> Result<()> {
+) -> crate::Result<()> {
-    {
+    if segment_entry.meta().delete_opstamp() == Some(target_opstamp) {
-        if segment_entry.meta().delete_opstamp() == Some(target_opstamp) {
+        // We are already up-to-date here.
-            // We are already up-to-date here.
+        return Ok(());
-            return Ok(());
+    }
-        }
 
-        let segment_reader = SegmentReader::open(&segment)?;
+    if segment_entry.delete_bitset().is_none() && segment_entry.delete_cursor().get().is_none() {
+        // There has been no `DeleteOperation` between the segment status and `target_opstamp`.
+        return Ok(());
+    }
 
-        let max_doc = segment_reader.max_doc();
+    let segment_reader = SegmentReader::open(&segment)?;
-        let mut delete_bitset: BitSet = match segment_entry.delete_bitset() {
-            Some(previous_delete_bitset) => (*previous_delete_bitset).clone(),
-            None => BitSet::with_capacity(max_doc as usize),
-        };
 
-        let delete_cursor = segment_entry.delete_cursor();
+    let max_doc = segment_reader.max_doc();
-        compute_deleted_bitset(
+    let mut delete_bitset: BitSet = match segment_entry.delete_bitset() {
-            &mut delete_bitset,
+        Some(previous_delete_bitset) => (*previous_delete_bitset).clone(),
-            &segment_reader,
+        None => BitSet::with_max_value(max_doc),
-            delete_cursor,
+    };
-            &DocToOpstampMapping::None,
-            target_opstamp,
-        )?;
 
-        // TODO optimize
+    let num_deleted_docs_before = segment.meta().num_deleted_docs();
+
+    compute_deleted_bitset(
+        &mut delete_bitset,
+        &segment_reader,
+        segment_entry.delete_cursor(),
+        &DocToOpstampMapping::None,
+        target_opstamp,
+    )?;
+
+    // TODO optimize
+    // It should be possible to do something smarter by manipulation bitsets directly
+    // to compute this union.
+    if let Some(seg_delete_bitset) = segment_reader.delete_bitset() {
         for doc in 0u32..max_doc {
-            if segment_reader.is_deleted(doc) {
+            if seg_delete_bitset.is_deleted(doc) {
-                delete_bitset.insert(doc as usize);
+                delete_bitset.insert(doc);
             }
         }
-
-        let num_deleted_docs = delete_bitset.len();
-        if num_deleted_docs > 0 {
-            segment = segment.with_delete_meta(num_deleted_docs as u32, target_opstamp);
-            let mut delete_file = segment.open_write(SegmentComponent::DELETE)?;
-            write_delete_bitset(&delete_bitset, max_doc, &mut delete_file)?;
-            delete_file.terminate()?;
-        }
     }
+
+    let num_deleted_docs: u32 = delete_bitset.len() as u32;
+    if num_deleted_docs > num_deleted_docs_before {
+        // There are new deletes. We need to write a new delete file.
+        segment = segment.with_delete_meta(num_deleted_docs as u32, target_opstamp);
+        let mut delete_file = segment.open_write(SegmentComponent::DELETE)?;
+        write_delete_bitset(&delete_bitset, max_doc, &mut delete_file)?;
+        delete_file.terminate()?;
+    }
+
     segment_entry.set_meta(segment.meta().clone());
     Ok(())
 }
@@ -181,7 +195,7 @@ fn index_documents(
     grouped_document_iterator: &mut dyn Iterator<Item = OperationGroup>,
     segment_updater: &mut SegmentUpdater,
     mut delete_cursor: DeleteCursor,
-) -> Result<bool> {
+) -> crate::Result<bool> {
     let schema = segment.schema();
 
     let mut segment_writer = SegmentWriter::for_segment(memory_budget, segment.clone(), &schema)?;
@@ -227,7 +241,7 @@ fn index_documents(
         delete_cursor,
         delete_bitset_opt,
     );
-    segment_updater.add_segment(segment_entry);
+    block_on(segment_updater.schedule_add_segment(segment_entry))?;
     Ok(true)
 }
 
@@ -236,7 +250,7 @@ fn apply_deletes(
     mut delete_cursor: &mut DeleteCursor,
     doc_opstamps: &[Opstamp],
     last_docstamp: Opstamp,
-) -> Result<Option<BitSet<u32>>> {
+) -> crate::Result<Option<BitSet>> {
     if delete_cursor.get().is_none() {
         // if there are no delete operation in the queue, no need
         // to even open the segment.
@@ -246,7 +260,7 @@ fn apply_deletes(
     let doc_to_opstamps = DocToOpstampMapping::from(doc_opstamps);
 
     let max_doc = segment.meta().max_doc();
-    let mut deleted_bitset = BitSet::with_capacity(max_doc as usize);
+    let mut deleted_bitset = BitSet::with_max_value(max_doc);
     let may_have_deletes = compute_deleted_bitset(
         &mut deleted_bitset,
         &segment_reader,
@@ -281,7 +295,7 @@ impl IndexWriter {
         num_threads: usize,
         heap_size_in_bytes_per_thread: usize,
         directory_lock: DirectoryLock,
-    ) -> Result<IndexWriter> {
+    ) -> crate::Result<IndexWriter> {
         if heap_size_in_bytes_per_thread < HEAP_SIZE_MIN {
             let err_msg = format!(
                 "The heap size per thread needs to be at least {}.",
@@ -330,12 +344,17 @@ impl IndexWriter {
         Ok(index_writer)
     }
 
+    fn drop_sender(&mut self) {
+        let (sender, _receiver) = channel::bounded(1);
+        self.operation_sender = sender;
+    }
+
     /// If there are some merging threads, blocks until they all finish their work and
     /// then drop the `IndexWriter`.
-    pub fn wait_merging_threads(mut self) -> Result<()> {
+    pub fn wait_merging_threads(mut self) -> crate::Result<()> {
         // this will stop the indexing thread,
         // dropping the last reference to the segment_updater.
-        drop(self.operation_sender);
+        self.drop_sender();
 
         let former_workers_handles = mem::replace(&mut self.workers_join_handle, vec![]);
         for join_handle in former_workers_handles {
@@ -346,7 +365,6 @@ impl IndexWriter {
                     TantivyError::ErrorInThread("Error in indexing worker thread.".into())
                 })?;
         }
-        drop(self.workers_join_handle);
 
         let result = self
             .segment_updater
@@ -361,10 +379,10 @@ impl IndexWriter {
     }
 
     #[doc(hidden)]
-    pub fn add_segment(&mut self, segment_meta: SegmentMeta) {
+    pub fn add_segment(&self, segment_meta: SegmentMeta) -> crate::Result<()> {
         let delete_cursor = self.delete_queue.cursor();
         let segment_entry = SegmentEntry::new(segment_meta, delete_cursor, None);
-        self.segment_updater.add_segment(segment_entry);
+        block_on(self.segment_updater.schedule_add_segment(segment_entry))
     }
 
     /// Creates a new segment.
@@ -381,7 +399,7 @@ impl IndexWriter {
 
     /// Spawns a new worker thread for indexing.
     /// The thread consumes documents from the pipeline.
-    fn add_indexing_worker(&mut self) -> Result<()> {
+    fn add_indexing_worker(&mut self) -> crate::Result<()> {
         let document_receiver_clone = self.operation_receiver.clone();
         let mut segment_updater = self.segment_updater.clone();
 
@@ -389,7 +407,7 @@ impl IndexWriter {
 
         let mem_budget = self.heap_size_in_bytes_per_thread;
         let index = self.index.clone();
-        let join_handle: JoinHandle<Result<()>> = thread::Builder::new()
+        let join_handle: JoinHandle<crate::Result<()>> = thread::Builder::new()
             .name(format!("thrd-tantivy-index{}", self.worker_id))
             .spawn(move || {
                 loop {
@@ -435,22 +453,23 @@ impl IndexWriter {
         self.segment_updater.get_merge_policy()
     }
 
-    /// Set the merge policy.
+    /// Setter for the merge policy.
     pub fn set_merge_policy(&self, merge_policy: Box<dyn MergePolicy>) {
         self.segment_updater.set_merge_policy(merge_policy);
     }
 
-    fn start_workers(&mut self) -> Result<()> {
+    fn start_workers(&mut self) -> crate::Result<()> {
         for _ in 0..self.num_threads {
             self.add_indexing_worker()?;
         }
         Ok(())
     }
 
-    /// Detects and removes the files that
+    /// Detects and removes the files that are not used by the index anymore.
-    /// are not used by the index anymore.
+    pub fn garbage_collect_files(
-    pub fn garbage_collect_files(&mut self) -> Result<()> {
+        &self,
-        self.segment_updater.garbage_collect_files().wait()
+    ) -> impl Future<Output = crate::Result<GarbageCollectionResult>> {
+        self.segment_updater.schedule_garbage_collect()
     }
 
     /// Deletes all documents from the index
@@ -489,7 +508,7 @@ impl IndexWriter {
     ///     Ok(())
     /// }
     /// ```
-    pub fn delete_all_documents(&mut self) -> Result<Opstamp> {
+    pub fn delete_all_documents(&self) -> crate::Result<Opstamp> {
         // Delete segments
         self.segment_updater.remove_all_segments();
         // Return new stamp - reverted stamp
@@ -503,8 +522,10 @@ impl IndexWriter {
     pub fn merge(
         &mut self,
         segment_ids: &[SegmentId],
-    ) -> Result<impl Future<Item = SegmentMeta, Error = Canceled>> {
+    ) -> impl Future<Output = crate::Result<SegmentMeta>> {
-        self.segment_updater.start_merge(segment_ids)
+        let merge_operation = self.segment_updater.make_merge_operation(segment_ids);
+        let segment_updater = self.segment_updater.clone();
+        async move { segment_updater.start_merge(merge_operation)?.await }
     }
 
     /// Closes the current document channel send.
@@ -515,6 +536,7 @@ impl IndexWriter {
     /// when no documents are remaining.
     ///
     /// Returns the former segment_ready channel.
+    #[allow(unused_must_use)]
     fn recreate_document_channel(&mut self) -> OperationReceiver {
         let (document_sender, document_receiver): (OperationSender, OperationReceiver) =
             channel::bounded(PIPELINE_MAX_SIZE_IN_DOCS);
@@ -530,13 +552,8 @@ impl IndexWriter {
     /// state as it was after the last commit.
     ///
     /// The opstamp at the last commit is returned.
-    pub fn rollback(&mut self) -> Result<Opstamp> {
+    pub fn rollback(&mut self) -> crate::Result<Opstamp> {
         info!("Rolling back to opstamp {}", self.committed_opstamp);
-        self.rollback_impl()
-    }
-
-    /// Private, implementation of rollback
-    fn rollback_impl(&mut self) -> Result<Opstamp> {
         // marks the segment updater as killed. From now on, all
         // segment updates will be ignored.
         self.segment_updater.kill();
@@ -559,7 +576,7 @@ impl IndexWriter {
         //
         // This will drop the document queue, and the thread
         // should terminate.
-        mem::replace(self, new_index_writer);
+        *self = new_index_writer;
 
         // Drains the document receiver pipeline :
         // Workers don't need to index the pending documents.
@@ -592,7 +609,7 @@ impl IndexWriter {
     /// It is also possible to add a payload to the `commit`
     /// using this API.
     /// See [`PreparedCommit::set_payload()`](PreparedCommit.html)
-    pub fn prepare_commit(&mut self) -> Result<PreparedCommit<'_>> {
+    pub fn prepare_commit(&mut self) -> crate::Result<PreparedCommit> {
         // Here, because we join all of the worker threads,
         // all of the segment update for this commit have been
         // sent.
@@ -639,7 +656,7 @@ impl IndexWriter {
     /// Commit returns the `opstamp` of the last document
     /// that made it in the commit.
     ///
-    pub fn commit(&mut self) -> Result<Opstamp> {
+    pub fn commit(&mut self) -> crate::Result<Opstamp> {
         self.prepare_commit()?.commit()
     }
 
@@ -680,9 +697,6 @@ impl IndexWriter {
     /// The opstamp is an increasing `u64` that can
     /// be used by the client to align commits with its own
     /// document queue.
-    ///
-    /// Currently it represents the number of documents that
-    /// have been added since the creation of the index.
     pub fn add_document(&self, document: Document) -> Opstamp {
         let opstamp = self.stamper.stamp();
         let add_operation = AddOperation { opstamp, document };
@@ -756,6 +770,16 @@ impl IndexWriter {
     }
 }
 
+impl Drop for IndexWriter {
+    fn drop(&mut self) {
+        self.segment_updater.kill();
+        self.drop_sender();
+        for work in self.workers_join_handle.drain(..) {
+            let _ = work.join();
+        }
+    }
+}
+
 #[cfg(test)]
 mod tests {
 
@@ -765,7 +789,7 @@ mod tests {
     use crate::error::*;
     use crate::indexer::NoMergePolicy;
     use crate::query::TermQuery;
-    use crate::schema::{self, IndexRecordOption};
+    use crate::schema::{self, IndexRecordOption, STRING};
     use crate::Index;
     use crate::ReloadPolicy;
     use crate::Term;
@@ -785,6 +809,46 @@ mod tests {
         assert_eq!(batch_opstamp1, 2u64);
     }
 
+    #[test]
+    fn test_no_need_to_rewrite_delete_file_if_no_new_deletes() {
+        let mut schema_builder = schema::Schema::builder();
+        let text_field = schema_builder.add_text_field("text", schema::TEXT);
+        let index = Index::create_in_ram(schema_builder.build());
+
+        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        index_writer.add_document(doc!(text_field => "hello1"));
+        index_writer.add_document(doc!(text_field => "hello2"));
+        assert!(index_writer.commit().is_ok());
+
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+        assert_eq!(searcher.segment_readers().len(), 1);
+        assert_eq!(searcher.segment_reader(0u32).num_deleted_docs(), 0);
+
+        index_writer.delete_term(Term::from_field_text(text_field, "hello1"));
+        assert!(index_writer.commit().is_ok());
+
+        assert!(reader.reload().is_ok());
+        let searcher = reader.searcher();
+        assert_eq!(searcher.segment_readers().len(), 1);
+        assert_eq!(searcher.segment_reader(0u32).num_deleted_docs(), 1);
+
+        let previous_delete_opstamp = index.load_metas().unwrap().segments[0].delete_opstamp();
+
+        // All docs containing hello1 have been already removed.
+        // We should not update the delete meta.
+        index_writer.delete_term(Term::from_field_text(text_field, "hello1"));
+        assert!(index_writer.commit().is_ok());
+
+        assert!(reader.reload().is_ok());
+        let searcher = reader.searcher();
+        assert_eq!(searcher.segment_readers().len(), 1);
+        assert_eq!(searcher.segment_reader(0u32).num_deleted_docs(), 1);
+
+        let after_delete_opstamp = index.load_metas().unwrap().segments[0].delete_opstamp();
+        assert_eq!(after_delete_opstamp, previous_delete_opstamp);
+    }
+
     #[test]
     fn test_ordered_batched_operations() {
         // * one delete for `doc!(field=>"a")`
@@ -879,7 +943,7 @@ mod tests {
         let index_writer = index.writer(3_000_000).unwrap();
         assert_eq!(
             format!("{:?}", index_writer.get_merge_policy()),
-            "LogMergePolicy { min_merge_size: 8, min_layer_size: 10000, \
+            "LogMergePolicy { min_merge_size: 8, max_merge_size: 10000000, min_layer_size: 10000, \
              level_log_size: 0.75 }"
         );
         let merge_policy = Box::new(NoMergePolicy::default());
@@ -1190,4 +1254,16 @@ mod tests {
         assert!(clear_again.is_ok());
         assert!(commit_again.is_ok());
     }
+
+    #[test]
+    fn test_index_doc_missing_field() {
+        let mut schema_builder = schema::Schema::builder();
+        let idfield = schema_builder.add_text_field("id", STRING);
+        schema_builder.add_text_field("optfield", STRING);
+        let index = Index::create_in_ram(schema_builder.build());
+        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        index_writer.add_document(doc!(idfield=>"myid"));
+        let commit = index_writer.commit();
+        assert!(commit.is_ok());
+    }
 }

src/indexer/log_merge_policy.rs

@@ -6,12 +6,14 @@ use std::f64;
 const DEFAULT_LEVEL_LOG_SIZE: f64 = 0.75;
 const DEFAULT_MIN_LAYER_SIZE: u32 = 10_000;
 const DEFAULT_MIN_MERGE_SIZE: usize = 8;
+const DEFAULT_MAX_MERGE_SIZE: usize = 10_000_000;
 
 /// `LogMergePolicy` tries tries to merge segments that have a similar number of
 /// documents.
 #[derive(Debug, Clone)]
 pub struct LogMergePolicy {
     min_merge_size: usize,
+    max_merge_size: usize,
     min_layer_size: u32,
     level_log_size: f64,
 }
@@ -26,6 +28,12 @@ impl LogMergePolicy {
         self.min_merge_size = min_merge_size;
     }
 
+    /// Set the maximum number docs in a segment for it to be considered for
+    /// merging.
+    pub fn set_max_merge_size(&mut self, max_merge_size: usize) {
+        self.max_merge_size = max_merge_size;
+    }
+
     /// Set the minimum segment size under which all segment belong
     /// to the same level.
     pub fn set_min_layer_size(&mut self, min_layer_size: u32) {
@@ -46,39 +54,44 @@ impl LogMergePolicy {
 
 impl MergePolicy for LogMergePolicy {
     fn compute_merge_candidates(&self, segments: &[SegmentMeta]) -> Vec<MergeCandidate> {
-        if segments.is_empty() {
-            return Vec::new();
-        }
-
         let mut size_sorted_tuples = segments
             .iter()
             .map(SegmentMeta::num_docs)
+            .filter(|s| s <= &(self.max_merge_size as u32))
             .enumerate()
             .collect::<Vec<(usize, u32)>>();
 
         size_sorted_tuples.sort_by(|x, y| y.1.cmp(&(x.1)));
 
+        if size_sorted_tuples.len() <= 1 {
+            return Vec::new();
+        }
+
         let size_sorted_log_tuples: Vec<_> = size_sorted_tuples
             .into_iter()
             .map(|(ind, num_docs)| (ind, f64::from(self.clip_min_size(num_docs)).log2()))
             .collect();
 
-        let (first_ind, first_score) = size_sorted_log_tuples[0];
+        if let Some(&(first_ind, first_score)) = size_sorted_log_tuples.first() {
-        let mut current_max_log_size = first_score;
+            let mut current_max_log_size = first_score;
-        let mut levels = vec![vec![first_ind]];
+            let mut levels = vec![vec![first_ind]];
-        for &(ind, score) in (&size_sorted_log_tuples).iter().skip(1) {
+            for &(ind, score) in (&size_sorted_log_tuples).iter().skip(1) {
-            if score < (current_max_log_size - self.level_log_size) {
+                if score < (current_max_log_size - self.level_log_size) {
-                current_max_log_size = score;
+                    current_max_log_size = score;
-                levels.push(Vec::new());
+                    levels.push(Vec::new());
+                }
+                levels.last_mut().unwrap().push(ind);
             }
-            levels.last_mut().unwrap().push(ind);
+            levels
+                .iter()
+                .filter(|level| level.len() >= self.min_merge_size)
+                .map(|ind_vec| {
+                    MergeCandidate(ind_vec.iter().map(|&ind| segments[ind].id()).collect())
+                })
+                .collect()
+        } else {
+            return vec![];
         }
-
-        levels
-            .iter()
-            .filter(|level| level.len() >= self.min_merge_size)
-            .map(|ind_vec| MergeCandidate(ind_vec.iter().map(|&ind| segments[ind].id()).collect()))
-            .collect()
     }
 }
 
@@ -86,6 +99,7 @@ impl Default for LogMergePolicy {
     fn default() -> LogMergePolicy {
         LogMergePolicy {
             min_merge_size: DEFAULT_MIN_MERGE_SIZE,
+            max_merge_size: DEFAULT_MAX_MERGE_SIZE,
             min_layer_size: DEFAULT_MIN_LAYER_SIZE,
             level_log_size: DEFAULT_LEVEL_LOG_SIZE,
         }
@@ -104,6 +118,7 @@ mod tests {
     fn test_merge_policy() -> LogMergePolicy {
         let mut log_merge_policy = LogMergePolicy::default();
         log_merge_policy.set_min_merge_size(3);
+        log_merge_policy.set_max_merge_size(100_000);
         log_merge_policy.set_min_layer_size(2);
         log_merge_policy
     }
@@ -141,11 +156,11 @@ mod tests {
             create_random_segment_meta(10),
             create_random_segment_meta(10),
             create_random_segment_meta(10),
-            create_random_segment_meta(1000),
+            create_random_segment_meta(1_000),
-            create_random_segment_meta(1000),
+            create_random_segment_meta(1_000),
-            create_random_segment_meta(1000),
+            create_random_segment_meta(1_000),
-            create_random_segment_meta(10000),
+            create_random_segment_meta(10_000),
-            create_random_segment_meta(10000),
+            create_random_segment_meta(10_000),
             create_random_segment_meta(10),
             create_random_segment_meta(10),
             create_random_segment_meta(10),
@@ -168,6 +183,7 @@ mod tests {
         let result_list = test_merge_policy().compute_merge_candidates(&test_input);
         assert_eq!(result_list.len(), 2);
     }
+
     #[test]
     fn test_log_merge_policy_small_segments() {
         // segments under min_layer_size are merged together
@@ -182,4 +198,30 @@ mod tests {
         let result_list = test_merge_policy().compute_merge_candidates(&test_input);
         assert_eq!(result_list.len(), 1);
     }
+
+    #[test]
+    fn test_log_merge_policy_all_segments_too_large_to_merge() {
+        let eight_large_segments: Vec<SegmentMeta> =
+            std::iter::repeat_with(|| create_random_segment_meta(100_001))
+                .take(8)
+                .collect();
+        assert!(test_merge_policy()
+            .compute_merge_candidates(&eight_large_segments)
+            .is_empty());
+    }
+
+    #[test]
+    fn test_large_merge_segments() {
+        let test_input = vec![
+            create_random_segment_meta(1_000_000),
+            create_random_segment_meta(100_001),
+            create_random_segment_meta(100_000),
+            create_random_segment_meta(100_000),
+            create_random_segment_meta(100_000),
+        ];
+        let result_list = test_merge_policy().compute_merge_candidates(&test_input);
+        // Do not include large segments
+        assert_eq!(result_list.len(), 1);
+        assert_eq!(result_list[0].0.len(), 3)
+    }
 }

src/indexer/merge_operation.rs

@@ -2,14 +2,23 @@ use crate::Opstamp;
 use crate::SegmentId;
 use census::{Inventory, TrackedObject};
 use std::collections::HashSet;
+use std::ops::Deref;
 
 #[derive(Default)]
-pub struct MergeOperationInventory(Inventory<InnerMergeOperation>);
+pub(crate) struct MergeOperationInventory(Inventory<InnerMergeOperation>);
+
+impl Deref for MergeOperationInventory {
+    type Target = Inventory<InnerMergeOperation>;
+
+    fn deref(&self) -> &Self::Target {
+        &self.0
+    }
+}
 
 impl MergeOperationInventory {
     pub fn segment_in_merge(&self) -> HashSet<SegmentId> {
         let mut segment_in_merge = HashSet::default();
-        for merge_op in self.0.list() {
+        for merge_op in self.list() {
             for &segment_id in &merge_op.segment_ids {
                 segment_in_merge.insert(segment_id);
             }
@@ -35,13 +44,13 @@ pub struct MergeOperation {
     inner: TrackedObject<InnerMergeOperation>,
 }
 
-struct InnerMergeOperation {
+pub(crate) struct InnerMergeOperation {
     target_opstamp: Opstamp,
     segment_ids: Vec<SegmentId>,
 }
 
 impl MergeOperation {
-    pub fn new(
+    pub(crate) fn new(
         inventory: &MergeOperationInventory,
         target_opstamp: Opstamp,
         segment_ids: Vec<SegmentId>,
@@ -51,7 +60,7 @@ impl MergeOperation {
             segment_ids,
         };
         MergeOperation {
-            inner: inventory.0.track(inner_merge_operation),
+            inner: inventory.track(inner_merge_operation),
         }
     }
 

src/indexer/merger.rs

@@ -2,30 +2,28 @@ use crate::common::MAX_DOC_LIMIT;
 use crate::core::Segment;
 use crate::core::SegmentReader;
 use crate::core::SerializableSegment;
-use crate::docset::DocSet;
+use crate::docset::{DocSet, TERMINATED};
 use crate::fastfield::BytesFastFieldReader;
 use crate::fastfield::DeleteBitSet;
 use crate::fastfield::FastFieldReader;
 use crate::fastfield::FastFieldSerializer;
 use crate::fastfield::MultiValueIntFastFieldReader;
-use crate::fieldnorm::FieldNormReader;
 use crate::fieldnorm::FieldNormsSerializer;
 use crate::fieldnorm::FieldNormsWriter;
+use crate::fieldnorm::{FieldNormReader, FieldNormReaders};
 use crate::indexer::SegmentSerializer;
-use crate::postings::InvertedIndexSerializer;
 use crate::postings::Postings;
+use crate::postings::{InvertedIndexSerializer, SegmentPostings};
 use crate::schema::Cardinality;
 use crate::schema::FieldType;
 use crate::schema::{Field, Schema};
 use crate::store::StoreWriter;
 use crate::termdict::TermMerger;
 use crate::termdict::TermOrdinal;
-use crate::DocId;
+use crate::{DocId, InvertedIndexReader, SegmentComponent};
-use crate::Result;
-use crate::TantivyError;
-use itertools::Itertools;
 use std::cmp;
 use std::collections::HashMap;
+use std::sync::Arc;
 
 fn compute_total_num_tokens(readers: &[SegmentReader], field: Field) -> u64 {
     let mut total_tokens = 0u64;
@@ -72,11 +70,11 @@ fn compute_min_max_val(
             Some(delete_bitset) => {
                 // some deleted documents,
                 // we need to recompute the max / min
-                (0..max_doc)
+                crate::common::minmax(
-                    .filter(|doc_id| delete_bitset.is_alive(*doc_id))
+                    (0..max_doc)
-                    .map(|doc_id| u64_reader.get(doc_id))
+                        .filter(|doc_id| delete_bitset.is_alive(*doc_id))
-                    .minmax()
+                        .map(|doc_id| u64_reader.get(doc_id)),
-                    .into_option()
+                )
             }
             None => {
                 // no deleted documents,
@@ -143,7 +141,7 @@ impl DeltaComputer {
 }
 
 impl IndexMerger {
-    pub fn open(schema: Schema, segments: &[Segment]) -> Result<IndexMerger> {
+    pub fn open(schema: Schema, segments: &[Segment]) -> crate::Result<IndexMerger> {
         let mut readers = vec![];
         let mut max_doc: u32 = 0u32;
         for segment in segments {
@@ -159,7 +157,7 @@ impl IndexMerger {
                  which exceeds the limit {}.",
                 max_doc, MAX_DOC_LIMIT
             );
-            return Err(TantivyError::InvalidArgument(err_msg));
+            return Err(crate::TantivyError::InvalidArgument(err_msg));
         }
         Ok(IndexMerger {
             schema,
@@ -168,7 +166,10 @@ impl IndexMerger {
         })
     }
 
-    fn write_fieldnorms(&self, fieldnorms_serializer: &mut FieldNormsSerializer) -> Result<()> {
+    fn write_fieldnorms(
+        &self,
+        mut fieldnorms_serializer: FieldNormsSerializer,
+    ) -> crate::Result<()> {
         let fields = FieldNormsWriter::fields_with_fieldnorm(&self.schema);
         let mut fieldnorms_data = Vec::with_capacity(self.max_doc as usize);
         for field in fields {
@@ -182,6 +183,7 @@ impl IndexMerger {
             }
             fieldnorms_serializer.serialize_field(field, &fieldnorms_data[..])?;
         }
+        fieldnorms_serializer.close()?;
         Ok(())
     }
 
@@ -189,7 +191,7 @@ impl IndexMerger {
         &self,
         fast_field_serializer: &mut FastFieldSerializer,
         mut term_ord_mappings: HashMap<Field, TermOrdinalMapping>,
-    ) -> Result<()> {
+    ) -> crate::Result<()> {
         for (field, field_entry) in self.schema.fields() {
             let field_type = field_entry.field_type();
             match *field_type {
@@ -234,7 +236,7 @@ impl IndexMerger {
         &self,
         field: Field,
         fast_field_serializer: &mut FastFieldSerializer,
-    ) -> Result<()> {
+    ) -> crate::Result<()> {
         let mut u64_readers = vec![];
         let mut min_value = u64::max_value();
         let mut max_value = u64::min_value();
@@ -284,7 +286,7 @@ impl IndexMerger {
         &self,
         field: Field,
         fast_field_serializer: &mut FastFieldSerializer,
-    ) -> Result<()> {
+    ) -> crate::Result<()> {
         let mut total_num_vals = 0u64;
         let mut u64s_readers: Vec<MultiValueIntFastFieldReader<u64>> = Vec::new();
 
@@ -331,7 +333,7 @@ impl IndexMerger {
         field: Field,
         term_ordinal_mappings: &TermOrdinalMapping,
         fast_field_serializer: &mut FastFieldSerializer,
-    ) -> Result<()> {
+    ) -> crate::Result<()> {
         // Multifastfield consists in 2 fastfields.
         // The first serves as an index into the second one and is stricly increasing.
         // The second contains the actual values.
@@ -371,7 +373,7 @@ impl IndexMerger {
         &self,
         field: Field,
         fast_field_serializer: &mut FastFieldSerializer,
-    ) -> Result<()> {
+    ) -> crate::Result<()> {
         // Multifastfield consists in 2 fastfields.
         // The first serves as an index into the second one and is stricly increasing.
         // The second contains the actual values.
@@ -436,7 +438,7 @@ impl IndexMerger {
         &self,
         field: Field,
         fast_field_serializer: &mut FastFieldSerializer,
-    ) -> Result<()> {
+    ) -> crate::Result<()> {
         let mut total_num_vals = 0u64;
         let mut bytes_readers: Vec<BytesFastFieldReader> = Vec::new();
 
@@ -492,10 +494,11 @@ impl IndexMerger {
         indexed_field: Field,
         field_type: &FieldType,
         serializer: &mut InvertedIndexSerializer,
-    ) -> Result<Option<TermOrdinalMapping>> {
+        fieldnorm_reader: Option<FieldNormReader>,
+    ) -> crate::Result<Option<TermOrdinalMapping>> {
         let mut positions_buffer: Vec<u32> = Vec::with_capacity(1_000);
         let mut delta_computer = DeltaComputer::new();
-        let field_readers = self
+        let field_readers: Vec<Arc<InvertedIndexReader>> = self
             .readers
             .iter()
             .map(|reader| reader.inverted_index(indexed_field))
@@ -550,7 +553,8 @@ impl IndexMerger {
         // - Segment 2's doc ids become  [seg0.max_doc + seg1.max_doc,
         //                                seg0.max_doc + seg1.max_doc + seg2.max_doc]
         // ...
-        let mut field_serializer = serializer.new_field(indexed_field, total_num_tokens)?;
+        let mut field_serializer =
+            serializer.new_field(indexed_field, total_num_tokens, fieldnorm_reader)?;
 
         let field_entry = self.schema.get_field_entry(indexed_field);
 
@@ -560,84 +564,78 @@ impl IndexMerger {
                          indexed. Have you modified the schema?",
         );
 
+        let mut segment_postings_containing_the_term: Vec<(usize, SegmentPostings)> = vec![];
+
         while merged_terms.advance() {
+            segment_postings_containing_the_term.clear();
             let term_bytes: &[u8] = merged_terms.key();
 
+            let mut total_doc_freq = 0;
+
             // Let's compute the list of non-empty posting lists
-            let segment_postings: Vec<_> = merged_terms
+            for heap_item in merged_terms.current_kvs() {
-                .current_kvs()
+                let segment_ord = heap_item.segment_ord;
-                .iter()
+                let term_info = heap_item.streamer.value();
-                .flat_map(|heap_item| {
+                let segment_reader = &self.readers[heap_item.segment_ord];
-                    let segment_ord = heap_item.segment_ord;
+                let inverted_index: &InvertedIndexReader = &*field_readers[segment_ord];
-                    let term_info = heap_item.streamer.value();
+                let segment_postings =
-                    let segment_reader = &self.readers[heap_item.segment_ord];
+                    inverted_index.read_postings_from_terminfo(term_info, segment_postings_option);
-                    let inverted_index = segment_reader.inverted_index(indexed_field);
+                let delete_bitset_opt = segment_reader.delete_bitset();
-                    let mut segment_postings = inverted_index
+                let doc_freq = if let Some(delete_bitset) = delete_bitset_opt {
-                        .read_postings_from_terminfo(term_info, segment_postings_option);
+                    segment_postings.doc_freq_given_deletes(delete_bitset)
-                    while segment_postings.advance() {
+                } else {
-                        if !segment_reader.is_deleted(segment_postings.doc()) {
+                    segment_postings.doc_freq()
-                            return Some((segment_ord, segment_postings));
+                };
-                        }
+                if doc_freq > 0u32 {
-                    }
+                    total_doc_freq += doc_freq;
-                    None
+                    segment_postings_containing_the_term.push((segment_ord, segment_postings));
-                })
+                }
-                .collect();
+            }
 
             // At this point, `segment_postings` contains the posting list
-            // of all of the segments containing the given term.
+            // of all of the segments containing the given term (and that are non-empty)
             //
             // These segments are non-empty and advance has already been called.
-            if !segment_postings.is_empty() {
+            if total_doc_freq == 0u32 {
-                // If not, the `term` will be entirely removed.
+                // All docs that used to contain the term have been deleted. The `term` will be
-
+                // entirely removed.
-                // We know that there is at least one document containing
+                continue;
-                // the term, so we add it.
-                let to_term_ord = field_serializer.new_term(term_bytes)?;
-
-                if let Some(ref mut term_ord_mapping) = term_ord_mapping_opt {
-                    for (segment_ord, from_term_ord) in merged_terms.matching_segments() {
-                        term_ord_mapping.register_from_to(segment_ord, from_term_ord, to_term_ord);
-                    }
-                }
-
-                // We can now serialize this postings, by pushing each document to the
-                // postings serializer.
-                for (segment_ord, mut segment_postings) in segment_postings {
-                    let old_to_new_doc_id = &merged_doc_id_map[segment_ord];
-                    loop {
-                        let doc = segment_postings.doc();
-
-                        // `.advance()` has been called once before the loop.
-                        //
-                        // It was required to make sure we only consider segments
-                        // that effectively contain at least one non-deleted document
-                        // and remove terms that do not have documents associated.
-                        //
-                        //  For this reason, we cannot use a `while segment_postings.advance()` loop.
-
-                        // deleted doc are skipped as they do not have a `remapped_doc_id`.
-                        if let Some(remapped_doc_id) = old_to_new_doc_id[doc as usize] {
-                            // we make sure to only write the term iff
-                            // there is at least one document.
-                            let term_freq = segment_postings.term_freq();
-                            segment_postings.positions(&mut positions_buffer);
-
-                            let delta_positions = delta_computer.compute_delta(&positions_buffer);
-                            field_serializer.write_doc(
-                                remapped_doc_id,
-                                term_freq,
-                                delta_positions,
-                            )?;
-                        }
-                        if !segment_postings.advance() {
-                            break;
-                        }
-                    }
-                }
-
-                // closing the term.
-                field_serializer.close_term()?;
             }
+
+            let to_term_ord = field_serializer.new_term(term_bytes, total_doc_freq)?;
+
+            if let Some(ref mut term_ord_mapping) = term_ord_mapping_opt {
+                for (segment_ord, from_term_ord) in merged_terms.matching_segments() {
+                    term_ord_mapping.register_from_to(segment_ord, from_term_ord, to_term_ord);
+                }
+            }
+
+            // We can now serialize this postings, by pushing each document to the
+            // postings serializer.
+            for (segment_ord, mut segment_postings) in
+                segment_postings_containing_the_term.drain(..)
+            {
+                let old_to_new_doc_id = &merged_doc_id_map[segment_ord];
+
+                let mut doc = segment_postings.doc();
+                while doc != TERMINATED {
+                    // deleted doc are skipped as they do not have a `remapped_doc_id`.
+                    if let Some(remapped_doc_id) = old_to_new_doc_id[doc as usize] {
+                        // we make sure to only write the term iff
+                        // there is at least one document.
+                        let term_freq = segment_postings.term_freq();
+                        segment_postings.positions(&mut positions_buffer);
+
+                        let delta_positions = delta_computer.compute_delta(&positions_buffer);
+                        field_serializer.write_doc(remapped_doc_id, term_freq, delta_positions)?;
+                    }
+
+                    doc = segment_postings.advance();
+                }
+            }
+
+            // closing the term.
+            field_serializer.close_term()?;
         }
         field_serializer.close()?;
         Ok(term_ord_mapping_opt)
@@ -646,13 +644,18 @@ impl IndexMerger {
     fn write_postings(
         &self,
         serializer: &mut InvertedIndexSerializer,
-    ) -> Result<HashMap<Field, TermOrdinalMapping>> {
+        fieldnorm_readers: FieldNormReaders,
+    ) -> crate::Result<HashMap<Field, TermOrdinalMapping>> {
         let mut term_ordinal_mappings = HashMap::new();
         for (field, field_entry) in self.schema.fields() {
+            let fieldnorm_reader = fieldnorm_readers.get_field(field);
             if field_entry.is_indexed() {
-                if let Some(term_ordinal_mapping) =
+                if let Some(term_ordinal_mapping) = self.write_postings_for_field(
-                    self.write_postings_for_field(field, field_entry.field_type(), serializer)?
+                    field,
-                {
+                    field_entry.field_type(),
+                    serializer,
+                    fieldnorm_reader,
+                )? {
                     term_ordinal_mappings.insert(field, term_ordinal_mapping);
                 }
             }
@@ -660,7 +663,7 @@ impl IndexMerger {
         Ok(term_ordinal_mappings)
     }
 
-    fn write_storable_fields(&self, store_writer: &mut StoreWriter) -> Result<()> {
+    fn write_storable_fields(&self, store_writer: &mut StoreWriter) -> crate::Result<()> {
         for reader in &self.readers {
             let store_reader = reader.get_store_reader();
             if reader.num_deleted_docs() > 0 {
@@ -677,9 +680,16 @@ impl IndexMerger {
 }
 
 impl SerializableSegment for IndexMerger {
-    fn write(&self, mut serializer: SegmentSerializer) -> Result<u32> {
+    fn write(&self, mut serializer: SegmentSerializer) -> crate::Result<u32> {
-        let term_ord_mappings = self.write_postings(serializer.get_postings_serializer())?;
+        if let Some(fieldnorms_serializer) = serializer.extract_fieldnorms_serializer() {
-        self.write_fieldnorms(serializer.get_fieldnorms_serializer())?;
+            self.write_fieldnorms(fieldnorms_serializer)?;
+        }
+        let fieldnorm_data = serializer
+            .segment()
+            .open_read(SegmentComponent::FIELDNORMS)?;
+        let fieldnorm_readers = FieldNormReaders::open(fieldnorm_data)?;
+        let term_ord_mappings =
+            self.write_postings(serializer.get_postings_serializer(), fieldnorm_readers)?;
         self.write_fast_fields(serializer.get_fast_field_serializer(), term_ord_mappings)?;
         self.write_storable_fields(serializer.get_store_writer())?;
         serializer.close()?;
@@ -689,15 +699,15 @@ impl SerializableSegment for IndexMerger {
 
 #[cfg(test)]
 mod tests {
+    use crate::assert_nearly_equals;
     use crate::collector::tests::TEST_COLLECTOR_WITH_SCORE;
     use crate::collector::tests::{BytesFastFieldTestCollector, FastFieldTestCollector};
     use crate::collector::{Count, FacetCollector};
     use crate::core::Index;
     use crate::query::AllQuery;
     use crate::query::BooleanQuery;
+    use crate::query::Scorer;
     use crate::query::TermQuery;
-    use crate::schema;
-    use crate::schema::Cardinality;
     use crate::schema::Document;
     use crate::schema::Facet;
     use crate::schema::IndexRecordOption;
@@ -705,11 +715,13 @@ mod tests {
     use crate::schema::Term;
     use crate::schema::TextFieldIndexing;
     use crate::schema::INDEXED;
+    use crate::schema::{Cardinality, TEXT};
     use crate::DocAddress;
     use crate::IndexWriter;
     use crate::Searcher;
+    use crate::{schema, DocSet, SegmentId};
     use byteorder::{BigEndian, ReadBytesExt, WriteBytesExt};
-    use futures::Future;
+    use futures::executor::block_on;
     use std::io::Cursor;
 
