Rebased

CHANGELOG.md

@@ -1,22 +1,6 @@
-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.
-- Change in the DocSet and Scorer API. (@fulmicoton). 
-A freshly created DocSet point directly to their first doc. A sentinel value called TERMINATED marks the end of a DocSet.
-`.advance()` returns the new DocId. `Scorer::skip(target)` has been replaced by `Scorer::seek(target)` and returns the resulting DocId.
-As a result, iterating through DocSet now looks as follows
-```rust
-let mut doc = docset.doc();
-while doc != TERMINATED {
-   // ...
-   doc = docset.advance();
-}
-```
-The change made it possible to greatly simplify a lot of the docset's code.
-- Misc internal optimization and introduction of the `Scorer::for_each_pruning` function. (@fulmicoton)
-- Added an offset option to the Top(.*)Collectors. (@robyoung)
+Tantivy 0.12.1
+=====================
+- By default IndexReader are in `Manual` mode.
 
 Tantivy 0.12.0
 ======================

Cargo.toml

@@ -13,23 +13,25 @@ keywords = ["search", "information", "retrieval"]
 edition = "2018"
 
 [dependencies]
-base64 = "0.12.0"
+base64 = "0.11.0"
 byteorder = "1.0"
 crc32fast = "1.2.0"
 once_cell = "1.0"
 regex ={version = "1.3.0", default-features = false, features = ["std"]}
-tantivy-fst = "0.3"
+tantivy-fst = "0.2.1"
 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 = {version="1.0", features=["derive"]}
+serde = "1.0"
+serde_derive = "1.0"
 serde_json = "1.0"
 num_cpus = "1.2"
 fs2={version="0.4", optional=true}
-levenshtein_automata = "0.2"
+itertools = "0.8"
+levenshtein_automata = "0.1"
 notify = {version="4", optional=true}
 uuid = { version = "0.8", features = ["v4", "serde"] }
 crossbeam = "0.7"
@@ -38,14 +40,14 @@ owning_ref = "0.4"
 stable_deref_trait = "1.0.0"
 rust-stemmers = "1.2"
 downcast-rs = { version="1.0" }
-tantivy-query-grammar = { version="0.13", path="./query-grammar" }
+tantivy-query-grammar = { version="0.12", 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 = "0.4"
+fail = "0.3"
 murmurhash32 = "0.2"
 chrono = "0.4"
 smallvec = "1.0"
@@ -60,7 +62,7 @@ maplit = "1"
 matches = "0.1.8"
 
 [dev-dependencies.fail]
-version = "0.4"
+version = "0.3"
 features = ["failpoints"]
 
 [profile.release]

README.md

@@ -31,20 +31,16 @@ Tantivy is, in fact, strongly inspired by Lucene's design.
 
 # Benchmark
 
+Tantivy is typically faster than Lucene, but the results depend on 
+the nature of the queries in your workload.
+
 The following [benchmark](https://tantivy-search.github.io/bench/) break downs 
 performance for different type of queries / collection.
 
-
-In general, Tantivy tends to be 
-- slower than Lucene on union with a Top-K due to Block-WAND optimization.
-- faster than Lucene on intersection and phrase queries. 
-
-Your mileage WILL vary depending on the nature of queries and their load.
-
 # Features
 
 - Full-text search
-- 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))
+- 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))
 - 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 --all --verbose --no-default-features --features mmap
+  - REM SET RUST_LOG=tantivy,test & cargo test --verbose --no-default-features --features mmap
   - REM SET RUST_BACKTRACE=1 & cargo build --examples

examples/faceted_search_with_tweaked_score.rs

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

examples/iterating_docs_and_positions.rs

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

examples/pre_tokenized_text.rs

@@ -9,10 +9,11 @@
 // - 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 = "0.13.0"
+version = "0.12.0"
 authors = ["Paul Masurel <paul.masurel@gmail.com>"]
 license = "MIT"
 categories = ["database-implementations", "data-structures"]

query-grammar/src/query_grammar.rs

@@ -154,11 +154,17 @@ 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(ast())
-            .skip(char(')'))
+        char('-')
+            .with(leaf())
+            .map(negate)
+            .or(char('+').with(leaf()).map(must))
+            .or(char('(').with(ast()).skip(char(')')))
             .or(char('*').map(|_| UserInputAST::from(UserInputLeaf::All)))
             .or(attempt(
                 string("NOT").skip(spaces1()).with(leaf()).map(negate),
@@ -170,16 +176,6 @@ 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)>)| {
@@ -243,29 +239,21 @@ fn aggregate_binary_expressions(
     }
 }
 
-fn operand_leaf<'a>() -> impl Parser<&'a str, Output = (BinaryOperand, UserInputAST)> {
-    (
+pub fn ast<'a>() -> impl Parser<&'a str, Output = UserInputAST> {
+    let operand_leaf = (
         binary_operand().skip(spaces()),
         boosted_leaf().skip(spaces()),
-    )
-}
-
-pub fn ast<'a>() -> impl Parser<&'a str, Output = UserInputAST> {
-    let boolean_expr = (boosted_leaf().skip(spaces()), many1(operand_leaf()))
+    );
+    let boolean_expr = (boosted_leaf().skip(spaces().silent()), many1(operand_leaf))
         .map(|(left, right)| aggregate_binary_expressions(left, right));
-    let whitespace_separated_leaves = many1(occur_leaf().skip(spaces().silent())).map(
-        |subqueries: Vec<(Option<Occur>, UserInputAST)>| {
+    let whitespace_separated_leaves =
+        many1(boosted_leaf().skip(spaces().silent())).map(|subqueries: Vec<UserInputAST>| {
             if subqueries.len() == 1 {
-                let (occur_opt, ast) = subqueries.into_iter().next().unwrap();
-                match occur_opt.unwrap_or(Occur::Should) {
-                    Occur::Must | Occur::Should => ast,
-                    Occur::MustNot => UserInputAST::Clause(vec![(Some(Occur::MustNot), ast)]),
-                }
+                subqueries.into_iter().next().unwrap()
             } else {
                 UserInputAST::Clause(subqueries.into_iter().collect())
             }
-        },
-    );
+        });
     let expr = attempt(boolean_expr).or(whitespace_separated_leaves);
     spaces().with(expr).skip(spaces())
 }
@@ -295,12 +283,6 @@ 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) {
@@ -348,7 +330,7 @@ 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]
@@ -356,16 +338,16 @@ 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)"
@@ -403,13 +385,6 @@ 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
@@ -438,67 +413,32 @@ 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_single_term() {
+    fn test_parse_query_to_ast() {
         test_parse_query_to_ast_helper("abc", "\"abc\"");
-    }
-
-    #[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("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_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]
-    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:toto", "+(abc:\"toto\")");
+        test_parse_query_to_ast_helper("(+abc:toto -titi)", "(+(abc:\"toto\") -(\"titi\"))");
+        test_parse_query_to_ast_helper("-abc:toto", "-(abc:\"toto\")");
+        test_parse_query_to_ast_helper("abc:a b", "(abc:\"a\" \"b\")");
         test_parse_query_to_ast_helper("abc:\"a b\"", "abc:\"a b\"");
         test_parse_query_to_ast_helper("foo:[1 TO 5]", "foo:[\"1\" TO \"5\"]");
-    }
-
-    #[test]
-    fn test_parse_query_with_range() {
         test_parse_query_to_ast_helper("[1 TO 5]", "[\"1\" TO \"5\"]");
         test_parse_query_to_ast_helper("foo:{a TO z}", "foo:{\"a\" TO \"z\"}");
         test_parse_query_to_ast_helper("foo:[1 TO toto}", "foo:[\"1\" TO \"toto\"}");

query-grammar/src/user_input_ast.rs

@@ -85,14 +85,15 @@ impl UserInputBound {
 }
 
 pub enum UserInputAST {
-    Clause(Vec<(Option<Occur>, UserInputAST)>),
+    Clause(Vec<UserInputAST>),
+    Unary(Occur, Box<UserInputAST>),
     Leaf(Box<UserInputLeaf>),
     Boost(Box<UserInputAST>, f32),
 }
 
 impl UserInputAST {
     pub fn unary(self, occur: Occur) -> UserInputAST {
-        UserInputAST::Clause(vec![(Some(occur), self)])
+        UserInputAST::Unary(occur, Box::new(self))
     }
 
     fn compose(occur: Occur, asts: Vec<UserInputAST>) -> UserInputAST {
@@ -103,7 +104,7 @@ impl UserInputAST {
         } else {
             UserInputAST::Clause(
                 asts.into_iter()
-                    .map(|ast: UserInputAST| (Some(occur), ast))
+                    .map(|ast: UserInputAST| ast.unary(occur))
                     .collect::<Vec<_>>(),
             )
         }
@@ -134,36 +135,25 @@ 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) -> fmt::Result {
+    fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> Result<(), fmt::Error> {
         match *self {
             UserInputAST::Clause(ref subqueries) => {
                 if subqueries.is_empty() {
                     write!(formatter, "<emptyclause>")?;
                 } else {
                     write!(formatter, "(")?;
-                    print_occur_ast(subqueries[0].0, &subqueries[0].1, formatter)?;
+                    write!(formatter, "{:?}", &subqueries[0])?;
                     for subquery in &subqueries[1..] {
-                        write!(formatter, " ")?;
-                        print_occur_ast(subquery.0, &subquery.1, formatter)?;
+                        write!(formatter, " {:?}", subquery)?;
                     }
                     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

@@ -11,13 +11,13 @@ impl<TCustomScorer, TScore> CustomScoreTopCollector<TCustomScorer, TScore>
 where
     TScore: Clone + PartialOrd,
 {
-    pub(crate) fn new(
+    pub fn new(
         custom_scorer: TCustomScorer,
-        collector: TopCollector<TScore>,
+        limit: usize,
     ) -> CustomScoreTopCollector<TCustomScorer, TScore> {
         CustomScoreTopCollector {
             custom_scorer,
-            collector,
+            collector: TopCollector::with_limit(limit),
         }
     }
 }
@@ -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(&mut self, doc: DocId) -> TScore;
+    fn score(&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 + FnMut(DocId) -> TScore,
+    F: 'static + Sync + Send + Fn(DocId) -> TScore,
 {
-    fn score(&mut self, doc: DocId) -> TScore {
+    fn score(&self, doc: DocId) -> TScore {
         (self)(doc)
     }
 }

src/collector/facet_collector.rs

@@ -1,5 +1,6 @@
 use crate::collector::Collector;
 use crate::collector::SegmentCollector;
+use crate::docset::SkipResult;
 use crate::fastfield::FacetReader;
 use crate::schema::Facet;
 use crate::schema::Field;
@@ -187,11 +188,6 @@ pub struct FacetSegmentCollector {
     collapse_facet_ords: Vec<u64>,
 }
 
-enum SkipResult {
-    Found,
-    NotFound,
-}
-
 fn skip<'a, I: Iterator<Item = &'a Facet>>(
     target: &[u8],
     collapse_it: &mut Peekable<I>,
@@ -201,14 +197,14 @@ fn skip<'a, I: Iterator<Item = &'a Facet>>(
             Some(facet_bytes) => match facet_bytes.encoded_str().as_bytes().cmp(target) {
                 Ordering::Less => {}
                 Ordering::Greater => {
-                    return SkipResult::NotFound;
+                    return SkipResult::OverStep;
                 }
                 Ordering::Equal => {
-                    return SkipResult::Found;
+                    return SkipResult::Reached;
                 }
             },
             None => {
-                return SkipResult::NotFound;
+                return SkipResult::End;
             }
         }
         collapse_it.next();
@@ -285,7 +281,7 @@ impl Collector for FacetCollector {
                     // is positionned on a term that has not been processed yet.
                     let skip_result = skip(facet_streamer.key(), &mut collapse_facet_it);
                     match skip_result {
-                        SkipResult::Found => {
+                        SkipResult::Reached => {
                             // we reach a facet we decided to collapse.
                             let collapse_depth = facet_depth(facet_streamer.key());
                             let mut collapsed_id = 0;
@@ -305,7 +301,7 @@ impl Collector for FacetCollector {
                             }
                             break;
                         }
-                        SkipResult::NotFound => {
+                        SkipResult::End | SkipResult::OverStep => {
                             collapse_mapping.push(0);
                             if !facet_streamer.advance() {
                                 break;

src/collector/mod.rs

@@ -109,7 +109,6 @@ pub use self::tweak_score_top_collector::{ScoreSegmentTweaker, ScoreTweaker};
 
 mod facet_collector;
 pub use self::facet_collector::FacetCollector;
-use crate::query::Weight;
 
 /// `Fruit` is the type for the result of our collection.
 /// e.g. `usize` for the `Count` collector.
@@ -155,29 +154,6 @@ pub trait Collector: Sync {
     /// Combines the fruit associated to the collection of each segments
     /// into one fruit.
     fn merge_fruits(&self, segment_fruits: Vec<Self::Fruit>) -> crate::Result<Self::Fruit>;
-
-    /// Created a segment collector and
-    fn collect_segment(
-        &self,
-        weight: &dyn Weight,
-        segment_ord: u32,
-        reader: &SegmentReader,
-    ) -> crate::Result<<Self::Child as SegmentCollector>::Fruit> {
-        let mut segment_collector = self.for_segment(segment_ord as u32, reader)?;
-
-        if let Some(delete_bitset) = reader.delete_bitset() {
-            weight.for_each(reader, &mut |doc, score| {
-                if delete_bitset.is_alive(doc) {
-                    segment_collector.collect(doc, score);
-                }
-            })?;
-        } else {
-            weight.for_each(reader, &mut |doc, score| {
-                segment_collector.collect(doc, score);
-            })?;
-        }
-        Ok(segment_collector.harvest())
-    }
 }
 
 /// The `SegmentCollector` is the trait in charge of defining the

src/collector/top_collector.rs

@@ -18,9 +18,9 @@ use std::collections::BinaryHeap;
 /// Two elements are equal if their feature is equal, and regardless of whether `doc`
 /// is equal. This should be perfectly fine for this usage, but let's make sure this
 /// struct is never public.
-pub(crate) struct ComparableDoc<T, D> {
-    pub feature: T,
-    pub doc: D,
+struct ComparableDoc<T, D> {
+    feature: T,
+    doc: D,
 }
 
 impl<T: PartialOrd, D: PartialOrd> PartialOrd for ComparableDoc<T, D> {
@@ -56,8 +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> {
-    pub limit: usize,
-    pub offset: usize,
+    limit: usize,
     _marker: PhantomData<T>,
 }
 
@@ -73,20 +72,14 @@ where
         if limit < 1 {
             panic!("Limit must be strictly greater than 0.");
         }
-        Self {
+        TopCollector {
             limit,
-            offset: 0,
             _marker: PhantomData,
         }
     }
 
