Bumped version

* Assert nearly equals macro

* Renamed specialized_scorer in TermScorer

.gitignore

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

CHANGELOG.md

@@ -1,3 +1,32 @@
+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 `Scorer::skip(target)` has been replaced by `Scorer::seek(target)` and returns the resulting DocId.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)

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "tantivy"
-version = "0.12.0"
+version = "0.13.0"
 authors = ["Paul Masurel <paul.masurel@gmail.com>"]
 license = "MIT"
 categories = ["database-implementations", "data-structures"]
@@ -18,7 +18,7 @@ byteorder = "1.0"
 crc32fast = "1.2.0"
 once_cell = "1.0"
 regex ={version = "1.3.0", default-features = false, features = ["std"]}
-tantivy-fst = "0.2.1"
+tantivy-fst = "0.3"
 memmap = {version = "0.7", optional=true}
 lz4 = {version="1.20", optional=true}
 snap = "1"
@@ -29,7 +29,7 @@ serde = {version="1.0", features=["derive"]}
 serde_json = "1.0"
 num_cpus = "1.2"
 fs2={version="0.4", optional=true}
-levenshtein_automata = "0.1"
+levenshtein_automata = "0.2"
 notify = {version="4", optional=true}
 uuid = { version = "0.8", features = ["v4", "serde"] }
 crossbeam = "0.7"
@@ -38,14 +38,14 @@ owning_ref = "0.4"
 stable_deref_trait = "1.0.0"
 rust-stemmers = "1.2"
 downcast-rs = { version="1.0" }
-tantivy-query-grammar = { version="0.12", path="./query-grammar" }
+tantivy-query-grammar = { version="0.13", path="./query-grammar" }
 bitpacking = {version="0.8", default-features = false, features=["bitpacker4x"]}
-census =  {path="../census"}
+census = "0.4"
 fnv = "1.0.6"
 owned-read = "0.4"
 failure = "0.1"
 htmlescape = "0.3.1"
-fail = "0.3"
+fail = "0.4"
 murmurhash32 = "0.2"
 chrono = "0.4"
 smallvec = "1.0"
@@ -58,9 +58,10 @@ winapi = "0.3"
 rand = "0.7"
 maplit = "1"
 matches = "0.1.8"
+proptest = "0.10"
 
 [dev-dependencies.fail]
-version = "0.3"
+version = "0.4"
 features = ["failpoints"]
 
 [profile.release]
@@ -79,6 +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

@@ -31,12 +31,16 @@ 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

TODO.md

@@ -1,9 +0,0 @@
-- segment writer serializes if already flush
-- flush if exceed max doc. 
-- unit tests
-- reader from `IndexWriter`
-- configurable limit to flush 
-- bundle directory
-- add index worker when exceeds some doc limit
-- flush before prepare commit.
-- segment_writer should not receive a segment 

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_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 mut 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() {
-                // the number of time the term appears in the document.
-                let doc_id: DocId = segment_postings.doc(); //< do not try to access this before calling advance once.
-
+            let mut doc_id = segment_postings.doc();
+            while doc_id != TERMINATED {
                 // This MAY contains deleted documents as well.
                 if segment_reader.is_deleted(doc_id) {
+                    doc_id = segment_postings.advance();
                     continue;
                 }
 
@@ -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();
             }
         }
     }

query-grammar/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "tantivy-query-grammar"
-version = "0.12.0"
+version = "0.13.0"
 authors = ["Paul Masurel <paul.masurel@gmail.com>"]
 license = "MIT"
 categories = ["database-implementations", "data-structures"]

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 {
-            Occur::Should => right,
-            Occur::Must => {
-                if right == Occur::MustNot {
-                    Occur::MustNot
-                } else {
-                    Occur::Must
-                }
-            }
-            Occur::MustNot => {
-                if right == Occur::MustNot {
-                    Occur::Must
-                } else {
-                    Occur::MustNot
-                }
-            }
+        match (left, right) {
+            (Occur::Should, _) => right,
+            (Occur::Must, Occur::MustNot) => Occur::MustNot,
+            (Occur::Must, _) => Occur::Must,
+            (Occur::MustNot, Occur::MustNot) => Occur::Must,
+            (Occur::MustNot, _) => Occur::MustNot,
         }
     }
 }
@@ -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

@@ -154,17 +154,11 @@ fn negate(expr: UserInputAST) -> UserInputAST {
     expr.unary(Occur::MustNot)
 }
 
-fn must(expr: UserInputAST) -> UserInputAST {
-    expr.unary(Occur::Must)
-}
-
 fn leaf<'a>() -> impl Parser<&'a str, Output = UserInputAST> {
     parser(|input| {
-        char('-')
-            .with(leaf())
-            .map(negate)
-            .or(char('+').with(leaf()).map(must))
-            .or(char('(').with(ast()).skip(char(')')))
+        char('(')
+            .with(ast())
+            .skip(char(')'))
             .or(char('*').map(|_| UserInputAST::from(UserInputLeaf::All)))
             .or(attempt(
                 string("NOT").skip(spaces1()).with(leaf()).map(negate),
@@ -176,7 +170,17 @@ fn leaf<'a>() -> impl Parser<&'a str, Output = UserInputAST> {
     })
 }
 
-fn positive_float_number<'a>() -> impl Parser<&'a str, Output = f32> {
+fn occur_symbol<'a>() -> impl Parser<&'a str, Output = Occur> {
+    char('-')
+        .map(|_| Occur::MustNot)
+        .or(char('+').map(|_| Occur::Must))
+}
+
+fn occur_leaf<'a>() -> impl Parser<&'a str, Output = (Option<Occur>, UserInputAST)> {
+    (optional(occur_symbol()), boosted_leaf())
+}
+
+fn positive_float_number<'a>() -> impl Parser<&'a str, Output = f64> {
     (many1(digit()), optional((char('.'), many1(digit())))).map(
         |(int_part, decimal_part_opt): (String, Option<(char, String)>)| {
             let mut float_str = int_part;
@@ -184,18 +188,18 @@ fn positive_float_number<'a>() -> impl Parser<&'a str, Output = f32> {
                 float_str.push(chr);
                 float_str.push_str(&decimal_str);
             }
-            float_str.parse::<f32>().unwrap()
+            float_str.parse::<f64>().unwrap()
         },
     )
 }
 
-fn boost<'a>() -> impl Parser<&'a str, Output = f32> {
+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::f32::EPSILON => {
+        Some(boost) if (boost - 1.0).abs() > std::f64::EPSILON => {
             UserInputAST::Boost(Box::new(leaf), boost)
         }
         _ => leaf,
@@ -239,21 +243,29 @@ fn aggregate_binary_expressions(
     }
 }
 
-pub fn ast<'a>() -> impl Parser<&'a str, Output = UserInputAST> {
-    let operand_leaf = (
+fn operand_leaf<'a>() -> impl Parser<&'a str, Output = (BinaryOperand, UserInputAST)> {
+    (
         binary_operand().skip(spaces()),
         boosted_leaf().skip(spaces()),
-    );
-    let boolean_expr = (boosted_leaf().skip(spaces().silent()), many1(operand_leaf))
+    )
+}
+
+pub fn ast<'a>() -> impl Parser<&'a str, Output = UserInputAST> {
+    let boolean_expr = (boosted_leaf().skip(spaces()), many1(operand_leaf()))
         .map(|(left, right)| aggregate_binary_expressions(left, right));
-    let whitespace_separated_leaves =
-        many1(boosted_leaf().skip(spaces().silent())).map(|subqueries: Vec<UserInputAST>| {
+    let whitespace_separated_leaves = many1(occur_leaf().skip(spaces().silent())).map(
+        |subqueries: Vec<(Option<Occur>, UserInputAST)>| {
             if subqueries.len() == 1 {
-                subqueries.into_iter().next().unwrap()
+                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())
 }
@@ -270,11 +282,11 @@ mod test {
     use super::*;
     use combine::parser::Parser;
 
-    pub fn nearly_equals(a: f32, b: f32) -> bool {
+    pub fn nearly_equals(a: f64, b: f64) -> bool {
         (a - b).abs() < 0.0005 * (a + b).abs()
     }
 
-    fn assert_nearly_equals(expected: f32, val: f32) {
+    fn assert_nearly_equals(expected: f64, val: f64) {
         assert!(
             nearly_equals(val, expected),
             "Got {}, expected {}.",
@@ -283,9 +295,15 @@ mod test {
         );
     }
 
+    #[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: f32, expected_remaining: &str) {
+        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);
@@ -293,9 +311,9 @@ mod test {
         fn error_parse(float_str: &str) {
             assert!(positive_float_number().parse(float_str).is_err());
         }
-        valid_parse("1.0", 1.0f32, "");
-        valid_parse("1", 1.0f32, "");
-        valid_parse("0.234234 aaa", 0.234234f32, " aaa");
+        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.");
@@ -330,7 +348,7 @@ 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]
@@ -338,16 +356,16 @@ mod test {
         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^3 b^2", "(*(\"a\")^3 *(\"b\")^2)");
         test_parse_query_to_ast_helper("a^1", "\"a\"");
     }
 
     #[test]
     fn test_parse_query_to_ast_binary_op() {
-        test_parse_query_to_ast_helper("a AND b", "(+(\"a\") +(\"b\"))");
-        test_parse_query_to_ast_helper("a OR b", "(?(\"a\") ?(\"b\"))");
-        test_parse_query_to_ast_helper("a OR b AND c", "(?(\"a\") ?((+(\"b\") +(\"c\"))))");
-        test_parse_query_to_ast_helper("a AND b         AND c", "(+(\"a\") +(\"b\") +(\"c\"))");
+        test_parse_query_to_ast_helper("a AND b", "(+\"a\" +\"b\")");
+        test_parse_query_to_ast_helper("a OR b", "(?\"a\" ?\"b\")");
+        test_parse_query_to_ast_helper("a OR b AND c", "(?\"a\" ?(+\"b\" +\"c\"))");
+        test_parse_query_to_ast_helper("a AND b         AND c", "(+\"a\" +\"b\" +\"c\")");
         assert_eq!(
             format!("{:?}", parse_to_ast().parse("a OR b aaa")),
             "Err(UnexpectedParse)"
@@ -385,6 +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
@@ -413,32 +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_parse_query_to_ast_helper("+(a b) +d", "(+((\"a\" \"b\")) +(\"d\"))");
-        test_parse_query_to_ast_helper("(+a +b) d", "((+(\"a\") +(\"b\")) \"d\")");
-        test_parse_query_to_ast_helper("(+a)", "+(\"a\")");
-        test_parse_query_to_ast_helper("(+a +b)", "(+(\"a\") +(\"b\"))");
+    }
+
+    #[test]
+    fn test_parse_query_default_clause() {
+        test_parse_query_to_ast_helper("a b", "(*\"a\" *\"b\")");
+    }
+
+    #[test]
+    fn test_parse_query_must_default_clause() {
+        test_parse_query_to_ast_helper("+(a b)", "(*\"a\" *\"b\")");
+    }
+
+    #[test]
+    fn test_parse_query_must_single_term() {
+        test_parse_query_to_ast_helper("+d", "\"d\"");
+    }
+
+    #[test]
+    fn test_single_term_with_field() {
         test_parse_query_to_ast_helper("abc:toto", "abc:\"toto\"");
+    }
+
+    #[test]
+    fn test_single_term_with_float() {
         test_parse_query_to_ast_helper("abc:1.1", "abc:\"1.1\"");
-        test_parse_query_to_ast_helper("+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]
+    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

@@ -85,15 +85,14 @@ impl UserInputBound {
 }
 
 pub enum UserInputAST {
-    Clause(Vec<UserInputAST>),
-    Unary(Occur, Box<UserInputAST>),
+    Clause(Vec<(Option<Occur>, UserInputAST)>),
     Leaf(Box<UserInputLeaf>),
-    Boost(Box<UserInputAST>, f32),
+    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 {
@@ -104,7 +103,7 @@ impl UserInputAST {
         } else {
             UserInputAST::Clause(
                 asts.into_iter()
-                    .map(|ast: UserInputAST| ast.unary(occur))
+                    .map(|ast: UserInputAST| (Some(occur), ast))
                     .collect::<Vec<_>>(),
             )
         }
@@ -135,25 +134,36 @@ 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

@@ -96,18 +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(1u32, 1f32);
+            count_collector.collect(0u32, 1.0);
+            count_collector.collect(1u32, 1.0);
             assert_eq!(count_collector.harvest(), 2);
         }
     }

src/collector/custom_score_top_collector.rs

@@ -11,13 +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,
         }
     }
 }
@@ -28,7 +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.
@@ -117,9 +117,9 @@ where
 
 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

@@ -1,6 +1,5 @@
 use crate::collector::Collector;
 use crate::collector::SegmentCollector;
-use crate::docset::SkipResult;
 use crate::fastfield::FacetReader;
 use crate::schema::Facet;
 use crate::schema::Field;
@@ -188,6 +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>,
@@ -197,14 +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();
@@ -281,7 +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;
@@ -301,7 +305,7 @@ impl Collector for FacetCollector {
                             }
                             break;
                         }
-                        SkipResult::End | SkipResult::OverStep => {
+                        SkipResult::NotFound => {
                             collapse_mapping.push(0);
                             if !facet_streamer.advance() {
                                 break;

src/collector/mod.rs

@@ -109,6 +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.
@@ -154,6 +155,29 @@ pub trait Collector: Sync {
     /// Combines the fruit associated to the collection of each segments
     /// into one 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

src/collector/multi_collector.rs

@@ -55,7 +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);
     }
 
@@ -65,7 +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>;
 }
 
@@ -74,7 +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);
     }
 

src/collector/tests.rs

@@ -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

@@ -18,9 +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> {
-    feature: T,
-    doc: D,
+pub(crate) struct ComparableDoc<T, D> {
+    pub feature: T,
+    pub doc: D,
 }
 
 impl<T: PartialOrd, D: PartialOrd> PartialOrd for ComparableDoc<T, D> {
@@ -56,7 +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>,
 }
 
@@ -72,14 +73,20 @@ where
         if limit < 1 {
             panic!("Limit must be strictly greater than 0.");
         }
-        TopCollector {
+        Self {
             limit,
+            offset: 0,
             _marker: PhantomData,
         }
     }
 
-    pub fn limit(&self) -> usize {
-        self.limit
+    /// Skip the first "offset" documents when collecting.
+    ///
+    /// 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(
@@ -92,7 +99,7 @@ where
         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 {
@@ -104,6 +111,7 @@ where
         Ok(top_collector
             .into_sorted_vec()
             .into_iter()
+            .skip(self.offset)
             .map(|cdoc| (cdoc.feature, cdoc.doc))
             .collect())
     }
@@ -113,7 +121,23 @@ where
         segment_id: SegmentLocalId,
         _: &SegmentReader,
     ) -> 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,
+        }
     }
 }
 
@@ -187,7 +211,7 @@ impl<T: PartialOrd + Clone> TopSegmentCollector<T> {
 
 #[cfg(test)]
 mod tests {
-    use super::TopSegmentCollector;
+    use super::{TopCollector, TopSegmentCollector};
     use crate::DocAddress;
 
     #[test]
@@ -248,6 +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::Score;
 use crate::SegmentLocalId;
 use crate::SegmentReader;
+use std::collections::BinaryHeap;
 use std::fmt;
 
 /// The `TopDocs` collector keeps track of the top `K` documents
@@ -50,14 +52,18 @@ use std::fmt;
 /// let query = query_parser.parse_query("diary").unwrap();
 /// let top_docs = searcher.search(&query, &TopDocs::with_limit(2)).unwrap();
 ///
-/// assert_eq!(&top_docs[0], &(0.7261542, DocAddress(0, 1)));
-/// assert_eq!(&top_docs[1], &(0.6099695, DocAddress(0, 3)));
+/// 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
+        )
     }
 }
 
@@ -66,7 +72,7 @@ struct ScorerByFastFieldReader {
 }
 
 impl CustomSegmentScorer<u64> for ScorerByFastFieldReader {
-    fn score(&self, doc: DocId) -> u64 {
+    fn score(&mut self, doc: DocId) -> u64 {
         self.ff_reader.get_u64(u64::from(doc))
     }
 }
@@ -101,6 +107,45 @@ 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
@@ -258,7 +303,7 @@ 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
     ///             }
     ///           });
@@ -281,7 +326,7 @@ impl TopDocs {
         TScoreSegmentTweaker: ScoreSegmentTweaker<TScore> + 'static,
         TScoreTweaker: ScoreTweaker<TScore, Child = TScoreSegmentTweaker>,
     {
-        TweakedScoreTopCollector::new(score_tweaker, self.0.limit())
+        TweakedScoreTopCollector::new(score_tweaker, self.0.into_tscore())
     }
 
     /// Ranks the documents using a custom score.
@@ -395,7 +440,7 @@ impl TopDocs {
         TCustomSegmentScorer: CustomSegmentScorer<TScore> + 'static,
         TCustomScorer: CustomScorer<TScore, Child = TCustomSegmentScorer>,
     {
-        CustomScoreTopCollector::new(custom_score, self.0.limit())
+        CustomScoreTopCollector::new(custom_score, self.0.into_tscore())
     }
 }
 
@@ -423,6 +468,64 @@ impl Collector for TopDocs {
     ) -> 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`.
@@ -432,7 +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)> {
@@ -446,10 +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 crate::{DocAddress, DocId, SegmentReader};
 
     fn make_index() -> Index {
         let mut schema_builder = Schema::builder();
@@ -467,6 +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();
@@ -479,16 +589,31 @@ mod tests {
             .searcher()
             .search(&text_query, &TopDocs::with_limit(4))
             .unwrap();
-        assert_eq!(
-            score_docs,
-            vec![
+        assert_results_equals(
+            &score_docs,
+            &[
                 (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();
@@ -501,12 +626,33 @@ mod tests {
             .searcher()
             .search(&text_query, &TopDocs::with_limit(2))
             .unwrap();
-        assert_eq!(
-            score_docs,
-            vec![
+        assert_results_equals(
+            &score_docs,
+            &[
                 (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)),
+            ],
         );
     }
 
@@ -567,8 +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,
-            vec![
+            &top_docs[..],
+            &[
                 (64, DocAddress(0, 1)),
                 (16, DocAddress(0, 2)),
                 (12, DocAddress(0, 0))
@@ -623,6 +769,50 @@ mod tests {
         }
     }
 
+    #[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(
         query: &str,
         query_field: Field,

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
@@ -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
-    // count multiple inserts.
-    // `usize` guards us from overflow.
+    len: usize,
     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/composite_file.rs

@@ -190,12 +190,11 @@ mod test {
     use crate::schema::Field;
     use std::io::Write;
     use std::path::Path;
-    
 
     #[test]
     fn test_composite_file() {
         let path = Path::new("test_path");
-        let mut directory = RAMDirectory::default();
+        let mut directory = RAMDirectory::create();
         {
             let w = directory.open_write(path).unwrap();
             let mut composite_write = CompositeWrite::wrap(w);

src/common/mod.rs

@@ -2,7 +2,6 @@ pub mod bitpacker;
 mod bitset;
 mod composite_file;
 mod counting_writer;
-mod mutable_enum;
 mod serialize;
 mod vint;
 
@@ -10,9 +9,10 @@ pub use self::bitset::BitSet;
 pub(crate) use self::bitset::TinySet;
 pub(crate) use self::composite_file::{CompositeFile, CompositeWrite};
 pub use self::counting_writer::CountingWriter;
-pub(crate) use self::mutable_enum::MutableEnum;
 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`.

src/common/mutable_enum.rs

@@ -1,37 +0,0 @@
-use std::ops::{Deref, DerefMut};
-
-pub(crate) struct MutableEnum<T>(Option<T>);
-
-impl<T> MutableEnum<T> {
-    pub fn wrap(val: T) -> Self {
-        MutableEnum(Some(val))
-    }
-
-    pub fn into(self) -> T {
-        self.0.unwrap()
-    }
-}
-
-impl<T> MutableEnum<T> {
-    pub fn map_mutate<E, F>(&mut self, transformation: F) -> Result<(), E>
-    where
-        F: FnOnce(T) -> Result<T, E>,
-    {
-        self.0 = self.0.take().map(transformation).transpose()?;
-        Ok(())
-    }
-}
-
-impl<T> Deref for MutableEnum<T> {
-    type Target = T;
-
-    fn deref(&self) -> &Self::Target {
-        self.0.as_ref().unwrap()
-    }
-}
-
-impl<T> DerefMut for MutableEnum<T> {
-    fn deref_mut(&mut self) -> &mut Self::Target {
-        self.0.as_mut().unwrap()
-    }
-}

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
+    (result, vlen)
 }
-
 /// Write a `u32` as a vint payload.
 pub fn write_u32_vint<W: io::Write>(val: u32, writer: &mut W) -> io::Result<()> {
-    let (val, num_bytes) = serialize_vint_u32(val);
-    let mut buffer = [0u8; 8];
-    LittleEndian::write_u64(&mut buffer, val);
-    writer.write_all(&buffer[..num_bytes])
+    let mut buf = [0u8; 8];
+    let data = serialize_vint_u32(val, &mut buf);
+    writer.write_all(&data)
 }
 
 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);
-        assert_eq!(len, len_vint, "len wrong for val {}", val);
-        LittleEndian::write_u64(&mut buffer2, vint);
-        assert_eq!(&buffer[..len], &buffer2[..len], "array wrong for {}", val);
+        let res2 = serialize_vint_u32(val, &mut buffer2);
+        assert_eq!(&buffer[..len_vint], res2, "array wrong for {}", val);
     }
 
     #[test]

src/core/index.rs

@@ -12,8 +12,8 @@ use crate::directory::INDEX_WRITER_LOCK;
 use crate::directory::{Directory, RAMDirectory};
 use crate::error::DataCorruption;
 use crate::error::TantivyError;
+use crate::indexer::index_writer::HEAP_SIZE_MIN;
 use crate::indexer::segment_updater::save_new_metas;
-use crate::indexer::IndexWriterConfig;
 use crate::reader::IndexReader;
 use crate::reader::IndexReaderBuilder;
 use crate::schema::Field;
@@ -21,12 +21,13 @@ use crate::schema::FieldType;
 use crate::schema::Schema;
 use crate::tokenizer::{TextAnalyzer, TokenizerManager};
 use crate::IndexWriter;
-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(
@@ -273,40 +274,8 @@ impl Index {
     pub fn writer_with_num_threads(
         &self,
         num_threads: usize,
-        overall_heap_size_in_bytes: u64,
+        overall_heap_size_in_bytes: usize,
     ) -> crate::Result<IndexWriter> {
-        let config = IndexWriterConfig {
-            max_indexing_threads: num_threads,
-            memory_budget: overall_heap_size_in_bytes,
-            ..Default::default()
-        };
-        self.writer_from_config(config)
-    }
-
-    /// Creates a multithreaded writer
-    ///
-    /// Tantivy will automatically define the number of threads to use.
-    /// `overall_heap_size_in_bytes` is the total target memory usage that will be split
-    /// between a given number of threads.
-    ///
-    /// # Errors
-    /// 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: u64) -> crate::Result<IndexWriter> {
-        let config = IndexWriterConfig {
-            max_indexing_threads: num_cpus::get(),
-            memory_budget: overall_heap_size_in_bytes,
-            ..Default::default()
-        };
-        self.writer_from_config(config)
-    }
-
-    /// Creates a new writer with a given configuration.
-    ///
-    /// See [`IndexWriterConfig`](./struct.IndexWriterConfig.html) for more information.
-    pub fn writer_from_config(&self, mut config: IndexWriterConfig) -> crate::Result<IndexWriter> {
-        config.validate()?;
         let directory_lock = self
             .directory
             .acquire_lock(&INDEX_WRITER_LOCK)
@@ -322,7 +291,32 @@ impl Index {
                     ),
                 )
             })?;
-        IndexWriter::new(self, config, directory_lock)
+        let heap_size_in_bytes_per_thread = overall_heap_size_in_bytes / num_threads;
+        IndexWriter::new(
+            self,
+            num_threads,
+            heap_size_in_bytes_per_thread,
+            directory_lock,
+        )
+    }
+
+    /// Creates a multithreaded writer
+    ///
+    /// Tantivy will automatically define the number of threads to use.
+    /// `overall_heap_size_in_bytes` is the total target memory usage that will be split
+    /// between a given number of threads.
+    ///
+    /// # Errors
+    /// 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) -> 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 {
+            num_threads = (overall_heap_size_in_bytes / HEAP_SIZE_MIN).max(1);
+        }
+        self.writer_with_num_threads(num_threads, overall_heap_size_in_bytes)
     }
 
     /// Accessor to the index schema
@@ -476,7 +470,7 @@ mod tests {
             .try_into()
             .unwrap();
         assert_eq!(reader.searcher().num_docs(), 0);
-        test_index_on_commit_reload_policy_aux(field, index.clone(), &index, &reader);
+        test_index_on_commit_reload_policy_aux(field, &index, &reader);
     }
 
     #[cfg(feature = "mmap")]
@@ -500,7 +494,7 @@ mod tests {
                 .try_into()
                 .unwrap();
             assert_eq!(reader.searcher().num_docs(), 0);
-            test_index_on_commit_reload_policy_aux(field, index.clone(), &index, &reader);
+            test_index_on_commit_reload_policy_aux(field, &index, &reader);
         }
 
         #[test]
@@ -542,16 +536,12 @@ mod tests {
                 .try_into()
                 .unwrap();
             assert_eq!(reader.searcher().num_docs(), 0);
-            test_index_on_commit_reload_policy_aux(field, read_index, &write_index, &reader);
+            test_index_on_commit_reload_policy_aux(field, &write_index, &reader);
         }
     }
 
-    fn test_index_on_commit_reload_policy_aux(
-        field: Field,
-        mut reader_index: Index,
-        index: &Index,
-        reader: &IndexReader,
-    ) {
+    fn test_index_on_commit_reload_policy_aux(field: Field, index: &Index, reader: &IndexReader) {
+        let mut reader_index = reader.index();
         let (sender, receiver) = crossbeam::channel::unbounded();
         let _watch_handle = reader_index.directory_mut().watch(Box::new(move || {
             let _ = sender.send(());

src/core/index_meta.rs

@@ -3,9 +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;
@@ -215,7 +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>,

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.
@@ -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_reader);
+        block_postings.reset(term_info.doc_freq, postings_slice);
     }
 
     /// 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};
@@ -17,26 +14,6 @@ use crate::Index;
 use std::fmt;
 use std::sync::Arc;
 
-fn collect_segment<C: Collector>(
-    collector: &C,
-    weight: &dyn Weight,
-    segment_ord: u32,
-    segment_reader: &SegmentReader,
-) -> crate::Result<C::Fruit> {
-    let mut scorer = weight.scorer(segment_reader, 1.0f32)?;
-    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
@@ -163,12 +140,7 @@ impl Searcher {
         let segment_readers = self.segment_readers();
         let fruits = executor.map(
             |(segment_ord, segment_reader)| {
-                collect_segment(
-                    collector,
-                    weight.as_ref(),
-                    segment_ord as u32,
-                    segment_reader,
-                )
+                collector.collect_segment(weight.as_ref(), segment_ord as u32, segment_reader)
             },
             segment_readers.iter().enumerate(),
         )?;

src/core/segment.rs

@@ -3,140 +3,41 @@ use crate::core::Index;
 use crate::core::SegmentId;
 use crate::core::SegmentMeta;
 use crate::directory::error::{OpenReadError, OpenWriteError};
-use crate::directory::{Directory, ManagedDirectory, RAMDirectory};
+use crate::directory::Directory;
 use crate::directory::{ReadOnlySource, WritePtr};
 use crate::indexer::segment_serializer::SegmentSerializer;
 use crate::schema::Schema;
 use crate::Opstamp;
 use std::fmt;
-use std::ops::{Deref, DerefMut};
 use std::path::PathBuf;
-use crate::indexer::{ResourceManager};
-
-#[derive(Clone)]
-pub(crate) enum SegmentDirectory {
-    Persisted(ManagedDirectory),
-    Volatile {
-        directory: RAMDirectory,
-    },
-}
-
-impl SegmentDirectory {
-    fn new_volatile(memory_manager: ResourceManager) -> SegmentDirectory {
-        SegmentDirectory::Volatile {
-            directory: RAMDirectory::create_with_memory_manager(memory_manager),
-        }
-    }
-}
-
-impl fmt::Debug for SegmentDirectory {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        match self {
-            SegmentDirectory::Volatile { .. } => write!(f, "volatile")?,
-            SegmentDirectory::Persisted(dir) => write!(f, "Persisted({:?})", dir)?,
-        }
-        Ok(())
-    }
-}
-
-impl From<ManagedDirectory> for SegmentDirectory {
-    fn from(directory: ManagedDirectory) -> Self {
-        SegmentDirectory::Persisted(directory)
-    }
-}
-
-impl Deref for SegmentDirectory {
-    type Target = dyn Directory;
-
-    fn deref(&self) -> &Self::Target {
-        match self {
-            SegmentDirectory::Volatile {
-                directory, ..
-            } => directory,
-            SegmentDirectory::Persisted(dir) => dir,
-        }
-    }
-}
-
-impl DerefMut for SegmentDirectory {
-    fn deref_mut(&mut self) -> &mut Self::Target {
-        match self {
-            SegmentDirectory::Volatile { directory, .. } => directory,
-            SegmentDirectory::Persisted(dir) => dir,
-        }
-    }
-}
 