     #[test]
@@ -792,11 +804,7 @@ mod tests {
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
             let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-            index_writer
+            block_on(index_writer.merge(&segment_ids)).expect("Merging failed");
-                .merge(&segment_ids)
-                .expect("Failed to initiate merge")
-                .wait()
-                .expect("Merging failed");
             index_writer.wait_merging_threads().unwrap();
         }
         {
@@ -1040,11 +1048,7 @@ mod tests {
             let segment_ids = index
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
-            index_writer
+            block_on(index_writer.merge(&segment_ids)).expect("Merging failed");
-                .merge(&segment_ids)
-                .expect("Failed to initiate merge")
-                .wait()
-                .expect("Merging failed");
             reader.reload().unwrap();
             let searcher = reader.searcher();
             assert_eq!(searcher.segment_readers().len(), 1);
@@ -1139,11 +1143,7 @@ mod tests {
             let segment_ids = index
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
-            index_writer
+            block_on(index_writer.merge(&segment_ids)).expect("Merging failed");
-                .merge(&segment_ids)
-                .expect("Failed to initiate merge")
-                .wait()
-                .expect("Merging failed");
             reader.reload().unwrap();
 
             let searcher = reader.searcher();
@@ -1277,11 +1277,7 @@ mod tests {
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
             let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-            index_writer
+            block_on(index_writer.merge(&segment_ids)).expect("Merging failed");
-                .merge(&segment_ids)
-                .expect("Failed to initiate merge")
-                .wait()
-                .expect("Merging failed");
             index_writer.wait_merging_threads().unwrap();
             reader.reload().unwrap();
             test_searcher(
@@ -1336,11 +1332,7 @@ mod tests {
         let segment_ids = index
             .searchable_segment_ids()
             .expect("Searchable segments failed.");
-        index_writer
+        block_on(index_writer.merge(&segment_ids)).expect("Merging failed");
-            .merge(&segment_ids)
-            .expect("Failed to initiate merge")
-            .wait()
-            .expect("Merging failed");
         reader.reload().unwrap();
         // commit has not been called yet. The document should still be
         // there.
@@ -1361,22 +1353,18 @@ mod tests {
             let mut doc = Document::default();
             doc.add_u64(int_field, 1);
             index_writer.add_document(doc.clone());
-            index_writer.commit().expect("commit failed");
+            assert!(index_writer.commit().is_ok());
             index_writer.add_document(doc);
-            index_writer.commit().expect("commit failed");
+            assert!(index_writer.commit().is_ok());
             index_writer.delete_term(Term::from_field_u64(int_field, 1));
 
             let segment_ids = index
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
-            index_writer
+            assert!(block_on(index_writer.merge(&segment_ids)).is_ok());
-                .merge(&segment_ids)
-                .expect("Failed to initiate merge")
-                .wait()
-                .expect("Merging failed");
 
             // assert delete has not been committed
-            reader.reload().expect("failed to load searcher 1");
+            assert!(reader.reload().is_ok());
             let searcher = reader.searcher();
             assert_eq!(searcher.num_docs(), 2);
 
@@ -1415,12 +1403,12 @@ mod tests {
             index_doc(&mut index_writer, &[1, 5]);
             index_doc(&mut index_writer, &[3]);
             index_doc(&mut index_writer, &[17]);
-            index_writer.commit().expect("committed");
+            assert!(index_writer.commit().is_ok());
             index_doc(&mut index_writer, &[20]);
-            index_writer.commit().expect("committed");
+            assert!(index_writer.commit().is_ok());
             index_doc(&mut index_writer, &[28, 27]);
             index_doc(&mut index_writer, &[1_000]);
-            index_writer.commit().expect("committed");
+            assert!(index_writer.commit().is_ok());
         }
         let reader = index.reader().unwrap();
         let searcher = reader.searcher();
@@ -1452,15 +1440,6 @@ mod tests {
             assert_eq!(&vals, &[17]);
         }
 
-        println!(
-            "{:?}",
-            searcher
-                .segment_readers()
-                .iter()
-                .map(|reader| reader.max_doc())
-                .collect::<Vec<_>>()
-        );
-
         {
             let segment = searcher.segment_reader(1u32);
             let ff_reader = segment.fast_fields().u64s(int_field).unwrap();
@@ -1484,27 +1463,13 @@ mod tests {
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
             let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-            index_writer
+            assert!(block_on(index_writer.merge(&segment_ids)).is_ok());
-                .merge(&segment_ids)
+            assert!(index_writer.wait_merging_threads().is_ok());
-                .expect("Failed to initiate merge")
-                .wait()
-                .expect("Merging failed");
-            index_writer
-                .wait_merging_threads()
-                .expect("Wait for merging threads");
         }
-        reader.reload().expect("Load searcher");
+        assert!(reader.reload().is_ok());
 
         {
             let searcher = reader.searcher();
-            println!(
-                "{:?}",
-                searcher
-                    .segment_readers()
-                    .iter()
-                    .map(|reader| reader.max_doc())
-                    .collect::<Vec<_>>()
-            );
             let segment = searcher.segment_reader(0u32);
             let ff_reader = segment.fast_fields().u64s(int_field).unwrap();
 
@@ -1539,4 +1504,109 @@ mod tests {
             assert_eq!(&vals, &[20]);
         }
     }
+
+    #[test]
+    fn merges_f64_fast_fields_correctly() -> crate::Result<()> {
+        let mut builder = schema::SchemaBuilder::new();
+
+        let fast_multi = IntOptions::default().set_fast(Cardinality::MultiValues);
+
+        let field = builder.add_f64_field("f64", schema::FAST);
+        let multi_field = builder.add_f64_field("f64s", fast_multi);
+
+        let index = Index::create_in_ram(builder.build());
+
+        let mut writer = index.writer_with_num_threads(1, 3_000_000)?;
+
+        // Make sure we'll attempt to merge every created segment
+        let mut policy = crate::indexer::LogMergePolicy::default();
+        policy.set_min_merge_size(2);
+        writer.set_merge_policy(Box::new(policy));
+
+        for i in 0..100 {
+            let mut doc = Document::new();
+            doc.add_f64(field, 42.0);
+            doc.add_f64(multi_field, 0.24);
+            doc.add_f64(multi_field, 0.27);
+            writer.add_document(doc);
+            if i % 5 == 0 {
+                writer.commit()?;
+            }
+        }
+
+        writer.commit()?;
+        writer.wait_merging_threads()?;
+
+        // If a merging thread fails, we should end up with more
+        // than one segment here
+        assert_eq!(1, index.searchable_segments()?.len());
+        Ok(())
+    }
+
+    #[test]
+    fn test_merged_index_has_blockwand() -> crate::Result<()> {
+        let mut builder = schema::SchemaBuilder::new();
+        let text = builder.add_text_field("text", TEXT);
+        let index = Index::create_in_ram(builder.build());
+        let mut writer = index.writer_with_num_threads(1, 3_000_000)?;
+        let happy_term = Term::from_field_text(text, "happy");
+        let term_query = TermQuery::new(happy_term, IndexRecordOption::WithFreqs);
+        for _ in 0..62 {
+            writer.add_document(doc!(text=>"hello happy tax payer"));
+        }
+        writer.commit()?;
+        let reader = index.reader()?;
+        let searcher = reader.searcher();
+        let mut term_scorer = term_query
+            .specialized_weight(&searcher, true)
+            .specialized_scorer(searcher.segment_reader(0u32), 1.0)?;
+        assert_eq!(term_scorer.doc(), 0);
+        assert_nearly_equals!(term_scorer.block_max_score(), 0.0079681855);
+        assert_nearly_equals!(term_scorer.score(), 0.0079681855);
+        for _ in 0..81 {
+            writer.add_document(doc!(text=>"hello happy tax payer"));
+        }
+        writer.commit()?;
+        reader.reload()?;
+        let searcher = reader.searcher();
+
+        assert_eq!(searcher.segment_readers().len(), 2);
+        for segment_reader in searcher.segment_readers() {
+            let mut term_scorer = term_query
+                .specialized_weight(&searcher, true)
+                .specialized_scorer(segment_reader, 1.0)?;
+            // the difference compared to before is instrinsic to the bm25 formula. no worries there.
+            for doc in segment_reader.doc_ids_alive() {
+                assert_eq!(term_scorer.doc(), doc);
+                assert_nearly_equals!(term_scorer.block_max_score(), 0.003478312);
+                assert_nearly_equals!(term_scorer.score(), 0.003478312);
+                term_scorer.advance();
+            }
+        }
+
+        let segment_ids: Vec<SegmentId> = searcher
+            .segment_readers()
+            .iter()
+            .map(|reader| reader.segment_id())
+            .collect();
+        block_on(writer.merge(&segment_ids[..]))?;
+
+        reader.reload()?;
+        let searcher = reader.searcher();
+        assert_eq!(searcher.segment_readers().len(), 1);
+
+        let segment_reader = searcher.segment_reader(0u32);
+        let mut term_scorer = term_query
+            .specialized_weight(&searcher, true)
+            .specialized_scorer(segment_reader, 1.0)?;
+        // the difference compared to before is instrinsic to the bm25 formula. no worries there.
+        for doc in segment_reader.doc_ids_alive() {
+            assert_eq!(term_scorer.doc(), doc);
+            assert_nearly_equals!(term_scorer.block_max_score(), 0.003478312);
+            assert_nearly_equals!(term_scorer.score(), 0.003478312);
+            term_scorer.advance();
+        }
+
+        Ok(())
+    }
 }

src/indexer/mod.rs

@@ -18,7 +18,7 @@ mod stamper;
 
 pub use self::index_writer::IndexWriter;
 pub use self::log_merge_policy::LogMergePolicy;
-pub use self::merge_operation::{MergeOperation, MergeOperationInventory};
+pub use self::merge_operation::MergeOperation;
 pub use self::merge_policy::{MergeCandidate, MergePolicy, NoMergePolicy};
 pub use self::prepared_commit::PreparedCommit;
 pub use self::segment_entry::SegmentEntry;
@@ -33,6 +33,7 @@ pub type DefaultMergePolicy = LogMergePolicy;
 mod tests {
     use crate::schema::{self, Schema};
     use crate::{Index, Term};
+
     #[test]
     fn test_advance_delete_bug() {
         let mut schema_builder = Schema::builder();

src/indexer/operation.rs

@@ -19,6 +19,8 @@ pub struct AddOperation {
 /// UserOperation is an enum type that encapsulates other operation types.
 #[derive(Eq, PartialEq, Debug)]
 pub enum UserOperation {
+    /// Add operation
     Add(Document),
+    /// Delete operation
     Delete(Term),
 }

src/indexer/prepared_commit.rs

@@ -1,6 +1,6 @@
 use super::IndexWriter;
 use crate::Opstamp;
-use crate::Result;
+use futures::executor::block_on;
 
 /// A prepared commit
 pub struct PreparedCommit<'a> {
@@ -26,15 +26,17 @@ impl<'a> PreparedCommit<'a> {
         self.payload = Some(payload.to_string())
     }
 
-    pub fn abort(self) -> Result<Opstamp> {
+    pub fn abort(self) -> crate::Result<Opstamp> {
         self.index_writer.rollback()
     }
 
-    pub fn commit(self) -> Result<Opstamp> {
+    pub fn commit(self) -> crate::Result<Opstamp> {
         info!("committing {}", self.opstamp);
-        self.index_writer
+        let _ = block_on(
-            .segment_updater()
+            self.index_writer
-            .commit(self.opstamp, self.payload)?;
+                .segment_updater()
+                .schedule_commit(self.opstamp, self.payload),
+        );
         Ok(self.opstamp)
     }
 }

src/indexer/segment_entry.rs

@@ -1,7 +1,7 @@
+use crate::common::BitSet;
 use crate::core::SegmentId;
 use crate::core::SegmentMeta;
 use crate::indexer::delete_queue::DeleteCursor;
-use bit_set::BitSet;
 use std::fmt;
 
 /// A segment entry describes the state of

src/indexer/segment_manager.rs

@@ -4,7 +4,6 @@ use crate::core::SegmentMeta;
 use crate::error::TantivyError;
 use crate::indexer::delete_queue::DeleteCursor;
 use crate::indexer::SegmentEntry;
-use crate::Result as TantivyResult;
 use std::collections::hash_set::HashSet;
 use std::fmt::{self, Debug, Formatter};
 use std::sync::RwLock;
@@ -16,6 +15,28 @@ struct SegmentRegisters {
     committed: SegmentRegister,
 }
 
+#[derive(PartialEq, Eq)]
+pub(crate) enum SegmentsStatus {
+    Committed,
+    Uncommitted,
+}
+
+impl SegmentRegisters {
+    /// Check if all the segments are committed or uncommited.
+    ///
+    /// If some segment is missing or segments are in a different state (this should not happen
+    /// if tantivy is used correctly), returns `None`.
+    fn segments_status(&self, segment_ids: &[SegmentId]) -> Option<SegmentsStatus> {
+        if self.uncommitted.contains_all(segment_ids) {
+            Some(SegmentsStatus::Uncommitted)
+        } else if self.committed.contains_all(segment_ids) {
+            Some(SegmentsStatus::Committed)
+        } else {
+            None
+        }
+    }
+}
+
 /// The segment manager stores the list of segments
 /// as well as their state.
 ///
@@ -27,7 +48,7 @@ pub struct SegmentManager {
 }
 
 impl Debug for SegmentManager {
-    fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), fmt::Error> {
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
         let lock = self.read();
         write!(
             f,
@@ -123,7 +144,7 @@ impl SegmentManager {
     /// Returns an error if some segments are missing, or if
     /// the `segment_ids` are not either all committed or all
     /// uncommitted.
-    pub fn start_merge(&self, segment_ids: &[SegmentId]) -> TantivyResult<Vec<SegmentEntry>> {
+    pub fn start_merge(&self, segment_ids: &[SegmentId]) -> crate::Result<Vec<SegmentEntry>> {
         let registers_lock = self.read();
         let mut segment_entries = vec![];
         if registers_lock.uncommitted.contains_all(segment_ids) {
@@ -153,33 +174,35 @@ impl SegmentManager {
         let mut registers_lock = self.write();
         registers_lock.uncommitted.add_segment_entry(segment_entry);
     }
-
+    // Replace a list of segments for their equivalent merged segment.
-    pub fn end_merge(
+    //
+    // Returns true if these segments are committed, false if the merge segments are uncommited.
+    pub(crate) fn end_merge(
         &self,
         before_merge_segment_ids: &[SegmentId],
         after_merge_segment_entry: SegmentEntry,
-    ) {
+    ) -> crate::Result<SegmentsStatus> {
         let mut registers_lock = self.write();
-        let target_register: &mut SegmentRegister = {
+        let segments_status = registers_lock
-            if registers_lock
+            .segments_status(before_merge_segment_ids)
-                .uncommitted
+            .ok_or_else(|| {
-                .contains_all(before_merge_segment_ids)
-            {
-                &mut registers_lock.uncommitted
-            } else if registers_lock
-                .committed
-                .contains_all(before_merge_segment_ids)
-            {
-                &mut registers_lock.committed
-            } else {
                 warn!("couldn't find segment in SegmentManager");
-                return;
+                crate::TantivyError::InvalidArgument(
-            }
+                    "The segments that were merged could not be found in the SegmentManager. \
+                     This is not necessarily a bug, and can happen after a rollback for instance."
+                        .to_string(),
+                )
+            })?;
+
+        let target_register: &mut SegmentRegister = match segments_status {
+            SegmentsStatus::Uncommitted => &mut registers_lock.uncommitted,
+            SegmentsStatus::Committed => &mut registers_lock.committed,
         };
         for segment_id in before_merge_segment_ids {
             target_register.remove_segment(segment_id);
         }
         target_register.add_segment_entry(after_merge_segment_entry);
+        Ok(segments_status)
     }
 
     pub fn committed_segment_metas(&self) -> Vec<SegmentMeta> {

src/indexer/segment_serializer.rs

@@ -1,5 +1,3 @@
-use crate::Result;
-
 use crate::core::Segment;
 use crate::core::SegmentComponent;
 use crate::fastfield::FastFieldSerializer;
@@ -10,15 +8,16 @@ use crate::store::StoreWriter;
 /// Segment serializer is in charge of laying out on disk
 /// the data accumulated and sorted by the `SegmentWriter`.
 pub struct SegmentSerializer {
+    segment: Segment,
     store_writer: StoreWriter,
     fast_field_serializer: FastFieldSerializer,
-    fieldnorms_serializer: FieldNormsSerializer,
+    fieldnorms_serializer: Option<FieldNormsSerializer>,
     postings_serializer: InvertedIndexSerializer,
 }
 
 impl SegmentSerializer {
     /// Creates a new `SegmentSerializer`.
-    pub fn for_segment(segment: &mut Segment) -> Result<SegmentSerializer> {
+    pub fn for_segment(mut segment: Segment) -> crate::Result<SegmentSerializer> {
         let store_write = segment.open_write(SegmentComponent::STORE)?;
 
         let fast_field_write = segment.open_write(SegmentComponent::FASTFIELDS)?;
@@ -27,15 +26,20 @@ impl SegmentSerializer {
         let fieldnorms_write = segment.open_write(SegmentComponent::FIELDNORMS)?;
         let fieldnorms_serializer = FieldNormsSerializer::from_write(fieldnorms_write)?;
 
-        let postings_serializer = InvertedIndexSerializer::open(segment)?;
+        let postings_serializer = InvertedIndexSerializer::open(&mut segment)?;
         Ok(SegmentSerializer {
+            segment,
             store_writer: StoreWriter::new(store_write),
             fast_field_serializer,
-            fieldnorms_serializer,
+            fieldnorms_serializer: Some(fieldnorms_serializer),
             postings_serializer,
         })
     }
 
+    pub fn segment(&self) -> &Segment {
+        &self.segment
+    }
+
     /// Accessor to the `PostingsSerializer`.
     pub fn get_postings_serializer(&mut self) -> &mut InvertedIndexSerializer {
         &mut self.postings_serializer
@@ -46,9 +50,11 @@ impl SegmentSerializer {
         &mut self.fast_field_serializer
     }
 
-    /// Accessor to the field norm serializer.
+    /// Extract the field norm serializer.
-    pub fn get_fieldnorms_serializer(&mut self) -> &mut FieldNormsSerializer {
+    ///
-        &mut self.fieldnorms_serializer
+    /// Note the fieldnorms serializer can only be extracted once.
+    pub fn extract_fieldnorms_serializer(&mut self) -> Option<FieldNormsSerializer> {
+        self.fieldnorms_serializer.take()
     }
 
     /// Accessor to the `StoreWriter`.
@@ -57,11 +63,13 @@ impl SegmentSerializer {
     }
 
     /// Finalize the segment serialization.
-    pub fn close(self) -> Result<()> {
+    pub fn close(mut self) -> crate::Result<()> {
+        if let Some(fieldnorms_serializer) = self.extract_fieldnorms_serializer() {
+            fieldnorms_serializer.close()?;
+        }
         self.fast_field_serializer.close()?;
         self.postings_serializer.close()?;
         self.store_writer.close()?;
-        self.fieldnorms_serializer.close()?;
         Ok(())
     }
 }

src/indexer/segment_updater.rs

@@ -6,39 +6,33 @@ use crate::core::SegmentId;
 use crate::core::SegmentMeta;
 use crate::core::SerializableSegment;
 use crate::core::META_FILEPATH;
-use crate::directory::{Directory, DirectoryClone};
+use crate::directory::{Directory, DirectoryClone, GarbageCollectionResult};
-use crate::error::TantivyError;
 use crate::indexer::delete_queue::DeleteCursor;
 use crate::indexer::index_writer::advance_deletes;
 use crate::indexer::merge_operation::MergeOperationInventory;
 use crate::indexer::merger::IndexMerger;
+use crate::indexer::segment_manager::SegmentsStatus;
 use crate::indexer::stamper::Stamper;
-use crate::indexer::MergeOperation;
 use crate::indexer::SegmentEntry;
 use crate::indexer::SegmentSerializer;
 use crate::indexer::{DefaultMergePolicy, MergePolicy};
+use crate::indexer::{MergeCandidate, MergeOperation};
 use crate::schema::Schema;
 use crate::Opstamp;
-use crate::Result;
+use futures::channel::oneshot;
-use futures::oneshot;
+use futures::executor::{ThreadPool, ThreadPoolBuilder};
-use futures::sync::oneshot::Receiver;
+use futures::future::Future;
-use futures::Future;
+use futures::future::TryFutureExt;
-use futures_cpupool::Builder as CpuPoolBuilder;
-use futures_cpupool::CpuFuture;
-use futures_cpupool::CpuPool;
-use serde_json;
 use std::borrow::BorrowMut;
-use std::collections::HashMap;
 use std::collections::HashSet;
 use std::io::Write;
-use std::mem;
+use std::ops::Deref;
-use std::ops::DerefMut;
 use std::path::PathBuf;
-use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
+use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::Arc;
 use std::sync::RwLock;
-use std::thread;
+
-use std::thread::JoinHandle;
+const NUM_MERGE_THREADS: usize = 4;
 
 /// Save the index meta file.
 /// This operation is atomic :
@@ -49,7 +43,7 @@ use std::thread::JoinHandle;
 /// and flushed.
 ///
 /// This method is not part of tantivy's public API
-pub fn save_new_metas(schema: Schema, directory: &mut dyn Directory) -> Result<()> {
+pub fn save_new_metas(schema: Schema, directory: &mut dyn Directory) -> crate::Result<()> {
     save_metas(
         &IndexMeta {
             segments: Vec::new(),
@@ -70,7 +64,7 @@ pub fn save_new_metas(schema: Schema, directory: &mut dyn Directory) -> Result<(
 /// and flushed.
 ///
 /// This method is not part of tantivy's public API
-fn save_metas(metas: &IndexMeta, directory: &mut dyn Directory) -> Result<()> {
+fn save_metas(metas: &IndexMeta, directory: &mut dyn Directory) -> crate::Result<()> {
     info!("save metas");
     let mut buffer = serde_json::to_vec_pretty(metas)?;
     // Just adding a new line at the end of the buffer.
@@ -89,21 +83,38 @@ fn save_metas(metas: &IndexMeta, directory: &mut dyn Directory) -> Result<()> {
 // We voluntarily pass a merge_operation ref to guarantee that
 // the merge_operation is alive during the process
 #[derive(Clone)]
-pub struct SegmentUpdater(Arc<InnerSegmentUpdater>);
+pub(crate) struct SegmentUpdater(Arc<InnerSegmentUpdater>);
 
-fn perform_merge(
+impl Deref for SegmentUpdater {
-    merge_operation: &MergeOperation,
+    type Target = InnerSegmentUpdater;
+
+    #[inline]
+    fn deref(&self) -> &Self::Target {
+        &self.0
+    }
+}
+
+async fn garbage_collect_files(
+    segment_updater: SegmentUpdater,
+) -> crate::Result<GarbageCollectionResult> {
+    info!("Running garbage collection");
+    let mut index = segment_updater.index.clone();
+    index
+        .directory_mut()
+        .garbage_collect(move || segment_updater.list_files())
+}
+
+/// Merges a list of segments the list of segment givens in the `segment_entries`.
+/// This function happens in the calling thread and is computationally expensive.
+fn merge(
     index: &Index,
     mut segment_entries: Vec<SegmentEntry>,
-) -> Result<SegmentEntry> {
+    target_opstamp: Opstamp,
-    let target_opstamp = merge_operation.target_opstamp();
+) -> crate::Result<SegmentEntry> {
-
     // first we need to apply deletes to our segment.
-    let mut merged_segment = index.new_segment();
+    let merged_segment = index.new_segment();
-
-    // TODO add logging
-    let schema = index.schema();
 
+    // First we apply all of the delet to the merged segment, up to the target opstamp.
     for segment_entry in &mut segment_entries {
         let segment = index.segment(segment_entry.meta().clone());
         advance_deletes(segment, segment_entry, target_opstamp)?;
@@ -117,22 +128,20 @@ fn perform_merge(
         .collect();
 
     // An IndexMerger is like a "view" of our merged segments.
-    let merger: IndexMerger = IndexMerger::open(schema, &segments[..])?;
+    let merger: IndexMerger = IndexMerger::open(index.schema(), &segments[..])?;
 
-    // ... we just serialize this index merger in our new segment
+    // ... we just serialize this index merger in our new segment to merge the two segments.
-    // to merge the two segments.
+    let segment_serializer = SegmentSerializer::for_segment(merged_segment.clone())?;
-
-    let segment_serializer = SegmentSerializer::for_segment(&mut merged_segment)?;
 
     let num_docs = merger.write(segment_serializer)?;
 
-    let segment_meta = index.new_segment_meta(merged_segment.id(), num_docs);
+    let merged_segment_id = merged_segment.id();
 
-    let after_merge_segment_entry = SegmentEntry::new(segment_meta.clone(), delete_cursor, None);
+    let segment_meta = index.new_segment_meta(merged_segment_id, num_docs);
-    Ok(after_merge_segment_entry)
+    Ok(SegmentEntry::new(segment_meta, delete_cursor, None))
 }
 
-struct InnerSegmentUpdater {
+pub(crate) struct InnerSegmentUpdater {
     // we keep a copy of the current active IndexMeta to
     // avoid loading the file everytime we need it in the
     // `SegmentUpdater`.
@@ -140,12 +149,12 @@ struct InnerSegmentUpdater {
     // This should be up to date as all update happen through
     // the unique active `SegmentUpdater`.
     active_metas: RwLock<Arc<IndexMeta>>,
-    pool: CpuPool,
+    pool: ThreadPool,
+    merge_thread_pool: ThreadPool,
+
     index: Index,
     segment_manager: SegmentManager,
     merge_policy: RwLock<Arc<Box<dyn MergePolicy>>>,
-    merging_thread_id: AtomicUsize,
-    merging_threads: RwLock<HashMap<usize, JoinHandle<Result<()>>>>,
     killed: AtomicBool,
     stamper: Stamper,
     merge_operations: MergeOperationInventory,
@@ -156,22 +165,35 @@ impl SegmentUpdater {
         index: Index,
         stamper: Stamper,
         delete_cursor: &DeleteCursor,
-    ) -> Result<SegmentUpdater> {
+    ) -> crate::Result<SegmentUpdater> {
         let segments = index.searchable_segment_metas()?;
         let segment_manager = SegmentManager::from_segments(segments, delete_cursor);
-        let pool = CpuPoolBuilder::new()
+        let pool = ThreadPoolBuilder::new()
             .name_prefix("segment_updater")
             .pool_size(1)
-            .create();
+            .create()
+            .map_err(|_| {
+                crate::TantivyError::SystemError(
+                    "Failed to spawn segment updater thread".to_string(),
+                )
+            })?;
+        let merge_thread_pool = ThreadPoolBuilder::new()
+            .name_prefix("merge_thread")
+            .pool_size(NUM_MERGE_THREADS)
+            .create()
+            .map_err(|_| {
+                crate::TantivyError::SystemError(
+                    "Failed to spawn segment merging thread".to_string(),
+                )
+            })?;
         let index_meta = index.load_metas()?;
         Ok(SegmentUpdater(Arc::new(InnerSegmentUpdater {
             active_metas: RwLock::new(Arc::new(index_meta)),
             pool,
+            merge_thread_pool,
             index,
             segment_manager,
             merge_policy: RwLock::new(Arc::new(Box::new(DefaultMergePolicy::default()))),
-            merging_thread_id: AtomicUsize::default(),
-            merging_threads: RwLock::new(HashMap::new()),
             killed: AtomicBool::new(false),
             stamper,
             merge_operations: Default::default(),
@@ -179,65 +201,82 @@ impl SegmentUpdater {
     }
 
     pub fn get_merge_policy(&self) -> Arc<Box<dyn MergePolicy>> {
-        self.0.merge_policy.read().unwrap().clone()
+        self.merge_policy.read().unwrap().clone()
     }
 
     pub fn set_merge_policy(&self, merge_policy: Box<dyn MergePolicy>) {
         let arc_merge_policy = Arc::new(merge_policy);
-        *self.0.merge_policy.write().unwrap() = arc_merge_policy;
+        *self.merge_policy.write().unwrap() = arc_merge_policy;
     }
 
-    fn get_merging_thread_id(&self) -> usize {
+    fn schedule_future<T: 'static + Send, F: Future<Output = crate::Result<T>> + 'static + Send>(
-        self.0.merging_thread_id.fetch_add(1, Ordering::SeqCst)
-    }
-
-    fn run_async<T: 'static + Send, F: 'static + Send + FnOnce(SegmentUpdater) -> T>(
         &self,
         f: F,
-    ) -> CpuFuture<T, TantivyError> {
+    ) -> impl Future<Output = crate::Result<T>> {
-        let me_clone = self.clone();
+        let (sender, receiver) = oneshot::channel();
-        self.0.pool.spawn_fn(move || Ok(f(me_clone)))
+        if self.is_alive() {
+            self.pool.spawn_ok(async move {
+                let _ = sender.send(f.await);
+            });
+        } else {
+            let _ = sender.send(Err(crate::TantivyError::SystemError(
+                "Segment updater killed".to_string(),
+            )));
+        }
+        receiver.unwrap_or_else(|_| {
+            let err_msg =
+                "A segment_updater future did not success. This should never happen.".to_string();
+            Err(crate::TantivyError::SystemError(err_msg))
+        })
     }
 
-    pub fn add_segment(&self, segment_entry: SegmentEntry) {
+    pub fn schedule_add_segment(
-        self.run_async(|segment_updater| {
+        &self,
-            segment_updater.0.segment_manager.add_segment(segment_entry);
+        segment_entry: SegmentEntry,
-            segment_updater.consider_merge_options();
+    ) -> impl Future<Output = crate::Result<()>> {
+        let segment_updater = self.clone();
+        self.schedule_future(async move {
+            segment_updater.segment_manager.add_segment(segment_entry);
+            segment_updater.consider_merge_options().await;
+            Ok(())
         })
-        .forget();
     }
 
     /// Orders `SegmentManager` to remove all segments
     pub(crate) fn remove_all_segments(&self) {
-        self.0.segment_manager.remove_all_segments();
+        self.segment_manager.remove_all_segments();
     }
 
     pub fn kill(&mut self) {
-        self.0.killed.store(true, Ordering::Release);
+        self.killed.store(true, Ordering::Release);
     }
 
     pub fn is_alive(&self) -> bool {
-        !self.0.killed.load(Ordering::Acquire)
+        !self.killed.load(Ordering::Acquire)
     }
 
     /// Apply deletes up to the target opstamp to all segments.
     ///
     /// The method returns copies of the segment entries,
     /// updated with the delete information.
-    fn purge_deletes(&self, target_opstamp: Opstamp) -> Result<Vec<SegmentEntry>> {
+    fn purge_deletes(&self, target_opstamp: Opstamp) -> crate::Result<Vec<SegmentEntry>> {
-        let mut segment_entries = self.0.segment_manager.segment_entries();
+        let mut segment_entries = self.segment_manager.segment_entries();
         for segment_entry in &mut segment_entries {
-            let segment = self.0.index.segment(segment_entry.meta().clone());
+            let segment = self.index.segment(segment_entry.meta().clone());
             advance_deletes(segment, segment_entry, target_opstamp)?;
         }
         Ok(segment_entries)
     }
 
-    pub fn save_metas(&self, opstamp: Opstamp, commit_message: Option<String>) {
+    pub fn save_metas(
+        &self,
+        opstamp: Opstamp,
+        commit_message: Option<String>,
+    ) -> crate::Result<()> {
         if self.is_alive() {
-            let index = &self.0.index;
+            let index = &self.index;
             let directory = index.directory();
-            let mut commited_segment_metas = self.0.segment_manager.committed_segment_metas();
+            let mut commited_segment_metas = self.segment_manager.committed_segment_metas();
 
             // We sort segment_readers by number of documents.
             // This is an heuristic to make multithreading more efficient.
@@ -259,16 +298,18 @@ impl SegmentUpdater {
                 opstamp,
                 payload: commit_message,
             };
-            save_metas(&index_meta, directory.box_clone().borrow_mut())
+            // TODO add context to the error.
-                .expect("Could not save metas.");
+            save_metas(&index_meta, directory.box_clone().borrow_mut())?;
             self.store_meta(&index_meta);
         }
+        Ok(())
     }
 
-    pub fn garbage_collect_files(&self) -> CpuFuture<(), TantivyError> {
+    pub fn schedule_garbage_collect(
-        self.run_async(move |segment_updater| {
+        &self,
-            segment_updater.garbage_collect_files_exec();
+    ) -> impl Future<Output = crate::Result<GarbageCollectionResult>> {
-        })
+        let garbage_collect_future = garbage_collect_files(self.clone());
+        self.schedule_future(garbage_collect_future)
     }
 
     /// List the files that are useful to the index.
@@ -276,148 +317,130 @@ impl SegmentUpdater {
     /// This does not include lock files, or files that are obsolete
     /// but have not yet been deleted by the garbage collector.
     fn list_files(&self) -> HashSet<PathBuf> {
-        let mut files = HashSet::new();
+        let mut files: HashSet<PathBuf> = self
+            .index
+            .list_all_segment_metas()
+            .into_iter()
+            .flat_map(|segment_meta| segment_meta.list_files())
+            .collect();
         files.insert(META_FILEPATH.to_path_buf());
-        for segment_meta in self.0.index.list_all_segment_metas() {
-            files.extend(segment_meta.list_files());
-        }
         files
     }
 
-    fn garbage_collect_files_exec(&self) {
+    pub fn schedule_commit(
-        info!("Running garbage collection");
+        &self,
-        let mut index = self.0.index.clone();
+        opstamp: Opstamp,
-        index.directory_mut().garbage_collect(|| self.list_files());
+        payload: Option<String>,
-    }
+    ) -> impl Future<Output = crate::Result<()>> {
-
+        let segment_updater: SegmentUpdater = self.clone();
-    pub fn commit(&self, opstamp: Opstamp, payload: Option<String>) -> Result<()> {
+        self.schedule_future(async move {
-        self.run_async(move |segment_updater| {
+            let segment_entries = segment_updater.purge_deletes(opstamp)?;
-            if segment_updater.is_alive() {
+            segment_updater.segment_manager.commit(segment_entries);
-                let segment_entries = segment_updater
+            segment_updater.save_metas(opstamp, payload)?;
-                    .purge_deletes(opstamp)
+            let _ = garbage_collect_files(segment_updater.clone()).await;
-                    .expect("Failed purge deletes");
+            segment_updater.consider_merge_options().await;
-                segment_updater.0.segment_manager.commit(segment_entries);
+            Ok(())
-                segment_updater.save_metas(opstamp, payload);
-                segment_updater.garbage_collect_files_exec();
-                segment_updater.consider_merge_options();
-            }
         })
-        .wait()
-    }
-
-    pub fn start_merge(&self, segment_ids: &[SegmentId]) -> Result<Receiver<SegmentMeta>> {
-        let commit_opstamp = self.load_metas().opstamp;
-        let merge_operation = MergeOperation::new(
-            &self.0.merge_operations,
-            commit_opstamp,
-            segment_ids.to_vec(),
-        );
-        self.run_async(move |segment_updater| segment_updater.start_merge_impl(merge_operation))
-            .wait()?
     }
 
     fn store_meta(&self, index_meta: &IndexMeta) {
-        *self.0.active_metas.write().unwrap() = Arc::new(index_meta.clone());
+        *self.active_metas.write().unwrap() = Arc::new(index_meta.clone());
-    }
-    fn load_metas(&self) -> Arc<IndexMeta> {
-        self.0.active_metas.read().unwrap().clone()
     }
 
+    fn load_metas(&self) -> Arc<IndexMeta> {
+        self.active_metas.read().unwrap().clone()
+    }
+
+    pub(crate) fn make_merge_operation(&self, segment_ids: &[SegmentId]) -> MergeOperation {
+        let commit_opstamp = self.load_metas().opstamp;
+        MergeOperation::new(&self.merge_operations, commit_opstamp, segment_ids.to_vec())
+    }
+
+    // Starts a merge operation. This function will block until the merge operation is effectively
+    // started. Note that it does not wait for the merge to terminate.
+    // The calling thread should not be block for a long time, as this only involve waiting for the
+    // `SegmentUpdater` queue which in turns only contains lightweight operations.
+    //
+    // The merge itself happens on a different thread.
+    //
+    // When successful, this function returns a `Future` for a `Result<SegmentMeta>` that represents
+    // the actual outcome of the merge operation.
+    //
+    // It returns an error if for some reason the merge operation could not be started.
+    //
+    // At this point an error is not necessarily the sign of a malfunction.
+    // (e.g. A rollback could have happened, between the instant when the merge operaiton was
+    // suggested and the moment when it ended up being executed.)
+    //
     // `segment_ids` is required to be non-empty.
-    fn start_merge_impl(&self, merge_operation: MergeOperation) -> Result<Receiver<SegmentMeta>> {
+    pub fn start_merge(
+        &self,
+        merge_operation: MergeOperation,
+    ) -> crate::Result<impl Future<Output = crate::Result<SegmentMeta>>> {
         assert!(
             !merge_operation.segment_ids().is_empty(),
             "Segment_ids cannot be empty."
         );
 
-        let segment_updater_clone = self.clone();
+        let segment_updater = self.clone();
         let segment_entries: Vec<SegmentEntry> = self
-            .0
             .segment_manager
             .start_merge(merge_operation.segment_ids())?;
 
-        //        let segment_ids_vec = merge_operation.segment_ids.to_vec();
+        info!("Starting merge  - {:?}", merge_operation.segment_ids());
 
-        let merging_thread_id = self.get_merging_thread_id();
+        let (merging_future_send, merging_future_recv) =
-        info!(
+            oneshot::channel::<crate::Result<SegmentMeta>>();
-            "Starting merge thread #{} - {:?}",
-            merging_thread_id,
-            merge_operation.segment_ids()
-        );
-        let (merging_future_send, merging_future_recv) = oneshot();
 
-        // first we need to apply deletes to our segment.
+        self.merge_thread_pool.spawn_ok(async move {
-        let merging_join_handle = thread::Builder::new()
+            // The fact that `merge_operation` is moved here is important.
-            .name(format!("mergingthread-{}", merging_thread_id))
+            // Its lifetime is used to track how many merging thread are currently running,
-            .spawn(move || {
+            // as well as which segment is currently in merge and therefore should not be
-                // first we need to apply deletes to our segment.
+            // candidate for another merge.
-                let merge_result = perform_merge(
+            match merge(
-                    &merge_operation,
+                &segment_updater.index,
-                    &segment_updater_clone.0.index,
+                segment_entries,
-                    segment_entries,
+                merge_operation.target_opstamp(),
-                );
+            ) {
-
+                Ok(after_merge_segment_entry) => {
-                match merge_result {
+                    let segment_meta = segment_updater
-                    Ok(after_merge_segment_entry) => {
+                        .end_merge(merge_operation, after_merge_segment_entry)
-                        let merged_segment_meta = after_merge_segment_entry.meta().clone();
+                        .await;
-                        segment_updater_clone
+                    let _send_result = merging_future_send.send(segment_meta);
-                            .end_merge(merge_operation, after_merge_segment_entry)
+                }
-                            .expect("Segment updater thread is corrupted.");
+                Err(e) => {
-
+                    warn!(
-                        // the future may fail if the listener of the oneshot future
+                        "Merge of {:?} was cancelled: {:?}",
-                        // has been destroyed.
+                        merge_operation.segment_ids().to_vec(),
-                        //
+                        e
-                        // This is not a problem here, so we just ignore any
+                    );
-                        // possible error.
+                    // ... cancel merge
-                        let _merging_future_res = merging_future_send.send(merged_segment_meta);
+                    if cfg!(test) {
-                    }
+                        panic!("Merge failed.");
-                    Err(e) => {
-                        warn!(
-                            "Merge of {:?} was cancelled: {:?}",
-                            merge_operation.segment_ids(),
-                            e
-                        );
-                        // ... cancel merge
-                        if cfg!(test) {
-                            panic!("Merge failed.");
-                        }
-                        // As `merge_operation` will be dropped, the segment in merge state will
-                        // be available for merge again.
-                        // `merging_future_send` will be dropped, sending an error to the future.
                     }
                 }
-                segment_updater_clone
+            }
-                    .0
+        });
-                    .merging_threads
+
-                    .write()
+        Ok(merging_future_recv
-                    .unwrap()
+            .unwrap_or_else(|_| Err(crate::TantivyError::SystemError("Merge failed".to_string()))))
-                    .remove(&merging_thread_id);
-                Ok(())
-            })
-            .expect("Failed to spawn a thread.");
-        self.0
-            .merging_threads
-            .write()
-            .unwrap()
-            .insert(merging_thread_id, merging_join_handle);
-        Ok(merging_future_recv)
     }
 