-    /// Skip the first "offset" documents when collecting.
-    ///
-    /// This is equivalent to `OFFSET` in MySQL or PostgreSQL and `start` in
-    /// Lucene's TopDocsCollector.
-    pub fn and_offset(mut self, offset: usize) -> TopCollector<T> {
-        self.offset = offset;
-        self
+    pub fn limit(&self) -> usize {
+        self.limit
     }
 
     pub fn merge_fruits(
@@ -99,7 +92,7 @@ where
         let mut top_collector = BinaryHeap::new();
         for child_fruit in children {
             for (feature, doc) in child_fruit {
-                if top_collector.len() < (self.limit + self.offset) {
+                if top_collector.len() < self.limit {
                     top_collector.push(ComparableDoc { feature, doc });
                 } else if let Some(mut head) = top_collector.peek_mut() {
                     if head.feature < feature {
@@ -111,7 +104,6 @@ where
         Ok(top_collector
             .into_sorted_vec()
             .into_iter()
-            .skip(self.offset)
             .map(|cdoc| (cdoc.feature, cdoc.doc))
             .collect())
     }
@@ -121,23 +113,7 @@ where
         segment_id: SegmentLocalId,
         _: &SegmentReader,
     ) -> crate::Result<TopSegmentCollector<F>> {
-        Ok(TopSegmentCollector::new(
-            segment_id,
-            self.limit + self.offset,
-        ))
-    }
-
-    /// Create a new TopCollector with the same limit and offset.
-    ///
-    /// Ideally we would use Into but the blanket implementation seems to cause the Scorer traits
-    /// to fail.
-    #[doc(hidden)]
-    pub(crate) fn into_tscore<TScore: PartialOrd + Clone>(self) -> TopCollector<TScore> {
-        TopCollector {
-            limit: self.limit,
-            offset: self.offset,
-            _marker: PhantomData,
-        }
+        Ok(TopSegmentCollector::new(segment_id, self.limit))
     }
 }
 
@@ -211,7 +187,7 @@ impl<T: PartialOrd + Clone> TopSegmentCollector<T> {
 
 #[cfg(test)]
 mod tests {
-    use super::{TopCollector, TopSegmentCollector};
+    use super::TopSegmentCollector;
     use crate::DocAddress;
 
     #[test]
@@ -272,48 +248,6 @@ mod tests {
             top_collector_limit_3.harvest()[..2].to_vec(),
         );
     }
-
-    #[test]
-    fn test_top_collector_with_limit_and_offset() {
-        let collector = TopCollector::with_limit(2).and_offset(1);
-
-        let results = collector
-            .merge_fruits(vec![vec![
-                (0.9, DocAddress(0, 1)),
-                (0.8, DocAddress(0, 2)),
-                (0.7, DocAddress(0, 3)),
-                (0.6, DocAddress(0, 4)),
-                (0.5, DocAddress(0, 5)),
-            ]])
-            .unwrap();
-
-        assert_eq!(
-            results,
-            vec![(0.8, DocAddress(0, 2)), (0.7, DocAddress(0, 3)),]
-        );
-    }
-
-    #[test]
-    fn test_top_collector_with_limit_larger_than_set_and_offset() {
-        let collector = TopCollector::with_limit(2).and_offset(1);
-
-        let results = collector
-            .merge_fruits(vec![vec![(0.9, DocAddress(0, 1)), (0.8, DocAddress(0, 2))]])
-            .unwrap();
-
-        assert_eq!(results, vec![(0.8, DocAddress(0, 2)),]);
-    }
-
-    #[test]
-    fn test_top_collector_with_limit_and_offset_larger_than_set() {
-        let collector = TopCollector::with_limit(2).and_offset(20);
-
-        let results = collector
-            .merge_fruits(vec![vec![(0.9, DocAddress(0, 1)), (0.8, DocAddress(0, 2))]])
-            .unwrap();
-
-        assert_eq!(results, vec![]);
-    }
 }
 
 #[cfg(all(test, feature = "unstable"))]

src/collector/top_score_collector.rs

@@ -1,20 +1,18 @@
 use super::Collector;
 use crate::collector::custom_score_top_collector::CustomScoreTopCollector;
+use crate::collector::top_collector::TopCollector;
 use crate::collector::top_collector::TopSegmentCollector;
-use crate::collector::top_collector::{ComparableDoc, TopCollector};
 use crate::collector::tweak_score_top_collector::TweakedScoreTopCollector;
 use crate::collector::{
     CustomScorer, CustomSegmentScorer, ScoreSegmentTweaker, ScoreTweaker, SegmentCollector,
 };
 use crate::fastfield::FastFieldReader;
-use crate::query::Weight;
 use crate::schema::Field;
 use crate::DocAddress;
 use crate::DocId;
 use crate::Score;
 use crate::SegmentLocalId;
 use crate::SegmentReader;
-use std::collections::BinaryHeap;
 use std::fmt;
 
 /// The `TopDocs` collector keeps track of the top `K` documents
@@ -59,11 +57,7 @@ pub struct TopDocs(TopCollector<Score>);
 
 impl fmt::Debug for TopDocs {
     fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        write!(
-            f,
-            "TopDocs(limit={}, offset={})",
-            self.0.limit, self.0.offset
-        )
+        write!(f, "TopDocs({})", self.0.limit())
     }
 }
 
@@ -72,7 +66,7 @@ struct ScorerByFastFieldReader {
 }
 
 impl CustomSegmentScorer<u64> for ScorerByFastFieldReader {
-    fn score(&mut self, doc: DocId) -> u64 {
+    fn score(&self, doc: DocId) -> u64 {
         self.ff_reader.get_u64(u64::from(doc))
     }
 }
@@ -107,45 +101,6 @@ impl TopDocs {
         TopDocs(TopCollector::with_limit(limit))
     }
 
-    /// Skip the first "offset" documents when collecting.
-    ///
-    /// This is equivalent to `OFFSET` in MySQL or PostgreSQL and `start` in
-    /// Lucene's TopDocsCollector.
-    ///
-    /// ```rust
-    /// use tantivy::collector::TopDocs;
-    /// use tantivy::query::QueryParser;
-    /// use tantivy::schema::{Schema, TEXT};
-    /// use tantivy::{doc, DocAddress, Index};
-    ///
-    /// let mut schema_builder = Schema::builder();
-    /// let title = schema_builder.add_text_field("title", TEXT);
-    /// let schema = schema_builder.build();
-    /// let index = Index::create_in_ram(schema);
-    ///
-    /// let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-    /// index_writer.add_document(doc!(title => "The Name of the Wind"));
-    /// index_writer.add_document(doc!(title => "The Diary of Muadib"));
-    /// index_writer.add_document(doc!(title => "A Dairy Cow"));
-    /// index_writer.add_document(doc!(title => "The Diary of a Young Girl"));
-    /// index_writer.add_document(doc!(title => "The Diary of Lena Mukhina"));
-    /// assert!(index_writer.commit().is_ok());
-    ///
-    /// let reader = index.reader().unwrap();
-    /// let searcher = reader.searcher();
-    ///
-    /// let query_parser = QueryParser::for_index(&index, vec![title]);
-    /// let query = query_parser.parse_query("diary").unwrap();
-    /// let top_docs = searcher.search(&query, &TopDocs::with_limit(2).and_offset(1)).unwrap();
-    ///
-    /// assert_eq!(top_docs.len(), 2);
-    /// assert_eq!(&top_docs[0], &(0.5204813, DocAddress(0, 4)));
-    /// assert_eq!(&top_docs[1], &(0.4793185, DocAddress(0, 3)));
-    /// ```
-    pub fn and_offset(self, offset: usize) -> TopDocs {
-        TopDocs(self.0.and_offset(offset))
-    }
-
     /// Set top-K to rank documents by a given fast field.
     ///
     /// ```rust
@@ -326,7 +281,7 @@ impl TopDocs {
         TScoreSegmentTweaker: ScoreSegmentTweaker<TScore> + 'static,
         TScoreTweaker: ScoreTweaker<TScore, Child = TScoreSegmentTweaker>,
     {
-        TweakedScoreTopCollector::new(score_tweaker, self.0.into_tscore())
+        TweakedScoreTopCollector::new(score_tweaker, self.0.limit())
     }
 
     /// Ranks the documents using a custom score.
@@ -440,7 +395,7 @@ impl TopDocs {
         TCustomSegmentScorer: CustomSegmentScorer<TScore> + 'static,
         TCustomScorer: CustomScorer<TScore, Child = TCustomSegmentScorer>,
     {
-        CustomScoreTopCollector::new(custom_score, self.0.into_tscore())
+        CustomScoreTopCollector::new(custom_score, self.0.limit())
     }
 }
 
@@ -468,64 +423,6 @@ impl Collector for TopDocs {
     ) -> crate::Result<Self::Fruit> {
         self.0.merge_fruits(child_fruits)
     }
-
-    fn collect_segment(
-        &self,
-        weight: &dyn Weight,
-        segment_ord: u32,
-        reader: &SegmentReader,
-    ) -> crate::Result<<Self::Child as SegmentCollector>::Fruit> {
-        let heap_len = self.0.limit + self.0.offset;
-        let mut heap: BinaryHeap<ComparableDoc<Score, DocId>> = BinaryHeap::with_capacity(heap_len);
-
-        if let Some(delete_bitset) = reader.delete_bitset() {
-            let mut threshold = f32::MIN;
-            weight.for_each_pruning(threshold, reader, &mut |doc, score| {
-                if delete_bitset.is_deleted(doc) {
-                    return threshold;
-                }
-                let heap_item = ComparableDoc {
-                    feature: score,
-                    doc,
-                };
-                if heap.len() < heap_len {
-                    heap.push(heap_item);
-                    if heap.len() == heap_len {
-                        threshold = heap.peek().map(|el| el.feature).unwrap_or(f32::MIN);
-                    }
-                    return threshold;
-                }
-                *heap.peek_mut().unwrap() = heap_item;
-                threshold = heap.peek().map(|el| el.feature).unwrap_or(std::f32::MIN);
-                threshold
-            })?;
-        } else {
-            weight.for_each_pruning(f32::MIN, reader, &mut |doc, score| {
-                let heap_item = ComparableDoc {
-                    feature: score,
-                    doc,
-                };
-                if heap.len() < heap_len {
-                    heap.push(heap_item);
-                    // TODO the threshold is suboptimal for heap.len == heap_len
-                    if heap.len() == heap_len {
-                        return heap.peek().map(|el| el.feature).unwrap_or(f32::MIN);
-                    } else {
-                        return f32::MIN;
-                    }
-                }
-                *heap.peek_mut().unwrap() = heap_item;
-                heap.peek().map(|el| el.feature).unwrap_or(std::f32::MIN)
-            })?;
-        }
-
-        let fruit = heap
-            .into_sorted_vec()
-            .into_iter()
-            .map(|cid| (cid.feature, DocAddress(segment_ord, cid.doc)))
-            .collect();
-        Ok(fruit)
-    }
 }
 
 /// Segment Collector associated to `TopDocs`.
@@ -535,7 +432,7 @@ impl SegmentCollector for TopScoreSegmentCollector {
     type Fruit = Vec<(Score, DocAddress)>;
 
     fn collect(&mut self, doc: DocId, score: Score) {
-        self.0.collect(doc, score);
+        self.0.collect(doc, score)
     }
 
     fn harvest(self) -> Vec<(Score, DocAddress)> {
@@ -549,10 +446,11 @@ mod tests {
     use crate::collector::Collector;
     use crate::query::{AllQuery, Query, QueryParser};
     use crate::schema::{Field, Schema, FAST, STORED, TEXT};
+    use crate::DocAddress;
     use crate::Index;
     use crate::IndexWriter;
     use crate::Score;
-    use crate::{DocAddress, DocId, SegmentReader};
+    use itertools::Itertools;
 
     fn make_index() -> Index {
         let mut schema_builder = Schema::builder();
@@ -592,21 +490,6 @@ mod tests {
         );
     }
 
-    #[test]
-    fn test_top_collector_not_at_capacity_with_offset() {
-        let index = make_index();
-        let field = index.schema().get_field("text").unwrap();
-        let query_parser = QueryParser::for_index(&index, vec![field]);
-        let text_query = query_parser.parse_query("droopy tax").unwrap();
-        let score_docs: Vec<(Score, DocAddress)> = index
-            .reader()
-            .unwrap()
-            .searcher()
-            .search(&text_query, &TopDocs::with_limit(4).and_offset(2))
-            .unwrap();
-        assert_eq!(score_docs, vec![(0.48527452, DocAddress(0, 0))]);
-    }
-
     #[test]
     fn test_top_collector_at_capacity() {
         let index = make_index();
@@ -628,27 +511,6 @@ mod tests {
         );
     }
 
-    #[test]
-    fn test_top_collector_at_capacity_with_offset() {
-        let index = make_index();
-        let field = index.schema().get_field("text").unwrap();
-        let query_parser = QueryParser::for_index(&index, vec![field]);
-        let text_query = query_parser.parse_query("droopy tax").unwrap();
-        let score_docs: Vec<(Score, DocAddress)> = index
-            .reader()
-            .unwrap()
-            .searcher()
-            .search(&text_query, &TopDocs::with_limit(2).and_offset(1))
-            .unwrap();
-        assert_eq!(
-            score_docs,
-            vec![
-                (0.5376842, DocAddress(0u32, 2)),
-                (0.48527452, DocAddress(0, 0))
-            ]
-        );
-    }
-
     #[test]
     fn test_top_collector_stable_sorting() {
         let index = make_index();
@@ -662,8 +524,8 @@ mod tests {
 
         // precondition for the test to be meaningful: we did get documents
         // with the same score
-        assert!(page_1.iter().all(|result| result.0 == page_1[0].0));
-        assert!(page_2.iter().all(|result| result.0 == page_2[0].0));
+        assert!(page_1.iter().map(|result| result.0).all_equal());
+        assert!(page_2.iter().map(|result| result.0).all_equal());
 
         // sanity check since we're relying on make_index()
         assert_eq!(page_1.len(), 2);
@@ -762,50 +624,6 @@ mod tests {
         }
     }
 
