Moved BlockSegmentPostings to a different file.

Added TryInto implementation for IndexReaderBuilder

CHANGELOG.md

@@ -1,6 +1,8 @@
-Tantivy 0.12.1
-=====================
-- By default IndexReader are in `Manual` mode.
+Tantivy 0.13.0
+======================
+- 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.
 
 Tantivy 0.12.0
 ======================

Cargo.toml

@@ -13,25 +13,23 @@ keywords = ["search", "information", "retrieval"]
 edition = "2018"
 
 [dependencies]
-base64 = "0.11.0"
+base64 = "0.12.0"
 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"
 atomicwrites = {version="0.2.2", optional=true}
 tempfile = "3.0"
 log = "0.4"
-serde = "1.0"
-serde_derive = "1.0"
+serde = {version="1.0", features=["derive"]}
 serde_json = "1.0"
 num_cpus = "1.2"
 fs2={version="0.4", optional=true}
-itertools = "0.8"
-levenshtein_automata = "0.1"
+levenshtein_automata = "0.2"
 notify = {version="4", optional=true}
 uuid = { version = "0.8", features = ["v4", "serde"] }
 crossbeam = "0.7"
@@ -40,18 +38,19 @@ 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 = "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"
 rayon = "1"
+# ordered-float = "1"
 
 [target.'cfg(windows)'.dependencies]
 winapi = "0.3"
@@ -60,9 +59,11 @@ winapi = "0.3"
 rand = "0.7"
 maplit = "1"
 matches = "0.1.8"
+proptest = "0.9"
+float-cmp = "0.6"
 
 [dev-dependencies.fail]
-version = "0.3"
+version = "0.4"
 features = ["failpoints"]
 
 [profile.release]

README.md

@@ -31,16 +31,20 @@ Tantivy is, in fact, strongly inspired by Lucene's design.
 
 # Benchmark
 
-Tantivy is typically faster than Lucene, but the results depend on 
-the nature of the queries in your workload.
-
 The following [benchmark](https://tantivy-search.github.io/bench/) break downs 
 performance for different type of queries / collection.
 
+
+In general, Tantivy tends to be 
+- slower than Lucene on union with a Top-K due to Block-WAND optimization.
+- faster than Lucene on intersection and phrase queries. 
+
+Your mileage WILL vary depending on the nature of queries and their load.
+
 # Features
 
 - Full-text search
-- Configurable tokenizer (stemming available for 17 Latin languages with third party support for Chinese ([tantivy-jieba](https://crates.io/crates/tantivy-jieba) and [cang-jie](https://crates.io/crates/cang-jie)) and [Japanese](https://crates.io/crates/tantivy-tokenizer-tiny-segmenter))
+- Configurable tokenizer (stemming available for 17 Latin languages with third party support for Chinese ([tantivy-jieba](https://crates.io/crates/tantivy-jieba) and [cang-jie](https://crates.io/crates/cang-jie)), Japanese ([lindera](https://github.com/lindera-morphology/lindera-tantivy) and [tantivy-tokenizer-tiny-segmente](https://crates.io/crates/tantivy-tokenizer-tiny-segmenter)) and Korean ([lindera](https://github.com/lindera-morphology/lindera-tantivy) + [lindera-ko-dic-builder](https://github.com/lindera-morphology/lindera-ko-dic-builder))
 - Fast (check out the :racehorse: :sparkles: [benchmark](https://tantivy-search.github.io/bench/) :sparkles: :racehorse:)
 - Tiny startup time (<10ms), perfect for command line tools
 - BM25 scoring (the same as Lucene)

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/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/pre_tokenized_text.rs

@@ -9,11 +9,10 @@
 // - import tokenized text straight from json,
 // - perform a search on documents with pre-tokenized text
 
-use tantivy::tokenizer::{PreTokenizedString, SimpleTokenizer, Token, Tokenizer};
-
 use tantivy::collector::{Count, TopDocs};
 use tantivy::query::TermQuery;
 use tantivy::schema::*;
+use tantivy::tokenizer::{PreTokenizedString, SimpleTokenizer, Token, Tokenizer};
 use tantivy::{doc, Index, ReloadPolicy};
 use tempfile::TempDir;
 

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/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),
@@ -176,6 +170,16 @@ fn leaf<'a>() -> impl Parser<&'a str, Output = UserInputAST> {
     })
 }
 
+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 = f32> {
     (many1(digit()), optional((char('.'), many1(digit())))).map(
         |(int_part, decimal_part_opt): (String, Option<(char, String)>)| {
@@ -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())
 }
@@ -283,6 +295,12 @@ 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) {
@@ -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),
 }
 
 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/custom_score_top_collector.rs

@@ -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/mod.rs

@@ -84,7 +84,7 @@ See the `custom_collector` example.
 
 */
 
-use crate::DocId;
+use crate::{DocId, Searcher, Executor};
 use crate::Score;
 use crate::SegmentLocalId;
 use crate::SegmentReader;
@@ -100,6 +100,9 @@ mod top_collector;
 
 mod top_score_collector;
 pub use self::top_score_collector::TopDocs;
+#[cfg(test)]
+pub(crate) use self::top_score_collector::TopScoreSegmentCollector;
+
 
 mod custom_score_top_collector;
 pub use self::custom_score_top_collector::{CustomScorer, CustomSegmentScorer};
@@ -109,6 +112,9 @@ pub use self::tweak_score_top_collector::{ScoreSegmentTweaker, ScoreTweaker};
 
 mod facet_collector;
 pub use self::facet_collector::FacetCollector;
+use crate::fastfield::DeleteBitSet;
+use crate::query::{Scorer, Weight};
+use std::borrow::BorrowMut;
 
 /// `Fruit` is the type for the result of our collection.
 /// e.g. `usize` for the `Count` collector.
@@ -116,6 +122,8 @@ pub trait Fruit: Send + downcast_rs::Downcast {}
 
 impl<T> Fruit for T where T: Send + downcast_rs::Downcast {}
 
+
+
 /// Collectors are in charge of collecting and retaining relevant
 /// information from the document found and scored by the query.
 ///
@@ -154,6 +162,20 @@ 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>;
+
+    fn collect_weight(&self, searcher: &Searcher, weight: &dyn Weight, executor: &Executor) -> crate::Result<Self::Fruit> {
+        let segment_readers = searcher.segment_readers();
+        let fruits = executor.map(
+            |(segment_ord, segment_reader)| {
+                let mut scorer = weight.scorer(segment_reader, 1.0f32)?;
+                let segment_collector =
+                    self.for_segment(segment_ord as u32, segment_reader)?;
+                Ok(segment_collector.collect_scorer(scorer.borrow_mut(), segment_reader.delete_bitset()))
+            },
+            segment_readers.iter().enumerate(),
+        )?;
+        self.merge_fruits(fruits)
+    }
 }
 
 /// The `SegmentCollector` is the trait in charge of defining the
@@ -161,7 +183,7 @@ pub trait Collector: Sync {
 ///
 /// `.collect(doc, score)` will be called for every documents
 /// matching the query.
-pub trait SegmentCollector: 'static {
+pub trait SegmentCollector: 'static + Sized {
     /// `Fruit` is the type for the result of our collection.
     /// e.g. `usize` for the `Count` collector.
     type Fruit: Fruit;
@@ -171,6 +193,19 @@ pub trait SegmentCollector: 'static {
 
     /// Extract the fruit of the collection from the `SegmentCollector`.
     fn harvest(self) -> Self::Fruit;
+
+    fn collect_scorer(mut self, scorer: &mut dyn Scorer, delete_bitset: Option<&DeleteBitSet>) -> Self::Fruit {
+        if let Some(delete_bitset) = delete_bitset {
+            scorer.for_each(&mut |doc, score| {
+                if delete_bitset.is_alive(doc) {
+                    self.collect(doc, score);
+                }
+            });
+        } else {
+            scorer.for_each(&mut |doc, score| self.collect(doc, score));
+        }
+        self.harvest()
+    }
 }
 
 // -----------------------------------------------

src/collector/top_collector.rs

@@ -56,7 +56,7 @@ 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,
     _marker: PhantomData<T>,
 }
 
@@ -69,9 +69,7 @@ where
     /// # Panics
     /// The method panics if limit is 0
     pub fn with_limit(limit: usize) -> TopCollector<T> {
-        if limit < 1 {
-            panic!("Limit must be strictly greater than 0.");
-        }
+        assert!(limit > 0, "Limit must be strictly greater than 0.");
         TopCollector {
             limit,
             _marker: PhantomData,
@@ -124,13 +122,13 @@ where
 /// The theorical complexity for collecting the top `K` out of `n` documents
 /// is `O(n log K)`.
 pub(crate) struct TopSegmentCollector<T> {
-    limit: usize,
+    pub limit: usize,
     heap: BinaryHeap<ComparableDoc<T, DocId>>,
     segment_id: u32,
 }
 
 impl<T: PartialOrd> TopSegmentCollector<T> {
-    fn new(segment_id: SegmentLocalId, limit: usize) -> TopSegmentCollector<T> {
+    pub fn new(segment_id: SegmentLocalId, limit: usize) -> TopSegmentCollector<T> {
         TopSegmentCollector {
             limit,
             heap: BinaryHeap::with_capacity(limit),
@@ -161,6 +159,10 @@ impl<T: PartialOrd + Clone> TopSegmentCollector<T> {
         self.heap.len() >= self.limit
     }
 
+    pub fn pruning_score(&self) -> Option<T> {
+        self.heap.peek().map(|head| head.feature.clone())
+    }
+
     /// Collects a document scored by the given feature
     ///
     /// It collects documents until it has reached the max capacity. Once it reaches capacity, it

src/collector/top_score_collector.rs

@@ -8,12 +8,13 @@ use crate::collector::{
 };
 use crate::fastfield::FastFieldReader;
 use crate::schema::Field;
-use crate::DocAddress;
+use crate::{DocAddress, Executor, Searcher};
 use crate::DocId;
 use crate::Score;
 use crate::SegmentLocalId;
 use crate::SegmentReader;
 use std::fmt;
+use crate::query::{Weight, PruningScorerIfPossible};
 
 /// The `TopDocs` collector keeps track of the top `K` documents
 /// sorted by their score.
@@ -66,7 +67,7 @@ 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))
     }
 }
@@ -417,6 +418,42 @@ impl Collector for TopDocs {
         true
     }
 
+    fn collect_weight(&self, searcher: &Searcher, weight: &dyn Weight, executor: &Executor) -> crate::Result<Self::Fruit> {
+        let segment_readers = searcher.segment_readers();
+        let fruits = executor.map(
+            |(segment_ord, segment_reader)| {
+                match weight.pruning_scorer(segment_reader, 1.0f32)? {
+                    PruningScorerIfPossible::NonPruning(mut scorer) => {
+                        let segment_collector =
+                            self.for_segment(segment_ord as u32, segment_reader)?;
+                        let fruit =
+                            segment_collector.collect_scorer(scorer.as_mut(), segment_reader.delete_bitset());
+                        Ok(fruit)
+                    }
+                    PruningScorerIfPossible::Pruning(mut pruning_scorer) => {
+                        let limit = self.0.limit;
+                        let mut segment_collector =
+                            self.for_segment(segment_ord as u32, segment_reader)?;
+                        for _ in 0..limit {
+                            if !pruning_scorer.advance() {
+                                return Ok(segment_collector.harvest());
+                            }
+                            segment_collector.collect(pruning_scorer.doc(), pruning_scorer.score());
+                        }
+                        let mut pruning_score = segment_collector.0.pruning_score().unwrap_or(0.0f32);
+                        while pruning_scorer.advance_with_pruning(pruning_score) {
+                            segment_collector.0.collect(pruning_scorer.doc(), pruning_scorer.score());
+                            pruning_score = segment_collector.0.pruning_score().unwrap_or(0.0f32);
+                        }
+                        Ok(segment_collector.harvest())
+                    }
+                }
+            },
+            segment_readers.iter().enumerate(),
+        )?;
+        self.merge_fruits(fruits)
+    }
+
     fn merge_fruits(
         &self,
         child_fruits: Vec<Vec<(Score, DocAddress)>>,
@@ -428,6 +465,12 @@ impl Collector for TopDocs {
 /// Segment Collector associated to `TopDocs`.
 pub struct TopScoreSegmentCollector(TopSegmentCollector<Score>);
 
+impl TopScoreSegmentCollector {
+    pub fn new(segment_id: SegmentLocalId, limit: usize) -> Self {
+        TopScoreSegmentCollector(TopSegmentCollector::new(segment_id, limit))
+    }
+}
+
 impl SegmentCollector for TopScoreSegmentCollector {
     type Fruit = Vec<(Score, DocAddress)>;
 
@@ -450,7 +493,6 @@ mod tests {
     use crate::Index;
     use crate::IndexWriter;
     use crate::Score;
-    use itertools::Itertools;
 
     fn make_index() -> Index {
         let mut schema_builder = Schema::builder();
@@ -524,8 +566,8 @@ mod tests {
 
         // precondition for the test to be meaningful: we did get documents
         // with the same score
-        assert!(page_1.iter().map(|result| result.0).all_equal());
-        assert!(page_2.iter().map(|result| result.0).all_equal());
+        assert!(page_1.iter().all(|result| result.0 == page_1[0].0));
+        assert!(page_2.iter().all(|result| result.0 == page_2[0].0));
 
         // sanity check since we're relying on make_index()
         assert_eq!(page_1.len(), 2);

src/collector/tweak_score_top_collector.rs

@@ -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/mod.rs

@@ -18,6 +18,19 @@ pub use byteorder::LittleEndian as Endianness;
 /// We do not allow segments with more than
 pub const MAX_DOC_LIMIT: u32 = 1 << 31;
 
+pub fn minmax<I, T>(mut vals: I) -> Option<(T, T)>
+where
+    I: Iterator<Item = T>,
+    T: Copy + Ord,
+{
+    if let Some(first_el) = vals.next() {
+        return Some(vals.fold((first_el, first_el), |(min_val, max_val), el| {
+            (min_val.min(el), max_val.max(el))
+        }));
+    }
+    None
+}
+
 /// Computes the number of bits that will be used for bitpacking.
 ///
 /// In general the target is the minimum number of bits
@@ -134,6 +147,7 @@ pub fn u64_to_f64(val: u64) -> f64 {
 #[cfg(test)]
 pub(crate) mod test {
 