-    fn consider_merge_options(&self) {
+    async fn consider_merge_options(&self) {
-        let merge_segment_ids: HashSet<SegmentId> = self.0.merge_operations.segment_in_merge();
+        let merge_segment_ids: HashSet<SegmentId> = self.merge_operations.segment_in_merge();
         let (committed_segments, uncommitted_segments) =
-            get_mergeable_segments(&merge_segment_ids, &self.0.segment_manager);
+            get_mergeable_segments(&merge_segment_ids, &self.segment_manager);
 
         // Committed segments cannot be merged with uncommitted_segments.
         // We therefore consider merges using these two sets of segments independently.
         let merge_policy = self.get_merge_policy();
 
-        let current_opstamp = self.0.stamper.stamp();
+        let current_opstamp = self.stamper.stamp();
         let mut merge_candidates: Vec<MergeOperation> = merge_policy
             .compute_merge_candidates(&uncommitted_segments)
             .into_iter()
             .map(|merge_candidate| {
-                MergeOperation::new(&self.0.merge_operations, current_opstamp, merge_candidate.0)
+                MergeOperation::new(&self.merge_operations, current_opstamp, merge_candidate.0)
             })
             .collect();
 
@@ -425,25 +448,18 @@ impl SegmentUpdater {
         let committed_merge_candidates = merge_policy
             .compute_merge_candidates(&committed_segments)
             .into_iter()
-            .map(|merge_candidate| {
+            .map(|merge_candidate: MergeCandidate| {
-                MergeOperation::new(&self.0.merge_operations, commit_opstamp, merge_candidate.0)
+                MergeOperation::new(&self.merge_operations, commit_opstamp, merge_candidate.0)
             })
             .collect::<Vec<_>>();
         merge_candidates.extend(committed_merge_candidates.into_iter());
 
         for merge_operation in merge_candidates {
-            match self.start_merge_impl(merge_operation) {
+            if let Err(err) = self.start_merge(merge_operation) {
-                Ok(merge_future) => {
+                warn!(
-                    if let Err(e) = merge_future.fuse().poll() {
+                    "Starting the merge failed for the following reason. This is not fatal. {}",
-                        error!("The merge task failed quickly after starting: {:?}", e);
+                    err
-                    }
+                );
-                }
-                Err(err) => {
-                    warn!(
-                        "Starting the merge failed for the following reason. This is not fatal. {}",
-                        err
-                    );
-                }
             }
         }
     }
@@ -452,15 +468,17 @@ impl SegmentUpdater {
         &self,
         merge_operation: MergeOperation,
         mut after_merge_segment_entry: SegmentEntry,
-    ) -> Result<()> {
+    ) -> impl Future<Output = crate::Result<SegmentMeta>> {
-        self.run_async(move |segment_updater| {
+        let segment_updater = self.clone();
+        let after_merge_segment_meta = after_merge_segment_entry.meta().clone();
+        let end_merge_future = self.schedule_future(async move {
             info!("End merge {:?}", after_merge_segment_entry.meta());
             {
                 let mut delete_cursor = after_merge_segment_entry.delete_cursor().clone();
                 if let Some(delete_operation) = delete_cursor.get() {
                     let committed_opstamp = segment_updater.load_metas().opstamp;
                     if delete_operation.opstamp < committed_opstamp {
-                        let index = &segment_updater.0.index;
+                        let index = &segment_updater.index;
                         let segment = index.segment(after_merge_segment_entry.meta().clone());
                         if let Err(e) = advance_deletes(
                             segment,
@@ -478,28 +496,33 @@ impl SegmentUpdater {
                             // ... cancel merge
                             // `merge_operations` are tracked. As it is dropped, the
                             // the segment_ids will be available again for merge.
-                            return;
+                            return Err(e);
                         }
                     }
                 }
                 let previous_metas = segment_updater.load_metas();
-                segment_updater
+                let segments_status = segment_updater
-                    .0
                     .segment_manager
-                    .end_merge(merge_operation.segment_ids(), after_merge_segment_entry);
+                    .end_merge(merge_operation.segment_ids(), after_merge_segment_entry)?;
-                segment_updater.consider_merge_options();
+
-                segment_updater.save_metas(previous_metas.opstamp, previous_metas.payload.clone());
+                if segments_status == SegmentsStatus::Committed {
+                    segment_updater
+                        .save_metas(previous_metas.opstamp, previous_metas.payload.clone())?;
+                }
+
+                segment_updater.consider_merge_options().await;
             } // we drop all possible handle to a now useless `SegmentMeta`.
-            segment_updater.garbage_collect_files_exec();
+            let _ = garbage_collect_files(segment_updater).await;
-        })
+            Ok(())
-        .wait()
+        });
+        end_merge_future.map_ok(|_| after_merge_segment_meta)
     }
 
     /// Wait for current merging threads.
     ///
     /// Upon termination of the current merging threads,
     /// merge opportunity may appear.
-    //
+    ///
     /// We keep waiting until the merge policy judges that
     /// no opportunity is available.
     ///
@@ -510,26 +533,9 @@ impl SegmentUpdater {
     ///
     /// Obsolete files will eventually be cleaned up
     /// by the directory garbage collector.
-    pub fn wait_merging_thread(&self) -> Result<()> {
+    pub fn wait_merging_thread(&self) -> crate::Result<()> {
-        loop {
+        self.merge_operations.wait_until_empty();
-            let merging_threads: HashMap<usize, JoinHandle<Result<()>>> = {
+        Ok(())
-                let mut merging_threads = self.0.merging_threads.write().unwrap();
-                mem::replace(merging_threads.deref_mut(), HashMap::new())
-            };
-            if merging_threads.is_empty() {
-                return Ok(());
-            }
-            debug!("wait merging thread {}", merging_threads.len());
-            for (_, merging_thread_handle) in merging_threads {
-                merging_thread_handle
-                    .join()
-                    .map(|_| ())
-                    .map_err(|_| TantivyError::ErrorInThread("Merging thread failed.".into()))?;
-            }
-            // Our merging thread may have queued their completed merged segment.
-            // Let's wait for that too.
-            self.run_async(move |_| {}).wait()?;
-        }
     }
 }
 
@@ -685,7 +691,6 @@ mod tests {
         index_writer.segment_updater().remove_all_segments();
         let seg_vec = index_writer
             .segment_updater()
-            .0
             .segment_manager
             .segment_entries();
         assert!(seg_vec.is_empty());

src/indexer/segment_writer.rs

@@ -2,7 +2,7 @@ use super::operation::AddOperation;
 use crate::core::Segment;
 use crate::core::SerializableSegment;
 use crate::fastfield::FastFieldsWriter;
-use crate::fieldnorm::FieldNormsWriter;
+use crate::fieldnorm::{FieldNormReaders, FieldNormsWriter};
 use crate::indexer::segment_serializer::SegmentSerializer;
 use crate::postings::compute_table_size;
 use crate::postings::MultiFieldPostingsWriter;
@@ -11,21 +11,18 @@ use crate::schema::Schema;
 use crate::schema::Term;
 use crate::schema::Value;
 use crate::schema::{Field, FieldEntry};
-use crate::tokenizer::BoxedTokenizer;
+use crate::tokenizer::{BoxTokenStream, PreTokenizedStream};
-use crate::tokenizer::FacetTokenizer;
+use crate::tokenizer::{FacetTokenizer, TextAnalyzer};
-use crate::tokenizer::PreTokenizedStream;
+use crate::tokenizer::{TokenStreamChain, Tokenizer};
-use crate::tokenizer::{TokenStream, TokenStreamChain, Tokenizer};
-use crate::DocId;
 use crate::Opstamp;
-use crate::Result;
+use crate::{DocId, SegmentComponent};
-use crate::TantivyError;
 use std::io;
 use std::str;
 
 /// Computes the initial size of the hash table.
 ///
 /// Returns a number of bit `b`, such that the recommended initial table size is 2^b.
-fn initial_table_size(per_thread_memory_budget: usize) -> Result<usize> {
+fn initial_table_size(per_thread_memory_budget: usize) -> crate::Result<usize> {
     let table_memory_upper_bound = per_thread_memory_budget / 3;
     if let Some(limit) = (10..)
         .take_while(|num_bits: &usize| compute_table_size(*num_bits) < table_memory_upper_bound)
@@ -33,7 +30,7 @@ fn initial_table_size(per_thread_memory_budget: usize) -> Result<usize> {
     {
         Ok(limit.min(19)) // we cap it at 2^19 = 512K.
     } else {
-        Err(TantivyError::InvalidArgument(
+        Err(crate::TantivyError::InvalidArgument(
             format!("per thread memory budget (={}) is too small. Raise the memory budget or lower the number of threads.", per_thread_memory_budget)))
     }
 }
@@ -50,7 +47,7 @@ pub struct SegmentWriter {
     fast_field_writers: FastFieldsWriter,
     fieldnorms_writer: FieldNormsWriter,
     doc_opstamps: Vec<Opstamp>,
-    tokenizers: Vec<Option<BoxedTokenizer>>,
+    tokenizers: Vec<Option<TextAnalyzer>>,
 }
 
 impl SegmentWriter {
@@ -65,11 +62,12 @@ impl SegmentWriter {
     /// - schema
     pub fn for_segment(
         memory_budget: usize,
-        mut segment: Segment,
+        segment: Segment,
         schema: &Schema,
-    ) -> Result<SegmentWriter> {
+    ) -> crate::Result<SegmentWriter> {
+        let tokenizer_manager = segment.index().tokenizers().clone();
         let table_num_bits = initial_table_size(memory_budget)?;
-        let segment_serializer = SegmentSerializer::for_segment(&mut segment)?;
+        let segment_serializer = SegmentSerializer::for_segment(segment)?;
         let multifield_postings = MultiFieldPostingsWriter::new(schema, table_num_bits);
         let tokenizers = schema
             .fields()
@@ -79,7 +77,7 @@ impl SegmentWriter {
                         .get_indexing_options()
                         .and_then(|text_index_option| {
                             let tokenizer_name = &text_index_option.tokenizer();
-                            segment.index().tokenizers().get(tokenizer_name)
+                            tokenizer_manager.get(tokenizer_name)
                         }),
                     _ => None,
                 },
@@ -100,7 +98,7 @@ impl SegmentWriter {
     ///
     /// Finalize consumes the `SegmentWriter`, so that it cannot
     /// be used afterwards.
-    pub fn finalize(mut self) -> Result<Vec<u64>> {
+    pub fn finalize(mut self) -> crate::Result<Vec<u64>> {
         self.fieldnorms_writer.fill_up_to_max_doc(self.max_doc);
         write(
             &self.multifield_postings,
@@ -153,13 +151,13 @@ impl SegmentWriter {
                         if let Some(unordered_term_id) = unordered_term_id_opt {
                             self.fast_field_writers
                                 .get_multivalue_writer(field)
-                                .expect("multified writer for facet missing")
+                                .expect("writer for facet missing")
                                 .add_val(unordered_term_id);
                         }
                     }
                 }
                 FieldType::Str(_) => {
-                    let mut token_streams: Vec<Box<dyn TokenStream>> = vec![];
+                    let mut token_streams: Vec<BoxTokenStream> = vec![];
                     let mut offsets = vec![];
                     let mut total_offset = 0;
 
@@ -172,7 +170,7 @@ impl SegmentWriter {
                                 }
 
                                 token_streams
-                                    .push(Box::new(PreTokenizedStream::from(tok_str.clone())));
+                                    .push(PreTokenizedStream::from(tok_str.clone()).into());
                             }
                             Value::Str(ref text) => {
                                 if let Some(ref mut tokenizer) =
@@ -191,8 +189,7 @@ impl SegmentWriter {
                     let num_tokens = if token_streams.is_empty() {
                         0
                     } else {
-                        let mut token_stream: Box<dyn TokenStream> =
+                        let mut token_stream = TokenStreamChain::new(offsets, token_streams);
-                            Box::new(TokenStreamChain::new(offsets, token_streams));
                         self.multifield_postings
                             .index_text(doc_id, field, &mut token_stream)
                     };
@@ -249,6 +246,7 @@ impl SegmentWriter {
             }
         }
         doc.filter_fields(|field| schema.get_field_entry(field).is_stored());
+        doc.prepare_for_store();
         let doc_writer = self.segment_serializer.get_store_writer();
         doc_writer.store(&doc)?;
         self.max_doc += 1;
@@ -282,16 +280,23 @@ fn write(
     fast_field_writers: &FastFieldsWriter,
     fieldnorms_writer: &FieldNormsWriter,
     mut serializer: SegmentSerializer,
-) -> Result<()> {
+) -> crate::Result<()> {
-    let term_ord_map = multifield_postings.serialize(serializer.get_postings_serializer())?;
+    if let Some(fieldnorms_serializer) = serializer.extract_fieldnorms_serializer() {
+        fieldnorms_writer.serialize(fieldnorms_serializer)?;
+    }
+    let fieldnorm_data = serializer
+        .segment()
+        .open_read(SegmentComponent::FIELDNORMS)?;
+    let fieldnorm_readers = FieldNormReaders::open(fieldnorm_data)?;
+    let term_ord_map =
+        multifield_postings.serialize(serializer.get_postings_serializer(), fieldnorm_readers)?;
     fast_field_writers.serialize(serializer.get_fast_field_serializer(), &term_ord_map)?;
-    fieldnorms_writer.serialize(serializer.get_fieldnorms_serializer())?;
     serializer.close()?;
     Ok(())
 }
 
 impl SerializableSegment for SegmentWriter {
-    fn write(&self, serializer: SegmentSerializer) -> Result<u32> {
+    fn write(&self, serializer: SegmentSerializer) -> crate::Result<u32> {
         let max_doc = self.max_doc;
         write(
             &self.multifield_postings,

src/indexer/stamper.rs

@@ -1,18 +1,76 @@
 use crate::Opstamp;
 use std::ops::Range;
-use std::sync::atomic::{AtomicU64, Ordering};
+use std::sync::atomic::Ordering;
 use std::sync::Arc;
 
+#[cfg(not(target_arch = "arm"))]
+mod atomic_impl {
+
+    use crate::Opstamp;
+    use std::sync::atomic::{AtomicU64, Ordering};
+
+    #[derive(Default)]
+    pub struct AtomicU64Wrapper(AtomicU64);
+
+    impl AtomicU64Wrapper {
+        pub fn new(first_opstamp: Opstamp) -> AtomicU64Wrapper {
+            AtomicU64Wrapper(AtomicU64::new(first_opstamp as u64))
+        }
+
+        pub fn fetch_add(&self, val: u64, order: Ordering) -> u64 {
+            self.0.fetch_add(val as u64, order) as u64
+        }
+
+        pub fn revert(&self, val: u64, order: Ordering) -> u64 {
+            self.0.store(val, order);
+            val
+        }
+    }
+}
+
+#[cfg(target_arch = "arm")]
+mod atomic_impl {
+
+    use crate::Opstamp;
+    /// Under other architecture, we rely on a mutex.
+    use std::sync::atomic::Ordering;
+    use std::sync::RwLock;
+
+    #[derive(Default)]
+    pub struct AtomicU64Wrapper(RwLock<u64>);
+
+    impl AtomicU64Wrapper {
+        pub fn new(first_opstamp: Opstamp) -> AtomicU64Wrapper {
+            AtomicU64Wrapper(RwLock::new(first_opstamp))
+        }
+
+        pub fn fetch_add(&self, incr: u64, _order: Ordering) -> u64 {
+            let mut lock = self.0.write().unwrap();
+            let previous_val = *lock;
+            *lock = previous_val + incr;
+            previous_val
+        }
+
+        pub fn revert(&self, val: u64, _order: Ordering) -> u64 {
+            let mut lock = self.0.write().unwrap();
+            *lock = val;
+            val
+        }
+    }
+}
+
+use self::atomic_impl::AtomicU64Wrapper;
+
 /// Stamper provides Opstamps, which is just an auto-increment id to label
 /// an operation.
 ///
 /// Cloning does not "fork" the stamp generation. The stamper actually wraps an `Arc`.
 #[derive(Clone, Default)]
-pub struct Stamper(Arc<AtomicU64>);
+pub struct Stamper(Arc<AtomicU64Wrapper>);
 
 impl Stamper {
     pub fn new(first_opstamp: Opstamp) -> Stamper {
-        Stamper(Arc::new(AtomicU64::new(first_opstamp)))
+        Stamper(Arc::new(AtomicU64Wrapper::new(first_opstamp)))
     }
 
     pub fn stamp(&self) -> Opstamp {
@@ -31,8 +89,7 @@ impl Stamper {
 
     /// Reverts the stamper to a given `Opstamp` value and returns it
     pub fn revert(&self, to_opstamp: Opstamp) -> Opstamp {
-        self.0.store(to_opstamp, Ordering::SeqCst);
+        self.0.revert(to_opstamp, Ordering::SeqCst)
-        to_opstamp
     }
 }
 

src/lib.rs

@@ -98,9 +98,6 @@
 //! [literate programming](https://tantivy-search.github.io/examples/basic_search.html) /
 //! [source code](https://github.com/tantivy-search/tantivy/blob/master/examples/basic_search.rs))
 
-#[macro_use]
-extern crate serde_derive;
-
 #[cfg_attr(test, macro_use)]
 extern crate serde_json;
 
@@ -121,13 +118,13 @@ mod functional_test;
 mod macros;
 
 pub use crate::error::TantivyError;
-
-#[deprecated(since = "0.7.0", note = "please use `tantivy::TantivyError` instead")]
-pub use crate::error::TantivyError as Error;
 pub use chrono;
 
 /// Tantivy result.
-pub type Result<T> = std::result::Result<T, error::TantivyError>;
+///
+/// Within tantivy, please avoid importing `Result` using `use crate::Result`
+/// and instead, refer to this as `crate::Result<T>`.
+pub type Result<T> = std::result::Result<T, TantivyError>;
 
 /// Tantivy DateTime
 pub type DateTime = chrono::DateTime<chrono::Utc>;
@@ -159,22 +156,70 @@ mod snippet;
 pub use self::snippet::{Snippet, SnippetGenerator};
 
 mod docset;
-pub use self::docset::{DocSet, SkipResult};
+pub use self::docset::{DocSet, TERMINATED};
-
 pub use crate::common::{f64_to_u64, i64_to_u64, u64_to_f64, u64_to_i64};
-pub use crate::core::SegmentComponent;
+pub use crate::core::{Executor, SegmentComponent};
 pub use crate::core::{Index, IndexMeta, Searcher, Segment, SegmentId, SegmentMeta};
 pub use crate::core::{InvertedIndexReader, SegmentReader};
 pub use crate::directory::Directory;
+pub use crate::indexer::operation::UserOperation;
 pub use crate::indexer::IndexWriter;
 pub use crate::postings::Postings;
 pub use crate::reader::LeasedItem;
 pub use crate::schema::{Document, Term};
+use std::fmt;
 
-/// Expose the current version of tantivy, as well
+use once_cell::sync::Lazy;
-/// whether it was compiled with the simd compression.
+use serde::{Deserialize, Serialize};
-pub fn version() -> &'static str {
+
-    env!("CARGO_PKG_VERSION")
+/// Index format version.
+const INDEX_FORMAT_VERSION: u32 = 2;
+
+/// Structure version for the index.
+#[derive(Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct Version {
+    major: u32,
+    minor: u32,
+    patch: u32,
+    index_format_version: u32,
+    store_compression: String,
+}
+
+impl fmt::Debug for Version {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        write!(f, "{}", self.to_string())
+    }
+}
+
+static VERSION: Lazy<Version> = Lazy::new(|| Version {
+    major: env!("CARGO_PKG_VERSION_MAJOR").parse().unwrap(),
+    minor: env!("CARGO_PKG_VERSION_MINOR").parse().unwrap(),
+    patch: env!("CARGO_PKG_VERSION_PATCH").parse().unwrap(),
+    index_format_version: INDEX_FORMAT_VERSION,
+    store_compression: crate::store::COMPRESSION.to_string(),
+});
+
+impl ToString for Version {
+    fn to_string(&self) -> String {
+        format!(
+            "tantivy v{}.{}.{}, index_format v{}, store_compression: {}",
+            self.major, self.minor, self.patch, self.index_format_version, self.store_compression
+        )
+    }
+}
+
+static VERSION_STRING: Lazy<String> = Lazy::new(|| VERSION.to_string());
+
+/// Expose the current version of tantivy as found in Cargo.toml during compilation.
+/// eg. "0.11.0" as well as the compression scheme used in the docstore.
+pub fn version() -> &'static Version {
+    &VERSION
+}
+
+/// Exposes the complete version of tantivy as found in Cargo.toml during compilation as a string.
+/// eg. "tantivy v0.11.0, index_format v1, store_compression: lz4".
+pub fn version_string() -> &'static str {
+    VERSION_STRING.as_str()
 }
 
 /// Defines tantivy's merging strategy
@@ -200,11 +245,18 @@ pub type DocId = u32;
 /// with opstamp `n+1`.
 pub type Opstamp = u64;
 
-/// A f32 that represents the relevance of the document to the query
+/// A Score that represents the relevance of the document to the query
 ///
-/// This is modelled internally as a `f32`. The
+/// This is modelled internally as a `f64`, because tantivy was compiled with the `scoref64`
-/// larger the number, the more relevant the document
+/// feature. The larger the number, the more relevant the document is to the search query.
-/// to the search
+#[cfg(feature = "scoref64")]
+pub type Score = f64;
+
+/// A Score that represents the relevance of the document to the query
+///
+/// This is modelled internally as a `f32`. The larger the number, the more relevant
+/// the document to the search query.
+#[cfg(not(feature = "scoref64"))]
 pub type Score = f32;
 
 /// A `SegmentLocalId` identifies a segment.
@@ -237,10 +289,9 @@ pub struct DocAddress(pub SegmentLocalId, pub DocId);
 
 #[cfg(test)]
 mod tests {
-
     use crate::collector::tests::TEST_COLLECTOR_WITH_SCORE;
     use crate::core::SegmentReader;
-    use crate::docset::DocSet;
+    use crate::docset::{DocSet, TERMINATED};
     use crate::query::BooleanQuery;
     use crate::schema::*;
     use crate::DocAddress;
@@ -253,17 +304,26 @@ mod tests {
     use rand::rngs::StdRng;
     use rand::{Rng, SeedableRng};
 
-    pub fn assert_nearly_equals(expected: f32, val: f32) {
+    /// Checks if left and right are close one to each other.
-        assert!(
+    /// Panics if the two values are more than 0.5% apart.
-            nearly_equals(val, expected),
+    #[macro_export]
-            "Got {}, expected {}.",
+    macro_rules! assert_nearly_equals {
-            val,
+        ($left:expr, $right:expr) => {{
-            expected
+            match (&$left, &$right) {
-        );
+                (left_val, right_val) => {
-    }
+                    let diff = (left_val - right_val).abs();
-
+                    let add = left_val.abs() + right_val.abs();
-    pub fn nearly_equals(a: f32, b: f32) -> bool {
+                    if diff > 0.0005 * add {
-        (a - b).abs() < 0.0005 * (a + b).abs()
+                        panic!(
+                            r#"assertion failed: `(left ~= right)`
+  left: `{:?}`,
+ right: `{:?}`"#,
+                            &*left_val, &*right_val
+                        )
+                    }
+                }
+            }
+        }};
     }
 
     pub fn generate_nonunique_unsorted(max_value: u32, n_elems: usize) -> Vec<u32> {
@@ -287,6 +347,18 @@ mod tests {
         sample_with_seed(n, ratio, 4)
     }
 
+    #[test]
+    #[cfg(not(feature = "lz4"))]
+    fn test_version_string() {
+        use regex::Regex;
+        let regex_ptn = Regex::new(
+            "tantivy v[0-9]{1,3}\\.[0-9]{1,3}\\.[0-9]{1,3}\\.{0,10}, index_format v[0-9]{1,5}",
+        )
+        .unwrap();
+        let version = super::version().to_string();
+        assert!(regex_ptn.find(&version).is_some());
+    }
+
     #[test]
     #[cfg(feature = "mmap")]
     fn test_indexing() {
@@ -324,19 +396,12 @@ mod tests {
             index_writer.commit().unwrap();
         }
         {
-            {
+            index_writer.add_document(doc!(text_field=>"a"));
-                let doc = doc!(text_field=>"a");
+            index_writer.add_document(doc!(text_field=>"a a"));
-                index_writer.add_document(doc);
-            }
-            {
-                let doc = doc!(text_field=>"a a");
-                index_writer.add_document(doc);
-            }
             index_writer.commit().unwrap();
         }
         {
-            let doc = doc!(text_field=>"c");
+            index_writer.add_document(doc!(text_field=>"c"));
-            index_writer.add_document(doc);
             index_writer.commit().unwrap();
         }
         {
@@ -415,10 +480,12 @@ mod tests {
     }
 
     fn advance_undeleted(docset: &mut dyn DocSet, reader: &SegmentReader) -> bool {
-        while docset.advance() {
+        let mut doc = docset.advance();
-            if !reader.is_deleted(docset.doc()) {
+        while doc != TERMINATED {
+            if !reader.is_deleted(doc) {
                 return true;
             }
+            doc = docset.advance();
         }
         false
     }
@@ -584,9 +651,8 @@ mod tests {
             .inverted_index(term.field())
             .read_postings(&term, IndexRecordOption::Basic)
             .unwrap();
-        assert!(postings.advance());
         assert_eq!(postings.doc(), 0);
-        assert!(!postings.advance());
+        assert_eq!(postings.advance(), TERMINATED);
     }
 
     #[test]
@@ -608,9 +674,8 @@ mod tests {
             .inverted_index(term.field())
             .read_postings(&term, IndexRecordOption::Basic)
             .unwrap();
-        assert!(postings.advance());
         assert_eq!(postings.doc(), 0);
-        assert!(!postings.advance());
+        assert_eq!(postings.advance(), TERMINATED);
     }
 
     #[test]
@@ -632,9 +697,8 @@ mod tests {
             .inverted_index(term.field())
             .read_postings(&term, IndexRecordOption::Basic)
             .unwrap();
-        assert!(postings.advance());
         assert_eq!(postings.doc(), 0);
-        assert!(!postings.advance());
+        assert_eq!(postings.advance(), TERMINATED);
     }
 
     #[test]
@@ -703,10 +767,8 @@ mod tests {
         {
             // writing the segment
             let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-            {
+            let doc = doc!(text_field=>"af af af bc bc");
-                let doc = doc!(text_field=>"af af af bc bc");
+            index_writer.add_document(doc);
-                index_writer.add_document(doc);
-            }
             index_writer.commit().unwrap();
         }
         {
@@ -722,10 +784,9 @@ mod tests {
             let mut postings = inverted_index
                 .read_postings(&term_af, IndexRecordOption::WithFreqsAndPositions)
                 .unwrap();
-            assert!(postings.advance());
             assert_eq!(postings.doc(), 0);
             assert_eq!(postings.term_freq(), 3);
-            assert!(!postings.advance());
+            assert_eq!(postings.advance(), TERMINATED);
         }
     }
 
@@ -882,4 +943,73 @@ mod tests {
             assert_eq!(fast_field_reader.get(0), 4f64)
         }
     }
+
+    // motivated by #729
+    #[test]
+    fn test_update_via_delete_insert() {
+        use crate::collector::Count;
+        use crate::indexer::NoMergePolicy;
+        use crate::query::AllQuery;
+        use crate::SegmentId;
+        use futures::executor::block_on;
+
+        const DOC_COUNT: u64 = 2u64;
+
+        let mut schema_builder = SchemaBuilder::default();
+        let id = schema_builder.add_u64_field("id", INDEXED);
+        let schema = schema_builder.build();
+
+        let index = Index::create_in_ram(schema.clone());
+        let index_reader = index.reader().unwrap();
+
+        let mut index_writer = index.writer(3_000_000).unwrap();
+        index_writer.set_merge_policy(Box::new(NoMergePolicy));
+
+        for doc_id in 0u64..DOC_COUNT {
+            index_writer.add_document(doc!(id => doc_id));
+        }
+        index_writer.commit().unwrap();
+
+        index_reader.reload().unwrap();
+        let searcher = index_reader.searcher();
+
+        assert_eq!(
+            searcher.search(&AllQuery, &Count).unwrap(),
+            DOC_COUNT as usize
+        );
+
+        // update the 10 elements by deleting and re-adding
+        for doc_id in 0u64..DOC_COUNT {
+            index_writer.delete_term(Term::from_field_u64(id, doc_id));
+            index_writer.commit().unwrap();
+            index_reader.reload().unwrap();
+            let doc = doc!(id =>  doc_id);
+            index_writer.add_document(doc);
+            index_writer.commit().unwrap();
+            index_reader.reload().unwrap();
+            let searcher = index_reader.searcher();
+            // The number of document should be stable.
+            assert_eq!(
+                searcher.search(&AllQuery, &Count).unwrap(),
+                DOC_COUNT as usize
+            );
+        }
+
+        index_reader.reload().unwrap();
+        let searcher = index_reader.searcher();
+        let segment_ids: Vec<SegmentId> = searcher
+            .segment_readers()
+            .into_iter()
+            .map(|reader| reader.segment_id())
+            .collect();
+        block_on(index_writer.merge(&segment_ids)).unwrap();
+
+        index_reader.reload().unwrap();
+        let searcher = index_reader.searcher();
+
+        assert_eq!(
+            searcher.search(&AllQuery, &Count).unwrap(),
+            DOC_COUNT as usize
+        );
+    }
 }

src/macros.rs

@@ -35,9 +35,9 @@
 /// let likes = schema_builder.add_u64_field("num_u64", FAST);
 /// let schema = schema_builder.build();
 /// let doc = doc!(
-/// 	title => "Life Aquatic",
+///     title => "Life Aquatic",
-/// 	author => "Wes Anderson",
+///     author => "Wes Anderson",
-/// 	likes => 4u64
+///     likes => 4u64
 /// );
 /// # }
 /// ```

src/positions/mod.rs

@@ -37,9 +37,9 @@ const LONG_SKIP_INTERVAL: u64 = (LONG_SKIP_IN_BLOCKS * COMPRESSION_BLOCK_SIZE) a
 #[cfg(test)]
 pub mod tests {
 
-    use super::{PositionReader, PositionSerializer};
+    use super::PositionSerializer;
     use crate::directory::ReadOnlySource;
-    use crate::positions::COMPRESSION_BLOCK_SIZE;
+    use crate::positions::reader::PositionReader;
     use std::iter;
 
     fn create_stream_buffer(vals: &[u32]) -> (ReadOnlySource, ReadOnlySource) {
@@ -68,7 +68,7 @@ pub mod tests {
         let mut position_reader = PositionReader::new(stream, skip, 0u64);
         for &n in &[1, 10, 127, 128, 130, 312] {
             let mut v = vec![0u32; n];
-            position_reader.read(&mut v[..n]);
+            position_reader.read(0, &mut v[..]);
             for i in 0..n {
                 assert_eq!(v[i], i as u32);
             }
@@ -76,19 +76,19 @@ pub mod tests {
     }
 
     #[test]
-    fn test_position_skip() {
+    fn test_position_read_with_offset() {
-        let v: Vec<u32> = (0..1_000).collect();
+        let v: Vec<u32> = (0..1000).collect();
         let (stream, skip) = create_stream_buffer(&v[..]);
         assert_eq!(skip.len(), 12);
         assert_eq!(stream.len(), 1168);
-
         let mut position_reader = PositionReader::new(stream, skip, 0u64);
-        position_reader.skip(10);
+        for &offset in &[1u64, 10u64, 127u64, 128u64, 130u64, 312u64] {
-        for &n in &[10, 127, COMPRESSION_BLOCK_SIZE, 130, 312] {
+            for &len in &[1, 10, 130, 500] {
-            let mut v = vec![0u32; n];
+                let mut v = vec![0u32; len];
-            position_reader.read(&mut v[..n]);
+                position_reader.read(offset, &mut v[..]);
-            for i in 0..n {
+                for i in 0..len {
-                assert_eq!(v[i], 10u32 + i as u32);
+                    assert_eq!(v[i], i as u32 + offset as u32);
+                }
             }
         }
     }
@@ -103,11 +103,12 @@ pub mod tests {
         let mut position_reader = PositionReader::new(stream, skip, 0u64);
         let mut buf = [0u32; 7];
         let mut c = 0;
+
+        let mut offset = 0;
         for _ in 0..100 {
-            position_reader.read(&mut buf);
+            position_reader.read(offset, &mut buf);
-            position_reader.read(&mut buf);
+            position_reader.read(offset, &mut buf);
-            position_reader.skip(4);
+            offset += 7;
-            position_reader.skip(3);
             for &el in &buf {
                 assert_eq!(c, el);
                 c += 1;
@@ -115,6 +116,58 @@ pub mod tests {
         }
     }
 
+    #[test]
+    fn test_position_reread_anchor_different_than_block() {
+        let v: Vec<u32> = (0..2_000_000).collect();
+        let (stream, skip) = create_stream_buffer(&v[..]);
+        assert_eq!(skip.len(), 15_749);
+        assert_eq!(stream.len(), 4_987_872);
+        let mut position_reader = PositionReader::new(stream.clone(), skip.clone(), 0);
+        let mut buf = [0u32; 256];
+        position_reader.read(128, &mut buf);
+        for i in 0..256 {
+            assert_eq!(buf[i], (128 + i) as u32);
+        }
+        position_reader.read(128, &mut buf);
+        for i in 0..256 {
+            assert_eq!(buf[i], (128 + i) as u32);
+        }
+    }
+
+    #[test]
+    #[should_panic(expected = "offset arguments should be increasing.")]
+    fn test_position_panic_if_called_previous_anchor() {
+        let v: Vec<u32> = (0..2_000_000).collect();
+        let (stream, skip) = create_stream_buffer(&v[..]);
+        assert_eq!(skip.len(), 15_749);
+        assert_eq!(stream.len(), 4_987_872);
+        let mut buf = [0u32; 1];
+        let mut position_reader = PositionReader::new(stream.clone(), skip.clone(), 200_000);
+        position_reader.read(230, &mut buf);
+        position_reader.read(9, &mut buf);
+    }
+
+    #[test]
+    fn test_positions_bug() {
+        let mut v: Vec<u32> = vec![];
+        for i in 1..200 {
+            for j in 0..i {
+                v.push(j);
+            }
+        }
+        let (stream, skip) = create_stream_buffer(&v[..]);
+        let mut buf = Vec::new();
+        let mut position_reader = PositionReader::new(stream.clone(), skip.clone(), 0);
+        let mut offset = 0;
+        for i in 1..24 {
+            buf.resize(i, 0);
+            position_reader.read(offset, &mut buf[..]);
+            offset += i as u64;
+            let r: Vec<u32> = (0..i).map(|el| el as u32).collect();
+            assert_eq!(buf, &r[..]);
+        }
+    }
+
     #[test]
     fn test_position_long_skip_const() {
         const CONST_VAL: u32 = 9u32;
@@ -124,7 +177,7 @@ pub mod tests {
         assert_eq!(stream.len(), 1_000_000);
         let mut position_reader = PositionReader::new(stream, skip, 128 * 1024);
         let mut buf = [0u32; 1];
-        position_reader.read(&mut buf);
+        position_reader.read(0, &mut buf);
         assert_eq!(buf[0], CONST_VAL);
     }
 
@@ -143,7 +196,7 @@ pub mod tests {
         ] {
             let mut position_reader = PositionReader::new(stream.clone(), skip.clone(), offset);
             let mut buf = [0u32; 1];
-            position_reader.read(&mut buf);
+            position_reader.read(0, &mut buf);
             assert_eq!(buf[0], offset as u32);
         }
     }

src/positions/reader.rs

@@ -3,7 +3,6 @@ use crate::directory::ReadOnlySource;
 use crate::positions::COMPRESSION_BLOCK_SIZE;
 use crate::positions::LONG_SKIP_INTERVAL;
 use crate::positions::LONG_SKIP_IN_BLOCKS;
-use crate::postings::compression::compressed_block_size;
 /// Positions works as a long sequence of compressed block.
 /// All terms are chained one after the other.
 ///
@@ -36,11 +35,10 @@ struct Positions {
 
 impl Positions {
     pub fn new(position_source: ReadOnlySource, skip_source: ReadOnlySource) -> Positions {
-        let skip_len = skip_source.len();
+        let (body, footer) = skip_source.split_from_end(u32::SIZE_IN_BYTES);
-        let (body, footer) = skip_source.split(skip_len - u32::SIZE_IN_BYTES);
         let num_long_skips = u32::deserialize(&mut footer.as_slice()).expect("Index corrupted");
-        let body_split = body.len() - u64::SIZE_IN_BYTES * (num_long_skips as usize);
+        let (skip_source, long_skip_source) =
-        let (skip_source, long_skip_source) = body.split(body_split);
+            body.split_from_end(u64::SIZE_IN_BYTES * (num_long_skips as usize));
         Positions {
             bit_packer: BitPacker4x::new(),
             skip_source,
@@ -63,74 +61,34 @@ impl Positions {
 
     fn reader(&self, offset: u64) -> PositionReader {
         let long_skip_id = (offset / LONG_SKIP_INTERVAL) as usize;
-        let small_skip = (offset % LONG_SKIP_INTERVAL) as usize;
         let offset_num_bytes: u64 = self.long_skip(long_skip_id);
         let mut position_read = OwnedRead::new(self.position_source.clone());
         position_read.advance(offset_num_bytes as usize);
         let mut skip_read = OwnedRead::new(self.skip_source.clone());
         skip_read.advance(long_skip_id * LONG_SKIP_IN_BLOCKS);
-        let mut position_reader = PositionReader {
+        PositionReader {
             bit_packer: self.bit_packer,
             skip_read,
             position_read,
-            inner_offset: 0,
             buffer: Box::new([0u32; 128]),
-            ahead: None,
+            block_offset: std::i64::MAX as u64,
-        };
+            anchor_offset: (long_skip_id as u64) * LONG_SKIP_INTERVAL,
-        position_reader.skip(small_skip);
+            abs_offset: offset,
-        position_reader
+        }
     }
 }
 
+#[derive(Clone)]
 pub struct PositionReader {
     skip_read: OwnedRead,
     position_read: OwnedRead,
     bit_packer: BitPacker4x,
-    inner_offset: usize,
+    buffer: Box<[u32; COMPRESSION_BLOCK_SIZE]>,
-    buffer: Box<[u32; 128]>,
-    ahead: Option<usize>, // if None, no block is loaded.
-                          // if Some(num_blocks), the block currently loaded is num_blocks ahead
-                          // of the block of the next int to read.
-}
 
-// `ahead` represents the offset of the block currently loaded
+    block_offset: u64,
-// compared to the cursor of the actual stream.
+    anchor_offset: u64,
-//
+
-// By contract, when this function is called, the current block has to be
+    abs_offset: u64,
-// decompressed.
-//
-// If the requested number of els ends exactly at a given block, the next
-// block is not decompressed.
-fn read_impl(
-    bit_packer: BitPacker4x,
-    mut position: &[u8],
-    buffer: &mut [u32; 128],
-    mut inner_offset: usize,
-    num_bits: &[u8],
-    output: &mut [u32],
-) -> usize {
-    let mut output_start = 0;
-    let mut output_len = output.len();
-    let mut ahead = 0;
-    loop {
-        let available_len = COMPRESSION_BLOCK_SIZE - inner_offset;
-        // We have enough elements in the current block.
-        // Let's copy the requested elements in the output buffer,
-        // and return.
-        if output_len <= available_len {
-            output[output_start..].copy_from_slice(&buffer[inner_offset..][..output_len]);
-            return ahead;
-        }
-        output[output_start..][..available_len].copy_from_slice(&buffer[inner_offset..]);
-        output_len -= available_len;
-        output_start += available_len;
-        inner_offset = 0;
-        let num_bits = num_bits[ahead];
-        bit_packer.decompress(position, &mut buffer[..], num_bits);
-        let block_len = compressed_block_size(num_bits);
-        position = &position[block_len..];
-        ahead += 1;
-    }
 }
 
 impl PositionReader {
@@ -142,57 +100,65 @@ impl PositionReader {
         Positions::new(position_source, skip_source).reader(offset)
     }
 