-    #[test]
-    fn test_tweak_score_top_collector_with_offset() {
-        let index = make_index();
-        let field = index.schema().get_field("text").unwrap();
-        let query_parser = QueryParser::for_index(&index, vec![field]);
-        let text_query = query_parser.parse_query("droopy tax").unwrap();
-        let collector = TopDocs::with_limit(2).and_offset(1).tweak_score(
-            move |_segment_reader: &SegmentReader| move |doc: DocId, _original_score: Score| doc,
-        );
-        let score_docs: Vec<(u32, DocAddress)> = index
-            .reader()
-            .unwrap()
-            .searcher()
-            .search(&text_query, &collector)
-            .unwrap();
-
-        assert_eq!(
-            score_docs,
-            vec![(1, DocAddress(0, 1)), (0, DocAddress(0, 0)),]
-        );
-    }
-
-    #[test]
-    fn test_custom_score_top_collector_with_offset() {
-        let index = make_index();
-        let field = index.schema().get_field("text").unwrap();
-        let query_parser = QueryParser::for_index(&index, vec![field]);
-        let text_query = query_parser.parse_query("droopy tax").unwrap();
-        let collector = TopDocs::with_limit(2)
-            .and_offset(1)
-            .custom_score(move |_segment_reader: &SegmentReader| move |doc: DocId| doc);
-        let score_docs: Vec<(u32, DocAddress)> = index
-            .reader()
-            .unwrap()
-            .searcher()
-            .search(&text_query, &collector)
-            .unwrap();
-
-        assert_eq!(
-            score_docs,
-            vec![(1, DocAddress(0, 1)), (0, DocAddress(0, 0)),]
-        );
-    }
-
     fn index(
         query: &str,
         query_field: Field,

src/collector/tweak_score_top_collector.rs

@@ -14,11 +14,11 @@ where
 {
     pub fn new(
         score_tweaker: TScoreTweaker,
-        collector: TopCollector<TScore>,
+        limit: usize,
     ) -> TweakedScoreTopCollector<TScoreTweaker, TScore> {
         TweakedScoreTopCollector {
             score_tweaker,
-            collector,
+            collector: TopCollector::with_limit(limit),
         }
     }
 }
@@ -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(&mut self, doc: DocId, score: Score) -> TScore;
+    fn score(&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 + FnMut(DocId, Score) -> TScore,
+    F: 'static + Sync + Send + Fn(DocId, Score) -> TScore,
 {
-    fn score(&mut self, doc: DocId, score: Score) -> TScore {
+    fn score(&self, doc: DocId, score: Score) -> TScore {
         (self)(doc, score)
     }
 }

src/common/bitset.rs

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

src/common/mod.rs

@@ -18,19 +18,6 @@ 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
@@ -147,7 +134,6 @@ 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;
@@ -213,21 +199,4 @@ 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

@@ -21,13 +21,12 @@ use crate::schema::FieldType;
 use crate::schema::Schema;
 use crate::tokenizer::{TextAnalyzer, TokenizerManager};
 use crate::IndexWriter;
+use num_cpus;
 use std::borrow::BorrowMut;
 use std::collections::HashSet;
 use std::fmt;
-
 #[cfg(feature = "mmap")]
-use std::path::Path;
-use std::path::PathBuf;
+use std::path::{Path, PathBuf};
 use std::sync::Arc;
 
 fn load_metas(
@@ -283,7 +282,7 @@ 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."
@@ -335,8 +334,7 @@ impl Index {
             .collect())
     }
 
-    #[doc(hidden)]
-    pub fn segment(&self, segment_meta: SegmentMeta) -> Segment {
+    pub(crate) fn segment(&self, segment_meta: SegmentMeta) -> Segment {
         Segment::for_index(self.clone(), segment_meta)
     }
 
@@ -348,6 +346,13 @@ 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
@@ -433,7 +438,7 @@ mod tests {
     }
 
     #[test]
-    fn create_should_wipeoff_existing() {
+    fn create_should_wipe_off_existing() {
         let directory = RAMDirectory::create();
         assert!(Index::create(directory.clone(), throw_away_schema()).is_ok());
         assert!(Index::exists(&directory));
@@ -470,7 +475,7 @@ mod tests {
             .try_into()
             .unwrap();
         assert_eq!(reader.searcher().num_docs(), 0);
-        test_index_on_commit_reload_policy_aux(field, &index, &reader);
+        test_index_on_commit_reload_policy_aux(field, index.clone(), &index, &reader);
     }
 
     #[cfg(feature = "mmap")]
@@ -494,7 +499,7 @@ mod tests {
                 .try_into()
                 .unwrap();
             assert_eq!(reader.searcher().num_docs(), 0);
-            test_index_on_commit_reload_policy_aux(field, &index, &reader);
+            test_index_on_commit_reload_policy_aux(field, index.clone(), &index, &reader);
         }
 
         #[test]
@@ -536,12 +541,16 @@ mod tests {
                 .try_into()
                 .unwrap();
             assert_eq!(reader.searcher().num_docs(), 0);
-            test_index_on_commit_reload_policy_aux(field, &write_index, &reader);
+            test_index_on_commit_reload_policy_aux(field, read_index, &write_index, &reader);
         }
     }
 
-    fn test_index_on_commit_reload_policy_aux(field: Field, index: &Index, reader: &IndexReader) {
-        let mut reader_index = reader.index();
+    fn test_index_on_commit_reload_policy_aux(
+        field: Field,
+        mut reader_index: Index,
+        index: &Index,
+        reader: &IndexReader,
+    ) {
         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

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

src/core/inverted_index_reader.rs

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

src/core/searcher.rs

@@ -1,8 +1,11 @@
 use crate::collector::Collector;
+use crate::collector::SegmentCollector;
 use crate::core::Executor;
 use crate::core::InvertedIndexReader;
 use crate::core::SegmentReader;
 use crate::query::Query;
+use crate::query::Scorer;
+use crate::query::Weight;
 use crate::schema::Document;
 use crate::schema::Schema;
 use crate::schema::{Field, Term};
@@ -14,6 +17,26 @@ use crate::Index;
 use std::fmt;
 use std::sync::Arc;
 
+fn collect_segment<C: Collector>(
+    collector: &C,
+    weight: &dyn Weight,
+    segment_ord: u32,
+    segment_reader: &SegmentReader,
+) -> crate::Result<C::Fruit> {
+    let mut scorer = weight.scorer(segment_reader, 1.0f32)?;
+    let mut segment_collector = collector.for_segment(segment_ord as u32, segment_reader)?;
+    if let Some(delete_bitset) = segment_reader.delete_bitset() {
+        scorer.for_each(&mut |doc, score| {
+            if delete_bitset.is_alive(doc) {
+                segment_collector.collect(doc, score);
+            }
+        });
+    } else {
+        scorer.for_each(&mut |doc, score| segment_collector.collect(doc, score));
+    }
+    Ok(segment_collector.harvest())
+}
+
 /// Holds a list of `SegmentReader`s ready for search.
 ///
 /// It guarantees that the `Segment` will not be removed before
@@ -140,7 +163,12 @@ impl Searcher {
         let segment_readers = self.segment_readers();
         let fruits = executor.map(
             |(segment_ord, segment_reader)| {
-                collector.collect_segment(weight.as_ref(), segment_ord as u32, segment_reader)
+                collect_segment(
+                    collector,
+                    weight.as_ref(),
+                    segment_ord as u32,
+                    segment_reader,
+                )
             },
             segment_readers.iter().enumerate(),
         )?;

src/core/segment.rs

@@ -3,19 +3,59 @@ use crate::core::Index;
 use crate::core::SegmentId;
 use crate::core::SegmentMeta;
 use crate::directory::error::{OpenReadError, OpenWriteError};
-use crate::directory::Directory;
+use crate::directory::{Directory, ManagedDirectory, RAMDirectory};
 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 {
-    index: Index,
+    schema: Schema,
     meta: SegmentMeta,
+    directory: SegmentDirectory,
 }
 
 impl fmt::Debug for Segment {
@@ -25,19 +65,55 @@ 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 { index, meta }
+        Segment {
+            directory: SegmentDirectory::Persisted(index.directory().clone()),
+            schema: index.schema(),
+            meta,
+        }
     }
 
-    /// 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()
+    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 segment meta-information
@@ -51,7 +127,8 @@ impl Segment {
     /// as we finalize a fresh new segment.
     pub(crate) fn with_max_doc(self, max_doc: u32) -> Segment {
         Segment {
-            index: self.index,
+            directory: self.directory,
+            schema: self.schema,
             meta: self.meta.with_max_doc(max_doc),
         }
     }
@@ -59,7 +136,8 @@ impl Segment {
     #[doc(hidden)]
     pub fn with_delete_meta(self, num_deleted_docs: u32, opstamp: Opstamp) -> Segment {
         Segment {
-            index: self.index,
+            directory: self.directory,
+            schema: self.schema,
             meta: self.meta.with_delete_meta(num_deleted_docs, opstamp),
         }
     }
@@ -80,14 +158,14 @@ 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.index.directory().open_read(&path)?;
+        let source = self.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.index.directory_mut().open_write(&path)?;
+        let write = self.directory.open_write(&path)?;
         Ok(write)
     }
 }

src/core/segment_id.rs

@@ -4,7 +4,6 @@ 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

@@ -56,6 +56,68 @@ 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
@@ -142,68 +204,68 @@ 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

@@ -11,6 +11,7 @@ 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

@@ -1,3 +1,10 @@
+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::{
@@ -13,12 +20,8 @@ use crate::directory::WatchCallback;
 use crate::directory::WatchCallbackList;
 use crate::directory::WatchHandle;
 use crate::directory::{TerminatingWrite, WritePtr};
-use fs2::FileExt;
+use atomicwrites;
 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;
@@ -220,13 +223,17 @@ struct MmapDirectoryInner {
 }
 
 impl MmapDirectoryInner {
-    fn new(root_path: PathBuf, temp_directory: Option<TempDir>) -> MmapDirectoryInner {
-        MmapDirectoryInner {
+    fn new(
+        root_path: PathBuf,
+        temp_directory: Option<TempDir>,
+    ) -> Result<MmapDirectoryInner, OpenDirectoryError> {
+        let mmap_directory_inner = MmapDirectoryInner {
             root_path,
             mmap_cache: Default::default(),
             _temp_directory: temp_directory,
             watcher: RwLock::new(None),
-        }
+        };
+        Ok(mmap_directory_inner)
     }
 
     fn watch(&self, watch_callback: WatchCallback) -> crate::Result<WatchHandle> {
@@ -260,11 +267,14 @@ impl fmt::Debug for MmapDirectory {
 }
 
 impl MmapDirectory {
-    fn new(root_path: PathBuf, temp_directory: Option<TempDir>) -> MmapDirectory {
-        let inner = MmapDirectoryInner::new(root_path, temp_directory);
-        MmapDirectory {
+    fn new(
+        root_path: PathBuf,
+        temp_directory: Option<TempDir>,
+    ) -> Result<MmapDirectory, OpenDirectoryError> {
+        let inner = MmapDirectoryInner::new(root_path, temp_directory)?;
+        Ok(MmapDirectory {
             inner: Arc::new(inner),
-        }
+        })
     }
 
     /// Creates a new MmapDirectory in a temporary directory.
@@ -274,7 +284,7 @@ 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());
-        Ok(MmapDirectory::new(tempdir_path, Some(tempdir)))
+        MmapDirectory::new(tempdir_path, Some(tempdir))
     }
 
     /// Opens a MmapDirectory in a directory.
@@ -292,7 +302,7 @@ 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

@@ -145,12 +145,6 @@ 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() {
@@ -220,28 +214,3 @@ 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/docset.rs

@@ -1,47 +1,58 @@
+use crate::common::BitSet;
 use crate::fastfield::DeleteBitSet;
 use crate::DocId;
 use std::borrow::Borrow;
 use std::borrow::BorrowMut;
+use std::cmp::Ordering;
 
-/// Sentinel value returned when a DocSet has been entirely consumed.
-///
-/// This is not u32::MAX as one would have expected, due to the lack of SSE2 instructions
-/// to compare [u32; 4].
-pub const TERMINATED: DocId = std::i32::MAX as u32;
+/// Expresses the outcome of a call to `DocSet`'s `.skip_next(...)`.
+#[derive(PartialEq, Eq, Debug)]
+pub enum SkipResult {
+    /// target was in the docset
+    Reached,
+    /// target was not in the docset, skipping stopped as a greater element was found
+    OverStep,
+    /// the docset was entirely consumed without finding the target, nor any
+    /// element greater than the target.
+    End,
+}
 
 /// Represents an iterable set of sorted doc ids.
 pub trait DocSet {
     /// Goes to the next element.
-    ///
-    /// The DocId of the next element is returned.
-    /// In other words we should always have :
-    /// ```ignore
-    /// let doc = docset.advance();
-    /// assert_eq!(doc, docset.doc());
-    /// ```
-    ///
-    /// If we reached the end of the DocSet, TERMINATED should be returned.
-    ///
-    /// Calling `.advance()` on a terminated DocSet should be supported, and TERMINATED should
-    /// be returned.
-    /// TODO Test existing docsets.
-    fn advance(&mut self) -> DocId;
+    /// `.advance(...)` needs to be called a first time to point to the correct
+    /// element.
+    fn advance(&mut self) -> bool;
 
-    /// Advances the DocSet forward until reaching the target, or going to the
-    /// lowest DocId greater than the target.
+    /// After skipping, position the iterator in such a way that `.doc()`
+    /// will return a value greater than or equal to target.
     ///
-    /// If the end of the DocSet is reached, TERMINATED is returned.
+    /// SkipResult expresses whether the `target value` was reached, overstepped,
+    /// or if the `DocSet` was entirely consumed without finding any value
+    /// greater or equal to the `target`.
     ///
-    /// Calling `.seek(target)` on a terminated DocSet is legal. Implementation
-    /// of DocSet should support it.
+    /// WARNING: Calling skip always advances the docset.
+    /// More specifically, if the docset is already positionned on the target
+    /// skipping will advance to the next position and return SkipResult::Overstep.
     ///
-    /// Calling `seek(TERMINATED)` is also legal and is the normal way to consume a DocSet.
-    fn seek(&mut self, target: DocId) -> DocId {
-        let mut doc = self.doc();
-        while doc < target {
-            doc = self.advance();
+    /// If `.skip_next()` oversteps, then the docset must be positionned correctly
+    /// on an existing document. In other words, `.doc()` should return the first document
+    /// greater than `DocId`.
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        if !self.advance() {
+            return SkipResult::End;
+        }
+        loop {
+            match self.doc().cmp(&target) {
+                Ordering::Less => {
+                    if !self.advance() {
+                        return SkipResult::End;
+                    }
+                }
+                Ordering::Equal => return SkipResult::Reached,
+                Ordering::Greater => return SkipResult::OverStep,
+            }
         }
-        doc
     }
 
     /// Fills a given mutable buffer with the next doc ids from the
@@ -60,38 +71,38 @@ pub trait DocSet {
     /// use case where batching. The normal way to
     /// go through the `DocId`'s is to call `.advance()`.
     fn fill_buffer(&mut self, buffer: &mut [DocId]) -> usize {
-        if self.doc() == TERMINATED {
-            return 0;
-        }
         for (i, buffer_val) in buffer.iter_mut().enumerate() {
-            *buffer_val = self.doc();
-            if self.advance() == TERMINATED {
-                return i + 1;
+            if self.advance() {
+                *buffer_val = self.doc();
+            } else {
+                return i;
             }
         }
         buffer.len()
     }
 