 /// A segment is a piece of the index.
 #[derive(Clone)]
 pub struct Segment {
-    schema: Schema,
+    index: Index,
     meta: SegmentMeta,
-    directory: SegmentDirectory,
 }
 
 impl fmt::Debug for Segment {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        write!(
-            f,
-            "Segment(id={:?}, directory={:?})",
-            self.id().uuid_string(),
-            self.directory
-        )
+        write!(f, "Segment({:?})", self.id().uuid_string())
     }
 }
 
 impl Segment {
-    pub(crate) fn new_persisted(
-        meta: SegmentMeta,
-        directory: ManagedDirectory,
-        schema: Schema,
-    ) -> Segment {
-        Segment {
-            meta,
-            schema,
-            directory: SegmentDirectory::from(directory),
-        }
-    }
-
-    /// Creates a new segment that embeds its own `RAMDirectory`.
-    ///
-    /// That segment is entirely dissociated from the index directory.
-    /// It will be persisted by a background thread in charge of IO.
-    pub fn new_volatile(meta: SegmentMeta, schema: Schema, resource_manager: ResourceManager) -> Segment {
-        Segment {
-            schema,
-            meta,
-            directory: SegmentDirectory::new_volatile(resource_manager),
-        }
-    }
-
     /// Creates a new segment given an `Index` and a `SegmentId`
     pub(crate) fn for_index(index: Index, meta: SegmentMeta) -> Segment {
-        let segment_directory = index.directory().clone();
-        Segment::new_persisted(meta, segment_directory, index.schema())
+        Segment { index, meta }
     }
 
-    /// Persists a given `Segment` to a directory.
-    pub fn persist(&mut self, mut dest_directory: ManagedDirectory) -> crate::Result<()> {
-        if let SegmentDirectory::Persisted(_) = self.directory {
-            // this segment is already persisted.
-            return Ok(());
-        }
-        if let SegmentDirectory::Volatile { directory, ..} = &self.directory {
-            directory.persist(&mut dest_directory)?;
-        }
-        self.directory = SegmentDirectory::Persisted(dest_directory);
-        Ok(())
-    }
-
-    pub fn into_volatile(&self, memory_manager: ResourceManager) -> Segment {
-        Segment::new_volatile(self.meta.clone(), self.schema.clone(), memory_manager)
+    /// Returns the index the segment belongs to.
+    pub fn index(&self) -> &Index {
+        &self.index
     }
 
     /// Returns our index's schema.
     pub fn schema(&self) -> Schema {
-        self.schema.clone()
+        self.index.schema()
     }
 
     /// Returns the segment meta-information
@@ -150,18 +51,16 @@ impl Segment {
     /// as we finalize a fresh new segment.
     pub(crate) fn with_max_doc(self, max_doc: u32) -> Segment {
         Segment {
-            schema: self.schema,
+            index: self.index,
             meta: self.meta.with_max_doc(max_doc),
-            directory: self.directory,
         }
     }
 
     #[doc(hidden)]
     pub fn with_delete_meta(self, num_deleted_docs: u32, opstamp: Opstamp) -> Segment {
         Segment {
-            schema: self.schema,
+            index: self.index,
             meta: self.meta.with_delete_meta(num_deleted_docs, opstamp),
-            directory: self.directory,
         }
     }
 
@@ -181,15 +80,15 @@ impl Segment {
     /// Open one of the component file for a *regular* read.
     pub fn open_read(&self, component: SegmentComponent) -> Result<ReadOnlySource, OpenReadError> {
         let path = self.relative_path(component);
-        let source = self.directory.open_read(&path)?;
+        let source = self.index.directory().open_read(&path)?;
         Ok(source)
     }
 
     /// Open one of the component file for *regular* write.
     pub fn open_write(&mut self, component: SegmentComponent) -> Result<WritePtr, OpenWriteError> {
         let path = self.relative_path(component);
-        let wrt = self.directory.open_write(&path)?;
-        Ok(wrt)
+        let write = self.index.directory_mut().open_write(&path)?;
+        Ok(write)
     }
 }
 

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;
@@ -48,7 +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>,
@@ -126,8 +126,8 @@ 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) {
-            FieldNormReader::open(fieldnorm_source)
+        if let Some(fieldnorm_reader) = self.fieldnorm_readers.get_field(field) {
+            fieldnorm_reader
         } else {
             let field_name = self.schema.get_field_name(field);
             let err_msg = format!(
@@ -178,8 +178,8 @@ impl SegmentReader {
         let fast_field_readers =
             Arc::new(FastFieldReaders::load_all(&schema, &fast_fields_composite)?);
 
-        let fieldnorms_data = segment.open_read(SegmentComponent::FIELDNORMS)?;
-        let fieldnorms_composite = CompositeFile::open(&fieldnorms_data)?;
+        let fieldnorm_data = segment.open_read(SegmentComponent::FIELDNORMS)?;
+        let fieldnorm_readers = FieldNormReaders::open(fieldnorm_data)?;
 
         let delete_bitset_opt = if segment.meta().has_deletes() {
             let delete_data = segment.open_read(SegmentComponent::DELETE)?;
@@ -195,7 +195,7 @@ 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,
@@ -295,8 +295,8 @@ impl SegmentReader {
     }
 
     /// Returns an iterator that will iterate over the alive document ids
-    pub fn doc_ids_alive(&self) -> SegmentReaderAliveDocsIterator<'_> {
-        SegmentReaderAliveDocsIterator::new(&self)
+    pub fn doc_ids_alive<'a>(&'a self) -> impl Iterator<Item = DocId> + 'a {
+        (0u32..self.max_doc).filter(move |doc| !self.is_deleted(*doc))
     }
 
     /// Summarize total space usage of this segment.
@@ -308,7 +308,7 @@ 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()
@@ -324,52 +324,6 @@ 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/footer.rs

@@ -94,12 +94,24 @@ impl Footer {
         match &self.versioned_footer {
             VersionedFooter::V1 {
                 crc32: _crc,
-                store_compression: compression,
+                store_compression,
             } => {
-                if &library_version.store_compression != compression {
+                if &library_version.store_compression != store_compression {
                     return Err(Incompatibility::CompressionMismatch {
                         library_compression_format: library_version.store_compression.to_string(),
-                        index_compression_format: compression.to_string(),
+                        index_compression_format: store_compression.to_string(),
+                    });
+                }
+                Ok(())
+            }
+            VersionedFooter::V2 {
+                crc32: _crc,
+                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(())
@@ -120,24 +132,29 @@ pub enum VersionedFooter {
         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::V1 {
+            VersionedFooter::V2 {
                 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(&1u32, &mut buf)?;
+                BinarySerializable::serialize(&2u32, &mut buf)?;
                 BinarySerializable::serialize(crc32, &mut buf)?;
                 BinarySerializable::serialize(compression, &mut buf)?;
             }
-            VersionedFooter::UnknownVersion => {
+            VersionedFooter::V1 { .. } | VersionedFooter::UnknownVersion => {
                 return Err(io::Error::new(
                     io::ErrorKind::InvalidInput,
                     "Cannot serialize an unknown versioned footer ",
@@ -166,22 +183,30 @@ impl BinarySerializable for VersionedFooter {
         reader.read_exact(&mut buf[..])?;
         let mut cursor = &buf[..];
         let version = u32::deserialize(&mut cursor)?;
-        if version == 1 {
-            let crc32 = u32::deserialize(&mut cursor)?;
-            let compression = String::deserialize(&mut cursor)?;
-            Ok(VersionedFooter::V1 {
-                crc32,
-                store_compression: compression,
-            })
-        } else {
-            Ok(VersionedFooter::UnknownVersion)
+        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,
+            }
+        })
     }
 }
 
 impl VersionedFooter {
     pub fn crc(&self) -> Option<CrcHashU32> {
         match self {
+            VersionedFooter::V2 { crc32, .. } => Some(*crc32),
             VersionedFooter::V1 { crc32, .. } => Some(*crc32),
             VersionedFooter::UnknownVersion { .. } => None,
         }
@@ -219,7 +244,7 @@ impl<W: TerminatingWrite> Write for FooterProxy<W> {
 impl<W: TerminatingWrite> TerminatingWrite for FooterProxy<W> {
     fn terminate_ref(&mut self, _: AntiCallToken) -> io::Result<()> {
         let crc32 = self.hasher.take().unwrap().finalize();
-        let footer = Footer::new(VersionedFooter::V1 {
+        let footer = Footer::new(VersionedFooter::V2 {
             crc32,
             store_compression: crate::store::COMPRESSION.to_string(),
         });
@@ -248,15 +273,11 @@ mod tests {
         assert!(footer_proxy.terminate().is_ok());
         assert_eq!(vec.len(), 167);
         let footer = Footer::deserialize(&mut &vec[..]).unwrap();
-        if let VersionedFooter::V1 {
-            crc32: _,
-            store_compression,
-        } = footer.versioned_footer
-        {
-            assert_eq!(store_compression, crate::store::COMPRESSION);
-        } else {
-            panic!("Versioned footer should be V1.");
-        }
+        assert!(matches!(
+           footer.versioned_footer,
+           VersionedFooter::V2 { store_compression, .. }
+           if store_compression == crate::store::COMPRESSION
+        ));
         assert_eq!(&footer.version, crate::version());
     }
 
@@ -264,7 +285,7 @@ mod tests {
     fn test_serialize_deserialize_footer() {
         let mut buffer = Vec::new();
         let crc32 = 123456u32;
-        let footer: Footer = Footer::new(VersionedFooter::V1 {
+        let footer: Footer = Footer::new(VersionedFooter::V2 {
             crc32,
             store_compression: "lz4".to_string(),
         });
@@ -276,7 +297,7 @@ mod tests {
     #[test]
     fn footer_length() {
         let crc32 = 1111111u32;
-        let versioned_footer = VersionedFooter::V1 {
+        let versioned_footer = VersionedFooter::V2 {
             crc32,
             store_compression: "lz4".to_string(),
         };
@@ -297,7 +318,7 @@ mod tests {
             // versionned footer length
             12 | 128,
             // index format version
-            1,
+            2,
             0,
             0,
             0,
@@ -316,7 +337,7 @@ mod tests {
         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::V1 {
+        let expected_versioned_footer: VersionedFooter = VersionedFooter::V2 {
             crc32: expected_crc,
             store_compression: "lz4".to_string(),
         };

src/directory/managed_directory.rs

@@ -11,7 +11,6 @@ use crate::error::DataCorruption;
 use crate::Directory;
 
 use crc32fast::Hasher;
-use serde_json;
 use std::collections::HashSet;
 use std::io;
 use std::io::Write;

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 +13,11 @@ 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;
@@ -224,17 +220,13 @@ struct MmapDirectoryInner {
 }
 
 impl MmapDirectoryInner {
-    fn new(
-        root_path: PathBuf,
-        temp_directory: Option<TempDir>,
-    ) -> Result<MmapDirectoryInner, OpenDirectoryError> {
-        let mmap_directory_inner = MmapDirectoryInner {
+    fn new(root_path: PathBuf, temp_directory: Option<TempDir>) -> MmapDirectoryInner {
+        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(
-        root_path: PathBuf,
-        temp_directory: Option<TempDir>,
-    ) -> Result<MmapDirectory, OpenDirectoryError> {
-        let inner = MmapDirectoryInner::new(root_path, temp_directory)?;
-        Ok(MmapDirectory {
+    fn new(root_path: PathBuf, temp_directory: Option<TempDir>) -> MmapDirectory {
+        let inner = MmapDirectoryInner::new(root_path, temp_directory);
+        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))
         }
     }
 

src/directory/mod.rs

@@ -13,10 +13,7 @@ mod footer;
 mod managed_directory;
 mod ram_directory;
 mod read_only_source;
-mod spilling_writer;
 mod watch_event_router;
-mod persistor;
-
 
 /// Errors specific to the directory module.
 pub mod error;
@@ -26,7 +23,6 @@ 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::spilling_writer::{SpillingResult, SpillingWriter};
 pub use self::watch_event_router::{WatchCallback, WatchCallbackList, WatchHandle};
 use std::io::{self, BufWriter, Write};
 use std::path::PathBuf;
@@ -83,16 +79,10 @@ impl<W: TerminatingWrite> TerminatingWrite for BufWriter<W> {
     }
 }
 
-impl TerminatingWrite for Vec<u8> {
-    fn terminate_ref(&mut self, _a: AntiCallToken) -> io::Result<()> {
-        Ok(())
-    }
-}
-
 #[cfg(test)]
 impl<'a> TerminatingWrite for &'a mut Vec<u8> {
     fn terminate_ref(&mut self, _a: AntiCallToken) -> io::Result<()> {
-        Ok(())
+        self.flush()
     }
 }
 

src/directory/persistor.rs

@@ -1,38 +0,0 @@
-use crate::indexer::{SegmentManager, ResourceManager, MergeOperationInventory};
-use std::thread::JoinHandle;
-use crate::{IndexWriterConfig, SegmentId};
-use std::collections::HashSet;
-
-pub(crate) struct Persistor {
-    memory_manager: ResourceManager,
-    thread_handle: JoinHandle<()>,
-}
-
-impl Persistor {
-    pub(crate) fn create_and_start(segment_manager: SegmentManager,
-                            memory_manager: ResourceManager,
-                            config: IndexWriterConfig) -> crate::Result<Persistor> {
-        let memory_manager_clone = memory_manager.clone();
-        let thread_handle = std::thread::Builder::new()
-            .name("persistor-thread".to_string())
-            .spawn(move || {
-                while let Ok(_) = memory_manager_clone.wait_until_in_range(config.persist_low..) {
-                    segment_manager.largest_segment_not_in_merge();
-                }
-            }).map_err(|_err| crate::TantivyError::ErrorInThread("Failed to start persistor thread.".to_string()))?;
-        Ok(Persistor {
-            memory_manager,
-            thread_handle
-        })
-    }
-
-    /// Stop the persisting thread.
-    ///
-    /// The memory manager will be terminated, which will unlock the thread from any waiting
-    /// position.
-    /// This method blocks for a short amount of tim until the persistor thread has terminated.
-    pub fn stop(self) {
-        self.memory_manager.terminate();
-        let _ = self.thread_handle.join();
-    }
-}

src/directory/ram_directory.rs

@@ -11,7 +11,6 @@ use std::io::{self, BufWriter, Cursor, Seek, SeekFrom, Write};
 use std::path::{Path, PathBuf};
 use std::result;
 use std::sync::{Arc, RwLock};
-use crate::indexer::ResourceManager;
 
 /// Writer associated with the `RAMDirectory`
 ///
@@ -83,12 +82,11 @@ impl TerminatingWrite for VecWriter {
 struct InnerDirectory {
     fs: HashMap<PathBuf, ReadOnlySource>,
     watch_router: WatchCallbackList,
-    memory_manager: ResourceManager,
 }
 
 impl InnerDirectory {
     fn write(&mut self, path: PathBuf, data: &[u8]) -> bool {
-        let data = ReadOnlySource::new_with_allocation(Vec::from(data), &self.memory_manager);
+        let data = ReadOnlySource::new(Vec::from(data));
         self.fs.insert(path, data).is_some()
     }
 
@@ -114,8 +112,8 @@ impl InnerDirectory {
         self.watch_router.subscribe(watch_handle)
     }
 
-    fn total_mem_usage(&self) -> u64 {
-        self.fs.values().map(|source| source.len() as u64).sum()
+    fn total_mem_usage(&self) -> usize {
+        self.fs.values().map(|f| f.len()).sum()
     }
 }
 
@@ -136,30 +134,14 @@ pub struct RAMDirectory {
 }
 
 impl RAMDirectory {
-
-    /// Creates a new RAMDirectory.
-    ///
-    /// Check `.create_with_memory_manager(..)` if you want to associate an external memory
-    /// manager to your RAMDirectory.
-    pub fn create() -> RAMDirectory {
-        RAMDirectory::default()
-    }
-
     /// Constructor
-    pub fn create_with_memory_manager(memory_manager: ResourceManager) -> RAMDirectory {
-        let inner_directory = InnerDirectory {
-            fs: Default::default(),
-            watch_router: Default::default(),
-            memory_manager
-        };
-        RAMDirectory {
-            fs: Arc::new(RwLock::new(inner_directory))
-        }
+    pub fn create() -> RAMDirectory {
+        Self::default()
     }
 
     /// Returns the sum of the size of the different files
     /// in the RAMDirectory.
-    pub fn total_mem_usage(&self) -> u64 {
+    pub fn total_mem_usage(&self) -> usize {
         self.fs.read().unwrap().total_mem_usage()
     }
 
@@ -245,9 +227,6 @@ mod tests {
     use crate::Directory;
     use std::io::Write;
     use std::path::Path;
-    use crate::indexer::ResourceManager;
-    use crate::directory::TerminatingWrite;
-    use std::mem;
 
     #[test]
     fn test_persist() {
@@ -265,59 +244,4 @@ mod tests {
         assert_eq!(directory_copy.atomic_read(path_atomic).unwrap(), msg_atomic);
         assert_eq!(directory_copy.atomic_read(path_seq).unwrap(), msg_seq);
     }
-
-    #[test]
-    fn test_memory_manager_several_path() {
-        let memory_manager = ResourceManager::default();
-        let mut ram_directory = RAMDirectory::create_with_memory_manager(memory_manager.clone());
-        assert!(ram_directory.atomic_write(Path::new("/titi"), b"abcd").is_ok());
-        assert_eq!(memory_manager.total_amount(), 4u64);
-        assert!(ram_directory.atomic_write(Path::new("/toto"), b"abcde").is_ok());
-        assert_eq!(memory_manager.total_amount(), 9u64);
-    }
-
-    #[test]
-    fn test_memory_manager_override() {
-        let memory_manager = ResourceManager::default();
-        let mut ram_directory = RAMDirectory::create_with_memory_manager(memory_manager.clone());
-        assert!(ram_directory.atomic_write(Path::new("/titi"), b"abcde").is_ok());
-        assert_eq!(memory_manager.total_amount(), 5u64);
-        assert!(ram_directory.atomic_write(Path::new("/titi"), b"abcdef").is_ok());
-        assert_eq!(memory_manager.total_amount(), 6u64);
-    }
-
-    #[test]
-    fn test_memory_manager_seq_wrt() {
-        let memory_manager = ResourceManager::default();
-        let mut ram_directory = RAMDirectory::create_with_memory_manager(memory_manager.clone());
-        let mut wrt = ram_directory.open_write(Path::new("/titi")).unwrap();
-        assert!(wrt.write_all(b"abcde").is_ok());
-        assert!(wrt.terminate().is_ok());
-        assert_eq!(memory_manager.total_amount(), 5u64);
-        assert!(ram_directory.atomic_write(Path::new("/titi"), b"abcdef").is_ok());
-        assert_eq!(memory_manager.total_amount(), 6u64);
-    }
-
-    #[test]
-    fn test_release_on_drop() {
-        let memory_manager = ResourceManager::default();
-        let mut ram_directory = RAMDirectory::create_with_memory_manager(memory_manager.clone());
-        let mut wrt = ram_directory.open_write(Path::new("/titi")).unwrap();
-        assert!(wrt.write_all(b"abcde").is_ok());
-        assert!(wrt.terminate().is_ok());
-        assert_eq!(memory_manager.total_amount(), 5u64);
-        let mut wrt2 = ram_directory.open_write(Path::new("/toto")).unwrap();
-        assert!(wrt2.write_all(b"abcdefghijkl").is_ok());
-        assert!(wrt2.terminate().is_ok());
-        assert_eq!(memory_manager.total_amount(), 17u64);
-        let source = ram_directory.open_read(Path::new("/titi")).unwrap();
-        let source_clone = source.clone();
-        assert_eq!(memory_manager.total_amount(), 17u64);
-        mem::drop(ram_directory);
-        assert_eq!(memory_manager.total_amount(), 5u64);
-        mem::drop(source);
-        assert_eq!(memory_manager.total_amount(), 5u64);
-        mem::drop(source_clone);
-        assert_eq!(memory_manager.total_amount(), 0u64);
-    }
 }

src/directory/read_only_source.rs

@@ -2,7 +2,6 @@ use crate::common::HasLen;
 use stable_deref_trait::{CloneStableDeref, StableDeref};
 use std::ops::Deref;
 use std::sync::Arc;
-use crate::indexer::{Allocation, ResourceManager};
 
 pub type BoxedData = Box<dyn Deref<Target = [u8]> + Send + Sync + 'static>;
 
@@ -16,7 +15,6 @@ pub struct ReadOnlySource {
     data: Arc<BoxedData>,
     start: usize,
     stop: usize,
-    allocation: Option<Arc<Allocation>>
 }
 
 unsafe impl StableDeref for ReadOnlySource {}
@@ -37,13 +35,11 @@ impl From<Arc<BoxedData>> for ReadOnlySource {
             data,
             start: 0,
             stop: len,
-            allocation: None
         }
     }
 }
 
 impl ReadOnlySource {
-
     pub(crate) fn new<D>(data: D) -> ReadOnlySource
     where
         D: Deref<Target = [u8]> + Send + Sync + 'static,
@@ -53,24 +49,9 @@ impl ReadOnlySource {
             data: Arc::new(Box::new(data)),
             start: 0,
             stop: len,
-            allocation: None
         }
     }
 
-    pub(crate) fn new_with_allocation<D>(data: D, memory_manager: &ResourceManager) -> ReadOnlySource
-        where
-            D: Deref<Target = [u8]> + Send + Sync + 'static,
-    {
-        let len = data.len();
-        ReadOnlySource {
-            data: Arc::new(Box::new(data)),
-            start: 0,
-            stop: len,
-            allocation: Some(Arc::new(memory_manager.allocate(len as u64)))
-        }
-    }
-
-
     /// Creates an empty ReadOnlySource
     pub fn empty() -> ReadOnlySource {
         ReadOnlySource::new(&[][..])
@@ -117,7 +98,6 @@ impl ReadOnlySource {
             data: self.data.clone(),
             start: self.start + start,
             stop: self.start + stop,
-            allocation: self.allocation.clone()
         }
     }
 

src/directory/spilling_writer.rs

@@ -1,186 +0,0 @@
-use crate::common::MutableEnum;
-use crate::directory::{TerminatingWrite, WritePtr};
-use std::io::{self, Write};
-
-/// Represents the state of the `SpillingWriter`.
-enum SpillingState {
-    Buffer {
-        buffer: Vec<u8>,
-        capacity: usize,
-        write_factory: Box<dyn FnOnce() -> io::Result<WritePtr>>,
-    },
-    Spilled(WritePtr),
-}
-
-impl SpillingState {
-    fn new(
-        limit: usize,
-        write_factory: Box<dyn FnOnce() -> io::Result<WritePtr>>,
-    ) -> SpillingState {
-        SpillingState::Buffer {
-            buffer: Vec::with_capacity(limit),
-            capacity: limit,
-            write_factory,
-        }
-    }
-
-    // Change the state in such a way that it is ready to accept
-    // `extra_capacity` bytes.
-    //
-    fn reserve(self, extra_capacity: usize) -> io::Result<SpillingState> {
-        match self {
-            SpillingState::Buffer {
-                buffer,
-                capacity,
-                write_factory,
-            } => {
-                if capacity >= extra_capacity {
-                    Ok(SpillingState::Buffer {
-                        buffer,
-                        capacity: capacity - extra_capacity,
-                        write_factory,
-                    })
-                } else {
-                    let mut wrt = write_factory()?;
-                    wrt.write_all(&buffer[..])?;
-                    Ok(SpillingState::Spilled(wrt))
-                }
-            }
-            SpillingState::Spilled(wrt) => Ok(SpillingState::Spilled(wrt)),
-        }
-    }
-}
-
-/// The `SpillingWriter` is a writer that start by writing in a
-/// buffer.
-///
-/// Once a memory limit is reached, the spilling writer will
-/// call a given `WritePtr` factory and start spilling into it.
-///
-/// Spilling here includes:
-/// - writing all of the data that were written in the in-memory buffer so far
-/// - writing subsequent data as well.
-///
-/// Once entering "spilling" mode, the `SpillingWriter` stays in this mode.
-pub struct SpillingWriter {
-    state: MutableEnum<SpillingState>,
-}
-
-impl SpillingWriter {
-    //// Creates a new `Spilling Writer`.
-    pub fn new(
-        limit: usize,
-        write_factory: Box<dyn FnOnce() -> io::Result<WritePtr>>,
-    ) -> SpillingWriter {
-        let state = SpillingState::new(limit, write_factory);
-        SpillingWriter {
-            state: MutableEnum::wrap(state),
-        }
-    }
-
-    /// Finalizes the `SpillingWriter`.
-    ///
-    /// The `SpillingResult` object is an enum specific
-    /// to whether the `SpillingWriter` reached the spilling limit
-    /// (In that case, the buffer is returned).
-    ///
-    /// If the writer reached the spilling mode, the underlying `WritePtr`
-    /// is terminated and SpillingResult::Spilled is returned.
-    pub fn finalize(self) -> io::Result<SpillingResult> {
-        match self.state.into() {
-            SpillingState::Spilled(wrt) => {
-                wrt.terminate()?;
-                Ok(SpillingResult::Spilled)
-            }
-            SpillingState::Buffer { buffer, .. } => Ok(SpillingResult::Buffer(buffer)),
-        }
-    }
-}
-
-/// enum used as the result of `.finalize()`.
-pub enum SpillingResult {
-    Spilled,
-    Buffer(Vec<u8>),
-}
-
-impl io::Write for SpillingWriter {
-    fn write(&mut self, payload: &[u8]) -> io::Result<usize> {
-        self.write_all(payload)?;
-        Ok(payload.len())
-    }
-
-    fn flush(&mut self) -> io::Result<()> {
-        if let SpillingState::Spilled(wrt) = &mut *self.state {
-            wrt.flush()?;
-        }
-        Ok(())
-    }
-
-    fn write_all(&mut self, payload: &[u8]) -> io::Result<()> {
-        self.state.map_mutate(|mut state| {
-            state = state.reserve(payload.len())?;
-            match &mut state {
-                SpillingState::Buffer { buffer, .. } => {
-                    buffer.extend_from_slice(payload);
-                }
-                SpillingState::Spilled(wrt) => {
-                    wrt.write_all(payload)?;
-                }
-            }
-            Ok(state)
-        })
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::SpillingWriter;
-    use crate::directory::spilling_writer::SpillingResult;
-    use crate::directory::RAMDirectory;
-    use crate::Directory;
-    use std::io::{self, Write};
-    use std::path::Path;
-
-    #[test]
-    fn test_no_spilling() {
-        let ram_directory = RAMDirectory::create();
-        let mut ram_directory_clone = ram_directory.clone();
-        let path = Path::new("test");
-        let write_factory = Box::new(move || {
-            ram_directory_clone
-                .open_write(path)
-                .map_err(|err| io::Error::new(io::ErrorKind::Other, err))
-        });
-        let mut spilling_wrt = SpillingWriter::new(10, write_factory);
-        assert!(spilling_wrt.write_all(b"abcd").is_ok());
-        if let SpillingResult::Buffer(buf) = spilling_wrt.finalize().unwrap() {
-            assert_eq!(buf, b"abcd")
-        } else {
-            panic!("spill writer should not have spilled");
-        }
-        assert!(!ram_directory.exists(path));
-    }
-
-    #[test]
-    fn test_spilling() {
-        let ram_directory = RAMDirectory::create();
-        let mut ram_directory_clone = ram_directory.clone();
-        let path = Path::new("test");
-        let write_factory = Box::new(move || {
-            ram_directory_clone
-                .open_write(path)
-                .map_err(|err| io::Error::new(io::ErrorKind::Other, err))
-        });
-        let mut spilling_wrt = SpillingWriter::new(10, write_factory);
-        assert!(spilling_wrt.write_all(b"abcd").is_ok());
-        assert!(spilling_wrt.write_all(b"efghijklmnop").is_ok());
-        if let SpillingResult::Spilled = spilling_wrt.finalize().unwrap() {
-        } else {
-            panic!("spill writer should have spilled");
-        }
-        assert_eq!(
-            ram_directory.atomic_read(path).unwrap(),
-            b"abcdefghijklmnop"
-        );
-    }
-}

src/directory/watch_event_router.rs

@@ -5,7 +5,7 @@ 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.
 ///
@@ -32,7 +32,7 @@ impl WatchHandle {
 }
 
 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);

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(...)`.
-#[derive(PartialEq, Eq, Debug)]
-pub enum SkipResult {
-    /// target was in the docset
-    Reached,
-    /// 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,
-}
+/// Sentinel value returned when a DocSet has been entirely consumed.
+///
+/// This is not u32::MAX as one would have expected, due to the lack of SSE2 instructions
+/// to compare [u32; 4].
+pub const TERMINATED: DocId = std::i32::MAX as u32;
 