-    /// Fills a buffer with the next `output.len()` integers.
+    fn advance_num_blocks(&mut self, num_blocks: usize) {
-    /// This does not consume / advance the stream.
+        let num_bits: usize = self.skip_read.as_ref()[..num_blocks]
-    pub fn read(&mut self, output: &mut [u32]) {
+            .iter()
-        let skip_data = self.skip_read.as_ref();
+            .cloned()
-        let position_data = self.position_read.as_ref();
+            .map(|num_bits| num_bits as usize)
-        let num_bits = self.skip_read.get(0);
+            .sum();
-        if self.ahead != Some(0) {
+        let num_bytes_to_skip = num_bits * COMPRESSION_BLOCK_SIZE / 8;
-            // the block currently available is not the block
+        self.skip_read.advance(num_blocks as usize);
-            // for the current position
+        self.position_read.advance(num_bytes_to_skip);
+    }
+
+    /// Fills a buffer with the positions `[offset..offset+output.len())` integers.
+    ///
+    /// `offset` is required to have a value >= to the offsets given in previous calls
+    /// for the given `PositionReaderAbsolute` instance.
+    pub fn read(&mut self, mut offset: u64, mut output: &mut [u32]) {
+        offset += self.abs_offset;
+        assert!(
+            offset >= self.anchor_offset,
+            "offset arguments should be increasing."
+        );
+        let delta_to_block_offset = offset as i64 - self.block_offset as i64;
+        if delta_to_block_offset < 0 || delta_to_block_offset >= 128 {
+            // The first position is not within the first block.
+            // We need to decompress the first block.
+            let delta_to_anchor_offset = offset - self.anchor_offset;
+            let num_blocks_to_skip =
+                (delta_to_anchor_offset / (COMPRESSION_BLOCK_SIZE as u64)) as usize;
+            self.advance_num_blocks(num_blocks_to_skip);
+            self.anchor_offset = offset - (offset % COMPRESSION_BLOCK_SIZE as u64);
+            self.block_offset = self.anchor_offset;
+            let num_bits = self.skip_read.get(0);
+            self.bit_packer
+                .decompress(self.position_read.as_ref(), self.buffer.as_mut(), num_bits);
+        } else {
+            let num_blocks_to_skip =
+                ((self.block_offset - self.anchor_offset) / COMPRESSION_BLOCK_SIZE as u64) as usize;
+            self.advance_num_blocks(num_blocks_to_skip);
+            self.anchor_offset = self.block_offset;
+        }
+
+        let mut num_bits = self.skip_read.get(0);
+        let mut position_data = self.position_read.as_ref();
+
+        for i in 1.. {
+            let offset_in_block = (offset as usize) % COMPRESSION_BLOCK_SIZE;
+            let remaining_in_block = COMPRESSION_BLOCK_SIZE - offset_in_block;
+            if remaining_in_block >= output.len() {
+                output.copy_from_slice(&self.buffer[offset_in_block..][..output.len()]);
+                break;
+            }
+            output[..remaining_in_block].copy_from_slice(&self.buffer[offset_in_block..]);
+            output = &mut output[remaining_in_block..];
+            offset += remaining_in_block as u64;
+            position_data = &position_data[(num_bits as usize * COMPRESSION_BLOCK_SIZE / 8)..];
+            num_bits = self.skip_read.get(i);
             self.bit_packer
                 .decompress(position_data, self.buffer.as_mut(), num_bits);
-            self.ahead = Some(0);
+            self.block_offset += COMPRESSION_BLOCK_SIZE as u64;
         }
-        let block_len = compressed_block_size(num_bits);
-        self.ahead = Some(read_impl(
-            self.bit_packer,
-            &position_data[block_len..],
-            self.buffer.as_mut(),
-            self.inner_offset,
-            &skip_data[1..],
-            output,
-        ));
-    }
-
-    /// Skip the next `skip_len` integer.
-    ///
-    /// If a full block is skipped, calling
-    /// `.skip(...)` will avoid decompressing it.
-    ///
-    /// May panic if the end of the stream is reached.
-    pub fn skip(&mut self, skip_len: usize) {
-        let skip_len_plus_inner_offset = skip_len + self.inner_offset;
-
-        let num_blocks_to_advance = skip_len_plus_inner_offset / COMPRESSION_BLOCK_SIZE;
-        self.inner_offset = skip_len_plus_inner_offset % COMPRESSION_BLOCK_SIZE;
-
-        self.ahead = self.ahead.and_then(|num_blocks| {
-            if num_blocks >= num_blocks_to_advance {
-                Some(num_blocks - num_blocks_to_advance)
-            } else {
-                None
-            }
-        });
-
-        let skip_len_in_bits = self.skip_read.as_ref()[..num_blocks_to_advance]
-            .iter()
-            .map(|num_bits| *num_bits as usize)
-            .sum::<usize>()
-            * COMPRESSION_BLOCK_SIZE;
-        let skip_len_in_bytes = skip_len_in_bits / 8;
-        self.skip_read.advance(num_blocks_to_advance);
-        self.position_read.advance(skip_len_in_bytes);
     }
 }

src/postings/block_search.rs

@@ -87,6 +87,7 @@ fn exponential_search(arr: &[u32], target: u32) -> (usize, usize) {
     (begin, end)
 }
 
+#[inline(never)]
 fn galloping(block_docs: &[u32], target: u32) -> usize {
     let (start, end) = exponential_search(&block_docs, target);
     start + linear_search(&block_docs[start..end], target)
@@ -106,7 +107,7 @@ impl BlockSearcher {
     /// the target.
     ///
     /// The results should be equivalent to
-    /// ```ignore
+    /// ```compile_fail
     /// block[..]
     //       .iter()
     //       .take_while(|&&val| val < target)
@@ -129,23 +130,18 @@ impl BlockSearcher {
     ///
     /// If SSE2 instructions are available in the `(platform, running CPU)`,
     /// then we use a different implementation that does an exhaustive linear search over
-    /// the full block whenever the block is full (`len == 128`). It is surprisingly faster, most likely because of the lack
+    /// the block regardless of whether the block is full or not.
-    /// of branch.
+    ///
-    pub(crate) fn search_in_block(
+    /// Indeed, if the block is not full, the remaining items are TERMINATED.
-        self,
+    /// It is surprisingly faster, most likely because of the lack of branch misprediction.
-        block_docs: &AlignedBuffer,
+    pub(crate) fn search_in_block(self, block_docs: &AlignedBuffer, target: u32) -> usize {
-        len: usize,
-        start: usize,
-        target: u32,
-    ) -> usize {
         #[cfg(target_arch = "x86_64")]
         {
-            use crate::postings::compression::COMPRESSION_BLOCK_SIZE;
+            if self == BlockSearcher::SSE2 {
-            if self == BlockSearcher::SSE2 && len == COMPRESSION_BLOCK_SIZE {
                 return sse2::linear_search_sse2_128(block_docs, target);
             }
         }
-        start + galloping(&block_docs.0[start..len], target)
+        galloping(&block_docs.0[..], target)
     }
 }
 
@@ -166,6 +162,7 @@ mod tests {
     use super::exponential_search;
     use super::linear_search;
     use super::BlockSearcher;
+    use crate::docset::TERMINATED;
     use crate::postings::compression::{AlignedBuffer, COMPRESSION_BLOCK_SIZE};
 
     #[test]
@@ -196,19 +193,12 @@ mod tests {
     fn util_test_search_in_block(block_searcher: BlockSearcher, block: &[u32], target: u32) {
         let cursor = search_in_block_trivial_but_slow(block, target);
         assert!(block.len() < COMPRESSION_BLOCK_SIZE);
-        let mut output_buffer = [u32::max_value(); COMPRESSION_BLOCK_SIZE];
+        let mut output_buffer = [TERMINATED; COMPRESSION_BLOCK_SIZE];
         output_buffer[..block.len()].copy_from_slice(block);
-        for i in 0..cursor {
+        assert_eq!(
-            assert_eq!(
+            block_searcher.search_in_block(&AlignedBuffer(output_buffer), target),
-                block_searcher.search_in_block(
+            cursor
-                    &AlignedBuffer(output_buffer),
+        );
-                    block.len(),
-                    i,
-                    target
-                ),
-                cursor
-            );
-        }
     }
 
     fn util_test_search_in_block_all(block_searcher: BlockSearcher, block: &[u32]) {

src/postings/block_segment_postings.rs

@@ -0,0 +1,527 @@
+use crate::common::{BinarySerializable, VInt};
+use crate::directory::ReadOnlySource;
+use crate::fieldnorm::FieldNormReader;
+use crate::postings::compression::{
+    AlignedBuffer, BlockDecoder, VIntDecoder, COMPRESSION_BLOCK_SIZE,
+};
+use crate::postings::{BlockInfo, FreqReadingOption, SkipReader};
+use crate::query::BM25Weight;
+use crate::schema::IndexRecordOption;
+use crate::{DocId, Score, TERMINATED};
+
+fn max_score<I: Iterator<Item = Score>>(mut it: I) -> Option<Score> {
+    if let Some(first) = it.next() {
+        Some(it.fold(first, Score::max))
+    } else {
+        None
+    }
+}
+
+/// `BlockSegmentPostings` is a cursor iterating over blocks
+/// of documents.
+///
+/// # Warning
+///
+/// While it is useful for some very specific high-performance
+/// use cases, you should prefer using `SegmentPostings` for most usage.
+#[derive(Clone)]
+pub struct BlockSegmentPostings {
+    pub(crate) doc_decoder: BlockDecoder,
+    loaded_offset: usize,
+    freq_decoder: BlockDecoder,
+    freq_reading_option: FreqReadingOption,
+    block_max_score_cache: Option<Score>,
+
+    doc_freq: u32,
+
+    data: ReadOnlySource,
+    pub(crate) skip_reader: SkipReader,
+}
+
+fn decode_bitpacked_block(
+    doc_decoder: &mut BlockDecoder,
+    freq_decoder_opt: Option<&mut BlockDecoder>,
+    data: &[u8],
+    doc_offset: DocId,
+    doc_num_bits: u8,
+    tf_num_bits: u8,
+) {
+    let num_consumed_bytes = doc_decoder.uncompress_block_sorted(data, doc_offset, doc_num_bits);
+    if let Some(freq_decoder) = freq_decoder_opt {
+        freq_decoder.uncompress_block_unsorted(&data[num_consumed_bytes..], tf_num_bits);
+    }
+}
+
+fn decode_vint_block(
+    doc_decoder: &mut BlockDecoder,
+    freq_decoder_opt: Option<&mut BlockDecoder>,
+    data: &[u8],
+    doc_offset: DocId,
+    num_vint_docs: usize,
+) {
+    let num_consumed_bytes =
+        doc_decoder.uncompress_vint_sorted(data, doc_offset, num_vint_docs, TERMINATED);
+    if let Some(freq_decoder) = freq_decoder_opt {
+        freq_decoder.uncompress_vint_unsorted(
+            &data[num_consumed_bytes..],
+            num_vint_docs,
+            TERMINATED,
+        );
+    }
+}
+
+fn split_into_skips_and_postings(
+    doc_freq: u32,
+    data: ReadOnlySource,
+) -> (Option<ReadOnlySource>, ReadOnlySource) {
+    if doc_freq < COMPRESSION_BLOCK_SIZE as u32 {
+        return (None, data);
+    }
+    let mut data_byte_arr = data.as_slice();
+    let skip_len = VInt::deserialize(&mut data_byte_arr)
+        .expect("Data corrupted")
+        .0 as usize;
+    let vint_len = data.len() - data_byte_arr.len();
+    let (skip_data, postings_data) = data.slice_from(vint_len).split(skip_len);
+    (Some(skip_data), postings_data)
+}
+
+impl BlockSegmentPostings {
+    pub(crate) fn from_data(
+        doc_freq: u32,
+        data: ReadOnlySource,
+        record_option: IndexRecordOption,
+        requested_option: IndexRecordOption,
+    ) -> BlockSegmentPostings {
+        let freq_reading_option = match (record_option, requested_option) {
+            (IndexRecordOption::Basic, _) => FreqReadingOption::NoFreq,
+            (_, IndexRecordOption::Basic) => FreqReadingOption::SkipFreq,
+            (_, _) => FreqReadingOption::ReadFreq,
+        };
+
+        let (skip_data_opt, postings_data) = split_into_skips_and_postings(doc_freq, data);
+        let skip_reader = match skip_data_opt {
+            Some(skip_data) => SkipReader::new(skip_data, doc_freq, record_option),
+            None => SkipReader::new(ReadOnlySource::empty(), doc_freq, record_option),
+        };
+
+        let mut block_segment_postings = BlockSegmentPostings {
+            doc_decoder: BlockDecoder::with_val(TERMINATED),
+            loaded_offset: std::usize::MAX,
+            freq_decoder: BlockDecoder::with_val(1),
+            freq_reading_option,
+            block_max_score_cache: None,
+            doc_freq,
+            data: postings_data,
+            skip_reader,
+        };
+        block_segment_postings.load_block();
+        block_segment_postings
+    }
+
+    /// Returns the block_max_score for the current block.
+    /// It does not require the block to be loaded. For instance, it is ok to call this method
+    /// after having called `.shallow_advance(..)`.
+    ///
+    /// See `TermScorer::block_max_score(..)` for more information.
+    pub fn block_max_score(
+        &mut self,
+        fieldnorm_reader: &FieldNormReader,
+        bm25_weight: &BM25Weight,
+    ) -> Score {
+        if let Some(score) = self.block_max_score_cache {
+            return score;
+        }
+        if let Some(skip_reader_max_score) = self.skip_reader.block_max_score(bm25_weight) {
+            // if we are on a full block, the skip reader should have the block max information
+            // for us
+            self.block_max_score_cache = Some(skip_reader_max_score);
+            return skip_reader_max_score;
+        }
+        // this is the last block of the segment posting list.
+        // If it is actually loaded, we can compute block max manually.
+        if self.block_is_loaded() {
+            let docs = self.doc_decoder.output_array().iter().cloned();
+            let freqs = self.freq_decoder.output_array().iter().cloned();
+            let bm25_scores = docs.zip(freqs).map(|(doc, term_freq)| {
+                let fieldnorm_id = fieldnorm_reader.fieldnorm_id(doc);
+                bm25_weight.score(fieldnorm_id, term_freq)
+            });
+            let block_max_score = max_score(bm25_scores).unwrap_or(0.0);
+            self.block_max_score_cache = Some(block_max_score);
+            return block_max_score;
+        }
+        // We do not have access to any good block max value. We return bm25_weight.max_score()
+        // as it is a valid upperbound.
+        //
+        // We do not cache it however, so that it gets computed when once block is loaded.
+        bm25_weight.max_score()
+    }
+
+    pub(crate) fn freq_reading_option(&self) -> FreqReadingOption {
+        self.freq_reading_option
+    }
+
+    // Resets the block segment postings on another position
+    // in the postings file.
+    //
+    // This is useful for enumerating through a list of terms,
+    // and consuming the associated posting lists while avoiding
+    // reallocating a `BlockSegmentPostings`.
+    //
+    // # Warning
+    //
+    // This does not reset the positions list.
+    pub(crate) fn reset(&mut self, doc_freq: u32, postings_data: ReadOnlySource) {
+        let (skip_data_opt, postings_data) = split_into_skips_and_postings(doc_freq, postings_data);
+        self.data = ReadOnlySource::new(postings_data);
+        self.block_max_score_cache = None;
+        self.loaded_offset = std::usize::MAX;
+        if let Some(skip_data) = skip_data_opt {
+            self.skip_reader.reset(skip_data, doc_freq);
+        } else {
+            self.skip_reader.reset(ReadOnlySource::empty(), doc_freq);
+        }
+        self.doc_freq = doc_freq;
+        self.load_block();
+    }
+
+    /// Returns the overall number of documents in the block postings.
+    /// It does not take in account whether documents are deleted or not.
+    ///
+    /// This `doc_freq` is simply the sum of the length of all of the blocks
+    /// length, and it does not take in account deleted documents.
+    pub fn doc_freq(&self) -> u32 {
+        self.doc_freq
+    }
+
+    /// Returns the array of docs in the current block.
+    ///
+    /// Before the first call to `.advance()`, the block
+    /// returned by `.docs()` is empty.
+    #[inline]
+    pub fn docs(&self) -> &[DocId] {
+        debug_assert!(self.block_is_loaded());
+        self.doc_decoder.output_array()
+    }
+
+    /// Returns a full block, regardless of whetehr the block is complete or incomplete (
+    /// as it happens for the last block of the posting list).
+    ///
+    /// In the latter case, the block is guaranteed to be padded with the sentinel value:
+    /// `TERMINATED`. The array is also guaranteed to be aligned on 16 bytes = 128 bits.
+    ///
+    /// This method is useful to run SSE2 linear search.
+    #[inline(always)]
+    pub(crate) fn docs_aligned(&self) -> &AlignedBuffer {
+        debug_assert!(self.block_is_loaded());
+        self.doc_decoder.output_aligned()
+    }
+
+    /// Return the document at index `idx` of the block.
+    #[inline(always)]
+    pub fn doc(&self, idx: usize) -> u32 {
+        self.doc_decoder.output(idx)
+    }
+
+    /// Return the array of `term freq` in the block.
+    #[inline]
+    pub fn freqs(&self) -> &[u32] {
+        debug_assert!(self.block_is_loaded());
+        self.freq_decoder.output_array()
+    }
+
+    /// Return the frequency at index `idx` of the block.
+    #[inline]
+    pub fn freq(&self, idx: usize) -> u32 {
+        debug_assert!(self.block_is_loaded());
+        self.freq_decoder.output(idx)
+    }
+
+    /// Returns the length of the current block.
+    ///
+    /// All blocks have a length of `NUM_DOCS_PER_BLOCK`,
+    /// except the last block that may have a length
+    /// of any number between 1 and `NUM_DOCS_PER_BLOCK - 1`
+    #[inline]
+    pub fn block_len(&self) -> usize {
+        debug_assert!(self.block_is_loaded());
+        self.doc_decoder.output_len
+    }
+
+    /// Position on a block that may contains `target_doc`.
+    ///
+    /// If all docs are smaller than target, the block loaded may be empty,
+    /// or be the last an incomplete VInt block.
+    pub fn seek(&mut self, target_doc: DocId) {
+        self.shallow_seek(target_doc);
+        self.load_block();
+    }
+
+    pub(crate) fn position_offset(&self) -> u64 {
+        self.skip_reader.position_offset()
+    }
+
+    /// Dangerous API! This calls seek on the skip list,
+    /// but does not `.load_block()` afterwards.
+    ///
+    /// `.load_block()` needs to be called manually afterwards.
+    /// If all docs are smaller than target, the block loaded may be empty,
+    /// or be the last an incomplete VInt block.
+    pub(crate) fn shallow_seek(&mut self, target_doc: DocId) {
+        if self.skip_reader.seek(target_doc) {
+            self.block_max_score_cache = None;
+        }
+    }
+
+    pub(crate) fn block_is_loaded(&self) -> bool {
+        self.loaded_offset == self.skip_reader.byte_offset()
+    }
+
+    pub(crate) fn load_block(&mut self) {
+        let offset = self.skip_reader.byte_offset();
+        if self.loaded_offset == offset {
+            return;
+        }
+        self.loaded_offset = offset;
+        match self.skip_reader.block_info() {
+            BlockInfo::BitPacked {
+                doc_num_bits,
+                tf_num_bits,
+                ..
+            } => {
+                decode_bitpacked_block(
+                    &mut self.doc_decoder,
+                    if let FreqReadingOption::ReadFreq = self.freq_reading_option {
+                        Some(&mut self.freq_decoder)
+                    } else {
+                        None
+                    },
+                    &self.data.as_slice()[offset..],
+                    self.skip_reader.last_doc_in_previous_block,
+                    doc_num_bits,
+                    tf_num_bits,
+                );
+            }
+            BlockInfo::VInt { num_docs } => {
+                let data = {
+                    if num_docs == 0 {
+                        &[]
+                    } else {
+                        &self.data.as_slice()[offset..]
+                    }
+                };
+                decode_vint_block(
+                    &mut self.doc_decoder,
+                    if let FreqReadingOption::ReadFreq = self.freq_reading_option {
+                        Some(&mut self.freq_decoder)
+                    } else {
+                        None
+                    },
+                    data,
+                    self.skip_reader.last_doc_in_previous_block,
+                    num_docs as usize,
+                );
+            }
+        }
+    }
+
+    /// Advance to the next block.
+    ///
+    /// Returns false iff there was no remaining blocks.
+    pub fn advance(&mut self) {
+        self.skip_reader.advance();
+        self.block_max_score_cache = None;
+        self.load_block();
+    }
+
+    /// Returns an empty segment postings object
+    pub fn empty() -> BlockSegmentPostings {
+        BlockSegmentPostings {
+            doc_decoder: BlockDecoder::with_val(TERMINATED),
+            loaded_offset: 0,
+            freq_decoder: BlockDecoder::with_val(1),
+            freq_reading_option: FreqReadingOption::NoFreq,
+            block_max_score_cache: None,
+            doc_freq: 0,
+            data: ReadOnlySource::new(vec![]),
+            skip_reader: SkipReader::new(ReadOnlySource::new(vec![]), 0, IndexRecordOption::Basic),
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::BlockSegmentPostings;
+    use crate::common::HasLen;
+    use crate::core::Index;
+    use crate::docset::{DocSet, TERMINATED};
+    use crate::postings::compression::COMPRESSION_BLOCK_SIZE;
+    use crate::postings::postings::Postings;
+    use crate::postings::SegmentPostings;
+    use crate::schema::IndexRecordOption;
+    use crate::schema::Schema;
+    use crate::schema::Term;
+    use crate::schema::INDEXED;
+    use crate::DocId;
+
+    #[test]
+    fn test_empty_segment_postings() {
+        let mut postings = SegmentPostings::empty();
+        assert_eq!(postings.doc(), TERMINATED);
+        assert_eq!(postings.advance(), TERMINATED);
+        assert_eq!(postings.advance(), TERMINATED);
+        assert_eq!(postings.doc_freq(), 0);
+        assert_eq!(postings.len(), 0);
+    }
+
+    #[test]
+    fn test_empty_postings_doc_returns_terminated() {
+        let mut postings = SegmentPostings::empty();
+        assert_eq!(postings.doc(), TERMINATED);
+        assert_eq!(postings.advance(), TERMINATED);
+    }
+
+    #[test]
+    fn test_empty_postings_doc_term_freq_returns_0() {
+        let postings = SegmentPostings::empty();
+        assert_eq!(postings.term_freq(), 1);
+    }
+
+    #[test]
+    fn test_empty_block_segment_postings() {
+        let mut postings = BlockSegmentPostings::empty();
+        assert!(postings.docs().is_empty());
+        assert_eq!(postings.doc_freq(), 0);
+        postings.advance();
+        assert!(postings.docs().is_empty());
+        assert_eq!(postings.doc_freq(), 0);
+    }
+
+    #[test]
+    fn test_block_segment_postings() {
+        let mut block_segments = build_block_postings(&(0..100_000).collect::<Vec<u32>>());
+        let mut offset: u32 = 0u32;
+        // checking that the `doc_freq` is correct
+        assert_eq!(block_segments.doc_freq(), 100_000);
+        loop {
+            let block = block_segments.docs();
+            if block.is_empty() {
+                break;
+            }
+            for (i, doc) in block.iter().cloned().enumerate() {
+                assert_eq!(offset + (i as u32), doc);
+            }
+            offset += block.len() as u32;
+            block_segments.advance();
+        }
+    }
+
+    #[test]
+    fn test_skip_right_at_new_block() {
+        let mut doc_ids = (0..128).collect::<Vec<u32>>();
+        // 128 is missing
+        doc_ids.push(129);
+        doc_ids.push(130);
+        {
+            let block_segments = build_block_postings(&doc_ids);
+            let mut docset = SegmentPostings::from_block_postings(block_segments, None);
+            assert_eq!(docset.seek(128), 129);
+            assert_eq!(docset.doc(), 129);
+            assert_eq!(docset.advance(), 130);
+            assert_eq!(docset.doc(), 130);
+            assert_eq!(docset.advance(), TERMINATED);
+        }
+        {
+            let block_segments = build_block_postings(&doc_ids);
+            let mut docset = SegmentPostings::from_block_postings(block_segments, None);
+            assert_eq!(docset.seek(129), 129);
+            assert_eq!(docset.doc(), 129);
+            assert_eq!(docset.advance(), 130);
+            assert_eq!(docset.doc(), 130);
+            assert_eq!(docset.advance(), TERMINATED);
+        }
+        {
+            let block_segments = build_block_postings(&doc_ids);
+            let mut docset = SegmentPostings::from_block_postings(block_segments, None);
+            assert_eq!(docset.doc(), 0);
+            assert_eq!(docset.seek(131), TERMINATED);
+            assert_eq!(docset.doc(), TERMINATED);
+        }
+    }
+
+    fn build_block_postings(docs: &[DocId]) -> BlockSegmentPostings {
+        let mut schema_builder = Schema::builder();
+        let int_field = schema_builder.add_u64_field("id", INDEXED);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        let mut last_doc = 0u32;
+        for &doc in docs {
+            for _ in last_doc..doc {
+                index_writer.add_document(doc!(int_field=>1u64));
+            }
+            index_writer.add_document(doc!(int_field=>0u64));
+            last_doc = doc + 1;
+        }
+        index_writer.commit().unwrap();
+        let searcher = index.reader().unwrap().searcher();
+        let segment_reader = searcher.segment_reader(0);
+        let inverted_index = segment_reader.inverted_index(int_field);
+        let term = Term::from_field_u64(int_field, 0u64);
+        let term_info = inverted_index.get_term_info(&term).unwrap();
+        inverted_index.read_block_postings_from_terminfo(&term_info, IndexRecordOption::Basic)
+    }
+
+    #[test]
+    fn test_block_segment_postings_seek() {
+        let mut docs = vec![0];
+        for i in 0..1300 {
+            docs.push((i * i / 100) + i);
+        }
+        let mut block_postings = build_block_postings(&docs[..]);
+        for i in vec![0, 424, 10000] {
+            block_postings.seek(i);
+            let docs = block_postings.docs();
+            assert!(docs[0] <= i);
+            assert!(docs.last().cloned().unwrap_or(0u32) >= i);
+        }
+        block_postings.seek(100_000);
+        assert_eq!(block_postings.doc(COMPRESSION_BLOCK_SIZE - 1), TERMINATED);
+    }
+
+    #[test]
+    fn test_reset_block_segment_postings() {
+        let mut schema_builder = Schema::builder();
+        let int_field = schema_builder.add_u64_field("id", INDEXED);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        // create two postings list, one containg even number,
+        // the other containing odd numbers.
+        for i in 0..6 {
+            let doc = doc!(int_field=> (i % 2) as u64);
+            index_writer.add_document(doc);
+        }
+        index_writer.commit().unwrap();
+        let searcher = index.reader().unwrap().searcher();
+        let segment_reader = searcher.segment_reader(0);
+
+        let mut block_segments;
+        {
+            let term = Term::from_field_u64(int_field, 0u64);
+            let inverted_index = segment_reader.inverted_index(int_field);
+            let term_info = inverted_index.get_term_info(&term).unwrap();
+            block_segments = inverted_index
+                .read_block_postings_from_terminfo(&term_info, IndexRecordOption::Basic);
+        }
+        assert_eq!(block_segments.docs(), &[0, 2, 4]);
+        {
+            let term = Term::from_field_u64(int_field, 1u64);
+            let inverted_index = segment_reader.inverted_index(int_field);
+            let term_info = inverted_index.get_term_info(&term).unwrap();
+            inverted_index.reset_block_postings_from_terminfo(&term_info, &mut block_segments);
+        }
+        assert_eq!(block_segments.docs(), &[1, 3, 5]);
+    }
+}

src/postings/compression/mod.rs

@@ -17,6 +17,12 @@ pub struct BlockEncoder {
     pub output_len: usize,
 }
 
+impl Default for BlockEncoder {
+    fn default() -> Self {
+        BlockEncoder::new()
+    }
+}
+
 impl BlockEncoder {
     pub fn new() -> BlockEncoder {
         BlockEncoder {
@@ -46,19 +52,23 @@ impl BlockEncoder {
 /// We ensure that the OutputBuffer is align on 128 bits
 /// in order to run SSE2 linear search on it.
 #[repr(align(128))]
+#[derive(Clone)]
 pub(crate) struct AlignedBuffer(pub [u32; COMPRESSION_BLOCK_SIZE]);
 
+#[derive(Clone)]
 pub struct BlockDecoder {
     bitpacker: BitPacker4x,
     output: AlignedBuffer,
     pub output_len: usize,
 }
 
-impl BlockDecoder {
+impl Default for BlockDecoder {
-    pub fn new() -> BlockDecoder {
+    fn default() -> Self {
         BlockDecoder::with_val(0u32)
     }
+}
 
+impl BlockDecoder {
     pub fn with_val(val: u32) -> BlockDecoder {
         BlockDecoder {
             bitpacker: BitPacker4x::new(),
@@ -90,8 +100,8 @@ impl BlockDecoder {
     }
 
     #[inline]
-    pub(crate) fn output_aligned(&self) -> (&AlignedBuffer, usize) {
+    pub(crate) fn output_aligned(&self) -> &AlignedBuffer {
-        (&self.output, self.output_len)
+        &self.output
     }
 
     #[inline]
@@ -134,11 +144,14 @@ pub trait VIntDecoder {
     /// For instance, if delta encoded are `1, 3, 9`, and the
     /// `offset` is 5, then the output will be:
     /// `5 + 1 = 6, 6 + 3= 9, 9 + 9 = 18`
-    fn uncompress_vint_sorted<'a>(
+    ///
+    /// The value given in `padding` will be used to fill the remaining `128 - num_els` values.
+    fn uncompress_vint_sorted(
         &mut self,
-        compressed_data: &'a [u8],
+        compressed_data: &[u8],
         offset: u32,
         num_els: usize,
+        padding: u32,
     ) -> usize;
 
     /// Uncompress an array of `u32s`, compressed using variable
@@ -146,7 +159,14 @@ pub trait VIntDecoder {
     ///
     /// The method takes a number of int to decompress, and returns
     /// the amount of bytes that were read to decompress them.
-    fn uncompress_vint_unsorted<'a>(&mut self, compressed_data: &'a [u8], num_els: usize) -> usize;
+    ///
+    /// The value given in `padding` will be used to fill the remaining `128 - num_els` values.
+    fn uncompress_vint_unsorted(
+        &mut self,
+        compressed_data: &[u8],
+        num_els: usize,
+        padding: u32,
+    ) -> usize;
 }
 
 impl VIntEncoder for BlockEncoder {
@@ -160,18 +180,26 @@ impl VIntEncoder for BlockEncoder {
 }
 
 impl VIntDecoder for BlockDecoder {
-    fn uncompress_vint_sorted<'a>(
+    fn uncompress_vint_sorted(
         &mut self,
-        compressed_data: &'a [u8],
+        compressed_data: &[u8],
         offset: u32,
         num_els: usize,
+        padding: u32,
     ) -> usize {
         self.output_len = num_els;
+        self.output.0.iter_mut().for_each(|el| *el = padding);
         vint::uncompress_sorted(compressed_data, &mut self.output.0[..num_els], offset)
     }
 
-    fn uncompress_vint_unsorted<'a>(&mut self, compressed_data: &'a [u8], num_els: usize) -> usize {
+    fn uncompress_vint_unsorted(
+        &mut self,
+        compressed_data: &[u8],
+        num_els: usize,
+        padding: u32,
+    ) -> usize {
         self.output_len = num_els;
+        self.output.0.iter_mut().for_each(|el| *el = padding);
         vint::uncompress_unsorted(compressed_data, &mut self.output.0[..num_els])
     }
 }
@@ -180,13 +208,14 @@ impl VIntDecoder for BlockDecoder {
 pub mod tests {
 
     use super::*;
+    use crate::TERMINATED;
 
     #[test]
     fn test_encode_sorted_block() {
         let vals: Vec<u32> = (0u32..128u32).map(|i| i * 7).collect();
         let mut encoder = BlockEncoder::new();
         let (num_bits, compressed_data) = encoder.compress_block_sorted(&vals, 0);
-        let mut decoder = BlockDecoder::new();
+        let mut decoder = BlockDecoder::default();
         {
             let consumed_num_bytes = decoder.uncompress_block_sorted(compressed_data, 0, num_bits);
             assert_eq!(consumed_num_bytes, compressed_data.len());
@@ -199,9 +228,9 @@ pub mod tests {
     #[test]
     fn test_encode_sorted_block_with_offset() {
         let vals: Vec<u32> = (0u32..128u32).map(|i| 11 + i * 7).collect();
-        let mut encoder = BlockEncoder::new();
+        let mut encoder = BlockEncoder::default();
         let (num_bits, compressed_data) = encoder.compress_block_sorted(&vals, 10);
-        let mut decoder = BlockDecoder::new();
+        let mut decoder = BlockDecoder::default();
         {
             let consumed_num_bytes = decoder.uncompress_block_sorted(compressed_data, 10, num_bits);
             assert_eq!(consumed_num_bytes, compressed_data.len());
@@ -216,11 +245,11 @@ pub mod tests {
         let mut compressed: Vec<u8> = Vec::new();
         let n = 128;
         let vals: Vec<u32> = (0..n).map(|i| 11u32 + (i as u32) * 7u32).collect();
-        let mut encoder = BlockEncoder::new();
+        let mut encoder = BlockEncoder::default();
         let (num_bits, compressed_data) = encoder.compress_block_sorted(&vals, 10);
         compressed.extend_from_slice(compressed_data);
         compressed.push(173u8);
-        let mut decoder = BlockDecoder::new();
+        let mut decoder = BlockDecoder::default();
         {
             let consumed_num_bytes = decoder.uncompress_block_sorted(&compressed, 10, num_bits);
             assert_eq!(consumed_num_bytes, compressed.len() - 1);
@@ -236,11 +265,11 @@ pub mod tests {
         let mut compressed: Vec<u8> = Vec::new();
         let n = 128;
         let vals: Vec<u32> = (0..n).map(|i| 11u32 + (i as u32) * 7u32 % 12).collect();
-        let mut encoder = BlockEncoder::new();
+        let mut encoder = BlockEncoder::default();
         let (num_bits, compressed_data) = encoder.compress_block_unsorted(&vals);
         compressed.extend_from_slice(compressed_data);
         compressed.push(173u8);
-        let mut decoder = BlockDecoder::new();
+        let mut decoder = BlockDecoder::default();
         {
             let consumed_num_bytes = decoder.uncompress_block_unsorted(&compressed, num_bits);
             assert_eq!(consumed_num_bytes + 1, compressed.len());
@@ -251,20 +280,27 @@ pub mod tests {
         }
     }
 
+    #[test]
+    fn test_block_decoder_initialization() {
+        let block = BlockDecoder::with_val(TERMINATED);
+        assert_eq!(block.output(0), TERMINATED);
+    }
     #[test]
     fn test_encode_vint() {
-        {
+        const PADDING_VALUE: u32 = 234_234_345u32;
-            let expected_length = 154;
+        let expected_length = 154;
-            let mut encoder = BlockEncoder::new();
+        let mut encoder = BlockEncoder::new();
-            let input: Vec<u32> = (0u32..123u32).map(|i| 4 + i * 7 / 2).into_iter().collect();
+        let input: Vec<u32> = (0u32..123u32).map(|i| 4 + i * 7 / 2).into_iter().collect();
-            for offset in &[0u32, 1u32, 2u32] {
+        for offset in &[0u32, 1u32, 2u32] {
-                let encoded_data = encoder.compress_vint_sorted(&input, *offset);
+            let encoded_data = encoder.compress_vint_sorted(&input, *offset);
-                assert!(encoded_data.len() <= expected_length);
+            assert!(encoded_data.len() <= expected_length);
-                let mut decoder = BlockDecoder::new();
+            let mut decoder = BlockDecoder::default();
-                let consumed_num_bytes =
+            let consumed_num_bytes =
-                    decoder.uncompress_vint_sorted(&encoded_data, *offset, input.len());
+                decoder.uncompress_vint_sorted(&encoded_data, *offset, input.len(), PADDING_VALUE);
-                assert_eq!(consumed_num_bytes, encoded_data.len());
+            assert_eq!(consumed_num_bytes, encoded_data.len());
-                assert_eq!(input, decoder.output_array());
+            assert_eq!(input, decoder.output_array());
+            for i in input.len()..COMPRESSION_BLOCK_SIZE {
+                assert_eq!(decoder.output(i), PADDING_VALUE);
             }
         }
     }

src/postings/compression/vint.rs

@@ -42,7 +42,7 @@ pub(crate) fn compress_unsorted<'a>(input: &[u32], output: &'a mut [u8]) -> &'a
 }
 
 #[inline(always)]
-pub fn uncompress_sorted<'a>(compressed_data: &'a [u8], output: &mut [u32], offset: u32) -> usize {
+pub fn uncompress_sorted(compressed_data: &[u8], output: &mut [u32], offset: u32) -> usize {
     let mut read_byte = 0;
     let mut result = offset;
     for output_mut in output.iter_mut() {

src/postings/mod.rs

@@ -3,11 +3,8 @@ Postings module (also called inverted index)
 */
 
 mod block_search;
+mod block_segment_postings;
 pub(crate) mod compression;
-/// Postings module
-///
-/// Postings, also called inverted lists, is the key datastructure
-/// to full-text search.
 mod postings;
 mod postings_writer;
 mod recorder;
@@ -22,18 +19,17 @@ pub(crate) use self::block_search::BlockSearcher;
 pub(crate) use self::postings_writer::MultiFieldPostingsWriter;
 pub use self::serializer::{FieldSerializer, InvertedIndexSerializer};
 
-use self::compression::COMPRESSION_BLOCK_SIZE;
 pub use self::postings::Postings;
-pub(crate) use self::skip::SkipReader;
+pub(crate) use self::skip::{BlockInfo, SkipReader};
 pub use self::term_info::TermInfo;
 
-pub use self::segment_postings::{BlockSegmentPostings, SegmentPostings};
+pub use self::block_segment_postings::BlockSegmentPostings;
+pub use self::segment_postings::SegmentPostings;
 
 pub(crate) use self::stacker::compute_table_size;
 
 pub use crate::common::HasLen;
 
-pub(crate) const USE_SKIP_INFO_LIMIT: u32 = COMPRESSION_BLOCK_SIZE as u32;
 pub(crate) type UnorderedTermId = u64;
 
 #[cfg_attr(feature = "cargo-clippy", allow(clippy::enum_variant_names))]
@@ -51,7 +47,7 @@ pub mod tests {
     use crate::core::Index;
     use crate::core::SegmentComponent;
     use crate::core::SegmentReader;
-    use crate::docset::{DocSet, SkipResult};
+    use crate::docset::{DocSet, TERMINATED};
     use crate::fieldnorm::FieldNormReader;
     use crate::indexer::operation::AddOperation;
     use crate::indexer::SegmentWriter;
@@ -69,45 +65,42 @@ pub mod tests {
     use std::iter;
 