     /// Returns the current document
-    /// Right after creating a new DocSet, the docset points to the first document.
-    ///
-    /// If the DocSet is empty, .doc() should return `TERMINATED`.
     fn doc(&self) -> DocId;
 
     /// Returns a best-effort hint of the
     /// length of the docset.
     fn size_hint(&self) -> u32;
 
+    /// Appends all docs to a `bitset`.
+    fn append_to_bitset(&mut self, bitset: &mut BitSet) {
+        while self.advance() {
+            bitset.insert(self.doc());
+        }
+    }
+
     /// Returns the number documents matching.
     /// Calling this method consumes the `DocSet`.
     fn count(&mut self, delete_bitset: &DeleteBitSet) -> u32 {
         let mut count = 0u32;
-        let mut doc = self.doc();
-        while doc != TERMINATED {
-            if !delete_bitset.is_deleted(doc) {
+        while self.advance() {
+            if !delete_bitset.is_deleted(self.doc()) {
                 count += 1u32;
             }
-            doc = self.advance();
         }
         count
     }
@@ -103,42 +114,22 @@ pub trait DocSet {
     /// given by `count()`.
     fn count_including_deleted(&mut self) -> u32 {
         let mut count = 0u32;
-        let mut doc = self.doc();
-        while doc != TERMINATED {
+        while self.advance() {
             count += 1u32;
-            doc = self.advance();
         }
         count
     }
 }
 
-impl<'a> DocSet for &'a mut dyn DocSet {
-    fn advance(&mut self) -> u32 {
-        (**self).advance()
-    }
-
-    fn seek(&mut self, target: DocId) -> DocId {
-        (**self).seek(target)
-    }
-
-    fn doc(&self) -> u32 {
-        (**self).doc()
-    }
-
-    fn size_hint(&self) -> u32 {
-        (**self).size_hint()
-    }
-}
-
 impl<TDocSet: DocSet + ?Sized> DocSet for Box<TDocSet> {
-    fn advance(&mut self) -> DocId {
+    fn advance(&mut self) -> bool {
         let unboxed: &mut TDocSet = self.borrow_mut();
         unboxed.advance()
     }
 
-    fn seek(&mut self, target: DocId) -> DocId {
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
         let unboxed: &mut TDocSet = self.borrow_mut();
-        unboxed.seek(target)
+        unboxed.skip_next(target)
     }
 
     fn doc(&self) -> DocId {
@@ -160,4 +151,9 @@ impl<TDocSet: DocSet + ?Sized> DocSet for Box<TDocSet> {
         let unboxed: &mut TDocSet = self.borrow_mut();
         unboxed.count_including_deleted()
     }
+
+    fn append_to_bitset(&mut self, bitset: &mut BitSet) {
+        let unboxed: &mut TDocSet = self.borrow_mut();
+        unboxed.append_to_bitset(bitset);
+    }
 }

src/error.rs

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

@@ -6,6 +6,7 @@ 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)
@@ -150,8 +151,8 @@ impl MultiValueIntFastFieldWriter {
                     }
                 }
                 None => {
-                    let val_min_max = crate::common::minmax(self.vals.iter().cloned());
-                    let (val_min, val_max) = val_min_max.unwrap_or((0u64, 0u64));
+                    let val_min_max = self.vals.iter().cloned().minmax();
+                    let (val_min, val_max) = val_min_max.into_option().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

@@ -8,30 +8,33 @@ 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::docset::{DocSet, TERMINATED};
+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::Opstamp;
+use crate::tokenizer::TokenizerManager;
+use crate::{IndexReader, Opstamp};
 use crossbeam::channel;
 use futures::executor::block_on;
 use futures::future::Future;
-use smallvec::smallvec;
-use smallvec::SmallVec;
+use smallvec::{smallvec, SmallVec};
 use std::mem;
 use std::ops::Range;
-use std::sync::Arc;
+use std::sync::{Arc, RwLock};
 use std::thread;
 use std::thread::JoinHandle;
 
@@ -68,6 +71,8 @@ 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,
@@ -87,6 +92,8 @@ pub struct IndexWriter {
 
     stamper: Stamper,
     committed_opstamp: Opstamp,
+
+    on_commit: WatchCallbackList,
 }
 
 fn compute_deleted_bitset(
@@ -112,15 +119,15 @@ fn compute_deleted_bitset(
         if let Some(mut docset) =
             inverted_index.read_postings(&delete_op.term, IndexRecordOption::Basic)
         {
-            let mut deleted_doc = docset.doc();
-            while deleted_doc != TERMINATED {
+            while docset.advance() {
+                let deleted_doc = docset.doc();
                 if deleted_doc < limit_doc {
                     delete_bitset.insert(deleted_doc);
                     might_have_changed = true;
                 }
-                deleted_doc = docset.advance();
             }
         }
+
         delete_cursor.advance();
     }
     Ok(might_have_changed)
@@ -133,7 +140,6 @@ 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<()> {
@@ -142,32 +148,38 @@ pub(crate) fn advance_deletes(
         return Ok(());
     }
 
-    if segment_entry.delete_bitset().is_none() && segment_entry.delete_cursor().get().is_none() {
+    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() {
         // 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 = match segment_entry.delete_bitset() {
-        Some(previous_delete_bitset) => (*previous_delete_bitset).clone(),
-        None => BitSet::with_max_value(max_doc),
-    };
+    let mut delete_bitset: BitSet =
+        delete_bitset_opt.unwrap_or_else(|| BitSet::with_max_value(max_doc));
 
     let num_deleted_docs_before = segment.meta().num_deleted_docs();
 
     compute_deleted_bitset(
         &mut delete_bitset,
         &segment_reader,
-        segment_entry.delete_cursor(),
+        &mut delete_cursor,
         &DocToOpstampMapping::None,
         target_opstamp,
     )?;
 
-    // TODO optimize
-    // It should be possible to do something smarter by manipulation bitsets directly
-    // to compute this union.
+    // TODO optimize... We are simply manipulating bitsets here.
+    // We should be able to compute the union much faster.
     if let Some(seg_delete_bitset) = segment_reader.delete_bitset() {
         for doc in 0u32..max_doc {
             if seg_delete_bitset.is_deleted(doc) {
@@ -176,16 +188,23 @@ pub(crate) fn advance_deletes(
         }
     }
 
-    let num_deleted_docs: u32 = delete_bitset.len() as u32;
+    let num_deleted_docs = delete_bitset.len() as u32;
+
     if num_deleted_docs > num_deleted_docs_before {
-        // There are new deletes. We need to write a new delete file.
-        segment = segment.with_delete_meta(num_deleted_docs as u32, target_opstamp);
-        let mut delete_file = segment.open_write(SegmentComponent::DELETE)?;
+        // 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)?;
         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);
     }
 
-    segment_entry.set_meta(segment.meta().clone());
+    // 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);
+
     Ok(())
 }
 
@@ -194,11 +213,12 @@ 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)?;
+    let mut segment_writer =
+        SegmentWriter::for_segment(memory_budget, segment.clone(), &schema, tokenizers)?;
     for document_group in grouped_document_iterator {
         for doc in document_group {
             segment_writer.add_document(doc, &schema)?;
@@ -236,11 +256,7 @@ fn index_documents(
         last_docstamp,
     )?;
 
-    let segment_entry = SegmentEntry::new(
-        segment_with_max_doc.meta().clone(),
-        delete_cursor,
-        delete_bitset_opt,
-    );
+    let segment_entry = SegmentEntry::new(segment_with_max_doc, delete_cursor, delete_bitset_opt);
     block_on(segment_updater.schedule_add_segment(segment_entry))?;
     Ok(true)
 }
@@ -312,16 +328,24 @@ impl IndexWriter {
 
         let delete_queue = DeleteQueue::new();
 
-        let current_opstamp = index.load_metas()?.opstamp;
+        let meta = index.load_metas()?;
 
-        let stamper = Stamper::new(current_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 segment_updater =
-            SegmentUpdater::create(index.clone(), stamper.clone(), &delete_queue.cursor())?;
+            SegmentUpdater::create(segment_registers.clone(), index.clone(), stamper.clone())?;
 
         let mut index_writer = IndexWriter {
             _directory_lock: Some(directory_lock),
-
+            segment_registers,
             heap_size_in_bytes_per_thread,
             index: index.clone(),
 
@@ -335,10 +359,12 @@ impl IndexWriter {
 
             delete_queue,
 
-            committed_opstamp: current_opstamp,
+            committed_opstamp: meta.opstamp,
             stamper,
 
             worker_id: 0,
+
+            on_commit: Default::default(),
         };
         index_writer.start_workers()?;
         Ok(index_writer)
@@ -346,7 +372,7 @@ impl IndexWriter {
 
     fn drop_sender(&mut self) {
         let (sender, _receiver) = channel::bounded(1);
-        self.operation_sender = sender;
+        mem::replace(&mut self.operation_sender, sender);
     }
 
     /// If there are some merging threads, blocks until they all finish their work and
@@ -378,13 +404,6 @@ 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
@@ -433,12 +452,13 @@ impl IndexWriter {
                         // was dropped.
                         return Ok(());
                     }
-                    let segment = index.new_segment();
+                    let segment = index.new_segment_unpersisted();
                     index_documents(
                         mem_budget,
                         segment,
                         &mut document_iterator,
                         &mut segment_updater,
+                        index.tokenizers(),
                         delete_cursor.clone(),
                     )?;
                 }
@@ -465,6 +485,21 @@ 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,
@@ -608,7 +643,7 @@ 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) -> crate::Result<PreparedCommit> {
+    pub fn prepare_commit(&mut self, soft_commit: bool) -> crate::Result<PreparedCommit> {
         // Here, because we join all of the worker threads,
         // all of the segment update for this commit have been
         // sent.
@@ -636,7 +671,7 @@ impl IndexWriter {
         }
 
         let commit_opstamp = self.stamper.stamp();
-        let prepared_commit = PreparedCommit::new(self, commit_opstamp);
+        let prepared_commit = PreparedCommit::new(self, commit_opstamp, soft_commit);
         info!("Prepared commit {}", commit_opstamp);
         Ok(prepared_commit)
     }
@@ -656,7 +691,11 @@ impl IndexWriter {
     /// that made it in the commit.
     ///
     pub fn commit(&mut self) -> crate::Result<Opstamp> {
-        self.prepare_commit()?.commit()
+        self.prepare_commit(false)?.commit()
+    }
+
+    pub fn soft_commit(&mut self) -> crate::Result<Opstamp> {
+        self.prepare_commit(true)?.commit()
     }
 
     pub(crate) fn segment_updater(&self) -> &SegmentUpdater {
@@ -781,7 +820,6 @@ impl Drop for IndexWriter {
 
 #[cfg(test)]
 mod tests {
-
     use super::super::operation::UserOperation;
     use crate::collector::TopDocs;
     use crate::directory::error::LockError;
@@ -808,46 +846,6 @@ 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")`
@@ -1054,7 +1052,8 @@ mod tests {
                 index_writer.add_document(doc!(text_field => "a"));
             }
             {
-                let mut prepared_commit = index_writer.prepare_commit().expect("commit failed");
+                let mut prepared_commit =
+                    index_writer.prepare_commit(false).expect("commit failed");
                 prepared_commit.set_payload("first commit");
                 prepared_commit.commit().expect("commit failed");
             }
@@ -1087,7 +1086,8 @@ mod tests {
                 index_writer.add_document(doc!(text_field => "a"));
             }
             {
-                let mut prepared_commit = index_writer.prepare_commit().expect("commit failed");
+                let mut prepared_commit =
+                    index_writer.prepare_commit(false).expect("commit failed");
                 prepared_commit.set_payload("first commit");
                 prepared_commit.abort().expect("commit failed");
             }
@@ -1262,7 +1262,43 @@ 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"));
-        let commit = index_writer.commit();
-        assert!(commit.is_ok());
+        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);
     }
 }

src/indexer/log_merge_policy.rs

@@ -54,6 +54,10 @@ impl LogMergePolicy {
 
 impl MergePolicy for LogMergePolicy {
     fn compute_merge_candidates(&self, segments: &[SegmentMeta]) -> Vec<MergeCandidate> {
+        if segments.is_empty() {
+            return Vec::new();
+        }
+
         let mut size_sorted_tuples = segments
             .iter()
             .map(SegmentMeta::num_docs)
@@ -63,35 +67,27 @@ impl MergePolicy for LogMergePolicy {
 
         size_sorted_tuples.sort_by(|x, y| y.1.cmp(&(x.1)));
 
-        if size_sorted_tuples.len() <= 1 {
-            return Vec::new();
-        }
-
         let size_sorted_log_tuples: Vec<_> = size_sorted_tuples
             .into_iter()
             .map(|(ind, num_docs)| (ind, f64::from(self.clip_min_size(num_docs)).log2()))
             .collect();
 
-        if let Some(&(first_ind, first_score)) = size_sorted_log_tuples.first() {
-            let mut current_max_log_size = first_score;
-            let mut levels = vec![vec![first_ind]];
-            for &(ind, score) in (&size_sorted_log_tuples).iter().skip(1) {
-                if score < (current_max_log_size - self.level_log_size) {
-                    current_max_log_size = score;
-                    levels.push(Vec::new());
-                }
-                levels.last_mut().unwrap().push(ind);
+        let (first_ind, first_score) = size_sorted_log_tuples[0];
+        let mut current_max_log_size = first_score;
+        let mut levels = vec![vec![first_ind]];
+        for &(ind, score) in (&size_sorted_log_tuples).iter().skip(1) {
+            if score < (current_max_log_size - self.level_log_size) {
+                current_max_log_size = score;
+                levels.push(Vec::new());
             }
-            levels
-                .iter()
-                .filter(|level| level.len() >= self.min_merge_size)
-                .map(|ind_vec| {
-                    MergeCandidate(ind_vec.iter().map(|&ind| segments[ind].id()).collect())
-                })
-                .collect()
-        } else {
-            return vec![];
+            levels.last_mut().unwrap().push(ind);
         }
+
+        levels
+            .iter()
+            .filter(|level| level.len() >= self.min_merge_size)
+            .map(|ind_vec| MergeCandidate(ind_vec.iter().map(|&ind| segments[ind].id()).collect()))
+            .collect()
     }
 }
 
@@ -183,7 +179,6 @@ mod tests {
         let result_list = test_merge_policy().compute_merge_candidates(&test_input);
         assert_eq!(result_list.len(), 2);
     }
-
     #[test]
     fn test_log_merge_policy_small_segments() {
         // segments under min_layer_size are merged together
@@ -199,17 +194,6 @@ mod tests {
         assert_eq!(result_list.len(), 1);
     }
 