 /// 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.
-    fn advance(&mut self) -> bool;
+    ///
+    /// The DocId of the next element is returned.
+    /// 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()`
-    /// will return a value greater than or equal to target.
+    /// Advances the DocSet forward until reaching the target, or going to the
+    /// lowest DocId greater than the target.
     ///
-    /// SkipResult expresses whether the `target value` was reached, overstepped,
-    /// or if the `DocSet` was entirely consumed without finding any value
-    /// greater or equal to the `target`.
+    /// If the end of the DocSet is reached, TERMINATED is returned.
     ///
-    /// WARNING: Calling skip always advances the docset.
-    /// More specifically, if the docset is already positionned on the target
-    /// skipping will advance to the next position and return SkipResult::Overstep.
+    /// Calling `.seek(target)` on a terminated DocSet is legal. Implementation
+    /// of DocSet should support it.
     ///
-    /// If `.skip_next()` oversteps, then the docset must be positionned correctly
-    /// on an existing document. In other words, `.doc()` should return the first document
-    /// greater than `DocId`.
-    fn skip_next(&mut self, target: DocId) -> SkipResult {
-        if !self.advance() {
-            return SkipResult::End;
-        }
-        loop {
-            match self.doc().cmp(&target) {
-                Ordering::Less => {
-                    if !self.advance() {
-                        return SkipResult::End;
-                    }
-                }
-                Ordering::Equal => return SkipResult::Reached,
-                Ordering::Greater => return SkipResult::OverStep,
-            }
+    /// Calling `seek(TERMINATED)` is also legal and is the normal way to consume a DocSet.
+    fn seek(&mut self, target: DocId) -> DocId {
+        let mut doc = self.doc();
+        debug_assert!(doc <= target);
+        while doc < target {
+            doc = self.advance();
         }
+        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();
-            } else {
-                return i;
+            *buffer_val = self.doc();
+            if self.advance() == TERMINATED {
+                return i + 1;
             }
         }
         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() {
-            if !delete_bitset.is_deleted(self.doc()) {
+        let mut doc = 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

@@ -7,7 +7,6 @@ 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;

src/fastfield/delete.rs

@@ -9,6 +9,8 @@ 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,
@@ -42,6 +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
@@ -83,42 +103,35 @@ impl HasLen for DeleteBitSet {
 
 #[cfg(test)]
 mod tests {
-    use super::*;
-    use crate::directory::*;
-    use std::path::PathBuf;
+    use super::DeleteBitSet;
+    use crate::common::HasLen;
 
-    fn test_delete_bitset_helper(bitset: &BitSet, max_doc: u32) {
-        let test_path = PathBuf::from("test");
-        let mut directory = RAMDirectory::default();
-        {
-            let mut writer = directory.open_write(&*test_path).unwrap();
-            write_delete_bitset(bitset, max_doc, &mut writer).unwrap();
-            writer.terminate().unwrap();
+    #[test]
+    fn test_delete_bitset_empty() {
+        let delete_bitset = DeleteBitSet::for_test(&[], 10);
+        for doc in 0..10 {
+            assert_eq!(delete_bitset.is_deleted(doc), !delete_bitset.is_alive(doc));
         }
-        let source = directory.open_read(&test_path).unwrap();
-        let delete_bitset = DeleteBitSet::open(source);
-        for doc in 0..max_doc {
-            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 mut bitset = BitSet::with_max_value(10);
-            bitset.insert(1);
-            bitset.insert(9);
-            test_delete_bitset_helper(&bitset, 10);
-        }
-        {
-            let mut bitset = BitSet::with_max_value(8);
-            bitset.insert(1);
-            bitset.insert(2);
-            bitset.insert(3);
-            bitset.insert(5);
-            bitset.insert(7);
-            test_delete_bitset_helper(&bitset, 8);
+        let delete_bitset = DeleteBitSet::for_test(&[1, 9], 10);
+        assert!(delete_bitset.is_alive(0));
+        assert!(delete_bitset.is_deleted(1));
+        assert!(delete_bitset.is_alive(2));
+        assert!(delete_bitset.is_alive(3));
+        assert!(delete_bitset.is_alive(4));
+        assert!(delete_bitset.is_alive(5));
+        assert!(delete_bitset.is_alive(6));
+        assert!(delete_bitset.is_alive(6));
+        assert!(delete_bitset.is_alive(7));
+        assert!(delete_bitset.is_alive(8));
+        assert!(delete_bitset.is_deleted(9));
+        for doc in 0..10 {
+            assert_eq!(delete_bitset.is_deleted(doc), !delete_bitset.is_alive(doc));
         }
+        assert_eq!(delete_bitset.len(), 2);
     }
 }

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)]
-impl From<Vec<u32>> for FieldNormReader {
-    fn from(field_norms: Vec<u32>) -> FieldNormReader {
+    #[cfg(test)]
+    pub fn for_test(field_norms: &[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/index_writer.rs

@@ -8,26 +8,22 @@ use crate::core::SegmentComponent;
 use crate::core::SegmentId;
 use crate::core::SegmentMeta;
 use crate::core::SegmentReader;
+use crate::directory::TerminatingWrite;
 use crate::directory::{DirectoryLock, GarbageCollectionResult};
-use crate::directory::{TerminatingWrite, WatchCallbackList};
-use crate::docset::DocSet;
+use crate::docset::{DocSet, TERMINATED};
 use crate::error::TantivyError;
 use crate::fastfield::write_delete_bitset;
 use crate::indexer::delete_queue::{DeleteCursor, DeleteQueue};
 use crate::indexer::doc_opstamp_mapping::DocToOpstampMapping;
 use crate::indexer::operation::DeleteOperation;
-use crate::indexer::segment_manager::SegmentRegisters;
-use crate::indexer::segment_register::SegmentRegister;
 use crate::indexer::stamper::Stamper;
-use crate::indexer::{SegmentEntry, ResourceManager};
+use crate::indexer::MergePolicy;
+use crate::indexer::SegmentEntry;
 use crate::indexer::SegmentWriter;
-use crate::indexer::{IndexWriterConfig, MergePolicy};
-use crate::reader::NRTReader;
 use crate::schema::Document;
 use crate::schema::IndexRecordOption;
 use crate::schema::Term;
-use crate::tokenizer::TokenizerManager;
-use crate::{IndexReader, Opstamp};
+use crate::Opstamp;
 use crossbeam::channel;
 use futures::executor::block_on;
 use futures::future::Future;
@@ -35,10 +31,18 @@ use smallvec::smallvec;
 use smallvec::SmallVec;
 use std::mem;
 use std::ops::Range;
-use std::sync::{Arc, RwLock};
+use std::sync::Arc;
 use std::thread;
 use std::thread::JoinHandle;
 
+// Size of the margin for the heap. A segment is closed when the remaining memory
+// in the heap goes below MARGIN_IN_BYTES.
+pub const MARGIN_IN_BYTES: usize = 1_000_000;
+
+// We impose the memory per thread to be at least 3 MB.
+pub const HEAP_SIZE_MIN: usize = ((MARGIN_IN_BYTES as u32) * 3u32) as usize;
+pub const HEAP_SIZE_MAX: usize = u32::max_value() as usize - MARGIN_IN_BYTES;
+
 // Add document will block if the number of docs waiting in the queue to be indexed
 // reaches `PIPELINE_MAX_SIZE_IN_DOCS`
 const PIPELINE_MAX_SIZE_IN_DOCS: usize = 10_000;
@@ -65,9 +69,8 @@ pub struct IndexWriter {
     _directory_lock: Option<DirectoryLock>,
 
     index: Index,
-    config: IndexWriterConfig,
 
-    segment_registers: Arc<RwLock<SegmentRegisters>>,
+    heap_size_in_bytes_per_thread: usize,
 
     workers_join_handle: Vec<JoinHandle<crate::Result<()>>>,
 
@@ -77,14 +80,13 @@ pub struct IndexWriter {
     segment_updater: SegmentUpdater,
 
     worker_id: usize,
+
+    num_threads: usize,
+
     delete_queue: DeleteQueue,
 
     stamper: Stamper,
     committed_opstamp: Opstamp,
-
-    on_commit: WatchCallbackList,
-
-    memory_manager: ResourceManager,
 }
 
 fn compute_deleted_bitset(
@@ -110,15 +112,15 @@ fn compute_deleted_bitset(
         if let Some(mut docset) =
             inverted_index.read_postings(&delete_op.term, IndexRecordOption::Basic)
         {
-            while docset.advance() {
-                let deleted_doc = docset.doc();
+            let mut deleted_doc = docset.doc();
+            while deleted_doc != TERMINATED {
                 if deleted_doc < limit_doc {
                     delete_bitset.insert(deleted_doc);
                     might_have_changed = true;
                 }
+                deleted_doc = docset.advance();
             }
         }
-
         delete_cursor.advance();
     }
     Ok(might_have_changed)
@@ -131,6 +133,7 @@ fn compute_deleted_bitset(
 /// 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,
 ) -> crate::Result<()> {
@@ -139,33 +142,25 @@ pub(crate) fn advance_deletes(
         return Ok(());
     }
 
-    let delete_bitset_opt = segment_entry.take_delete_bitset();
-
-    // We avoid directly advancing the `SegmentEntry` delete cursor, because
-    // we do not want to end up in an invalid state if the delete bitset
-    // serialization fails.
-    let mut delete_cursor = segment_entry.delete_cursor();
-
-    if delete_bitset_opt.is_none() && segment_entry.delete_cursor().get().is_none() {
+    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(());
     }
 
-    // We open our current serialized segment to compute the new deleted bitset.
-    let segment = segment_entry.segment().clone();
     let segment_reader = SegmentReader::open(&segment)?;
 
     let max_doc = segment_reader.max_doc();
-
-    let mut delete_bitset: BitSet =
-        delete_bitset_opt.unwrap_or_else(|| BitSet::with_max_value(max_doc));
+    let mut delete_bitset: BitSet = match segment_entry.delete_bitset() {
+        Some(previous_delete_bitset) => (*previous_delete_bitset).clone(),
+        None => BitSet::with_max_value(max_doc),
+    };
 
     let num_deleted_docs_before = segment.meta().num_deleted_docs();
 
     compute_deleted_bitset(
         &mut delete_bitset,
         &segment_reader,
-        &mut delete_cursor,
+        segment_entry.delete_cursor(),
         &DocToOpstampMapping::None,
         target_opstamp,
     )?;
@@ -184,40 +179,32 @@ pub(crate) fn advance_deletes(
     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.
-        let mut delete_file = segment
-            .with_delete_meta(num_deleted_docs as u32, target_opstamp)
-            .open_write(SegmentComponent::DELETE)?;
+        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.reset_delete_meta(num_deleted_docs as u32, target_opstamp);
     }
 
-    // Regardless of whether we did end up having to write a new file or not
-    // we advance the `delete_cursor`. This is an optimisation. We want to ensure we do not
-    // check that a given deleted term does not match any of our docs more than once.
-    segment_entry.set_delete_cursor(delete_cursor);
+    segment_entry.set_meta(segment.meta().clone());
     Ok(())
 }
 
 fn index_documents(
-    config: IndexWriterConfig,
+    memory_budget: usize,
     segment: Segment,
     grouped_document_iterator: &mut dyn Iterator<Item = OperationGroup>,
     segment_updater: &mut SegmentUpdater,
-    tokenizers: &TokenizerManager,
     mut delete_cursor: DeleteCursor,
-    memory_manager: ResourceManager
 ) -> crate::Result<bool> {
     let schema = segment.schema();
 
-    let mut segment_writer = SegmentWriter::for_segment(
-        &config, segment, &schema, tokenizers, memory_manager)?;
+    let mut segment_writer = SegmentWriter::for_segment(memory_budget, segment.clone(), &schema)?;
     for document_group in grouped_document_iterator {
         for doc in document_group {
             segment_writer.add_document(doc, &schema)?;
         }
         let mem_usage = segment_writer.mem_usage();
-        if mem_usage >= config.heap_size_before_flushing() {
+        if mem_usage >= memory_budget - MARGIN_IN_BYTES {
             info!(
                 "Buffer limit reached, flushing segment with maxdoc={}.",
                 segment_writer.max_doc()
@@ -236,14 +223,24 @@ fn index_documents(
     // the worker thread.
     assert!(max_doc > 0);
 
-    let (segment, doc_opstamps): (Segment, Vec<Opstamp>) = segment_writer.finalize()?;
+    let doc_opstamps: Vec<Opstamp> = segment_writer.finalize()?;
+
+    let segment_with_max_doc = segment.with_max_doc(max_doc);
 
     let last_docstamp: Opstamp = *(doc_opstamps.last().unwrap());
 
-    let delete_bitset_opt =
-        apply_deletes(&segment, &mut delete_cursor, &doc_opstamps, last_docstamp)?;
+    let delete_bitset_opt = apply_deletes(
+        &segment_with_max_doc,
+        &mut delete_cursor,
+        &doc_opstamps,
+        last_docstamp,
+    )?;
 
-    let segment_entry = SegmentEntry::new(segment, delete_cursor, delete_bitset_opt);
+    let segment_entry = SegmentEntry::new(
+        segment_with_max_doc.meta().clone(),
+        delete_cursor,
+        delete_bitset_opt,
+    );
     block_on(segment_updater.schedule_add_segment(segment_entry))?;
     Ok(true)
 }
@@ -295,10 +292,21 @@ impl IndexWriter {
     /// If the heap size per thread is too small, panics.
     pub(crate) fn new(
         index: &Index,
-        mut config: IndexWriterConfig,
+        num_threads: usize,
+        heap_size_in_bytes_per_thread: usize,
         directory_lock: DirectoryLock,
     ) -> crate::Result<IndexWriter> {
-        config.validate()?;
+        if heap_size_in_bytes_per_thread < HEAP_SIZE_MIN {
+            let err_msg = format!(
+                "The heap size per thread needs to be at least {}.",
+                HEAP_SIZE_MIN
+            );
+            return Err(TantivyError::InvalidArgument(err_msg));
+        }
+        if heap_size_in_bytes_per_thread >= HEAP_SIZE_MAX {
+            let err_msg = format!("The heap size per thread cannot exceed {}", HEAP_SIZE_MAX);
+            return Err(TantivyError::InvalidArgument(err_msg));
+        }
         let (document_sender, document_receiver): (OperationSender, OperationReceiver) =
             channel::bounded(PIPELINE_MAX_SIZE_IN_DOCS);
 
@@ -306,26 +314,16 @@ impl IndexWriter {
 
         let current_opstamp = index.load_metas()?.opstamp;
 
-        let meta = index.load_metas()?;
-
         let stamper = Stamper::new(current_opstamp);
 
-        let commited_segments = SegmentRegister::new(
-            index.directory(),
-            &index.schema(),
-            meta.segments,
-            &delete_queue.cursor(),
-        );
-        let segment_registers = Arc::new(RwLock::new(SegmentRegisters::new(commited_segments)));
-
         let segment_updater =
-            SegmentUpdater::create(segment_registers.clone(), index.clone(), stamper.clone())?;
+            SegmentUpdater::create(index.clone(), stamper.clone(), &delete_queue.cursor())?;
 
         let mut index_writer = IndexWriter {
             _directory_lock: Some(directory_lock),
 
+            heap_size_in_bytes_per_thread,
             index: index.clone(),
-            config,
 
             operation_receiver: document_receiver,
             operation_sender: document_sender,
@@ -333,6 +331,7 @@ impl IndexWriter {
             segment_updater,
 
             workers_join_handle: vec![],
+            num_threads,
 
             delete_queue,
 
@@ -340,10 +339,6 @@ impl IndexWriter {
             stamper,
 
             worker_id: 0,
-            segment_registers,
-            on_commit: Default::default(),
-
-            memory_manager: Default::default()
         };
         index_writer.start_workers()?;
         Ok(index_writer)
@@ -351,7 +346,7 @@ impl IndexWriter {
 
     fn drop_sender(&mut self) {
         let (sender, _receiver) = channel::bounded(1);
-        mem::replace(&mut self.operation_sender, sender);
+        self.operation_sender = sender;
     }
 
     /// If there are some merging threads, blocks until they all finish their work and
@@ -371,11 +366,23 @@ impl IndexWriter {
                 })?;
         }
 
-        self
+        let result = self
             .segment_updater
-            .wait_merging_thread();
+            .wait_merging_thread()
+            .map_err(|_| TantivyError::ErrorInThread("Failed to join merging thread.".into()));
 
-        Ok(())
+        if let Err(ref e) = result {
+            error!("Some merging thread failed {:?}", e);
+        }
+
+        result
+    }
+
+    #[doc(hidden)]
+    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);
+        block_on(self.segment_updater.schedule_add_segment(segment_entry))
     }
 
     /// Creates a new segment.
@@ -398,9 +405,8 @@ impl IndexWriter {
 
         let mut delete_cursor = self.delete_queue.cursor();
 
+        let mem_budget = self.heap_size_in_bytes_per_thread;
         let index = self.index.clone();
-        let config = self.config.clone();
-        let memory_manager = self.memory_manager.clone();
         let join_handle: JoinHandle<crate::Result<()>> = thread::Builder::new()
             .name(format!("thrd-tantivy-index{}", self.worker_id))
             .spawn(move || {
@@ -429,13 +435,11 @@ impl IndexWriter {
                     }
                     let segment = index.new_segment();
                     index_documents(
-                        config.clone(),
+                        mem_budget,
                         segment,
                         &mut document_iterator,
                         &mut segment_updater,
-                        index.tokenizers(),
                         delete_cursor.clone(),
-                        memory_manager.clone()
                     )?;
                 }
             })?;
@@ -455,7 +459,7 @@ impl IndexWriter {
     }
 
     fn start_workers(&mut self) -> crate::Result<()> {
-        for _ in 0..self.config.max_indexing_threads {
+        for _ in 0..self.num_threads {
             self.add_indexing_worker()?;
         }
         Ok(())
@@ -532,6 +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);
@@ -560,14 +565,18 @@ impl IndexWriter {
             .take()
             .expect("The IndexWriter does not have any lock. This is a bug, please report.");
 
-        let new_index_writer: IndexWriter =
-            IndexWriter::new(&self.index, self.config.clone(), directory_lock)?;
+        let new_index_writer: IndexWriter = IndexWriter::new(
+            &self.index,
+            self.num_threads,
+            self.heap_size_in_bytes_per_thread,
+            directory_lock,
+        )?;
 
         // the current `self` is dropped right away because of this call.
         //
         // 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.
@@ -600,7 +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, soft_commit: bool) -> crate::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.
@@ -628,7 +637,7 @@ impl IndexWriter {
         }
 
         let commit_opstamp = self.stamper.stamp();
-        let prepared_commit = PreparedCommit::new(self, commit_opstamp, soft_commit);
+        let prepared_commit = PreparedCommit::new(self, commit_opstamp);
         info!("Prepared commit {}", commit_opstamp);
         Ok(prepared_commit)
     }
@@ -648,25 +657,7 @@ impl IndexWriter {
     /// that made it in the commit.
     ///
     pub fn commit(&mut self) -> crate::Result<Opstamp> {
-        self.prepare_commit(false)?.commit()
-    }
-
-    pub fn soft_commit(&mut self) -> crate::Result<Opstamp> {
-        self.prepare_commit(true)?.commit()
-    }
-
-    pub(crate) fn trigger_commit(&self) -> impl Future<Output = ()> {
-        self.on_commit.broadcast()
-    }
-
-    pub fn reader(&self, num_searchers: usize) -> crate::Result<IndexReader> {
-        let nrt_reader = NRTReader::create(
-            num_searchers,
-            self.index.clone(),
-            self.segment_registers.clone(),
-            &self.on_commit,
-        )?;
-        Ok(IndexReader::NRT(nrt_reader))
+        self.prepare_commit()?.commit()
     }
 
     pub(crate) fn segment_updater(&self) -> &SegmentUpdater {
@@ -1064,8 +1055,7 @@ mod tests {
                 index_writer.add_document(doc!(text_field => "a"));
             }
             {
-                let mut prepared_commit =
-                    index_writer.prepare_commit(false).expect("commit failed");
+                let mut prepared_commit = index_writer.prepare_commit().expect("commit failed");
                 prepared_commit.set_payload("first commit");
                 prepared_commit.commit().expect("commit failed");
             }
@@ -1098,8 +1088,7 @@ mod tests {
                 index_writer.add_document(doc!(text_field => "a"));
             }
             {
-                let mut prepared_commit =
-                    index_writer.prepare_commit(false).expect("commit failed");
+                let mut prepared_commit = index_writer.prepare_commit().expect("commit failed");
                 prepared_commit.set_payload("first commit");
                 prepared_commit.abort().expect("commit failed");
             }
@@ -1277,41 +1266,4 @@ mod tests {
         let commit = index_writer.commit();
         assert!(commit.is_ok());
     }
-
-    #[test]
-    fn test_index_writer_reader() {
-        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"));
-        assert!(index_writer.commit().is_ok());
-        let reader = index_writer.reader(2).unwrap();
-        let searcher = reader.searcher();
-        assert_eq!(searcher.num_docs(), 1u64);
-        index_writer.add_document(doc!(idfield=>"myid"));
-        assert!(index_writer.commit().is_ok());
-        assert_eq!(reader.searcher().num_docs(), 2u64);
-        assert_eq!(searcher.num_docs(), 1u64);
-    }
-
-    #[test]
-    fn test_index_writer_reader_soft_commit() {
-        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"));
-        assert!(index_writer.soft_commit().is_ok());
-        let nrt_reader = index_writer.reader(2).unwrap();
-        let normal_reader = index.reader_builder().try_into().unwrap();
-        assert_eq!(nrt_reader.searcher().num_docs(), 1u64);
-        assert_eq!(normal_reader.searcher().num_docs(), 0u64);
-        assert!(index_writer.commit().is_ok());
-        assert!(normal_reader.reload().is_ok());
-        assert_eq!(nrt_reader.searcher().num_docs(), 1u64);
-        assert_eq!(normal_reader.searcher().num_docs(), 1u64);
-    }
 }

src/indexer/index_writer_config.rs

@@ -1,133 +0,0 @@
-use serde::{Deserialize, Serialize};
-
-// Size of the margin for the heap. A segment is closed when the remaining memory
-// in the heap goes below MARGIN_IN_BYTES.
-const MARGIN_IN_BYTES: u64 = 1_000_000;
-
-// We impose the memory per thread to be at least 3 MB.
-const HEAP_SIZE_MIN: u64 = MARGIN_IN_BYTES * 3u64;
-const HEAP_SIZE_MAX: u64 = u32::max_value() as u64 - MARGIN_IN_BYTES;
-
-#[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct IndexWriterConfig {
-    pub max_indexing_threads: usize,
-    pub max_merging_threads: usize,
-    pub memory_budget: u64,
-    pub store_flush_num_bytes: u64,
-    pub persist_low: u64,
-    pub persist_high: u64,
-}
-
-impl Default for IndexWriterConfig {
-    fn default() -> Self {
-        IndexWriterConfig {
-            max_indexing_threads: 1,
-            max_merging_threads: 3,
-            memory_budget: 50_000_000u64,
-            store_flush_num_bytes: 10_000_000u64,
-            persist_low: 10_000_000u64,
-            persist_high: 50_000_000u64
-        }
-    }
-}
-
-impl IndexWriterConfig {
-    #[cfg(test)]
-    pub fn for_test() -> IndexWriterConfig {
-        IndexWriterConfig {
-            max_indexing_threads: 1,
-            max_merging_threads: 5,
-            memory_budget: 4_000_000u64,
-            store_flush_num_bytes: 500_000u64,
-            persist_low: 2_000_000u64,
-            persist_high: 3_000_000u64,
-        }
-    }
-
-    // Ensures the `IndexWriterConfig` is correct.
-    //
-    // This method checks that the values in the `IndexWriterConfig`
-    // are valid. If it is not, it may mutate some of the values (like `max_num_threads`) to
-    // fit the contracts or return an error with an explicit error message.
-    //
-    // If called twice, the config is guaranteed to not be updated the second time.
-    pub fn validate(&mut self) -> crate::Result<()> {
-        if self.memory_budget < HEAP_SIZE_MIN {
-            let err_msg = format!(
-                "The heap size per thread needs to be at least {}.",
-                HEAP_SIZE_MIN
-            );
-            return Err(crate::TantivyError::InvalidArgument(err_msg));
-        }
-        let heap_size_in_bytes_per_thread = self.heap_size_in_byte_per_thread();
-        if heap_size_in_bytes_per_thread >= HEAP_SIZE_MAX {
-            let err_msg = format!("The heap size per thread cannot exceed {}", HEAP_SIZE_MAX);
-            return Err(crate::TantivyError::InvalidArgument(err_msg));
-        }
-        if heap_size_in_bytes_per_thread < HEAP_SIZE_MIN {
-            self.max_indexing_threads = (self.memory_budget / HEAP_SIZE_MIN) as usize;
-        }
-        Ok(())
-    }
-
-    pub fn heap_size_in_byte_per_thread(&self) -> u64 {
-        self.memory_budget / self.max_indexing_threads as u64
-    }
-
-    pub fn heap_size_before_flushing(&self) -> u64 {
-        self.heap_size_in_byte_per_thread() - MARGIN_IN_BYTES
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use crate::IndexWriterConfig;
-
-    #[test]
-    fn test_index_writer_config_simple() {
-        let mut index = IndexWriterConfig {
-            max_indexing_threads: 3,
-            memory_budget: super::HEAP_SIZE_MIN * 3,
-            ..Default::default()
-        };
-        assert!(index.validate().is_ok());
-        assert_eq!(index.max_indexing_threads, 3);
-        assert_eq!(index.heap_size_in_byte_per_thread(), super::HEAP_SIZE_MIN);
-    }
-
-    #[test]
-    fn test_index_writer_config_reduce_num_threads() {
-        let mut index = IndexWriterConfig {
-            max_indexing_threads: 3,
-            memory_budget: super::HEAP_SIZE_MIN,
-            ..Default::default()
-        };
-        assert!(index.validate().is_ok());
-        assert_eq!(index.max_indexing_threads, 1);
-        assert_eq!(index.heap_size_in_byte_per_thread(), super::HEAP_SIZE_MIN);
-    }
-
-    #[test]
-    fn test_index_writer_config_not_enough_memory() {
-        let mut index = IndexWriterConfig {
-            max_indexing_threads: 1,
-            memory_budget: super::HEAP_SIZE_MIN - 1,
-            ..Default::default()
-        };
-        assert!(
-            matches!(index.validate(), Err(crate::TantivyError::InvalidArgument(msg) ) if msg.contains("The heap size per thread needs to be at least"))
-        );
-    }
-
-    #[test]
-    fn test_index_writer_config_too_much_memory() {
-        let mut index = IndexWriterConfig {
-            max_indexing_threads: 1,
-            memory_budget: (u32::max_value() as u64) + 1,
-            ..Default::default()
-        };
-        assert!(
-            matches!(index.validate(), Err(crate::TantivyError::InvalidArgument(msg) ) if msg.contains("The heap size per thread cannot exceed"))
-        );
-    }
-}

src/indexer/log_merge_policy.rs

@@ -54,10 +54,6 @@ 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)
@@ -67,27 +63,35 @@ impl MergePolicy for LogMergePolicy {
 
         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];
-        let mut current_max_log_size = first_score;
-        let mut levels = vec![vec![first_ind]];
-        for &(ind, score) in (&size_sorted_log_tuples).iter().skip(1) {
-            if score < (current_max_log_size - self.level_log_size) {
-                current_max_log_size = score;
-                levels.push(Vec::new());
+        if let Some(&(first_ind, first_score)) = size_sorted_log_tuples.first() {
+            let mut current_max_log_size = first_score;
+            let mut levels = vec![vec![first_ind]];
+            for &(ind, score) in (&size_sorted_log_tuples).iter().skip(1) {
+                if score < (current_max_log_size - self.level_log_size) {
+                    current_max_log_size = score;
+                    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()
     }
 }
 
@@ -179,6 +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
@@ -194,6 +199,17 @@ mod tests {
         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![

src/indexer/merge_operation.rs

@@ -1,22 +1,17 @@
-use crate::indexer::resource_manager::{Allocation, ResourceManager};
 use crate::Opstamp;
 use crate::SegmentId;
 use census::{Inventory, TrackedObject};
 use std::collections::HashSet;
-use std::fmt;
 use std::ops::Deref;
 
-#[derive(Default, Clone)]
-pub(crate) struct MergeOperationInventory {
-    inventory: Inventory<InnerMergeOperation>,
-    num_merge_watcher: ResourceManager,
-}
+#[derive(Default)]
+pub(crate) struct MergeOperationInventory(Inventory<InnerMergeOperation>);
 
 impl Deref for MergeOperationInventory {
     type Target = Inventory<InnerMergeOperation>;
 
     fn deref(&self) -> &Self::Target {
-        &self.inventory
+        &self.0
     }
 }
 
@@ -30,10 +25,6 @@ impl MergeOperationInventory {
         }
         segment_in_merge
     }
-
-    pub fn wait_until_empty(&self) {
-        let _ = self.num_merge_watcher.wait_until_in_range(0..1);
-    }
 }
 