     #[test]
-    pub fn test_position_write() {
+    pub fn test_position_write() -> crate::Result<()> {
         let mut schema_builder = Schema::builder();
         let text_field = schema_builder.add_text_field("text", TEXT);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
         let mut segment = index.new_segment();
-        let mut posting_serializer = InvertedIndexSerializer::open(&mut segment).unwrap();
+        let mut posting_serializer = InvertedIndexSerializer::open(&mut segment)?;
-        {
+        let mut field_serializer = posting_serializer.new_field(text_field, 120 * 4, None)?;
-            let mut field_serializer = posting_serializer.new_field(text_field, 120 * 4).unwrap();
+        field_serializer.new_term("abc".as_bytes(), 12u32)?;
-            field_serializer.new_term("abc".as_bytes()).unwrap();
+        for doc_id in 0u32..120u32 {
-            for doc_id in 0u32..120u32 {
+            let delta_positions = vec![1, 2, 3, 2];
-                let delta_positions = vec![1, 2, 3, 2];
+            field_serializer.write_doc(doc_id, 4, &delta_positions)?;
-                field_serializer
-                    .write_doc(doc_id, 4, &delta_positions)
-                    .unwrap();
-            }
-            field_serializer.close_term().unwrap();
         }
-        posting_serializer.close().unwrap();
+        field_serializer.close_term()?;
-        let read = segment.open_read(SegmentComponent::POSITIONS).unwrap();
+        posting_serializer.close()?;
+        let read = segment.open_read(SegmentComponent::POSITIONS)?;
         assert!(read.len() <= 140);
+        Ok(())
     }
 
     #[test]
-    pub fn test_skip_positions() {
+    pub fn test_skip_positions() -> crate::Result<()> {
         let mut schema_builder = Schema::builder();
         let title = schema_builder.add_text_field("title", TEXT);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_with_num_threads(1, 30_000_000).unwrap();
+        let mut index_writer = index.writer_with_num_threads(1, 30_000_000)?;
         index_writer.add_document(doc!(title => r#"abc abc abc"#));
         index_writer.add_document(doc!(title => r#"abc be be be be abc"#));
         for _ in 0..1_000 {
             index_writer.add_document(doc!(title => r#"abc abc abc"#));
         }
         index_writer.add_document(doc!(title => r#"abc be be be be abc"#));
-        index_writer.commit().unwrap();
+        index_writer.commit()?;
 
-        let searcher = index.reader().unwrap().searcher();
+        let searcher = index.reader()?.searcher();
         let inverted_index = searcher.segment_reader(0u32).inverted_index(title);
         let term = Term::from_field_text(title, "abc");
         let mut positions = Vec::new();
@@ -115,29 +108,12 @@ pub mod tests {
             let mut postings = inverted_index
                 .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
                 .unwrap();
-            postings.advance();
+            assert_eq!(postings.doc(), 0);
             postings.positions(&mut positions);
             assert_eq!(&[0, 1, 2], &positions[..]);
             postings.positions(&mut positions);
             assert_eq!(&[0, 1, 2], &positions[..]);
-            postings.advance();
+            assert_eq!(postings.advance(), 1);
-            postings.positions(&mut positions);
-            assert_eq!(&[0, 5], &positions[..]);
-        }
-        {
-            let mut postings = inverted_index
-                .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
-                .unwrap();
-            postings.advance();
-            postings.advance();
-            postings.positions(&mut positions);
-            assert_eq!(&[0, 5], &positions[..]);
-        }
-        {
-            let mut postings = inverted_index
-                .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
-                .unwrap();
-            assert_eq!(postings.skip_next(1), SkipResult::Reached);
             assert_eq!(postings.doc(), 1);
             postings.positions(&mut positions);
             assert_eq!(&[0, 5], &positions[..]);
@@ -146,7 +122,25 @@ pub mod tests {
             let mut postings = inverted_index
                 .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
                 .unwrap();
-            assert_eq!(postings.skip_next(1002), SkipResult::Reached);
+            assert_eq!(postings.doc(), 0);
+            assert_eq!(postings.advance(), 1);
+            postings.positions(&mut positions);
+            assert_eq!(&[0, 5], &positions[..]);
+        }
+        {
+            let mut postings = inverted_index
+                .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
+                .unwrap();
+            assert_eq!(postings.seek(1), 1);
+            assert_eq!(postings.doc(), 1);
+            postings.positions(&mut positions);
+            assert_eq!(&[0, 5], &positions[..]);
+        }
+        {
+            let mut postings = inverted_index
+                .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
+                .unwrap();
+            assert_eq!(postings.seek(1002), 1002);
             assert_eq!(postings.doc(), 1002);
             postings.positions(&mut positions);
             assert_eq!(&[0, 5], &positions[..]);
@@ -155,12 +149,13 @@ pub mod tests {
             let mut postings = inverted_index
                 .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
                 .unwrap();
-            assert_eq!(postings.skip_next(100), SkipResult::Reached);
+            assert_eq!(postings.seek(100), 100);
-            assert_eq!(postings.skip_next(1002), SkipResult::Reached);
+            assert_eq!(postings.seek(1002), 1002);
             assert_eq!(postings.doc(), 1002);
             postings.positions(&mut positions);
             assert_eq!(&[0, 5], &positions[..]);
         }
+        Ok(())
     }
 
     #[test]
@@ -281,22 +276,21 @@ pub mod tests {
                     .read_postings(&term_a, IndexRecordOption::WithFreqsAndPositions)
                     .unwrap();
                 assert_eq!(postings_a.len(), 1000);
-                assert!(postings_a.advance());
                 assert_eq!(postings_a.doc(), 0);
                 assert_eq!(postings_a.term_freq(), 6);
                 postings_a.positions(&mut positions);
                 assert_eq!(&positions[..], [0, 2, 4, 6, 7, 13]);
-                assert!(postings_a.advance());
+                assert_eq!(postings_a.advance(), 1u32);
                 assert_eq!(postings_a.doc(), 1u32);
                 assert_eq!(postings_a.term_freq(), 1);
                 for i in 2u32..1000u32 {
-                    assert!(postings_a.advance());
+                    assert_eq!(postings_a.advance(), i);
                     assert_eq!(postings_a.term_freq(), 1);
                     postings_a.positions(&mut positions);
                     assert_eq!(&positions[..], [i]);
                     assert_eq!(postings_a.doc(), i);
                 }
-                assert!(!postings_a.advance());
+                assert_eq!(postings_a.advance(), TERMINATED);
             }
             {
                 let term_e = Term::from_field_text(text_field, "e");
@@ -306,7 +300,6 @@ pub mod tests {
                     .unwrap();
                 assert_eq!(postings_e.len(), 1000 - 2);
                 for i in 2u32..1000u32 {
-                    assert!(postings_e.advance());
                     assert_eq!(postings_e.term_freq(), i);
                     postings_e.positions(&mut positions);
                     assert_eq!(positions.len(), i as usize);
@@ -314,8 +307,9 @@ pub mod tests {
                         assert_eq!(positions[j], (j as u32));
                     }
                     assert_eq!(postings_e.doc(), i);
+                    postings_e.advance();
                 }
-                assert!(!postings_e.advance());
+                assert_eq!(postings_e.doc(), TERMINATED);
             }
         }
     }
@@ -329,16 +323,8 @@ pub mod tests {
         let index = Index::create_in_ram(schema);
         {
             let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-            {
+            index_writer.add_document(doc!(text_field => "g b b d c g c"));
-                let mut doc = Document::default();
+            index_writer.add_document(doc!(text_field => "g a b b a d c g c"));
-                doc.add_text(text_field, "g b b d c g c");
-                index_writer.add_document(doc);
-            }
-            {
-                let mut doc = Document::default();
-                doc.add_text(text_field, "g a b b a d c g c");
-                index_writer.add_document(doc);
-            }
             assert!(index_writer.commit().is_ok());
         }
         let term_a = Term::from_field_text(text_field, "a");
@@ -348,7 +334,6 @@ pub mod tests {
             .inverted_index(text_field)
             .read_postings(&term_a, IndexRecordOption::WithFreqsAndPositions)
             .unwrap();
-        assert!(postings.advance());
         assert_eq!(postings.doc(), 1u32);
         postings.positions(&mut positions);
         assert_eq!(&positions[..], &[1u32, 4]);
@@ -370,11 +355,8 @@ pub mod tests {
             let index = Index::create_in_ram(schema);
             {
                 let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-                for i in 0..num_docs {
+                for i in 0u64..num_docs as u64 {
-                    let mut doc = Document::default();
+                    let doc = doc!(value_field => 2u64, value_field => i % 2u64);
-                    doc.add_u64(value_field, 2);
-                    doc.add_u64(value_field, (i % 2) as u64);
-
                     index_writer.add_document(doc);
                 }
                 assert!(index_writer.commit().is_ok());
@@ -391,11 +373,10 @@ pub mod tests {
                     .inverted_index(term_2.field())
                     .read_postings(&term_2, IndexRecordOption::Basic)
                     .unwrap();
-
+                assert_eq!(segment_postings.seek(i), i);
-                assert_eq!(segment_postings.skip_next(i), SkipResult::Reached);
                 assert_eq!(segment_postings.doc(), i);
 
-                assert_eq!(segment_postings.skip_next(j), SkipResult::Reached);
+                assert_eq!(segment_postings.seek(j), j);
                 assert_eq!(segment_postings.doc(), j);
             }
         }
@@ -407,17 +388,16 @@ pub mod tests {
                 .unwrap();
 
             // check that `skip_next` advances the iterator
-            assert!(segment_postings.advance());
             assert_eq!(segment_postings.doc(), 0);
 
-            assert_eq!(segment_postings.skip_next(1), SkipResult::Reached);
+            assert_eq!(segment_postings.seek(1), 1);
             assert_eq!(segment_postings.doc(), 1);
 
-            assert_eq!(segment_postings.skip_next(1), SkipResult::OverStep);
+            assert_eq!(segment_postings.seek(1), 1);
-            assert_eq!(segment_postings.doc(), 2);
+            assert_eq!(segment_postings.doc(), 1);
 
             // check that going beyond the end is handled
-            assert_eq!(segment_postings.skip_next(num_docs), SkipResult::End);
+            assert_eq!(segment_postings.seek(num_docs), TERMINATED);
         }
 
         // check that filtering works
@@ -428,7 +408,7 @@ pub mod tests {
                 .unwrap();
 
             for i in 0..num_docs / 2 {
-                assert_eq!(segment_postings.skip_next(i * 2), SkipResult::Reached);
+                assert_eq!(segment_postings.seek(i * 2), i * 2);
                 assert_eq!(segment_postings.doc(), i * 2);
             }
 
@@ -438,7 +418,7 @@ pub mod tests {
                 .unwrap();
 
             for i in 0..num_docs / 2 - 1 {
-                assert_eq!(segment_postings.skip_next(i * 2 + 1), SkipResult::OverStep);
+                assert!(segment_postings.seek(i * 2 + 1) > (i * 1) * 2);
                 assert_eq!(segment_postings.doc(), (i + 1) * 2);
             }
         }
@@ -450,6 +430,7 @@ pub mod tests {
             assert!(index_writer.commit().is_ok());
         }
         let searcher = index.reader().unwrap().searcher();
+        assert_eq!(searcher.segment_readers().len(), 1);
         let segment_reader = searcher.segment_reader(0);
 
         // make sure seeking still works
@@ -460,11 +441,11 @@ pub mod tests {
                 .unwrap();
 
             if i % 2 == 0 {
-                assert_eq!(segment_postings.skip_next(i), SkipResult::Reached);
+                assert_eq!(segment_postings.seek(i), i);
                 assert_eq!(segment_postings.doc(), i);
                 assert!(segment_reader.is_deleted(i));
             } else {
-                assert_eq!(segment_postings.skip_next(i), SkipResult::Reached);
+                assert_eq!(segment_postings.seek(i), i);
                 assert_eq!(segment_postings.doc(), i);
             }
         }
@@ -479,12 +460,16 @@ pub mod tests {
             let mut last = 2; // start from 5 to avoid seeking to 3 twice
             let mut cur = 3;
             loop {
-                match segment_postings.skip_next(cur) {
+                let seek = segment_postings.seek(cur);
-                    SkipResult::End => break,
+                if seek == TERMINATED {
-                    SkipResult::Reached => assert_eq!(segment_postings.doc(), cur),
+                    break;
-                    SkipResult::OverStep => assert_eq!(segment_postings.doc(), cur + 1),
+                }
+                assert_eq!(seek, segment_postings.doc());
+                if seek == cur {
+                    assert_eq!(segment_postings.doc(), cur);
+                } else {
+                    assert_eq!(segment_postings.doc(), cur + 1);
                 }
-
                 let next = cur + last;
                 last = cur;
                 cur = next;
@@ -570,7 +555,7 @@ pub mod tests {
     }
 
     impl<TDocSet: DocSet> DocSet for UnoptimizedDocSet<TDocSet> {
-        fn advance(&mut self) -> bool {
+        fn advance(&mut self) -> DocId {
             self.0.advance()
         }
 
@@ -595,31 +580,26 @@ pub mod tests {
     ) {
         for target in targets {
             let mut postings_opt = postings_factory();
+            if target < postings_opt.doc() {
+                continue;
+            }
             let mut postings_unopt = UnoptimizedDocSet::wrap(postings_factory());
-            let skip_result_opt = postings_opt.skip_next(target);
+            let skip_result_opt = postings_opt.seek(target);
-            let skip_result_unopt = postings_unopt.skip_next(target);
+            let skip_result_unopt = postings_unopt.seek(target);
             assert_eq!(
                 skip_result_unopt, skip_result_opt,
                 "Failed while skipping to {}",
                 target
             );
-            match skip_result_opt {
+            assert!(skip_result_opt >= target);
-                SkipResult::Reached => assert_eq!(postings_opt.doc(), target),
+            assert_eq!(skip_result_opt, postings_opt.doc());
-                SkipResult::OverStep => assert!(postings_opt.doc() > target),
+            if skip_result_opt == TERMINATED {
-                SkipResult::End => {
+                return;
-                    return;
-                }
             }
-            while postings_opt.advance() {
+            while postings_opt.doc() != TERMINATED {
-                assert!(postings_unopt.advance());
+                assert_eq!(postings_opt.doc(), postings_unopt.doc());
-                assert_eq!(
+                assert_eq!(postings_opt.advance(), postings_unopt.advance());
-                    postings_opt.doc(),
-                    postings_unopt.doc(),
-                    "Failed while skipping to {}",
-                    target
-                );
             }
-            assert!(!postings_unopt.advance());
         }
     }
 }
@@ -628,7 +608,7 @@ pub mod tests {
 mod bench {
 
     use super::tests::*;
-    use crate::docset::SkipResult;
+    use crate::docset::TERMINATED;
     use crate::query::Intersection;
     use crate::schema::IndexRecordOption;
     use crate::tests;
@@ -646,7 +626,7 @@ mod bench {
                 .inverted_index(TERM_A.field())
                 .read_postings(&*TERM_A, IndexRecordOption::Basic)
                 .unwrap();
-            while segment_postings.advance() {}
+            while segment_postings.advance() != TERMINATED {}
         });
     }
 
@@ -678,7 +658,7 @@ mod bench {
                 segment_postings_c,
                 segment_postings_d,
             ]);
-            while intersection.advance() {}
+            while intersection.advance() != TERMINATED {}
         });
     }
 
@@ -694,11 +674,10 @@ mod bench {
             .unwrap();
 
         let mut existing_docs = Vec::new();
-        segment_postings.advance();
         for doc in &docs {
             if *doc >= segment_postings.doc() {
                 existing_docs.push(*doc);
-                if segment_postings.skip_next(*doc) == SkipResult::End {
+                if segment_postings.seek(*doc) == TERMINATED {
                     break;
                 }
             }
@@ -710,7 +689,7 @@ mod bench {
                 .read_postings(&*TERM_A, IndexRecordOption::Basic)
                 .unwrap();
             for doc in &existing_docs {
-                if segment_postings.skip_next(*doc) == SkipResult::End {
+                if segment_postings.seek(*doc) == TERMINATED {
                     break;
                 }
             }
@@ -749,8 +728,9 @@ mod bench {
                 .read_postings(&*TERM_A, IndexRecordOption::Basic)
                 .unwrap();
             let mut s = 0u32;
-            while segment_postings.advance() {
+            while segment_postings.doc() != TERMINATED {
                 s += (segment_postings.doc() & n) % 1024;
+                segment_postings.advance()
             }
             s
         });

src/postings/postings_writer.rs

@@ -1,5 +1,6 @@
 use super::stacker::{Addr, MemoryArena, TermHashMap};
 
+use crate::fieldnorm::FieldNormReaders;
 use crate::postings::recorder::{
     BufferLender, NothingRecorder, Recorder, TFAndPositionRecorder, TermFrequencyRecorder,
 };
@@ -11,7 +12,6 @@ use crate::termdict::TermOrdinal;
 use crate::tokenizer::TokenStream;
 use crate::tokenizer::{Token, MAX_TOKEN_LEN};
 use crate::DocId;
-use crate::Result;
 use fnv::FnvHashMap;
 use std::collections::HashMap;
 use std::io;
@@ -129,7 +129,8 @@ impl MultiFieldPostingsWriter {
     pub fn serialize(
         &self,
         serializer: &mut InvertedIndexSerializer,
-    ) -> Result<HashMap<Field, FnvHashMap<UnorderedTermId, TermOrdinal>>> {
+        fieldnorm_readers: FieldNormReaders,
+    ) -> crate::Result<HashMap<Field, FnvHashMap<UnorderedTermId, TermOrdinal>>> {
         let mut term_offsets: Vec<(&[u8], Addr, UnorderedTermId)> =
             self.term_index.iter().collect();
         term_offsets.sort_unstable_by_key(|&(k, _, _)| k);
@@ -162,8 +163,12 @@ impl MultiFieldPostingsWriter {
             }
 
             let postings_writer = &self.per_field_postings_writers[field.field_id() as usize];
-            let mut field_serializer =
+            let fieldnorm_reader = fieldnorm_readers.get_field(field);
-                serializer.new_field(field, postings_writer.total_num_tokens())?;
+            let mut field_serializer = serializer.new_field(
+                field,
+                postings_writer.total_num_tokens(),
+                fieldnorm_reader,
+            )?;
             postings_writer.serialize(
                 &term_offsets[start..stop],
                 &mut field_serializer,
@@ -298,7 +303,8 @@ impl<Rec: Recorder + 'static> PostingsWriter for SpecializedPostingsWriter<Rec>
         let mut buffer_lender = BufferLender::default();
         for &(term_bytes, addr, _) in term_addrs {
             let recorder: Rec = termdict_heap.read(addr);
-            serializer.new_term(&term_bytes[4..])?;
+            let term_doc_freq = recorder.term_doc_freq().unwrap_or(0u32);
+            serializer.new_term(&term_bytes[4..], term_doc_freq)?;
             recorder.serialize(&mut buffer_lender, serializer, heap)?;
             serializer.close_term()?;
         }

src/postings/recorder.rs

@@ -75,6 +75,10 @@ pub(crate) trait Recorder: Copy + 'static {
         serializer: &mut FieldSerializer<'_>,
         heap: &MemoryArena,
     ) -> io::Result<()>;
+    /// Returns the number of document containing this term.
+    ///
+    /// Returns `None` if not available.
+    fn term_doc_freq(&self) -> Option<u32>;
 }
 
 /// Only records the doc ids
@@ -113,11 +117,16 @@ impl Recorder for NothingRecorder {
     ) -> io::Result<()> {
         let buffer = buffer_lender.lend_u8();
         self.stack.read_to_end(heap, buffer);
+        // TODO avoid reading twice.
         for doc in VInt32Reader::new(&buffer[..]) {
             serializer.write_doc(doc as u32, 0u32, &[][..])?;
         }
         Ok(())
     }
+
+    fn term_doc_freq(&self) -> Option<u32> {
+        None
+    }
 }
 
 /// Recorder encoding document ids, and term frequencies
@@ -126,6 +135,7 @@ pub struct TermFrequencyRecorder {
     stack: ExpUnrolledLinkedList,
     current_doc: DocId,
     current_tf: u32,
+    term_doc_freq: u32,
 }
 
 impl Recorder for TermFrequencyRecorder {
@@ -134,6 +144,7 @@ impl Recorder for TermFrequencyRecorder {
             stack: ExpUnrolledLinkedList::new(),
             current_doc: u32::max_value(),
             current_tf: 0u32,
+            term_doc_freq: 0u32,
         }
     }
 
@@ -142,6 +153,7 @@ impl Recorder for TermFrequencyRecorder {
     }
 
     fn new_doc(&mut self, doc: DocId, heap: &mut MemoryArena) {
+        self.term_doc_freq += 1;
         self.current_doc = doc;
         let _ = write_u32_vint(doc, &mut self.stack.writer(heap));
     }
@@ -172,6 +184,10 @@ impl Recorder for TermFrequencyRecorder {
 
         Ok(())
     }
+
+    fn term_doc_freq(&self) -> Option<u32> {
+        Some(self.term_doc_freq)
+    }
 }
 
 /// Recorder encoding term frequencies as well as positions.
@@ -179,12 +195,14 @@ impl Recorder for TermFrequencyRecorder {
 pub struct TFAndPositionRecorder {
     stack: ExpUnrolledLinkedList,
     current_doc: DocId,
+    term_doc_freq: u32,
 }
 impl Recorder for TFAndPositionRecorder {
     fn new() -> Self {
         TFAndPositionRecorder {
             stack: ExpUnrolledLinkedList::new(),
             current_doc: u32::max_value(),
+            term_doc_freq: 0u32,
         }
     }
 
@@ -194,6 +212,7 @@ impl Recorder for TFAndPositionRecorder {
 
     fn new_doc(&mut self, doc: DocId, heap: &mut MemoryArena) {
         self.current_doc = doc;
+        self.term_doc_freq += 1u32;
         let _ = write_u32_vint(doc, &mut self.stack.writer(heap));
     }
 
@@ -233,6 +252,10 @@ impl Recorder for TFAndPositionRecorder {
         }
         Ok(())
     }
+
+    fn term_doc_freq(&self) -> Option<u32> {
+        Some(self.term_doc_freq)
+    }
 }
 
 #[cfg(test)]

src/postings/segment_postings.rs

@@ -1,81 +1,70 @@
-use crate::common::BitSet;
 use crate::common::HasLen;
-use crate::common::{BinarySerializable, VInt};
+
-use crate::docset::{DocSet, SkipResult};
+use crate::docset::DocSet;
 use crate::positions::PositionReader;
-use crate::postings::compression::{compressed_block_size, AlignedBuffer};
+
-use crate::postings::compression::{BlockDecoder, VIntDecoder, COMPRESSION_BLOCK_SIZE};
+use crate::postings::compression::COMPRESSION_BLOCK_SIZE;
 use crate::postings::serializer::PostingsSerializer;
 use crate::postings::BlockSearcher;
-use crate::postings::FreqReadingOption;
+
 use crate::postings::Postings;
-use crate::postings::SkipReader;
+
-use crate::postings::USE_SKIP_INFO_LIMIT;
 use crate::schema::IndexRecordOption;
-use crate::DocId;
+use crate::{DocId, TERMINATED};
-use owned_read::OwnedRead;
-use std::cmp::Ordering;
-use tantivy_fst::Streamer;
 
-struct PositionComputer {
+use crate::directory::ReadOnlySource;
-    // store the amount of position int
+use crate::fastfield::DeleteBitSet;
-    // before reading positions.
+use crate::postings::BlockSegmentPostings;
-    //
-    // if none, position are already loaded in
-    // the positions vec.
-    position_to_skip: usize,
-    position_reader: PositionReader,
-}
-
-impl PositionComputer {
-    pub fn new(position_reader: PositionReader) -> PositionComputer {
-        PositionComputer {
-            position_to_skip: 0,
-            position_reader,
-        }
-    }
-
-    pub fn add_skip(&mut self, num_skip: usize) {
-        self.position_to_skip += num_skip;
-    }
-
-    // Positions can only be read once.
-    pub fn positions_with_offset(&mut self, offset: u32, output: &mut [u32]) {
-        self.position_reader.skip(self.position_to_skip);
-        self.position_to_skip = 0;
-        self.position_reader.read(output);
-        let mut cum = offset;
-        for output_mut in output.iter_mut() {
-            cum += *output_mut;
-            *output_mut = cum;
-        }
-    }
-}
 
 /// `SegmentPostings` represents the inverted list or postings associated to
 /// a term in a `Segment`.
 ///
 /// As we iterate through the `SegmentPostings`, the frequencies are optionally decoded.
 /// Positions on the other hand, are optionally entirely decoded upfront.
+#[derive(Clone)]
 pub struct SegmentPostings {
-    block_cursor: BlockSegmentPostings,
+    pub(crate) block_cursor: BlockSegmentPostings,
     cur: usize,
-    position_computer: Option<PositionComputer>,
+    position_reader: Option<PositionReader>,
     block_searcher: BlockSearcher,
 }
 
 impl SegmentPostings {
     /// Returns an empty segment postings object
     pub fn empty() -> Self {
-        let empty_block_cursor = BlockSegmentPostings::empty();
         SegmentPostings {
-            block_cursor: empty_block_cursor,
+            block_cursor: BlockSegmentPostings::empty(),
-            cur: COMPRESSION_BLOCK_SIZE,
+            cur: 0,
-            position_computer: None,
+            position_reader: None,
             block_searcher: BlockSearcher::default(),
         }
     }
 
+    /// Compute the number of non-deleted documents.
+    ///
+    /// This method will clone and scan through the posting lists.
+    /// (this is a rather expensive operation).
+    pub fn doc_freq_given_deletes(&self, delete_bitset: &DeleteBitSet) -> u32 {
+        let mut docset = self.clone();
+        let mut doc_freq = 0;
+        loop {
+            let doc = docset.doc();
+            if doc == TERMINATED {
+                return doc_freq;
+            }
+            if delete_bitset.is_alive(doc) {
+                doc_freq += 1u32;
+            }
+            docset.advance();
+        }
+    }
+
+    /// Returns the overall number of documents in the block postings.
+    /// It does not take in account whether documents are deleted or not.
+    pub fn doc_freq(&self) -> u32 {
+        self.block_cursor.doc_freq()
+    }
+
     /// Creates a segment postings object with the given documents
     /// and no frequency encoded.
     ///
@@ -87,7 +76,9 @@ impl SegmentPostings {
     pub fn create_from_docs(docs: &[u32]) -> SegmentPostings {
         let mut buffer = Vec::new();
         {
-            let mut postings_serializer = PostingsSerializer::new(&mut buffer, false, false);
+            let mut postings_serializer =
+                PostingsSerializer::new(&mut buffer, 0.0, false, false, None);
+            postings_serializer.new_term(docs.len() as u32);
             for &doc in docs {
                 postings_serializer.write_doc(doc, 1u32);
             }
@@ -97,15 +88,58 @@ impl SegmentPostings {
         }
         let block_segment_postings = BlockSegmentPostings::from_data(
             docs.len() as u32,
-            OwnedRead::new(buffer),
+            ReadOnlySource::from(buffer),
             IndexRecordOption::Basic,
             IndexRecordOption::Basic,
         );
         SegmentPostings::from_block_postings(block_segment_postings, None)
     }
-}
 
-impl SegmentPostings {
+    /// Helper functions to create `SegmentPostings` for tests.
+    #[cfg(test)]
+    pub fn create_from_docs_and_tfs(
+        doc_and_tfs: &[(u32, u32)],
+        fieldnorms: Option<&[u32]>,
+    ) -> SegmentPostings {
+        use crate::fieldnorm::FieldNormReader;
+        use crate::Score;
+        let mut buffer: Vec<u8> = Vec::new();
+        let fieldnorm_reader = fieldnorms.map(FieldNormReader::for_test);
+        let average_field_norm = fieldnorms
+            .map(|fieldnorms| {
+                if fieldnorms.len() == 0 {
+                    return 0.0;
+                }
+                let total_num_tokens: u64 = fieldnorms
+                    .iter()
+                    .map(|&fieldnorm| fieldnorm as u64)
+                    .sum::<u64>();
+                total_num_tokens as Score / fieldnorms.len() as f32
+            })
+            .unwrap_or(0.0);
+        let mut postings_serializer = PostingsSerializer::new(
+            &mut buffer,
+            average_field_norm,
+            true,
+            false,
+            fieldnorm_reader,
+        );
+        postings_serializer.new_term(doc_and_tfs.len() as u32);
+        for &(doc, tf) in doc_and_tfs {
+            postings_serializer.write_doc(doc, tf);
+        }
+        postings_serializer
+            .close_term(doc_and_tfs.len() as u32)
+            .unwrap();
+        let block_segment_postings = BlockSegmentPostings::from_data(
+            doc_and_tfs.len() as u32,
+            ReadOnlySource::from(buffer),
+            IndexRecordOption::WithFreqs,
+            IndexRecordOption::WithFreqs,
+        );
+        SegmentPostings::from_block_postings(block_segment_postings, None)
+    }
+
     /// Reads a Segment postings from an &[u8]
     ///
     /// * `len` - number of document in the posting lists.
@@ -114,12 +148,12 @@ impl SegmentPostings {
     ///   frequencies and/or positions
     pub(crate) fn from_block_postings(
         segment_block_postings: BlockSegmentPostings,
-        positions_stream_opt: Option<PositionReader>,
+        position_reader: Option<PositionReader>,
     ) -> SegmentPostings {
         SegmentPostings {
             block_cursor: segment_block_postings,
-            cur: COMPRESSION_BLOCK_SIZE, // cursor within the block
+            cur: 0, // cursor within the block
-            position_computer: positions_stream_opt.map(PositionComputer::new),
+            position_reader,
             block_searcher: BlockSearcher::default(),
         }
     }
@@ -129,139 +163,60 @@ impl DocSet for SegmentPostings {
     // goes to the next element.
     // next needs to be called a first time to point to the correct element.
     #[inline]
-    fn advance(&mut self) -> bool {
+    fn advance(&mut self) -> DocId {
-        if self.position_computer.is_some() && self.cur < COMPRESSION_BLOCK_SIZE {
+        debug_assert!(self.block_cursor.block_is_loaded());
-            let term_freq = self.term_freq() as usize;
+        if self.cur == COMPRESSION_BLOCK_SIZE - 1 {
-            if let Some(position_computer) = self.position_computer.as_mut() {
-                position_computer.add_skip(term_freq);
-            }
-        }
-        self.cur += 1;
-        if self.cur >= self.block_cursor.block_len() {
             self.cur = 0;
-            if !self.block_cursor.advance() {
+            self.block_cursor.advance();
-                self.cur = COMPRESSION_BLOCK_SIZE;
+        } else {
-                return false;
+            self.cur += 1;
-            }
         }
-        true
+        self.doc()
     }
 
-    fn skip_next(&mut self, target: DocId) -> SkipResult {
+    fn seek(&mut self, target: DocId) -> DocId {
-        if !self.advance() {
+        debug_assert!(self.doc() <= target);
-            return SkipResult::End;
+        if self.doc() >= target {
-        }
+            return self.doc();
-        match self.doc().cmp(&target) {
-            Ordering::Equal => {
-                return SkipResult::Reached;
-            }
-            Ordering::Greater => {
-                return SkipResult::OverStep;
-            }
-            _ => {
-                // ...
-            }
         }
 
-        // In the following, thanks to the call to advance above,
+        self.block_cursor.seek(target);
-        // we know that the position is not loaded and we need
-        // to skip every doc_freq we cross.
 
-        // skip blocks until one that might contain the target
+        // At this point we are on the block, that might contain our document.
-        // check if we need to go to the next block
+        let output = self.block_cursor.docs_aligned();
-        let mut sum_freqs_skipped: u32 = 0;
+        self.cur = self.block_searcher.search_in_block(&output, target);
-        if !self
-            .block_cursor
-            .docs()
-            .last()
-            .map(|doc| *doc >= target)
-            .unwrap_or(false)
-        // there should always be at least a document in the block
-        // since advance returned.
-        {
-            // we are not in the right block.
-            //
-            // First compute all of the freqs skipped from the current block.
-            if self.position_computer.is_some() {
-                sum_freqs_skipped = self.block_cursor.freqs()[self.cur..].iter().sum();
-                match self.block_cursor.skip_to(target) {
-                    BlockSegmentPostingsSkipResult::Success(block_skip_freqs) => {
-                        sum_freqs_skipped += block_skip_freqs;
-                    }
-                    BlockSegmentPostingsSkipResult::Terminated => {
-                        return SkipResult::End;
-                    }
-                }
-            } else if self.block_cursor.skip_to(target)
-                == BlockSegmentPostingsSkipResult::Terminated
-            {
-                // no positions needed. no need to sum freqs.
-                return SkipResult::End;
-            }
-            self.cur = 0;
-        }
 
-        let cur = self.cur;
+        // The last block is not full and padded with the value TERMINATED,
-
+        // so that we are guaranteed to have at least doc in the block (a real one or the padding)
-        // we're in the right block now, start with an exponential search
+        // that is greater or equal to the target.
-        let (output, len) = self.block_cursor.docs_aligned();
+        debug_assert!(self.cur < COMPRESSION_BLOCK_SIZE);
-        let new_cur = self
-            .block_searcher
-            .search_in_block(&output, len, cur, target);
-        if let Some(position_computer) = self.position_computer.as_mut() {
-            sum_freqs_skipped += self.block_cursor.freqs()[cur..new_cur].iter().sum::<u32>();
-            position_computer.add_skip(sum_freqs_skipped as usize);
-        }
-        self.cur = new_cur;
 
         // `doc` is now the first element >= `target`
-        let doc = output.0[new_cur];
+
+        // If all docs are smaller than target the current block should be incomplemented and padded
+        // with the value `TERMINATED`.
+        //
+        // After the search, the cursor should point to the first value of TERMINATED.
+        let doc = output.0[self.cur];
         debug_assert!(doc >= target);
-        if doc == target {
+        debug_assert_eq!(doc, self.doc());
-            SkipResult::Reached
+        doc
-        } else {
-            SkipResult::OverStep
-        }
     }
 
     /// Return the current document's `DocId`.
-    ///
-    /// # Panics
-    ///
-    /// Will panics if called without having called advance before.
     #[inline]
     fn doc(&self) -> DocId {
-        let docs = self.block_cursor.docs();
+        self.block_cursor.doc(self.cur)
-        debug_assert!(
-            self.cur < docs.len(),
-            "Have you forgotten to call `.advance()` at least once before calling `.doc()`                                      ."
-        );
-        docs[self.cur]
     }
 
     fn size_hint(&self) -> u32 {
         self.len() as u32
     }
-
-    fn append_to_bitset(&mut self, bitset: &mut BitSet) {
-        // finish the current block
-        if self.advance() {
-            for &doc in &self.block_cursor.docs()[self.cur..] {
-                bitset.insert(doc);
-            }
-            // ... iterate through the remaining blocks.
-            while self.block_cursor.advance() {
-                for &doc in self.block_cursor.docs() {
-                    bitset.insert(doc);
-                }
-            }
-        }
-    }
 }
 
 impl HasLen for SegmentPostings {
     fn len(&self) -> usize {
-        self.block_cursor.doc_freq()
+        self.block_cursor.doc_freq() as usize
     }
 }
 
@@ -290,515 +245,63 @@ impl Postings for SegmentPostings {
 
     fn positions_with_offset(&mut self, offset: u32, output: &mut Vec<u32>) {
         let term_freq = self.term_freq() as usize;
-        if let Some(position_comp) = self.position_computer.as_mut() {
+        if let Some(position_reader) = self.position_reader.as_mut() {
+            let read_offset = self.block_cursor.position_offset()
+                + (self.block_cursor.freqs()[..self.cur]
+                    .iter()
+                    .cloned()
+                    .sum::<u32>() as u64);
             output.resize(term_freq, 0u32);
-            position_comp.positions_with_offset(offset, &mut output[..]);
+            position_reader.read(read_offset, &mut output[..]);
+            let mut cum = offset;
+            for output_mut in output.iter_mut() {
+                cum += *output_mut;
+                *output_mut = cum;
+            }
         } else {
             output.clear();
         }
     }
 }
 