+    pub use super::minmax;
     pub use super::serialize::test::fixed_size_test;
     use super::{compute_num_bits, f64_to_u64, i64_to_u64, u64_to_f64, u64_to_i64};
     use std::f64;
@@ -199,4 +213,21 @@ pub(crate) mod test {
         assert!(((super::MAX_DOC_LIMIT - 1) as i32) >= 0);
         assert!((super::MAX_DOC_LIMIT as i32) < 0);
     }
+
+    #[test]
+    fn test_minmax_empty() {
+        let vals: Vec<u32> = vec![];
+        assert_eq!(minmax(vals.into_iter()), None);
+    }
+
+    #[test]
+    fn test_minmax_one() {
+        assert_eq!(minmax(vec![1].into_iter()), Some((1, 1)));
+    }
+
+    #[test]
+    fn test_minmax_two() {
+        assert_eq!(minmax(vec![1, 2].into_iter()), Some((1, 2)));
+        assert_eq!(minmax(vec![2, 1].into_iter()), Some((1, 2)));
+    }
 }

src/core/index.rs

@@ -21,7 +21,6 @@ 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;
@@ -282,7 +281,7 @@ impl Index {
                 TantivyError::LockFailure(
                     err,
                     Some(
-                        "Failed to acquire index lock. If you are using \
+                        "Failed to acquire index lock. If you are using\
                          a regular directory, this means there is already an \
                          `IndexWriter` working on this `Directory`, in this process \
                          or in a different process."
@@ -334,7 +333,8 @@ impl Index {
             .collect())
     }
 
-    pub(crate) fn segment(&self, segment_meta: SegmentMeta) -> Segment {
+    #[doc(hidden)]
+    pub fn segment(&self, segment_meta: SegmentMeta) -> Segment {
         Segment::for_index(self.clone(), segment_meta)
     }
 
@@ -346,13 +346,6 @@ impl Index {
         self.segment(segment_meta)
     }
 
-    /// Creates a new segment.
-    pub(crate) fn new_segment_unpersisted(&self) -> Segment {
-        let meta = self
-            .inventory
-            .new_segment_meta(SegmentId::generate_random(), 0);
-        Segment::new_volatile(meta, self.schema())
-    }
     /// Return a reference to the index directory.
     pub fn directory(&self) -> &ManagedDirectory {
         &self.directory
@@ -438,7 +431,7 @@ mod tests {
     }
 
     #[test]
-    fn create_should_wipe_off_existing() {
+    fn create_should_wipeoff_existing() {
         let directory = RAMDirectory::create();
         assert!(Index::create(directory.clone(), throw_away_schema()).is_ok());
         assert!(Index::exists(&directory));
@@ -475,7 +468,7 @@ 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")]
@@ -499,7 +492,7 @@ 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]
@@ -541,16 +534,12 @@ 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

@@ -3,8 +3,7 @@ use crate::core::SegmentId;
 use crate::schema::Schema;
 use crate::Opstamp;
 use census::{Inventory, TrackedObject};
-use serde;
-use serde_json;
+use serde::{Deserialize, Serialize};
 use std::collections::HashSet;
 use std::fmt;
 use std::path::PathBuf;

src/core/searcher.rs

@@ -4,7 +4,6 @@ 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;
@@ -24,17 +23,9 @@ fn collect_segment<C: Collector>(
     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())
+    let segment_collector =
+        collector.for_segment(segment_ord as u32, segment_reader)?;
+    Ok(segment_collector.collect_scorer(&mut scorer, segment_reader.delete_bitset()))
 }
 
 /// Holds a list of `SegmentReader`s ready for search.

src/core/segment.rs

@@ -3,59 +3,19 @@ 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;
 
-#[derive(Clone)]
-pub(crate) enum SegmentDirectory {
-    Persisted(ManagedDirectory),
-    Volatile(RAMDirectory),
-}
-
-impl SegmentDirectory {
-    pub fn new_volatile() -> SegmentDirectory {
-        SegmentDirectory::Volatile(RAMDirectory::default())
-    }
-}
-
-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(dir) => dir,
-            SegmentDirectory::Persisted(dir) => dir,
-        }
-    }
-}
-
-impl DerefMut for SegmentDirectory {
-    fn deref_mut(&mut self) -> &mut Self::Target {
-        match self {
-            SegmentDirectory::Volatile(dir) => dir,
-            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 {
@@ -65,55 +25,19 @@ impl fmt::Debug for Segment {
 }
 
 impl Segment {
-    /// Returns our index's schema.
-    // TODO return a ref.
-    pub fn schema(&self) -> Schema {
-        self.schema.clone()
-    }
-
-    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) -> Segment {
-        Segment {
-            schema,
-            meta,
-            directory: SegmentDirectory::new_volatile(),
-        }
-    }
-
     /// Creates a new segment given an `Index` and a `SegmentId`
     pub(crate) fn for_index(index: Index, meta: SegmentMeta) -> Segment {
-        Segment {
-            directory: SegmentDirectory::Persisted(index.directory().clone()),
-            schema: index.schema(),
-            meta,
-        }
+        Segment { index, meta }
     }
 
-    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(ram_directory) = &self.directory {
-            ram_directory.persist(&mut dest_directory)?;
-        }
-        self.directory = SegmentDirectory::Persisted(dest_directory);
-        Ok(())
+    /// 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.index.schema()
     }
 
     /// Returns the segment meta-information
@@ -127,8 +51,7 @@ impl Segment {
     /// as we finalize a fresh new segment.
     pub(crate) fn with_max_doc(self, max_doc: u32) -> Segment {
         Segment {
-            directory: self.directory,
-            schema: self.schema,
+            index: self.index,
             meta: self.meta.with_max_doc(max_doc),
         }
     }
@@ -136,8 +59,7 @@ impl Segment {
     #[doc(hidden)]
     pub fn with_delete_meta(self, num_deleted_docs: u32, opstamp: Opstamp) -> Segment {
         Segment {
-            directory: self.directory,
-            schema: self.schema,
+            index: self.index,
             meta: self.meta.with_delete_meta(num_deleted_docs, opstamp),
         }
     }
@@ -158,14 +80,14 @@ 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 write = self.directory.open_write(&path)?;
+        let write = self.index.directory_mut().open_write(&path)?;
         Ok(write)
     }
 }

src/core/segment_id.rs

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

src/core/segment_reader.rs

@@ -56,68 +56,6 @@ pub struct SegmentReader {
 }
 
 impl SegmentReader {
-    /// Open a new segment for reading.
-    pub fn open(segment: &Segment) -> crate::Result<SegmentReader> {
-        let termdict_source = segment.open_read(SegmentComponent::TERMS)?;
-        let termdict_composite = CompositeFile::open(&termdict_source)?;
-
-        let store_source = segment.open_read(SegmentComponent::STORE)?;
-
-        fail_point!("SegmentReader::open#middle");
-
-        let postings_source = segment.open_read(SegmentComponent::POSTINGS)?;
-        let postings_composite = CompositeFile::open(&postings_source)?;
-
-        let positions_composite = {
-            if let Ok(source) = segment.open_read(SegmentComponent::POSITIONS) {
-                CompositeFile::open(&source)?
-            } else {
-                CompositeFile::empty()
-            }
-        };
-
-        let positions_idx_composite = {
-            if let Ok(source) = segment.open_read(SegmentComponent::POSITIONSSKIP) {
-                CompositeFile::open(&source)?
-            } else {
-                CompositeFile::empty()
-            }
-        };
-
-        let schema = segment.schema();
-
-        let fast_fields_data = segment.open_read(SegmentComponent::FASTFIELDS)?;
-        let fast_fields_composite = CompositeFile::open(&fast_fields_data)?;
-        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 delete_bitset_opt = if segment.meta().has_deletes() {
-            let delete_data = segment.open_read(SegmentComponent::DELETE)?;
-            Some(DeleteBitSet::open(delete_data))
-        } else {
-            None
-        };
-
-        Ok(SegmentReader {
-            inv_idx_reader_cache: Arc::new(RwLock::new(HashMap::new())),
-            max_doc: segment.meta().max_doc(),
-            num_docs: segment.meta().num_docs(),
-            termdict_composite,
-            postings_composite,
-            fast_fields_readers: fast_field_readers,
-            fieldnorms_composite,
-            segment_id: segment.id(),
-            store_source,
-            delete_bitset_opt,
-            positions_composite,
-            positions_idx_composite,
-            schema,
-        })
-    }
-
     /// Returns the highest document id ever attributed in
     /// this segment + 1.
     /// Today, `tantivy` does not handle deletes, so it happens
@@ -204,68 +142,68 @@ impl SegmentReader {
     pub fn get_store_reader(&self) -> StoreReader {
         StoreReader::from_source(self.store_source.clone())
     }
-    //
-    //    /// Open a new segment for reading.
-    //    pub fn open(segment: &Segment) -> crate::Result<SegmentReader> {
-    //        let termdict_source = segment.open_read(SegmentComponent::TERMS)?;
-    //        let termdict_composite = CompositeFile::open(&termdict_source)?;
-    //
-    //        let store_source = segment.open_read(SegmentComponent::STORE)?;
-    //
-    //        fail_point!("SegmentReader::open#middle");
-    //
-    //        let postings_source = segment.open_read(SegmentComponent::POSTINGS)?;
-    //        let postings_composite = CompositeFile::open(&postings_source)?;
-    //
-    //        let positions_composite = {
-    //            if let Ok(source) = segment.open_read(SegmentComponent::POSITIONS) {
-    //                CompositeFile::open(&source)?
-    //            } else {
-    //                CompositeFile::empty()
-    //            }
-    //        };
-    //
-    //        let positions_idx_composite = {
-    //            if let Ok(source) = segment.open_read(SegmentComponent::POSITIONSSKIP) {
-    //                CompositeFile::open(&source)?
-    //            } else {
-    //                CompositeFile::empty()
-    //            }
-    //        };
-    //
-    //        let schema = segment.schema();
-    //
-    //        let fast_fields_data = segment.open_read(SegmentComponent::FASTFIELDS)?;
-    //        let fast_fields_composite = CompositeFile::open(&fast_fields_data)?;
-    //        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 delete_bitset_opt = if segment.meta().has_deletes() {
-    //            let delete_data = segment.open_read(SegmentComponent::DELETE)?;
-    //            Some(DeleteBitSet::open(delete_data))
-    //        } else {
-    //            None
-    //        };
-    //
-    //        Ok(SegmentReader {
-    //            inv_idx_reader_cache: Arc::new(RwLock::new(HashMap::new())),
-    //            max_doc: segment.meta().max_doc(),
-    //            num_docs: segment.meta().num_docs(),
-    //            termdict_composite,
-    //            postings_composite,
-    //            fast_fields_readers: fast_field_readers,
-    //            fieldnorms_composite,
-    //            segment_id: segment.id(),
-    //            store_source,
-    //            delete_bitset_opt,
-    //            positions_composite,
-    //            positions_idx_composite,
-    //            schema,
-    //        })
-    //    }
+
+    /// Open a new segment for reading.
+    pub fn open(segment: &Segment) -> crate::Result<SegmentReader> {
+        let termdict_source = segment.open_read(SegmentComponent::TERMS)?;
+        let termdict_composite = CompositeFile::open(&termdict_source)?;
+
+        let store_source = segment.open_read(SegmentComponent::STORE)?;
+
+        fail_point!("SegmentReader::open#middle");
+
+        let postings_source = segment.open_read(SegmentComponent::POSTINGS)?;
+        let postings_composite = CompositeFile::open(&postings_source)?;
+
+        let positions_composite = {
+            if let Ok(source) = segment.open_read(SegmentComponent::POSITIONS) {
+                CompositeFile::open(&source)?
+            } else {
+                CompositeFile::empty()
+            }
+        };
+
+        let positions_idx_composite = {
+            if let Ok(source) = segment.open_read(SegmentComponent::POSITIONSSKIP) {
+                CompositeFile::open(&source)?
+            } else {
+                CompositeFile::empty()
+            }
+        };
+
+        let schema = segment.schema();
+
+        let fast_fields_data = segment.open_read(SegmentComponent::FASTFIELDS)?;
+        let fast_fields_composite = CompositeFile::open(&fast_fields_data)?;
+        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 delete_bitset_opt = if segment.meta().has_deletes() {
+            let delete_data = segment.open_read(SegmentComponent::DELETE)?;
+            Some(DeleteBitSet::open(delete_data))
+        } else {
+            None
+        };
+
+        Ok(SegmentReader {
+            inv_idx_reader_cache: Arc::new(RwLock::new(HashMap::new())),
+            max_doc: segment.meta().max_doc(),
+            num_docs: segment.meta().num_docs(),
+            termdict_composite,
+            postings_composite,
+            fast_fields_readers: fast_field_readers,
+            fieldnorms_composite,
+            segment_id: segment.id(),
+            store_source,
+            delete_bitset_opt,
+            positions_composite,
+            positions_idx_composite,
+            schema,
+        })
+    }
 
     /// Returns a field reader associated to the field given in argument.
     /// If the field was not present in the index during indexing time,

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 +13,12 @@ use crate::directory::WatchCallback;
 use crate::directory::WatchCallbackList;
 use crate::directory::WatchHandle;
 use crate::directory::{TerminatingWrite, WritePtr};
-use atomicwrites;
+use fs2::FileExt;
 use memmap::Mmap;
+use notify::RawEvent;
+use notify::RecursiveMode;
+use notify::Watcher;
+use serde::{Deserialize, Serialize};
 use std::collections::HashMap;
 use std::convert::From;
 use std::fmt;
@@ -223,17 +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> {
@@ -267,14 +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.
@@ -284,7 +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.
@@ -302,7 +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/ram_directory.rs

@@ -145,6 +145,12 @@ impl RAMDirectory {
         self.fs.read().unwrap().total_mem_usage()
     }
 