 /// A `MergeOperation` has two roles.
@@ -56,17 +47,6 @@ pub struct MergeOperation {
 pub(crate) struct InnerMergeOperation {
     target_opstamp: Opstamp,
     segment_ids: Vec<SegmentId>,
-    _allocation: Allocation,
-}
-
-impl fmt::Debug for InnerMergeOperation {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        write!(
-            f,
-            "MergeOp(target_opstamp={:?}, segment_ids={:?})",
-            self.target_opstamp, self.segment_ids
-        )
-    }
 }
 
 impl MergeOperation {
@@ -75,11 +55,9 @@ impl MergeOperation {
         target_opstamp: Opstamp,
         segment_ids: Vec<SegmentId>,
     ) -> MergeOperation {
-        let allocation = inventory.num_merge_watcher.allocate(1);
         let inner_merge_operation = InnerMergeOperation {
             target_opstamp,
             segment_ids,
-            _allocation: allocation,
         };
         MergeOperation {
             inner: inventory.track(inner_merge_operation),

src/indexer/merger.rs

@@ -2,29 +2,28 @@ use crate::common::MAX_DOC_LIMIT;
 use crate::core::Segment;
 use crate::core::SegmentReader;
 use crate::core::SerializableSegment;
-use crate::directory::TerminatingWrite;
-use crate::directory::WritePtr;
-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 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;
@@ -169,7 +168,7 @@ impl IndexMerger {
 
     fn write_fieldnorms(
         &self,
-        fieldnorms_serializer: &mut FieldNormsSerializer,
+        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);
@@ -184,6 +183,7 @@ impl IndexMerger {
             }
             fieldnorms_serializer.serialize_field(field, &fieldnorms_data[..])?;
         }
+        fieldnorms_serializer.close()?;
         Ok(())
     }
 
@@ -494,10 +494,11 @@ impl IndexMerger {
         indexed_field: Field,
         field_type: &FieldType,
         serializer: &mut InvertedIndexSerializer,
+        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))
@@ -552,7 +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);
 
@@ -562,84 +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
-                .current_kvs()
-                .iter()
-                .flat_map(|heap_item| {
-                    let segment_ord = heap_item.segment_ord;
-                    let term_info = heap_item.streamer.value();
-                    let segment_reader = &self.readers[heap_item.segment_ord];
-                    let inverted_index = segment_reader.inverted_index(indexed_field);
-                    let mut segment_postings = inverted_index
-                        .read_postings_from_terminfo(term_info, segment_postings_option);
-                    while segment_postings.advance() {
-                        if !segment_reader.is_deleted(segment_postings.doc()) {
-                            return Some((segment_ord, segment_postings));
-                        }
-                    }
-                    None
-                })
-                .collect();
+            for heap_item in merged_terms.current_kvs() {
+                let segment_ord = heap_item.segment_ord;
+                let term_info = heap_item.streamer.value();
+                let segment_reader = &self.readers[heap_item.segment_ord];
+                let inverted_index: &InvertedIndexReader = &*field_readers[segment_ord];
+                let segment_postings =
+                    inverted_index.read_postings_from_terminfo(term_info, segment_postings_option);
+                let delete_bitset_opt = segment_reader.delete_bitset();
+                let doc_freq = if let Some(delete_bitset) = delete_bitset_opt {
+                    segment_postings.doc_freq_given_deletes(delete_bitset)
+                } else {
+                    segment_postings.doc_freq()
+                };
+                if doc_freq > 0u32 {
+                    total_doc_freq += doc_freq;
+                    segment_postings_containing_the_term.push((segment_ord, segment_postings));
+                }
+            }
 
             // 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 not, the `term` will be entirely removed.
-
-                // We know that there is at least one document containing
-                // 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()?;
+            if total_doc_freq == 0u32 {
+                // All docs that used to contain the term have been deleted. The `term` will be
+                // entirely removed.
+                continue;
             }
+
+            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)
@@ -648,13 +644,18 @@ impl IndexMerger {
     fn write_postings(
         &self,
         serializer: &mut InvertedIndexSerializer,
+        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) =
-                    self.write_postings_for_field(field, field_entry.field_type(), serializer)?
-                {
+                if let Some(term_ordinal_mapping) = self.write_postings_for_field(
+                    field,
+                    field_entry.field_type(),
+                    serializer,
+                    fieldnorm_reader,
+                )? {
                     term_ordinal_mappings.insert(field, term_ordinal_mapping);
                 }
             }
@@ -662,8 +663,7 @@ impl IndexMerger {
         Ok(term_ordinal_mappings)
     }
 
-    pub fn write_storable_fields(&self, store_wrt: WritePtr) -> crate::Result<()> {
-        let mut store_writer = StoreWriter::new(store_wrt);
+    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 {
@@ -675,17 +675,23 @@ impl IndexMerger {
                 store_writer.stack(&store_reader)?;
             }
         }
-        let store_wrt = store_writer.close()?;
-        store_wrt.terminate()?;
         Ok(())
     }
 }
 
 impl SerializableSegment for IndexMerger {
     fn write(&self, mut serializer: SegmentSerializer) -> crate::Result<u32> {
-        let term_ord_mappings = self.write_postings(serializer.get_postings_serializer())?;
-        self.write_fieldnorms(serializer.get_fieldnorms_serializer())?;
+        if let Some(fieldnorms_serializer) = serializer.extract_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()?;
         Ok(self.max_doc)
     }
@@ -693,15 +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;
@@ -709,9 +715,11 @@ 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::executor::block_on;
     use std::io::Cursor;
@@ -1518,12 +1526,9 @@ mod tests {
         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()?;
             }
@@ -1535,6 +1540,72 @@ mod tests {
         // 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

@@ -2,14 +2,12 @@ pub mod delete_queue;
 
 mod doc_opstamp_mapping;
 pub mod index_writer;
-mod index_writer_config;
 mod log_merge_policy;
 mod merge_operation;
 pub mod merge_policy;
 pub mod merger;
 pub mod operation;
 mod prepared_commit;
-mod resource_manager;
 mod segment_entry;
 mod segment_manager;
 mod segment_register;
@@ -18,17 +16,13 @@ pub mod segment_updater;
 mod segment_writer;
 mod stamper;
 
-pub(crate) use self::resource_manager::{Allocation, ResourceManager};
-pub(crate) use self::merge_operation::MergeOperationInventory;
 pub use self::index_writer::IndexWriter;
-pub use self::index_writer_config::IndexWriterConfig;
 pub use self::log_merge_policy::LogMergePolicy;
 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;
 pub use self::segment_manager::SegmentManager;
-pub(crate) use self::segment_manager::SegmentRegisters;
 pub use self::segment_serializer::SegmentSerializer;
 pub use self::segment_writer::SegmentWriter;
 

src/indexer/prepared_commit.rs

@@ -7,20 +7,14 @@ pub struct PreparedCommit<'a> {
     index_writer: &'a mut IndexWriter,
     payload: Option<String>,
     opstamp: Opstamp,
-    soft_commit: bool,
 }
 
 impl<'a> PreparedCommit<'a> {
-    pub(crate) fn new(
-        index_writer: &'a mut IndexWriter,
-        opstamp: Opstamp,
-        soft_commit: bool,
-    ) -> PreparedCommit<'_> {
+    pub(crate) fn new(index_writer: &'a mut IndexWriter, opstamp: Opstamp) -> PreparedCommit<'_> {
         PreparedCommit {
             index_writer,
             payload: None,
             opstamp,
-            soft_commit,
         }
     }
 
@@ -38,12 +32,11 @@ impl<'a> PreparedCommit<'a> {
 
     pub fn commit(self) -> crate::Result<Opstamp> {
         info!("committing {}", self.opstamp);
-        block_on(self.index_writer.segment_updater().schedule_commit(
-            self.opstamp,
-            self.payload,
-            self.soft_commit,
-        ))?;
-        block_on(self.index_writer.trigger_commit());
+        let _ = block_on(
+            self.index_writer
+                .segment_updater()
+                .schedule_commit(self.opstamp, self.payload),
+        );
         Ok(self.opstamp)
     }
 }

src/indexer/resource_manager.rs

@@ -1,213 +0,0 @@
-use std::ops::RangeBounds;
-use std::sync::{Arc, Condvar, Mutex, MutexGuard, RwLock};
-
-struct LockedData {
-    count: u64,
-    enabled: bool
-}
-
-impl Default for LockedData {
-    fn default() -> Self {
-        LockedData {
-            count: 0u64,
-            enabled: true
-        }
-    }
-}
-
-#[derive(Default)]
-struct Inner {
-    resource_level: Mutex<LockedData>,
-    convdvar: Condvar,
-}
-
-
-/// The resource manager makes it possible to track the amount of level of a given resource.
-/// There is no magic here : it is to the description of the user to declare how much
-/// of the resource is being held.
-///
-/// Allocation of a resource is bound to the lifetime of a `Allocation` instance.
-///
-/// ```rust
-/// let resource_manager = ResourceManager::default();
-///
-/// ```
-///
-/// In tantivy, this is used to check the number of merging thread and the number of memory
-/// used by the volatile segments.
-///
-#[derive(Clone, Default)]
-pub struct ResourceManager {
-    inner: Arc<Inner>,
-}
-
-impl ResourceManager {
-    /// Return the total amount of reousrce allocated
-    pub fn total_amount(&self) -> u64 {
-        self.lock().count
-    }
-
-    fn lock(&self) -> MutexGuard<LockedData> {
-        self.inner
-            .resource_level
-            .lock()
-            .expect("Failed to obtain lock for ReservedMemory. This should never happen.")
-    }
-
-    fn record_delta(&self, delta: i64) {
-        if delta == 0i64 {
-            return;
-        }
-        let mut lock = self.lock();
-        let new_val = lock.count as i64 + delta;
-        lock.count = new_val as u64;
-        self.inner.convdvar.notify_all();
-    }
-
-    /// Records a new allocation.
-    ///
-    /// The returned allocate object is used to automatically release the allocated resource
-    /// on drop.
-    pub fn allocate(&self, amount: u64) -> Allocation {
-        self.record_delta(amount as i64);
-        Allocation {
-            resource_manager: self.clone(),
-            amount: RwLock::new(amount),
-        }
-    }
-
-    /// Stops the resource manager.
-    ///
-    /// If any thread is waiting via `.wait_until_in_range(...)`, the method will stop
-    /// being blocking and will return an error.
-    pub fn terminate(&self) {
-        self.lock().enabled = false;
-        self.inner.convdvar.notify_all();
-    }
-
-    /// Blocks the current thread until the resource level reaches the given range,
-    /// in a cpu-efficient way.
-    ///
-    /// This method does not necessarily wakes up the current thread at every transition
-    /// into the targetted range, but any durable entry in the range will be detected.
-    pub fn wait_until_in_range<R: RangeBounds<u64>>(&self, range: R) -> Result<u64, u64> {
-        let mut levels = self.lock();
-        if !levels.enabled {
-            return Err(levels.count)
-        }
-        while !range.contains(&levels.count) {
-           levels = self.inner.convdvar.wait(levels).unwrap();
-            if !levels.enabled {
-                return Err(levels.count)
-            }
-        }
-        Ok(levels.count)
-    }
-}
-
-pub struct Allocation {
-    resource_manager: ResourceManager,
-    amount: RwLock<u64>,
-}
-
-impl Allocation {
-    pub fn amount(&self) -> u64 {
-        *self.amount.read().unwrap()
-    }
-
-    pub fn modify(&self, new_amount: u64) {
-        let mut wlock = self.amount.write().unwrap();
-        let delta = new_amount as i64 - *wlock as i64;
-        *wlock = new_amount;
-        self.resource_manager.record_delta(delta);
-    }
-}
-
-impl Drop for Allocation {
-    fn drop(&mut self) {
-        let amount = self.amount();
-        self.resource_manager.record_delta(-(amount as i64))
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::ResourceManager;
-    use futures::channel::oneshot;
-    use futures::executor::block_on;
-    use std::{mem, thread};
-
-    #[test]
-    fn test_simple_allocation() {
-        let memory = ResourceManager::default();
-        assert_eq!(memory.total_amount(), 0u64);
-        let _allocation = memory.allocate(10u64);
-        assert_eq!(memory.total_amount(), 10u64);
-    }
-
-    #[test]
-    fn test_multiple_allocation() {
-        let memory = ResourceManager::default();
-        assert_eq!(memory.total_amount(), 0u64);
-        let _allocation = memory.allocate(10u64);
-        let _allocation_2 = memory.allocate(11u64);
-        assert_eq!(memory.total_amount(), 21u64);
-    }
-
-    #[test]
-    fn test_release_on_drop() {
-        let memory = ResourceManager::default();
-        assert_eq!(memory.total_amount(), 0u64);
-        let allocation = memory.allocate(10u64);
-        let allocation_2 = memory.allocate(11u64);
-        assert_eq!(memory.total_amount(), 21u64);
-        mem::drop(allocation);
-        assert_eq!(memory.total_amount(), 11u64);
-        mem::drop(allocation_2);
-        assert_eq!(memory.total_amount(), 0u64);
-    }
-
-    #[test]
-    fn test_wait_until() {
-        let memory = ResourceManager::default();
-        let (send, recv) = oneshot::channel::<()>();
-        let memory_clone = memory.clone();
-        thread::spawn(move || {
-            let _allocation1 = memory_clone.allocate(2u64);
-            let _allocation2 = memory_clone.allocate(3u64);
-            let _allocation3 = memory_clone.allocate(4u64);
-            std::mem::drop(_allocation3);
-            assert!(block_on(recv).is_ok());
-        });
-        assert_eq!(memory.wait_until_in_range(5u64..8u64), Ok(5u64));
-        assert!(send.send(()).is_ok());
-    }
-
-    #[test]
-    fn test_modify_amount() {
-        let memory = ResourceManager::default();
-        let alloc = memory.allocate(2u64);
-        assert_eq!(memory.total_amount(), 2u64);
-        assert_eq!(alloc.amount(), 2u64);
-        let alloc2 = memory.allocate(3u64);
-        assert_eq!(memory.total_amount(), 2u64 + 3u64);
-        assert_eq!(alloc2.amount(), 3u64);
-        alloc.modify(14u64);
-        assert_eq!(alloc.amount(), 14u64);
-        assert_eq!(memory.total_amount(), 14u64 + 3u64)
-    }
-    
-    #[test]
-    fn test_stop_resource_manager() {
-        let resource_manager = ResourceManager::default();
-        let resource_manager_clone = resource_manager.clone();
-        let (sender, recv) = oneshot::channel();
-        let join_handle = thread::spawn(move || {
-            assert!(sender.send(()).is_ok());
-            resource_manager_clone.wait_until_in_range(10..20)
-        });
-        let _ = block_on(recv);
-        resource_manager.terminate();
-        assert_eq!(join_handle.join().unwrap(), Err(0u64));
-    }
-}

src/indexer/segment_entry.rs

@@ -1,9 +1,7 @@
 use crate::common::BitSet;
 use crate::core::SegmentId;
 use crate::core::SegmentMeta;
-use crate::directory::ManagedDirectory;
 use crate::indexer::delete_queue::DeleteCursor;
-use crate::{Opstamp, Segment};
 use std::fmt;
 
 /// A segment entry describes the state of
@@ -21,7 +19,7 @@ use std::fmt;
 /// in the .del file or in the `delete_bitset`.
 #[derive(Clone)]
 pub struct SegmentEntry {
-    segment: Segment,
+    meta: SegmentMeta,
     delete_bitset: Option<BitSet>,
     delete_cursor: DeleteCursor,
 }
@@ -29,67 +27,47 @@ pub struct SegmentEntry {
 impl SegmentEntry {
     /// Create a new `SegmentEntry`
     pub fn new(
-        segment: Segment,
+        segment_meta: SegmentMeta,
         delete_cursor: DeleteCursor,
         delete_bitset: Option<BitSet>,
     ) -> SegmentEntry {
         SegmentEntry {
-            segment,
+            meta: segment_meta,
             delete_bitset,
             delete_cursor,
         }
     }
 
-    pub fn persist(&mut self, dest_directory: ManagedDirectory) -> crate::Result<()> {
-        // TODO take in account delete bitset?
-        self.segment.persist(dest_directory)?;
-        Ok(())
-    }
-
-    pub fn set_delete_cursor(&mut self, delete_cursor: DeleteCursor) {
-        self.delete_cursor = delete_cursor;
-    }
-
-    /// `Takes` (as in Option::take) the delete bitset of a segment entry.
-    /// `DocId` in this bitset are flagged as deleted.
-    pub fn take_delete_bitset(&mut self) -> Option<BitSet> {
-        self.delete_bitset.take()
-    }
-
-    /// Reset the delete information in this segment.
+    /// Return a reference to the segment entry deleted bitset.
     ///
-    /// The `SegmentEntry` segment's `SegmentMeta` gets updated, and
-    /// any delete bitset is drop and set to None.
-    pub fn reset_delete_meta(&mut self, num_deleted_docs: u32, target_opstamp: Opstamp) {
-        self.segment = self
-            .segment
-            .clone()
-            .with_delete_meta(num_deleted_docs, target_opstamp);
-        self.delete_bitset = None;
+    /// `DocId` in this bitset are flagged as deleted.
+    pub fn delete_bitset(&self) -> Option<&BitSet> {
+        self.delete_bitset.as_ref()
+    }
+
+    /// Set the `SegmentMeta` for this segment.
+    pub fn set_meta(&mut self, segment_meta: SegmentMeta) {
+        self.meta = segment_meta;
     }
 
     /// Return a reference to the segment_entry's delete cursor
-    pub fn delete_cursor(&mut self) -> DeleteCursor {
-        self.delete_cursor.clone()
+    pub fn delete_cursor(&mut self) -> &mut DeleteCursor {
+        &mut self.delete_cursor
     }
 
     /// Returns the segment id.
     pub fn segment_id(&self) -> SegmentId {
-        self.meta().id()
+        self.meta.id()
     }
 
-    /// Returns the `segment` associated to the `SegmentEntry`.
-    pub fn segment(&self) -> &Segment {
-        &self.segment
-    }
     /// Accessor to the `SegmentMeta`
     pub fn meta(&self) -> &SegmentMeta {
-        self.segment.meta()
+        &self.meta
     }
 }
 
 impl fmt::Debug for SegmentEntry {
     fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
-        write!(formatter, "SegmentEntry({:?})", self.meta())
+        write!(formatter, "SegmentEntry({:?})", self.meta)
     }
 }

src/indexer/segment_manager.rs

@@ -2,15 +2,15 @@ use super::segment_register::SegmentRegister;
 use crate::core::SegmentId;
 use crate::core::SegmentMeta;
 use crate::error::TantivyError;
-use crate::indexer::{SegmentEntry, MergeOperationInventory, MergeCandidate, MergeOperation};
-use crate::{Segment, Opstamp};
+use crate::indexer::delete_queue::DeleteCursor;
+use crate::indexer::SegmentEntry;
 use std::collections::hash_set::HashSet;
 use std::fmt::{self, Debug, Formatter};
-use std::sync::{Arc, RwLock};
+use std::sync::RwLock;
 use std::sync::{RwLockReadGuard, RwLockWriteGuard};
 
 #[derive(Default)]
-pub(crate) struct SegmentRegisters {
+struct SegmentRegisters {
     uncommitted: SegmentRegister,
     committed: SegmentRegister,
 }
@@ -22,17 +22,6 @@ pub(crate) enum SegmentsStatus {
 }
 
 impl SegmentRegisters {
-    pub fn new(committed: SegmentRegister) -> SegmentRegisters {
-        SegmentRegisters {
-            uncommitted: Default::default(),
-            committed,
-        }
-    }
-
-    pub fn committed_segment(&self) -> Vec<Segment> {
-        self.committed.segments()
-    }
-
     /// 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
@@ -55,8 +44,7 @@ impl SegmentRegisters {
 /// changes (merges especially)
 #[derive(Default)]
 pub struct SegmentManager {
-    registers: Arc<RwLock<SegmentRegisters>>,
-    merge_operations: MergeOperationInventory,
+    registers: RwLock<SegmentRegisters>,
 }
 
 impl Debug for SegmentManager {
@@ -70,28 +58,34 @@ impl Debug for SegmentManager {
     }
 }
 
+pub fn get_mergeable_segments(
+    in_merge_segment_ids: &HashSet<SegmentId>,
+    segment_manager: &SegmentManager,
+) -> (Vec<SegmentMeta>, Vec<SegmentMeta>) {
+    let registers_lock = segment_manager.read();
+    (
+        registers_lock
+            .committed
+            .get_mergeable_segments(in_merge_segment_ids),
+        registers_lock
+            .uncommitted
+            .get_mergeable_segments(in_merge_segment_ids),
+    )
+}
 
 impl SegmentManager {
-    pub(crate) fn new(registers: Arc<RwLock<SegmentRegisters>>) -> SegmentManager {
+    pub fn from_segments(
+        segment_metas: Vec<SegmentMeta>,
+        delete_cursor: &DeleteCursor,
+    ) -> SegmentManager {
         SegmentManager {
-            registers,
-            merge_operations: Default::default()
+            registers: RwLock::new(SegmentRegisters {
+                uncommitted: SegmentRegister::default(),
+                committed: SegmentRegister::new(segment_metas, delete_cursor),
+            }),
         }
     }
 
-    pub fn new_merge_operation(&self, opstamp: Opstamp, merge_candidate: MergeCandidate) -> MergeOperation {
-        MergeOperation::new(
-            &self.merge_operations,
-            opstamp,
-            merge_candidate.0
-
-        )
-    }
-
-    pub fn wait_merging_thread(&self) {
-        self.merge_operations.wait_until_empty()
-    }
-
     /// Returns all of the segment entries (committed or uncommitted)
     pub fn segment_entries(&self) -> Vec<SegmentEntry> {
         let registers_lock = self.read();
@@ -100,34 +94,6 @@ impl SegmentManager {
         segment_entries
     }
 
-    /// Returns the segments that are currently not in merge.
-    ///
-    /// They are split over two `Vec`. The committed segments on one hand
-    /// and the uncommitted ones on the other hand.
-    ///
-    /// This method is useful when searching for merge candidate or segments
-    /// to persists.
-    pub fn segments_not_in_merge(&self) -> (Vec<SegmentMeta>, Vec<SegmentMeta>) {
-        let in_merge_segment_ids: HashSet<SegmentId> = self.merge_operations.segment_in_merge();
-        let registers_lock = self.read();
-        (
-            registers_lock
-                .committed
-                .get_mergeable_segments(&in_merge_segment_ids),
-            registers_lock
-                .uncommitted
-                .get_mergeable_segments(&in_merge_segment_ids),
-        )
-    }
-
-    pub fn largest_segment_not_in_merge(&self) -> Option<SegmentMeta> {
-        let (committed, uncommitted) = self.segments_not_in_merge();
-        let mut segments = vec![];
-        segments.extend(committed);
-        segments.extend(uncommitted);
-        None
-    }
-
     // Lock poisoning should never happen :
     // The lock is acquired and released within this class,
     // and the operations cannot panic.

src/indexer/segment_register.rs

@@ -1,10 +1,7 @@
 use crate::core::SegmentId;
 use crate::core::SegmentMeta;
-use crate::directory::ManagedDirectory;
 use crate::indexer::delete_queue::DeleteCursor;
 use crate::indexer::segment_entry::SegmentEntry;
-use crate::schema::Schema;
-use crate::Segment;
 use std::collections::HashMap;
 use std::collections::HashSet;
 use std::fmt::{self, Debug, Formatter};
@@ -49,13 +46,6 @@ impl SegmentRegister {
             .collect()
     }
 
-    pub fn segments(&self) -> Vec<Segment> {
-        self.segment_states
-            .values()
-            .map(|segment_entry| segment_entry.segment().clone())
-            .collect()
-    }
-
     pub fn segment_entries(&self) -> Vec<SegmentEntry> {
         self.segment_states.values().cloned().collect()
     }
@@ -89,17 +79,11 @@ impl SegmentRegister {
         self.segment_states.get(segment_id).cloned()
     }
 
-    pub fn new(
-        directory: &ManagedDirectory,
-        schema: &Schema,
-        segment_metas: Vec<SegmentMeta>,
-        delete_cursor: &DeleteCursor,
-    ) -> SegmentRegister {
+    pub fn new(segment_metas: Vec<SegmentMeta>, delete_cursor: &DeleteCursor) -> SegmentRegister {
         let mut segment_states = HashMap::new();
         for segment_meta in segment_metas {
             let segment_id = segment_meta.id();
-            let segment = Segment::new_persisted(segment_meta, directory.clone(), schema.clone());
-            let segment_entry = SegmentEntry::new(segment, delete_cursor.clone(), None);
+            let segment_entry = SegmentEntry::new(segment_meta, delete_cursor.clone(), None);
             segment_states.insert(segment_id, segment_entry);
         }
         SegmentRegister { segment_states }
@@ -111,7 +95,6 @@ mod tests {
     use super::*;
     use crate::core::{SegmentId, SegmentMetaInventory};
     use crate::indexer::delete_queue::*;
-    use crate::indexer::ResourceManager;
 
     fn segment_ids(segment_register: &SegmentRegister) -> Vec<SegmentId> {
         segment_register
@@ -125,34 +108,28 @@ mod tests {
     fn test_segment_register() {
         let inventory = SegmentMetaInventory::default();
         let delete_queue = DeleteQueue::new();
-        let schema = Schema::builder().build();
 
         let mut segment_register = SegmentRegister::default();
         let segment_id_a = SegmentId::generate_random();
         let segment_id_b = SegmentId::generate_random();
         let segment_id_merged = SegmentId::generate_random();
 
-        let memory_manager = ResourceManager::default();
-
         {
-            let meta = inventory.new_segment_meta(segment_id_a, 0u32);
-            let segment = Segment::new_volatile(meta, schema.clone(), memory_manager.clone());
-            let segment_entry = SegmentEntry::new(segment, delete_queue.cursor(), None);
+            let segment_meta = inventory.new_segment_meta(segment_id_a, 0u32);
+            let segment_entry = SegmentEntry::new(segment_meta, delete_queue.cursor(), None);
             segment_register.add_segment_entry(segment_entry);
         }
         assert_eq!(segment_ids(&segment_register), vec![segment_id_a]);
         {
-            let meta = inventory.new_segment_meta(segment_id_b, 0u32);
-            let segment = Segment::new_volatile(meta, schema.clone(), memory_manager.clone());
-            let segment_entry = SegmentEntry::new(segment, delete_queue.cursor(), None);
+            let segment_meta = inventory.new_segment_meta(segment_id_b, 0u32);
+            let segment_entry = SegmentEntry::new(segment_meta, delete_queue.cursor(), None);
             segment_register.add_segment_entry(segment_entry);
         }
         segment_register.remove_segment(&segment_id_a);
         segment_register.remove_segment(&segment_id_b);
         {
             let segment_meta_merged = inventory.new_segment_meta(segment_id_merged, 0u32);
-            let segment_merged = Segment::new_volatile(segment_meta_merged, schema.clone(), memory_manager.clone());
-            let segment_entry = SegmentEntry::new(segment_merged, delete_queue.cursor(), None);
+            let segment_entry = SegmentEntry::new(segment_meta_merged, delete_queue.cursor(), None);
             segment_register.add_segment_entry(segment_entry);
         }
         assert_eq!(segment_ids(&segment_register), vec![segment_id_merged]);

src/indexer/segment_serializer.rs

@@ -3,32 +3,43 @@ use crate::core::SegmentComponent;
 use crate::fastfield::FastFieldSerializer;
 use crate::fieldnorm::FieldNormsSerializer;
 use crate::postings::InvertedIndexSerializer;
+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) -> crate::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)?;
         let fast_field_serializer = FastFieldSerializer::from_write(fast_field_write)?;
 
         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
@@ -39,16 +50,26 @@ impl SegmentSerializer {
         &mut self.fast_field_serializer
     }
 
-    /// Accessor to the field norm serializer.
-    pub fn get_fieldnorms_serializer(&mut self) -> &mut FieldNormsSerializer {
-        &mut self.fieldnorms_serializer
+    /// Extract the field norm 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`.
+    pub fn get_store_writer(&mut self) -> &mut StoreWriter {
+        &mut self.store_writer
     }
 
     /// Finalize the segment serialization.
-    pub fn close(self) -> crate::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.fieldnorms_serializer.close()?;
+        self.store_writer.close()?;
         Ok(())
     }
 }

src/indexer/segment_updater.rs

@@ -1,4 +1,4 @@
-use super::segment_manager::SegmentManager;
+use super::segment_manager::{get_mergeable_segments, SegmentManager};
 use crate::core::Index;
 use crate::core::IndexMeta;
 use crate::core::Segment;
@@ -7,21 +7,22 @@ use crate::core::SegmentMeta;
 use crate::core::SerializableSegment;
 use crate::core::META_FILEPATH;
 use crate::directory::{Directory, DirectoryClone, GarbageCollectionResult};
+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::{SegmentRegisters, SegmentsStatus};
+use crate::indexer::segment_manager::SegmentsStatus;
 use crate::indexer::stamper::Stamper;
 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, SegmentComponent};
+use crate::Opstamp;
 use futures::channel::oneshot;
 use futures::executor::{ThreadPool, ThreadPoolBuilder};
 use futures::future::Future;
 use futures::future::TryFutureExt;
-use serde_json;
 use std::borrow::BorrowMut;
 use std::collections::HashSet;
 use std::io::Write;
@@ -111,36 +112,33 @@ fn merge(
     target_opstamp: 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();
 