-/// `BlockSegmentPostings` is a cursor iterating over blocks
-/// of documents.
-///
-/// # Warning
-///
-/// While it is useful for some very specific high-performance
-/// use cases, you should prefer using `SegmentPostings` for most usage.
-pub struct BlockSegmentPostings {
-    doc_decoder: BlockDecoder,
-    freq_decoder: BlockDecoder,
-    freq_reading_option: FreqReadingOption,
-
-    doc_freq: usize,
-    doc_offset: DocId,
-
-    num_vint_docs: usize,
-
-    remaining_data: OwnedRead,
-    skip_reader: SkipReader,
-}
-
-fn split_into_skips_and_postings(
-    doc_freq: u32,
-    mut data: OwnedRead,
-) -> (Option<OwnedRead>, OwnedRead) {
-    if doc_freq >= USE_SKIP_INFO_LIMIT {
-        let skip_len = VInt::deserialize(&mut data).expect("Data corrupted").0 as usize;
-        let mut postings_data = data.clone();
-        postings_data.advance(skip_len);
-        data.clip(skip_len);
-        (Some(data), postings_data)
-    } else {
-        (None, data)
-    }
-}
-
-#[derive(Debug, Eq, PartialEq)]
-pub enum BlockSegmentPostingsSkipResult {
-    Terminated,
-    Success(u32), //< number of term freqs to skip
-}
-
-impl BlockSegmentPostings {
-    pub(crate) fn from_data(
-        doc_freq: u32,
-        data: OwnedRead,
-        record_option: IndexRecordOption,
-        requested_option: IndexRecordOption,
-    ) -> BlockSegmentPostings {
-        let freq_reading_option = match (record_option, requested_option) {
-            (IndexRecordOption::Basic, _) => FreqReadingOption::NoFreq,
-            (_, IndexRecordOption::Basic) => FreqReadingOption::SkipFreq,
-            (_, _) => FreqReadingOption::ReadFreq,
-        };
-
-        let (skip_data_opt, postings_data) = split_into_skips_and_postings(doc_freq, data);
-        let skip_reader = match skip_data_opt {
-            Some(skip_data) => SkipReader::new(skip_data, record_option),
-            None => SkipReader::new(OwnedRead::new(&[][..]), record_option),
-        };
-        let doc_freq = doc_freq as usize;
-        let num_vint_docs = doc_freq % COMPRESSION_BLOCK_SIZE;
-        BlockSegmentPostings {
-            num_vint_docs,
-            doc_decoder: BlockDecoder::new(),
-            freq_decoder: BlockDecoder::with_val(1),
-            freq_reading_option,
-            doc_offset: 0,
-            doc_freq,
-            remaining_data: postings_data,
-            skip_reader,
-        }
-    }
-
-    // Resets the block segment postings on another position
-    // in the postings file.
-    //
-    // This is useful for enumerating through a list of terms,
-    // and consuming the associated posting lists while avoiding
-    // reallocating a `BlockSegmentPostings`.
-    //
-    // # Warning
-    //
-    // This does not reset the positions list.
-    pub(crate) fn reset(&mut self, doc_freq: u32, postings_data: OwnedRead) {
-        let (skip_data_opt, postings_data) = split_into_skips_and_postings(doc_freq, postings_data);
-        let num_vint_docs = (doc_freq as usize) & (COMPRESSION_BLOCK_SIZE - 1);
-        self.num_vint_docs = num_vint_docs;
-        self.remaining_data = postings_data;
-        if let Some(skip_data) = skip_data_opt {
-            self.skip_reader.reset(skip_data);
-        } else {
-            self.skip_reader.reset(OwnedRead::new(&[][..]))
-        }
-        self.doc_offset = 0;
-        self.doc_freq = doc_freq as usize;
-    }
-
-    /// Returns the document frequency associated to this block postings.
-    ///
-    /// This `doc_freq` is simply the sum of the length of all of the blocks
-    /// length, and it does not take in account deleted documents.
-    pub fn doc_freq(&self) -> usize {
-        self.doc_freq
-    }
-
-    /// Returns the array of docs in the current block.
-    ///
-    /// Before the first call to `.advance()`, the block
-    /// returned by `.docs()` is empty.
-    #[inline]
-    pub fn docs(&self) -> &[DocId] {
-        self.doc_decoder.output_array()
-    }
-
-    pub(crate) fn docs_aligned(&self) -> (&AlignedBuffer, usize) {
-        self.doc_decoder.output_aligned()
-    }
-
-    /// Return the document at index `idx` of the block.
-    #[inline]
-    pub fn doc(&self, idx: usize) -> u32 {
-        self.doc_decoder.output(idx)
-    }
-
-    /// Return the array of `term freq` in the block.
-    #[inline]
-    pub fn freqs(&self) -> &[u32] {
-        self.freq_decoder.output_array()
-    }
-
-    /// Return the frequency at index `idx` of the block.
-    #[inline]
-    pub fn freq(&self, idx: usize) -> u32 {
-        self.freq_decoder.output(idx)
-    }
-
-    /// Returns the length of the current block.
-    ///
-    /// All blocks have a length of `NUM_DOCS_PER_BLOCK`,
-    /// except the last block that may have a length
-    /// of any number between 1 and `NUM_DOCS_PER_BLOCK - 1`
-    #[inline]
-    fn block_len(&self) -> usize {
-        self.doc_decoder.output_len
-    }
-
-    /// position on a block that may contains `doc_id`.
-    /// Always advance the current block.
-    ///
-    /// Returns true if a block that has an element greater or equal to the target is found.
-    /// Returning true does not guarantee that the smallest element of the block is smaller
-    /// than the target. It only guarantees that the last element is greater or equal.
-    ///
-    /// Returns false iff all of the document remaining are smaller than
-    /// `doc_id`. In that case, all of these document are consumed.
-    ///
-    pub fn skip_to(&mut self, target_doc: DocId) -> BlockSegmentPostingsSkipResult {
-        let mut skip_freqs = 0u32;
-        while self.skip_reader.advance() {
-            if self.skip_reader.doc() >= target_doc {
-                // the last document of the current block is larger
-                // than the target.
-                //
-                // We found our block!
-                let num_bits = self.skip_reader.doc_num_bits();
-                let num_consumed_bytes = self.doc_decoder.uncompress_block_sorted(
-                    self.remaining_data.as_ref(),
-                    self.doc_offset,
-                    num_bits,
-                );
-                self.remaining_data.advance(num_consumed_bytes);
-                let tf_num_bits = self.skip_reader.tf_num_bits();
-                match self.freq_reading_option {
-                    FreqReadingOption::NoFreq => {}
-                    FreqReadingOption::SkipFreq => {
-                        let num_bytes_to_skip = compressed_block_size(tf_num_bits);
-                        self.remaining_data.advance(num_bytes_to_skip);
-                    }
-                    FreqReadingOption::ReadFreq => {
-                        let num_consumed_bytes = self
-                            .freq_decoder
-                            .uncompress_block_unsorted(self.remaining_data.as_ref(), tf_num_bits);
-                        self.remaining_data.advance(num_consumed_bytes);
-                    }
-                }
-                self.doc_offset = self.skip_reader.doc();
-                return BlockSegmentPostingsSkipResult::Success(skip_freqs);
-            } else {
-                skip_freqs += self.skip_reader.tf_sum();
-                let advance_len = self.skip_reader.total_block_len();
-                self.doc_offset = self.skip_reader.doc();
-                self.remaining_data.advance(advance_len);
-            }
-        }
-
-        // we are now on the last, incomplete, variable encoded block.
-        if self.num_vint_docs > 0 {
-            let num_compressed_bytes = self.doc_decoder.uncompress_vint_sorted(
-                self.remaining_data.as_ref(),
-                self.doc_offset,
-                self.num_vint_docs,
-            );
-            self.remaining_data.advance(num_compressed_bytes);
-            match self.freq_reading_option {
-                FreqReadingOption::NoFreq | FreqReadingOption::SkipFreq => {}
-                FreqReadingOption::ReadFreq => {
-                    self.freq_decoder
-                        .uncompress_vint_unsorted(self.remaining_data.as_ref(), self.num_vint_docs);
-                }
-            }
-            self.num_vint_docs = 0;
-            return self
-                .docs()
-                .last()
-                .map(|last_doc| {
-                    if *last_doc >= target_doc {
-                        BlockSegmentPostingsSkipResult::Success(skip_freqs)
-                    } else {
-                        BlockSegmentPostingsSkipResult::Terminated
-                    }
-                })
-                .unwrap_or(BlockSegmentPostingsSkipResult::Terminated);
-        }
-        BlockSegmentPostingsSkipResult::Terminated
-    }
-
-    /// Advance to the next block.
-    ///
-    /// Returns false iff there was no remaining blocks.
-    pub fn advance(&mut self) -> bool {
-        if self.skip_reader.advance() {
-            let num_bits = self.skip_reader.doc_num_bits();
-            let num_consumed_bytes = self.doc_decoder.uncompress_block_sorted(
-                self.remaining_data.as_ref(),
-                self.doc_offset,
-                num_bits,
-            );
-            self.remaining_data.advance(num_consumed_bytes);
-            let tf_num_bits = self.skip_reader.tf_num_bits();
-            match self.freq_reading_option {
-                FreqReadingOption::NoFreq => {}
-                FreqReadingOption::SkipFreq => {
-                    let num_bytes_to_skip = compressed_block_size(tf_num_bits);
-                    self.remaining_data.advance(num_bytes_to_skip);
-                }
-                FreqReadingOption::ReadFreq => {
-                    let num_consumed_bytes = self
-                        .freq_decoder
-                        .uncompress_block_unsorted(self.remaining_data.as_ref(), tf_num_bits);
-                    self.remaining_data.advance(num_consumed_bytes);
-                }
-            }
-            // it will be used as the next offset.
-            self.doc_offset = self.doc_decoder.output(COMPRESSION_BLOCK_SIZE - 1);
-            true
-        } else if self.num_vint_docs > 0 {
-            let num_compressed_bytes = self.doc_decoder.uncompress_vint_sorted(
-                self.remaining_data.as_ref(),
-                self.doc_offset,
-                self.num_vint_docs,
-            );
-            self.remaining_data.advance(num_compressed_bytes);
-            match self.freq_reading_option {
-                FreqReadingOption::NoFreq | FreqReadingOption::SkipFreq => {}
-                FreqReadingOption::ReadFreq => {
-                    self.freq_decoder
-                        .uncompress_vint_unsorted(self.remaining_data.as_ref(), self.num_vint_docs);
-                }
-            }
-            self.num_vint_docs = 0;
-            true
-        } else {
-            false
-        }
-    }
-
-    /// Returns an empty segment postings object
-    pub fn empty() -> BlockSegmentPostings {
-        BlockSegmentPostings {
-            num_vint_docs: 0,
-
-            doc_decoder: BlockDecoder::new(),
-            freq_decoder: BlockDecoder::with_val(1),
-            freq_reading_option: FreqReadingOption::NoFreq,
-
-            doc_offset: 0,
-            doc_freq: 0,
-
-            remaining_data: OwnedRead::new(vec![]),
-            skip_reader: SkipReader::new(OwnedRead::new(vec![]), IndexRecordOption::Basic),
-        }
-    }
-}
-
-impl<'b> Streamer<'b> for BlockSegmentPostings {
-    type Item = &'b [DocId];
-
-    fn next(&'b mut self) -> Option<&'b [DocId]> {
-        if self.advance() {
-            Some(self.docs())
-        } else {
-            None
-        }
-    }
-}
-
 #[cfg(test)]
 mod tests {
-    use super::BlockSegmentPostings;
+
-    use super::BlockSegmentPostingsSkipResult;
     use super::SegmentPostings;
     use crate::common::HasLen;
-    use crate::core::Index;
+
-    use crate::docset::DocSet;
+    use crate::docset::{DocSet, TERMINATED};
+    use crate::fastfield::DeleteBitSet;
     use crate::postings::postings::Postings;
-    use crate::schema::IndexRecordOption;
-    use crate::schema::Schema;
-    use crate::schema::Term;
-    use crate::schema::INDEXED;
-    use crate::DocId;
-    use crate::SkipResult;
-    use tantivy_fst::Streamer;
 
     #[test]
     fn test_empty_segment_postings() {
         let mut postings = SegmentPostings::empty();
-        assert!(!postings.advance());
+        assert_eq!(postings.advance(), TERMINATED);
-        assert!(!postings.advance());
+        assert_eq!(postings.advance(), TERMINATED);
         assert_eq!(postings.len(), 0);
     }
 
     #[test]
-    #[should_panic(expected = "Have you forgotten to call `.advance()`")]
+    fn test_empty_postings_doc_returns_terminated() {
-    fn test_panic_if_doc_called_before_advance() {
+        let mut postings = SegmentPostings::empty();
-        SegmentPostings::empty().doc();
+        assert_eq!(postings.doc(), TERMINATED);
+        assert_eq!(postings.advance(), TERMINATED);
     }
 
     #[test]
-    #[should_panic(expected = "Have you forgotten to call `.advance()`")]
+    fn test_empty_postings_doc_term_freq_returns_0() {
-    fn test_panic_if_freq_called_before_advance() {
+        let postings = SegmentPostings::empty();
-        SegmentPostings::empty().term_freq();
+        assert_eq!(postings.term_freq(), 1);
     }
 
     #[test]
-    fn test_empty_block_segment_postings() {
+    fn test_doc_freq() {
-        let mut postings = BlockSegmentPostings::empty();
+        let docs = SegmentPostings::create_from_docs(&[0, 2, 10]);
-        assert!(!postings.advance());
+        assert_eq!(docs.doc_freq(), 3);
-        assert_eq!(postings.doc_freq(), 0);
+        let delete_bitset = DeleteBitSet::for_test(&[2], 12);
-    }
+        assert_eq!(docs.doc_freq_given_deletes(&delete_bitset), 2);
-
+        let all_deleted = DeleteBitSet::for_test(&[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11], 12);
-    #[test]
+        assert_eq!(docs.doc_freq_given_deletes(&all_deleted), 0);
-    fn test_block_segment_postings() {
-        let mut block_segments = build_block_postings(&(0..100_000).collect::<Vec<u32>>());
-        let mut offset: u32 = 0u32;
-        // checking that the block before calling advance is empty
-        assert!(block_segments.docs().is_empty());
-        // checking that the `doc_freq` is correct
-        assert_eq!(block_segments.doc_freq(), 100_000);
-        while let Some(block) = block_segments.next() {
-            for (i, doc) in block.iter().cloned().enumerate() {
-                assert_eq!(offset + (i as u32), doc);
-            }
-            offset += block.len() as u32;
-        }
-    }
-
-    #[test]
-    fn test_skip_right_at_new_block() {
-        let mut doc_ids = (0..128).collect::<Vec<u32>>();
-        doc_ids.push(129);
-        doc_ids.push(130);
-        {
-            let block_segments = build_block_postings(&doc_ids);
-            let mut docset = SegmentPostings::from_block_postings(block_segments, None);
-            assert_eq!(docset.skip_next(128), SkipResult::OverStep);
-            assert_eq!(docset.doc(), 129);
-            assert!(docset.advance());
-            assert_eq!(docset.doc(), 130);
-            assert!(!docset.advance());
-        }
-        {
-            let block_segments = build_block_postings(&doc_ids);
-            let mut docset = SegmentPostings::from_block_postings(block_segments, None);
-            assert_eq!(docset.skip_next(129), SkipResult::Reached);
-            assert_eq!(docset.doc(), 129);
-            assert!(docset.advance());
-            assert_eq!(docset.doc(), 130);
-            assert!(!docset.advance());
-        }
-        {
-            let block_segments = build_block_postings(&doc_ids);
-            let mut docset = SegmentPostings::from_block_postings(block_segments, None);
-            assert_eq!(docset.skip_next(131), SkipResult::End);
-        }
-    }
-
-    fn build_block_postings(docs: &[DocId]) -> BlockSegmentPostings {
-        let mut schema_builder = Schema::builder();
-        let int_field = schema_builder.add_u64_field("id", INDEXED);
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-        let mut last_doc = 0u32;
-        for &doc in docs {
-            for _ in last_doc..doc {
-                index_writer.add_document(doc!(int_field=>1u64));
-            }
-            index_writer.add_document(doc!(int_field=>0u64));
-            last_doc = doc + 1;
-        }
-        index_writer.commit().unwrap();
-        let searcher = index.reader().unwrap().searcher();
-        let segment_reader = searcher.segment_reader(0);
-        let inverted_index = segment_reader.inverted_index(int_field);
-        let term = Term::from_field_u64(int_field, 0u64);
-        let term_info = inverted_index.get_term_info(&term).unwrap();
-        inverted_index.read_block_postings_from_terminfo(&term_info, IndexRecordOption::Basic)
-    }
-
-    #[test]
-    fn test_block_segment_postings_skip() {
-        for i in 0..4 {
-            let mut block_postings = build_block_postings(&[3]);
-            assert_eq!(
-                block_postings.skip_to(i),
-                BlockSegmentPostingsSkipResult::Success(0u32)
-            );
-            assert_eq!(
-                block_postings.skip_to(i),
-                BlockSegmentPostingsSkipResult::Terminated
-            );
-        }
-        let mut block_postings = build_block_postings(&[3]);
-        assert_eq!(
-            block_postings.skip_to(4u32),
-            BlockSegmentPostingsSkipResult::Terminated
-        );
-    }
-
-    #[test]
-    fn test_block_segment_postings_skip2() {
-        let mut docs = vec![0];
-        for i in 0..1300 {
-            docs.push((i * i / 100) + i);
-        }
-        let mut block_postings = build_block_postings(&docs[..]);
-        for i in vec![0, 424, 10000] {
-            assert_eq!(
-                block_postings.skip_to(i),
-                BlockSegmentPostingsSkipResult::Success(0u32)
-            );
-            let docs = block_postings.docs();
-            assert!(docs[0] <= i);
-            assert!(docs.last().cloned().unwrap_or(0u32) >= i);
-        }
-        assert_eq!(
-            block_postings.skip_to(100_000),
-            BlockSegmentPostingsSkipResult::Terminated
-        );
-        assert_eq!(
-            block_postings.skip_to(101_000),
-            BlockSegmentPostingsSkipResult::Terminated
-        );
-    }
-
-    #[test]
-    fn test_reset_block_segment_postings() {
-        let mut schema_builder = Schema::builder();
-        let int_field = schema_builder.add_u64_field("id", INDEXED);
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-        // create two postings list, one containg even number,
-        // the other containing odd numbers.
-        for i in 0..6 {
-            let doc = doc!(int_field=> (i % 2) as u64);
-            index_writer.add_document(doc);
-        }
-        index_writer.commit().unwrap();
-        let searcher = index.reader().unwrap().searcher();
-        let segment_reader = searcher.segment_reader(0);
-
-        let mut block_segments;
-        {
-            let term = Term::from_field_u64(int_field, 0u64);
-            let inverted_index = segment_reader.inverted_index(int_field);
-            let term_info = inverted_index.get_term_info(&term).unwrap();
-            block_segments = inverted_index
-                .read_block_postings_from_terminfo(&term_info, IndexRecordOption::Basic);
-        }
-        assert!(block_segments.advance());
-        assert_eq!(block_segments.docs(), &[0, 2, 4]);
-        {
-            let term = Term::from_field_u64(int_field, 1u64);
-            let inverted_index = segment_reader.inverted_index(int_field);
-            let term_info = inverted_index.get_term_info(&term).unwrap();
-            inverted_index.reset_block_postings_from_terminfo(&term_info, &mut block_segments);
-        }
-        assert!(block_segments.advance());
-        assert_eq!(block_segments.docs(), &[1, 3, 5]);
     }
 }

src/postings/serializer.rs

@@ -3,15 +3,16 @@ use crate::common::{BinarySerializable, VInt};
 use crate::common::{CompositeWrite, CountingWriter};
 use crate::core::Segment;
 use crate::directory::WritePtr;
+use crate::fieldnorm::FieldNormReader;
 use crate::positions::PositionSerializer;
 use crate::postings::compression::{BlockEncoder, VIntEncoder, COMPRESSION_BLOCK_SIZE};
 use crate::postings::skip::SkipSerializer;
-use crate::postings::USE_SKIP_INFO_LIMIT;
+use crate::query::BM25Weight;
 use crate::schema::Schema;
 use crate::schema::{Field, FieldEntry, FieldType};
 use crate::termdict::{TermDictionaryBuilder, TermOrdinal};
-use crate::DocId;
+use crate::{DocId, Score};
-use crate::Result;
+use std::cmp::Ordering;
 use std::io::{self, Write};
 
 /// `InvertedIndexSerializer` is in charge of serializing
@@ -61,7 +62,7 @@ impl InvertedIndexSerializer {
         positions_write: CompositeWrite<WritePtr>,
         positionsidx_write: CompositeWrite<WritePtr>,
         schema: Schema,
-    ) -> Result<InvertedIndexSerializer> {
+    ) -> crate::Result<InvertedIndexSerializer> {
         Ok(InvertedIndexSerializer {
             terms_write,
             postings_write,
@@ -72,7 +73,7 @@ impl InvertedIndexSerializer {
     }
 
     /// Open a new `PostingsSerializer` for the given segment
-    pub fn open(segment: &mut Segment) -> Result<InvertedIndexSerializer> {
+    pub fn open(segment: &mut Segment) -> crate::Result<InvertedIndexSerializer> {
         use crate::SegmentComponent::{POSITIONS, POSITIONSSKIP, POSTINGS, TERMS};
         InvertedIndexSerializer::create(
             CompositeWrite::wrap(segment.open_write(TERMS)?),
@@ -91,20 +92,22 @@ impl InvertedIndexSerializer {
         &mut self,
         field: Field,
         total_num_tokens: u64,
+        fieldnorm_reader: Option<FieldNormReader>,
     ) -> io::Result<FieldSerializer<'_>> {
         let field_entry: &FieldEntry = self.schema.get_field_entry(field);
         let term_dictionary_write = self.terms_write.for_field(field);
         let postings_write = self.postings_write.for_field(field);
-        total_num_tokens.serialize(postings_write)?;
         let positions_write = self.positions_write.for_field(field);
         let positionsidx_write = self.positionsidx_write.for_field(field);
         let field_type: FieldType = (*field_entry.field_type()).clone();
         FieldSerializer::create(
             &field_type,
+            total_num_tokens,
             term_dictionary_write,
             postings_write,
             positions_write,
             positionsidx_write,
+            fieldnorm_reader,
         )
     }
 
@@ -132,11 +135,14 @@ pub struct FieldSerializer<'a> {
 impl<'a> FieldSerializer<'a> {
     fn create(
         field_type: &FieldType,
+        total_num_tokens: u64,
         term_dictionary_write: &'a mut CountingWriter<WritePtr>,
         postings_write: &'a mut CountingWriter<WritePtr>,
         positions_write: &'a mut CountingWriter<WritePtr>,
         positionsidx_write: &'a mut CountingWriter<WritePtr>,
+        fieldnorm_reader: Option<FieldNormReader>,
     ) -> io::Result<FieldSerializer<'a>> {
+        total_num_tokens.serialize(postings_write)?;
         let (term_freq_enabled, position_enabled): (bool, bool) = match field_type {
             FieldType::Str(ref text_options) => {
                 if let Some(text_indexing_options) = text_options.get_indexing_options() {
@@ -148,10 +154,18 @@ impl<'a> FieldSerializer<'a> {
             }
             _ => (false, false),
         };
-        let term_dictionary_builder =
+        let term_dictionary_builder = TermDictionaryBuilder::create(term_dictionary_write)?;
-            TermDictionaryBuilder::create(term_dictionary_write, &field_type)?;
+        let average_fieldnorm = fieldnorm_reader
-        let postings_serializer =
+            .as_ref()
-            PostingsSerializer::new(postings_write, term_freq_enabled, position_enabled);
+            .map(|ff_reader| (total_num_tokens as Score / ff_reader.num_docs() as Score))
+            .unwrap_or(0.0);
+        let postings_serializer = PostingsSerializer::new(
+            postings_write,
+            average_fieldnorm,
+            term_freq_enabled,
+            position_enabled,
+            fieldnorm_reader,
+        );
         let positions_serializer_opt = if position_enabled {
             Some(PositionSerializer::new(positions_write, positionsidx_write))
         } else {
@@ -184,18 +198,20 @@ impl<'a> FieldSerializer<'a> {
     /// Starts the postings for a new term.
     /// * term - the term. It needs to come after the previous term according
     ///   to the lexicographical order.
-    /// * doc_freq - return the number of document containing the term.
+    /// * term_doc_freq - return the number of document containing the term.
-    pub fn new_term(&mut self, term: &[u8]) -> io::Result<TermOrdinal> {
+    pub fn new_term(&mut self, term: &[u8], term_doc_freq: u32) -> io::Result<TermOrdinal> {
         assert!(
             !self.term_open,
             "Called new_term, while the previous term was not closed."
         );
+
         self.term_open = true;
         self.postings_serializer.clear();
         self.current_term_info = self.current_term_info();
         self.term_dictionary_builder.insert_key(term)?;
         let term_ordinal = self.num_terms;
         self.num_terms += 1;
+        self.postings_serializer.new_term(term_doc_freq);
         Ok(term_ordinal)
     }
 
@@ -309,14 +325,27 @@ pub struct PostingsSerializer<W: Write> {
 
     termfreq_enabled: bool,
     termfreq_sum_enabled: bool,
+    fieldnorm_reader: Option<FieldNormReader>,
+
+    bm25_weight: Option<BM25Weight>,
+
+    num_docs: u32, // Number of docs in the segment
+    avg_fieldnorm: Score, // Average number of term in the field for that segment.
+                   // this value is used to compute the block wand information.
 }
 
 impl<W: Write> PostingsSerializer<W> {
     pub fn new(
         write: W,
+        avg_fieldnorm: Score,
         termfreq_enabled: bool,
         termfreq_sum_enabled: bool,
+        fieldnorm_reader: Option<FieldNormReader>,
     ) -> PostingsSerializer<W> {
+        let num_docs = fieldnorm_reader
+            .as_ref()
+            .map(|fieldnorm_reader| fieldnorm_reader.num_docs())
+            .unwrap_or(0u32);
         PostingsSerializer {
             output_write: CountingWriter::wrap(write),
 
@@ -329,6 +358,23 @@ impl<W: Write> PostingsSerializer<W> {
             last_doc_id_encoded: 0u32,
             termfreq_enabled,
             termfreq_sum_enabled,
+
+            fieldnorm_reader,
+            bm25_weight: None,
+
+            num_docs,
+            avg_fieldnorm,
+        }
+    }
+
+    pub fn new_term(&mut self, term_doc_freq: u32) {
+        if self.termfreq_enabled && self.num_docs > 0 {
+            let bm25_weight = BM25Weight::for_one_term(
+                term_doc_freq as u64,
+                self.num_docs as u64,
+                self.avg_fieldnorm,
+            );
+            self.bm25_weight = Some(bm25_weight);
         }
     }
 
@@ -345,7 +391,6 @@ impl<W: Write> PostingsSerializer<W> {
             self.postings_write.extend(block_encoded);
         }
         if self.termfreq_enabled {
-            // encode the term_freqs
             let (num_bits, block_encoded): (u8, &[u8]) = self
                 .block_encoder
                 .compress_block_unsorted(&self.block.term_freqs());
@@ -355,6 +400,31 @@ impl<W: Write> PostingsSerializer<W> {
                 let sum_freq = self.block.term_freqs().iter().cloned().sum();
                 self.skip_write.write_total_term_freq(sum_freq);
             }
+            let mut blockwand_params = (0u8, 0u32);
+            if let Some(bm25_weight) = self.bm25_weight.as_ref() {
+                if let Some(fieldnorm_reader) = self.fieldnorm_reader.as_ref() {
+                    let docs = self.block.doc_ids().iter().cloned();
+                    let term_freqs = self.block.term_freqs().iter().cloned();
+                    let fieldnorms = docs.map(|doc| fieldnorm_reader.fieldnorm_id(doc));
+                    blockwand_params = fieldnorms
+                        .zip(term_freqs)
+                        .max_by(
+                            |(left_fieldnorm_id, left_term_freq),
+                             (right_fieldnorm_id, right_term_freq)| {
+                                let left_score =
+                                    bm25_weight.tf_factor(*left_fieldnorm_id, *left_term_freq);
+                                let right_score =
+                                    bm25_weight.tf_factor(*right_fieldnorm_id, *right_term_freq);
+                                left_score
+                                    .partial_cmp(&right_score)
+                                    .unwrap_or(Ordering::Equal)
+                            },
+                        )
+                        .unwrap();
+                }
+            }
+            let (fieldnorm_id, term_freq) = blockwand_params;
+            self.skip_write.write_blockwand_max(fieldnorm_id, term_freq);
         }
         self.block.clear();
     }
@@ -393,7 +463,7 @@ impl<W: Write> PostingsSerializer<W> {
             }
             self.block.clear();
         }
-        if doc_freq >= USE_SKIP_INFO_LIMIT {
+        if doc_freq >= COMPRESSION_BLOCK_SIZE as u32 {
             let skip_data = self.skip_write.data();
             VInt(skip_data.len() as u64).serialize(&mut self.output_write)?;
             self.output_write.write_all(skip_data)?;
@@ -403,6 +473,7 @@ impl<W: Write> PostingsSerializer<W> {
         }
         self.skip_write.clear();
         self.postings_write.clear();
+        self.bm25_weight = None;
         Ok(())
     }
 

src/postings/skip.rs

@@ -1,7 +1,9 @@
-use crate::common::BinarySerializable;
+use crate::common::{read_u32_vint_no_advance, serialize_vint_u32, BinarySerializable, VInt};
-use crate::postings::compression::COMPRESSION_BLOCK_SIZE;
+use crate::directory::ReadOnlySource;
+use crate::postings::compression::{compressed_block_size, COMPRESSION_BLOCK_SIZE};
+use crate::query::BM25Weight;
 use crate::schema::IndexRecordOption;
-use crate::DocId;
+use crate::{DocId, Score, TERMINATED};
 use owned_read::OwnedRead;
 
 pub struct SkipSerializer {
@@ -39,6 +41,13 @@ impl SkipSerializer {
             .expect("Should never fail");
     }
 
+    pub fn write_blockwand_max(&mut self, fieldnorm_id: u8, term_freq: u32) {
+        self.buffer.push(fieldnorm_id);
+        let mut buf = [0u8; 8];
+        let bytes = serialize_vint_u32(term_freq, &mut buf);
+        self.buffer.extend_from_slice(bytes);
+    }
+
     pub fn data(&self) -> &[u8] {
         &self.buffer[..]
     }
@@ -49,81 +58,200 @@ impl SkipSerializer {
     }
 }
 
+#[derive(Clone)]
 pub(crate) struct SkipReader {
-    doc: DocId,
+    last_doc_in_block: DocId,
+    pub(crate) last_doc_in_previous_block: DocId,
     owned_read: OwnedRead,
-    doc_num_bits: u8,
-    tf_num_bits: u8,
-    tf_sum: u32,
     skip_info: IndexRecordOption,
+    byte_offset: usize,
+    remaining_docs: u32, // number of docs remaining, including the
+    // documents in the current block.
+    block_info: BlockInfo,
+
+    position_offset: u64,
+}
+
+#[derive(Clone, Eq, PartialEq, Copy, Debug)]
+pub(crate) enum BlockInfo {
+    BitPacked {
+        doc_num_bits: u8,
+        tf_num_bits: u8,
+        tf_sum: u32,
+        block_wand_fieldnorm_id: u8,
+        block_wand_term_freq: u32,
+    },
+    VInt {
+        num_docs: u32,
+    },
+}
+
+impl Default for BlockInfo {
+    fn default() -> Self {
+        BlockInfo::VInt { num_docs: 0u32 }
+    }
 }
 
 impl SkipReader {
-    pub fn new(data: OwnedRead, skip_info: IndexRecordOption) -> SkipReader {
+    pub fn new(data: ReadOnlySource, doc_freq: u32, skip_info: IndexRecordOption) -> SkipReader {
-        SkipReader {
+        let mut skip_reader = SkipReader {
-            doc: 0u32,
+            last_doc_in_block: if doc_freq >= COMPRESSION_BLOCK_SIZE as u32 {
-            owned_read: data,
+                0
+            } else {
+                TERMINATED
+            },
+            last_doc_in_previous_block: 0u32,
+            owned_read: OwnedRead::new(data),
             skip_info,
-            doc_num_bits: 0u8,
+            block_info: BlockInfo::VInt { num_docs: doc_freq },
-            tf_num_bits: 0u8,
+            byte_offset: 0,
-            tf_sum: 0u32,
+            remaining_docs: doc_freq,
+            position_offset: 0u64,
+        };
+        if doc_freq >= COMPRESSION_BLOCK_SIZE as u32 {
+            skip_reader.read_block_info();
+        }
+        skip_reader
+    }
+
+    pub fn reset(&mut self, data: ReadOnlySource, doc_freq: u32) {
+        self.last_doc_in_block = if doc_freq >= COMPRESSION_BLOCK_SIZE as u32 {
+            0
+        } else {
+            TERMINATED
+        };
+        self.last_doc_in_previous_block = 0u32;
+        self.owned_read = OwnedRead::new(data);
+        self.block_info = BlockInfo::VInt { num_docs: doc_freq };
+        self.byte_offset = 0;
+        self.remaining_docs = doc_freq;
+        self.position_offset = 0u64;
+        if doc_freq >= COMPRESSION_BLOCK_SIZE as u32 {
+            self.read_block_info();
         }
     }
 
-    pub fn reset(&mut self, data: OwnedRead) {
+    // Returns the block max score for this block if available.
-        self.doc = 0u32;
+    //
-        self.owned_read = data;
+    // The block max score is available for all full bitpacked block,
-        self.doc_num_bits = 0u8;
+    // but no available for the last VInt encoded incomplete block.
-        self.tf_num_bits = 0u8;
+    pub fn block_max_score(&self, bm25_weight: &BM25Weight) -> Option<Score> {
-        self.tf_sum = 0u32;
+        match self.block_info {
+            BlockInfo::BitPacked {
+                block_wand_fieldnorm_id,
+                block_wand_term_freq,
+                ..
+            } => Some(bm25_weight.score(block_wand_fieldnorm_id, block_wand_term_freq)),
+            BlockInfo::VInt { .. } => None,
+        }
     }
 
-    pub fn total_block_len(&self) -> usize {
+    pub(crate) fn last_doc_in_block(&self) -> DocId {
-        (self.doc_num_bits + self.tf_num_bits) as usize * COMPRESSION_BLOCK_SIZE / 8
+        self.last_doc_in_block
     }
 
-    pub fn doc(&self) -> DocId {
+    pub fn position_offset(&self) -> u64 {
-        self.doc
+        self.position_offset
     }
 
-    pub fn doc_num_bits(&self) -> u8 {
+    pub fn byte_offset(&self) -> usize {
-        self.doc_num_bits
+        self.byte_offset
     }
 
-    /// Number of bits used to encode term frequencies
+    fn read_block_info(&mut self) {
-    ///
+        let doc_delta = u32::deserialize(&mut self.owned_read).expect("Skip data corrupted");
-    /// 0 if term frequencies are not enabled.
+        self.last_doc_in_block += doc_delta as DocId;
-    pub fn tf_num_bits(&self) -> u8 {
+        let doc_num_bits = self.owned_read.get(0);
-        self.tf_num_bits
+        match self.skip_info {
-    }
+            IndexRecordOption::Basic => {
-
+                self.owned_read.advance(1);
-    pub fn tf_sum(&self) -> u32 {
+                self.block_info = BlockInfo::BitPacked {
-        self.tf_sum
+                    doc_num_bits,
-    }
+                    tf_num_bits: 0,
-
+                    tf_sum: 0,
-    pub fn advance(&mut self) -> bool {
+                    block_wand_fieldnorm_id: 0,
-        if self.owned_read.as_ref().is_empty() {
+                    block_wand_term_freq: 0,
-            false
+                };
-        } else {
-            let doc_delta = u32::deserialize(&mut self.owned_read).expect("Skip data corrupted");
-            self.doc += doc_delta as DocId;
-            self.doc_num_bits = self.owned_read.get(0);
-            match self.skip_info {
-                IndexRecordOption::Basic => {
-                    self.owned_read.advance(1);
-                }
-                IndexRecordOption::WithFreqs => {
-                    self.tf_num_bits = self.owned_read.get(1);
-                    self.owned_read.advance(2);
-                }
-                IndexRecordOption::WithFreqsAndPositions => {
-                    self.tf_num_bits = self.owned_read.get(1);
-                    self.owned_read.advance(2);
-                    self.tf_sum =
-                        u32::deserialize(&mut self.owned_read).expect("Failed reading tf_sum");
-                }
             }
-            true
+            IndexRecordOption::WithFreqs => {
+                let tf_num_bits = self.owned_read.get(1);
+                let block_wand_fieldnorm_id = self.owned_read.get(2);
+                let data = &self.owned_read.as_ref()[3..];
+                let (block_wand_term_freq, num_bytes) = read_u32_vint_no_advance(data);
+                self.owned_read.advance(3 + num_bytes);
+                self.block_info = BlockInfo::BitPacked {
+                    doc_num_bits,
+                    tf_num_bits,
+                    tf_sum: 0,
+                    block_wand_fieldnorm_id,
+                    block_wand_term_freq,
+                };
+            }
+            IndexRecordOption::WithFreqsAndPositions => {
+                let tf_num_bits = self.owned_read.get(1);
+                self.owned_read.advance(2);
+                let tf_sum = u32::deserialize(&mut self.owned_read).expect("Failed reading tf_sum");
+                let block_wand_fieldnorm_id = self.owned_read.get(0);
+                self.owned_read.advance(1);
+                let block_wand_term_freq =
+                    VInt::deserialize_u64(&mut self.owned_read).unwrap() as u32;
+                self.block_info = BlockInfo::BitPacked {
+                    doc_num_bits,
+                    tf_num_bits,
+                    tf_sum,
+                    block_wand_fieldnorm_id,
+                    block_wand_term_freq,
+                };
+            }
+        }
+    }
+
+    pub fn block_info(&self) -> BlockInfo {
+        self.block_info
+    }
+
+    /// Advance the skip reader to the block that may contain the target.
+    ///
+    /// If the target is larger than all documents, the skip_reader
+    /// then advance to the last Variable In block.
+    pub fn seek(&mut self, target: DocId) -> bool {
+        if self.last_doc_in_block() >= target {
+            return false;
+        }
+        loop {
+            self.advance();
+            if self.last_doc_in_block() >= target {
+                return true;
+            }
+        }
+    }
+
+    pub fn advance(&mut self) {
+        match self.block_info {
+            BlockInfo::BitPacked {
+                doc_num_bits,
+                tf_num_bits,
+                tf_sum,
+                ..
+            } => {
+                self.remaining_docs -= COMPRESSION_BLOCK_SIZE as u32;
+                self.byte_offset += compressed_block_size(doc_num_bits + tf_num_bits);
+                self.position_offset += tf_sum as u64;
+            }
+            BlockInfo::VInt { num_docs } => {
+                debug_assert_eq!(num_docs, self.remaining_docs);
+                self.remaining_docs = 0;
+                self.byte_offset = std::usize::MAX;
+            }
+        }
+        self.last_doc_in_previous_block = self.last_doc_in_block;
+        if self.remaining_docs >= COMPRESSION_BLOCK_SIZE as u32 {
+            self.read_block_info();
+        } else {
+            self.last_doc_in_block = TERMINATED;
+            self.block_info = BlockInfo::VInt {
+                num_docs: self.remaining_docs,
+            };
         }
     }
 }
@@ -131,9 +259,11 @@ impl SkipReader {
 #[cfg(test)]
 mod tests {
 
+    use super::BlockInfo;
     use super::IndexRecordOption;
     use super::{SkipReader, SkipSerializer};
-    use owned_read::OwnedRead;
+    use crate::directory::ReadOnlySource;
+    use crate::postings::compression::COMPRESSION_BLOCK_SIZE;
 
     #[test]
     fn test_skip_with_freq() {
@@ -141,20 +271,47 @@ mod tests {
             let mut skip_serializer = SkipSerializer::new();
             skip_serializer.write_doc(1u32, 2u8);
             skip_serializer.write_term_freq(3u8);
+            skip_serializer.write_blockwand_max(13u8, 3u32);
             skip_serializer.write_doc(5u32, 5u8);
             skip_serializer.write_term_freq(2u8);
+            skip_serializer.write_blockwand_max(8u8, 2u32);
             skip_serializer.data().to_owned()
         };
-        let mut skip_reader = SkipReader::new(OwnedRead::new(buf), IndexRecordOption::WithFreqs);
+        let doc_freq = 3u32 + (COMPRESSION_BLOCK_SIZE * 2) as u32;
-        assert!(skip_reader.advance());
+        let mut skip_reader = SkipReader::new(
-        assert_eq!(skip_reader.doc(), 1u32);
+            ReadOnlySource::new(buf),
-        assert_eq!(skip_reader.doc_num_bits(), 2u8);
+            doc_freq,
-        assert_eq!(skip_reader.tf_num_bits(), 3u8);
+            IndexRecordOption::WithFreqs,
-        assert!(skip_reader.advance());
+        );
-        assert_eq!(skip_reader.doc(), 5u32);
+        assert_eq!(skip_reader.last_doc_in_block(), 1u32);
-        assert_eq!(skip_reader.doc_num_bits(), 5u8);
+        assert_eq!(
-        assert_eq!(skip_reader.tf_num_bits(), 2u8);
+            skip_reader.block_info,
-        assert!(!skip_reader.advance());
+            BlockInfo::BitPacked {
+                doc_num_bits: 2u8,
+                tf_num_bits: 3u8,
+                tf_sum: 0,
+                block_wand_fieldnorm_id: 13,
+                block_wand_term_freq: 3
+            }
+        );
+        skip_reader.advance();
+        assert_eq!(skip_reader.last_doc_in_block(), 5u32);
+        assert_eq!(
+            skip_reader.block_info(),
+            BlockInfo::BitPacked {
+                doc_num_bits: 5u8,
+                tf_num_bits: 2u8,
+                tf_sum: 0,
+                block_wand_fieldnorm_id: 8,
+                block_wand_term_freq: 2
+            }
+        );
+        skip_reader.advance();
+        assert_eq!(skip_reader.block_info(), BlockInfo::VInt { num_docs: 3u32 });
+        skip_reader.advance();
+        assert_eq!(skip_reader.block_info(), BlockInfo::VInt { num_docs: 0u32 });
+        skip_reader.advance();
+        assert_eq!(skip_reader.block_info(), BlockInfo::VInt { num_docs: 0u32 });
     }
 
     #[test]
@@ -165,13 +322,68 @@ mod tests {
             skip_serializer.write_doc(5u32, 5u8);
             skip_serializer.data().to_owned()
         };
-        let mut skip_reader = SkipReader::new(OwnedRead::new(buf), IndexRecordOption::Basic);
+        let doc_freq = 3u32 + (COMPRESSION_BLOCK_SIZE * 2) as u32;
-        assert!(skip_reader.advance());
+        let mut skip_reader = SkipReader::new(
-        assert_eq!(skip_reader.doc(), 1u32);
+            ReadOnlySource::from(buf),
-        assert_eq!(skip_reader.doc_num_bits(), 2u8);
+            doc_freq,
-        assert!(skip_reader.advance());
+            IndexRecordOption::Basic,
-        assert_eq!(skip_reader.doc(), 5u32);
+        );
-        assert_eq!(skip_reader.doc_num_bits(), 5u8);
+        assert_eq!(skip_reader.last_doc_in_block(), 1u32);
-        assert!(!skip_reader.advance());
+        assert_eq!(
+            skip_reader.block_info(),
+            BlockInfo::BitPacked {
+                doc_num_bits: 2u8,
+                tf_num_bits: 0,
+                tf_sum: 0u32,
+                block_wand_fieldnorm_id: 0,
+                block_wand_term_freq: 0
+            }
+        );
+        skip_reader.advance();
+        assert_eq!(skip_reader.last_doc_in_block(), 5u32);
+        assert_eq!(
+            skip_reader.block_info(),
+            BlockInfo::BitPacked {
+                doc_num_bits: 5u8,
+                tf_num_bits: 0,
+                tf_sum: 0u32,
+                block_wand_fieldnorm_id: 0,
+                block_wand_term_freq: 0
+            }
+        );
+        skip_reader.advance();
+        assert_eq!(skip_reader.block_info(), BlockInfo::VInt { num_docs: 3u32 });
+        skip_reader.advance();
+        assert_eq!(skip_reader.block_info(), BlockInfo::VInt { num_docs: 0u32 });
+        skip_reader.advance();
+        assert_eq!(skip_reader.block_info(), BlockInfo::VInt { num_docs: 0u32 });
+    }
+
+    #[test]
+    fn test_skip_multiple_of_block_size() {
+        let buf = {
+            let mut skip_serializer = SkipSerializer::new();
+            skip_serializer.write_doc(1u32, 2u8);
+            skip_serializer.data().to_owned()
+        };
+        let doc_freq = COMPRESSION_BLOCK_SIZE as u32;
+        let mut skip_reader = SkipReader::new(
+            ReadOnlySource::from(buf),
+            doc_freq,
+            IndexRecordOption::Basic,
+        );
+        assert_eq!(skip_reader.last_doc_in_block(), 1u32);
+        assert_eq!(
+            skip_reader.block_info(),
+            BlockInfo::BitPacked {
+                doc_num_bits: 2u8,
+                tf_num_bits: 0,
+                tf_sum: 0u32,
+                block_wand_fieldnorm_id: 0,
+                block_wand_term_freq: 0
+            }
+        );
+        skip_reader.advance();
+        assert_eq!(skip_reader.block_info(), BlockInfo::VInt { num_docs: 0u32 });
     }
 }

src/postings/stacker/term_hashmap.rs

@@ -1,6 +1,4 @@
-use murmurhash32;
+use murmurhash32::murmurhash2;
-
-use self::murmurhash32::murmurhash2;
 
 use super::{Addr, MemoryArena};
 use crate::postings::stacker::memory_arena::store;

src/query/all_query.rs

@@ -1,20 +1,20 @@
 use crate::core::Searcher;
 use crate::core::SegmentReader;
-use crate::docset::DocSet;
+use crate::docset::{DocSet, TERMINATED};
+use crate::query::boost_query::BoostScorer;
 use crate::query::explanation::does_not_match;
 use crate::query::{Explanation, Query, Scorer, Weight};
 use crate::DocId;
-use crate::Result;
 use crate::Score;
 