+    /// Write a copy of all of the files saved in the RAMDirectory in the target `Directory`.
+    ///
+    /// Files are all written using the `Directory::write` meaning, even if they were
+    /// written using the `atomic_write` api.
+    ///
+    /// If an error is encounterred, files may be persisted partially.
     pub fn persist(&self, dest: &mut dyn Directory) -> crate::Result<()> {
         let wlock = self.fs.write().unwrap();
         for (path, source) in wlock.fs.iter() {
@@ -214,3 +220,28 @@ impl Directory for RAMDirectory {
         Ok(self.fs.write().unwrap().watch(watch_callback))
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::RAMDirectory;
+    use crate::Directory;
+    use std::io::Write;
+    use std::path::Path;
+
+    #[test]
+    fn test_persist() {
+        let msg_atomic: &'static [u8] = b"atomic is the way";
+        let msg_seq: &'static [u8] = b"sequential is the way";
+        let path_atomic: &'static Path = Path::new("atomic");
+        let path_seq: &'static Path = Path::new("seq");
+        let mut directory = RAMDirectory::create();
+        assert!(directory.atomic_write(path_atomic, msg_atomic).is_ok());
+        let mut wrt = directory.open_write(path_seq).unwrap();
+        assert!(wrt.write_all(msg_seq).is_ok());
+        assert!(wrt.flush().is_ok());
+        let mut directory_copy = RAMDirectory::create();
+        assert!(directory.persist(&mut directory_copy).is_ok());
+        assert_eq!(directory_copy.atomic_read(path_atomic).unwrap(), msg_atomic);
+        assert_eq!(directory_copy.atomic_read(path_seq).unwrap(), msg_seq);
+    }
+}

src/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/multivalued/writer.rs

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

src/indexer/index_writer.rs

@@ -8,33 +8,30 @@ 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::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::MergePolicy;
 use crate::indexer::SegmentEntry;
 use crate::indexer::SegmentWriter;
-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;
-use smallvec::{smallvec, SmallVec};
+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;
 
@@ -71,8 +68,6 @@ pub struct IndexWriter {
     // lifetime of the lock with that of the IndexWriter.
     _directory_lock: Option<DirectoryLock>,
 
-    segment_registers: Arc<RwLock<SegmentRegisters>>,
-
     index: Index,
 
     heap_size_in_bytes_per_thread: usize,
@@ -92,8 +87,6 @@ pub struct IndexWriter {
 
     stamper: Stamper,
     committed_opstamp: Opstamp,
-
-    on_commit: WatchCallbackList,
 }
 
 fn compute_deleted_bitset(
@@ -140,6 +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<()> {
@@ -148,38 +142,32 @@ 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() && 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,
     )?;
 
-    // TODO optimize... We are simply manipulating bitsets here.
-    // We should be able to compute the union much faster.
+    // TODO optimize
+    // It should be possible to do something smarter by manipulation bitsets directly
+    // to compute this union.
     if let Some(seg_delete_bitset) = segment_reader.delete_bitset() {
         for doc in 0u32..max_doc {
             if seg_delete_bitset.is_deleted(doc) {
@@ -188,23 +176,16 @@ pub(crate) fn advance_deletes(
         }
     }
 
-    let num_deleted_docs = delete_bitset.len() as u32;
-
+    let num_deleted_docs: u32 = delete_bitset.len() as u32;
     if num_deleted_docs > num_deleted_docs_before {
-        // We need to write a new delete file.
-        let mut delete_file = segment
-            .with_delete_meta(num_deleted_docs, target_opstamp)
-            .open_write(SegmentComponent::DELETE)?;
+        // There are new deletes. We need to write a new delete file.
+        segment = segment.with_delete_meta(num_deleted_docs as u32, target_opstamp);
+        let mut delete_file = segment.open_write(SegmentComponent::DELETE)?;
         write_delete_bitset(&delete_bitset, max_doc, &mut delete_file)?;
         delete_file.terminate()?;
-        segment_entry.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(())
 }
 
@@ -213,12 +194,11 @@ fn index_documents(
     segment: Segment,
     grouped_document_iterator: &mut dyn Iterator<Item = OperationGroup>,
     segment_updater: &mut SegmentUpdater,
-    tokenizers: &TokenizerManager,
     mut delete_cursor: DeleteCursor,
 ) -> crate::Result<bool> {
     let schema = segment.schema();
-    let mut segment_writer =
-        SegmentWriter::for_segment(memory_budget, segment.clone(), &schema, tokenizers)?;
+
+    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)?;
@@ -256,7 +236,11 @@ fn index_documents(
         last_docstamp,
     )?;
 
-    let segment_entry = SegmentEntry::new(segment_with_max_doc, 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)
 }
@@ -328,24 +312,16 @@ impl IndexWriter {
 
         let delete_queue = DeleteQueue::new();
 
-        let meta = index.load_metas()?;
+        let current_opstamp = index.load_metas()?.opstamp;
 
-        let stamper = Stamper::new(meta.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 stamper = Stamper::new(current_opstamp);
 
         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),
-            segment_registers,
+
             heap_size_in_bytes_per_thread,
             index: index.clone(),
 
@@ -359,12 +335,10 @@ impl IndexWriter {
 
             delete_queue,
 
-            committed_opstamp: meta.opstamp,
+            committed_opstamp: current_opstamp,
             stamper,
 
             worker_id: 0,
-
-            on_commit: Default::default(),
         };
         index_writer.start_workers()?;
         Ok(index_writer)
@@ -404,6 +378,13 @@ impl IndexWriter {
         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.
     ///
     /// This method is useful only for users trying to do complex
@@ -452,13 +433,12 @@ impl IndexWriter {
                         // was dropped.
                         return Ok(());
                     }
-                    let segment = index.new_segment_unpersisted();
+                    let segment = index.new_segment();
                     index_documents(
                         mem_budget,
                         segment,
                         &mut document_iterator,
                         &mut segment_updater,
-                        index.tokenizers(),
                         delete_cursor.clone(),
                     )?;
                 }
@@ -485,21 +465,6 @@ impl IndexWriter {
         Ok(())
     }
 
-    // TODO move me
-    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))
-    }
-
     /// Detects and removes the files that are not used by the index anymore.
     pub fn garbage_collect_files(
         &self,
@@ -643,7 +608,7 @@ 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.
@@ -671,7 +636,7 @@ 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)
     }
@@ -691,11 +656,7 @@ 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()
+        self.prepare_commit()?.commit()
     }
 
     pub(crate) fn segment_updater(&self) -> &SegmentUpdater {
@@ -820,6 +781,7 @@ impl Drop for IndexWriter {
 
 #[cfg(test)]
 mod tests {
+
     use super::super::operation::UserOperation;
     use crate::collector::TopDocs;
     use crate::directory::error::LockError;
@@ -846,6 +808,46 @@ mod tests {
         assert_eq!(batch_opstamp1, 2u64);
     }
 
+    #[test]
+    fn test_no_need_to_rewrite_delete_file_if_no_new_deletes() {
+        let mut schema_builder = schema::Schema::builder();
+        let text_field = schema_builder.add_text_field("text", schema::TEXT);
+        let index = Index::create_in_ram(schema_builder.build());
+
+        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        index_writer.add_document(doc!(text_field => "hello1"));
+        index_writer.add_document(doc!(text_field => "hello2"));
+        assert!(index_writer.commit().is_ok());
+
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+        assert_eq!(searcher.segment_readers().len(), 1);
+        assert_eq!(searcher.segment_reader(0u32).num_deleted_docs(), 0);
+
+        index_writer.delete_term(Term::from_field_text(text_field, "hello1"));
+        assert!(index_writer.commit().is_ok());
+
+        assert!(reader.reload().is_ok());
+        let searcher = reader.searcher();
+        assert_eq!(searcher.segment_readers().len(), 1);
+        assert_eq!(searcher.segment_reader(0u32).num_deleted_docs(), 1);
+
+        let previous_delete_opstamp = index.load_metas().unwrap().segments[0].delete_opstamp();
+
+        // All docs containing hello1 have been already removed.
+        // We should not update the delete meta.
+        index_writer.delete_term(Term::from_field_text(text_field, "hello1"));
+        assert!(index_writer.commit().is_ok());
+
+        assert!(reader.reload().is_ok());
+        let searcher = reader.searcher();
+        assert_eq!(searcher.segment_readers().len(), 1);
+        assert_eq!(searcher.segment_reader(0u32).num_deleted_docs(), 1);
+
+        let after_delete_opstamp = index.load_metas().unwrap().segments[0].delete_opstamp();
+        assert_eq!(after_delete_opstamp, previous_delete_opstamp);
+    }
+
     #[test]
     fn test_ordered_batched_operations() {
         // * one delete for `doc!(field=>"a")`
@@ -1052,8 +1054,7 @@ 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");
             }
@@ -1086,8 +1087,7 @@ 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");
             }
@@ -1262,43 +1262,7 @@ mod tests {
         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());
-    }
-
-    #[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);
+        let commit = index_writer.commit();
+        assert!(commit.is_ok());
     }
 }

src/indexer/merger.rs

@@ -21,7 +21,6 @@ use crate::store::StoreWriter;
 use crate::termdict::TermMerger;
 use crate::termdict::TermOrdinal;
 use crate::DocId;
-use itertools::Itertools;
 use std::cmp;
 use std::collections::HashMap;
 
@@ -70,11 +69,11 @@ fn compute_min_max_val(
             Some(delete_bitset) => {
                 // some deleted documents,
                 // we need to recompute the max / min
-                (0..max_doc)
-                    .filter(|doc_id| delete_bitset.is_alive(*doc_id))
-                    .map(|doc_id| u64_reader.get(doc_id))
-                    .minmax()
-                    .into_option()
+                crate::common::minmax(
+                    (0..max_doc)
+                        .filter(|doc_id| delete_bitset.is_alive(*doc_id))
+                        .map(|doc_id| u64_reader.get(doc_id)),
+                )
             }
             None => {
                 // no deleted documents,

src/indexer/mod.rs

@@ -23,7 +23,6 @@ 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/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
@@ -21,81 +19,55 @@ 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,
 }
 
 impl SegmentEntry {
     /// Create a new `SegmentEntry`
-    pub(crate) fn new(
-        segment: Segment,
+    pub fn new(
+        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 segment(&self) -> &Segment {
-        &self.segment
-    }
-
-    /// `Takes` (as in Option::take) the delete bitset of
-    /// a segment entry.
+    /// Return a reference to the segment entry deleted bitset.
     ///
     /// `DocId` in this bitset are flagged as deleted.
-    pub fn take_delete_bitset(&mut self) -> Option<BitSet> {
-        self.delete_bitset.take()
+    pub fn delete_bitset(&self) -> Option<&BitSet> {
+        self.delete_bitset.as_ref()
     }
 
-    /// Reset the delete informmationo in this segment.
-    ///
-    /// 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;
+    /// Set the `SegmentMeta` for this segment.
+    pub fn set_meta(&mut self, segment_meta: SegmentMeta) {
+        self.meta = segment_meta;
     }
 
-    pub fn set_delete_cursor(&mut self, delete_cursor: DeleteCursor) {
-        self.delete_cursor = delete_cursor;
-    }
     /// 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.segment.id()
+        self.meta.id()
     }
 
     /// 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 {
-        let num_deletes = self.delete_bitset.as_ref().map(|bitset| bitset.len());
-        write!(
-            formatter,
-            "SegmentEntry(seg={:?}, ndel={:?})",
-            self.segment, num_deletes
-        )
+        write!(formatter, "SegmentEntry({:?})", self.meta)
     }
 }

src/indexer/segment_manager.rs

@@ -1,14 +1,16 @@
 use super::segment_register::SegmentRegister;
 use crate::core::SegmentId;
 use crate::core::SegmentMeta;
+use crate::error::TantivyError;
+use crate::indexer::delete_queue::DeleteCursor;
 use crate::indexer::SegmentEntry;
-use crate::Segment;
 use std::collections::hash_set::HashSet;
-use std::sync::{Arc, RwLock};
+use std::fmt::{self, Debug, Formatter};
+use std::sync::RwLock;
 use std::sync::{RwLockReadGuard, RwLockWriteGuard};
 
 #[derive(Default)]
-pub(crate) struct SegmentRegisters {
+struct SegmentRegisters {
     uncommitted: SegmentRegister,
     committed: SegmentRegister,
 }
@@ -20,17 +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
@@ -53,7 +44,18 @@ impl SegmentRegisters {
 /// changes (merges especially)
 #[derive(Default)]
 pub struct SegmentManager {
-    registers: Arc<RwLock<SegmentRegisters>>,
+    registers: RwLock<SegmentRegisters>,
+}
+
+impl Debug for SegmentManager {
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        let lock = self.read();
+        write!(
+            f,
+            "{{ uncommitted: {:?}, committed: {:?} }}",
+            lock.uncommitted, lock.committed
+        )
+    }
 }
 
 pub fn get_mergeable_segments(
@@ -72,8 +74,16 @@ pub fn get_mergeable_segments(
 }
 
 impl SegmentManager {
-    pub(crate) fn new(registers: Arc<RwLock<SegmentRegisters>>) -> SegmentManager {
-        SegmentManager { registers }
+    pub fn from_segments(
+        segment_metas: Vec<SegmentMeta>,
+        delete_cursor: &DeleteCursor,
+    ) -> SegmentManager {
+        SegmentManager {
+            registers: RwLock::new(SegmentRegisters {
+                uncommitted: SegmentRegister::default(),
+                committed: SegmentRegister::new(segment_metas, delete_cursor),
+            }),
+        }
     }
 
     /// Returns all of the segment entries (committed or uncommitted)
@@ -155,7 +165,7 @@ impl SegmentManager {
             let error_msg = "Merge operation sent for segments that are not \
                              all uncommited or commited."
                 .to_string();
-            return Err(crate::Error::InvalidArgument(error_msg));
+            return Err(TantivyError::InvalidArgument(error_msg));
         }
         Ok(segment_entries)
     }

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 meta in segment_metas {
-            let segment_id = meta.id();
-            let segment = Segment::new_persisted(meta, directory.clone(), schema.clone());
-            let segment_entry = SegmentEntry::new(segment, delete_cursor.clone(), None);
+        for segment_meta in segment_metas {
+            let segment_id = segment_meta.id();
+            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::schema::Schema;
 
     fn segment_ids(segment_register: &SegmentRegister) -> Vec<SegmentId> {
         segment_register
@@ -125,7 +108,6 @@ 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();
@@ -133,24 +115,21 @@ mod tests {
         let segment_id_merged = SegmentId::generate_random();
 
         {
-            let meta = inventory.new_segment_meta(segment_id_a, 0u32);
-            let segment = Segment::new_volatile(meta, schema.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 segment_meta = inventory.new_segment_meta(segment_id_b, 0u32);
-            let segment = Segment::new_volatile(segment_meta, schema.clone());
-            let segment_entry = SegmentEntry::new(segment, delete_queue.cursor(), None);
+            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 = Segment::new_volatile(segment_meta_merged, schema);
-            let segment_entry = SegmentEntry::new(segment, 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_updater.rs