     // First we apply all of the delet to the merged segment, up to the target opstamp.
     for segment_entry in &mut segment_entries {
-        advance_deletes(segment_entry, target_opstamp)?;
+        let segment = index.segment(segment_entry.meta().clone());
+        advance_deletes(segment, segment_entry, target_opstamp)?;
     }
 
-    let delete_cursor = segment_entries[0].delete_cursor();
+    let delete_cursor = segment_entries[0].delete_cursor().clone();
 
     let segments: Vec<Segment> = segment_entries
         .iter()
-        .map(|segment_entry| segment_entry.segment().clone())
+        .map(|segment_entry| index.segment(segment_entry.meta().clone()))
         .collect();
 
     // An IndexMerger is like a "view" of our merged segments.
     let merger: IndexMerger = IndexMerger::open(index.schema(), &segments[..])?;
 
     // ... we just serialize this index merger in our new segment to merge the two segments.
-    let segment_serializer = SegmentSerializer::for_segment(&mut merged_segment)?;
+    let segment_serializer = SegmentSerializer::for_segment(merged_segment.clone())?;
 
-    let store_wrt = merged_segment.open_write(SegmentComponent::STORE)?;
-    merger.write_storable_fields(store_wrt)?;
+    let num_docs = merger.write(segment_serializer)?;
 
-    let max_doc = merger.write(segment_serializer)?;
+    let merged_segment_id = merged_segment.id();
 
-    Ok(SegmentEntry::new(
-        merged_segment.with_max_doc(max_doc),
-        delete_cursor,
-        None,
-    ))
+    let segment_meta = index.new_segment_meta(merged_segment_id, num_docs);
+    Ok(SegmentEntry::new(segment_meta, delete_cursor, None))
 }
 
 pub(crate) struct InnerSegmentUpdater {
@@ -159,15 +157,17 @@ pub(crate) struct InnerSegmentUpdater {
     merge_policy: RwLock<Arc<Box<dyn MergePolicy>>>,
     killed: AtomicBool,
     stamper: Stamper,
+    merge_operations: MergeOperationInventory,
 }
 
 impl SegmentUpdater {
     pub fn create(
-        segment_registers: Arc<RwLock<SegmentRegisters>>,
         index: Index,
         stamper: Stamper,
+        delete_cursor: &DeleteCursor,
     ) -> crate::Result<SegmentUpdater> {
-        let segment_manager = SegmentManager::new(segment_registers);
+        let segments = index.searchable_segment_metas()?;
+        let segment_manager = SegmentManager::from_segments(segments, delete_cursor);
         let pool = ThreadPoolBuilder::new()
             .name_prefix("segment_updater")
             .pool_size(1)
@@ -196,6 +196,7 @@ impl SegmentUpdater {
             merge_policy: RwLock::new(Arc::new(Box::new(DefaultMergePolicy::default()))),
             killed: AtomicBool::new(false),
             stamper,
+            merge_operations: Default::default(),
         })))
     }
 
@@ -261,7 +262,8 @@ impl SegmentUpdater {
     fn purge_deletes(&self, target_opstamp: Opstamp) -> crate::Result<Vec<SegmentEntry>> {
         let mut segment_entries = self.segment_manager.segment_entries();
         for segment_entry in &mut segment_entries {
-            advance_deletes(segment_entry, target_opstamp)?;
+            let segment = self.index.segment(segment_entry.meta().clone());
+            advance_deletes(segment, segment_entry, target_opstamp)?;
         }
         Ok(segment_entries)
     }
@@ -329,21 +331,12 @@ impl SegmentUpdater {
         &self,
         opstamp: Opstamp,
         payload: Option<String>,
-        soft_commit: bool,
     ) -> impl Future<Output = crate::Result<()>> {
         let segment_updater: SegmentUpdater = self.clone();
-        let directory = self.index.directory().clone();
         self.schedule_future(async move {
-            let mut segment_entries = segment_updater.purge_deletes(opstamp)?;
-            if !soft_commit {
-                for segment_entry in &mut segment_entries {
-                    segment_entry.persist(directory.clone())?;
-                }
-            }
+            let segment_entries = segment_updater.purge_deletes(opstamp)?;
             segment_updater.segment_manager.commit(segment_entries);
-            if !soft_commit {
-                segment_updater.save_metas(opstamp, payload)?;
-            }
+            segment_updater.save_metas(opstamp, payload)?;
             let _ = garbage_collect_files(segment_updater.clone()).await;
             segment_updater.consider_merge_options().await;
             Ok(())
@@ -360,7 +353,7 @@ impl SegmentUpdater {
 
     pub(crate) fn make_merge_operation(&self, segment_ids: &[SegmentId]) -> MergeOperation {
         let commit_opstamp = self.load_metas().opstamp;
-        self.segment_manager.new_merge_operation(commit_opstamp, MergeCandidate(segment_ids.to_vec()))
+        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
@@ -434,8 +427,9 @@ impl SegmentUpdater {
     }
 
     async fn consider_merge_options(&self) {
+        let merge_segment_ids: HashSet<SegmentId> = self.merge_operations.segment_in_merge();
         let (committed_segments, uncommitted_segments) =
-            self.segment_manager.segments_not_in_merge();
+            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.
@@ -446,7 +440,7 @@ impl SegmentUpdater {
             .compute_merge_candidates(&uncommitted_segments)
             .into_iter()
             .map(|merge_candidate| {
-                self.segment_manager.new_merge_operation(current_opstamp, merge_candidate)
+                MergeOperation::new(&self.merge_operations, current_opstamp, merge_candidate.0)
             })
             .collect();
 
@@ -455,7 +449,7 @@ impl SegmentUpdater {
             .compute_merge_candidates(&committed_segments)
             .into_iter()
             .map(|merge_candidate: MergeCandidate| {
-                self.segment_manager.new_merge_operation(commit_opstamp, merge_candidate)
+                MergeOperation::new(&self.merge_operations, commit_opstamp, merge_candidate.0)
             })
             .collect::<Vec<_>>();
         merge_candidates.extend(committed_merge_candidates.into_iter());
@@ -480,13 +474,17 @@ impl SegmentUpdater {
         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();
+                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 {
-                        if let Err(e) =
-                            advance_deletes(&mut after_merge_segment_entry, committed_opstamp)
-                        {
+                        let index = &segment_updater.index;
+                        let segment = index.segment(after_merge_segment_entry.meta().clone());
+                        if let Err(e) = advance_deletes(
+                            segment,
+                            &mut after_merge_segment_entry,
+                            committed_opstamp,
+                        ) {
                             error!(
                                 "Merge of {:?} was cancelled (advancing deletes failed): {:?}",
                                 merge_operation.segment_ids(),
@@ -524,7 +522,7 @@ impl SegmentUpdater {
     ///
     /// Upon termination of the current merging threads,
     /// merge opportunity may appear.
-    //
+    ///
     /// We keep waiting until the merge policy judges that
     /// no opportunity is available.
     ///
@@ -535,8 +533,9 @@ impl SegmentUpdater {
     ///
     /// Obsolete files will eventually be cleaned up
     /// by the directory garbage collector.
-    pub fn wait_merging_thread(&self) {
-        self.segment_manager.wait_merging_thread()
+    pub fn wait_merging_thread(&self) -> crate::Result<()> {
+        self.merge_operations.wait_until_empty();
+        Ok(())
     }
 }
 
@@ -545,8 +544,7 @@ mod tests {
 
     use crate::indexer::merge_policy::tests::MergeWheneverPossible;
     use crate::schema::*;
-    use crate::{Index, SegmentId};
-    use futures::executor::block_on;
+    use crate::Index;
 
     #[test]
     fn test_delete_during_merge() {
@@ -697,27 +695,4 @@ mod tests {
             .segment_entries();
         assert!(seg_vec.is_empty());
     }
-
-    #[test]
-    fn test_merge_over_soft_commit() {
-        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);
-        // writing the segment
-        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-        index_writer.add_document(doc!(text_field=>"a"));
-        assert!(index_writer.soft_commit().is_ok());
-        index_writer.add_document(doc!(text_field=>"a"));
-        assert!(index_writer.soft_commit().is_ok());
-
-        let reader = index_writer.reader(1).unwrap();
-        let segment_ids: Vec<SegmentId> = reader
-            .searcher()
-            .segment_readers()
-            .iter()
-            .map(|reader| reader.segment_id())
-            .collect();
-        assert!(block_on(index_writer.merge(&segment_ids)).is_ok());
-    }
 }

src/indexer/segment_writer.rs

@@ -1,11 +1,9 @@
 use super::operation::AddOperation;
 use crate::core::Segment;
 use crate::core::SerializableSegment;
-use crate::directory::{SpillingResult, SpillingWriter, TerminatingWrite};
 use crate::fastfield::FastFieldsWriter;
-use crate::fieldnorm::FieldNormsWriter;
+use crate::fieldnorm::{FieldNormReaders, FieldNormsWriter};
 use crate::indexer::segment_serializer::SegmentSerializer;
-use crate::indexer::IndexWriterConfig;
 use crate::postings::compute_table_size;
 use crate::postings::MultiFieldPostingsWriter;
 use crate::schema::FieldType;
@@ -13,24 +11,21 @@ use crate::schema::Schema;
 use crate::schema::Term;
 use crate::schema::Value;
 use crate::schema::{Field, FieldEntry};
-use crate::store::StoreWriter;
 use crate::tokenizer::{BoxTokenStream, PreTokenizedStream};
-use crate::tokenizer::{FacetTokenizer, TextAnalyzer, TokenizerManager};
+use crate::tokenizer::{FacetTokenizer, TextAnalyzer};
 use crate::tokenizer::{TokenStreamChain, Tokenizer};
 use crate::Opstamp;
 use crate::{DocId, SegmentComponent};
 use std::io;
-use std::io::Write;
 use std::str;
-use crate::indexer::resource_manager::ResourceManager;
 
 /// 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: u64) -> crate::Result<usize> {
-    let table_memory_upper_bound = per_thread_memory_budget / 3u64;
+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| compute_table_size(*num_bits) < table_memory_upper_bound)
+        .take_while(|num_bits: &usize| compute_table_size(*num_bits) < table_memory_upper_bound)
         .last()
     {
         Ok(limit.min(19)) // we cap it at 2^19 = 512K.
@@ -48,13 +43,11 @@ fn initial_table_size(per_thread_memory_budget: u64) -> crate::Result<usize> {
 pub struct SegmentWriter {
     max_doc: DocId,
     multifield_postings: MultiFieldPostingsWriter,
-    segment: Segment,
+    segment_serializer: SegmentSerializer,
     fast_field_writers: FastFieldsWriter,
     fieldnorms_writer: FieldNormsWriter,
     doc_opstamps: Vec<Opstamp>,
     tokenizers: Vec<Option<TextAnalyzer>>,
-    store_writer: StoreWriter<SpillingWriter>,
-    memory_manager: ResourceManager,
 }
 
 impl SegmentWriter {
@@ -68,13 +61,13 @@ impl SegmentWriter {
     /// - segment: The segment being written
     /// - schema
     pub fn for_segment(
-        config: &IndexWriterConfig,
+        memory_budget: usize,
         segment: Segment,
         schema: &Schema,
-        tokenizer_manager: &TokenizerManager,
-        memory_manager: ResourceManager
     ) -> crate::Result<SegmentWriter> {
-        let table_num_bits = initial_table_size(config.heap_size_in_byte_per_thread())?;
+        let tokenizer_manager = segment.index().tokenizers().clone();
+        let table_num_bits = initial_table_size(memory_budget)?;
+        let segment_serializer = SegmentSerializer::for_segment(segment)?;
         let multifield_postings = MultiFieldPostingsWriter::new(schema, table_num_bits);
         let tokenizers = schema
             .fields()
@@ -90,26 +83,14 @@ impl SegmentWriter {
                 },
             )
             .collect();
-        let mut segment_clone = segment.clone();
-        let spilling_wrt = SpillingWriter::new(
-            50_000_000,
-            Box::new(move || {
-                segment_clone
-                    .open_write(SegmentComponent::STORE)
-                    .map_err(|err| io::Error::new(io::ErrorKind::Other, err))
-            }),
-        );
-        let store_writer = StoreWriter::new(spilling_wrt);
         Ok(SegmentWriter {
             max_doc: 0,
             multifield_postings,
             fieldnorms_writer: FieldNormsWriter::for_schema(schema),
-            segment,
+            segment_serializer,
             fast_field_writers: FastFieldsWriter::from_schema(schema),
             doc_opstamps: Vec::with_capacity(1_000),
             tokenizers,
-            store_writer,
-            memory_manager
         })
     }
 
@@ -117,33 +98,18 @@ impl SegmentWriter {
     ///
     /// Finalize consumes the `SegmentWriter`, so that it cannot
     /// be used afterwards.
-    pub fn finalize(mut self) -> crate::Result<(Segment, Vec<u64>)> {
+    pub fn finalize(mut self) -> crate::Result<Vec<u64>> {
         self.fieldnorms_writer.fill_up_to_max_doc(self.max_doc);
-        let spilling_wrt = self.store_writer.close()?;
-        let mut segment: Segment;
-        match spilling_wrt.finalize()? {
-            SpillingResult::Spilled => {
-                segment = self.segment.clone();
-            }
-            SpillingResult::Buffer(buf) => {
-                segment = self.segment.into_volatile(self.memory_manager.clone());
-                let mut store_wrt = segment.open_write(SegmentComponent::STORE)?;
-                store_wrt.write_all(&buf[..])?;
-                store_wrt.terminate()?;
-            }
-        }
-        let segment_serializer = SegmentSerializer::for_segment(&mut segment)?;
-        segment = segment.with_max_doc(self.max_doc);
         write(
             &self.multifield_postings,
             &self.fast_field_writers,
             &self.fieldnorms_writer,
-            segment_serializer,
+            self.segment_serializer,
         )?;
-        Ok((segment, self.doc_opstamps))
+        Ok(self.doc_opstamps)
     }
 
-    pub fn mem_usage(&self) -> u64 {
+    pub fn mem_usage(&self) -> usize {
         self.multifield_postings.mem_usage()
     }
 
@@ -281,7 +247,8 @@ impl SegmentWriter {
         }
         doc.filter_fields(|field| schema.get_field_entry(field).is_stored());
         doc.prepare_for_store();
-        self.store_writer.store(&doc)?;
+        let doc_writer = self.segment_serializer.get_store_writer();
+        doc_writer.store(&doc)?;
         self.max_doc += 1;
         Ok(())
     }
@@ -314,9 +281,16 @@ fn write(
     fieldnorms_writer: &FieldNormsWriter,
     mut serializer: SegmentSerializer,
 ) -> 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(())
 }

src/lib.rs

@@ -156,14 +156,14 @@ 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::{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, IndexWriterConfig};
+pub use crate::indexer::IndexWriter;
 pub use crate::postings::Postings;
 pub use crate::reader::LeasedItem;
 pub use crate::schema::{Document, Term};
@@ -173,7 +173,7 @@ use once_cell::sync::Lazy;
 use serde::{Deserialize, Serialize};
 
 /// Index format version.
-const INDEX_FORMAT_VERSION: u32 = 1;
+const INDEX_FORMAT_VERSION: u32 = 2;
 
 /// Structure version for the index.
 #[derive(Clone, PartialEq, Eq, Serialize, Deserialize)]
@@ -245,11 +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
-/// larger the number, the more relevant the document
-/// to the search
+/// This is modelled internally as a `f64`, because tantivy was compiled with the `scoref64`
+/// feature. The larger the number, the more relevant the document is to the search query.
+#[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.
@@ -282,10 +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;
@@ -298,17 +304,26 @@ mod tests {
     use rand::rngs::StdRng;
     use rand::{Rng, SeedableRng};
 
-    pub fn assert_nearly_equals(expected: f32, val: f32) {
-        assert!(
-            nearly_equals(val, expected),
-            "Got {}, expected {}.",
-            val,
-            expected
-        );
-    }
-
-    pub fn nearly_equals(a: f32, b: f32) -> bool {
-        (a - b).abs() < 0.0005 * (a + b).abs()
+    /// Checks if left and right are close one to each other.
+    /// Panics if the two values are more than 0.5% apart.
+    #[macro_export]
+    macro_rules! assert_nearly_equals {
+        ($left:expr, $right:expr) => {{
+            match (&$left, &$right) {
+                (left_val, right_val) => {
+                    let diff = (left_val - right_val).abs();
+                    let add = left_val.abs() + right_val.abs();
+                    if diff > 0.0005 * add {
+                        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> {
@@ -381,19 +396,12 @@ mod tests {
             index_writer.commit().unwrap();
         }
         {
-            {
-                let doc = doc!(text_field=>"a");
-                index_writer.add_document(doc);
-            }
-            {
-                let doc = doc!(text_field=>"a a");
-                index_writer.add_document(doc);
-            }
+            index_writer.add_document(doc!(text_field=>"a"));
+            index_writer.add_document(doc!(text_field=>"a a"));
             index_writer.commit().unwrap();
         }
         {
-            let doc = doc!(text_field=>"c");
-            index_writer.add_document(doc);
+            index_writer.add_document(doc!(text_field=>"c"));
             index_writer.commit().unwrap();
         }
         {
@@ -472,10 +480,12 @@ mod tests {
     }
 
     fn advance_undeleted(docset: &mut dyn DocSet, reader: &SegmentReader) -> bool {
-        while docset.advance() {
-            if !reader.is_deleted(docset.doc()) {
+        let mut doc = docset.advance();
+        while doc != TERMINATED {
+            if !reader.is_deleted(doc) {
                 return true;
             }
+            doc = docset.advance();
         }
         false
     }
@@ -641,9 +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]
@@ -665,9 +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]
@@ -689,9 +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]
@@ -760,10 +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");
-                index_writer.add_document(doc);
-            }
+            let doc = doc!(text_field=>"af af af bc bc");
+            index_writer.add_document(doc);
             index_writer.commit().unwrap();
         }
         {
@@ -779,10 +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);
         }
     }
 

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() {
-        let v: Vec<u32> = (0..1_000).collect();
+    fn test_position_read_with_offset() {
+        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 &n in &[10, 127, COMPRESSION_BLOCK_SIZE, 130, 312] {
-            let mut v = vec![0u32; n];
-            position_reader.read(&mut v[..n]);
-            for i in 0..n {
-                assert_eq!(v[i], 10u32 + i as u32);
+        for &offset in &[1u64, 10u64, 127u64, 128u64, 130u64, 312u64] {
+            for &len in &[1, 10, 130, 500] {
+                let mut v = vec![0u32; len];
+                position_reader.read(offset, &mut v[..]);
+                for i in 0..len {
+                    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(&mut buf);
-            position_reader.skip(4);
-            position_reader.skip(3);
+            position_reader.read(offset, &mut buf);
+            position_reader.read(offset, &mut buf);
+            offset += 7;
             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.
 ///
@@ -62,74 +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,
-        };
-        position_reader.skip(small_skip);
-        position_reader
+            block_offset: std::i64::MAX as u64,
+            anchor_offset: (long_skip_id as u64) * LONG_SKIP_INTERVAL,
+            abs_offset: offset,
+        }
     }
 }
 
+#[derive(Clone)]
 pub struct PositionReader {
     skip_read: OwnedRead,
     position_read: OwnedRead,
     bit_packer: BitPacker4x,
-    inner_offset: usize,
-    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.
-}
+    buffer: Box<[u32; COMPRESSION_BLOCK_SIZE]>,
 
-// `ahead` represents the offset of the block currently loaded
-// compared to the cursor of the actual stream.
-//
-// By contract, when this function is called, the current block has to be
-// 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;
-    }
+    block_offset: u64,
+    anchor_offset: u64,
+
+    abs_offset: u64,
 }
 
 impl PositionReader {
@@ -141,57 +100,65 @@ impl PositionReader {
         Positions::new(position_source, skip_source).reader(offset)
     }
 
-    /// Fills a buffer with the next `output.len()` integers.
-    /// This does not consume / advance the stream.
-    pub fn read(&mut self, output: &mut [u32]) {
-        let skip_data = self.skip_read.as_ref();
-        let position_data = self.position_read.as_ref();
-        let num_bits = self.skip_read.get(0);
-        if self.ahead != Some(0) {
-            // the block currently available is not the block
-            // for the current position
+    fn advance_num_blocks(&mut self, num_blocks: usize) {
+        let num_bits: usize = self.skip_read.as_ref()[..num_blocks]
+            .iter()
+            .cloned()
+            .map(|num_bits| num_bits as usize)
+            .sum();
+        let num_bytes_to_skip = num_bits * COMPRESSION_BLOCK_SIZE / 8;
+        self.skip_read.advance(num_blocks as usize);
+        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)
@@ -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
-    /// of branch.
-    pub(crate) fn search_in_block(
-        self,
-        block_docs: &AlignedBuffer,
-        len: usize,
-        start: usize,
-        target: u32,
-    ) -> usize {
+    /// the block regardless of whether the block is full or not.
+    ///
+    /// Indeed, if the block is not full, the remaining items are TERMINATED.
+    /// It is surprisingly faster, most likely because of the lack of branch misprediction.
+    pub(crate) fn search_in_block(self, block_docs: &AlignedBuffer, target: u32) -> usize {
         #[cfg(target_arch = "x86_64")]
         {
-            use crate::postings::compression::COMPRESSION_BLOCK_SIZE;
-            if self == BlockSearcher::SSE2 && len == COMPRESSION_BLOCK_SIZE {
+            if self == BlockSearcher::SSE2 {
                 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!(
-                block_searcher.search_in_block(
-                    &AlignedBuffer(output_buffer),
-                    block.len(),
-                    i,
-                    target
-                ),
-                cursor
-            );
-        }
+        assert_eq!(
+            block_searcher.search_in_block(&AlignedBuffer(output_buffer), 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 {
-    pub fn new() -> BlockDecoder {
+impl Default for 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) {
-        (&self.output, self.output_len)
+    pub(crate) fn output_aligned(&self) -> &AlignedBuffer {
+        &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() {
-        {
-            let expected_length = 154;
-            let mut encoder = BlockEncoder::new();
-            let input: Vec<u32> = (0u32..123u32).map(|i| 4 + i * 7 / 2).into_iter().collect();
-            for offset in &[0u32, 1u32, 2u32] {
-                let encoded_data = encoder.compress_vint_sorted(&input, *offset);
-                assert!(encoded_data.len() <= expected_length);
-                let mut decoder = BlockDecoder::new();
-                let consumed_num_bytes =
-                    decoder.uncompress_vint_sorted(&encoded_data, *offset, input.len());
-                assert_eq!(consumed_num_bytes, encoded_data.len());
-                assert_eq!(input, decoder.output_array());
+        const PADDING_VALUE: u32 = 234_234_345u32;
+        let expected_length = 154;
+        let mut encoder = BlockEncoder::new();
+        let input: Vec<u32> = (0u32..123u32).map(|i| 4 + i * 7 / 2).into_iter().collect();
+        for offset in &[0u32, 1u32, 2u32] {
+            let encoded_data = encoder.compress_vint_sorted(&input, *offset);
+            assert!(encoded_data.len() <= expected_length);
+            let mut decoder = BlockDecoder::default();
+            let consumed_num_bytes =
+                decoder.uncompress_vint_sorted(&encoded_data, *offset, input.len(), PADDING_VALUE);
+            assert_eq!(consumed_num_bytes, encoded_data.len());
+            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))]
@@ -51,10 +47,10 @@ 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::{IndexWriterConfig, SegmentWriter, ResourceManager};
+    use crate::indexer::SegmentWriter;
     use crate::merge_policy::NoMergePolicy;
     use crate::query::Scorer;
     use crate::schema::{Document, Schema, Term, INDEXED, STRING, TEXT};
@@ -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 field_serializer = posting_serializer.new_field(text_field, 120 * 4).unwrap();
-            field_serializer.new_term("abc".as_bytes()).unwrap();
-            for doc_id in 0u32..120u32 {
-                let delta_positions = vec![1, 2, 3, 2];
-                field_serializer
-                    .write_doc(doc_id, 4, &delta_positions)
-                    .unwrap();
-            }
-            field_serializer.close_term().unwrap();
+        let mut posting_serializer = InvertedIndexSerializer::open(&mut segment)?;
+        let mut field_serializer = posting_serializer.new_field(text_field, 120 * 4, None)?;
+        field_serializer.new_term("abc".as_bytes(), 12u32)?;
+        for doc_id in 0u32..120u32 {
+            let delta_positions = vec![1, 2, 3, 2];
+            field_serializer.write_doc(doc_id, 4, &delta_positions)?;
         }
-        posting_serializer.close().unwrap();
-        let read = segment.open_read(SegmentComponent::POSITIONS).unwrap();
+        field_serializer.close_term()?;
+        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();
-            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.advance(), 1);
             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.skip_next(1002), SkipResult::Reached);
+            assert_eq!(postings.seek(100), 100);
+            assert_eq!(postings.seek(1002), 1002);
             assert_eq!(postings.doc(), 1002);
             postings.positions(&mut positions);
             assert_eq!(&[0, 5], &positions[..]);
         }
+        Ok(())
     }
 
     #[test]
@@ -216,17 +211,11 @@ pub mod tests {
         let text_field = schema_builder.add_text_field("text", TEXT);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema.clone());
+        let segment = index.new_segment();
 
-        let resource_manager = ResourceManager::default();
-        let segment = {
-            let mut segment_writer = SegmentWriter::for_segment(
-                &IndexWriterConfig::for_test(),
-                index.new_segment(),
-                &schema,
-                index.tokenizers(),
-                resource_manager
-            )
-            .unwrap();
+        {
+            let mut segment_writer =
+                SegmentWriter::for_segment(3_000_000, segment.clone(), &schema).unwrap();
             {
                 let mut doc = Document::default();
                 // checking that position works if the field has two values
@@ -258,9 +247,8 @@ pub mod tests {
                 };
                 segment_writer.add_document(op, &schema).unwrap();
             }
-            let (segment, _) = segment_writer.finalize().unwrap();
-            segment
-        };
+            segment_writer.finalize().unwrap();
+        }
         {
             let segment_reader = SegmentReader::open(&segment).unwrap();
             {
@@ -288,22 +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");
@@ -313,7 +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);
@@ -321,8 +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);
             }
         }
     }
@@ -336,16 +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();
-            {
-                let mut doc = Document::default();
-                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);
-            }
+            index_writer.add_document(doc!(text_field => "g b b d c g c"));
+            index_writer.add_document(doc!(text_field => "g a b b a d c g c"));
             assert!(index_writer.commit().is_ok());
         }
         let term_a = Term::from_field_text(text_field, "a");
@@ -355,7 +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]);
@@ -377,11 +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 {
-                    let mut doc = Document::default();
-                    doc.add_u64(value_field, 2);
-                    doc.add_u64(value_field, (i % 2) as u64);
-
+                for i in 0u64..num_docs as u64 {
+                    let doc = doc!(value_field => 2u64, value_field => i % 2u64);
                     index_writer.add_document(doc);
                 }
                 assert!(index_writer.commit().is_ok());
@@ -398,11 +373,10 @@ pub mod tests {
                     .inverted_index(term_2.field())
                     .read_postings(&term_2, IndexRecordOption::Basic)
                     .unwrap();
-
-                assert_eq!(segment_postings.skip_next(i), SkipResult::Reached);
+                assert_eq!(segment_postings.seek(i), i);
                 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);
             }
         }
@@ -414,17 +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.doc(), 2);
+            assert_eq!(segment_postings.seek(1), 1);
+            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
@@ -435,7 +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);
             }
 
@@ -445,7 +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);
             }
         }
@@ -457,6 +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
@@ -467,11 +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);
             }
         }
@@ -486,12 +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) {
-                    SkipResult::End => break,
-                    SkipResult::Reached => assert_eq!(segment_postings.doc(), cur),
-                    SkipResult::OverStep => assert_eq!(segment_postings.doc(), cur + 1),
+                let seek = segment_postings.seek(cur);
+                if seek == TERMINATED {
+                    break;
+                }
+                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;
@@ -577,7 +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()
         }
 
@@ -602,31 +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_unopt = postings_unopt.skip_next(target);
+            let skip_result_opt = postings_opt.seek(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 {
-                SkipResult::Reached => assert_eq!(postings_opt.doc(), target),
-                SkipResult::OverStep => assert!(postings_opt.doc() > target),
-                SkipResult::End => {
-                    return;
-                }
+            assert!(skip_result_opt >= target);
+            assert_eq!(skip_result_opt, postings_opt.doc());
+            if skip_result_opt == TERMINATED {
+                return;
             }
-            while postings_opt.advance() {
-                assert!(postings_unopt.advance());
-                assert_eq!(
-                    postings_opt.doc(),
-                    postings_unopt.doc(),
-                    "Failed while skipping to {}",
-                    target
-                );
+            while postings_opt.doc() != TERMINATED {
+                assert_eq!(postings_opt.doc(), postings_unopt.doc());
+                assert_eq!(postings_opt.advance(), postings_unopt.advance());
             }
-            assert!(!postings_unopt.advance());
         }
     }
 }
@@ -635,7 +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;
@@ -653,7 +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 {}
         });
     }
 