-    #[test]
-    fn test_log_merge_policy_all_segments_too_large_to_merge() {
-        let eight_large_segments: Vec<SegmentMeta> =
-            std::iter::repeat_with(|| create_random_segment_meta(100_001))
-                .take(8)
-                .collect();
-        assert!(test_merge_policy()
-            .compute_merge_candidates(&eight_large_segments)
-            .is_empty());
-    }
-
     #[test]
     fn test_large_merge_segments() {
         let test_input = vec![

src/indexer/merger.rs

@@ -2,7 +2,7 @@ use crate::common::MAX_DOC_LIMIT;
 use crate::core::Segment;
 use crate::core::SegmentReader;
 use crate::core::SerializableSegment;
-use crate::docset::{DocSet, TERMINATED};
+use crate::docset::DocSet;
 use crate::fastfield::BytesFastFieldReader;
 use crate::fastfield::DeleteBitSet;
 use crate::fastfield::FastFieldReader;
@@ -21,6 +21,7 @@ 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;
 
@@ -69,11 +70,11 @@ fn compute_min_max_val(
             Some(delete_bitset) => {
                 // some deleted documents,
                 // we need to recompute the max / min
-                crate::common::minmax(
-                    (0..max_doc)
-                        .filter(|doc_id| delete_bitset.is_alive(*doc_id))
-                        .map(|doc_id| u64_reader.get(doc_id)),
-                )
+                (0..max_doc)
+                    .filter(|doc_id| delete_bitset.is_alive(*doc_id))
+                    .map(|doc_id| u64_reader.get(doc_id))
+                    .minmax()
+                    .into_option()
             }
             None => {
                 // no deleted documents,
@@ -574,12 +575,10 @@ impl IndexMerger {
                     let inverted_index = segment_reader.inverted_index(indexed_field);
                     let mut segment_postings = inverted_index
                         .read_postings_from_terminfo(term_info, segment_postings_option);
-                    let mut doc = segment_postings.doc();
-                    while doc != TERMINATED {
-                        if !segment_reader.is_deleted(doc) {
+                    while segment_postings.advance() {
+                        if !segment_reader.is_deleted(segment_postings.doc()) {
                             return Some((segment_ord, segment_postings));
                         }
-                        doc = segment_postings.advance();
                     }
                     None
                 })
@@ -589,45 +588,57 @@ impl IndexMerger {
             // of all of the segments containing the given term.
             //
             // These segments are non-empty and advance has already been called.
-            if segment_postings.is_empty() {
-                continue;
-            }
-            // If not, the `term` will be entirely removed.
+            if !segment_postings.is_empty() {
+                // If not, the `term` will be entirely removed.
 
-            // We know that there is at least one document containing
-            // the term, so we add it.
-            let to_term_ord = field_serializer.new_term(term_bytes)?;
+                // We know that there is at least one document containing
+                // the term, so we add it.
+                let to_term_ord = field_serializer.new_term(term_bytes)?;
 
-            if let Some(ref mut term_ord_mapping) = term_ord_mapping_opt {
-                for (segment_ord, from_term_ord) in merged_terms.matching_segments() {
-                    term_ord_mapping.register_from_to(segment_ord, from_term_ord, to_term_ord);
-                }
-            }
-
-            // We can now serialize this postings, by pushing each document to the
-            // postings serializer.
-            for (segment_ord, mut segment_postings) in segment_postings {
-                let old_to_new_doc_id = &merged_doc_id_map[segment_ord];
-
-                let mut doc = segment_postings.doc();
-                while doc != TERMINATED {
-                    // deleted doc are skipped as they do not have a `remapped_doc_id`.
-                    if let Some(remapped_doc_id) = old_to_new_doc_id[doc as usize] {
-                        // we make sure to only write the term iff
-                        // there is at least one document.
-                        let term_freq = segment_postings.term_freq();
-                        segment_postings.positions(&mut positions_buffer);
-
-                        let delta_positions = delta_computer.compute_delta(&positions_buffer);
-                        field_serializer.write_doc(remapped_doc_id, term_freq, delta_positions)?;
+                if let Some(ref mut term_ord_mapping) = term_ord_mapping_opt {
+                    for (segment_ord, from_term_ord) in merged_terms.matching_segments() {
+                        term_ord_mapping.register_from_to(segment_ord, from_term_ord, to_term_ord);
                     }
-
-                    doc = segment_postings.advance();
                 }
-            }
 
-            // closing the term.
-            field_serializer.close_term()?;
+                // We can now serialize this postings, by pushing each document to the
+                // postings serializer.
+                for (segment_ord, mut segment_postings) in segment_postings {
+                    let old_to_new_doc_id = &merged_doc_id_map[segment_ord];
+                    loop {
+                        let doc = segment_postings.doc();
+
+                        // `.advance()` has been called once before the loop.
+                        //
+                        // It was required to make sure we only consider segments
+                        // that effectively contain at least one non-deleted document
+                        // and remove terms that do not have documents associated.
+                        //
+                        //  For this reason, we cannot use a `while segment_postings.advance()` loop.
+
+                        // deleted doc are skipped as they do not have a `remapped_doc_id`.
+                        if let Some(remapped_doc_id) = old_to_new_doc_id[doc as usize] {
+                            // we make sure to only write the term iff
+                            // there is at least one document.
+                            let term_freq = segment_postings.term_freq();
+                            segment_postings.positions(&mut positions_buffer);
+
+                            let delta_positions = delta_computer.compute_delta(&positions_buffer);
+                            field_serializer.write_doc(
+                                remapped_doc_id,
+                                term_freq,
+                                delta_positions,
+                            )?;
+                        }
+                        if !segment_postings.advance() {
+                            break;
+                        }
+                    }
+                }
+
+                // closing the term.
+                field_serializer.close_term()?;
+            }
         }
         field_serializer.close()?;
         Ok(term_ord_mapping_opt)

src/indexer/mod.rs

@@ -23,6 +23,7 @@ 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

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

src/indexer/segment_entry.rs

@@ -1,7 +1,9 @@
 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
@@ -19,55 +21,81 @@ use std::fmt;
 /// in the .del file or in the `delete_bitset`.
 #[derive(Clone)]
 pub struct SegmentEntry {
-    meta: SegmentMeta,
+    segment: Segment,
     delete_bitset: Option<BitSet>,
     delete_cursor: DeleteCursor,
 }
 
 impl SegmentEntry {
     /// Create a new `SegmentEntry`
-    pub fn new(
-        segment_meta: SegmentMeta,
+    pub(crate) fn new(
+        segment: Segment,
         delete_cursor: DeleteCursor,
         delete_bitset: Option<BitSet>,
     ) -> SegmentEntry {
         SegmentEntry {
-            meta: segment_meta,
+            segment,
             delete_bitset,
             delete_cursor,
         }
     }
 
-    /// Return a reference to the segment entry deleted bitset.
+    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.
     ///
     /// `DocId` in this bitset are flagged as deleted.
-    pub fn delete_bitset(&self) -> Option<&BitSet> {
-        self.delete_bitset.as_ref()
+    pub fn take_delete_bitset(&mut self) -> Option<BitSet> {
+        self.delete_bitset.take()
     }
 
-    /// Set the `SegmentMeta` for this segment.
-    pub fn set_meta(&mut self, segment_meta: SegmentMeta) {
-        self.meta = segment_meta;
+    /// 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;
     }
 
+    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) -> &mut DeleteCursor {
-        &mut self.delete_cursor
+    pub fn delete_cursor(&mut self) -> DeleteCursor {
+        self.delete_cursor.clone()
     }
 
     /// Returns the segment id.
     pub fn segment_id(&self) -> SegmentId {
-        self.meta.id()
+        self.segment.id()
     }
 
     /// Accessor to the `SegmentMeta`
     pub fn meta(&self) -> &SegmentMeta {
-        &self.meta
+        self.segment.meta()
     }
 }
 
 impl fmt::Debug for SegmentEntry {
     fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
-        write!(formatter, "SegmentEntry({:?})", self.meta)
+        let num_deletes = self.delete_bitset.as_ref().map(|bitset| bitset.len());
+        write!(
+            formatter,
+            "SegmentEntry(seg={:?}, ndel={:?})",
+            self.segment, num_deletes
+        )
     }
 }

src/indexer/segment_manager.rs

@@ -1,16 +1,14 @@
 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::fmt::{self, Debug, Formatter};
-use std::sync::RwLock;
+use std::sync::{Arc, RwLock};
 use std::sync::{RwLockReadGuard, RwLockWriteGuard};
 
 #[derive(Default)]
-struct SegmentRegisters {
+pub(crate) struct SegmentRegisters {
     uncommitted: SegmentRegister,
     committed: SegmentRegister,
 }
@@ -22,6 +20,17 @@ 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
@@ -44,18 +53,7 @@ impl SegmentRegisters {
 /// changes (merges especially)
 #[derive(Default)]
 pub struct SegmentManager {
-    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
-        )
-    }
+    registers: Arc<RwLock<SegmentRegisters>>,
 }
 
 pub fn get_mergeable_segments(
@@ -74,16 +72,8 @@ pub fn get_mergeable_segments(
 }
 
 impl SegmentManager {
-    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),
-            }),
-        }
+    pub(crate) fn new(registers: Arc<RwLock<SegmentRegisters>>) -> SegmentManager {
+        SegmentManager { registers }
     }
 
     /// Returns all of the segment entries (committed or uncommitted)
@@ -165,7 +155,7 @@ impl SegmentManager {
             let error_msg = "Merge operation sent for segments that are not \
                              all uncommited or commited."
                 .to_string();
-            return Err(TantivyError::InvalidArgument(error_msg));
+            return Err(crate::Error::InvalidArgument(error_msg));
         }
         Ok(segment_entries)
     }

src/indexer/segment_register.rs

@@ -1,7 +1,10 @@
 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};
@@ -46,6 +49,13 @@ 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()
     }
@@ -79,11 +89,17 @@ impl SegmentRegister {
         self.segment_states.get(segment_id).cloned()
     }
 
-    pub fn new(segment_metas: Vec<SegmentMeta>, delete_cursor: &DeleteCursor) -> SegmentRegister {
+    pub fn new(
+        directory: &ManagedDirectory,
+        schema: &Schema,
+        segment_metas: Vec<SegmentMeta>,
+        delete_cursor: &DeleteCursor,
+    ) -> SegmentRegister {
         let mut segment_states = HashMap::new();
-        for segment_meta in segment_metas {
-            let segment_id = segment_meta.id();
-            let segment_entry = SegmentEntry::new(segment_meta, delete_cursor.clone(), None);
+        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);
             segment_states.insert(segment_id, segment_entry);
         }
         SegmentRegister { segment_states }
@@ -95,6 +111,7 @@ 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
@@ -108,6 +125,7 @@ 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();
@@ -115,21 +133,24 @@ mod tests {
         let segment_id_merged = SegmentId::generate_random();
 
         {
-            let segment_meta = inventory.new_segment_meta(segment_id_a, 0u32);
-            let segment_entry = SegmentEntry::new(segment_meta, delete_queue.cursor(), None);
+            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);
             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_entry = SegmentEntry::new(segment_meta, delete_queue.cursor(), None);
+            let segment = Segment::new_volatile(segment_meta, schema.clone());
+            let segment_entry = SegmentEntry::new(segment, 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_entry = SegmentEntry::new(segment_meta_merged, delete_queue.cursor(), None);
+            let segment = Segment::new_volatile(segment_meta_merged, schema);
+            let segment_entry = SegmentEntry::new(segment, 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

@@ -7,11 +7,10 @@ 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::SegmentsStatus;
+use crate::indexer::segment_manager::{SegmentRegisters, SegmentsStatus};
 use crate::indexer::stamper::Stamper;
 use crate::indexer::SegmentEntry;
 use crate::indexer::SegmentSerializer;
@@ -23,6 +22,7 @@ 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;
@@ -116,8 +116,7 @@ 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 {
-        let segment = index.segment(segment_entry.meta().clone());
-        advance_deletes(segment, segment_entry, target_opstamp)?;
+        advance_deletes(segment_entry, target_opstamp)?;
     }
 
     let delete_cursor = segment_entries[0].delete_cursor().clone();
@@ -133,11 +132,13 @@ 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 num_docs = merger.write(segment_serializer)?;
+    let max_doc = merger.write(segment_serializer)?;
 
-    let segment_meta = index.new_segment_meta(merged_segment.id(), num_docs);
-
-    Ok(SegmentEntry::new(segment_meta, delete_cursor, None))
+    Ok(SegmentEntry::new(
+        merged_segment.with_max_doc(max_doc),
+        delete_cursor,
+        None,
+    ))
 }
 
 pub(crate) struct InnerSegmentUpdater {
@@ -161,12 +162,11 @@ 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 segments = index.searchable_segment_metas()?;
-        let segment_manager = SegmentManager::from_segments(segments, delete_cursor);
+        let segment_manager = SegmentManager::new(segment_registers);
         let pool = ThreadPoolBuilder::new()
             .name_prefix("segment_updater")
             .pool_size(1)
@@ -233,6 +233,7 @@ 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);
@@ -261,8 +262,7 @@ 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 {
-            let segment = self.index.segment(segment_entry.meta().clone());
-            advance_deletes(segment, segment_entry, target_opstamp)?;
+            advance_deletes(segment_entry, target_opstamp)?;
         }
         Ok(segment_entries)
     }
@@ -330,12 +330,21 @@ 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 segment_entries = segment_updater.purge_deletes(opstamp)?;
+            let mut segment_entries = segment_updater.purge_deletes(opstamp)?;
+            if !soft_commit {
+                for segment_entry in &mut segment_entries {
+                    segment_entry.persist(directory.clone())?;
+                }
+            }
             segment_updater.segment_manager.commit(segment_entries);
-            segment_updater.save_metas(opstamp, payload)?;
+            if !soft_commit {
+                segment_updater.save_metas(opstamp, payload)?;
+            }
             let _ = garbage_collect_files(segment_updater.clone()).await;
             segment_updater.consider_merge_options().await;
             Ok(())
@@ -473,17 +482,14 @@ 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().clone();
+                let mut delete_cursor = after_merge_segment_entry.delete_cursor();
                 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;
-                        let segment = index.segment(after_merge_segment_entry.meta().clone());
-                        if let Err(e) = advance_deletes(
-                            segment,
-                            &mut after_merge_segment_entry,
-                            committed_opstamp,
-                        ) {
+                        let _index = &segment_updater.index;
+                        if let Err(e) =
+                            advance_deletes(&mut after_merge_segment_entry, committed_opstamp)
+                        {
                             error!(
                                 "Merge of {:?} was cancelled (advancing deletes failed): {:?}",
                                 merge_operation.segment_ids(),

src/indexer/segment_writer.rs

@@ -11,8 +11,9 @@ use crate::schema::Schema;
 use crate::schema::Term;
 use crate::schema::Value;
 use crate::schema::{Field, FieldEntry};
-use crate::tokenizer::{BoxTokenStream, PreTokenizedStream};
-use crate::tokenizer::{FacetTokenizer, TextAnalyzer};
+use crate::tokenizer::TokenizerManager;
+use crate::tokenizer::{BoxTokenStream, FacetTokenizer};
+use crate::tokenizer::{PreTokenizedStream, TextAnalyzer};
 use crate::tokenizer::{TokenStreamChain, Tokenizer};
 use crate::DocId;
 use crate::Opstamp;
@@ -64,10 +65,12 @@ 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(
@@ -76,7 +79,7 @@ impl SegmentWriter {
                         .get_indexing_options()
                         .and_then(|text_index_option| {
                             let tokenizer_name = &text_index_option.tokenizer();
-                            segment.index().tokenizers().get(tokenizer_name)
+                            tokenizer_manager.get(tokenizer_name)
                         }),
                     _ => None,
                 },
@@ -85,9 +88,9 @@ 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