@@ -7,10 +7,11 @@ 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;
@@ -22,7 +23,6 @@ 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;
@@ -116,7 +116,8 @@ fn merge(
 
     // 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().clone();
@@ -132,13 +133,11 @@ fn merge(
     // ... 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 max_doc = merger.write(segment_serializer)?;
+    let num_docs = merger.write(segment_serializer)?;
 
-    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 {
@@ -162,11 +161,12 @@ pub(crate) struct InnerSegmentUpdater {
 
 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)
@@ -233,7 +233,6 @@ impl SegmentUpdater {
         &self,
         segment_entry: SegmentEntry,
     ) -> impl Future<Output = crate::Result<()>> {
-        // TODO temporary: serializing the segment at this point.
         let segment_updater = self.clone();
         self.schedule_future(async move {
             segment_updater.segment_manager.add_segment(segment_entry);
@@ -262,7 +261,8 @@ 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)
     }
@@ -330,21 +330,12 @@ 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(())
@@ -482,14 +473,17 @@ 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 {
-                        let _index = &segment_updater.index;
-                        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(),

src/indexer/segment_writer.rs

@@ -11,9 +11,8 @@ use crate::schema::Schema;
 use crate::schema::Term;
 use crate::schema::Value;
 use crate::schema::{Field, FieldEntry};
-use crate::tokenizer::TokenizerManager;
-use crate::tokenizer::{BoxTokenStream, FacetTokenizer};
-use crate::tokenizer::{PreTokenizedStream, TextAnalyzer};
+use crate::tokenizer::{BoxTokenStream, PreTokenizedStream};
+use crate::tokenizer::{FacetTokenizer, TextAnalyzer};
 use crate::tokenizer::{TokenStreamChain, Tokenizer};
 use crate::DocId;
 use crate::Opstamp;
@@ -65,12 +64,10 @@ impl SegmentWriter {
         memory_budget: usize,
         mut segment: Segment,
         schema: &Schema,
-        tokenizer_manager: &TokenizerManager,
     ) -> crate::Result<SegmentWriter> {
         let table_num_bits = initial_table_size(memory_budget)?;
         let segment_serializer = SegmentSerializer::for_segment(&mut segment)?;
-        let multifield_postings = MultiFieldPostingsWriter::new(&schema, table_num_bits);
-
+        let multifield_postings = MultiFieldPostingsWriter::new(schema, table_num_bits);
         let tokenizers = schema
             .fields()
             .map(
@@ -79,7 +76,7 @@ impl SegmentWriter {
                         .get_indexing_options()
                         .and_then(|text_index_option| {
                             let tokenizer_name = &text_index_option.tokenizer();
-                            tokenizer_manager.get(tokenizer_name)
+                            segment.index().tokenizers().get(tokenizer_name)
                         }),
                     _ => None,
                 },
@@ -88,9 +85,9 @@ impl SegmentWriter {
         Ok(SegmentWriter {
             max_doc: 0,
             multifield_postings,
-            fieldnorms_writer: FieldNormsWriter::for_schema(&schema),
+            fieldnorms_writer: FieldNormsWriter::for_schema(schema),
             segment_serializer,
-            fast_field_writers: FastFieldsWriter::from_schema(&schema),
+            fast_field_writers: FastFieldsWriter::from_schema(schema),
             doc_opstamps: Vec::with_capacity(1_000),
             tokenizers,
         })

src/lib.rs

@@ -98,9 +98,6 @@
 //! [literate programming](https://tantivy-search.github.io/examples/basic_search.html) /
 //! [source code](https://github.com/tantivy-search/tantivy/blob/master/examples/basic_search.rs))
 
-#[macro_use]
-extern crate serde_derive;
-
 #[cfg_attr(test, macro_use)]
 extern crate serde_json;
 
@@ -121,7 +118,6 @@ mod functional_test;
 mod macros;
 
 pub use crate::error::TantivyError;
-pub use crate::error::TantivyError as Error;
 pub use chrono;
 
 /// Tantivy result.
@@ -174,6 +170,7 @@ pub use crate::schema::{Document, Term};
 use std::fmt;
 
 use once_cell::sync::Lazy;
+use serde::{Deserialize, Serialize};
 
 /// Index format version.
 const INDEX_FORMAT_VERSION: u32 = 1;

src/postings/block_max_postings.rs

@@ -0,0 +1,16 @@
+use crate::postings::Postings;
+use crate::DocId;
+
+/// Inverted list with additional information about the maximum term frequency
+/// within a block, as well as globally within the list.
+pub trait BlockMaxPostings: Postings {
+    /// Returns the maximum frequency in the entire list.
+    fn max_term_freq(&self) -> u32;
+    /// Returns the maximum frequency in the current block.
+    fn block_max_term_freq(&mut self) -> u32;
+    /// Returns the document with the largest frequency.
+    fn max_doc(&self) -> DocId;
+    /// Returns the document with the largest frequency within the current
+    /// block.
+    fn block_max_doc(&self) -> DocId;
+}

src/postings/block_max_segment_postings.rs

@@ -0,0 +1,76 @@
+use crate::postings::{BlockMaxPostings, Postings, SegmentPostings};
+use crate::{DocId, DocSet, SkipResult};
+
+/// A wrapper over [`SegmentPostings`](./struct.SegmentPostings.html)
+/// with max block frequencies.
+pub struct BlockMaxSegmentPostings {
+    postings: SegmentPostings,
+    max_blocks: SegmentPostings,
+    doc_with_max_term_freq: DocId,
+    max_term_freq: u32,
+}
+
+impl BlockMaxSegmentPostings {
+    /// Constructs a new segment postings with block-max information.
+    pub fn new(
+        postings: SegmentPostings,
+        max_blocks: SegmentPostings,
+        doc_with_max_term_freq: DocId,
+        max_term_freq: u32,
+    ) -> Self {
+        Self {
+            postings,
+            max_blocks,
+            doc_with_max_term_freq,
+            max_term_freq,
+        }
+    }
+}
+
+impl DocSet for BlockMaxSegmentPostings {
+    fn advance(&mut self) -> bool {
+        self.postings.advance()
+    }
+
+    fn doc(&self) -> DocId {
+        self.postings.doc()
+    }
+
+    fn size_hint(&self) -> u32 {
+        self.postings.size_hint()
+    }
+
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        self.postings.skip_next(target)
+    }
+}
+
+impl Postings for BlockMaxSegmentPostings {
+    fn term_freq(&self) -> u32 {
+        self.postings.term_freq()
+    }
+    fn positions_with_offset(&mut self, offset: u32, output: &mut Vec<u32>) {
+        self.postings.positions_with_offset(offset, output);
+    }
+    fn positions(&mut self, output: &mut Vec<u32>) {
+        self.postings.positions(output);
+    }
+}
+
+impl BlockMaxPostings for BlockMaxSegmentPostings {
+    fn max_term_freq(&self) -> u32 {
+        self.max_term_freq
+    }
+    fn block_max_term_freq(&mut self) -> u32 {
+        if let SkipResult::End = self.max_blocks.skip_next(self.doc()) {
+            panic!("Max blocks corrupted: reached end of max block");
+        }
+        self.max_blocks.term_freq()
+    }
+    fn max_doc(&self) -> DocId {
+        self.doc_with_max_term_freq
+    }
+    fn block_max_doc(&self) -> DocId {
+        self.max_blocks.doc()
+    }
+}

src/postings/block_segment_postings.rs

@@ -0,0 +1,316 @@
+use crate::DocId;
+use tantivy_fst::Streamer;
+use crate::postings::{SkipReader, FreqReadingOption, USE_SKIP_INFO_LIMIT};
+use owned_read::OwnedRead;
+use crate::postings::compression::{BlockDecoder, COMPRESSION_BLOCK_SIZE, VIntDecoder, compressed_block_size, AlignedBuffer};
+use crate::schema::IndexRecordOption;
+use crate::common::{VInt, BinarySerializable};
+
+
+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)
+    }
+}
+
+/// `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,
+}
+
+
+#[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]
+    pub(crate) 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<'a> Streamer<'a> for BlockSegmentPostings {
+    type Item = &'a [DocId];
+
+    fn next(&'a mut self) -> Option<&'a [DocId]> {
+        if self.advance() {
+            Some(self.docs())
+        } else {
+            None
+        }
+    }
+}

src/postings/mod.rs

@@ -2,6 +2,8 @@
 Postings module (also called inverted index)
 */
 
+mod block_max_postings;
+mod block_max_segment_postings;
 mod block_search;
 pub(crate) mod compression;
 /// Postings module
@@ -12,6 +14,7 @@ mod postings;
 mod postings_writer;
 mod recorder;
 mod segment_postings;
+mod block_segment_postings;
 mod serializer;
 mod skip;
 mod stacker;
@@ -27,7 +30,11 @@ pub use self::postings::Postings;
 pub(crate) use self::skip::SkipReader;
 pub use self::term_info::TermInfo;
 
-pub use self::segment_postings::{BlockSegmentPostings, SegmentPostings};
+pub use self::segment_postings::SegmentPostings;
+
+pub use self::block_segment_postings::BlockSegmentPostings;
+pub use self::block_max_postings::BlockMaxPostings;
+pub use self::block_max_segment_postings::BlockMaxSegmentPostings;
 
 pub(crate) use self::stacker::compute_table_size;
 
@@ -220,8 +227,7 @@ pub mod tests {
 
         {
             let mut segment_writer =
-                SegmentWriter::for_segment(3_000_000, segment.clone(), &schema, index.tokenizers())
-                    .unwrap();
+                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

src/postings/segment_postings.rs

@@ -1,21 +1,16 @@
 use crate::common::BitSet;
 use crate::common::HasLen;
-use crate::common::{BinarySerializable, VInt};
 use crate::docset::{DocSet, SkipResult};
 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::{BlockSearcher, BlockSegmentPostings};
 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::postings::block_segment_postings::BlockSegmentPostingsSkipResult;
 
 struct PositionComputer {
     // store the amount of position int
@@ -299,313 +294,6 @@ impl Postings for SegmentPostings {
     }
 }
 
-/// `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;

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;

src/query/block_max_scorer.rs

@@ -0,0 +1,31 @@
+use crate::docset::DocSet;
+use crate::DocId;
+use crate::Score;
+use downcast_rs::impl_downcast;
+use crate::query::Scorer;
+
+/// A set of documents matching a query within a specific segment
+/// and having a maximum score within certain blocks.
+///
+/// See [`Query`](./trait.Query.html) and [`Scorer`](./trait.Scorer.html).
+pub trait BlockMaxScorer: downcast_rs::Downcast + DocSet + Scorer + 'static {
+    /// Returns the maximum score within the current block.
+    ///
+    /// The blocks are defined when indexing. For example, blocks can be
+    /// have a specific number postings each, or can be optimized for
+    /// retrieval speed. Read more in
+    /// [Faster BlockMax WAND with Variable-sized Blocks][vbmw]
+    ///
+    /// This method will perform a bit of computation and is not cached.
+    ///
+    /// [vbmw]: https://dl.acm.org/doi/abs/10.1145/3077136.3080780
+    fn block_max_score(&mut self) -> Score;
+
+    /// Returns the last document in the current block.
+    fn block_max_doc(&mut self) -> DocId;
+
+    /// Returns the maximum possible score within the entire document set.
+    fn max_score(&self) -> Score;
+}
+
+impl_downcast!(BlockMaxScorer);