@@ -685,7 +658,7 @@ mod bench {
                 segment_postings_c,
                 segment_postings_d,
             ]);
-            while intersection.advance() {}
+            while intersection.advance() != TERMINATED {}
         });
     }
 
@@ -701,11 +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;
                 }
             }
@@ -717,7 +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;
                 }
             }
@@ -756,8 +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,
 };
@@ -95,7 +96,7 @@ impl MultiFieldPostingsWriter {
         }
     }
 
-    pub fn mem_usage(&self) -> u64 {
+    pub fn mem_usage(&self) -> usize {
         self.term_index.mem_usage() + self.heap.mem_usage()
     }
 
@@ -128,6 +129,7 @@ impl MultiFieldPostingsWriter {
     pub fn serialize(
         &self,
         serializer: &mut InvertedIndexSerializer,
+        fieldnorm_readers: FieldNormReaders,
     ) -> crate::Result<HashMap<Field, FnvHashMap<UnorderedTermId, TermOrdinal>>> {
         let mut term_offsets: Vec<(&[u8], Addr, UnorderedTermId)> =
             self.term_index.iter().collect();
@@ -161,8 +163,12 @@ impl MultiFieldPostingsWriter {
             }
 
             let postings_writer = &self.per_field_postings_writers[field.field_id() as usize];
-            let mut field_serializer =
-                serializer.new_field(field, postings_writer.total_num_tokens())?;
+            let fieldnorm_reader = fieldnorm_readers.get_field(field);
+            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,
@@ -297,7 +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 owned_read::OwnedRead;
-use std::cmp::Ordering;
-use tantivy_fst::Streamer;
+use crate::{DocId, TERMINATED};
 
-struct PositionComputer {
-    // store the amount of position int
-    // before reading positions.
-    //
-    // 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;
-        }
-    }
-}
+use crate::directory::ReadOnlySource;
+use crate::fastfield::DeleteBitSet;
+use crate::postings::BlockSegmentPostings;
 
 /// `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,
-            cur: COMPRESSION_BLOCK_SIZE,
-            position_computer: None,
+            block_cursor: BlockSegmentPostings::empty(),
+            cur: 0,
+            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
-            position_computer: positions_stream_opt.map(PositionComputer::new),
+            cur: 0, // cursor within the block
+            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 {
-        if self.position_computer.is_some() && self.cur < COMPRESSION_BLOCK_SIZE {
-            let term_freq = self.term_freq() as usize;
-            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() {
+    fn advance(&mut self) -> DocId {
+        debug_assert!(self.block_cursor.block_is_loaded());
+        if self.cur == COMPRESSION_BLOCK_SIZE - 1 {
             self.cur = 0;
-            if !self.block_cursor.advance() {
-                self.cur = COMPRESSION_BLOCK_SIZE;
-                return false;
-            }
+            self.block_cursor.advance();
+        } else {
+            self.cur += 1;
         }
-        true
+        self.doc()
     }
 
-    fn skip_next(&mut self, target: DocId) -> SkipResult {
-        if !self.advance() {
-            return SkipResult::End;
-        }
-        match self.doc().cmp(&target) {
-            Ordering::Equal => {
-                return SkipResult::Reached;
-            }
-            Ordering::Greater => {
-                return SkipResult::OverStep;
-            }
-            _ => {
-                // ...
-            }
+    fn seek(&mut self, target: DocId) -> DocId {
+        debug_assert!(self.doc() <= target);
+        if self.doc() >= target {
+            return self.doc();
         }
 
-        // In the following, thanks to the call to advance above,
-        // we know that the position is not loaded and we need
-        // to skip every doc_freq we cross.
+        self.block_cursor.seek(target);
 
-        // skip blocks until one that might contain the target
-        // check if we need to go to the next block
-        let mut sum_freqs_skipped: u32 = 0;
-        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;
-        }
+        // At this point we are on the block, that might contain our document.
+        let output = self.block_cursor.docs_aligned();
+        self.cur = self.block_searcher.search_in_block(&output, target);
 
-        let cur = self.cur;
-
-        // we're in the right block now, start with an exponential search
-        let (output, len) = self.block_cursor.docs_aligned();
-        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;
+        // 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)
+        // that is greater or equal to the target.
+        debug_assert!(self.cur < COMPRESSION_BLOCK_SIZE);
 
         // `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 {
-            SkipResult::Reached
-        } else {
-            SkipResult::OverStep
-        }
+        debug_assert_eq!(doc, self.doc());
+        doc
     }
 
     /// 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();
-        debug_assert!(
-            self.cur < docs.len(),
-            "Have you forgotten to call `.advance()` at least once before calling `.doc()`                                      ."
-        );
-        docs[self.cur]
+        self.block_cursor.doc(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!(!postings.advance());
+        assert_eq!(postings.advance(), TERMINATED);
+        assert_eq!(postings.advance(), TERMINATED);
         assert_eq!(postings.len(), 0);
     }
 
     #[test]
-    #[should_panic(expected = "Have you forgotten to call `.advance()`")]
-    fn test_panic_if_doc_called_before_advance() {
-        SegmentPostings::empty().doc();
+    fn test_empty_postings_doc_returns_terminated() {
+        let mut postings = SegmentPostings::empty();
+        assert_eq!(postings.doc(), TERMINATED);
+        assert_eq!(postings.advance(), TERMINATED);
     }
 
     #[test]
-    #[should_panic(expected = "Have you forgotten to call `.advance()`")]
-    fn test_panic_if_freq_called_before_advance() {
-        SegmentPostings::empty().term_freq();
+    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.advance());
-        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 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]);
+    fn test_doc_freq() {
+        let docs = SegmentPostings::create_from_docs(&[0, 2, 10]);
+        assert_eq!(docs.doc_freq(), 3);
+        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);
+        assert_eq!(docs.doc_freq_given_deletes(&all_deleted), 0);
     }
 }

src/postings/serializer.rs

@@ -3,14 +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 std::cmp::Ordering;
 use std::io::{self, Write};
 
 /// `InvertedIndexSerializer` is in charge of serializing
@@ -90,20 +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,
         )
     }
 
@@ -131,11 +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() {
@@ -148,8 +155,17 @@ impl<'a> FieldSerializer<'a> {
             _ => (false, false),
         };
         let term_dictionary_builder = TermDictionaryBuilder::create(term_dictionary_write)?;
-        let postings_serializer =
-            PostingsSerializer::new(postings_write, term_freq_enabled, position_enabled);
+        let average_fieldnorm = fieldnorm_reader
+            .as_ref()
+            .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 {
@@ -182,18 +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.
-    pub fn new_term(&mut self, term: &[u8]) -> io::Result<TermOrdinal> {
+    /// * term_doc_freq - return the number of document containing the term.
+    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)
     }
 
@@ -307,14 +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),
 
@@ -327,6 +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);
         }
     }
 
@@ -343,7 +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());
@@ -353,6 +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();
     }
@@ -391,7 +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)?;
@@ -401,6 +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::postings::compression::COMPRESSION_BLOCK_SIZE;
+use crate::common::{read_u32_vint_no_advance, serialize_vint_u32, BinarySerializable, VInt};
+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 {
-        SkipReader {
-            doc: 0u32,
-            owned_read: data,
+    pub fn new(data: ReadOnlySource, doc_freq: u32, skip_info: IndexRecordOption) -> SkipReader {
+        let mut skip_reader = SkipReader {
+            last_doc_in_block: if doc_freq >= COMPRESSION_BLOCK_SIZE as u32 {
+                0
+            } else {
+                TERMINATED
+            },
+            last_doc_in_previous_block: 0u32,
+            owned_read: OwnedRead::new(data),
             skip_info,
-            doc_num_bits: 0u8,
-            tf_num_bits: 0u8,
-            tf_sum: 0u32,
+            block_info: BlockInfo::VInt { num_docs: doc_freq },
+            byte_offset: 0,
+            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) {
-        self.doc = 0u32;
-        self.owned_read = data;
-        self.doc_num_bits = 0u8;
-        self.tf_num_bits = 0u8;
-        self.tf_sum = 0u32;
+    // Returns the block max score for this block if available.
+    //
+    // The block max score is available for all full bitpacked block,
+    // but no available for the last VInt encoded incomplete block.
+    pub fn block_max_score(&self, bm25_weight: &BM25Weight) -> Option<Score> {
+        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 {
-        (self.doc_num_bits + self.tf_num_bits) as usize * COMPRESSION_BLOCK_SIZE / 8
+    pub(crate) fn last_doc_in_block(&self) -> DocId {
+        self.last_doc_in_block
     }
 
-    pub fn doc(&self) -> DocId {
-        self.doc
+    pub fn position_offset(&self) -> u64 {
+        self.position_offset
     }
 
-    pub fn doc_num_bits(&self) -> u8 {
-        self.doc_num_bits
+    pub fn byte_offset(&self) -> usize {
+        self.byte_offset
     }
 
-    /// Number of bits used to encode term frequencies
-    ///
-    /// 0 if term frequencies are not enabled.
-    pub fn tf_num_bits(&self) -> u8 {
-        self.tf_num_bits
-    }
-
-    pub fn tf_sum(&self) -> u32 {
-        self.tf_sum
-    }
-
-    pub fn advance(&mut self) -> bool {
-        if self.owned_read.as_ref().is_empty() {
-            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");
-                }
+    fn read_block_info(&mut self) {
+        let doc_delta = u32::deserialize(&mut self.owned_read).expect("Skip data corrupted");
+        self.last_doc_in_block += doc_delta as DocId;
+        let doc_num_bits = self.owned_read.get(0);
+        match self.skip_info {
+            IndexRecordOption::Basic => {
+                self.owned_read.advance(1);
+                self.block_info = BlockInfo::BitPacked {
+                    doc_num_bits,
+                    tf_num_bits: 0,
+                    tf_sum: 0,
+                    block_wand_fieldnorm_id: 0,
+                    block_wand_term_freq: 0,
+                };
             }
-            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);
-        assert!(skip_reader.advance());
-        assert_eq!(skip_reader.doc(), 1u32);
-        assert_eq!(skip_reader.doc_num_bits(), 2u8);
-        assert_eq!(skip_reader.tf_num_bits(), 3u8);
-        assert!(skip_reader.advance());
-        assert_eq!(skip_reader.doc(), 5u32);
-        assert_eq!(skip_reader.doc_num_bits(), 5u8);
-        assert_eq!(skip_reader.tf_num_bits(), 2u8);
-        assert!(!skip_reader.advance());
+        let doc_freq = 3u32 + (COMPRESSION_BLOCK_SIZE * 2) as u32;
+        let mut skip_reader = SkipReader::new(
+            ReadOnlySource::new(buf),
+            doc_freq,
+            IndexRecordOption::WithFreqs,
+        );
+        assert_eq!(skip_reader.last_doc_in_block(), 1u32);
+        assert_eq!(
+            skip_reader.block_info,
+            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);
-        assert!(skip_reader.advance());
-        assert_eq!(skip_reader.doc(), 1u32);
-        assert_eq!(skip_reader.doc_num_bits(), 2u8);
-        assert!(skip_reader.advance());
-        assert_eq!(skip_reader.doc(), 5u32);
-        assert_eq!(skip_reader.doc_num_bits(), 5u8);
-        assert!(!skip_reader.advance());
+        let doc_freq = 3u32 + (COMPRESSION_BLOCK_SIZE * 2) 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.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/memory_arena.rs

@@ -106,8 +106,8 @@ impl MemoryArena {
     ///
     /// Internally, it counts a number of `1MB` pages
     /// and therefore delivers an upperbound.
-    pub fn mem_usage(&self) -> u64 {
-        (self.pages.len() as u64) * (PAGE_SIZE as u64)
+    pub fn mem_usage(&self) -> usize {
+        self.pages.len() * PAGE_SIZE
     }
 
     pub fn write_at<Item: Copy + 'static>(&mut self, addr: Addr, val: Item) {

src/postings/stacker/term_hashmap.rs

@@ -1,6 +1,4 @@
-use murmurhash32;
-
-use self::murmurhash32::murmurhash2;
+use murmurhash32::murmurhash2;
 
 use super::{Addr, MemoryArena};
 use crate::postings::stacker::memory_arena::store;
@@ -12,8 +10,8 @@ use std::slice;
 
 /// Returns the actual memory size in bytes
 /// required to create a table of size $2^num_bits$.
-pub fn compute_table_size(num_bits: usize) -> u64 {
-    (1u64 << num_bits as u64) * mem::size_of::<KeyValue>() as u64
+pub fn compute_table_size(num_bits: usize) -> usize {
+    (1 << num_bits) * mem::size_of::<KeyValue>()
 }
 
 /// `KeyValue` is the item stored in the hash table.
@@ -116,8 +114,8 @@ impl TermHashMap {
         QuadraticProbing::compute(hash as usize, self.mask)
     }
 
-    pub fn mem_usage(&self) -> u64 {
-        self.table.len() as u64 * mem::size_of::<KeyValue>() as u64
+    pub fn mem_usage(&self) -> usize {
+        self.table.len() * mem::size_of::<KeyValue>()
     }
 
     fn is_saturated(&self) -> bool {

src/query/all_query.rs

@@ -1,6 +1,6 @@
 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};
@@ -9,7 +9,7 @@ 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;
 
@@ -23,9 +23,8 @@ impl Query for AllQuery {
 pub struct AllWeight;
 
 impl Weight for AllWeight {
-    fn scorer(&self, reader: &SegmentReader, boost: f32) -> crate::Result<Box<dyn Scorer>> {
+    fn scorer(&self, reader: &SegmentReader, boost: Score) -> crate::Result<Box<dyn Scorer>> {
         let all_scorer = AllScorer {
-            state: State::NotStarted,
             doc: 0u32,
             max_doc: reader.max_doc(),
         };
@@ -36,43 +35,24 @@ impl Weight for AllWeight {
         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 {
-        match self.state {
-            State::NotStarted => {
-                self.state = State::Started;
-                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
+    fn advance(&mut self) -> DocId {
+        if self.doc + 1 >= self.max_doc {
+            self.doc = TERMINATED;
+            return TERMINATED;
         }
+        self.doc += 1;
+        self.doc
     }
 
     fn doc(&self) -> DocId {
@@ -86,13 +66,14 @@ 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;
@@ -119,19 +100,17 @@ mod tests {
         let weight = AllQuery.weight(&searcher, false).unwrap();
         {
             let reader = searcher.segment_reader(0);
-            let mut scorer = weight.scorer(reader, 1.0f32).unwrap();
-            assert!(scorer.advance());
+            let mut scorer = weight.scorer(reader, 1.0).unwrap();
             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, 1.0f32).unwrap();
-            assert!(scorer.advance());
+            let mut scorer = weight.scorer(reader, 1.0).unwrap();
             assert_eq!(scorer.doc(), 0u32);
-            assert!(!scorer.advance());
+            assert_eq!(scorer.advance(), TERMINATED);
         }
     }
 
@@ -143,16 +122,14 @@ mod tests {
         let weight = AllQuery.weight(&searcher, false).unwrap();
         let reader = searcher.segment_reader(0);
         {
-            let mut scorer = weight.scorer(reader, 2.0f32).unwrap();
-            assert!(scorer.advance());
+            let mut scorer = weight.scorer(reader, 2.0).unwrap();
             assert_eq!(scorer.doc(), 0u32);
-            assert_eq!(scorer.score(), 2.0f32);
+            assert_eq!(scorer.score(), 2.0);
         }
         {
-            let mut scorer = weight.scorer(reader, 1.5f32).unwrap();
-            assert!(scorer.advance());
+            let mut scorer = weight.scorer(reader, 1.5).unwrap();
             assert_eq!(scorer.doc(), 0u32);
-            assert_eq!(scorer.score(), 1.5f32);
+            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, boost: f32) -> 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);
@@ -51,10 +50,15 @@ 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() {
-                for &doc in block_segment_postings.docs() {
+            loop {
+                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);
@@ -63,9 +67,9 @@ where
     }
 
     fn explain(&self, reader: &SegmentReader, doc: DocId) -> Result<Explanation> {
-        let mut scorer = self.scorer(reader, 1.0f32)?;
-        if scorer.skip_next(doc) == SkipResult::Reached {
-            Ok(Explanation::new("AutomatonScorer", 1.0f32))
+        let mut scorer = self.scorer(reader, 1.0)?;
+        if scorer.seek(doc) == doc {
+            Ok(Explanation::new("AutomatonScorer", 1.0))
         } else {
             Err(TantivyError::InvalidArgument(
                 "Document does not exist".to_string(),
@@ -77,6 +81,7 @@ where
 #[cfg(test)]
 mod tests {
     use super::AutomatonWeight;
+    use crate::docset::TERMINATED;
     use crate::query::Weight;
     use crate::schema::{Schema, STRING};
     use crate::Index;
@@ -139,15 +144,14 @@ mod tests {
         let reader = index.reader().unwrap();
         let searcher = reader.searcher();
         let mut scorer = automaton_weight
-            .scorer(searcher.segment_reader(0u32), 1.0f32)
+            .scorer(searcher.segment_reader(0u32), 1.0)
             .unwrap();
-        assert!(scorer.advance());
         assert_eq!(scorer.doc(), 0u32);
-        assert_eq!(scorer.score(), 1.0f32);
-        assert!(scorer.advance());
+        assert_eq!(scorer.score(), 1.0);
+        assert_eq!(scorer.advance(), 2u32);
         assert_eq!(scorer.doc(), 2u32);
-        assert_eq!(scorer.score(), 1.0f32);
-        assert!(!scorer.advance());
+        assert_eq!(scorer.score(), 1.0);
+        assert_eq!(scorer.advance(), TERMINATED);
     }
 
     #[test]
@@ -158,10 +162,9 @@ mod tests {
         let reader = index.reader().unwrap();
         let searcher = reader.searcher();
         let mut scorer = automaton_weight
-            .scorer(searcher.segment_reader(0u32), 1.32f32)
+            .scorer(searcher.segment_reader(0u32), 1.32)
             .unwrap();
-        assert!(scorer.advance());
         assert_eq!(scorer.doc(), 0u32);
-        assert_eq!(scorer.score(), 1.32f32);
+        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 {
-        // skip is required to advance.
-        if !self.advance() {
-            return SkipResult::End;
+    fn seek(&mut self, target: DocId) -> DocId {
+        if target >= self.docs.max_value() {
+            self.doc = TERMINATED;
+            return TERMINATED;
         }
         let target_bucket = target / 64u32;
-
-        // Mask for all of the bits greater or equal
-        // to our target document.
-        match target_bucket.cmp(&self.cursor_bucket) {
-            Ordering::Greater => {
-                self.go_to_bucket(target_bucket);
-                let greater_filter: TinySet = TinySet::range_greater_or_equal(target);
-                self.cursor_tinybitset = self.cursor_tinybitset.intersect(greater_filter);
-                if !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
+        if target_bucket > self.cursor_bucket {
+            self.go_to_bucket(target_bucket);
+            let greater_filter: TinySet = TinySet::range_greater_or_equal(target);
+            self.cursor_tinybitset = self.cursor_tinybitset.intersect(greater_filter);
+            self.advance()
+        } else {
+            let mut doc = self.doc();
+            while doc < target {
+                doc = self.advance();
             }
+            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!(!docset.advance());
+        assert_eq!(docset.advance(), TERMINATED);
     }
 
     #[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!(!docset.advance());
+            assert_eq!(docset.seek(5113), TERMINATED);
+            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 B: f32 = 0.75;
+const K1: Score = 1.2;
+const B: Score = 0.75;
 
-fn idf(doc_freq: u64, doc_count: u64) -> f32 {
-    let x = ((doc_count - doc_freq) as f32 + 0.5) / (doc_freq as f32 + 0.5);
-    (1f32 + x).ln()
+fn idf(doc_freq: u64, doc_count: u64) -> Score {
+    assert!(doc_count >= doc_freq, "{} >= {}", doc_count, doc_freq);
+    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 {
-    K1 * (1f32 - B + B * fieldnorm as f32 / average_fieldnorm)
+fn cached_tf_component(fieldnorm: u32, average_fieldnorm: Score) -> Score {
+    K1 * (1.0 - B + B * fieldnorm as Score / average_fieldnorm)
 }
 
-fn compute_tf_cache(average_fieldnorm: f32) -> [f32; 256] {
-    let mut cache = [0f32; 256];
+fn compute_tf_cache(average_fieldnorm: Score) -> [Score; 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 +28,22 @@ 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,
-    cache: [f32; 256],
-    average_fieldnorm: f32,
+    weight: Score,
+    cache: [Score; 256],
+    average_fieldnorm: Score,
 }
 
 impl BM25Weight {
-    pub fn boost_by(&self, boost: f32) -> BM25Weight {
+    pub fn boost_by(&self, boost: Score) -> BM25Weight {
         BM25Weight {
             idf_explain: self.idf_explain.clone(),
             weight: self.weight * boost,
@@ -60,19 +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);
-            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);
+            BM25Weight::for_one_term(term_doc_freq, total_num_docs, average_fieldnorm)
         } else {
             let idf = terms
                 .iter()
@@ -80,14 +82,30 @@ impl BM25Weight {
                     let term_doc_freq = searcher.doc_freq(term);
                     idf(term_doc_freq, total_num_docs)
                 })
-                .sum::<f32>();
-            idf_explain = Explanation::new("idf", idf);
+                .sum::<Score>();
+            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 {
-        let weight = idf_explain.value() * (1f32 + K1);
+    pub fn for_one_term(
+        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,
@@ -98,19 +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;
-        self.weight * term_freq / (term_freq + norm)
+        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(
@@ -123,12 +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
@@ -139,10 +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_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;
@@ -10,16 +12,21 @@ use crate::query::Scorer;
 use crate::query::Union;
 use crate::query::Weight;
 use crate::query::{intersect_scorers, Explanation};
-use crate::{DocId, SkipResult};
+use crate::{DocId, Score};
 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
     }
 
     {
@@ -29,14 +36,30 @@ where
                 .into_iter()
                 .map(|scorer| *(scorer.downcast::<TermScorer>().map_err(|_| ()).unwrap()))
                 .collect();
-            let scorer: Box<dyn Scorer> =
-                Box::new(Union::<TermScorer, TScoreCombiner>::from(scorers));
-            return scorer;
+            if scorers
+                .iter()
+                .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));
-    scorer
+fn into_box_scorer<TScoreCombiner: ScoreCombiner>(scorer: SpecializedScorer) -> Box<dyn 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 {
@@ -55,7 +78,7 @@ impl BooleanWeight {
     fn per_occur_scorers(
         &self,
         reader: &SegmentReader,
-        boost: f32,
+        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 {
@@ -71,42 +94,52 @@ impl BooleanWeight {
     fn complex_scorer<TScoreCombiner: ScoreCombiner>(
         &self,
         reader: &SegmentReader,
-        boost: f32,
-    ) -> crate::Result<Box<dyn Scorer>> {
+        boost: Score,
+    ) -> 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)
         }
@@ -114,7 +147,7 @@ impl BooleanWeight {
 }
 
 impl Weight for BooleanWeight {
-    fn scorer(&self, reader: &SegmentReader, boost: f32) -> crate::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 {
@@ -126,18 +159,22 @@ impl Weight for BooleanWeight {
             }
         } else if self.scoring_enabled {
             self.complex_scorer::<SumWithCoordsCombiner>(reader, boost)
+                .map(|specialized_scorer| {
+                    into_box_scorer::<SumWithCoordsCombiner>(specialized_scorer)
+                })
         } else {
             self.complex_scorer::<DoNothingCombiner>(reader, boost)
+                .map(into_box_scorer::<DoNothingCombiner>)
         }
     }
 
     fn explain(&self, reader: &SegmentReader, doc: DocId) -> crate::Result<Explanation> {
-        let mut scorer = self.scorer(reader, 1.0f32)?;
-        if scorer.skip_next(doc) != SkipResult::Reached {
+        let mut scorer = self.scorer(reader, 1.0)?;
+        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());
@@ -150,6 +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;
@@ -18,9 +22,8 @@ mod tests {
     use crate::query::Scorer;
     use crate::query::TermQuery;
     use crate::schema::*;
-    use crate::tests::assert_nearly_equals;
     use crate::Index;
-    use crate::{DocAddress, DocId};
+    use crate::{DocAddress, DocId, Score};
 
     fn aux_test_helper() -> (Index, Field) {
         let mut schema_builder = Schema::builder();
@@ -31,24 +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);
-            }
-            {
-                let doc = doc!(text_field => "a c");
-                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);
+                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"));
+                index_writer.add_document(doc!(text_field => "a b c d"));
+                index_writer.add_document(doc!(text_field => "d"));
             }
             assert!(index_writer.commit().is_ok());
         }
@@ -71,9 +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), 1.0f32)
-            .unwrap();
+        let scorer = weight.scorer(searcher.segment_reader(0u32), 1.0).unwrap();
         assert!(scorer.is::<TermScorer>());
     }
 
@@ -85,17 +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), 1.0f32)
-                .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), 1.0f32)
-                .unwrap();
+            let scorer = weight.scorer(searcher.segment_reader(0u32), 1.0).unwrap();
             assert!(scorer.is::<Intersection<Box<dyn Scorer>>>());
         }
     }
@@ -108,9 +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), 1.0f32)
-                .unwrap();
+            let scorer = weight.scorer(searcher.segment_reader(0u32), 1.0).unwrap();
             assert!(scorer.is::<RequiredOptionalScorer<
                 Box<dyn Scorer>,
                 Box<dyn Scorer>,
@@ -120,9 +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), 1.0f32)
-                .unwrap();
+            let scorer = weight.scorer(searcher.segment_reader(0u32), 1.0).unwrap();
             assert!(scorer.is::<TermScorer>());
         }
     }
@@ -153,7 +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]);
@@ -190,6 +169,54 @@ 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();
@@ -218,19 +245,17 @@ mod tests {
         let boolean_weight = boolean_query.weight(&searcher, true).unwrap();
         {
             let mut boolean_scorer = boolean_weight
-                .scorer(searcher.segment_reader(0u32), 1.0f32)
+                .scorer(searcher.segment_reader(0u32), 1.0)
                 .unwrap();
-            assert!(boolean_scorer.advance());
             assert_eq!(boolean_scorer.doc(), 0u32);
-            assert_nearly_equals(boolean_scorer.score(), 0.84163445f32);
+            assert_nearly_equals!(boolean_scorer.score(), 0.84163445);
         }
         {
             let mut boolean_scorer = boolean_weight
-                .scorer(searcher.segment_reader(0u32), 2.0f32)
+                .scorer(searcher.segment_reader(0u32), 2.0)
                 .unwrap();
-            assert!(boolean_scorer.advance());
             assert_eq!(boolean_scorer.doc(), 0u32);
-            assert_nearly_equals(boolean_scorer.score(), 1.6832689f32);
+            assert_nearly_equals!(boolean_scorer.score(), 1.6832689);
         }
     }
 
@@ -260,170 +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), 1.0f32)
-            .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

@@ -1,8 +1,7 @@
-use crate::common::BitSet;
 use crate::fastfield::DeleteBitSet;
 use crate::query::explanation::does_not_match;
 use crate::query::{Explanation, Query, Scorer, Weight};
-use crate::{DocId, DocSet, Searcher, SegmentReader, SkipResult, Term};
+use crate::{DocId, DocSet, Score, Searcher, SegmentReader, Term};
 use std::collections::BTreeSet;
 use std::fmt;
 
@@ -13,12 +12,12 @@ use std::fmt;
 /// factor.
 pub struct BoostQuery {
     query: Box<dyn Query>,
-    boost: f32,
+    boost: Score,
 }
 
 impl BoostQuery {
     /// Builds a boost query.
-    pub fn new(query: Box<dyn Query>, boost: f32) -> BoostQuery {
+    pub fn new(query: Box<dyn Query>, boost: Score) -> BoostQuery {
         BoostQuery { query, boost }
     }
 }
@@ -56,23 +55,23 @@ impl Query for BoostQuery {
 
 pub(crate) struct BoostWeight {
     weight: Box<dyn Weight>,
-    boost: f32,
+    boost: Score,
 }
 
 impl BoostWeight {
-    pub fn new(weight: Box<dyn Weight>, boost: f32) -> Self {
+    pub fn new(weight: Box<dyn Weight>, boost: Score) -> Self {
         BoostWeight { weight, boost }
     }
 }
 
 impl Weight for BoostWeight {
-    fn scorer(&self, reader: &SegmentReader, boost: f32) -> crate::Result<Box<dyn Scorer>> {
+    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, 1.0f32)?;
-        if scorer.skip_next(doc) != SkipResult::Reached {
+        let mut scorer = self.scorer(reader, 1.0)?;
+        if scorer.seek(doc) != doc {
             return Err(does_not_match(doc));
         }
         let mut explanation =
@@ -89,22 +88,22 @@ impl Weight for BoostWeight {
 
 pub(crate) struct BoostScorer<S: Scorer> {
     underlying: S,
-    boost: f32,
+    boost: Score,
 }
 
 impl<S: Scorer> BoostScorer<S> {
-    pub fn new(underlying: S, boost: f32) -> 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) -> bool {
+    fn advance(&mut self) -> DocId {
         self.underlying.advance()
     }
 