@@ -98,6 +98,9 @@
 //! [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;
 
@@ -118,6 +121,7 @@ mod functional_test;
 mod macros;
 
 pub use crate::error::TantivyError;
+pub use crate::error::TantivyError as Error;
 pub use chrono;
 
 /// Tantivy result.
@@ -156,7 +160,7 @@ mod snippet;
 pub use self::snippet::{Snippet, SnippetGenerator};
 
 mod docset;
-pub use self::docset::{DocSet, TERMINATED};
+pub use self::docset::{DocSet, SkipResult};
 pub use crate::common::{f64_to_u64, i64_to_u64, u64_to_f64, u64_to_i64};
 pub use crate::core::{Executor, SegmentComponent};
 pub use crate::core::{Index, IndexMeta, Searcher, Segment, SegmentId, SegmentMeta};
@@ -170,7 +174,6 @@ 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;
@@ -285,7 +288,7 @@ mod tests {
 
     use crate::collector::tests::TEST_COLLECTOR_WITH_SCORE;
     use crate::core::SegmentReader;
-    use crate::docset::{DocSet, TERMINATED};
+    use crate::docset::DocSet;
     use crate::query::BooleanQuery;
     use crate::schema::*;
     use crate::DocAddress;
@@ -381,12 +384,19 @@ mod tests {
             index_writer.commit().unwrap();
         }
         {
-            index_writer.add_document(doc!(text_field=>"a"));
-            index_writer.add_document(doc!(text_field=>"a a"));
+            {
+                let doc = doc!(text_field=>"a");
+                index_writer.add_document(doc);
+            }
+            {
+                let doc = doc!(text_field=>"a a");
+                index_writer.add_document(doc);
+            }
             index_writer.commit().unwrap();
         }
         {
-            index_writer.add_document(doc!(text_field=>"c"));
+            let doc = doc!(text_field=>"c");
+            index_writer.add_document(doc);
             index_writer.commit().unwrap();
         }
         {
@@ -465,12 +475,10 @@ mod tests {
     }
 
     fn advance_undeleted(docset: &mut dyn DocSet, reader: &SegmentReader) -> bool {
-        let mut doc = docset.advance();
-        while doc != TERMINATED {
-            if !reader.is_deleted(doc) {
+        while docset.advance() {
+            if !reader.is_deleted(docset.doc()) {
                 return true;
             }
-            doc = docset.advance();
         }
         false
     }
@@ -636,8 +644,9 @@ mod tests {
             .inverted_index(term.field())
             .read_postings(&term, IndexRecordOption::Basic)
             .unwrap();
+        assert!(postings.advance());
         assert_eq!(postings.doc(), 0);
-        assert_eq!(postings.advance(), TERMINATED);
+        assert!(!postings.advance());
     }
 
     #[test]
@@ -659,8 +668,9 @@ mod tests {
             .inverted_index(term.field())
             .read_postings(&term, IndexRecordOption::Basic)
             .unwrap();
+        assert!(postings.advance());
         assert_eq!(postings.doc(), 0);
-        assert_eq!(postings.advance(), TERMINATED);
+        assert!(!postings.advance());
     }
 
     #[test]
@@ -682,8 +692,9 @@ mod tests {
             .inverted_index(term.field())
             .read_postings(&term, IndexRecordOption::Basic)
             .unwrap();
+        assert!(postings.advance());
         assert_eq!(postings.doc(), 0);
-        assert_eq!(postings.advance(), TERMINATED);
+        assert!(!postings.advance());
     }
 
     #[test]
@@ -752,8 +763,10 @@ mod tests {
         {
             // writing the segment
             let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-            let doc = doc!(text_field=>"af af af bc bc");
-            index_writer.add_document(doc);
+            {
+                let doc = doc!(text_field=>"af af af bc bc");
+                index_writer.add_document(doc);
+            }
             index_writer.commit().unwrap();
         }
         {
@@ -769,9 +782,10 @@ mod tests {
             let mut postings = inverted_index
                 .read_postings(&term_af, IndexRecordOption::WithFreqsAndPositions)
                 .unwrap();
+            assert!(postings.advance());
             assert_eq!(postings.doc(), 0);
             assert_eq!(postings.term_freq(), 3);
-            assert_eq!(postings.advance(), TERMINATED);
+            assert!(!postings.advance());
         }
     }
 

src/positions/mod.rs

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

src/positions/reader.rs

@@ -3,6 +3,7 @@ use crate::directory::ReadOnlySource;
 use crate::positions::COMPRESSION_BLOCK_SIZE;
 use crate::positions::LONG_SKIP_INTERVAL;
 use crate::positions::LONG_SKIP_IN_BLOCKS;
+use crate::postings::compression::compressed_block_size;
 /// Positions works as a long sequence of compressed block.
 /// All terms are chained one after the other.
 ///
@@ -61,20 +62,22 @@ impl Positions {
 
     fn reader(&self, offset: u64) -> PositionReader {
         let long_skip_id = (offset / LONG_SKIP_INTERVAL) as usize;
+        let small_skip = (offset % LONG_SKIP_INTERVAL) as usize;
         let offset_num_bytes: u64 = self.long_skip(long_skip_id);
         let mut position_read = OwnedRead::new(self.position_source.clone());
         position_read.advance(offset_num_bytes as usize);
         let mut skip_read = OwnedRead::new(self.skip_source.clone());
         skip_read.advance(long_skip_id * LONG_SKIP_IN_BLOCKS);
-        PositionReader {
+        let mut position_reader = PositionReader {
             bit_packer: self.bit_packer,
             skip_read,
             position_read,
+            inner_offset: 0,
             buffer: Box::new([0u32; 128]),
-            block_offset: std::i64::MAX as u64,
-            anchor_offset: (long_skip_id as u64) * LONG_SKIP_INTERVAL,
-            abs_offset: offset,
-        }
+            ahead: None,
+        };
+        position_reader.skip(small_skip);
+        position_reader
     }
 }
 
@@ -82,12 +85,51 @@ pub struct PositionReader {
     skip_read: OwnedRead,
     position_read: OwnedRead,
     bit_packer: BitPacker4x,
-    buffer: Box<[u32; COMPRESSION_BLOCK_SIZE]>,
+    inner_offset: usize,
+    buffer: Box<[u32; 128]>,
+    ahead: Option<usize>, // if None, no block is loaded.
+                          // if Some(num_blocks), the block currently loaded is num_blocks ahead
+                          // of the block of the next int to read.
+}
 
-    block_offset: u64,
-    anchor_offset: u64,
-
-    abs_offset: u64,
+// `ahead` represents the offset of the block currently loaded
+// compared to the cursor of the actual stream.
+//
+// By contract, when this function is called, the current block has to be
+// decompressed.
+//
+// If the requested number of els ends exactly at a given block, the next
+// block is not decompressed.
+fn read_impl(
+    bit_packer: BitPacker4x,
+    mut position: &[u8],
+    buffer: &mut [u32; 128],
+    mut inner_offset: usize,
+    num_bits: &[u8],
+    output: &mut [u32],
+) -> usize {
+    let mut output_start = 0;
+    let mut output_len = output.len();
+    let mut ahead = 0;
+    loop {
+        let available_len = COMPRESSION_BLOCK_SIZE - inner_offset;
+        // We have enough elements in the current block.
+        // Let's copy the requested elements in the output buffer,
+        // and return.
+        if output_len <= available_len {
+            output[output_start..].copy_from_slice(&buffer[inner_offset..][..output_len]);
+            return ahead;
+        }
+        output[output_start..][..available_len].copy_from_slice(&buffer[inner_offset..]);
+        output_len -= available_len;
+        output_start += available_len;
+        inner_offset = 0;
+        let num_bits = num_bits[ahead];
+        bit_packer.decompress(position, &mut buffer[..], num_bits);
+        let block_len = compressed_block_size(num_bits);
+        position = &position[block_len..];
+        ahead += 1;
+    }
 }
 
 impl PositionReader {
@@ -99,65 +141,57 @@ impl PositionReader {
         Positions::new(position_source, skip_source).reader(offset)
     }
 
-    fn advance_num_blocks(&mut self, num_blocks: usize) {
-        let num_bits: usize = self.skip_read.as_ref()[..num_blocks]
-            .iter()
-            .cloned()
-            .map(|num_bits| num_bits as usize)
-            .sum();
-        let num_bytes_to_skip = num_bits * COMPRESSION_BLOCK_SIZE / 8;
-        self.skip_read.advance(num_blocks as usize);
-        self.position_read.advance(num_bytes_to_skip);
-    }
-
-    /// Fills a buffer with the positions `[offset..offset+output.len())` integers.
-    ///
-    /// `offset` is required to have a value >= to the offsets given in previous calls
-    /// for the given `PositionReaderAbsolute` instance.
-    pub fn read(&mut self, mut offset: u64, mut output: &mut [u32]) {
-        offset += self.abs_offset;
-        assert!(
-            offset >= self.anchor_offset,
-            "offset arguments should be increasing."
-        );
-        let delta_to_block_offset = offset as i64 - self.block_offset as i64;
-        if delta_to_block_offset < 0 || delta_to_block_offset >= 128 {
-            // The first position is not within the first block.
-            // We need to decompress the first block.
-            let delta_to_anchor_offset = offset - self.anchor_offset;
-            let num_blocks_to_skip =
-                (delta_to_anchor_offset / (COMPRESSION_BLOCK_SIZE as u64)) as usize;
-            self.advance_num_blocks(num_blocks_to_skip);
-            self.anchor_offset = offset - (offset % COMPRESSION_BLOCK_SIZE as u64);
-            self.block_offset = self.anchor_offset;
-            let num_bits = self.skip_read.get(0);
-            self.bit_packer
-                .decompress(self.position_read.as_ref(), self.buffer.as_mut(), num_bits);
-        } else {
-            let num_blocks_to_skip =
-                ((self.block_offset - self.anchor_offset) / COMPRESSION_BLOCK_SIZE as u64) as usize;
-            self.advance_num_blocks(num_blocks_to_skip);
-            self.anchor_offset = self.block_offset;
-        }
-
-        let mut num_bits = self.skip_read.get(0);
-        let mut position_data = self.position_read.as_ref();
-
-        for i in 1.. {
-            let offset_in_block = (offset as usize) % COMPRESSION_BLOCK_SIZE;
-            let remaining_in_block = COMPRESSION_BLOCK_SIZE - offset_in_block;
-            if remaining_in_block >= output.len() {
-                output.copy_from_slice(&self.buffer[offset_in_block..][..output.len()]);
-                break;
-            }
-            output[..remaining_in_block].copy_from_slice(&self.buffer[offset_in_block..]);
-            output = &mut output[remaining_in_block..];
-            offset += remaining_in_block as u64;
-            position_data = &position_data[(num_bits as usize * COMPRESSION_BLOCK_SIZE / 8)..];
-            num_bits = self.skip_read.get(i);
+    /// Fills a buffer with the next `output.len()` integers.
+    /// This does not consume / advance the stream.
+    pub fn read(&mut self, output: &mut [u32]) {
+        let skip_data = self.skip_read.as_ref();
+        let position_data = self.position_read.as_ref();
+        let num_bits = self.skip_read.get(0);
+        if self.ahead != Some(0) {
+            // the block currently available is not the block
+            // for the current position
             self.bit_packer
                 .decompress(position_data, self.buffer.as_mut(), num_bits);
-            self.block_offset += COMPRESSION_BLOCK_SIZE as u64;
+            self.ahead = Some(0);
         }
+        let block_len = compressed_block_size(num_bits);
+        self.ahead = Some(read_impl(
+            self.bit_packer,
+            &position_data[block_len..],
+            self.buffer.as_mut(),
+            self.inner_offset,
+            &skip_data[1..],
+            output,
+        ));
+    }
+
+    /// Skip the next `skip_len` integer.
+    ///
+    /// If a full block is skipped, calling
+    /// `.skip(...)` will avoid decompressing it.
+    ///
+    /// May panic if the end of the stream is reached.
+    pub fn skip(&mut self, skip_len: usize) {
+        let skip_len_plus_inner_offset = skip_len + self.inner_offset;
+
+        let num_blocks_to_advance = skip_len_plus_inner_offset / COMPRESSION_BLOCK_SIZE;
+        self.inner_offset = skip_len_plus_inner_offset % COMPRESSION_BLOCK_SIZE;
+
+        self.ahead = self.ahead.and_then(|num_blocks| {
+            if num_blocks >= num_blocks_to_advance {
+                Some(num_blocks - num_blocks_to_advance)
+            } else {
+                None
+            }
+        });
+
+        let skip_len_in_bits = self.skip_read.as_ref()[..num_blocks_to_advance]
+            .iter()
+            .map(|num_bits| *num_bits as usize)
+            .sum::<usize>()
+            * COMPRESSION_BLOCK_SIZE;
+        let skip_len_in_bytes = skip_len_in_bits / 8;
+        self.skip_read.advance(num_blocks_to_advance);
+        self.position_read.advance(skip_len_in_bytes);
     }
 }