src/query/block_max_wand.rs

@@ -0,0 +1,613 @@
+use crate::docset::{DocSet, SkipResult};
+use crate::query::score_combiner::ScoreCombiner;
+use crate::query::{BlockMaxScorer, Scorer};
+use crate::DocId;
+use crate::Score;
+use crate::query::weight::PruningScorer;
+
+#[derive(Debug, Copy, Clone, PartialEq, Eq)]
+struct Pivot {
+    position: usize,
+    first_occurrence: usize,
+    doc: DocId,
+}
+
+
+/// Find the position in the sorted list of posting lists of the **pivot**.
+///
+/// docsets need to be advanced, and are required to be sorted by the doc they point to.
+///
+/// The pivot is then defined as the lowest DocId that has a chance of matching our condition.
+fn find_pivot_position<'a, TScorer>(
+    mut docsets: impl Iterator<Item = &'a TScorer>,
+    lower_bound_score: Score,
+) -> Option<Pivot>
+    where TScorer: BlockMaxScorer
+{
+    let mut position = 0;
+    let mut upper_bound = Score::default();
+    while let Some(docset) = docsets.next() {
+        upper_bound += docset.max_score();
+        if lower_bound_score < upper_bound {
+            let pivot_doc = docset.doc();
+            let first_occurrence = position;
+            while let Some(docset) = docsets.next() {
+                if docset.doc() != pivot_doc {
+                    break;
+                } else {
+                    position += 1;
+                }
+            }
+            return Some(Pivot {
+                position,
+                doc: pivot_doc,
+                first_occurrence,
+            });
+        }
+        position += 1;
+    }
+    None
+}
+
+/// Given an iterator over all ordered lists up to the pivot (inclusive) and the following list (if
+/// exists), it returns the next document ID that can be possibly relevant, based on the block max
+/// scores.
+fn find_next_relevant_doc<T, TScorer>(
+    docsets_up_to_pivot: &mut [T],
+    pivot_docset: &mut T,
+    docset_after_pivot: Option<&mut T>,
+) -> DocId
+    where
+        T: AsMut<TScorer>,
+        TScorer: BlockMaxScorer + Scorer,
+{
+    let pivot_docset = pivot_docset.as_mut();
+    let mut next_doc = 1 + docsets_up_to_pivot
+        .iter_mut()
+        .map(|docset| docset.as_mut().block_max_doc())
+        .chain(std::iter::once(pivot_docset.block_max_doc()))
+        .min()
+        .unwrap();
+    if let Some(docset) = docset_after_pivot {
+        let doc = docset.as_mut().doc();
+        if doc < next_doc {
+            next_doc = doc;
+        }
+    }
+    if next_doc <= pivot_docset.doc() {
+        pivot_docset.doc() + 1
+    } else {
+        next_doc
+    }
+}
+
+/// Sifts down the first element of the slice.
+///
+/// `docsets[1..]` are assumed sorted.
+/// This function swaps `docsets[0]` with its right
+/// neighbor successively -bubble sort style- until it reaches the first
+/// position such that `docsets` is sorted.
+fn sift_down<T, TScorer>(docsets: &mut [T])
+    where
+        T: AsRef<TScorer>,
+        TScorer: BlockMaxScorer + Scorer,
+{
+    for idx in 1..docsets.len() {
+        if docsets[idx].as_ref().doc() >= docsets[idx - 1].as_ref().doc() {
+            return;
+        }
+        docsets.swap(idx, idx - 1);
+    }
+}
+
+/// Creates a `DocSet` that iterates through the union of two or more `DocSet`s,
+/// applying [BlockMaxWand] dynamic pruning.
+///
+/// [BlockMaxWand]: https://dl.acm.org/doi/10.1145/2009916.2010048
+pub struct BlockMaxWand<TScorer, TScoreCombiner> {
+    docsets: Vec<Box<TScorer>>,
+    doc: DocId,
+    score: Score,
+    combiner: TScoreCombiner,
+}
+
+impl<TScorer, TScoreCombiner> BlockMaxWand<TScorer, TScoreCombiner>
+    where
+        TScoreCombiner: ScoreCombiner,
+        TScorer: BlockMaxScorer + Scorer,
+{
+    fn new(
+        docsets: Vec<TScorer>,
+        combiner: TScoreCombiner,
+    ) -> BlockMaxWand<TScorer, TScoreCombiner> {
+        let mut non_empty_docsets: Vec<_> = docsets
+            .into_iter()
+            .flat_map(|mut docset| {
+                if docset.advance() {
+                    Some(Box::new(docset))
+                } else {
+                    None
+                }
+            })
+            .collect();
+        non_empty_docsets.sort_by_key(Box::<TScorer>::doc);
+        BlockMaxWand {
+            docsets: non_empty_docsets,
+            combiner,
+            doc: 0u32,
+            score: 0f32
+        }
+    }
+
+    /// Find the position in the sorted list of posting lists of the **pivot**.
+    fn find_pivot_position(&self, lower_bound_score: Score) -> Option<Pivot> {
+        find_pivot_position(
+            self.docsets.iter().map(|docset| docset.as_ref()),
+            lower_bound_score)
+    }
+
+    fn advance_with_pivot(&mut self, pivot: Pivot, lower_bound_score: Score) -> SkipResult {
+        let block_upper_bound: Score = self.docsets[..=pivot.position]
+            .iter_mut()
+            .map(|docset| docset.block_max_score())
+            .sum();
+        if block_upper_bound > lower_bound_score {
+            if pivot.doc == self.docsets[0].doc() {
+                // Since self.docsets is sorted by their current doc, in this branch, all
+                // docsets in [0..=pivot] are positioned on pivot.doc.
+                //
+                // Lets compute the actual score for this doc.
+                //
+                // NOTE(elshize): One additional check needs to be done to improve performance:
+                // update block-wise bound while accumulating score with the actual score,
+                // and check each time if still above threshold.
+                self.combiner.clear();
+                for idx in (0..=pivot.position).rev() {
+                    self.combiner.update(self.docsets[idx].as_mut());
+                    if !self.docsets[idx].advance() {
+                        self.docsets.swap_remove(idx);
+                    }
+                }
+                self.score = self.combiner.score();
+                self.doc = pivot.doc;
+                self.docsets.sort_by_key(Box::<TScorer>::doc);
+                SkipResult::Reached
+            } else {
+                // The substraction does not underflow because otherwise we would go to the other
+                // branch.
+                //
+                // `advanced_idx` is the last idx that is not positionned on the pivot yet.
+                let advanced_idx = pivot.first_occurrence - 1;
+                if !self.docsets[advanced_idx].advance() {
+                    self.docsets.swap_remove(advanced_idx);
+                }
+                if self.docsets.is_empty() {
+                    return SkipResult::End;
+                }
+                sift_down(&mut self.docsets[advanced_idx..]);
+                SkipResult::OverStep
+            }
+        } else {
+            let (up_to_pivot, pivot_and_rest) = self.docsets.split_at_mut(pivot.position as usize);
+            let (pivot, after_pivot) = pivot_and_rest.split_first_mut().unwrap();
+            let next_doc = find_next_relevant_doc(up_to_pivot, pivot, after_pivot.first_mut());
+            // NOTE(elshize): It might be more efficient to advance the list with the higher
+            // max score, but let's advance the first one for now for simplicity.
+            if self.docsets[0].skip_next(next_doc) == SkipResult::End {
+                self.docsets.swap_remove(0);
+            }
+            if self.docsets.is_empty() {
+                return SkipResult::End;
+            }
+            sift_down(&mut self.docsets[..]);
+            SkipResult::OverStep
+        }
+    }
+}
+
+impl<TScorer, TScoreCombiner> PruningScorer
+for BlockMaxWand<TScorer, TScoreCombiner>
+    where
+        TScoreCombiner: ScoreCombiner,
+        TScorer: Scorer + BlockMaxScorer,
+{
+    fn doc(&self) -> DocId {
+        self.doc
+    }
+
+    fn score(&self) -> Score {
+        self.score
+    }
+
+    fn advance_with_pruning(&mut self, lower_bound_score: f32) -> bool {
+        while let Some(pivot) = self.find_pivot_position(lower_bound_score) {
+            match self.advance_with_pivot(pivot, lower_bound_score) {
+                SkipResult::End => { return false },
+                SkipResult::Reached=> { return true; }
+                SkipResult::OverStep => {}
+            }
+        }
+        false
+    }
+}
+
+
+#[cfg(test)]
+mod tests {
+
+    use super::*;
+    use crate::common::HasLen;
+    use crate::docset::DocSet;
+    use crate::query::score_combiner::SumCombiner;
+    use crate::query::Union;
+    use crate::query::{BlockMaxScorer, Scorer};
+    use crate::{DocId, Score};
+    use float_cmp::approx_eq;
+    use proptest::strategy::Strategy;
+    use std::cmp::Ordering;
+    use std::num::Wrapping;
+    use crate::collector::{SegmentCollector, TopScoreSegmentCollector};
+
+    /*
+    #[derive(Debug, Clone)]
+    pub struct VecDocSet {
+        postings: Vec<(DocId, Score)>,
+        cursor: Wrapping<usize>,
+        block_max_scores: Vec<(DocId, Score)>,
+        max_score: Score,
+        block_size: usize,
+    }
+
+    impl VecDocSet {
+        fn new(postings: Vec<(DocId, Score)>, block_size: usize) -> VecDocSet {
+            let block_max_scores: Vec<(DocId, f32)> = postings
+                .chunks(block_size)
+                .into_iter()
+                .map(|block| {
+                    (
+                        block.iter().last().unwrap().0,
+                        block
+                            .iter()
+                            .map(|(_, s)| *s)
+                            .fold(-f32::INFINITY, |left, right| left.max(right))
+                    )
+                })
+                .collect();
+            let max_score = block_max_scores
+                .iter()
+                .copied()
+                .map(|(_, s)| s)
+                .fold(-f32::INFINITY, |left, right| left.max(right));
+            VecDocSet {
+                postings,
+                cursor: Wrapping(0_usize) - Wrapping(1_usize),
+                block_max_scores,
+                max_score,
+                block_size,
+            }
+        }
+        /// Constructs a new set and advances it.
+        fn started(postings: Vec<(DocId, Score)>, block_size: usize) -> VecDocSet {
+            let mut docset = VecDocSet::new(postings, block_size);
+            docset.advance();
+            docset
+        }
+    }
+
+    impl DocSet for VecDocSet {
+        fn advance(&mut self) -> bool {
+            self.cursor += Wrapping(1);
+            self.postings.len() > self.cursor.0
+        }
+
+        fn doc(&self) -> DocId {
+            self.postings[self.cursor.0].0
+        }
+
+        fn size_hint(&self) -> u32 {
+            self.len() as u32
+        }
+    }
+
+    impl HasLen for VecDocSet {
+        fn len(&self) -> usize {
+            self.postings.len()
+        }
+    }
+
+    impl BlockMaxScorer for VecDocSet {
+        fn max_score(&self) -> Score {
+            self.max_score
+        }
+        fn block_max_score(&mut self) -> Score {
+            self.block_max_scores[self.cursor.0 / self.block_size].1
+        }
+        fn block_max_doc(&mut self) -> DocId {
+            self.block_max_scores[self.cursor.0 / self.block_size].0
+        }
+    }
+
+    impl Scorer for VecDocSet {
+        fn score(&mut self) -> Score {
+            self.postings[self.cursor.0].1
+        }
+    }
+
+    #[derive(Debug)]
+    struct ComparableDoc<T, D> {
+        feature: T,
+        doc: D,
+    }
+
+    impl<T: PartialOrd, D: PartialOrd> PartialOrd for ComparableDoc<T, D> {
+        fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
+            Some(self.cmp(other))
+        }
+    }
+
+    impl<T: PartialOrd, D: PartialOrd> Ord for ComparableDoc<T, D> {
+        #[inline]
+        fn cmp(&self, other: &Self) -> Ordering {
+            // Reversed to make BinaryHeap work as a min-heap
+            let by_feature = other
+                .feature
+                .partial_cmp(&self.feature)
+                .unwrap_or(Ordering::Equal);
+
+            let lazy_by_doc_address =
+                || self.doc.partial_cmp(&other.doc).unwrap_or(Ordering::Equal);
+
+            // In case of a tie on the feature, we sort by ascending
+            // `DocAddress` in order to ensure a stable sorting of the
+            // documents.
+            by_feature.then_with(lazy_by_doc_address)
+        }
+    }
+
+    impl<T: PartialOrd, D: PartialOrd> PartialEq for ComparableDoc<T, D> {
+        fn eq(&self, other: &Self) -> bool {
+            self.cmp(other) == Ordering::Equal
+        }
+    }
+
+    impl<T: PartialOrd, D: PartialOrd> Eq for ComparableDoc<T, D> {}
+
+    fn union_vs_bmw(posting_lists: Vec<VecDocSet>) {
+        let mut union = Union::<VecDocSet, SumCombiner>::from(posting_lists.clone());
+        let mut top_union = TopScoreSegmentCollector::new(0, 10);
+        while union.advance() {
+            top_union.collect(union.doc(), union.score());
+        }
+        let top_bmw = TopScoreSegmentCollector::new(0, 10 );
+        let mut bmw = BlockMaxWand::new(posting_lists, SumCombiner::default());
+        let top_docs_bnw = top_bmw.collect_scorer(&mut bmw, None);
+        for ((expected_score, expected_doc), (actual_score, actual_doc)) in
+        top_union.harvest().into_iter().zip( top_docs_bnw )
+        {
+            assert!(approx_eq!(
+                f32,
+                expected_score,
+                actual_score,
+                epsilon = 0.0001
+            ));
+            assert_eq!(expected_doc, actual_doc);
+        }
+    }
+
+    #[test]
+    fn test_bmw_0() {
+        union_vs_bmw(vec![
+            VecDocSet {
+                postings: vec![
+                    (0, 1.0),
+                    (23, 1.0),
+                    (28, 1.0),
+                    (56, 1.0),
+                    (59, 1.0),
+                    (66, 1.0),
+                    (93, 1.0),
+                ],
+                cursor: Wrapping(0_usize) - Wrapping(1_usize),
+                block_max_scores: vec![(93, 1.0)],
+                max_score: 1.0,
+                block_size: 16,
+            },
+            VecDocSet {
+                postings: vec![
+                    (2, 1.6549665),
+                    (43, 2.6958032),
+                    (53, 3.5309567),
+                    (71, 2.7688136),
+                    (87, 3.4279852),
+                    (96, 3.9028034),
+                ],
+                cursor: Wrapping(0_usize) - Wrapping(1_usize),
+                block_max_scores: vec![(96, 3.9028034)],
+                max_score: 3.9028034,
+                block_size: 16,
+            },
+        ])
+    }
+
+    #[test]
+    fn test_bmw_1() {
+        union_vs_bmw(vec![
+            VecDocSet {
+                postings: vec![(73, 1.0), (82, 1.0)],
+                cursor: Wrapping(0_usize) - Wrapping(1_usize),
+                block_max_scores: vec![(82, 1.0)],
+                max_score: 1.0,
+                block_size: 16,
+            },
+            VecDocSet {
+                postings: vec![
+                    (21, 3.582513),
+                    (23, 1.6928024),
+                    (27, 3.887647),
+                    (42, 1.5469292),
+                    (61, 1.7317574),
+                    (62, 1.2968783),
+                    (82, 2.4040694),
+                    (85, 3.1487892),
+                ],
+                cursor: Wrapping(0_usize) - Wrapping(1_usize),
+                block_max_scores: vec![(85, 3.887647)],
+                max_score: 3.887647,
+                block_size: 16,
+            },
+        ])
+    }
+
+    proptest::proptest! {
+        #[test]
+        fn test_union_vs_bmw(postings in proptest::collection::vec(
+            proptest::collection::vec(0_u32..100, 1..10)
+                .prop_flat_map(|v| {
+                    let scores = proptest::collection::vec(1_f32..4_f32, v.len()..=v.len());
+                    scores.prop_map(move |s| {
+                        let mut postings: Vec<_> = v.iter().copied().zip(s.iter().copied()).collect();
+                        postings.sort_by_key(|p| p.0);
+                        postings.dedup_by_key(|p| p.0);
+                        VecDocSet::new(postings, 16)
+                    })
+                }),
+            2..5)
+        ) {
+            union_vs_bmw(postings);
+        }
+    }
+
+    #[test]
+    fn test_find_pivot_position() {
+        let postings = vec![
+            VecDocSet::started(vec![(0, 2.0)], 1),
+            VecDocSet::started(vec![(1, 3.0)], 1),
+            VecDocSet::started(vec![(2, 4.0)], 1),
+            VecDocSet::started(vec![(3, 5.0)], 1),
+            VecDocSet::started(vec![(3, 6.0)], 1),
+        ];
+        assert_eq!(
+            find_pivot_position(postings.iter(), 2.0f32),
+            Some(Pivot {
+                position: 1,
+                doc: 1,
+                first_occurrence: 1,
+            })
+        );
+        assert_eq!(
+            find_pivot_position(postings.iter(), 5.0f32),
+            Some(Pivot {
+                position: 2,
+                doc: 2,
+                first_occurrence: 2,
+            })
+        );
+        assert_eq!(
+            find_pivot_position(postings.iter(), 9.0f32),
+            Some(Pivot {
+                position: 4,
+                doc: 3,
+                first_occurrence: 3,
+            })
+        );
+        assert_eq!(
+            find_pivot_position(postings.iter(), 20.0f32),
+            None
+        );
+    }
+
+    #[test]
+    fn test_find_next_relevant_doc_before_pivot() {
+        let mut postings = vec![
+            Box::new(VecDocSet::started(vec![(0, 0.0), (3, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(1, 0.0), (4, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(2, 0.0), (6, 0.0)], 2)), // pivot
+            Box::new(VecDocSet::started(vec![(6, 0.0), (8, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(6, 0.0), (8, 0.0)], 2)),
+        ];
+        let (up_to_pivot, rest) = postings.split_at_mut(2);
+        let (pivot, after_pivot) = rest.split_first_mut().unwrap();
+        let next_doc = find_next_relevant_doc(up_to_pivot, pivot, Some(&mut after_pivot[0]));
+        assert_eq!(next_doc, 4);
+    }
+
+    #[test]
+    fn test_find_next_relevant_doc_prefix_smaller_than_pivot() {
+        let mut postings = vec![
+            Box::new(VecDocSet::started(vec![(0, 0.0), (3, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(1, 0.0), (4, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(5, 0.0), (8, 0.0)], 2)), // pivot
+            Box::new(VecDocSet::started(vec![(6, 0.0), (8, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(6, 0.0), (8, 0.0)], 2)),
+        ];
+        let (up_to_pivot, rest) = postings.split_at_mut(2);
+        let (pivot, after_pivot) = rest.split_first_mut().unwrap();
+        let next_doc = find_next_relevant_doc(up_to_pivot, pivot, Some(&mut after_pivot[0]));
+        assert_eq!(next_doc, 6);
+    }
+
+    #[test]
+    fn test_find_next_relevant_doc_after_pivot() {
+        let mut postings = vec![
+            Box::new(VecDocSet::started(vec![(0, 0.0), (8, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(1, 0.0), (8, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(2, 0.0), (8, 0.0)], 2)), // pivot
+            Box::new(VecDocSet::started(vec![(5, 0.0), (7, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(6, 0.0), (7, 0.0)], 2)),
+        ];
+        let (up_to_pivot, rest) = postings.split_at_mut(2);
+        let (pivot, after_pivot) = rest.split_first_mut().unwrap();
+        let next_doc = find_next_relevant_doc(up_to_pivot, pivot, Some(&mut after_pivot[0]));
+        assert_eq!(next_doc, 5);
+    }
+
+    #[test]
+    fn test_sift_down_already_sifted() {
+        let mut postings = vec![
+            Box::new(VecDocSet::started(vec![(0, 0.0), (8, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(1, 0.0), (8, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(2, 0.0), (8, 0.0)], 2)), // pivot
+            Box::new(VecDocSet::started(vec![(5, 0.0), (7, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(6, 0.0), (7, 0.0)], 2)),
+        ];
+        sift_down(&mut postings[2..]);
+        assert_eq!(
+            postings.into_iter().map(|p| p.doc()).collect::<Vec<_>>(),
+            vec![0, 1, 2, 5, 6]
+        );
+    }
+
+    #[test]
+    fn test_sift_down_sift_one_down() {
+        let mut postings = vec![
+            Box::new(VecDocSet::started(vec![(0, 0.0), (8, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(1, 0.0), (8, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(6, 0.0), (8, 0.0)], 2)), // pivot
+            Box::new(VecDocSet::started(vec![(5, 0.0), (7, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(7, 0.0), (7, 0.0)], 2)),
+        ];
+        sift_down(&mut postings[2..]);
+        assert_eq!(
+            postings.into_iter().map(|p| p.doc()).collect::<Vec<_>>(),
+            vec![0, 1, 5, 6, 7]
+        );
+    }
+
+    #[test]
+    fn test_sift_down_to_bottom() {
+        let mut postings = vec![
+            Box::new(VecDocSet::started(vec![(0, 0.0), (8, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(1, 0.0), (8, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(7, 0.0), (8, 0.0)], 2)), // pivot
+            Box::new(VecDocSet::started(vec![(5, 0.0), (7, 0.0)], 2)),
+            Box::new(VecDocSet::started(vec![(6, 0.0), (7, 0.0)], 2)),
+        ];
+        sift_down(&mut postings[2..]);
+        assert_eq!(
+            postings.into_iter().map(|p| p.doc()).collect::<Vec<_>>(),
+            vec![0, 1, 5, 6, 7]
+        );
+    }
+    
+     */
+}