-    fn skip_next(&mut self, target: DocId) -> SkipResult {
-        self.underlying.skip_next(target)
+    fn seek(&mut self, target: DocId) -> DocId {
+        self.underlying.seek(target)
     }
 
     fn fill_buffer(&mut self, buffer: &mut [DocId]) -> usize {
@@ -119,10 +118,6 @@ impl<S: Scorer> DocSet for BoostScorer<S> {
         self.underlying.size_hint()
     }
 
-    fn append_to_bitset(&mut self, bitset: &mut BitSet) {
-        self.underlying.append_to_bitset(bitset)
-    }
-
     fn count(&mut self, delete_bitset: &DeleteBitSet) -> u32 {
         self.underlying.count(delete_bitset)
     }
@@ -133,7 +128,7 @@ impl<S: Scorer> DocSet for BoostScorer<S> {
 }
 
 impl<S: Scorer> Scorer for BoostScorer<S> {
-    fn score(&mut self) -> f32 {
+    fn score(&mut self) -> Score {
         self.underlying.score() * self.boost
     }
 }

src/query/empty_query.rs

@@ -1,4 +1,5 @@
 use super::Scorer;
+use crate::docset::TERMINATED;
 use crate::query::explanation::does_not_match;
 use crate::query::Weight;
 use crate::query::{Explanation, Query};
@@ -33,7 +34,7 @@ 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, _boost: f32) -> crate::Result<Box<dyn Scorer>> {
+    fn scorer(&self, _reader: &SegmentReader, _boost: Score) -> crate::Result<Box<dyn Scorer>> {
         Ok(Box::new(EmptyScorer))
     }
 
@@ -48,15 +49,12 @@ impl Weight for EmptyWeight {
 pub struct EmptyScorer;
 
 impl DocSet for EmptyScorer {
-    fn advance(&mut self) -> bool {
-        false
+    fn advance(&mut self) -> DocId {
+        TERMINATED
     }
 
     fn doc(&self) -> DocId {
-        panic!(
-            "You may not call .doc() on a scorer \
-             where the last call to advance() did not return true."
-        );
+        TERMINATED
     }
 
     fn size_hint(&self) -> u32 {
@@ -66,24 +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());
-    }
-
-    #[test]
-    #[should_panic]
-    fn test_empty_scorer_panic_on_doc_call() {
-        EmptyScorer.doc();
+        assert_eq!(empty_scorer.doc(), TERMINATED);
+        assert_eq!(empty_scorer.advance(), TERMINATED);
+        assert_eq!(empty_scorer.doc(), TERMINATED);
     }
 }

src/query/exclude.rs

@@ -1,12 +1,11 @@
-use crate::docset::{DocSet, SkipResult};
+use crate::docset::{DocSet, TERMINATED};
 use crate::query::Scorer;
 use crate::DocId;
 use crate::Score;
 
-#[derive(Clone, Copy, Debug)]
-enum State {
-    ExcludeOne(DocId),
-    Finished,
+#[inline(always)]
+fn is_within<TDocSetExclude: DocSet>(docset: &mut TDocSetExclude, doc: DocId) -> bool {
+    docset.doc() <= doc && docset.seek(doc) == doc
 }
 
 /// Filters a given `DocSet` by removing the docs from a given `DocSet`.
@@ -15,29 +14,6 @@ enum State {
 pub struct Exclude<TDocSet, TDocSetExclude> {
     underlying_docset: TDocSet,
     excluding_docset: TDocSetExclude,
-    excluding_state: State,
-}
-
-impl<TDocSet, TDocSetExclude> Exclude<TDocSet, TDocSetExclude>
-where
-    TDocSetExclude: DocSet,
-{
-    /// Creates a new `ExcludeScorer`
-    pub fn new(
-        underlying_docset: TDocSet,
-        mut excluding_docset: TDocSetExclude,
-    ) -> Exclude<TDocSet, TDocSetExclude> {
-        let state = if excluding_docset.advance() {
-            State::ExcludeOne(excluding_docset.doc())
-        } else {
-            State::Finished
-        };
-        Exclude {
-            underlying_docset,
-            excluding_docset,
-            excluding_state: state,
-        }
-    }
 }
 
 impl<TDocSet, TDocSetExclude> Exclude<TDocSet, TDocSetExclude>
@@ -45,33 +21,21 @@ where
     TDocSet: DocSet,
     TDocSetExclude: DocSet,
 {
-    /// Returns true iff the doc is not removed.
-    ///
-    /// The method has to be called with non strictly
-    /// increasing `doc`.
-    fn accept(&mut self) -> bool {
-        let doc = self.underlying_docset.doc();
-        match self.excluding_state {
-            State::ExcludeOne(excluded_doc) => {
-                if doc == excluded_doc {
-                    return false;
-                }
-                if excluded_doc > doc {
-                    return true;
-                }
-                match self.excluding_docset.skip_next(doc) {
-                    SkipResult::OverStep => {
-                        self.excluding_state = State::ExcludeOne(self.excluding_docset.doc());
-                        true
-                    }
-                    SkipResult::End => {
-                        self.excluding_state = State::Finished;
-                        true
-                    }
-                    SkipResult::Reached => false,
-                }
+    /// Creates a new `ExcludeScorer`
+    pub fn new(
+        mut underlying_docset: TDocSet,
+        mut excluding_docset: TDocSetExclude,
+    ) -> Exclude<TDocSet, TDocSetExclude> {
+        while underlying_docset.doc() != TERMINATED {
+            let target = underlying_docset.doc();
+            if !is_within(&mut excluding_docset, target) {
+                break;
             }
-            State::Finished => true,
+            underlying_docset.advance();
+        }
+        Exclude {
+            underlying_docset,
+            excluding_docset,
         }
     }
 }
@@ -81,27 +45,27 @@ where
     TDocSet: DocSet,
     TDocSetExclude: DocSet,
 {
-    fn advance(&mut self) -> bool {
-        while self.underlying_docset.advance() {
-            if self.accept() {
-                return true;
+    fn advance(&mut self) -> DocId {
+        loop {
+            let candidate = self.underlying_docset.advance();
+            if candidate == TERMINATED {
+                return TERMINATED;
+            }
+            if !is_within(&mut self.excluding_docset, candidate) {
+                return candidate;
             }
         }
-        false
     }
 
-    fn skip_next(&mut self, target: DocId) -> SkipResult {
-        let underlying_skip_result = self.underlying_docset.skip_next(target);
-        if underlying_skip_result == SkipResult::End {
-            return SkipResult::End;
+    fn seek(&mut self, target: DocId) -> DocId {
+        let candidate = self.underlying_docset.seek(target);
+        if candidate == TERMINATED {
+            return TERMINATED;
         }
-        if self.accept() {
-            underlying_skip_result
-        } else if self.advance() {
-            SkipResult::OverStep
-        } else {
-            SkipResult::End
+        if !is_within(&mut self.excluding_docset, candidate) {
+            return candidate;
         }
+        self.advance()
     }
 
     fn doc(&self) -> DocId {
@@ -141,8 +105,9 @@ mod tests {
             VecDocSet::from(vec![1, 2, 3, 10, 16, 24]),
         );
         let mut els = vec![];
-        while exclude_scorer.advance() {
+        while exclude_scorer.doc() != TERMINATED {
             els.push(exclude_scorer.doc());
+            exclude_scorer.advance();
         }
         assert_eq!(els, vec![5, 8, 15]);
     }
@@ -156,7 +121,7 @@ mod tests {
                     VecDocSet::from(vec![1, 2, 3, 10, 16, 24]),
                 ))
             },
-            vec![1, 2, 5, 8, 10, 15, 24],
+            vec![5, 8, 10, 15, 24],
         );
     }
 

src/query/explanation.rs

@@ -1,5 +1,6 @@
-use crate::{DocId, TantivyError};
+use crate::{DocId, Score, TantivyError};
 use serde::Serialize;
+use std::fmt;
 
 pub(crate) fn does_not_match(doc: DocId) -> TantivyError {
     TantivyError::InvalidArgument(format!("Document #({}) does not match", doc))
@@ -12,15 +13,21 @@ pub(crate) fn does_not_match(doc: DocId) -> TantivyError {
 /// representation of this tree when debugging a given score.
 #[derive(Clone, Serialize)]
 pub struct Explanation {
-    value: f32,
+    value: Score,
     description: String,
     #[serde(skip_serializing_if = "Vec::is_empty")]
     details: Vec<Explanation>,
 }
 
+impl fmt::Debug for Explanation {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        write!(f, "Explanation({})", self.to_pretty_json())
+    }
+}
+
 impl Explanation {
     /// Creates a new explanation object.
-    pub fn new<T: ToString>(description: T, value: f32) -> Explanation {
+    pub fn new<T: ToString>(description: T, value: Score) -> Explanation {
         Explanation {
             value,
             description: description.to_string(),
@@ -29,7 +36,7 @@ impl Explanation {
     }
 
     /// Returns the value associated to the current node.
-    pub fn value(&self) -> f32 {
+    pub fn value(&self) -> Score {
         self.value
     }
 
@@ -41,7 +48,7 @@ impl Explanation {
     }
 
     /// Shortcut for `self.details.push(Explanation::new(name, value));`
-    pub fn add_const<T: ToString>(&mut self, name: T, value: f32) {
+    pub fn add_const<T: ToString>(&mut self, name: T, value: Score) {
         self.details.push(Explanation::new(name, value));
     }
 

src/query/fuzzy_query.rs

@@ -2,10 +2,36 @@ use crate::query::{AutomatonWeight, Query, Weight};
 use crate::schema::Term;
 use crate::Searcher;
 use crate::TantivyError::InvalidArgument;
-use levenshtein_automata::{LevenshteinAutomatonBuilder, DFA};
+use levenshtein_automata::{Distance, LevenshteinAutomatonBuilder, DFA};
 use once_cell::sync::Lazy;
 use std::collections::HashMap;
 use std::ops::Range;
+use tantivy_fst::Automaton;
+
+pub(crate) struct DFAWrapper(pub DFA);
+
+impl Automaton for DFAWrapper {
+    type State = u32;
+
+    fn start(&self) -> Self::State {
+        self.0.initial_state()
+    }
+
+    fn is_match(&self, state: &Self::State) -> bool {
+        match self.0.distance(*state) {
+            Distance::Exact(_) => true,
+            Distance::AtLeast(_) => false,
+        }
+    }
+
+    fn can_match(&self, state: &u32) -> bool {
+        *state != levenshtein_automata::SINK_STATE
+    }
+
+    fn accept(&self, state: &Self::State, byte: u8) -> Self::State {
+        self.0.transition(*state, byte)
+    }
+}
 
 /// A range of Levenshtein distances that we will build DFAs for our terms
 /// The computation is exponential, so best keep it to low single digits
@@ -91,7 +117,7 @@ impl FuzzyTermQuery {
         }
     }
 
-    /// Creates a new Fuzzy Query that treats transpositions as cost one rather than two
+    /// Creates a new Fuzzy Query of the Term prefix
     pub fn new_prefix(term: Term, distance: u8, transposition_cost_one: bool) -> FuzzyTermQuery {
         FuzzyTermQuery {
             term,
@@ -101,13 +127,20 @@ impl FuzzyTermQuery {
         }
     }
 
-    fn specialized_weight(&self) -> crate::Result<AutomatonWeight<DFA>> {
+    fn specialized_weight(&self) -> crate::Result<AutomatonWeight<DFAWrapper>> {
         // LEV_BUILDER is a HashMap, whose `get` method returns an Option
         match LEV_BUILDER.get(&(self.distance, false)) {
             // Unwrap the option and build the Ok(AutomatonWeight)
             Some(automaton_builder) => {
-                let automaton = automaton_builder.build_dfa(self.term.text());
-                Ok(AutomatonWeight::new(self.term.field(), automaton))
+                let automaton = if self.prefix {
+                    automaton_builder.build_prefix_dfa(self.term.text())
+                } else {
+                    automaton_builder.build_dfa(self.term.text())
+                };
+                Ok(AutomatonWeight::new(
+                    self.term.field(),
+                    DFAWrapper(automaton),
+                ))
             }
             None => Err(InvalidArgument(format!(
                 "Levenshtein distance of {} is not allowed. Choose a value in the {:?} range",
@@ -130,10 +163,10 @@ impl Query for FuzzyTermQuery {
 #[cfg(test)]
 mod test {
     use super::FuzzyTermQuery;
+    use crate::assert_nearly_equals;
     use crate::collector::TopDocs;
     use crate::schema::Schema;
     use crate::schema::TEXT;
-    use crate::tests::assert_nearly_equals;
     use crate::Index;
     use crate::Term;
 
@@ -155,6 +188,8 @@ mod test {
         }
         let reader = index.reader().unwrap();
         let searcher = reader.searcher();
+
+        // passes because Levenshtein distance is 1 (substitute 'o' with 'a')
         {
             let term = Term::from_field_text(country_field, "japon");
 
@@ -164,7 +199,31 @@ mod test {
                 .unwrap();
             assert_eq!(top_docs.len(), 1, "Expected only 1 document");
             let (score, _) = top_docs[0];
-            assert_nearly_equals(1f32, score);
+            assert_nearly_equals!(1.0, score);
+        }
+
+        // fails because non-prefix Levenshtein distance is more than 1 (add 'a' and 'n')
+        {
+            let term = Term::from_field_text(country_field, "jap");
+
+            let fuzzy_query = FuzzyTermQuery::new(term, 1, true);
+            let top_docs = searcher
+                .search(&fuzzy_query, &TopDocs::with_limit(2))
+                .unwrap();
+            assert_eq!(top_docs.len(), 0, "Expected no document");
+        }
+
+        // passes because prefix Levenshtein distance is 0
+        {
+            let term = Term::from_field_text(country_field, "jap");
+
+            let fuzzy_query = FuzzyTermQuery::new_prefix(term, 1, true);
+            let top_docs = searcher
+                .search(&fuzzy_query, &TopDocs::with_limit(2))
+                .unwrap();
+            assert_eq!(top_docs.len(), 1, "Expected only 1 document");
+            let (score, _) = top_docs[0];
+            assert_nearly_equals!(1.0, score);
         }
     }
 }

src/query/intersection.rs

@@ -1,4 +1,4 @@
-use crate::docset::{DocSet, SkipResult};
+use crate::docset::{DocSet, TERMINATED};
 use crate::query::term_query::TermScorer;
 use crate::query::EmptyScorer;
 use crate::query::Scorer;
@@ -20,12 +20,14 @@ pub fn intersect_scorers(mut scorers: Vec<Box<dyn Scorer>>) -> Box<dyn Scorer> {
     if scorers.len() == 1 {
         return scorers.pop().unwrap();
     }
+    scorers.sort_by_key(|scorer| scorer.size_hint());
+    let doc = go_to_first_doc(&mut scorers[..]);
+    if doc == TERMINATED {
+        return Box::new(EmptyScorer);
+    }
     // We know that we have at least 2 elements.
-    let num_docsets = scorers.len();
-    scorers.sort_by(|left, right| right.size_hint().cmp(&left.size_hint()));
-    let left = scorers.pop().unwrap();
-    let right = scorers.pop().unwrap();
-    scorers.reverse();
+    let left = scorers.remove(0);
+    let right = scorers.remove(0);
     let all_term_scorers = [&left, &right]
         .iter()
         .all(|&scorer| scorer.is::<TermScorer>());
@@ -34,14 +36,12 @@ pub fn intersect_scorers(mut scorers: Vec<Box<dyn Scorer>>) -> Box<dyn Scorer> {
             left: *(left.downcast::<TermScorer>().map_err(|_| ()).unwrap()),
             right: *(right.downcast::<TermScorer>().map_err(|_| ()).unwrap()),
             others: scorers,
-            num_docsets,
         });
     }
     Box::new(Intersection {
         left,
         right,
         others: scorers,
-        num_docsets,
     })
 }
 
@@ -50,22 +50,35 @@ pub struct Intersection<TDocSet: DocSet, TOtherDocSet: DocSet = Box<dyn Scorer>>
     left: TDocSet,
     right: TDocSet,
     others: Vec<TOtherDocSet>,
-    num_docsets: usize,
+}
+
+fn go_to_first_doc<TDocSet: DocSet>(docsets: &mut [TDocSet]) -> DocId {
+    assert!(!docsets.is_empty());
+    let mut candidate = docsets.iter().map(TDocSet::doc).max().unwrap();
+    'outer: loop {
+        for docset in docsets.iter_mut() {
+            let seek_doc = docset.seek(candidate);
+            if seek_doc > candidate {
+                candidate = docset.doc();
+                continue 'outer;
+            }
+        }
+        return candidate;
+    }
 }
 
 impl<TDocSet: DocSet> Intersection<TDocSet, TDocSet> {
     pub(crate) fn new(mut docsets: Vec<TDocSet>) -> Intersection<TDocSet, TDocSet> {
         let num_docsets = docsets.len();
         assert!(num_docsets >= 2);
-        docsets.sort_by(|left, right| right.size_hint().cmp(&left.size_hint()));
-        let left = docsets.pop().unwrap();
-        let right = docsets.pop().unwrap();
-        docsets.reverse();
+        docsets.sort_by_key(|docset| docset.size_hint());
+        go_to_first_doc(&mut docsets);
+        let left = docsets.remove(0);
+        let right = docsets.remove(0);
         Intersection {
             left,
             right,
             others: docsets,
-            num_docsets,
         }
     }
 }
@@ -80,128 +93,49 @@ impl<TDocSet: DocSet> Intersection<TDocSet, TDocSet> {
     }
 }
 
-impl<TDocSet: DocSet, TOtherDocSet: DocSet> Intersection<TDocSet, TOtherDocSet> {
-    pub(crate) fn docset_mut(&mut self, ord: usize) -> &mut dyn DocSet {
-        match ord {
-            0 => &mut self.left,
-            1 => &mut self.right,
-            n => &mut self.others[n - 2],
-        }
-    }
-}
-
 impl<TDocSet: DocSet, TOtherDocSet: DocSet> DocSet for Intersection<TDocSet, TOtherDocSet> {
-    fn advance(&mut self) -> bool {
+    fn advance(&mut self) -> DocId {
         let (left, right) = (&mut self.left, &mut self.right);
-
-        if !left.advance() {
-            return false;
-        }
-
-        let mut candidate = left.doc();
-        let mut other_candidate_ord: usize = usize::max_value();
+        let mut candidate = left.advance();
 
         'outer: loop {
             // In the first part we look for a document in the intersection
             // of the two rarest `DocSet` in the intersection.
+
             loop {
-                match right.skip_next(candidate) {
-                    SkipResult::Reached => {
-                        break;
-                    }
-                    SkipResult::OverStep => {
-                        candidate = right.doc();
-                        other_candidate_ord = usize::max_value();
-                    }
-                    SkipResult::End => {
-                        return false;
-                    }
-                }
-                match left.skip_next(candidate) {
-                    SkipResult::Reached => {
-                        break;
-                    }
-                    SkipResult::OverStep => {
-                        candidate = left.doc();
-                        other_candidate_ord = usize::max_value();
-                    }
-                    SkipResult::End => {
-                        return false;
-                    }
+                let right_doc = right.seek(candidate);
+                candidate = left.seek(right_doc);
+                if candidate == right_doc {
+                    break;
                 }
             }
+
+            debug_assert_eq!(left.doc(), right.doc());
             // test the remaining scorers;
-            for (ord, docset) in self.others.iter_mut().enumerate() {
-                if ord == other_candidate_ord {
-                    continue;
-                }
-                // `candidate_ord` is already at the
-                // right position.
-                //
-                // Calling `skip_next` would advance this docset
-                // and miss it.
-                match docset.skip_next(candidate) {
-                    SkipResult::Reached => {}
-                    SkipResult::OverStep => {
-                        // this is not in the intersection,
-                        // let's update our candidate.
-                        candidate = docset.doc();
-                        match left.skip_next(candidate) {
-                            SkipResult::Reached => {
-                                other_candidate_ord = ord;
-                            }
-                            SkipResult::OverStep => {
-                                candidate = left.doc();
-                                other_candidate_ord = usize::max_value();
-                            }
-                            SkipResult::End => {
-                                return false;
-                            }
-                        }
-                        continue 'outer;
-                    }
-                    SkipResult::End => {
-                        return false;
-                    }
+            for docset in self.others.iter_mut() {
+                let seek_doc = docset.seek(candidate);
+                if seek_doc > candidate {
+                    candidate = left.seek(seek_doc);
+                    continue 'outer;
                 }
             }
-            return true;
+            debug_assert_eq!(candidate, self.left.doc());
+            debug_assert_eq!(candidate, self.right.doc());
+            debug_assert!(self.others.iter().all(|docset| docset.doc() == candidate));
+            return candidate;
         }
     }
 
-    fn skip_next(&mut self, target: DocId) -> SkipResult {
-        // We optimize skipping by skipping every single member
-        // of the intersection to target.
-        let mut current_target: DocId = target;
-        let mut current_ord = self.num_docsets;
-
-        'outer: loop {
-            for ord in 0..self.num_docsets {
-                let docset = self.docset_mut(ord);
-                if ord == current_ord {
-                    continue;
-                }
-                match docset.skip_next(current_target) {
-                    SkipResult::End => {
-                        return SkipResult::End;
-                    }
-                    SkipResult::OverStep => {
-                        // update the target
-                        // for the remaining members of the intersection.
-                        current_target = docset.doc();
-                        current_ord = ord;
-                        continue 'outer;
-                    }
-                    SkipResult::Reached => {}
-                }
-            }
-            if target == current_target {
-                return SkipResult::Reached;
-            } else {
-                assert!(current_target > target);
-                return SkipResult::OverStep;
-            }
+    fn seek(&mut self, target: DocId) -> DocId {
+        self.left.seek(target);
+        let mut docsets: Vec<&mut dyn DocSet> = vec![&mut self.left, &mut self.right];
+        for docset in &mut self.others {
+            docsets.push(docset);
         }
+        let doc = go_to_first_doc(&mut docsets[..]);
+        debug_assert!(docsets.iter().all(|docset| docset.doc() == doc));
+        debug_assert!(doc >= target);
+        doc
     }
 
     fn doc(&self) -> DocId {
@@ -228,7 +162,7 @@ where
 #[cfg(test)]
 mod tests {
     use super::Intersection;
-    use crate::docset::{DocSet, SkipResult};
+    use crate::docset::{DocSet, TERMINATED};
     use crate::postings::tests::test_skip_against_unoptimized;
     use crate::query::VecDocSet;
 
@@ -238,20 +172,18 @@ mod tests {
             let left = VecDocSet::from(vec![1, 3, 9]);
             let right = VecDocSet::from(vec![3, 4, 9, 18]);
             let mut intersection = Intersection::new(vec![left, right]);
-            assert!(intersection.advance());
             assert_eq!(intersection.doc(), 3);
-            assert!(intersection.advance());
+            assert_eq!(intersection.advance(), 9);
             assert_eq!(intersection.doc(), 9);
-            assert!(!intersection.advance());
+            assert_eq!(intersection.advance(), TERMINATED);
         }
         {
             let a = VecDocSet::from(vec![1, 3, 9]);
             let b = VecDocSet::from(vec![3, 4, 9, 18]);
             let c = VecDocSet::from(vec![1, 5, 9, 111]);
             let mut intersection = Intersection::new(vec![a, b, c]);
-            assert!(intersection.advance());
             assert_eq!(intersection.doc(), 9);
-            assert!(!intersection.advance());
+            assert_eq!(intersection.advance(), TERMINATED);
         }
     }
 
@@ -260,8 +192,8 @@ mod tests {
         let left = VecDocSet::from(vec![0]);
         let right = VecDocSet::from(vec![0]);
         let mut intersection = Intersection::new(vec![left, right]);
-        assert!(intersection.advance());
         assert_eq!(intersection.doc(), 0);
+        assert_eq!(intersection.advance(), TERMINATED);
     }
 
     #[test]
@@ -269,7 +201,7 @@ mod tests {
         let left = VecDocSet::from(vec![0, 1, 2, 4]);
         let right = VecDocSet::from(vec![2, 5]);
         let mut intersection = Intersection::new(vec![left, right]);
-        assert_eq!(intersection.skip_next(2), SkipResult::Reached);
+        assert_eq!(intersection.seek(2), 2);
         assert_eq!(intersection.doc(), 2);
     }
 
@@ -312,7 +244,7 @@ mod tests {
         let a = VecDocSet::from(vec![1, 3]);
         let b = VecDocSet::from(vec![1, 4]);
         let c = VecDocSet::from(vec![3, 9]);
-        let mut intersection = Intersection::new(vec![a, b, c]);
-        assert!(!intersection.advance());
+        let intersection = Intersection::new(vec![a, b, c]);
+        assert_eq!(intersection.doc(), TERMINATED);
     }
 }

src/query/mod.rs

@@ -1,6 +1,4 @@
-/*!
-Query
-*/
+/*! Query Module */
 
 mod all_query;
 mod automaton_weight;
@@ -28,6 +26,7 @@ mod weight;
 mod vec_docset;
 
 pub(crate) mod score_combiner;
+pub(crate) use self::bm25::BM25Weight;
 pub use self::intersection::Intersection;
 pub use self::union::Union;
 
@@ -42,6 +41,8 @@ pub use self::boost_query::BoostQuery;
 pub use self::empty_query::{EmptyQuery, EmptyScorer, EmptyWeight};
 pub use self::exclude::Exclude;
 pub use self::explanation::Explanation;
+#[cfg(test)]
+pub(crate) use self::fuzzy_query::DFAWrapper;
 pub use self::fuzzy_query::FuzzyTermQuery;
 pub use self::intersection::intersect_scorers;
 pub use self::phrase_query::PhraseQuery;

src/query/phrase_query/mod.rs

@@ -10,13 +10,13 @@ pub use self::phrase_weight::PhraseWeight;
 pub mod tests {
 
     use super::*;
+    use crate::assert_nearly_equals;
     use crate::collector::tests::{TEST_COLLECTOR_WITHOUT_SCORE, TEST_COLLECTOR_WITH_SCORE};
     use crate::core::Index;
-    use crate::error::TantivyError;
+    use crate::query::Weight;
     use crate::schema::{Schema, Term, TEXT};
-    use crate::tests::assert_nearly_equals;
-    use crate::DocAddress;
     use crate::DocId;
+    use crate::{DocAddress, TERMINATED};
 
     pub fn create_index(texts: &[&'static str]) -> Index {
         let mut schema_builder = Schema::builder();
@@ -61,13 +61,30 @@ pub mod tests {
                 .map(|docaddr| docaddr.1)
                 .collect::<Vec<_>>()
         };
-        assert_eq!(test_query(vec!["a", "b", "c"]), vec![2, 4]);
         assert_eq!(test_query(vec!["a", "b"]), vec![1, 2, 3, 4]);
+        assert_eq!(test_query(vec!["a", "b", "c"]), vec![2, 4]);
         assert_eq!(test_query(vec!["b", "b"]), vec![0, 1]);
         assert!(test_query(vec!["g", "ewrwer"]).is_empty());
         assert!(test_query(vec!["g", "a"]).is_empty());
     }
 
+    #[test]
+    pub fn test_phrase_query_simple() -> crate::Result<()> {
+        let index = create_index(&["a b b d c g c", "a b a b c"]);
+        let text_field = index.schema().get_field("text").unwrap();
+        let searcher = index.reader()?.searcher();
+        let terms: Vec<Term> = vec!["a", "b", "c"]
+            .iter()
+            .map(|text| Term::from_field_text(text_field, text))
+            .collect();
+        let phrase_query = PhraseQuery::new(terms);
+        let phrase_weight = phrase_query.phrase_weight(&searcher, false)?;
+        let mut phrase_scorer = phrase_weight.scorer(searcher.segment_reader(0), 1.0)?;
+        assert_eq!(phrase_scorer.doc(), 1);
+        assert_eq!(phrase_scorer.advance(), TERMINATED);
+        Ok(())
+    }
+
     #[test]
     pub fn test_phrase_query_no_score() {
         let index = create_index(&[
@@ -127,21 +144,16 @@ pub mod tests {
             Term::from_field_text(text_field, "a"),
             Term::from_field_text(text_field, "b"),
         ]);
-        match searcher
+
+        let search_result = searcher
             .search(&phrase_query, &TEST_COLLECTOR_WITH_SCORE)
-            .map(|_| ())
-            .unwrap_err()
-        {
-            TantivyError::SchemaError(ref msg) => {
-                assert_eq!(
-                    "Applied phrase query on field \"text\", which does not have positions indexed",
-                    msg.as_str()
-                );
-            }
-            _ => {
-                panic!("Should have returned an error");
-            }
-        }
+            .map(|_| ());
+        assert!(matches!(
+            search_result,
+            Err(crate::TantivyError::SchemaError(msg))
+            if msg == "Applied phrase query on field \"text\", which does not have positions \
+            indexed"
+        ));
     }
 