 /// Query that matches all of the documents.
 ///
-/// All of the document get the score 1f32.
+/// All of the document get the score 1.0.
 #[derive(Clone, Debug)]
 pub struct AllQuery;
 
 impl Query for AllQuery {
-    fn weight(&self, _: &Searcher, _: bool) -> Result<Box<dyn Weight>> {
+    fn weight(&self, _: &Searcher, _: bool) -> crate::Result<Box<dyn Weight>> {
         Ok(Box::new(AllWeight))
     }
 }
@@ -23,55 +23,36 @@ impl Query for AllQuery {
 pub struct AllWeight;
 
 impl Weight for AllWeight {
-    fn scorer(&self, reader: &SegmentReader) -> Result<Box<dyn Scorer>> {
+    fn scorer(&self, reader: &SegmentReader, boost: Score) -> crate::Result<Box<dyn Scorer>> {
-        Ok(Box::new(AllScorer {
+        let all_scorer = AllScorer {
-            state: State::NotStarted,
             doc: 0u32,
             max_doc: reader.max_doc(),
-        }))
+        };
+        Ok(Box::new(BoostScorer::new(all_scorer, boost)))
     }
 
-    fn explain(&self, reader: &SegmentReader, doc: DocId) -> Result<Explanation> {
+    fn explain(&self, reader: &SegmentReader, doc: DocId) -> crate::Result<Explanation> {
         if doc >= reader.max_doc() {
             return Err(does_not_match(doc));
         }
-        Ok(Explanation::new("AllQuery", 1f32))
+        Ok(Explanation::new("AllQuery", 1.0))
     }
 }
 
-enum State {
-    NotStarted,
-    Started,
-    Finished,
-}
-
 /// Scorer associated to the `AllQuery` query.
 pub struct AllScorer {
-    state: State,
     doc: DocId,
     max_doc: DocId,
 }
 
 impl DocSet for AllScorer {
-    fn advance(&mut self) -> bool {
+    fn advance(&mut self) -> DocId {
-        match self.state {
+        if self.doc + 1 >= self.max_doc {
-            State::NotStarted => {
+            self.doc = TERMINATED;
-                self.state = State::Started;
+            return TERMINATED;
-                self.doc = 0;
-            }
-            State::Started => {
-                self.doc += 1u32;
-            }
-            State::Finished => {
-                return false;
-            }
-        }
-        if self.doc < self.max_doc {
-            true
-        } else {
-            self.state = State::Finished;
-            false
         }
+        self.doc += 1;
+        self.doc
     }
 
     fn doc(&self) -> DocId {
@@ -85,20 +66,19 @@ impl DocSet for AllScorer {
 
 impl Scorer for AllScorer {
     fn score(&mut self) -> Score {
-        1f32
+        1.0
     }
 }
 
 #[cfg(test)]
 mod tests {
-
     use super::AllQuery;
+    use crate::docset::TERMINATED;
     use crate::query::Query;
     use crate::schema::{Schema, TEXT};
     use crate::Index;
 
-    #[test]
+    fn create_test_index() -> Index {
-    fn test_all_query() {
         let mut schema_builder = Schema::builder();
         let field = schema_builder.add_text_field("text", TEXT);
         let schema = schema_builder.build();
@@ -109,25 +89,47 @@ mod tests {
         index_writer.commit().unwrap();
         index_writer.add_document(doc!(field=>"ccc"));
         index_writer.commit().unwrap();
+        index
+    }
+
+    #[test]
+    fn test_all_query() {
+        let index = create_test_index();
         let reader = index.reader().unwrap();
-        reader.reload().unwrap();
         let searcher = reader.searcher();
         let weight = AllQuery.weight(&searcher, false).unwrap();
         {
             let reader = searcher.segment_reader(0);
-            let mut scorer = weight.scorer(reader).unwrap();
+            let mut scorer = weight.scorer(reader, 1.0).unwrap();
-            assert!(scorer.advance());
             assert_eq!(scorer.doc(), 0u32);
-            assert!(scorer.advance());
+            assert_eq!(scorer.advance(), 1u32);
             assert_eq!(scorer.doc(), 1u32);
-            assert!(!scorer.advance());
+            assert_eq!(scorer.advance(), TERMINATED);
         }
         {
             let reader = searcher.segment_reader(1);
-            let mut scorer = weight.scorer(reader).unwrap();
+            let mut scorer = weight.scorer(reader, 1.0).unwrap();
-            assert!(scorer.advance());
             assert_eq!(scorer.doc(), 0u32);
-            assert!(!scorer.advance());
+            assert_eq!(scorer.advance(), TERMINATED);
+        }
+    }
+
+    #[test]
+    fn test_all_query_with_boost() {
+        let index = create_test_index();
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+        let weight = AllQuery.weight(&searcher, false).unwrap();
+        let reader = searcher.segment_reader(0);
+        {
+            let mut scorer = weight.scorer(reader, 2.0).unwrap();
+            assert_eq!(scorer.doc(), 0u32);
+            assert_eq!(scorer.score(), 2.0);
+        }
+        {
+            let mut scorer = weight.scorer(reader, 1.5).unwrap();
+            assert_eq!(scorer.doc(), 0u32);
+            assert_eq!(scorer.score(), 1.5);
         }
     }
 }

src/query/automaton_weight.rs

@@ -5,9 +5,9 @@ use crate::query::{BitSetDocSet, Explanation};
 use crate::query::{Scorer, Weight};
 use crate::schema::{Field, IndexRecordOption};
 use crate::termdict::{TermDictionary, TermStreamer};
-use crate::DocId;
+use crate::Result;
 use crate::TantivyError;
-use crate::{Result, SkipResult};
+use crate::{DocId, Score};
 use std::sync::Arc;
 use tantivy_fst::Automaton;
 
@@ -40,10 +40,9 @@ impl<A> Weight for AutomatonWeight<A>
 where
     A: Automaton + Send + Sync + 'static,
 {
-    fn scorer(&self, reader: &SegmentReader) -> Result<Box<dyn Scorer>> {
+    fn scorer(&self, reader: &SegmentReader, boost: Score) -> Result<Box<dyn Scorer>> {
         let max_doc = reader.max_doc();
         let mut doc_bitset = BitSet::with_max_value(max_doc);
-
         let inverted_index = reader.inverted_index(self.field);
         let term_dict = inverted_index.terms();
         let mut term_stream = self.automaton_stream(term_dict);
@@ -51,20 +50,26 @@ where
             let term_info = term_stream.value();
             let mut block_segment_postings = inverted_index
                 .read_block_postings_from_terminfo(term_info, IndexRecordOption::Basic);
-            while block_segment_postings.advance() {
+            loop {
-                for &doc in block_segment_postings.docs() {
+                let docs = block_segment_postings.docs();
+                if docs.is_empty() {
+                    break;
+                }
+                for &doc in docs {
                     doc_bitset.insert(doc);
                 }
+                block_segment_postings.advance();
             }
         }
         let doc_bitset = BitSetDocSet::from(doc_bitset);
-        Ok(Box::new(ConstScorer::new(doc_bitset)))
+        let const_scorer = ConstScorer::new(doc_bitset, boost);
+        Ok(Box::new(const_scorer))
     }
 
     fn explain(&self, reader: &SegmentReader, doc: DocId) -> Result<Explanation> {
-        let mut scorer = self.scorer(reader)?;
+        let mut scorer = self.scorer(reader, 1.0)?;
-        if scorer.skip_next(doc) == SkipResult::Reached {
+        if scorer.seek(doc) == doc {
-            Ok(Explanation::new("AutomatonScorer", 1.0f32))
+            Ok(Explanation::new("AutomatonScorer", 1.0))
         } else {
             Err(TantivyError::InvalidArgument(
                 "Document does not exist".to_string(),
@@ -72,3 +77,94 @@ where
         }
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::AutomatonWeight;
+    use crate::docset::TERMINATED;
+    use crate::query::Weight;
+    use crate::schema::{Schema, STRING};
+    use crate::Index;
+    use tantivy_fst::Automaton;
+
+    fn create_index() -> Index {
+        let mut schema = Schema::builder();
+        let title = schema.add_text_field("title", STRING);
+        let index = Index::create_in_ram(schema.build());
+        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        index_writer.add_document(doc!(title=>"abc"));
+        index_writer.add_document(doc!(title=>"bcd"));
+        index_writer.add_document(doc!(title=>"abcd"));
+        assert!(index_writer.commit().is_ok());
+        index
+    }
+
+    enum State {
+        Start,
+        NotMatching,
+        AfterA,
+    }
+
+    struct PrefixedByA;
+
+    impl Automaton for PrefixedByA {
+        type State = State;
+
+        fn start(&self) -> Self::State {
+            State::Start
+        }
+
+        fn is_match(&self, state: &Self::State) -> bool {
+            match *state {
+                State::AfterA => true,
+                _ => false,
+            }
+        }
+
+        fn accept(&self, state: &Self::State, byte: u8) -> Self::State {
+            match *state {
+                State::Start => {
+                    if byte == b'a' {
+                        State::AfterA
+                    } else {
+                        State::NotMatching
+                    }
+                }
+                State::AfterA => State::AfterA,
+                State::NotMatching => State::NotMatching,
+            }
+        }
+    }
+
+    #[test]
+    fn test_automaton_weight() {
+        let index = create_index();
+        let field = index.schema().get_field("title").unwrap();
+        let automaton_weight = AutomatonWeight::new(field, PrefixedByA);
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+        let mut scorer = automaton_weight
+            .scorer(searcher.segment_reader(0u32), 1.0)
+            .unwrap();
+        assert_eq!(scorer.doc(), 0u32);
+        assert_eq!(scorer.score(), 1.0);
+        assert_eq!(scorer.advance(), 2u32);
+        assert_eq!(scorer.doc(), 2u32);
+        assert_eq!(scorer.score(), 1.0);
+        assert_eq!(scorer.advance(), TERMINATED);
+    }
+
+    #[test]
+    fn test_automaton_weight_boost() {
+        let index = create_index();
+        let field = index.schema().get_field("title").unwrap();
+        let automaton_weight = AutomatonWeight::new(field, PrefixedByA);
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+        let mut scorer = automaton_weight
+            .scorer(searcher.segment_reader(0u32), 1.32)
+            .unwrap();
+        assert_eq!(scorer.doc(), 0u32);
+        assert_eq!(scorer.score(), 1.32);
+    }
+}

src/query/bitset/mod.rs

@@ -1,7 +1,6 @@
 use crate::common::{BitSet, TinySet};
-use crate::docset::{DocSet, SkipResult};
+use crate::docset::{DocSet, TERMINATED};
 use crate::DocId;
-use std::cmp::Ordering;
 
 /// A `BitSetDocSet` makes it possible to iterate through a bitset as if it was a `DocSet`.
 ///
@@ -33,74 +32,51 @@ impl From<BitSet> for BitSetDocSet {
         } else {
             docs.tinyset(0)
         };
-        BitSetDocSet {
+        let mut docset = BitSetDocSet {
             docs,
             cursor_bucket: 0,
             cursor_tinybitset: first_tiny_bitset,
             doc: 0u32,
-        }
+        };
+        docset.advance();
+        docset
     }
 }
 
 impl DocSet for BitSetDocSet {
-    fn advance(&mut self) -> bool {
+    fn advance(&mut self) -> DocId {
         if let Some(lower) = self.cursor_tinybitset.pop_lowest() {
             self.doc = (self.cursor_bucket as u32 * 64u32) | lower;
-            return true;
+            return self.doc;
         }
         if let Some(cursor_bucket) = self.docs.first_non_empty_bucket(self.cursor_bucket + 1) {
             self.go_to_bucket(cursor_bucket);
             let lower = self.cursor_tinybitset.pop_lowest().unwrap();
             self.doc = (cursor_bucket * 64u32) | lower;
-            true
+            self.doc
         } else {
-            false
+            self.doc = TERMINATED;
+            TERMINATED
         }
     }
 
-    fn skip_next(&mut self, target: DocId) -> SkipResult {
+    fn seek(&mut self, target: DocId) -> DocId {
-        // skip is required to advance.
+        if target >= self.docs.max_value() {
-        if !self.advance() {
+            self.doc = TERMINATED;
-            return SkipResult::End;
+            return TERMINATED;
         }
         let target_bucket = target / 64u32;
-
+        if target_bucket > self.cursor_bucket {
-        // Mask for all of the bits greater or equal
+            self.go_to_bucket(target_bucket);
-        // to our target document.
+            let greater_filter: TinySet = TinySet::range_greater_or_equal(target);
-        match target_bucket.cmp(&self.cursor_bucket) {
+            self.cursor_tinybitset = self.cursor_tinybitset.intersect(greater_filter);
-            Ordering::Greater => {
+            self.advance()
-                self.go_to_bucket(target_bucket);
+        } else {
-                let greater_filter: TinySet = TinySet::range_greater_or_equal(target);
+            let mut doc = self.doc();
-                self.cursor_tinybitset = self.cursor_tinybitset.intersect(greater_filter);
+            while doc < target {
-                if !self.advance() {
+                doc = self.advance();
-                    SkipResult::End
-                } else if self.doc() == target {
-                    SkipResult::Reached
-                } else {
-                    debug_assert!(self.doc() > target);
-                    SkipResult::OverStep
-                }
-            }
-            Ordering::Equal => loop {
-                match self.doc().cmp(&target) {
-                    Ordering::Less => {
-                        if !self.advance() {
-                            return SkipResult::End;
-                        }
-                    }
-                    Ordering::Equal => {
-                        return SkipResult::Reached;
-                    }
-                    Ordering::Greater => {
-                        debug_assert!(self.doc() > target);
-                        return SkipResult::OverStep;
-                    }
-                }
-            },
-            Ordering::Less => {
-                debug_assert!(self.doc() > target);
-                SkipResult::OverStep
             }
+            doc
         }
     }
 
@@ -122,7 +98,7 @@ impl DocSet for BitSetDocSet {
 mod tests {
     use super::BitSetDocSet;
     use crate::common::BitSet;
-    use crate::docset::{DocSet, SkipResult};
+    use crate::docset::{DocSet, TERMINATED};
     use crate::DocId;
 
     fn create_docbitset(docs: &[DocId], max_doc: DocId) -> BitSetDocSet {
@@ -133,19 +109,31 @@ mod tests {
         BitSetDocSet::from(docset)
     }
 
+    #[test]
+    fn test_empty() {
+        let bitset = BitSet::with_max_value(1000);
+        let mut empty = BitSetDocSet::from(bitset);
+        assert_eq!(empty.advance(), TERMINATED)
+    }
+
+    #[test]
+    fn test_seek_terminated() {
+        let bitset = BitSet::with_max_value(1000);
+        let mut empty = BitSetDocSet::from(bitset);
+        assert_eq!(empty.seek(TERMINATED), TERMINATED)
+    }
+
     fn test_go_through_sequential(docs: &[DocId]) {
         let mut docset = create_docbitset(docs, 1_000u32);
         for &doc in docs {
-            assert!(docset.advance());
             assert_eq!(doc, docset.doc());
+            docset.advance();
         }
-        assert!(!docset.advance());
+        assert_eq!(docset.advance(), TERMINATED);
-        assert!(!docset.advance());
     }
 
     #[test]
     fn test_docbitset_sequential() {
-        test_go_through_sequential(&[]);
         test_go_through_sequential(&[1, 2, 3]);
         test_go_through_sequential(&[1, 2, 3, 4, 5, 63, 64, 65]);
         test_go_through_sequential(&[63, 64, 65]);
@@ -156,64 +144,64 @@ mod tests {
     fn test_docbitset_skip() {
         {
             let mut docset = create_docbitset(&[1, 5, 6, 7, 5112], 10_000);
-            assert_eq!(docset.skip_next(7), SkipResult::Reached);
+            assert_eq!(docset.seek(7), 7);
             assert_eq!(docset.doc(), 7);
-            assert!(docset.advance(), 7);
+            assert_eq!(docset.advance(), 5112);
             assert_eq!(docset.doc(), 5112);
-            assert!(!docset.advance());
+            assert_eq!(docset.advance(), TERMINATED);
         }
         {
             let mut docset = create_docbitset(&[1, 5, 6, 7, 5112], 10_000);
-            assert_eq!(docset.skip_next(3), SkipResult::OverStep);
+            assert_eq!(docset.seek(3), 5);
             assert_eq!(docset.doc(), 5);
-            assert!(docset.advance());
+            assert_eq!(docset.advance(), 6);
         }
         {
             let mut docset = create_docbitset(&[5112], 10_000);
-            assert_eq!(docset.skip_next(5112), SkipResult::Reached);
+            assert_eq!(docset.seek(5112), 5112);
             assert_eq!(docset.doc(), 5112);
-            assert!(!docset.advance());
+            assert_eq!(docset.advance(), TERMINATED);
         }
         {
             let mut docset = create_docbitset(&[5112], 10_000);
-            assert_eq!(docset.skip_next(5113), SkipResult::End);
+            assert_eq!(docset.seek(5113), TERMINATED);
-            assert!(!docset.advance());
+            assert_eq!(docset.advance(), TERMINATED);
         }
         {
             let mut docset = create_docbitset(&[5112], 10_000);
-            assert_eq!(docset.skip_next(5111), SkipResult::OverStep);
+            assert_eq!(docset.seek(5111), 5112);
             assert_eq!(docset.doc(), 5112);
-            assert!(!docset.advance());
+            assert_eq!(docset.advance(), TERMINATED);
         }
         {
             let mut docset = create_docbitset(&[1, 5, 6, 7, 5112, 5500, 6666], 10_000);
-            assert_eq!(docset.skip_next(5112), SkipResult::Reached);
+            assert_eq!(docset.seek(5112), 5112);
             assert_eq!(docset.doc(), 5112);
-            assert!(docset.advance());
+            assert_eq!(docset.advance(), 5500);
             assert_eq!(docset.doc(), 5500);
-            assert!(docset.advance());
+            assert_eq!(docset.advance(), 6666);
             assert_eq!(docset.doc(), 6666);
-            assert!(!docset.advance());
+            assert_eq!(docset.advance(), TERMINATED);
         }
         {
             let mut docset = create_docbitset(&[1, 5, 6, 7, 5112, 5500, 6666], 10_000);
-            assert_eq!(docset.skip_next(5111), SkipResult::OverStep);
+            assert_eq!(docset.seek(5111), 5112);
             assert_eq!(docset.doc(), 5112);
-            assert!(docset.advance());
+            assert_eq!(docset.advance(), 5500);
             assert_eq!(docset.doc(), 5500);
-            assert!(docset.advance());
+            assert_eq!(docset.advance(), 6666);
             assert_eq!(docset.doc(), 6666);
-            assert!(!docset.advance());
+            assert_eq!(docset.advance(), TERMINATED);
         }
         {
             let mut docset = create_docbitset(&[1, 5, 6, 7, 5112, 5513, 6666], 10_000);
-            assert_eq!(docset.skip_next(5111), SkipResult::OverStep);
+            assert_eq!(docset.seek(5111), 5112);
             assert_eq!(docset.doc(), 5112);
-            assert!(docset.advance());
+            assert_eq!(docset.advance(), 5513);
             assert_eq!(docset.doc(), 5513);
-            assert!(docset.advance());
+            assert_eq!(docset.advance(), 6666);
             assert_eq!(docset.doc(), 6666);
-            assert!(!docset.advance());
+            assert_eq!(docset.advance(), TERMINATED);
         }
     }
 }
@@ -223,6 +211,7 @@ mod bench {
 
     use super::BitSet;
     use super::BitSetDocSet;
+    use crate::docset::TERMINATED;
     use crate::test;
     use crate::tests;
     use crate::DocSet;
@@ -257,7 +246,7 @@ mod bench {
         }
         b.iter(|| {
             let mut docset = BitSetDocSet::from(bitset.clone());
-            while docset.advance() {}
+            while docset.advance() != TERMINATED {}
         });
     }
 }

src/query/bm25.rs

@@ -3,21 +3,24 @@ use crate::query::Explanation;
 use crate::Score;
 use crate::Searcher;
 use crate::Term;
+use serde::Deserialize;
+use serde::Serialize;
 
-const K1: f32 = 1.2;
+const K1: Score = 1.2;
-const B: f32 = 0.75;
+const B: Score = 0.75;
 
-fn idf(doc_freq: u64, doc_count: u64) -> f32 {
+fn idf(doc_freq: u64, doc_count: u64) -> Score {
-    let x = ((doc_count - doc_freq) as f32 + 0.5) / (doc_freq as f32 + 0.5);
+    assert!(doc_count >= doc_freq, "{} >= {}", doc_count, doc_freq);
-    (1f32 + x).ln()
+    let x = ((doc_count - doc_freq) as Score + 0.5) / (doc_freq as Score + 0.5);
+    (1.0 + x).ln()
 }
 
-fn cached_tf_component(fieldnorm: u32, average_fieldnorm: f32) -> f32 {
+fn cached_tf_component(fieldnorm: u32, average_fieldnorm: Score) -> Score {
-    K1 * (1f32 - B + B * fieldnorm as f32 / average_fieldnorm)
+    K1 * (1.0 - B + B * fieldnorm as Score / average_fieldnorm)
 }
 
-fn compute_tf_cache(average_fieldnorm: f32) -> [f32; 256] {
+fn compute_tf_cache(average_fieldnorm: Score) -> [Score; 256] {
-    let mut cache = [0f32; 256];
+    let mut cache: [Score; 256] = [0.0; 256];
     for (fieldnorm_id, cache_mut) in cache.iter_mut().enumerate() {
         let fieldnorm = FieldNormReader::id_to_fieldnorm(fieldnorm_id as u8);
         *cache_mut = cached_tf_component(fieldnorm, average_fieldnorm);
@@ -25,15 +28,30 @@ fn compute_tf_cache(average_fieldnorm: f32) -> [f32; 256] {
     cache
 }
 
+#[derive(Clone, PartialEq, Debug, Serialize, Deserialize)]
+pub struct BM25Params {
+    pub idf: Score,
+    pub avg_fieldnorm: Score,
+}
+
 #[derive(Clone)]
 pub struct BM25Weight {
     idf_explain: Explanation,
-    weight: f32,
+    weight: Score,
-    cache: [f32; 256],
+    cache: [Score; 256],
-    average_fieldnorm: f32,
+    average_fieldnorm: Score,
 }
 
 impl BM25Weight {
+    pub fn boost_by(&self, boost: Score) -> BM25Weight {
+        BM25Weight {
+            idf_explain: self.idf_explain.clone(),
+            weight: self.weight * boost,
+            cache: self.cache,
+            average_fieldnorm: self.average_fieldnorm,
+        }
+    }
+
     pub fn for_terms(searcher: &Searcher, terms: &[Term]) -> BM25Weight {
         assert!(!terms.is_empty(), "BM25 requires at least one term");
         let field = terms[0].field();
@@ -52,19 +70,11 @@ impl BM25Weight {
             total_num_tokens += inverted_index.total_num_tokens();
             total_num_docs += u64::from(segment_reader.max_doc());
         }
-        let average_fieldnorm = total_num_tokens as f32 / total_num_docs as f32;
+        let average_fieldnorm = total_num_tokens as Score / total_num_docs as Score;
 
-        let mut idf_explain: Explanation;
         if terms.len() == 1 {
             let term_doc_freq = searcher.doc_freq(&terms[0]);
-            let idf = idf(term_doc_freq, total_num_docs);
+            BM25Weight::for_one_term(term_doc_freq, total_num_docs, average_fieldnorm)
-            idf_explain =
-                Explanation::new("idf, computed as log(1 + (N - n + 0.5) / (n + 0.5))", idf);
-            idf_explain.add_const(
-                "n, number of docs containing this term",
-                term_doc_freq as f32,
-            );
-            idf_explain.add_const("N, total number of docs", total_num_docs as f32);
         } else {
             let idf = terms
                 .iter()
@@ -72,14 +82,30 @@ impl BM25Weight {
                     let term_doc_freq = searcher.doc_freq(term);
                     idf(term_doc_freq, total_num_docs)
                 })
-                .sum::<f32>();
+                .sum::<Score>();
-            idf_explain = Explanation::new("idf", idf);
+            let idf_explain = Explanation::new("idf", idf);
+            BM25Weight::new(idf_explain, average_fieldnorm)
         }
-        BM25Weight::new(idf_explain, average_fieldnorm)
     }
 
-    fn new(idf_explain: Explanation, average_fieldnorm: f32) -> BM25Weight {
+    pub fn for_one_term(
-        let weight = idf_explain.value() * (1f32 + K1);
+        term_doc_freq: u64,
+        total_num_docs: u64,
+        avg_fieldnorm: Score,
+    ) -> BM25Weight {
+        let idf = idf(term_doc_freq, total_num_docs);
+        let mut idf_explain =
+            Explanation::new("idf, computed as log(1 + (N - n + 0.5) / (n + 0.5))", idf);
+        idf_explain.add_const(
+            "n, number of docs containing this term",
+            term_doc_freq as Score,
+        );
+        idf_explain.add_const("N, total number of docs", total_num_docs as Score);
+        BM25Weight::new(idf_explain, avg_fieldnorm)
+    }
+
+    fn new(idf_explain: Explanation, average_fieldnorm: Score) -> BM25Weight {
+        let weight = idf_explain.value() * (1.0 + K1);
         BM25Weight {
             idf_explain,
             weight,
@@ -90,19 +116,27 @@ impl BM25Weight {
 
     #[inline(always)]
     pub fn score(&self, fieldnorm_id: u8, term_freq: u32) -> Score {
+        self.weight * self.tf_factor(fieldnorm_id, term_freq)
+    }
+
+    pub fn max_score(&self) -> Score {
+        self.score(255u8, 2_013_265_944)
+    }
+
+    #[inline(always)]
+    pub(crate) fn tf_factor(&self, fieldnorm_id: u8, term_freq: u32) -> Score {
+        let term_freq = term_freq as Score;
         let norm = self.cache[fieldnorm_id as usize];
-        let term_freq = term_freq as f32;
+        term_freq / (term_freq + norm)
-        self.weight * term_freq / (term_freq + norm)
     }
 
     pub fn explain(&self, fieldnorm_id: u8, term_freq: u32) -> Explanation {
         // The explain format is directly copied from Lucene's.
         // (So, Kudos to Lucene)
-
         let score = self.score(fieldnorm_id, term_freq);
 
         let norm = self.cache[fieldnorm_id as usize];
-        let term_freq = term_freq as f32;
+        let term_freq = term_freq as Score;
         let right_factor = term_freq / (term_freq + norm);
 
         let mut tf_explanation = Explanation::new(
@@ -115,12 +149,12 @@ impl BM25Weight {
         tf_explanation.add_const("b, length normalization parameter", B);
         tf_explanation.add_const(
             "dl, length of field",
-            FieldNormReader::id_to_fieldnorm(fieldnorm_id) as f32,
+            FieldNormReader::id_to_fieldnorm(fieldnorm_id) as Score,
         );
         tf_explanation.add_const("avgdl, average length of field", self.average_fieldnorm);
 
         let mut explanation = Explanation::new("TermQuery, product of...", score);
-        explanation.add_detail(Explanation::new("(K1+1)", K1 + 1f32));
+        explanation.add_detail(Explanation::new("(K1+1)", K1 + 1.0));
         explanation.add_detail(self.idf_explain.clone());
         explanation.add_detail(tf_explanation);
         explanation
@@ -131,10 +165,11 @@ impl BM25Weight {
 mod tests {
 
     use super::idf;
-    use crate::tests::assert_nearly_equals;
+    use crate::{assert_nearly_equals, Score};
 
     #[test]
     fn test_idf() {
-        assert_nearly_equals(idf(1, 2), 0.6931472);
+        let score: Score = 2.0;
+        assert_nearly_equals!(idf(1, 2), score.ln());
     }
 }

src/query/boolean_query/block_wand.rs

@@ -0,0 +1,432 @@
+use crate::query::term_query::TermScorer;
+use crate::query::Scorer;
+use crate::{DocId, DocSet, Score, TERMINATED};
+use std::ops::Deref;
+use std::ops::DerefMut;
+
+fn is_sorted<I: Iterator<Item = DocId>>(mut it: I) -> bool {
+    if let Some(first) = it.next() {
+        let mut prev = first;
+        for doc in it {
+            if doc < prev {
+                return false;
+            }
+            prev = doc;
+        }
+    }
+    true
+}
+
+/// Takes a term_scorers sorted by their current doc() and a threshold and returns
+/// Returns (pivot_len, pivot_ord) defined as follows:
+/// - `pivot_doc` lowest document that has a chance of exceeding (>) the threshold score.
+/// - `before_pivot_len` number of term_scorers such that term_scorer.doc() < pivot.
+/// - `pivot_len` number of term_scorers such that term_scorer.doc() <= pivot.
+///
+/// We always have `before_pivot_len` < `pivot_len`.
+///
+/// None is returned if we establish that no document can exceed the threshold.
+fn find_pivot_doc(
+    term_scorers: &[TermScorerWithMaxScore],
+    threshold: Score,
+) -> Option<(usize, usize, DocId)> {
+    let mut max_score = 0.0;
+    let mut before_pivot_len = 0;
+    let mut pivot_doc = TERMINATED;
+    while before_pivot_len < term_scorers.len() {
+        let term_scorer = &term_scorers[before_pivot_len];
+        max_score += term_scorer.max_score;
+        if max_score > threshold {
+            pivot_doc = term_scorer.doc();
+            break;
+        }
+        before_pivot_len += 1;
+    }
+    if pivot_doc == TERMINATED {
+        return None;
+    }
+    // Right now i is an ordinal, we want a len.
+    let mut pivot_len = before_pivot_len + 1;
+    // Some other term_scorer may be positioned on the same document.
+    pivot_len += term_scorers[pivot_len..]
+        .iter()
+        .take_while(|term_scorer| term_scorer.doc() == pivot_doc)
+        .count();
+    Some((before_pivot_len, pivot_len, pivot_doc))
+}
+
+struct TermScorerWithMaxScore<'a> {
+    scorer: &'a mut TermScorer,
+    max_score: Score,
+}
+
+impl<'a> From<&'a mut TermScorer> for TermScorerWithMaxScore<'a> {
+    fn from(scorer: &'a mut TermScorer) -> Self {
+        let max_score = scorer.max_score();
+        TermScorerWithMaxScore { scorer, max_score }
+    }
+}
+
+impl<'a> Deref for TermScorerWithMaxScore<'a> {
+    type Target = TermScorer;
+
+    fn deref(&self) -> &Self::Target {
+        self.scorer
+    }
+}
+
+impl<'a> DerefMut for TermScorerWithMaxScore<'a> {
+    fn deref_mut(&mut self) -> &mut Self::Target {
+        self.scorer
+    }
+}
+
+// Before and after calling this method, scorers need to be sorted by their `.doc()`.
+fn block_max_was_too_low_advance_one_scorer(
+    scorers: &mut Vec<TermScorerWithMaxScore>,
+    pivot_len: usize,
+) {
+    let mut scorer_to_seek = pivot_len - 1;
+    let mut doc_to_seek_after = scorers[scorer_to_seek].doc();
+    for scorer_ord in (0..pivot_len - 1).rev() {
+        let scorer = &scorers[scorer_ord];
+        if scorer.last_doc_in_block() <= doc_to_seek_after {
+            doc_to_seek_after = scorer.last_doc_in_block();
+            scorer_to_seek = scorer_ord;
+        }
+    }
+    for scorer in &scorers[pivot_len..] {
+        if scorer.doc() <= doc_to_seek_after {
+            doc_to_seek_after = scorer.doc();
+        }
+    }
+    scorers[scorer_to_seek].seek(doc_to_seek_after + 1);
+    restore_ordering(scorers, scorer_to_seek);
+}
+
+// Given a list of term_scorers and a `ord` and assuming that `term_scorers[ord]` is sorted
+// except term_scorers[ord] that might be in advance compared to its ranks,
+// bubble up term_scorers[ord] in order to restore the ordering.
+fn restore_ordering(term_scorers: &mut Vec<TermScorerWithMaxScore>, ord: usize) {
+    let doc = term_scorers[ord].doc();
+    for i in ord + 1..term_scorers.len() {
+        if term_scorers[i].doc() >= doc {
+            break;
+        }
+        term_scorers.swap(i, i - 1);
+    }
+    debug_assert!(is_sorted(term_scorers.iter().map(|scorer| scorer.doc())));
+}
+
+// Attempts to advance all term_scorers between `&term_scorers[0..before_len]` to the pivot.
+// If this works, return true.
+// If this fails (ie: one of the term_scorer does not contain `pivot_doc` and seek goes past the
+// pivot), reorder the term_scorers to ensure the list is still sorted and returns `false`.
+// If a term_scorer reach TERMINATED in the process return false remove the term_scorer and return.
+fn align_scorers(
+    term_scorers: &mut Vec<TermScorerWithMaxScore>,
+    pivot_doc: DocId,
+    before_pivot_len: usize,
+) -> bool {
+    debug_assert_ne!(pivot_doc, TERMINATED);
+    for i in (0..before_pivot_len).rev() {
+        let new_doc = term_scorers[i].seek(pivot_doc);
+        if new_doc != pivot_doc {
+            if new_doc == TERMINATED {
+                term_scorers.swap_remove(i);
+            }
+            // We went past the pivot.
+            // We just go through the outer loop mechanic (Note that pivot is
+            // still a possible candidate).
+            //
+            // Termination is still guaranteed since we can only consider the same
+            // pivot at most term_scorers.len() - 1 times.
+            restore_ordering(term_scorers, i);
+            return false;
+        }
+    }
+    true
+}
+
+// Assumes terms_scorers[..pivot_len] are positioned on the same doc (pivot_doc).
+// Advance term_scorers[..pivot_len] and out of these removes the terminated scores.
+// Restores the ordering of term_scorers.
+fn advance_all_scorers_on_pivot(term_scorers: &mut Vec<TermScorerWithMaxScore>, pivot_len: usize) {
+    for term_scorer in &mut term_scorers[..pivot_len] {
+        term_scorer.advance();
+    }
+    // TODO use drain_filter when available.
+    let mut i = 0;
+    while i != term_scorers.len() {
+        if term_scorers[i].doc() == TERMINATED {
+            term_scorers.swap_remove(i);
+        } else {
+            i += 1;
+        }
+    }
+    term_scorers.sort_by_key(|scorer| scorer.doc());
+}
+
+pub fn block_wand(
+    mut scorers: Vec<TermScorer>,
+    mut threshold: Score,
+    callback: &mut dyn FnMut(u32, Score) -> Score,
+) {
+    let mut scorers: Vec<TermScorerWithMaxScore> = scorers
+        .iter_mut()
+        .map(TermScorerWithMaxScore::from)
+        .collect();
+    scorers.sort_by_key(|scorer| scorer.doc());
+    loop {
+        // At this point we need to ensure that the scorers are sorted!
+        debug_assert!(is_sorted(scorers.iter().map(|scorer| scorer.doc())));
+        if let Some((before_pivot_len, pivot_len, pivot_doc)) =
+            find_pivot_doc(&scorers[..], threshold)
+        {
+            debug_assert_ne!(pivot_doc, TERMINATED);
+            debug_assert!(before_pivot_len < pivot_len);
+
+            let block_max_score_upperbound: Score = scorers[..pivot_len]
+                .iter_mut()
+                .map(|scorer| {
+                    scorer.shallow_seek(pivot_doc);
+                    scorer.block_max_score()
+                })
+                .sum();
+
+            // Beware after shallow advance, skip readers can be in advance compared to
+            // the segment posting lists.
+            //
+            // `block_segment_postings.load_block()` need to be called separately.
+            if block_max_score_upperbound <= threshold {
+                // Block max condition was not reached
+                // We could get away by simply advancing the scorers to DocId + 1 but it would
+                // be inefficient. The optimization requires proper explanation and was
+                // isolated in a different function.
+                block_max_was_too_low_advance_one_scorer(&mut scorers, pivot_len);
+                continue;
+            }
+
+            // Block max condition is observed.
+            //
+            // Let's try and advance all scorers before the pivot to the pivot.
+            if !align_scorers(&mut scorers, pivot_doc, before_pivot_len) {
+                // At least of the scorer does not contain the pivot.
+                //
+                // Let's stop scoring this pivot and go through the pivot selection again.
+                // Note that the current pivot is not necessarily a bad candidate and it
+                // may be picked again.
+                continue;
+            }
+
+            // At this point, all scorers are positioned on the doc.
+            let score = scorers[..pivot_len]
+                .iter_mut()
+                .map(|scorer| scorer.score())
+                .sum();
+            if score > threshold {
+                threshold = callback(pivot_doc, score);
+            }
+            // let's advance all of the scorers that are currently positioned on the pivot.
+            advance_all_scorers_on_pivot(&mut scorers, pivot_len);
+        } else {
+            return;
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::query::score_combiner::SumCombiner;
+    use crate::query::term_query::TermScorer;
+    use crate::query::Union;
+    use crate::query::{BM25Weight, Scorer};
+    use crate::{DocId, DocSet, Score, TERMINATED};
+    use proptest::prelude::*;
+    use std::cmp::Ordering;
+    use std::collections::BinaryHeap;
+    use std::iter;
+
+    struct Float(Score);
+    impl Eq for Float {}
+    impl PartialEq for Float {
+        fn eq(&self, other: &Self) -> bool {
+            self.cmp(&other) == Ordering::Equal
+        }
+    }
+    impl PartialOrd for Float {
+        fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
+            Some(self.cmp(other))
+        }
+    }
+    impl Ord for Float {
+        fn cmp(&self, other: &Self) -> Ordering {
+            other.0.partial_cmp(&self.0).unwrap_or(Ordering::Equal)
+        }
+    }
+
+    fn nearly_equals(left: Score, right: Score) -> bool {
+        (left - right).abs() < 0.000001 * (left + right).abs()
+    }
+
+    fn compute_checkpoints_for_each_pruning(
+        term_scorers: Vec<TermScorer>,
+        n: usize,
+    ) -> Vec<(DocId, Score)> {
+        let mut heap: BinaryHeap<Float> = BinaryHeap::with_capacity(n);
+        let mut checkpoints: Vec<(DocId, Score)> = Vec::new();
+        let mut limit: Score = 0.0;
+        super::block_wand(term_scorers, Score::MIN, &mut |doc, score| {
+            heap.push(Float(score));
+            if heap.len() > n {
+                heap.pop().unwrap();
+            }
+            if heap.len() == n {
+                limit = heap.peek().unwrap().0;
+            }
+            if !nearly_equals(score, limit) {
+                checkpoints.push((doc, score));
+            }
+            return limit;
+        });
+        checkpoints
+    }
+
+    fn compute_checkpoints_manual(term_scorers: Vec<TermScorer>, n: usize) -> Vec<(DocId, Score)> {
+        let mut heap: BinaryHeap<Float> = BinaryHeap::with_capacity(n);
+        let mut checkpoints: Vec<(DocId, Score)> = Vec::new();
+        let mut scorer: Union<TermScorer, SumCombiner> = Union::from(term_scorers);
+
+        let mut limit = Score::MIN;
+        loop {
+            if scorer.doc() == TERMINATED {
+                break;
+            }
+            let doc = scorer.doc();
+            let score = scorer.score();
+            if score > limit {
+                heap.push(Float(score));
+                if heap.len() > n {
+                    heap.pop().unwrap();
+                }
+                if heap.len() == n {
+                    limit = heap.peek().unwrap().0;
+                }
+                if !nearly_equals(score, limit) {
+                    checkpoints.push((doc, score));
+                }
+            }
+            scorer.advance();
+        }
+        checkpoints
+    }
+
+    const MAX_TERM_FREQ: u32 = 100u32;
+
+    fn posting_list(max_doc: u32) -> BoxedStrategy<Vec<(DocId, u32)>> {
+        (1..max_doc + 1)
+            .prop_flat_map(move |doc_freq| {
+                (
+                    proptest::bits::bitset::sampled(doc_freq as usize, 0..max_doc as usize),
+                    proptest::collection::vec(1u32..MAX_TERM_FREQ, doc_freq as usize),
+                )
+            })
+            .prop_map(|(docset, term_freqs)| {
+                docset
+                    .iter()
+                    .map(|doc| doc as u32)
+                    .zip(term_freqs.iter().cloned())
+                    .collect::<Vec<_>>()
+            })
+            .boxed()
+    }
+
+    fn gen_term_scorers(num_scorers: usize) -> BoxedStrategy<(Vec<Vec<(DocId, u32)>>, Vec<u32>)> {
+        (1u32..100u32)
+            .prop_flat_map(move |max_doc: u32| {
+                (
+                    proptest::collection::vec(posting_list(max_doc), num_scorers),
+                    proptest::collection::vec(2u32..10u32 * MAX_TERM_FREQ, max_doc as usize),
+                )
+            })
+            .boxed()
+    }
+
+    fn test_block_wand_aux(posting_lists: &[Vec<(DocId, u32)>], fieldnorms: &[u32]) {
+        // We virtually repeat all docs 64 times in order to emulate blocks of 2 documents
+        // and surface blogs more easily.
+        const REPEAT: usize = 64;
+        let fieldnorms_expanded = fieldnorms
+            .iter()
+            .cloned()
+            .flat_map(|fieldnorm| iter::repeat(fieldnorm).take(REPEAT))
+            .collect::<Vec<u32>>();
+
+        let postings_lists_expanded: Vec<Vec<(DocId, u32)>> = posting_lists
+            .iter()
+            .map(|posting_list| {
+                posting_list
+                    .into_iter()
+                    .cloned()
+                    .flat_map(|(doc, term_freq)| {
+                        (0 as u32..REPEAT as u32).map(move |offset| {
+                            (
+                                doc * (REPEAT as u32) + offset,
+                                if offset == 0 { term_freq } else { 1 },
+                            )
+                        })
+                    })
+                    .collect::<Vec<(DocId, u32)>>()
+            })
+            .collect::<Vec<_>>();
+
+        let total_fieldnorms: u64 = fieldnorms_expanded
+            .iter()
+            .cloned()
+            .map(|fieldnorm| fieldnorm as u64)
+            .sum();
+        let average_fieldnorm = (total_fieldnorms as Score) / (fieldnorms_expanded.len() as Score);
+        let max_doc = fieldnorms_expanded.len();
+
+        let term_scorers: Vec<TermScorer> = postings_lists_expanded
+            .iter()
+            .map(|postings| {
+                let bm25_weight = BM25Weight::for_one_term(
+                    postings.len() as u64,
+                    max_doc as u64,
+                    average_fieldnorm,
+                );
+                TermScorer::create_for_test(postings, &fieldnorms_expanded[..], bm25_weight)
+            })
+            .collect();
+        for top_k in 1..4 {
+            let checkpoints_for_each_pruning =
+                compute_checkpoints_for_each_pruning(term_scorers.clone(), top_k);
+            let checkpoints_manual = compute_checkpoints_manual(term_scorers.clone(), top_k);
+            assert_eq!(checkpoints_for_each_pruning.len(), checkpoints_manual.len());
+            for (&(left_doc, left_score), &(right_doc, right_score)) in checkpoints_for_each_pruning
+                .iter()
+                .zip(checkpoints_manual.iter())
+            {
+                assert_eq!(left_doc, right_doc);
+                assert!(nearly_equals(left_score, right_score));
+            }
+        }
+    }
+
+    proptest! {
+        #![proptest_config(ProptestConfig::with_cases(500))]
+        #[test]
+        fn test_block_wand_two_term_scorers((posting_lists, fieldnorms) in gen_term_scorers(2)) {
+            test_block_wand_aux(&posting_lists[..], &fieldnorms[..]);
+        }
+    }
+
+    proptest! {
+        #![proptest_config(ProptestConfig::with_cases(500))]
+        #[test]
+        fn test_block_wand_three_term_scorers((posting_lists, fieldnorms) in gen_term_scorers(3)) {
+            test_block_wand_aux(&posting_lists[..], &fieldnorms[..]);
+        }
+    }
+}

src/query/boolean_query/boolean_query.rs

@@ -5,7 +5,6 @@ use crate::query::TermQuery;
 use crate::query::Weight;
 use crate::schema::IndexRecordOption;
 use crate::schema::Term;
-use crate::Result;
 use crate::Searcher;
 use std::collections::BTreeSet;
 