src/query/boolean_query/mod.rs

@@ -31,24 +31,11 @@ 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());
         }

src/query/explanation.rs

@@ -1,4 +1,5 @@
 use crate::{DocId, TantivyError};
+use serde::Serialize;
 
 pub(crate) fn does_not_match(doc: DocId) -> TantivyError {
     TantivyError::InvalidArgument(format!("Document #({}) does not match", doc))

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
@@ -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",
@@ -166,5 +199,17 @@ mod test {
             let (score, _) = top_docs[0];
             assert_nearly_equals(1f32, score);
         }
+
+        {
+            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(1f32, score);
+        }
     }
 }

src/query/mod.rs

@@ -1,10 +1,10 @@
-/*!
-Query
-*/
+/*! Query Module */
 
 mod all_query;
 mod automaton_weight;
 mod bitset;
+mod block_max_scorer;
+mod block_max_wand;
 mod bm25;
 mod boolean_query;
 mod boost_query;
@@ -37,11 +37,14 @@ pub use self::vec_docset::VecDocSet;
 pub use self::all_query::{AllQuery, AllScorer, AllWeight};
 pub use self::automaton_weight::AutomatonWeight;
 pub use self::bitset::BitSetDocSet;
+pub use self::block_max_scorer::BlockMaxScorer;
 pub use self::boolean_query::BooleanQuery;
 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;
@@ -56,6 +59,8 @@ pub use self::scorer::Scorer;
 pub use self::term_query::TermQuery;
 pub use self::weight::Weight;
 pub use tantivy_query_grammar::Occur;
+pub use self::weight::PruningScorerIfPossible;
+
 
 #[cfg(test)]
 mod tests {

src/query/phrase_query/mod.rs

@@ -12,7 +12,6 @@ pub mod tests {
     use super::*;
     use crate::collector::tests::{TEST_COLLECTOR_WITHOUT_SCORE, TEST_COLLECTOR_WITH_SCORE};
     use crate::core::Index;
-    use crate::error::TantivyError;
     use crate::schema::{Schema, Term, TEXT};
     use crate::tests::assert_nearly_equals;
     use crate::DocAddress;
@@ -127,21 +126,16 @@ 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]

src/query/query_parser/query_parser.rs

@@ -174,6 +174,16 @@ pub struct QueryParser {
     boost: HashMap<Field, f32>,
 }
 
+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 {
     /// Creates a `QueryParser`, given
     /// * schema - index Schema
@@ -253,8 +263,13 @@ 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)
@@ -410,31 +425,23 @@ 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))
             }
+            UserInputAST::Leaf(leaf) => self.compute_logical_ast_from_leaf(*leaf),
         }
     }
 
@@ -782,6 +789,20 @@ 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 +822,13 @@ 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 +988,18 @@ 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 +1019,6 @@ 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 +1042,26 @@ 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/term_query/block_max_term_scorer.rs

@@ -0,0 +1,98 @@
+use crate::docset::{DocSet, SkipResult};
+use crate::query::{Explanation, Scorer};
+use crate::DocId;
+use crate::Score;
+
+use crate::fieldnorm::FieldNormReader;
+use crate::postings::Postings;
+use crate::postings::{BlockMaxPostings, BlockMaxSegmentPostings};
+use crate::query::bm25::BM25Weight;
+use crate::query::BlockMaxScorer;
+
+pub struct BlockMaxTermScorer {
+    postings: BlockMaxSegmentPostings,
+    fieldnorm_reader: FieldNormReader,
+    similarity_weight: BM25Weight,
+}
+
+impl BlockMaxTermScorer {
+    pub fn new(
+        postings: BlockMaxSegmentPostings,
+        fieldnorm_reader: FieldNormReader,
+        similarity_weight: BM25Weight,
+    ) -> Self {
+        Self {
+            postings,
+            fieldnorm_reader,
+            similarity_weight,
+        }
+    }
+}
+
+impl BlockMaxTermScorer {
+    fn _score(&self, fieldnorm_id: u8, term_freq: u32) -> Score {
+        self.similarity_weight.score(fieldnorm_id, term_freq)
+    }
+
+    pub fn term_freq(&self) -> u32 {
+        self.postings.term_freq()
+    }
+
+    pub fn fieldnorm_id(&self) -> u8 {
+        self.fieldnorm_reader.fieldnorm_id(self.doc())
+    }
+
+    pub fn explain(&self) -> Explanation {
+        let fieldnorm_id = self.fieldnorm_id();
+        let term_freq = self.term_freq();
+        self.similarity_weight.explain(fieldnorm_id, term_freq)
+    }
+}
+
+impl DocSet for BlockMaxTermScorer {
+    fn advance(&mut self) -> bool {
+        self.postings.advance()
+    }
+
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        self.postings.skip_next(target)
+    }
+
+    fn doc(&self) -> DocId {
+        self.postings.doc()
+    }
+
+    fn size_hint(&self) -> u32 {
+        self.postings.size_hint()
+    }
+}
+
+impl Scorer for BlockMaxTermScorer {
+    fn score(&mut self) -> Score {
+        self._score(
+            self.fieldnorm_reader.fieldnorm_id(self.doc()),
+            self.postings.term_freq(),
+        )
+    }
+}
+
+impl BlockMaxScorer for BlockMaxTermScorer {
+    fn block_max_score(&mut self) -> Score {
+        self._score(
+            self.fieldnorm_reader
+                .fieldnorm_id(self.postings.block_max_doc()),
+            self.postings.term_freq(),
+        )
+    }
+
+    fn block_max_doc(&mut self) -> DocId {
+        self.postings.block_max_doc()
+    }
+
+    fn max_score(&self) -> Score {
+        self._score(
+            self.fieldnorm_reader.fieldnorm_id(self.postings.max_doc()),
+            self.postings.max_term_freq(),
+        )
+    }
+}

src/query/term_query/mod.rs

@@ -1,7 +1,9 @@
+mod block_max_term_scorer;
 mod term_query;
 mod term_scorer;
 mod term_weight;
 
+pub use self::block_max_term_scorer::BlockMaxTermScorer;
 pub use self::term_query::TermQuery;
 pub use self::term_scorer::TermScorer;
 pub use self::term_weight::TermWeight;
@@ -112,7 +114,6 @@ mod tests {
         let term_a = Term::from_field_text(text_field, "a");
         let term_query = TermQuery::new(term_a, IndexRecordOption::Basic);
         let reader = index.reader().unwrap();
-        assert_eq!(reader.searcher().segment_readers().len(), 1);
         assert_eq!(term_query.count(&*reader.searcher()).unwrap(), 1);
     }
 

src/query/union.rs

@@ -1,7 +1,7 @@
 use crate::common::TinySet;
 use crate::docset::{DocSet, SkipResult};
 use crate::query::score_combiner::{DoNothingCombiner, ScoreCombiner};
-use crate::query::Scorer;
+use crate::query::{Scorer, BlockMaxScorer};
 use crate::DocId;
 use crate::Score;
 use std::cmp::Ordering;
@@ -9,6 +9,99 @@ use std::cmp::Ordering;
 const HORIZON_NUM_TINYBITSETS: usize = 64;
 const HORIZON: u32 = 64u32 * HORIZON_NUM_TINYBITSETS as u32;
 
+
+#[derive(Debug, Copy, Clone, PartialEq, Eq)]
+struct Pivot {
+    position: usize,
+    first_occurrence: usize,
+    doc: DocId,
+}
+
+/// Find the position in the sorted list of posting lists of the **pivot**.
+///
+/// docsets need to be advanced, and are required to be sorted by the doc they point to.
+///
+/// The pivot is then defined as the lowest DocId that has a chance of matching our condition.
+fn find_pivot_position<'a, TScorer>(
+    mut docsets: impl Iterator<Item = &'a TScorer>,
+    lower_bound_score: Score,
+) -> Option<Pivot>
+    where TScorer: BlockMaxScorer
+{
+    let mut position = 0;
+    let mut upper_bound = Score::default();
+    while let Some(docset) = docsets.next() {
+        upper_bound += docset.max_score();
+        if lower_bound_score < upper_bound {
+            let pivot_doc = docset.doc();
+            let first_occurrence = position;
+            while let Some(docset) = docsets.next() {
+                if docset.doc() != pivot_doc {
+                    break;
+                } else {
+                    position += 1;
+                }
+            }
+            return Some(Pivot {
+                position,
+                doc: pivot_doc,
+                first_occurrence,
+            });
+        }
+        position += 1;
+    }
+    None
+}
+
+/// Sifts down the first element of the slice.
+///
+/// `docsets[1..]` are assumed sorted.
+/// This function swaps `docsets[0]` with its right
+/// neighbor successively -bubble sort style- until it reaches the first
+/// position such that `docsets` is sorted.
+fn sift_down<TScorer>(docsets: &mut [TScorer])
+    where
+        TScorer: BlockMaxScorer + Scorer,
+{
+    for idx in 1..docsets.len() {
+        if docsets[idx].doc() >= docsets[idx - 1].doc() {
+            return;
+        }
+        docsets.swap(idx, idx - 1);
+    }
+}
+
+
+/// Given an iterator over all ordered lists up to the pivot (inclusive) and the following list (if
+/// exists), it returns the next document ID that can be possibly relevant, based on the block max
+/// scores.
+fn find_next_relevant_doc<TScorer>(
+    docsets_up_to_pivot: &mut [TScorer],
+    pivot_docset: &mut TScorer,
+    docset_after_pivot: Option<&mut TScorer>,
+) -> DocId
+    where
+        TScorer: BlockMaxScorer + Scorer,
+{
+    let mut next_doc = 1 + docsets_up_to_pivot
+        .iter_mut()
+        .map(|docset| docset.block_max_doc())
+        .chain(std::iter::once(pivot_docset.block_max_doc()))
+        .min()
+        .unwrap();
+    if let Some(docset) = docset_after_pivot {
+        let doc = docset.doc();
+        if doc < next_doc {
+            next_doc = doc;
+        }
+    }
+    if next_doc <= pivot_docset.doc() {
+        pivot_docset.doc() + 1
+    } else {
+        next_doc
+    }
+}
+
 // `drain_filter` is not stable yet.
 // This function is similar except that it does is not unstable, and
 // it does not keep the original vector ordering.
@@ -39,6 +132,7 @@ pub struct Union<TScorer, TScoreCombiner = DoNothingCombiner> {
     score: Score,
 }
 