src/postings/block_search.rs

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

src/postings/block_segment_postings.rs

@@ -1,427 +0,0 @@
-use crate::common::{BinarySerializable, VInt};
-use crate::directory::ReadOnlySource;
-use crate::postings::compression::{
-    AlignedBuffer, BlockDecoder, VIntDecoder, COMPRESSION_BLOCK_SIZE,
-};
-use crate::postings::{BlockInfo, FreqReadingOption, SkipReader};
-use crate::schema::IndexRecordOption;
-use crate::{DocId, TERMINATED};
-
-/// `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 {
-    pub(crate) doc_decoder: BlockDecoder,
-    loaded_offset: usize,
-    freq_decoder: BlockDecoder,
-    freq_reading_option: FreqReadingOption,
-
-    doc_freq: usize,
-
-    data: ReadOnlySource,
-    skip_reader: SkipReader,
-}
-
-fn decode_bitpacked_block(
-    doc_decoder: &mut BlockDecoder,
-    freq_decoder_opt: Option<&mut BlockDecoder>,
-    data: &[u8],
-    doc_offset: DocId,
-    doc_num_bits: u8,
-    tf_num_bits: u8,
-) {
-    let num_consumed_bytes = doc_decoder.uncompress_block_sorted(data, doc_offset, doc_num_bits);
-    if let Some(freq_decoder) = freq_decoder_opt {
-        freq_decoder.uncompress_block_unsorted(&data[num_consumed_bytes..], tf_num_bits);
-    }
-}
-
-fn decode_vint_block(
-    doc_decoder: &mut BlockDecoder,
-    freq_decoder_opt: Option<&mut BlockDecoder>,
-    data: &[u8],
-    doc_offset: DocId,
-    num_vint_docs: usize,
-) {
-    doc_decoder.clear();
-    let num_consumed_bytes = doc_decoder.uncompress_vint_sorted(data, doc_offset, num_vint_docs);
-    if let Some(freq_decoder) = freq_decoder_opt {
-        freq_decoder.uncompress_vint_unsorted(&data[num_consumed_bytes..], num_vint_docs);
-    }
-}
-
-fn split_into_skips_and_postings(
-    doc_freq: u32,
-    data: ReadOnlySource,
-) -> (Option<ReadOnlySource>, ReadOnlySource) {
-    if doc_freq < COMPRESSION_BLOCK_SIZE as u32 {
-        return (None, data);
-    }
-    let mut data_byte_arr = data.as_slice();
-    let skip_len = VInt::deserialize(&mut data_byte_arr)
-        .expect("Data corrupted")
-        .0 as usize;
-    let vint_len = data.len() - data_byte_arr.len();
-    let (skip_data, postings_data) = data.slice_from(vint_len).split(skip_len);
-    (Some(skip_data), postings_data)
-}
-
-impl BlockSegmentPostings {
-    pub(crate) fn from_data(
-        doc_freq: u32,
-        data: ReadOnlySource,
-        record_option: IndexRecordOption,
-        requested_option: IndexRecordOption,
-    ) -> BlockSegmentPostings {
-        let freq_reading_option = match (record_option, requested_option) {
-            (IndexRecordOption::Basic, _) => FreqReadingOption::NoFreq,
-            (_, IndexRecordOption::Basic) => FreqReadingOption::SkipFreq,
-            (_, _) => FreqReadingOption::ReadFreq,
-        };
-
-        let (skip_data_opt, postings_data) = split_into_skips_and_postings(doc_freq, data);
-        let skip_reader = match skip_data_opt {
-            Some(skip_data) => SkipReader::new(skip_data, doc_freq, record_option),
-            None => SkipReader::new(ReadOnlySource::empty(), doc_freq, record_option),
-        };
-
-        let doc_freq = doc_freq as usize;
-        let mut block_segment_postings = BlockSegmentPostings {
-            doc_decoder: BlockDecoder::with_val(TERMINATED),
-            loaded_offset: std::usize::MAX,
-            freq_decoder: BlockDecoder::with_val(1),
-            freq_reading_option,
-            doc_freq,
-            data: postings_data,
-            skip_reader,
-        };
-        block_segment_postings.advance();
-        block_segment_postings
-    }
-
-    // Resets the block segment postings on another position
-    // in the postings file.
-    //
-    // This is useful for enumerating through a list of terms,
-    // and consuming the associated posting lists while avoiding
-    // reallocating a `BlockSegmentPostings`.
-    //
-    // # Warning
-    //
-    // This does not reset the positions list.
-    pub(crate) fn reset(&mut self, doc_freq: u32, postings_data: ReadOnlySource) {
-        let (skip_data_opt, postings_data) = split_into_skips_and_postings(doc_freq, postings_data);
-        self.data = ReadOnlySource::new(postings_data);
-        self.loaded_offset = std::usize::MAX;
-        self.loaded_offset = std::usize::MAX;
-        if let Some(skip_data) = skip_data_opt {
-            self.skip_reader.reset(skip_data, doc_freq);
-        } else {
-            self.skip_reader.reset(ReadOnlySource::empty(), doc_freq);
-        }
-        self.doc_freq = doc_freq 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()
-    }
-
-    #[inline(always)]
-    pub(crate) fn docs_aligned(&self) -> &AlignedBuffer {
-        self.doc_decoder.output_aligned()
-    }
-
-    /// Return the document at index `idx` of the block.
-    #[inline(always)]
-    pub fn doc(&self, idx: usize) -> u32 {
-        self.doc_decoder.output(idx)
-    }
-
-    /// Return the array of `term freq` in the block.
-    #[inline]
-    pub fn freqs(&self) -> &[u32] {
-        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 fn block_len(&self) -> usize {
-        self.doc_decoder.output_len
-    }
-
-    pub(crate) fn position_offset(&self) -> u64 {
-        self.skip_reader.position_offset()
-    }
-
-    /// Position on a block that may contains `target_doc`.
-    ///
-    /// If all docs are smaller than target, the block loaded may be empty,
-    /// or be the last an incomplete VInt block.
-    pub fn seek(&mut self, target_doc: DocId) {
-        self.skip_reader.seek(target_doc);
-        self.load_block();
-    }
-
-    fn load_block(&mut self) {
-        let offset = self.skip_reader.byte_offset();
-        if self.loaded_offset == offset {
-            return;
-        }
-        self.loaded_offset = offset;
-        match self.skip_reader.block_info() {
-            BlockInfo::BitPacked {
-                doc_num_bits,
-                tf_num_bits,
-                ..
-            } => {
-                decode_bitpacked_block(
-                    &mut self.doc_decoder,
-                    if let FreqReadingOption::ReadFreq = self.freq_reading_option {
-                        Some(&mut self.freq_decoder)
-                    } else {
-                        None
-                    },
-                    &self.data.as_slice()[offset..],
-                    self.skip_reader.last_doc_in_previous_block,
-                    doc_num_bits,
-                    tf_num_bits,
-                );
-            }
-            BlockInfo::VInt(num_vint_docs) => {
-                decode_vint_block(
-                    &mut self.doc_decoder,
-                    if let FreqReadingOption::ReadFreq = self.freq_reading_option {
-                        Some(&mut self.freq_decoder)
-                    } else {
-                        None
-                    },
-                    &self.data.as_slice()[offset..],
-                    self.skip_reader.last_doc_in_previous_block,
-                    num_vint_docs as usize,
-                );
-            }
-        }
-    }
-
-    /// Advance to the next block.
-    ///
-    /// Returns false iff there was no remaining blocks.
-    pub fn advance(&mut self) -> bool {
-        if !self.skip_reader.advance() {
-            return false;
-        }
-        self.load_block();
-        true
-    }
-
-    /// Returns an empty segment postings object
-    pub fn empty() -> BlockSegmentPostings {
-        BlockSegmentPostings {
-            doc_decoder: BlockDecoder::with_val(TERMINATED),
-            loaded_offset: std::usize::MAX,
-            freq_decoder: BlockDecoder::with_val(1),
-            freq_reading_option: FreqReadingOption::NoFreq,
-            doc_freq: 0,
-            data: ReadOnlySource::new(vec![]),
-            skip_reader: SkipReader::new(ReadOnlySource::new(vec![]), 0, IndexRecordOption::Basic),
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::BlockSegmentPostings;
-    use crate::common::HasLen;
-    use crate::core::Index;
-    use crate::docset::{DocSet, TERMINATED};
-    use crate::postings::compression::COMPRESSION_BLOCK_SIZE;
-    use crate::postings::postings::Postings;
-    use crate::postings::SegmentPostings;
-    use crate::schema::IndexRecordOption;
-    use crate::schema::Schema;
-    use crate::schema::Term;
-    use crate::schema::INDEXED;
-    use crate::DocId;
-
-    #[test]
-    fn test_empty_segment_postings() {
-        let mut postings = SegmentPostings::empty();
-        assert_eq!(postings.advance(), TERMINATED);
-        assert_eq!(postings.advance(), TERMINATED);
-        assert_eq!(postings.len(), 0);
-    }
-
-    #[test]
-    fn test_empty_postings_doc_returns_terminated() {
-        let mut postings = SegmentPostings::empty();
-        assert_eq!(postings.doc(), TERMINATED);
-        assert_eq!(postings.advance(), TERMINATED);
-    }
-
-    #[test]
-    fn test_empty_postings_doc_term_freq_returns_0() {
-        let postings = SegmentPostings::empty();
-        assert_eq!(postings.term_freq(), 1);
-    }
-
-    #[test]
-    fn test_empty_block_segment_postings() {
-        let mut postings = BlockSegmentPostings::empty();
-        assert!(!postings.advance());
-        assert_eq!(postings.doc_freq(), 0);
-    }
-
-    #[test]
-    fn test_block_segment_postings() {
-        let mut block_segments = build_block_postings(&(0..100_000).collect::<Vec<u32>>());
-        let mut offset: u32 = 0u32;
-        // checking that the `doc_freq` is correct
-        assert_eq!(block_segments.doc_freq(), 100_000);
-        loop {
-            let block = block_segments.docs();
-            for (i, doc) in block.iter().cloned().enumerate() {
-                assert_eq!(offset + (i as u32), doc);
-            }
-            offset += block.len() as u32;
-            if block_segments.advance() {
-                break;
-            }
-        }
-    }
-
-    #[test]
-    fn test_skip_right_at_new_block() {
-        let mut doc_ids = (0..128).collect::<Vec<u32>>();
-        // 128 is missing
-        doc_ids.push(129);
-        doc_ids.push(130);
-        {
-            let block_segments = build_block_postings(&doc_ids);
-            let mut docset = SegmentPostings::from_block_postings(block_segments, None);
-            assert_eq!(docset.seek(128), 129);
-            assert_eq!(docset.doc(), 129);
-            assert_eq!(docset.advance(), 130);
-            assert_eq!(docset.doc(), 130);
-            assert_eq!(docset.advance(), TERMINATED);
-        }
-        {
-            let block_segments = build_block_postings(&doc_ids);
-            let mut docset = SegmentPostings::from_block_postings(block_segments, None);
-            assert_eq!(docset.seek(129), 129);
-            assert_eq!(docset.doc(), 129);
-            assert_eq!(docset.advance(), 130);
-            assert_eq!(docset.doc(), 130);
-            assert_eq!(docset.advance(), TERMINATED);
-        }
-        {
-            let block_segments = build_block_postings(&doc_ids);
-            let mut docset = SegmentPostings::from_block_postings(block_segments, None);
-            assert_eq!(docset.doc(), 0);
-            assert_eq!(docset.seek(131), TERMINATED);
-            assert_eq!(docset.doc(), TERMINATED);
-        }
-    }
-
-    fn build_block_postings(docs: &[DocId]) -> BlockSegmentPostings {
-        let mut schema_builder = Schema::builder();
-        let int_field = schema_builder.add_u64_field("id", INDEXED);
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-        let mut last_doc = 0u32;
-        for &doc in docs {
-            for _ in last_doc..doc {
-                index_writer.add_document(doc!(int_field=>1u64));
-            }
-            index_writer.add_document(doc!(int_field=>0u64));
-            last_doc = doc + 1;
-        }
-        index_writer.commit().unwrap();
-        let searcher = index.reader().unwrap().searcher();
-        let segment_reader = searcher.segment_reader(0);
-        let inverted_index = segment_reader.inverted_index(int_field);
-        let term = Term::from_field_u64(int_field, 0u64);
-        let term_info = inverted_index.get_term_info(&term).unwrap();
-        inverted_index.read_block_postings_from_terminfo(&term_info, IndexRecordOption::Basic)
-    }
-
-    #[test]
-    fn test_block_segment_postings_skip2() {
-        let mut docs = vec![0];
-        for i in 0..1300 {
-            docs.push((i * i / 100) + i);
-        }
-        let mut block_postings = build_block_postings(&docs[..]);
-        for i in vec![0, 424, 10000] {
-            block_postings.seek(i);
-            let docs = block_postings.docs();
-            assert!(docs[0] <= i);
-            assert!(docs.last().cloned().unwrap_or(0u32) >= i);
-        }
-        block_postings.seek(100_000);
-        assert_eq!(block_postings.doc(COMPRESSION_BLOCK_SIZE - 1), TERMINATED);
-    }
-
-    #[test]
-    fn test_reset_block_segment_postings() {
-        let mut schema_builder = Schema::builder();
-        let int_field = schema_builder.add_u64_field("id", INDEXED);
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-        // create two postings list, one containg even number,
-        // the other containing odd numbers.
-        for i in 0..6 {
-            let doc = doc!(int_field=> (i % 2) as u64);
-            index_writer.add_document(doc);
-        }
-        index_writer.commit().unwrap();
-        let searcher = index.reader().unwrap().searcher();
-        let segment_reader = searcher.segment_reader(0);
-
-        let mut block_segments;
-        {
-            let term = Term::from_field_u64(int_field, 0u64);
-            let inverted_index = segment_reader.inverted_index(int_field);
-            let term_info = inverted_index.get_term_info(&term).unwrap();
-            block_segments = inverted_index
-                .read_block_postings_from_terminfo(&term_info, IndexRecordOption::Basic);
-        }
-        assert_eq!(block_segments.docs(), &[0, 2, 4]);
-        {
-            let term = Term::from_field_u64(int_field, 1u64);
-            let inverted_index = segment_reader.inverted_index(int_field);
-            let term_info = inverted_index.get_term_info(&term).unwrap();
-            inverted_index.reset_block_postings_from_terminfo(&term_info, &mut block_segments);
-        }
-        assert!(block_segments.advance());
-        assert_eq!(block_segments.docs(), &[1, 3, 5]);
-    }
-}

src/postings/compression/mod.rs

@@ -1,5 +1,4 @@
 use crate::common::FixedSize;
-use crate::docset::TERMINATED;
 use bitpacking::{BitPacker, BitPacker4x};
 
 pub const COMPRESSION_BLOCK_SIZE: usize = BitPacker4x::BLOCK_LEN;
@@ -18,12 +17,6 @@ pub struct BlockEncoder {
     pub output_len: usize,
 }
 
-impl Default for BlockEncoder {
-    fn default() -> Self {
-        BlockEncoder::new()
-    }
-}
-
 impl BlockEncoder {
     pub fn new() -> BlockEncoder {
         BlockEncoder {
@@ -61,13 +54,11 @@ pub struct BlockDecoder {
     pub output_len: usize,
 }
 
-impl Default for BlockDecoder {
-    fn default() -> Self {
+impl BlockDecoder {
+    pub fn new() -> BlockDecoder {
         BlockDecoder::with_val(0u32)
     }
-}
 