@@ -30,9 +29,9 @@ use std::collections::BTreeSet;
 ///use tantivy::query::{BooleanQuery, Occur, PhraseQuery, Query, TermQuery};
 ///use tantivy::schema::{IndexRecordOption, Schema, TEXT};
 ///use tantivy::Term;
-///use tantivy::{Index, Result};
+///use tantivy::Index;
 ///
-///fn main() -> Result<()> {
+///fn main() -> tantivy::Result<()> {
 ///    let mut schema_builder = Schema::builder();
 ///    let title = schema_builder.add_text_field("title", TEXT);
 ///    let body = schema_builder.add_text_field("body", TEXT);
@@ -149,14 +148,14 @@ impl From<Vec<(Occur, Box<dyn Query>)>> for BooleanQuery {
 }
 
 impl Query for BooleanQuery {
-    fn weight(&self, searcher: &Searcher, scoring_enabled: bool) -> Result<Box<dyn Weight>> {
+    fn weight(&self, searcher: &Searcher, scoring_enabled: bool) -> crate::Result<Box<dyn Weight>> {
         let sub_weights = self
             .subqueries
             .iter()
             .map(|&(ref occur, ref subquery)| {
                 Ok((*occur, subquery.weight(searcher, scoring_enabled)?))
             })
-            .collect::<Result<_>>()?;
+            .collect::<crate::Result<_>>()?;
         Ok(Box::new(BooleanWeight::new(sub_weights, scoring_enabled)))
     }
 

src/query/boolean_query/boolean_weight.rs

@@ -1,7 +1,9 @@
 use crate::core::SegmentReader;
+use crate::postings::FreqReadingOption;
 use crate::query::explanation::does_not_match;
 use crate::query::score_combiner::{DoNothingCombiner, ScoreCombiner, SumWithCoordsCombiner};
 use crate::query::term_query::TermScorer;
+use crate::query::weight::{for_each_pruning_scorer, for_each_scorer};
 use crate::query::EmptyScorer;
 use crate::query::Exclude;
 use crate::query::Occur;
@@ -10,17 +12,21 @@ use crate::query::Scorer;
 use crate::query::Union;
 use crate::query::Weight;
 use crate::query::{intersect_scorers, Explanation};
-use crate::Result;
+use crate::{DocId, Score};
-use crate::{DocId, SkipResult};
 use std::collections::HashMap;
 
-fn scorer_union<TScoreCombiner>(scorers: Vec<Box<dyn Scorer>>) -> Box<dyn Scorer>
+enum SpecializedScorer {
+    TermUnion(Vec<TermScorer>),
+    Other(Box<dyn Scorer>),
+}
+
+fn scorer_union<TScoreCombiner>(scorers: Vec<Box<dyn Scorer>>) -> SpecializedScorer
 where
     TScoreCombiner: ScoreCombiner,
 {
     assert!(!scorers.is_empty());
     if scorers.len() == 1 {
-        return scorers.into_iter().next().unwrap(); //< we checked the size beforehands
+        return SpecializedScorer::Other(scorers.into_iter().next().unwrap()); //< we checked the size beforehands
     }
 
     {
@@ -30,14 +36,30 @@ where
                 .into_iter()
                 .map(|scorer| *(scorer.downcast::<TermScorer>().map_err(|_| ()).unwrap()))
                 .collect();
-            let scorer: Box<dyn Scorer> =
+            if scorers
-                Box::new(Union::<TermScorer, TScoreCombiner>::from(scorers));
+                .iter()
-            return scorer;
+                .all(|scorer| scorer.freq_reading_option() == FreqReadingOption::ReadFreq)
+            {
+                // Block wand is only available iff we read frequencies.
+                return SpecializedScorer::TermUnion(scorers);
+            } else {
+                return SpecializedScorer::Other(Box::new(Union::<_, TScoreCombiner>::from(
+                    scorers,
+                )));
+            }
         }
     }
+    SpecializedScorer::Other(Box::new(Union::<_, TScoreCombiner>::from(scorers)))
+}
 
-    let scorer: Box<dyn Scorer> = Box::new(Union::<_, TScoreCombiner>::from(scorers));
+fn into_box_scorer<TScoreCombiner: ScoreCombiner>(scorer: SpecializedScorer) -> Box<dyn Scorer> {
-    scorer
+    match scorer {
+        SpecializedScorer::TermUnion(term_scorers) => {
+            let union_scorer = Union::<TermScorer, TScoreCombiner>::from(term_scorers);
+            Box::new(union_scorer)
+        }
+        SpecializedScorer::Other(scorer) => scorer,
+    }
 }
 
 pub struct BooleanWeight {
@@ -56,10 +78,11 @@ impl BooleanWeight {
     fn per_occur_scorers(
         &self,
         reader: &SegmentReader,
-    ) -> Result<HashMap<Occur, Vec<Box<dyn Scorer>>>> {
+        boost: Score,
+    ) -> crate::Result<HashMap<Occur, Vec<Box<dyn Scorer>>>> {
         let mut per_occur_scorers: HashMap<Occur, Vec<Box<dyn Scorer>>> = HashMap::new();
         for &(ref occur, ref subweight) in &self.weights {
-            let sub_scorer: Box<dyn Scorer> = subweight.scorer(reader)?;
+            let sub_scorer: Box<dyn Scorer> = subweight.scorer(reader, boost)?;
             per_occur_scorers
                 .entry(*occur)
                 .or_insert_with(Vec::new)
@@ -71,41 +94,52 @@ impl BooleanWeight {
     fn complex_scorer<TScoreCombiner: ScoreCombiner>(
         &self,
         reader: &SegmentReader,
-    ) -> Result<Box<dyn Scorer>> {
+        boost: Score,
-        let mut per_occur_scorers = self.per_occur_scorers(reader)?;
+    ) -> crate::Result<SpecializedScorer> {
+        let mut per_occur_scorers = self.per_occur_scorers(reader, boost)?;
 
-        let should_scorer_opt: Option<Box<dyn Scorer>> = per_occur_scorers
+        let should_scorer_opt: Option<SpecializedScorer> = per_occur_scorers
             .remove(&Occur::Should)
             .map(scorer_union::<TScoreCombiner>);
 
         let exclude_scorer_opt: Option<Box<dyn Scorer>> = per_occur_scorers
             .remove(&Occur::MustNot)
-            .map(scorer_union::<TScoreCombiner>);
+            .map(scorer_union::<DoNothingCombiner>)
+            .map(into_box_scorer::<DoNothingCombiner>);
 
         let must_scorer_opt: Option<Box<dyn Scorer>> = per_occur_scorers
             .remove(&Occur::Must)
             .map(intersect_scorers);
 
-        let positive_scorer: Box<dyn Scorer> = match (should_scorer_opt, must_scorer_opt) {
+        let positive_scorer: SpecializedScorer = match (should_scorer_opt, must_scorer_opt) {
             (Some(should_scorer), Some(must_scorer)) => {
                 if self.scoring_enabled {
-                    Box::new(RequiredOptionalScorer::<_, _, TScoreCombiner>::new(
+                    SpecializedScorer::Other(Box::new(RequiredOptionalScorer::<
+                        Box<dyn Scorer>,
+                        Box<dyn Scorer>,
+                        TScoreCombiner,
+                    >::new(
                         must_scorer,
-                        should_scorer,
+                        into_box_scorer::<TScoreCombiner>(should_scorer),
-                    ))
+                    )))
                 } else {
-                    must_scorer
+                    SpecializedScorer::Other(must_scorer)
                 }
             }
-            (None, Some(must_scorer)) => must_scorer,
+            (None, Some(must_scorer)) => SpecializedScorer::Other(must_scorer),
             (Some(should_scorer), None) => should_scorer,
             (None, None) => {
-                return Ok(Box::new(EmptyScorer));
+                return Ok(SpecializedScorer::Other(Box::new(EmptyScorer)));
             }
         };
 
         if let Some(exclude_scorer) = exclude_scorer_opt {
-            Ok(Box::new(Exclude::new(positive_scorer, exclude_scorer)))
+            let positive_scorer_boxed: Box<dyn Scorer> =
+                into_box_scorer::<TScoreCombiner>(positive_scorer);
+            Ok(SpecializedScorer::Other(Box::new(Exclude::new(
+                positive_scorer_boxed,
+                exclude_scorer,
+            ))))
         } else {
             Ok(positive_scorer)
         }
@@ -113,7 +147,7 @@ impl BooleanWeight {
 }
 
 impl Weight for BooleanWeight {
-    fn scorer(&self, reader: &SegmentReader) -> Result<Box<dyn Scorer>> {
+    fn scorer(&self, reader: &SegmentReader, boost: Score) -> crate::Result<Box<dyn Scorer>> {
         if self.weights.is_empty() {
             Ok(Box::new(EmptyScorer))
         } else if self.weights.len() == 1 {
@@ -121,22 +155,26 @@ impl Weight for BooleanWeight {
             if occur == Occur::MustNot {
                 Ok(Box::new(EmptyScorer))
             } else {
-                weight.scorer(reader)
+                weight.scorer(reader, boost)
             }
         } else if self.scoring_enabled {
-            self.complex_scorer::<SumWithCoordsCombiner>(reader)
+            self.complex_scorer::<SumWithCoordsCombiner>(reader, boost)
+                .map(|specialized_scorer| {
+                    into_box_scorer::<SumWithCoordsCombiner>(specialized_scorer)
+                })
         } else {
-            self.complex_scorer::<DoNothingCombiner>(reader)
+            self.complex_scorer::<DoNothingCombiner>(reader, boost)
+                .map(into_box_scorer::<DoNothingCombiner>)
         }
     }
 
-    fn explain(&self, reader: &SegmentReader, doc: DocId) -> Result<Explanation> {
+    fn explain(&self, reader: &SegmentReader, doc: DocId) -> crate::Result<Explanation> {
-        let mut scorer = self.scorer(reader)?;
+        let mut scorer = self.scorer(reader, 1.0)?;
-        if scorer.skip_next(doc) != SkipResult::Reached {
+        if scorer.seek(doc) != doc {
             return Err(does_not_match(doc));
         }
         if !self.scoring_enabled {
-            return Ok(Explanation::new("BooleanQuery with no scoring", 1f32));
+            return Ok(Explanation::new("BooleanQuery with no scoring", 1.0));
         }
 
         let mut explanation = Explanation::new("BooleanClause. Sum of ...", scorer.score());
@@ -149,6 +187,53 @@ impl Weight for BooleanWeight {
         }
         Ok(explanation)
     }
+
+    fn for_each(
+        &self,
+        reader: &SegmentReader,
+        callback: &mut dyn FnMut(DocId, Score),
+    ) -> crate::Result<()> {
+        let scorer = self.complex_scorer::<SumWithCoordsCombiner>(reader, 1.0)?;
+        match scorer {
+            SpecializedScorer::TermUnion(term_scorers) => {
+                let mut union_scorer =
+                    Union::<TermScorer, SumWithCoordsCombiner>::from(term_scorers);
+                for_each_scorer(&mut union_scorer, callback);
+            }
+            SpecializedScorer::Other(mut scorer) => {
+                for_each_scorer(scorer.as_mut(), callback);
+            }
+        }
+        Ok(())
+    }
+
+    /// Calls `callback` with all of the `(doc, score)` for which score
+    /// is exceeding a given threshold.
+    ///
+    /// This method is useful for the TopDocs collector.
+    /// For all docsets, the blanket implementation has the benefit
+    /// of prefiltering (doc, score) pairs, avoiding the
+    /// virtual dispatch cost.
+    ///
+    /// More importantly, it makes it possible for scorers to implement
+    /// important optimization (e.g. BlockWAND for union).
+    fn for_each_pruning(
+        &self,
+        threshold: Score,
+        reader: &SegmentReader,
+        callback: &mut dyn FnMut(DocId, Score) -> Score,
+    ) -> crate::Result<()> {
+        let scorer = self.complex_scorer::<SumWithCoordsCombiner>(reader, 1.0)?;
+        match scorer {
+            SpecializedScorer::TermUnion(term_scorers) => {
+                super::block_wand(term_scorers, threshold, callback);
+            }
+            SpecializedScorer::Other(mut scorer) => {
+                for_each_pruning_scorer(scorer.as_mut(), threshold, callback);
+            }
+        }
+        Ok(())
+    }
 }
 
 fn is_positive_occur(occur: Occur) -> bool {

src/query/boolean_query/mod.rs

@@ -1,13 +1,17 @@
+mod block_wand;
 mod boolean_query;
 mod boolean_weight;
 
+pub(crate) use self::block_wand::block_wand;
 pub use self::boolean_query::BooleanQuery;
 
 #[cfg(test)]
 mod tests {
 
     use super::*;
+    use crate::assert_nearly_equals;
     use crate::collector::tests::TEST_COLLECTOR_WITH_SCORE;
+    use crate::collector::TopDocs;
     use crate::query::score_combiner::SumWithCoordsCombiner;
     use crate::query::term_query::TermScorer;
     use crate::query::Intersection;
@@ -19,7 +23,7 @@ mod tests {
     use crate::query::TermQuery;
     use crate::schema::*;
     use crate::Index;
-    use crate::{DocAddress, DocId};
+    use crate::{DocAddress, DocId, Score};
 
     fn aux_test_helper() -> (Index, Field) {
         let mut schema_builder = Schema::builder();
@@ -30,24 +34,11 @@ mod tests {
             // writing the segment
             let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
             {
-                let doc = doc!(text_field => "a b c");
+                index_writer.add_document(doc!(text_field => "a b c"));
-                index_writer.add_document(doc);
+                index_writer.add_document(doc!(text_field => "a c"));
-            }
+                index_writer.add_document(doc!(text_field => "b c"));
-            {
+                index_writer.add_document(doc!(text_field => "a b c d"));
-                let doc = doc!(text_field => "a c");
+                index_writer.add_document(doc!(text_field => "d"));
-                index_writer.add_document(doc);
-            }
-            {
-                let doc = doc!(text_field => "b c");
-                index_writer.add_document(doc);
-            }
-            {
-                let doc = doc!(text_field => "a b c d");
-                index_writer.add_document(doc);
-            }
-            {
-                let doc = doc!(text_field => "d");
-                index_writer.add_document(doc);
             }
             assert!(index_writer.commit().is_ok());
         }
@@ -70,7 +61,7 @@ mod tests {
         let query = query_parser.parse_query("+a").unwrap();
         let searcher = index.reader().unwrap().searcher();
         let weight = query.weight(&searcher, true).unwrap();
-        let scorer = weight.scorer(searcher.segment_reader(0u32)).unwrap();
+        let scorer = weight.scorer(searcher.segment_reader(0u32), 1.0).unwrap();
         assert!(scorer.is::<TermScorer>());
     }
 
@@ -82,13 +73,13 @@ mod tests {
         {
             let query = query_parser.parse_query("+a +b +c").unwrap();
             let weight = query.weight(&searcher, true).unwrap();
-            let scorer = weight.scorer(searcher.segment_reader(0u32)).unwrap();
+            let scorer = weight.scorer(searcher.segment_reader(0u32), 1.0).unwrap();
             assert!(scorer.is::<Intersection<TermScorer>>());
         }
         {
             let query = query_parser.parse_query("+a +(b c)").unwrap();
             let weight = query.weight(&searcher, true).unwrap();
-            let scorer = weight.scorer(searcher.segment_reader(0u32)).unwrap();
+            let scorer = weight.scorer(searcher.segment_reader(0u32), 1.0).unwrap();
             assert!(scorer.is::<Intersection<Box<dyn Scorer>>>());
         }
     }
@@ -101,7 +92,7 @@ mod tests {
         {
             let query = query_parser.parse_query("+a b").unwrap();
             let weight = query.weight(&searcher, true).unwrap();
-            let scorer = weight.scorer(searcher.segment_reader(0u32)).unwrap();
+            let scorer = weight.scorer(searcher.segment_reader(0u32), 1.0).unwrap();
             assert!(scorer.is::<RequiredOptionalScorer<
                 Box<dyn Scorer>,
                 Box<dyn Scorer>,
@@ -111,7 +102,7 @@ mod tests {
         {
             let query = query_parser.parse_query("+a b").unwrap();
             let weight = query.weight(&searcher, false).unwrap();
-            let scorer = weight.scorer(searcher.segment_reader(0u32)).unwrap();
+            let scorer = weight.scorer(searcher.segment_reader(0u32), 1.0).unwrap();
             assert!(scorer.is::<TermScorer>());
         }
     }
@@ -142,7 +133,6 @@ mod tests {
                 .map(|doc| doc.1)
                 .collect::<Vec<DocId>>()
         };
-
         {
             let boolean_query = BooleanQuery::from(vec![(Occur::Must, make_term_query("a"))]);
             assert_eq!(matching_docs(&boolean_query), vec![0, 1, 3]);
@@ -179,6 +169,96 @@ mod tests {
         }
     }
 
+    #[test]
+    pub fn test_boolean_query_two_excluded() {
+        let (index, text_field) = aux_test_helper();
+
+        let make_term_query = |text: &str| {
+            let term_query = TermQuery::new(
+                Term::from_field_text(text_field, text),
+                IndexRecordOption::Basic,
+            );
+            let query: Box<dyn Query> = Box::new(term_query);
+            query
+        };
+
+        let reader = index.reader().unwrap();
+
+        let matching_topdocs = |query: &dyn Query| {
+            reader
+                .searcher()
+                .search(query, &TopDocs::with_limit(3))
+                .unwrap()
+        };
+
+        let score_doc_4: Score; // score of doc 4 should not be influenced by exclusion
+        {
+            let boolean_query_no_excluded =
+                BooleanQuery::from(vec![(Occur::Must, make_term_query("d"))]);
+            let topdocs_no_excluded = matching_topdocs(&boolean_query_no_excluded);
+            assert_eq!(topdocs_no_excluded.len(), 2);
+            let (top_score, top_doc) = topdocs_no_excluded[0];
+            assert_eq!(top_doc, DocAddress(0, 4));
+            assert_eq!(topdocs_no_excluded[1].1, DocAddress(0, 3)); // ignore score of doc 3.
+            score_doc_4 = top_score;
+        }
+
+        {
+            let boolean_query_two_excluded = BooleanQuery::from(vec![
+                (Occur::Must, make_term_query("d")),
+                (Occur::MustNot, make_term_query("a")),
+                (Occur::MustNot, make_term_query("b")),
+            ]);
+            let topdocs_excluded = matching_topdocs(&boolean_query_two_excluded);
+            assert_eq!(topdocs_excluded.len(), 1);
+            let (top_score, top_doc) = topdocs_excluded[0];
+            assert_eq!(top_doc, DocAddress(0, 4));
+            assert_eq!(top_score, score_doc_4);
+        }
+    }
+
+    #[test]
+    pub fn test_boolean_query_with_weight() {
+        let mut schema_builder = Schema::builder();
+        let text_field = schema_builder.add_text_field("text", TEXT);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        {
+            let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+            index_writer.add_document(doc!(text_field => "a b c"));
+            index_writer.add_document(doc!(text_field => "a c"));
+            index_writer.add_document(doc!(text_field => "b c"));
+            assert!(index_writer.commit().is_ok());
+        }
+        let term_a: Box<dyn Query> = Box::new(TermQuery::new(
+            Term::from_field_text(text_field, "a"),
+            IndexRecordOption::WithFreqs,
+        ));
+        let term_b: Box<dyn Query> = Box::new(TermQuery::new(
+            Term::from_field_text(text_field, "b"),
+            IndexRecordOption::WithFreqs,
+        ));
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+        let boolean_query =
+            BooleanQuery::from(vec![(Occur::Should, term_a), (Occur::Should, term_b)]);
+        let boolean_weight = boolean_query.weight(&searcher, true).unwrap();
+        {
+            let mut boolean_scorer = boolean_weight
+                .scorer(searcher.segment_reader(0u32), 1.0)
+                .unwrap();
+            assert_eq!(boolean_scorer.doc(), 0u32);
+            assert_nearly_equals!(boolean_scorer.score(), 0.84163445);
+        }
+        {
+            let mut boolean_scorer = boolean_weight
+                .scorer(searcher.segment_reader(0u32), 2.0)
+                .unwrap();
+            assert_eq!(boolean_scorer.doc(), 0u32);
+            assert_nearly_equals!(boolean_scorer.score(), 1.6832689);
+        }
+    }
+
     #[test]
     pub fn test_intersection_score() {
         let (index, text_field) = aux_test_helper();
@@ -205,168 +285,9 @@ mod tests {
                 (Occur::Must, make_term_query("a")),
                 (Occur::Must, make_term_query("b")),
             ]);
-            assert_eq!(score_docs(&boolean_query), vec![0.977973, 0.84699446]);
+            let scores = score_docs(&boolean_query);
+            assert_nearly_equals!(scores[0], 0.977973);
+            assert_nearly_equals!(scores[1], 0.84699446);
         }
     }
-
-    // motivated by #554
-    #[test]
-    fn test_bm25_several_fields() {
-        let mut schema_builder = Schema::builder();
-        let title = schema_builder.add_text_field("title", TEXT);
-        let text = schema_builder.add_text_field("text", TEXT);
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-
-        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-        index_writer.add_document(doc!(
-            // tf = 1 0
-            title =>  "Законы притяжения   Оксана Кулакова",
-            // tf = 1 0
-            text => "Законы притяжения Оксана Кулакова]  \n\nТема: Сексуальное искусство, Женственность\nТип товара: Запись вебинара (аудио)\nПродолжительность: 1,5 часа\n\nСсылка на вебинар:\n  ",
-        ));
-        index_writer.add_document(doc!(
-            // tf = 1 0
-            title => "Любимые русские пироги (Оксана Путан)",
-            // tf = 2 0
-            text => "http://i95.fastpic.ru/big/2017/0628/9a/615b9c8504d94a3893d7f496ac53539a.jpg \n\nОт издателя\nОксана Путан   профессиональный повар, автор кулинарных книг и известный кулинарный блогер. Ее рецепты отличаются практичностью, доступностью и пользуются огромной популярностью в русскоязычном интернете. Это третья книга автора о самом вкусном и ароматном   настоящих русских пирогах и выпечке!\nДаже новички на кухне легко готовят по ее рецептам. Оксана описывает процесс приготовления настолько подробно и понятно, что вам остается только наслаждаться готовкой и не тратить время на лишние усилия. Готовьте легко и просто!\n\nhttps://www.ozon.ru/context/detail/id/139872462/"
-        ));
-        index_writer.add_document(doc!(
-            // tf = 1 1
-            title =>  "PDF Мастер Класс \"Морячок\" (Оксана Лифенко)",
-            // tf = 0 0 
-            text => "https://i.ibb.co/pzvHrDN/I3d U T6 Gg TM.jpg\nhttps://i.ibb.co/NFrb6v6/N0ls Z9nwjb U.jpg\nВ описание входит штаны, кофта, берет, матросский воротник. Описание продается в формате PDF, состоит из 12 страниц формата А4 и может быть напечатано на любом принтере.\nОписание предназначено для кукол BJD RealPuki от FairyLand, но может подойти и другим подобным куклам. Также вы можете вязать этот наряд из обычной пряжи, и он подойдет для куколок побольше.\nhttps://vk.com/market 95724412?w=product 95724412_2212"
-        ));
-        for _ in 0..1_000 {
-            index_writer.add_document(doc!(
-                title =>  "a b d e f g",
-                text => "maitre corbeau sur un arbre perche tenait dans son bec un fromage Maitre rnard par lodeur alleche lui tint a peu pres ce langage."
-            ));
-        }
-        index_writer.commit().unwrap();
-        let reader = index.reader().unwrap();
-        let searcher = reader.searcher();
-        let query_parser = QueryParser::for_index(&index, vec![title, text]);
-        let query = query_parser.parse_query("Оксана Лифенко").unwrap();
-        let weight = query.weight(&searcher, true).unwrap();
-        let mut scorer = weight.scorer(searcher.segment_reader(0u32)).unwrap();
-        scorer.advance();
-
-        let explanation = query.explain(&searcher, DocAddress(0u32, 0u32)).unwrap();
-        assert_eq!(
-            explanation.to_pretty_json(),
-            r#"{
-  "value": 12.997711,
-  "description": "BooleanClause. Sum of ...",
-  "details": [
-    {
-      "value": 12.997711,
-      "description": "BooleanClause. Sum of ...",
-      "details": [
-        {
-          "value": 6.551476,
-          "description": "TermQuery, product of...",
-          "details": [
-            {
-              "value": 2.2,
-              "description": "(K1+1)"
-            },
-            {
-              "value": 5.658984,
-              "description": "idf, computed as log(1 + (N - n + 0.5) / (n + 0.5))",
-              "details": [
-                {
-                  "value": 3.0,
-                  "description": "n, number of docs containing this term"
-                },
-                {
-                  "value": 1003.0,
-                  "description": "N, total number of docs"
-                }
-              ]
-            },
-            {
-              "value": 0.5262329,
-              "description": "freq / (freq + k1 * (1 - b + b * dl / avgdl))",
-              "details": [
-                {
-                  "value": 1.0,
-                  "description": "freq, occurrences of term within document"
-                },
-                {
-                  "value": 1.2,
-                  "description": "k1, term saturation parameter"
-                },
-                {
-                  "value": 0.75,
-                  "description": "b, length normalization parameter"
-                },
-                {
-                  "value": 4.0,
-                  "description": "dl, length of field"
-                },
-                {
-                  "value": 5.997009,
-                  "description": "avgdl, average length of field"
-                }
-              ]
-            }
-          ]
-        },
-        {
-          "value": 6.446235,
-          "description": "TermQuery, product of...",
-          "details": [
-            {
-              "value": 2.2,
-              "description": "(K1+1)"
-            },
-            {
-              "value": 5.9954567,
-              "description": "idf, computed as log(1 + (N - n + 0.5) / (n + 0.5))",
-              "details": [
-                {
-                  "value": 2.0,
-                  "description": "n, number of docs containing this term"
-                },
-                {
-                  "value": 1003.0,
-                  "description": "N, total number of docs"
-                }
-              ]
-            },
-            {
-              "value": 0.4887212,
-              "description": "freq / (freq + k1 * (1 - b + b * dl / avgdl))",
-              "details": [
-                {
-                  "value": 1.0,
-                  "description": "freq, occurrences of term within document"
-                },
-                {
-                  "value": 1.2,
-                  "description": "k1, term saturation parameter"
-                },
-                {
-                  "value": 0.75,
-                  "description": "b, length normalization parameter"
-                },
-                {
-                  "value": 20.0,
-                  "description": "dl, length of field"
-                },
-                {
-                  "value": 24.123629,
-                  "description": "avgdl, average length of field"
-                }
-              ]
-            }
-          ]
-        }
-      ]
-    }
-  ]
-}"#
-        );
-    }
 }

src/query/boost_query.rs

@@ -0,0 +1,159 @@
+use crate::fastfield::DeleteBitSet;
+use crate::query::explanation::does_not_match;
+use crate::query::{Explanation, Query, Scorer, Weight};
+use crate::{DocId, DocSet, Score, Searcher, SegmentReader, Term};
+use std::collections::BTreeSet;
+use std::fmt;
+
+/// `BoostQuery` is a wrapper over a query used to boost its score.
+///
+/// The document set matched by the `BoostQuery` is strictly the same as the underlying query.
+/// The score of each document, is the score of the underlying query multiplied by the `boost`
+/// factor.
+pub struct BoostQuery {
+    query: Box<dyn Query>,
+    boost: Score,
+}
+
+impl BoostQuery {
+    /// Builds a boost query.
+    pub fn new(query: Box<dyn Query>, boost: Score) -> BoostQuery {
+        BoostQuery { query, boost }
+    }
+}
+
+impl Clone for BoostQuery {
+    fn clone(&self) -> Self {
+        BoostQuery {
+            query: self.query.box_clone(),
+            boost: self.boost,
+        }
+    }
+}
+
+impl fmt::Debug for BoostQuery {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        write!(f, "Boost(query={:?}, boost={})", self.query, self.boost)
+    }
+}
+
+impl Query for BoostQuery {
+    fn weight(&self, searcher: &Searcher, scoring_enabled: bool) -> crate::Result<Box<dyn Weight>> {
+        let weight_without_boost = self.query.weight(searcher, scoring_enabled)?;
+        let boosted_weight = if scoring_enabled {
+            Box::new(BoostWeight::new(weight_without_boost, self.boost))
+        } else {
+            weight_without_boost
+        };
+        Ok(boosted_weight)
+    }
+
+    fn query_terms(&self, term_set: &mut BTreeSet<Term>) {
+        self.query.query_terms(term_set)
+    }
+}
+
+pub(crate) struct BoostWeight {
+    weight: Box<dyn Weight>,
+    boost: Score,
+}
+
+impl BoostWeight {
+    pub fn new(weight: Box<dyn Weight>, boost: Score) -> Self {
+        BoostWeight { weight, boost }
+    }
+}
+
+impl Weight for BoostWeight {
+    fn scorer(&self, reader: &SegmentReader, boost: Score) -> crate::Result<Box<dyn Scorer>> {
+        self.weight.scorer(reader, boost * self.boost)
+    }
+
+    fn explain(&self, reader: &SegmentReader, doc: u32) -> crate::Result<Explanation> {
+        let mut scorer = self.scorer(reader, 1.0)?;
+        if scorer.seek(doc) != doc {
+            return Err(does_not_match(doc));
+        }
+        let mut explanation =
+            Explanation::new(format!("Boost x{} of ...", self.boost), scorer.score());
+        let underlying_explanation = self.weight.explain(reader, doc)?;
+        explanation.add_detail(underlying_explanation);
+        Ok(explanation)
+    }
+
+    fn count(&self, reader: &SegmentReader) -> crate::Result<u32> {
+        self.weight.count(reader)
+    }
+}
+
+pub(crate) struct BoostScorer<S: Scorer> {
+    underlying: S,
+    boost: Score,
+}
+
+impl<S: Scorer> BoostScorer<S> {
+    pub fn new(underlying: S, boost: Score) -> BoostScorer<S> {
+        BoostScorer { underlying, boost }
+    }
+}
+
+impl<S: Scorer> DocSet for BoostScorer<S> {
+    fn advance(&mut self) -> DocId {
+        self.underlying.advance()
+    }
+
+    fn seek(&mut self, target: DocId) -> DocId {
+        self.underlying.seek(target)
+    }
+
+    fn fill_buffer(&mut self, buffer: &mut [DocId]) -> usize {
+        self.underlying.fill_buffer(buffer)
+    }
+
+    fn doc(&self) -> u32 {
+        self.underlying.doc()
+    }
+
+    fn size_hint(&self) -> u32 {
+        self.underlying.size_hint()
+    }
+
+    fn count(&mut self, delete_bitset: &DeleteBitSet) -> u32 {
+        self.underlying.count(delete_bitset)
+    }
+
+    fn count_including_deleted(&mut self) -> u32 {
+        self.underlying.count_including_deleted()
+    }
+}
+
+impl<S: Scorer> Scorer for BoostScorer<S> {
+    fn score(&mut self) -> Score {
+        self.underlying.score() * self.boost
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::BoostQuery;
+    use crate::query::{AllQuery, Query};
+    use crate::schema::Schema;
+    use crate::{DocAddress, Document, Index};
+
+    #[test]
+    fn test_boost_query_explain() {
+        let schema = Schema::builder().build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        index_writer.add_document(Document::new());
+        assert!(index_writer.commit().is_ok());
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+        let query = BoostQuery::new(Box::new(AllQuery), 0.2);
+        let explanation = query.explain(&searcher, DocAddress(0, 0u32)).unwrap();
+        assert_eq!(
+            explanation.to_pretty_json(),
+            "{\n  \"value\": 0.2,\n  \"description\": \"Boost x0.2 of ...\",\n  \"details\": [\n    {\n      \"value\": 1.0,\n      \"description\": \"AllQuery\"\n    }\n  ]\n}"
+        )
+    }
+}

src/query/empty_query.rs

@@ -1,10 +1,10 @@
 use super::Scorer;
+use crate::docset::TERMINATED;
 use crate::query::explanation::does_not_match;
 use crate::query::Weight;
 use crate::query::{Explanation, Query};
 use crate::DocId;
 use crate::DocSet;
-use crate::Result;
 use crate::Score;
 use crate::Searcher;
 use crate::SegmentReader;
@@ -16,11 +16,15 @@ use crate::SegmentReader;
 pub struct EmptyQuery;
 
 impl Query for EmptyQuery {
-    fn weight(&self, _searcher: &Searcher, _scoring_enabled: bool) -> Result<Box<dyn Weight>> {
+    fn weight(
+        &self,
+        _searcher: &Searcher,
+        _scoring_enabled: bool,
+    ) -> crate::Result<Box<dyn Weight>> {
         Ok(Box::new(EmptyWeight))
     }
 
-    fn count(&self, _searcher: &Searcher) -> Result<usize> {
+    fn count(&self, _searcher: &Searcher) -> crate::Result<usize> {
         Ok(0)
     }
 }
@@ -30,11 +34,11 @@ impl Query for EmptyQuery {
 /// It is useful for tests and handling edge cases.
 pub struct EmptyWeight;
 impl Weight for EmptyWeight {
-    fn scorer(&self, _reader: &SegmentReader) -> Result<Box<dyn Scorer>> {
+    fn scorer(&self, _reader: &SegmentReader, _boost: Score) -> crate::Result<Box<dyn Scorer>> {
         Ok(Box::new(EmptyScorer))
     }
 
-    fn explain(&self, _reader: &SegmentReader, doc: DocId) -> Result<Explanation> {
+    fn explain(&self, _reader: &SegmentReader, doc: DocId) -> crate::Result<Explanation> {
         Err(does_not_match(doc))
     }
 }
@@ -45,15 +49,12 @@ impl Weight for EmptyWeight {
 pub struct EmptyScorer;
 
 impl DocSet for EmptyScorer {
-    fn advance(&mut self) -> bool {
+    fn advance(&mut self) -> DocId {
-        false
+        TERMINATED
     }
 
     fn doc(&self) -> DocId {
-        panic!(
+        TERMINATED
-            "You may not call .doc() on a scorer \
-             where the last call to advance() did not return true."
-        );
     }
 
     fn size_hint(&self) -> u32 {
@@ -63,24 +64,21 @@ impl DocSet for EmptyScorer {
 
 impl Scorer for EmptyScorer {
     fn score(&mut self) -> Score {
-        0f32
+        0.0
     }
 }
 
 #[cfg(test)]
 mod tests {
+    use crate::docset::TERMINATED;
     use crate::query::EmptyScorer;
     use crate::DocSet;
 
     #[test]
     fn test_empty_scorer() {
         let mut empty_scorer = EmptyScorer;
-        assert!(!empty_scorer.advance());
+        assert_eq!(empty_scorer.doc(), TERMINATED);
-    }
+        assert_eq!(empty_scorer.advance(), TERMINATED);
-
+        assert_eq!(empty_scorer.doc(), TERMINATED);
-    #[test]
-    #[should_panic]
-    fn test_empty_scorer_panic_on_doc_call() {
-        EmptyScorer.doc();
     }
 }