+
 impl<TScorer, TScoreCombiner> From<Vec<TScorer>> for Union<TScorer, TScoreCombiner>
 where
     TScoreCombiner: ScoreCombiner,
@@ -126,6 +220,76 @@ impl<TScorer: Scorer, TScoreCombiner: ScoreCombiner> Union<TScorer, TScoreCombin
         }
         false
     }
+
+
+}
+
+
+impl<TScorer: BlockMaxScorer, TScoreCombiner: ScoreCombiner> Union<TScorer, TScoreCombiner> {
+    fn advance_with_pivot(&mut self, pivot: Pivot, lower_bound_score: Score) -> SkipResult {
+        let block_upper_bound: Score = self.docsets[..=pivot.position]
+            .iter_mut()
+            .map(|docset| docset.block_max_score())
+            .sum();
+        if block_upper_bound > lower_bound_score {
+            if pivot.doc == self.docsets[0].doc() {
+                // Since self.docsets is sorted by their current doc, in this branch, all
+                // docsets in [0..=pivot] are positioned on pivot.doc.
+                //
+                // Lets compute the actual score for this doc.
+                //
+                // NOTE(elshize): One additional check needs to be done to improve performance:
+                // update block-wise bound while accumulating score with the actual score,
+                // and check each time if still above threshold.
+                let mut combiner = TScoreCombiner::default();
+                for idx in (0..=pivot.position).rev() {
+                    combiner.update(&mut self.docsets[idx]);
+                    if !self.docsets[idx].advance() {
+                        self.docsets.swap_remove(idx);
+                    }
+                }
+                self.score = combiner.score();
+                self.doc = pivot.doc;
+                self.docsets.sort_by_key(TScorer::doc);
+                SkipResult::Reached
+            } else {
+                // The substraction does not underflow because otherwise we would go to the other
+                // branch.
+                //
+                // `advanced_idx` is the last idx that is not positionned on the pivot yet.
+                let advanced_idx = pivot.first_occurrence - 1;
+                if !self.docsets[advanced_idx].advance() {
+                    self.docsets.swap_remove(advanced_idx);
+                }
+                if self.docsets.is_empty() {
+                    return SkipResult::End;
+                }
+                sift_down(&mut self.docsets[advanced_idx..]);
+                SkipResult::OverStep
+            }
+        } else {
+            let (up_to_pivot, pivot_and_rest) = self.docsets.split_at_mut(pivot.position as usize);
+            let (pivot, after_pivot) = pivot_and_rest.split_first_mut().unwrap();
+            let next_doc = find_next_relevant_doc(up_to_pivot, pivot, after_pivot.first_mut());
+            // NOTE(elshize): It might be more efficient to advance the list with the higher
+            // max score, but let's advance the first one for now for simplicity.
+            if self.docsets[0].skip_next(next_doc) == SkipResult::End {
+                self.docsets.swap_remove(0);
+            }
+            if self.docsets.is_empty() {
+                return SkipResult::End;
+            }
+            sift_down(&mut self.docsets[..]);
+            SkipResult::OverStep
+        }
+    }
+
+    /// Find the position in the sorted list of posting lists of the **pivot**.
+    fn find_pivot_position(&self, lower_bound_score: Score) -> Option<Pivot> {
+        find_pivot_position(
+            self.docsets.iter().map(|docset| docset),
+            lower_bound_score)
+    }
 }
 
 impl<TScorer, TScoreCombiner> DocSet for Union<TScorer, TScoreCombiner>

src/query/weight.rs

@@ -1,7 +1,26 @@
 use super::Scorer;
 use crate::core::SegmentReader;
 use crate::query::Explanation;
-use crate::DocId;
+use crate::{DocId, Score};
+
+pub trait PruningScorer {
+
+    fn doc(&self) -> DocId;
+
+    fn score(&self) -> Score;
+
+    /// Advance to the next document that has a score strictly greater than
+    /// `lower_bound_score`.
+    fn advance_with_pruning(&mut self, score_lower_bound: f32) -> bool;
+    fn advance(&mut self) -> bool {
+        self.advance_with_pruning(std::f32::NEG_INFINITY)
+    }
+}
+
+pub enum PruningScorerIfPossible {
+    Pruning(Box<dyn PruningScorer>),
+    NonPruning(Box<dyn Scorer>)
+}
 
 /// A Weight is the specialization of a Query
 /// for a given set of segments.
@@ -15,6 +34,11 @@ pub trait Weight: Send + Sync + 'static {
     /// See [`Query`](./trait.Query.html).
     fn scorer(&self, reader: &SegmentReader, boost: f32) -> crate::Result<Box<dyn Scorer>>;
 
+    fn pruning_scorer(&self, reader: &SegmentReader, boost: f32) -> crate::Result<PruningScorerIfPossible> {
+        let scorer = self.scorer(reader, boost)?;
+        Ok(PruningScorerIfPossible::NonPruning(Box::new(scorer)))
+    }
+
     /// Returns an `Explanation` for the given document.
     fn explain(&self, reader: &SegmentReader, doc: DocId) -> crate::Result<Explanation>;
 

src/reader/index_writer_reader.rs

@@ -1,83 +0,0 @@
-use crate::directory::{WatchCallbackList, WatchHandle};
-use crate::indexer::SegmentRegisters;
-use crate::reader::pool::Pool;
-use crate::{Index, LeasedItem, Searcher, Segment, SegmentReader};
-use std::iter::repeat_with;
-use std::sync::{Arc, RwLock, Weak};
-
-struct InnerNRTReader {
-    num_searchers: usize,
-    index: Index,
-    searcher_pool: Pool<Searcher>,
-    segment_registers: Arc<RwLock<SegmentRegisters>>,
-}
-
-impl InnerNRTReader {
-    fn load_segment_readers(&self) -> crate::Result<Vec<SegmentReader>> {
-        let segments: Vec<Segment> = {
-            let rlock = self.segment_registers.read().unwrap();
-            rlock.committed_segment()
-        };
-        segments
-            .iter()
-            .map(SegmentReader::open)
-            .collect::<crate::Result<Vec<SegmentReader>>>()
-    }
-
-    pub fn reload(&self) -> crate::Result<()> {
-        let segment_readers: Vec<SegmentReader> = self.load_segment_readers()?;
-        let schema = self.index.schema();
-        let searchers = repeat_with(|| {
-            Searcher::new(schema.clone(), self.index.clone(), segment_readers.clone())
-        })
-        .take(self.num_searchers)
-        .collect();
-        self.searcher_pool.publish_new_generation(searchers);
-        Ok(())
-    }
-
-    pub fn searcher(&self) -> LeasedItem<Searcher> {
-        self.searcher_pool.acquire()
-    }
-}
-
-#[derive(Clone)]
-pub struct NRTReader {
-    inner: Arc<InnerNRTReader>,
-    watch_handle: WatchHandle,
-}
-
-impl NRTReader {
-    pub fn reload(&self) -> crate::Result<()> {
-        self.inner.reload()
-    }
-
-    pub fn searcher(&self) -> LeasedItem<Searcher> {
-        self.inner.searcher()
-    }
-
-    pub(crate) fn create(
-        num_searchers: usize,
-        index: Index,
-        segment_registers: Arc<RwLock<SegmentRegisters>>,
-        watch_callback_list: &WatchCallbackList,
-    ) -> crate::Result<Self> {
-        let inner_reader: Arc<InnerNRTReader> = Arc::new(InnerNRTReader {
-            num_searchers,
-            index,
-            searcher_pool: Pool::new(),
-            segment_registers,
-        });
-        let inner_reader_weak: Weak<InnerNRTReader> = Arc::downgrade(&inner_reader);
-        let watch_handle = watch_callback_list.subscribe(Box::new(move || {
-            if let Some(nrt_reader_arc) = inner_reader_weak.upgrade() {
-                let _ = nrt_reader_arc.reload();
-            }
-        }));
-        inner_reader.reload()?;
-        Ok(NRTReader {
-            inner: inner_reader,
-            watch_handle,
-        })
-    }
-}

src/reader/meta_file_reader.rs

@@ -1,180 +0,0 @@
-use super::pool::Pool;
-use crate::core::Segment;
-use crate::directory::Directory;
-use crate::directory::WatchHandle;
-use crate::directory::META_LOCK;
-use crate::Searcher;
-use crate::SegmentReader;
-use crate::{Index, LeasedItem};
-use crate::{IndexReader, Result};
-use std::iter::repeat_with;
-use std::sync::Arc;
-
-/// Defines when a new version of the index should be reloaded.
-///
-/// Regardless of whether you search and index in the same process, tantivy does not necessarily
-/// reflects the change that are commited to your index. `ReloadPolicy` precisely helps you define
-/// when you want your index to be reloaded.
-#[derive(Clone, Copy)]
-pub enum ReloadPolicy {
-    /// The index is entirely reloaded manually.
-    /// All updates of the index should be manual.
-    ///
-    /// No change is reflected automatically. You are required to call `.load_seacher()` manually.
-    Manual,
-    /// The index is reloaded within milliseconds after a new commit is available.
-    /// This is made possible by watching changes in the `meta.json` file.
-    OnCommit, // TODO add NEAR_REAL_TIME(target_ms)
-}
-
-/// `IndexReader` builder
-///
-/// It makes it possible to set the following values.
-///
-/// - `num_searchers` (by default, the number of detected CPU threads):
-///
-///   When `num_searchers` queries are requested at the same time, the `num_searchers` will block
-///   until the one of the searcher in-use gets released.
-/// - `reload_policy` (by default `ReloadPolicy::OnCommit`):
-///
-///   See [`ReloadPolicy`](./enum.ReloadPolicy.html) for more details.
-#[derive(Clone)]
-pub struct IndexReaderBuilder {
-    num_searchers: usize,
-    reload_policy: ReloadPolicy,
-    index: Index,
-}
-
-impl IndexReaderBuilder {
-    pub(crate) fn new(index: Index) -> IndexReaderBuilder {
-        IndexReaderBuilder {
-            num_searchers: num_cpus::get(),
-            reload_policy: ReloadPolicy::Manual,
-            index,
-        }
-    }
-
-    /// Builds the reader.
-    ///
-    /// Building the reader is a non-trivial operation that requires
-    /// to open different segment readers. It may take hundreds of milliseconds
-    /// of time and it may return an error.
-    /// TODO(pmasurel) Use the `TryInto` trait once it is available in stable.
-    pub fn try_into(self) -> Result<IndexReader> {
-        let inner_reader = MetaFileIndexReaderInner {
-            index: self.index,
-            num_searchers: self.num_searchers,
-            searcher_pool: Pool::new(),
-        };
-        inner_reader.reload()?;
-        let inner_reader_arc = Arc::new(inner_reader);
-        let watch_handle_opt: Option<WatchHandle>;
-        match self.reload_policy {
-            ReloadPolicy::Manual => {
-                // No need to set anything...
-                watch_handle_opt = None;
-            }
-            ReloadPolicy::OnCommit => {
-                let inner_reader_arc_clone = inner_reader_arc.clone();
-                let callback = move || {
-                    if let Err(err) = inner_reader_arc_clone.reload() {
-                        error!(
-                            "Error while loading searcher after commit was detected. {:?}",
-                            err
-                        );
-                    }
-                };
-                let watch_handle = inner_reader_arc
-                    .index
-                    .directory()
-                    .watch(Box::new(callback))?;
-                watch_handle_opt = Some(watch_handle);
-            }
-        }
-        Ok(IndexReader::from(MetaFileIndexReader {
-            inner: inner_reader_arc,
-            watch_handle_opt,
-        }))
-    }
-
-    /// Sets the reload_policy.
-    ///
-    /// See [`ReloadPolicy`](./enum.ReloadPolicy.html) for more details.
-    pub fn reload_policy(mut self, reload_policy: ReloadPolicy) -> IndexReaderBuilder {
-        self.reload_policy = reload_policy;
-        self
-    }
-
-    /// Sets the number of `Searcher` in the searcher pool.
-    pub fn num_searchers(mut self, num_searchers: usize) -> IndexReaderBuilder {
-        self.num_searchers = num_searchers;
-        self
-    }
-}
-
-struct MetaFileIndexReaderInner {
-    num_searchers: usize,
-    searcher_pool: Pool<Searcher>,
-    index: Index,
-}
-
-impl MetaFileIndexReaderInner {
-    fn load_segment_readers(&self) -> crate::Result<Vec<SegmentReader>> {
-        // We keep the lock until we have effectively finished opening the
-        // the `SegmentReader` because it prevents a diffferent process
-        // to garbage collect these file while we open them.
-        //
-        // Once opened, on linux & mac, the mmap will remain valid after
-        // the file has been deleted
-        // On windows, the file deletion will fail.
-        let _meta_lock = self.index.directory().acquire_lock(&META_LOCK)?;
-        let searchable_segments = self.searchable_segments()?;
-        searchable_segments
-            .iter()
-            .map(SegmentReader::open)
-            .collect::<Result<_>>()
-    }
-
-    fn reload(&self) -> crate::Result<()> {
-        let segment_readers: Vec<SegmentReader> = self.load_segment_readers()?;
-        let schema = self.index.schema();
-        let searchers = repeat_with(|| {
-            Searcher::new(schema.clone(), self.index.clone(), segment_readers.clone())
-        })
-        .take(self.num_searchers)
-        .collect();
-        self.searcher_pool.publish_new_generation(searchers);
-        Ok(())
-    }
-
-    /// Returns the list of segments that are searchable
-    fn searchable_segments(&self) -> crate::Result<Vec<Segment>> {
-        self.index.searchable_segments()
-    }
-
-    fn searcher(&self) -> LeasedItem<Searcher> {
-        self.searcher_pool.acquire()
-    }
-}
-
-/// `IndexReader` is your entry point to read and search the index.
-///
-/// It controls when a new version of the index should be loaded and lends
-/// you instances of `Searcher` for the last loaded version.
-///
-/// `Clone` does not clone the different pool of searcher. `IndexReader`
-/// just wraps and `Arc`.
-#[derive(Clone)]
-pub struct MetaFileIndexReader {
-    inner: Arc<MetaFileIndexReaderInner>,
-    watch_handle_opt: Option<WatchHandle>,
-}
-
-impl MetaFileIndexReader {
-    pub fn reload(&self) -> crate::Result<()> {
-        self.inner.reload()
-    }
-    pub fn searcher(&self) -> LeasedItem<Searcher> {
-        self.inner.searcher()
-    }
-}

src/reader/mod.rs

@@ -1,32 +1,16 @@
-mod index_writer_reader;
-mod meta_file_reader;
 mod pool;
 