     #[test]
@@ -163,8 +175,8 @@ pub mod tests {
                 .to_vec()
         };
         let scores = test_query(vec!["a", "b"]);
-        assert_nearly_equals(scores[0], 0.40618482);
-        assert_nearly_equals(scores[1], 0.46844664);
+        assert_nearly_equals!(scores[0], 0.40618482);
+        assert_nearly_equals!(scores[1], 0.46844664);
     }
 
     #[test] // motivated by #234

src/query/phrase_query/phrase_scorer.rs

@@ -1,9 +1,9 @@
-use crate::docset::{DocSet, SkipResult};
+use crate::docset::{DocSet, TERMINATED};
 use crate::fieldnorm::FieldNormReader;
 use crate::postings::Postings;
 use crate::query::bm25::BM25Weight;
 use crate::query::{Intersection, Scorer};
-use crate::DocId;
+use crate::{DocId, Score};
 use std::cmp::Ordering;
 
 struct PostingsWithOffset<TPostings> {
@@ -25,12 +25,12 @@ impl<TPostings: Postings> PostingsWithOffset<TPostings> {
 }
 
 impl<TPostings: Postings> DocSet for PostingsWithOffset<TPostings> {
-    fn advance(&mut self) -> bool {
+    fn advance(&mut self) -> DocId {
         self.postings.advance()
     }
 
-    fn skip_next(&mut self, target: DocId) -> SkipResult {
-        self.postings.skip_next(target)
+    fn seek(&mut self, target: DocId) -> DocId {
+        self.postings.seek(target)
     }
 
     fn doc(&self) -> DocId {
@@ -149,7 +149,7 @@ impl<TPostings: Postings> PhraseScorer<TPostings> {
                 PostingsWithOffset::new(postings, (max_offset - offset) as u32)
             })
             .collect::<Vec<_>>();
-        PhraseScorer {
+        let mut scorer = PhraseScorer {
             intersection_docset: Intersection::new(postings_with_offsets),
             num_terms: num_docsets,
             left: Vec::with_capacity(100),
@@ -158,7 +158,11 @@ impl<TPostings: Postings> PhraseScorer<TPostings> {
             similarity_weight,
             fieldnorm_reader,
             score_needed,
+        };
+        if scorer.doc() != TERMINATED && !scorer.phrase_match() {
+            scorer.advance();
         }
+        scorer
     }
 
     pub fn phrase_count(&self) -> u32 {
@@ -225,31 +229,22 @@ impl<TPostings: Postings> PhraseScorer<TPostings> {
 }
 
 impl<TPostings: Postings> DocSet for PhraseScorer<TPostings> {
-    fn advance(&mut self) -> bool {
-        while self.intersection_docset.advance() {
-            if self.phrase_match() {
-                return true;
+    fn advance(&mut self) -> DocId {
+        loop {
+            let doc = self.intersection_docset.advance();
+            if doc == TERMINATED || self.phrase_match() {
+                return doc;
             }
         }
-        false
     }
 
-    fn skip_next(&mut self, target: DocId) -> SkipResult {
-        if self.intersection_docset.skip_next(target) == SkipResult::End {
-            return SkipResult::End;
-        }
-        if self.phrase_match() {
-            if self.doc() == target {
-                return SkipResult::Reached;
-            } else {
-                return SkipResult::OverStep;
-            }
-        }
-        if self.advance() {
-            SkipResult::OverStep
-        } else {
-            SkipResult::End
+    fn seek(&mut self, target: DocId) -> DocId {
+        debug_assert!(target >= self.doc());
+        let doc = self.intersection_docset.seek(target);
+        if doc == TERMINATED || self.phrase_match() {
+            return doc;
         }
+        self.advance()
     }
 
     fn doc(&self) -> DocId {
@@ -262,7 +257,7 @@ impl<TPostings: Postings> DocSet for PhraseScorer<TPostings> {
 }
 
 impl<TPostings: Postings> Scorer for PhraseScorer<TPostings> {
-    fn score(&mut self) -> f32 {
+    fn score(&mut self) -> Score {
         let doc = self.doc();
         let fieldnorm_id = self.fieldnorm_reader.fieldnorm_id(doc);
         self.similarity_weight
@@ -272,7 +267,6 @@ impl<TPostings: Postings> Scorer for PhraseScorer<TPostings> {
 
 #[cfg(test)]
 mod tests {
-
     use super::{intersection, intersection_count};
 
     fn test_intersection_sym(left: &[u32], right: &[u32], expected: &[u32]) {

src/query/phrase_query/phrase_weight.rs

@@ -10,7 +10,7 @@ use crate::query::{EmptyScorer, Explanation};
 use crate::schema::IndexRecordOption;
 use crate::schema::Term;
 use crate::{DocId, DocSet};
-use crate::{Result, SkipResult};
+use crate::{Result, Score};
 
 pub struct PhraseWeight {
     phrase_terms: Vec<(usize, Term)>,
@@ -40,7 +40,7 @@ impl PhraseWeight {
     fn phrase_scorer(
         &self,
         reader: &SegmentReader,
-        boost: f32,
+        boost: Score,
     ) -> Result<Option<PhraseScorer<SegmentPostings>>> {
         let similarity_weight = self.similarity_weight.boost_by(boost);
         let fieldnorm_reader = self.fieldnorm_reader(reader);
@@ -85,7 +85,7 @@ impl PhraseWeight {
 }
 
 impl Weight for PhraseWeight {
-    fn scorer(&self, reader: &SegmentReader, boost: f32) -> Result<Box<dyn Scorer>> {
+    fn scorer(&self, reader: &SegmentReader, boost: Score) -> Result<Box<dyn Scorer>> {
         if let Some(scorer) = self.phrase_scorer(reader, boost)? {
             Ok(Box::new(scorer))
         } else {
@@ -94,12 +94,12 @@ impl Weight for PhraseWeight {
     }
 
     fn explain(&self, reader: &SegmentReader, doc: DocId) -> Result<Explanation> {
-        let scorer_opt = self.phrase_scorer(reader, 1.0f32)?;
+        let scorer_opt = self.phrase_scorer(reader, 1.0)?;
         if scorer_opt.is_none() {
             return Err(does_not_match(doc));
         }
         let mut scorer = scorer_opt.unwrap();
-        if scorer.skip_next(doc) != SkipResult::Reached {
+        if scorer.seek(doc) != doc {
             return Err(does_not_match(doc));
         }
         let fieldnorm_reader = self.fieldnorm_reader(reader);
@@ -114,6 +114,7 @@ impl Weight for PhraseWeight {
 #[cfg(test)]
 mod tests {
     use super::super::tests::create_index;
+    use crate::docset::TERMINATED;
     use crate::query::PhraseQuery;
     use crate::{DocSet, Term};
 
@@ -129,15 +130,14 @@ mod tests {
         ]);
         let phrase_weight = phrase_query.phrase_weight(&searcher, true).unwrap();
         let mut phrase_scorer = phrase_weight
-            .phrase_scorer(searcher.segment_reader(0u32), 1.0f32)
+            .phrase_scorer(searcher.segment_reader(0u32), 1.0)
             .unwrap()
             .unwrap();
-        assert!(phrase_scorer.advance());
         assert_eq!(phrase_scorer.doc(), 1);
         assert_eq!(phrase_scorer.phrase_count(), 2);
-        assert!(phrase_scorer.advance());
+        assert_eq!(phrase_scorer.advance(), 2);
         assert_eq!(phrase_scorer.doc(), 2);
         assert_eq!(phrase_scorer.phrase_count(), 1);
-        assert!(!phrase_scorer.advance());
+        assert_eq!(phrase_scorer.advance(), TERMINATED);
     }
 }

src/query/query_parser/logical_ast.rs

@@ -2,6 +2,7 @@ use crate::query::Occur;
 use crate::schema::Field;
 use crate::schema::Term;
 use crate::schema::Type;
+use crate::Score;
 use std::fmt;
 use std::ops::Bound;
 
@@ -21,12 +22,12 @@ pub enum LogicalLiteral {
 pub enum LogicalAST {
     Clause(Vec<(Occur, LogicalAST)>),
     Leaf(Box<LogicalLiteral>),
-    Boost(Box<LogicalAST>, f32),
+    Boost(Box<LogicalAST>, Score),
 }
 
 impl LogicalAST {
-    pub fn boost(self, boost: f32) -> LogicalAST {
-        if (boost - 1.0f32).abs() < std::f32::EPSILON {
+    pub fn boost(self, boost: Score) -> LogicalAST {
+        if (boost - 1.0).abs() < Score::EPSILON {
             self
         } else {
             LogicalAST::Boost(Box::new(self), boost)

src/query/query_parser/query_parser.rs

@@ -12,6 +12,7 @@ use crate::schema::{Facet, IndexRecordOption};
 use crate::schema::{Field, Schema};
 use crate::schema::{FieldType, Term};
 use crate::tokenizer::TokenizerManager;
+use crate::Score;
 use std::borrow::Cow;
 use std::collections::HashMap;
 use std::num::{ParseFloatError, ParseIntError};
@@ -113,8 +114,9 @@ fn trim_ast(logical_ast: LogicalAST) -> Option<LogicalAST> {
 /// The language covered by the current parser is extremely simple.
 ///
 /// * simple terms: "e.g.: `Barack Obama` are simply tokenized using
-///   tantivy's `StandardTokenizer`, hence becoming `["barack", "obama"]`.
-///   The terms are then searched within the default terms of the query parser.
+///   tantivy's [`SimpleTokenizer`](../tokenizer/struct.SimpleTokenizer.html), hence
+///   becoming `["barack", "obama"]`. The terms are then searched within
+///   the default terms of the query parser.
 ///
 ///   e.g. If `body` and `title` are default fields, our example terms are
 ///   `["title:barack", "body:barack", "title:obama", "body:obama"]`.
@@ -171,7 +173,17 @@ pub struct QueryParser {
     default_fields: Vec<Field>,
     conjunction_by_default: bool,
     tokenizer_manager: TokenizerManager,
-    boost: HashMap<Field, f32>,
+    boost: HashMap<Field, Score>,
+}
+
+fn all_negative(ast: &LogicalAST) -> bool {
+    match ast {
+        LogicalAST::Leaf(_) => false,
+        LogicalAST::Boost(ref child_ast, _) => all_negative(&*child_ast),
+        LogicalAST::Clause(children) => children
+            .iter()
+            .all(|(ref occur, child)| (*occur == Occur::MustNot) || all_negative(child)),
+    }
 }
 
 impl QueryParser {
@@ -217,7 +229,7 @@ impl QueryParser {
     /// If the query defines a query boost through the query language (e.g: `country:France^3.0`),
     /// the two boosts (the one defined in the query, and the one defined in the `QueryParser`)
     /// are multiplied together.
-    pub fn set_field_boost(&mut self, field: Field, boost: f32) {
+    pub fn set_field_boost(&mut self, field: Field, boost: Score) {
         self.boost.insert(field, boost);
     }
 
@@ -253,8 +265,13 @@ impl QueryParser {
         &self,
         user_input_ast: UserInputAST,
     ) -> Result<LogicalAST, QueryParserError> {
-        let (occur, ast) = self.compute_logical_ast_with_occur(user_input_ast)?;
-        if occur == Occur::MustNot {
+        let ast = self.compute_logical_ast_with_occur(user_input_ast)?;
+        if let LogicalAST::Clause(children) = &ast {
+            if children.is_empty() {
+                return Ok(ast);
+            }
+        }
+        if all_negative(&ast) {
             return Err(QueryParserError::AllButQueryForbidden);
         }
         Ok(ast)
@@ -410,36 +427,28 @@ impl QueryParser {
     fn compute_logical_ast_with_occur(
         &self,
         user_input_ast: UserInputAST,
-    ) -> Result<(Occur, LogicalAST), QueryParserError> {
+    ) -> Result<LogicalAST, QueryParserError> {
         match user_input_ast {
             UserInputAST::Clause(sub_queries) => {
                 let default_occur = self.default_occur();
                 let mut logical_sub_queries: Vec<(Occur, LogicalAST)> = Vec::new();
-                for sub_query in sub_queries {
-                    let (occur, sub_ast) = self.compute_logical_ast_with_occur(sub_query)?;
-                    let new_occur = Occur::compose(default_occur, occur);
-                    logical_sub_queries.push((new_occur, sub_ast));
+                for (occur_opt, sub_ast) in sub_queries {
+                    let sub_ast = self.compute_logical_ast_with_occur(sub_ast)?;
+                    let occur = occur_opt.unwrap_or(default_occur);
+                    logical_sub_queries.push((occur, sub_ast));
                 }
-                Ok((Occur::Should, LogicalAST::Clause(logical_sub_queries)))
-            }
-            UserInputAST::Unary(left_occur, subquery) => {
-                let (right_occur, logical_sub_queries) =
-                    self.compute_logical_ast_with_occur(*subquery)?;
-                Ok((Occur::compose(left_occur, right_occur), logical_sub_queries))
+                Ok(LogicalAST::Clause(logical_sub_queries))
             }
             UserInputAST::Boost(ast, boost) => {
-                let (occur, ast_without_occur) = self.compute_logical_ast_with_occur(*ast)?;
-                Ok((occur, ast_without_occur.boost(boost)))
-            }
-            UserInputAST::Leaf(leaf) => {
-                let result_ast = self.compute_logical_ast_from_leaf(*leaf)?;
-                Ok((Occur::Should, result_ast))
+                let ast = self.compute_logical_ast_with_occur(*ast)?;
+                Ok(ast.boost(boost as Score))
             }
+            UserInputAST::Leaf(leaf) => self.compute_logical_ast_from_leaf(*leaf),
         }
     }
 
-    fn field_boost(&self, field: Field) -> f32 {
-        self.boost.get(&field).cloned().unwrap_or(1.0f32)
+    fn field_boost(&self, field: Field) -> Score {
+        self.boost.get(&field).cloned().unwrap_or(1.0)
     }
 
     fn compute_logical_ast_from_leaf(
@@ -650,7 +659,7 @@ mod test {
         let mut query_parser = make_query_parser();
         let schema = make_schema();
         let text_field = schema.get_field("text").unwrap();
-        query_parser.set_field_boost(text_field, 2.0f32);
+        query_parser.set_field_boost(text_field, 2.0);
         let query = query_parser.parse_query("text:hello").unwrap();
         assert_eq!(
             format!("{:?}", query),
@@ -663,7 +672,7 @@ mod test {
         let mut query_parser = make_query_parser();
         let schema = make_schema();
         let title_field = schema.get_field("title").unwrap();
-        query_parser.set_field_boost(title_field, 2.0f32);
+        query_parser.set_field_boost(title_field, 2.0);
         let query = query_parser.parse_query("title:[A TO B]").unwrap();
         assert_eq!(
             format!("{:?}", query),
@@ -676,7 +685,7 @@ mod test {
         let mut query_parser = make_query_parser();
         let schema = make_schema();
         let text_field = schema.get_field("text").unwrap();
-        query_parser.set_field_boost(text_field, 2.0f32);
+        query_parser.set_field_boost(text_field, 2.0);
         let query = query_parser.parse_query("text:hello^2").unwrap();
         assert_eq!(
             format!("{:?}", query),
@@ -782,6 +791,20 @@ mod test {
         );
     }
 
+    #[test]
+    fn test_parse_query_to_ast_ab_c() {
+        test_parse_query_to_logical_ast_helper(
+            "(+title:a +title:b) title:c",
+            "((+Term(field=0,bytes=[97]) +Term(field=0,bytes=[98])) Term(field=0,bytes=[99]))",
+            false,
+        );
+        test_parse_query_to_logical_ast_helper(
+            "(+title:a +title:b) title:c",
+            "(+(+Term(field=0,bytes=[97]) +Term(field=0,bytes=[98])) +Term(field=0,bytes=[99]))",
+            true,
+        );
+    }
+
     #[test]
     pub fn test_parse_query_to_ast_single_term() {
         test_parse_query_to_logical_ast_helper(
@@ -801,11 +824,13 @@ mod test {
              Term(field=1,bytes=[116, 105, 116, 105])))",
             false,
         );
-        assert_eq!(
-            parse_query_to_logical_ast("-title:toto", false)
-                .err()
-                .unwrap(),
-            QueryParserError::AllButQueryForbidden
+    }
+
+    #[test]
+    fn test_single_negative_term() {
+        assert_matches!(
+            parse_query_to_logical_ast("-title:toto", false),
+            Err(QueryParserError::AllButQueryForbidden)
         );
     }
 
@@ -965,6 +990,18 @@ mod test {
         assert!(query_parser.parse_query("with_stop_words:the").is_ok());
     }
 
+    #[test]
+    pub fn test_parse_query_single_negative_term_through_error() {
+        assert_matches!(
+            parse_query_to_logical_ast("-title:toto", true),
+            Err(QueryParserError::AllButQueryForbidden)
+        );
+        assert_matches!(
+            parse_query_to_logical_ast("-title:toto", false),
+            Err(QueryParserError::AllButQueryForbidden)
+        );
+    }
+
     #[test]
     pub fn test_parse_query_to_ast_conjunction() {
         test_parse_query_to_logical_ast_helper(
@@ -984,12 +1021,6 @@ mod test {
              Term(field=1,bytes=[116, 105, 116, 105])))",
             true,
         );
-        assert_eq!(
-            parse_query_to_logical_ast("-title:toto", true)
-                .err()
-                .unwrap(),
-            QueryParserError::AllButQueryForbidden
-        );
         test_parse_query_to_logical_ast_helper(
             "title:a b",
             "(+Term(field=0,bytes=[97]) \
@@ -1013,4 +1044,26 @@ mod test {
             false
         );
     }
+
+    #[test]
+    fn test_and_default_regardless_of_default_conjunctive() {
+        for &default_conjunction in &[false, true] {
+            test_parse_query_to_logical_ast_helper(
+                "title:a AND title:b",
+                "(+Term(field=0,bytes=[97]) +Term(field=0,bytes=[98]))",
+                default_conjunction,
+            );
+        }
+    }
+
+    #[test]
+    fn test_or_default_conjunctive() {
+        for &default_conjunction in &[false, true] {
+            test_parse_query_to_logical_ast_helper(
+                "title:a OR title:b",
+                "(Term(field=0,bytes=[97]) Term(field=0,bytes=[98]))",
+                default_conjunction,
+            );
+        }
+    }
 }

src/query/range_query.rs

@@ -9,8 +9,8 @@ use crate::query::{Query, Scorer, Weight};
 use crate::schema::Type;
 use crate::schema::{Field, IndexRecordOption, Term};
 use crate::termdict::{TermDictionary, TermStreamer};
-use crate::DocId;
-use crate::{Result, SkipResult};
+use crate::Result;
+use crate::{DocId, Score};
 use std::collections::Bound;
 use std::ops::Range;
 
@@ -289,7 +289,7 @@ impl RangeWeight {
 }
 
 impl Weight for RangeWeight {
-    fn scorer(&self, reader: &SegmentReader, boost: f32) -> 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);
 
@@ -300,10 +300,15 @@ impl Weight for RangeWeight {
             let term_info = term_range.value();
             let mut block_segment_postings = inverted_index
                 .read_block_postings_from_terminfo(term_info, IndexRecordOption::Basic);
-            while block_segment_postings.advance() {
+            loop {
+                let docs = block_segment_postings.docs();
+                if docs.is_empty() {
+                    break;
+                }
                 for &doc in block_segment_postings.docs() {
                     doc_bitset.insert(doc);
                 }
+                block_segment_postings.advance();
             }
         }
         let doc_bitset = BitSetDocSet::from(doc_bitset);
@@ -311,11 +316,11 @@ impl Weight for RangeWeight {
     }
 
     fn explain(&self, reader: &SegmentReader, doc: DocId) -> Result<Explanation> {
-        let mut scorer = self.scorer(reader, 1.0f32)?;
-        if scorer.skip_next(doc) != SkipResult::Reached {
+        let mut scorer = self.scorer(reader, 1.0)?;
+        if scorer.seek(doc) != doc {
             return Err(does_not_match(doc));
         }
-        Ok(Explanation::new("RangeQuery", 1.0f32))
+        Ok(Explanation::new("RangeQuery", 1.0))
     }
 }
 
@@ -323,8 +328,9 @@ impl Weight for RangeWeight {
 mod tests {
 
     use super::RangeQuery;
-    use crate::collector::Count;
-    use crate::schema::{Document, Field, Schema, INDEXED};
+    use crate::collector::{Count, TopDocs};
+    use crate::query::QueryParser;
+    use crate::schema::{Document, Field, Schema, INDEXED, TEXT};
     use crate::Index;
     use std::collections::Bound;
 
@@ -471,4 +477,28 @@ mod tests {
             91
         );
     }
+
+    #[test]
+    fn test_bug_reproduce_range_query() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        schema_builder.add_text_field("title", TEXT);
+        schema_builder.add_i64_field("year", INDEXED);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema.clone());
+        let mut index_writer = index.writer_with_num_threads(1, 10_000_000)?;
+        let title = schema.get_field("title").unwrap();
+        let year = schema.get_field("year").unwrap();
+        index_writer.add_document(doc!(
+          title => "hemoglobin blood",
+          year => 1990 as i64
+        ));
+        index_writer.commit()?;
+        let reader = index.reader()?;
+        let searcher = reader.searcher();
+        let query_parser = QueryParser::for_index(&index, vec![title]);
+        let query = query_parser.parse_query("hemoglobin AND year:[1970 TO 1990]")?;
+        let top_docs = searcher.search(&query, &TopDocs::with_limit(10))?;
+        assert_eq!(top_docs.len(), 1);
+        Ok(())
+    }
 }

src/query/regex_query.rs

@@ -89,10 +89,10 @@ impl Query for RegexQuery {
 #[cfg(test)]
 mod test {
     use super::RegexQuery;
+    use crate::assert_nearly_equals;
     use crate::collector::TopDocs;
     use crate::schema::TEXT;
     use crate::schema::{Field, Schema};
-    use crate::tests::assert_nearly_equals;
     use crate::{Index, IndexReader};
     use std::sync::Arc;
     use tantivy_fst::Regex;
@@ -129,7 +129,7 @@ mod test {
                 .unwrap();
             assert_eq!(scored_docs.len(), 1, "Expected only 1 document");
             let (score, _) = scored_docs[0];
-            assert_nearly_equals(1f32, score);
+            assert_nearly_equals!(1.0, score);
         }
         let top_docs = searcher
             .search(&query_matching_zero, &TopDocs::with_limit(2))

src/query/reqopt_scorer.rs

@@ -1,9 +1,8 @@
-use crate::docset::{DocSet, SkipResult};
+use crate::docset::DocSet;
 use crate::query::score_combiner::ScoreCombiner;
 use crate::query::Scorer;
 use crate::DocId;
 use crate::Score;
-use std::cmp::Ordering;
 use std::marker::PhantomData;
 
 /// Given a required scorer and an optional scorer
@@ -17,7 +16,6 @@ pub struct RequiredOptionalScorer<TReqScorer, TOptScorer, TScoreCombiner> {
     req_scorer: TReqScorer,
     opt_scorer: TOptScorer,
     score_cache: Option<Score>,
-    opt_finished: bool,
     _phantom: PhantomData<TScoreCombiner>,
 }
 
@@ -29,14 +27,12 @@ where
     /// Creates a new `RequiredOptionalScorer`.
     pub fn new(
         req_scorer: TReqScorer,
-        mut opt_scorer: TOptScorer,
+        opt_scorer: TOptScorer,
     ) -> RequiredOptionalScorer<TReqScorer, TOptScorer, TScoreCombiner> {
-        let opt_finished = !opt_scorer.advance();
         RequiredOptionalScorer {
             req_scorer,
             opt_scorer,
             score_cache: None,
-            opt_finished,
             _phantom: PhantomData,
         }
     }
@@ -48,7 +44,7 @@ where
     TReqScorer: DocSet,
     TOptScorer: DocSet,
 {
-    fn advance(&mut self) -> bool {
+    fn advance(&mut self) -> DocId {
         self.score_cache = None;
         self.req_scorer.advance()
     }
@@ -76,22 +72,8 @@ where
         let doc = self.doc();
         let mut score_combiner = TScoreCombiner::default();
         score_combiner.update(&mut self.req_scorer);
-        if !self.opt_finished {
-            match self.opt_scorer.doc().cmp(&doc) {
-                Ordering::Greater => {}
-                Ordering::Equal => {
-                    score_combiner.update(&mut self.opt_scorer);
-                }
-                Ordering::Less => match self.opt_scorer.skip_next(doc) {
-                    SkipResult::Reached => {
-                        score_combiner.update(&mut self.opt_scorer);
-                    }
-                    SkipResult::End => {
-                        self.opt_finished = true;
-                    }
-                    SkipResult::OverStep => {}
-                },
-            }
+        if self.opt_scorer.doc() <= doc && self.opt_scorer.seek(doc) == doc {
+            score_combiner.update(&mut self.opt_scorer);
         }
         let score = score_combiner.score();
         self.score_cache = Some(score);
@@ -102,7 +84,7 @@ where
 #[cfg(test)]
 mod tests {
     use super::RequiredOptionalScorer;
-    use crate::docset::DocSet;
+    use crate::docset::{DocSet, TERMINATED};
     use crate::postings::tests::test_skip_against_unoptimized;
     use crate::query::score_combiner::{DoNothingCombiner, SumCombiner};
     use crate::query::ConstScorer;
@@ -119,8 +101,9 @@ mod tests {
                 ConstScorer::from(VecDocSet::from(vec![])),
             );
         let mut docs = vec![];
-        while reqoptscorer.advance() {
+        while reqoptscorer.doc() != TERMINATED {
             docs.push(reqoptscorer.doc());
+            reqoptscorer.advance();
         }
         assert_eq!(docs, req);
     }
@@ -129,50 +112,49 @@ mod tests {
     fn test_reqopt_scorer() {
         let mut reqoptscorer: RequiredOptionalScorer<_, _, SumCombiner> =
             RequiredOptionalScorer::new(
-                ConstScorer::new(VecDocSet::from(vec![1, 3, 7, 8, 9, 10, 13, 15]), 1.0f32),
-                ConstScorer::new(VecDocSet::from(vec![1, 2, 7, 11, 12, 15]), 1.0f32),
+                ConstScorer::new(VecDocSet::from(vec![1, 3, 7, 8, 9, 10, 13, 15]), 1.0),
+                ConstScorer::new(VecDocSet::from(vec![1, 2, 7, 11, 12, 15]), 1.0),
             );
         {
-            assert!(reqoptscorer.advance());
             assert_eq!(reqoptscorer.doc(), 1);
-            assert_eq!(reqoptscorer.score(), 2f32);
+            assert_eq!(reqoptscorer.score(), 2.0);
         }
         {
-            assert!(reqoptscorer.advance());
+            assert_eq!(reqoptscorer.advance(), 3);
             assert_eq!(reqoptscorer.doc(), 3);
-            assert_eq!(reqoptscorer.score(), 1f32);
+            assert_eq!(reqoptscorer.score(), 1.0);
         }
         {
-            assert!(reqoptscorer.advance());
+            assert_eq!(reqoptscorer.advance(), 7);
             assert_eq!(reqoptscorer.doc(), 7);
-            assert_eq!(reqoptscorer.score(), 2f32);
+            assert_eq!(reqoptscorer.score(), 2.0);
         }
         {
-            assert!(reqoptscorer.advance());
+            assert_eq!(reqoptscorer.advance(), 8);
             assert_eq!(reqoptscorer.doc(), 8);
-            assert_eq!(reqoptscorer.score(), 1f32);
+            assert_eq!(reqoptscorer.score(), 1.0);
         }
         {
-            assert!(reqoptscorer.advance());
+            assert_eq!(reqoptscorer.advance(), 9);
             assert_eq!(reqoptscorer.doc(), 9);
-            assert_eq!(reqoptscorer.score(), 1f32);
+            assert_eq!(reqoptscorer.score(), 1.0);
         }
         {
-            assert!(reqoptscorer.advance());
+            assert_eq!(reqoptscorer.advance(), 10);
             assert_eq!(reqoptscorer.doc(), 10);
-            assert_eq!(reqoptscorer.score(), 1f32);
+            assert_eq!(reqoptscorer.score(), 1.0);
         }
         {
-            assert!(reqoptscorer.advance());
+            assert_eq!(reqoptscorer.advance(), 13);
             assert_eq!(reqoptscorer.doc(), 13);
-            assert_eq!(reqoptscorer.score(), 1f32);
+            assert_eq!(reqoptscorer.score(), 1.0);
         }
         {
-            assert!(reqoptscorer.advance());
+            assert_eq!(reqoptscorer.advance(), 15);
             assert_eq!(reqoptscorer.doc(), 15);
-            assert_eq!(reqoptscorer.score(), 2f32);
+            assert_eq!(reqoptscorer.score(), 2.0);
         }
-        assert!(!reqoptscorer.advance());
+        assert_eq!(reqoptscorer.advance(), TERMINATED);
     }
 
     #[test]