-use self::meta_file_reader::MetaFileIndexReader;
-pub use self::meta_file_reader::{IndexReaderBuilder, ReloadPolicy};
 pub use self::pool::LeasedItem;
-
-
-
-
-
-pub(crate) use crate::reader::index_writer_reader::NRTReader;
-
+use self::pool::Pool;
+use crate::core::Segment;
+use crate::directory::Directory;
+use crate::directory::WatchHandle;
+use crate::directory::META_LOCK;
+use crate::Index;
 use crate::Searcher;
-
-/*
-//
-//enum SegmentSource {
-//    FromMetaFile,
-//    FromWriter(Arc<RwLock<SegmentRegisters>>),
-//}
-//
-//impl SegmentSource {
-//    fn from_meta_file() -> SegmentSource {
-//
-//    }
-//
-//}
+use crate::SegmentReader;
+use std::convert::TryInto;
+use std::sync::Arc;
 
 /// Defines when a new version of the index should be reloaded.
 ///
@@ -77,7 +61,6 @@ impl IndexReaderBuilder {
     /// Building the reader is a non-trivial operation that requires
     /// to open different segment readers. It may take hundreds of milliseconds
     /// of time and it may return an error.
-    /// TODO(pmasurel) Use the `TryInto` trait once it is available in stable.
     pub fn try_into(self) -> crate::Result<IndexReader> {
         let inner_reader = InnerIndexReader {
             index: self.index,
@@ -130,6 +113,14 @@ impl IndexReaderBuilder {
     }
 }
 
+impl TryInto<IndexReader> for IndexReaderBuilder {
+    type Error = crate::TantivyError;
+
+    fn try_into(self) -> crate::Result<IndexReader> {
+        IndexReaderBuilder::try_into(self)
+    }
+}
+
 struct InnerIndexReader {
     num_searchers: usize,
     searcher_pool: Pool<Searcher>,
@@ -137,24 +128,15 @@ struct InnerIndexReader {
 }
 
 impl InnerIndexReader {
-    fn load_segment_readers(&self) -> crate::Result<Vec<SegmentReader>> {
-        // We keep the lock until we have effectively finished opening the
-        // the `SegmentReader` because it prevents a diffferent process
-        // to garbage collect these file while we open them.
-        //
-        // Once opened, on linux & mac, the mmap will remain valid after
-        // the file has been deleted
-        // On windows, the file deletion will fail.
-        let _meta_lock = self.index.directory().acquire_lock(&META_LOCK)?;
-        let searchable_segments = self.searchable_segments()?;
-        searchable_segments
-            .iter()
-            .map(SegmentReader::open)
-            .collect::<crate::Result<_>>()
-    }
-
     fn reload(&self) -> crate::Result<()> {
-        let segment_readers: Vec<SegmentReader> = self.load_segment_readers()?;
+        let segment_readers: Vec<SegmentReader> = {
+            let _meta_lock = self.index.directory().acquire_lock(&META_LOCK)?;
+            let searchable_segments = self.searchable_segments()?;
+            searchable_segments
+                .iter()
+                .map(SegmentReader::open)
+                .collect::<crate::Result<_>>()?
+        };
         let schema = self.index.schema();
         let searchers = (0..self.num_searchers)
             .map(|_| Searcher::new(schema.clone(), self.index.clone(), segment_readers.clone()))
@@ -180,16 +162,18 @@ impl InnerIndexReader {
 ///
 /// `Clone` does not clone the different pool of searcher. `IndexReader`
 /// just wraps and `Arc`.
-=======
->>>>>>> Added NRTReader
-*/
 #[derive(Clone)]
-pub enum IndexReader {
-    FromMetaFile(MetaFileIndexReader),
-    NRT(NRTReader),
+pub struct IndexReader {
+    inner: Arc<InnerIndexReader>,
+    watch_handle_opt: Option<WatchHandle>,
 }
 
 impl IndexReader {
+    #[cfg(test)]
+    pub(crate) fn index(&self) -> Index {
+        self.inner.index.clone()
+    }
+
     /// Update searchers so that they reflect the state of the last
     /// `.commit()`.
     ///
@@ -200,10 +184,7 @@ impl IndexReader {
     /// This automatic reload can take 10s of milliseconds to kick in however, and in unit tests
     /// it can be nice to deterministically force the reload of searchers.
     pub fn reload(&self) -> crate::Result<()> {
-        match self {
-            IndexReader::FromMetaFile(meta_file_reader) => meta_file_reader.reload(),
-            IndexReader::NRT(nrt_reader) => nrt_reader.reload(),
-        }
+        self.inner.reload()
     }
 
     /// Returns a searcher
@@ -217,21 +198,6 @@ impl IndexReader {
     /// The same searcher must be used for a given query, as it ensures
     /// the use of a consistent segment set.
     pub fn searcher(&self) -> LeasedItem<Searcher> {
-        match self {
-            IndexReader::FromMetaFile(meta_file_reader) => meta_file_reader.searcher(),
-            IndexReader::NRT(nrt_reader) => nrt_reader.searcher(),
-        }
-    }
-}
-
-impl From<MetaFileIndexReader> for IndexReader {
-    fn from(meta_file_reader: MetaFileIndexReader) -> Self {
-        IndexReader::FromMetaFile(meta_file_reader)
-    }
-}
-
-impl From<NRTReader> for IndexReader {
-    fn from(nrt_reader: NRTReader) -> Self {
-        IndexReader::NRT(nrt_reader)
+        self.inner.searcher()
     }
 }

src/schema/document.rs

@@ -3,8 +3,8 @@ use crate::common::BinarySerializable;
 use crate::common::VInt;
 use crate::tokenizer::PreTokenizedString;
 use crate::DateTime;
-use itertools::Itertools;
 use std::io::{self, Read, Write};
+use std::mem;
 
 /// Tantivy's Document is the object that can
 /// be indexed and then searched for.
@@ -16,7 +16,7 @@ use std::io::{self, Read, Write};
 
 /// Documents are really just a list of couple `(field, value)`.
 /// In this list, one field may appear more than once.
-#[derive(Clone, Debug, Serialize, Deserialize, Default)]
+#[derive(Clone, Debug, serde::Serialize, serde::Deserialize, Default)]
 pub struct Document {
     field_values: Vec<FieldValue>,
 }
@@ -131,12 +131,34 @@ impl Document {
     pub fn get_sorted_field_values(&self) -> Vec<(Field, Vec<&FieldValue>)> {
         let mut field_values: Vec<&FieldValue> = self.field_values().iter().collect();
         field_values.sort_by_key(|field_value| field_value.field());
-        field_values
-            .into_iter()
-            .group_by(|field_value| field_value.field())
-            .into_iter()
-            .map(|(key, group)| (key, group.collect()))
-            .collect::<Vec<(Field, Vec<&FieldValue>)>>()
+
+        let mut grouped_field_values = vec![];
+
+        let mut current_field;
+        let mut current_group;
+
+        let mut field_values_it = field_values.into_iter();
+        if let Some(field_value) = field_values_it.next() {
+            current_field = field_value.field();
+            current_group = vec![field_value]
+        } else {
+            return grouped_field_values;
+        }
+
+        for field_value in field_values_it {
+            if field_value.field() == current_field {
+                current_group.push(field_value);
+            } else {
+                grouped_field_values.push((
+                    current_field,
+                    mem::replace(&mut current_group, vec![field_value]),
+                ));
+                current_field = field_value.field();
+            }
+        }
+
+        grouped_field_values.push((current_field, current_group));
+        grouped_field_values
     }
 
     /// Returns all of the `FieldValue`s associated the given field

src/schema/field.rs

@@ -5,7 +5,9 @@ use std::io::Write;
 
 /// `Field` is represented by an unsigned 32-bit integer type
 /// The schema holds the mapping between field names and `Field` objects.
-#[derive(Copy, Clone, Debug, PartialEq, PartialOrd, Eq, Ord, Hash, Serialize, Deserialize)]
+#[derive(
+    Copy, Clone, Debug, PartialEq, PartialOrd, Eq, Ord, Hash, serde::Serialize, serde::Deserialize,
+)]
 pub struct Field(u32);
 
 impl Field {

src/schema/field_value.rs

@@ -1,12 +1,10 @@
 use crate::common::BinarySerializable;
 use crate::schema::Field;
 use crate::schema::Value;
-use std::io;
-use std::io::Read;
-use std::io::Write;
+use std::io::{self, Read, Write};
 
 /// `FieldValue` holds together a `Field` and its `Value`.
-#[derive(Debug, Clone, Ord, PartialEq, Eq, PartialOrd, Serialize, Deserialize)]
+#[derive(Debug, Clone, Ord, PartialEq, Eq, PartialOrd, serde::Serialize, serde::Deserialize)]
 pub struct FieldValue {
     field: Field,
     value: Value,

src/schema/index_record_option.rs

@@ -1,3 +1,5 @@
+use serde::{Deserialize, Serialize};
+
 /// `IndexRecordOption` describes an amount information associated
 /// to a given indexed field.
 ///

src/schema/int_options.rs

@@ -1,4 +1,5 @@
 use crate::schema::flags::{FastFlag, IndexedFlag, SchemaFlagList, StoredFlag};
+use serde::{Deserialize, Serialize};
 use std::ops::BitOr;
 
 /// Express whether a field is single-value or multi-valued.

src/schema/named_field_document.rs

@@ -1,4 +1,5 @@
 use crate::schema::Value;
+use serde::Serialize;
 use std::collections::BTreeMap;
 
 /// Internal representation of a document used for JSON

src/schema/text_options.rs

@@ -1,6 +1,7 @@
 use crate::schema::flags::SchemaFlagList;
 use crate::schema::flags::StoredFlag;
 use crate::schema::IndexRecordOption;
+use serde::{Deserialize, Serialize};
 use std::borrow::Cow;
 use std::ops::BitOr;
 
@@ -155,30 +156,17 @@ mod tests {
 
     #[test]
     fn test_field_options() {
-        {
-            let field_options = STORED | TEXT;
-            assert!(field_options.is_stored());
-            assert!(field_options.get_indexing_options().is_some());
-        }
-        {
-            let mut schema_builder = Schema::builder();
-            schema_builder.add_text_field("body", TEXT);
-            let schema = schema_builder.build();
-            let field = schema.get_field("body").unwrap();
-            let field_entry = schema.get_field_entry(field);
-            match field_entry.field_type() {
-                &FieldType::Str(ref text_options) => {
-                    assert!(text_options.get_indexing_options().is_some());
-                    assert_eq!(
-                        text_options.get_indexing_options().unwrap().tokenizer(),
-                        "default"
-                    );
-                }
-                _ => {
-                    panic!("");
-                }
-            }
-        }
+        let field_options = STORED | TEXT;
+        assert!(field_options.is_stored());
+        assert!(field_options.get_indexing_options().is_some());
+        let mut schema_builder = Schema::builder();
+        schema_builder.add_text_field("body", TEXT);
+        let schema = schema_builder.build();
+        let field = schema.get_field("body").unwrap();
+        let field_entry = schema.get_field_entry(field);
+        assert!(matches!(field_entry.field_type(),
+                &FieldType::Str(ref text_options)
+                if text_options.get_indexing_options().unwrap().tokenizer() == "default"));
     }
 
     #[test]

src/space_usage/mod.rs

@@ -11,6 +11,7 @@ under-count actual resultant space usage by up to 4095 bytes per file.
 
 use crate::schema::Field;
 use crate::SegmentComponent;
+use serde::{Deserialize, Serialize};
 use std::collections::HashMap;
 
 /// Indicates space usage in bytes

src/store/compression_snap.rs

@@ -1,5 +1,3 @@
-use snap;
-
 use std::io::{self, Read, Write};
 
 /// Name of the compression scheme used in the doc store.

src/termdict/mod.rs

@@ -434,6 +434,7 @@ mod tests {
 
     #[test]
     fn test_automaton_search() {
+        use crate::query::DFAWrapper;
         use levenshtein_automata::LevenshteinAutomatonBuilder;
 
         const COUNTRIES: [&'static str; 7] = [
@@ -463,7 +464,7 @@ mod tests {
 
         // We can now build an entire dfa.
         let lev_automaton_builder = LevenshteinAutomatonBuilder::new(2, true);
-        let automaton = lev_automaton_builder.build_dfa("Spaen");
+        let automaton = DFAWrapper(lev_automaton_builder.build_dfa("Spaen"));
 
         let mut range = term_dict.search(automaton).into_stream();
 

src/termdict/termdict.rs

@@ -7,7 +7,6 @@ use crate::postings::TermInfo;
 use crate::termdict::TermOrdinal;
 use once_cell::sync::Lazy;
 use std::io::{self, Write};
-use tantivy_fst;
 use tantivy_fst::raw::Fst;
 use tantivy_fst::Automaton;
 

src/tokenizer/stemmer.rs

@@ -1,6 +1,7 @@
 use super::{Token, TokenFilter, TokenStream};
 use crate::tokenizer::BoxTokenStream;
 use rust_stemmers::{self, Algorithm};
+use serde::{Deserialize, Serialize};
 
 /// Available stemmer languages.
 #[derive(Debug, Serialize, Deserialize, Eq, PartialEq, Copy, Clone)]

src/tokenizer/tokenized_string.rs

@@ -1,4 +1,5 @@
 use crate::tokenizer::{BoxTokenStream, Token, TokenStream, TokenStreamChain};
+use serde::{Deserialize, Serialize};
 use std::cmp::Ordering;
 
 /// Struct representing pre-tokenized text

src/tokenizer/tokenizer.rs

@@ -1,4 +1,5 @@
 use crate::tokenizer::TokenStreamChain;
+use serde::{Deserialize, Serialize};
 /// The tokenizer module contains all of the tools used to process
 /// text in `tantivy`.
 use std::borrow::{Borrow, BorrowMut};