-impl BlockDecoder {
     pub fn with_val(val: u32) -> BlockDecoder {
         BlockDecoder {
             bitpacker: BitPacker4x::new(),
@@ -99,18 +90,14 @@ impl BlockDecoder {
     }
 
     #[inline]
-    pub(crate) fn output_aligned(&self) -> &AlignedBuffer {
-        &self.output
+    pub(crate) fn output_aligned(&self) -> (&AlignedBuffer, usize) {
+        (&self.output, self.output_len)
     }
 
     #[inline]
     pub fn output(&self, idx: usize) -> u32 {
         self.output.0[idx]
     }
-
-    pub fn clear(&mut self) {
-        self.output.0.iter_mut().for_each(|el| *el = TERMINATED);
-    }
 }
 
 pub trait VIntEncoder {
@@ -147,9 +134,9 @@ pub trait VIntDecoder {
     /// For instance, if delta encoded are `1, 3, 9`, and the
     /// `offset` is 5, then the output will be:
     /// `5 + 1 = 6, 6 + 3= 9, 9 + 9 = 18`
-    fn uncompress_vint_sorted(
+    fn uncompress_vint_sorted<'a>(
         &mut self,
-        compressed_data: &[u8],
+        compressed_data: &'a [u8],
         offset: u32,
         num_els: usize,
     ) -> usize;
@@ -159,7 +146,7 @@ pub trait VIntDecoder {
     ///
     /// The method takes a number of int to decompress, and returns
     /// the amount of bytes that were read to decompress them.
-    fn uncompress_vint_unsorted(&mut self, compressed_data: &[u8], num_els: usize) -> usize;
+    fn uncompress_vint_unsorted<'a>(&mut self, compressed_data: &'a [u8], num_els: usize) -> usize;
 }
 
 impl VIntEncoder for BlockEncoder {
@@ -173,9 +160,9 @@ impl VIntEncoder for BlockEncoder {
 }
 
 impl VIntDecoder for BlockDecoder {
-    fn uncompress_vint_sorted(
+    fn uncompress_vint_sorted<'a>(
         &mut self,
-        compressed_data: &[u8],
+        compressed_data: &'a [u8],
         offset: u32,
         num_els: usize,
     ) -> usize {
@@ -183,7 +170,7 @@ impl VIntDecoder for BlockDecoder {
         vint::uncompress_sorted(compressed_data, &mut self.output.0[..num_els], offset)
     }
 
-    fn uncompress_vint_unsorted(&mut self, compressed_data: &[u8], num_els: usize) -> usize {
+    fn uncompress_vint_unsorted<'a>(&mut self, compressed_data: &'a [u8], num_els: usize) -> usize {
         self.output_len = num_els;
         vint::uncompress_unsorted(compressed_data, &mut self.output.0[..num_els])
     }
@@ -199,7 +186,7 @@ pub mod tests {
         let vals: Vec<u32> = (0u32..128u32).map(|i| i * 7).collect();
         let mut encoder = BlockEncoder::new();
         let (num_bits, compressed_data) = encoder.compress_block_sorted(&vals, 0);
-        let mut decoder = BlockDecoder::default();
+        let mut decoder = BlockDecoder::new();
         {
             let consumed_num_bytes = decoder.uncompress_block_sorted(compressed_data, 0, num_bits);
             assert_eq!(consumed_num_bytes, compressed_data.len());
@@ -212,9 +199,9 @@ pub mod tests {
     #[test]
     fn test_encode_sorted_block_with_offset() {
         let vals: Vec<u32> = (0u32..128u32).map(|i| 11 + i * 7).collect();
-        let mut encoder = BlockEncoder::default();
+        let mut encoder = BlockEncoder::new();
         let (num_bits, compressed_data) = encoder.compress_block_sorted(&vals, 10);
-        let mut decoder = BlockDecoder::default();
+        let mut decoder = BlockDecoder::new();
         {
             let consumed_num_bytes = decoder.uncompress_block_sorted(compressed_data, 10, num_bits);
             assert_eq!(consumed_num_bytes, compressed_data.len());
@@ -229,11 +216,11 @@ pub mod tests {
         let mut compressed: Vec<u8> = Vec::new();
         let n = 128;
         let vals: Vec<u32> = (0..n).map(|i| 11u32 + (i as u32) * 7u32).collect();
-        let mut encoder = BlockEncoder::default();
+        let mut encoder = BlockEncoder::new();
         let (num_bits, compressed_data) = encoder.compress_block_sorted(&vals, 10);
         compressed.extend_from_slice(compressed_data);
         compressed.push(173u8);
-        let mut decoder = BlockDecoder::default();
+        let mut decoder = BlockDecoder::new();
         {
             let consumed_num_bytes = decoder.uncompress_block_sorted(&compressed, 10, num_bits);
             assert_eq!(consumed_num_bytes, compressed.len() - 1);
@@ -249,11 +236,11 @@ pub mod tests {
         let mut compressed: Vec<u8> = Vec::new();
         let n = 128;
         let vals: Vec<u32> = (0..n).map(|i| 11u32 + (i as u32) * 7u32 % 12).collect();
-        let mut encoder = BlockEncoder::default();
+        let mut encoder = BlockEncoder::new();
         let (num_bits, compressed_data) = encoder.compress_block_unsorted(&vals);
         compressed.extend_from_slice(compressed_data);
         compressed.push(173u8);
-        let mut decoder = BlockDecoder::default();
+        let mut decoder = BlockDecoder::new();
         {
             let consumed_num_bytes = decoder.uncompress_block_unsorted(&compressed, num_bits);
             assert_eq!(consumed_num_bytes + 1, compressed.len());
@@ -264,11 +251,6 @@ pub mod tests {
         }
     }
 
-    #[test]
-    fn test_block_decoder_initialization() {
-        let block = BlockDecoder::with_val(TERMINATED);
-        assert_eq!(block.output(0), TERMINATED);
-    }
     #[test]
     fn test_encode_vint() {
         {
@@ -278,7 +260,7 @@ pub mod tests {
             for offset in &[0u32, 1u32, 2u32] {
                 let encoded_data = encoder.compress_vint_sorted(&input, *offset);
                 assert!(encoded_data.len() <= expected_length);
-                let mut decoder = BlockDecoder::default();
+                let mut decoder = BlockDecoder::new();
                 let consumed_num_bytes =
                     decoder.uncompress_vint_sorted(&encoded_data, *offset, input.len());
                 assert_eq!(consumed_num_bytes, encoded_data.len());

src/postings/compression/vint.rs

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

src/postings/mod.rs

@@ -3,8 +3,11 @@ Postings module (also called inverted index)
 */
 
 mod block_search;
-mod block_segment_postings;
 pub(crate) mod compression;
+/// Postings module
+///
+/// Postings, also called inverted lists, is the key datastructure
+/// to full-text search.
 mod postings;
 mod postings_writer;
 mod recorder;
@@ -19,17 +22,18 @@ pub(crate) use self::block_search::BlockSearcher;
 pub(crate) use self::postings_writer::MultiFieldPostingsWriter;
 pub use self::serializer::{FieldSerializer, InvertedIndexSerializer};
 
+use self::compression::COMPRESSION_BLOCK_SIZE;
 pub use self::postings::Postings;
-pub(crate) use self::skip::{BlockInfo, SkipReader};
+pub(crate) use self::skip::SkipReader;
 pub use self::term_info::TermInfo;
 
-pub use self::block_segment_postings::BlockSegmentPostings;
-pub use self::segment_postings::SegmentPostings;
+pub use self::segment_postings::{BlockSegmentPostings, SegmentPostings};
 
 pub(crate) use self::stacker::compute_table_size;
 
 pub use crate::common::HasLen;
 
+pub(crate) const USE_SKIP_INFO_LIMIT: u32 = COMPRESSION_BLOCK_SIZE as u32;
 pub(crate) type UnorderedTermId = u64;
 
 #[cfg_attr(feature = "cargo-clippy", allow(clippy::enum_variant_names))]
@@ -47,7 +51,7 @@ pub mod tests {
     use crate::core::Index;
     use crate::core::SegmentComponent;
     use crate::core::SegmentReader;
-    use crate::docset::{DocSet, TERMINATED};
+    use crate::docset::{DocSet, SkipResult};
     use crate::fieldnorm::FieldNormReader;
     use crate::indexer::operation::AddOperation;
     use crate::indexer::SegmentWriter;
@@ -111,12 +115,29 @@ pub mod tests {
             let mut postings = inverted_index
                 .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
                 .unwrap();
-            assert_eq!(postings.doc(), 0);
+            postings.advance();
             postings.positions(&mut positions);
             assert_eq!(&[0, 1, 2], &positions[..]);
             postings.positions(&mut positions);
             assert_eq!(&[0, 1, 2], &positions[..]);
-            assert_eq!(postings.advance(), 1);
+            postings.advance();
+            postings.positions(&mut positions);
+            assert_eq!(&[0, 5], &positions[..]);
+        }
+        {
+            let mut postings = inverted_index
+                .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
+                .unwrap();
+            postings.advance();
+            postings.advance();
+            postings.positions(&mut positions);
+            assert_eq!(&[0, 5], &positions[..]);
+        }
+        {
+            let mut postings = inverted_index
+                .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
+                .unwrap();
+            assert_eq!(postings.skip_next(1), SkipResult::Reached);
             assert_eq!(postings.doc(), 1);
             postings.positions(&mut positions);
             assert_eq!(&[0, 5], &positions[..]);
@@ -125,25 +146,7 @@ pub mod tests {
             let mut postings = inverted_index
                 .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
                 .unwrap();
-            assert_eq!(postings.doc(), 0);
-            assert_eq!(postings.advance(), 1);
-            postings.positions(&mut positions);
-            assert_eq!(&[0, 5], &positions[..]);
-        }
-        {
-            let mut postings = inverted_index
-                .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
-                .unwrap();
-            assert_eq!(postings.seek(1), 1);
-            assert_eq!(postings.doc(), 1);
-            postings.positions(&mut positions);
-            assert_eq!(&[0, 5], &positions[..]);
-        }
-        {
-            let mut postings = inverted_index
-                .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
-                .unwrap();
-            assert_eq!(postings.seek(1002), 1002);
+            assert_eq!(postings.skip_next(1002), SkipResult::Reached);
             assert_eq!(postings.doc(), 1002);
             postings.positions(&mut positions);
             assert_eq!(&[0, 5], &positions[..]);
@@ -152,8 +155,8 @@ pub mod tests {
             let mut postings = inverted_index
                 .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
                 .unwrap();
-            assert_eq!(postings.seek(100), 100);
-            assert_eq!(postings.seek(1002), 1002);
+            assert_eq!(postings.skip_next(100), SkipResult::Reached);
+            assert_eq!(postings.skip_next(1002), SkipResult::Reached);
             assert_eq!(postings.doc(), 1002);
             postings.positions(&mut positions);
             assert_eq!(&[0, 5], &positions[..]);
@@ -217,7 +220,8 @@ pub mod tests {
 
         {
             let mut segment_writer =
-                SegmentWriter::for_segment(3_000_000, segment.clone(), &schema).unwrap();
+                SegmentWriter::for_segment(3_000_000, segment.clone(), &schema, index.tokenizers())
+                    .unwrap();
             {
                 let mut doc = Document::default();
                 // checking that position works if the field has two values
@@ -278,21 +282,22 @@ pub mod tests {
                     .read_postings(&term_a, IndexRecordOption::WithFreqsAndPositions)
                     .unwrap();
                 assert_eq!(postings_a.len(), 1000);
+                assert!(postings_a.advance());
                 assert_eq!(postings_a.doc(), 0);
                 assert_eq!(postings_a.term_freq(), 6);
                 postings_a.positions(&mut positions);
                 assert_eq!(&positions[..], [0, 2, 4, 6, 7, 13]);
-                assert_eq!(postings_a.advance(), 1u32);
+                assert!(postings_a.advance());
                 assert_eq!(postings_a.doc(), 1u32);
                 assert_eq!(postings_a.term_freq(), 1);
                 for i in 2u32..1000u32 {
-                    assert_eq!(postings_a.advance(), i);
+                    assert!(postings_a.advance());
                     assert_eq!(postings_a.term_freq(), 1);
                     postings_a.positions(&mut positions);
                     assert_eq!(&positions[..], [i]);
                     assert_eq!(postings_a.doc(), i);
                 }
-                assert_eq!(postings_a.advance(), TERMINATED);
+                assert!(!postings_a.advance());
             }
             {
                 let term_e = Term::from_field_text(text_field, "e");
@@ -302,6 +307,7 @@ pub mod tests {
                     .unwrap();
                 assert_eq!(postings_e.len(), 1000 - 2);
                 for i in 2u32..1000u32 {
+                    assert!(postings_e.advance());
                     assert_eq!(postings_e.term_freq(), i);
                     postings_e.positions(&mut positions);
                     assert_eq!(positions.len(), i as usize);
@@ -309,9 +315,8 @@ pub mod tests {
                         assert_eq!(positions[j], (j as u32));
                     }
                     assert_eq!(postings_e.doc(), i);
-                    postings_e.advance();
                 }
-                assert_eq!(postings_e.doc(), TERMINATED);
+                assert!(!postings_e.advance());
             }
         }
     }
@@ -325,8 +330,16 @@ pub mod tests {
         let index = Index::create_in_ram(schema);
         {
             let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-            index_writer.add_document(doc!(text_field => "g b b d c g c"));
-            index_writer.add_document(doc!(text_field => "g a b b a d c g c"));
+            {
+                let mut doc = Document::default();
+                doc.add_text(text_field, "g b b d c g c");
+                index_writer.add_document(doc);
+            }
+            {
+                let mut doc = Document::default();
+                doc.add_text(text_field, "g a b b a d c g c");
+                index_writer.add_document(doc);
+            }
             assert!(index_writer.commit().is_ok());
         }
         let term_a = Term::from_field_text(text_field, "a");
@@ -336,6 +349,7 @@ pub mod tests {
             .inverted_index(text_field)
             .read_postings(&term_a, IndexRecordOption::WithFreqsAndPositions)
             .unwrap();
+        assert!(postings.advance());
         assert_eq!(postings.doc(), 1u32);
         postings.positions(&mut positions);
         assert_eq!(&positions[..], &[1u32, 4]);
@@ -357,8 +371,11 @@ pub mod tests {
             let index = Index::create_in_ram(schema);
             {
                 let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-                for i in 0u64..num_docs as u64 {
-                    let doc = doc!(value_field => 2u64, value_field => i % 2u64);
+                for i in 0..num_docs {
+                    let mut doc = Document::default();
+                    doc.add_u64(value_field, 2);
+                    doc.add_u64(value_field, (i % 2) as u64);
+
                     index_writer.add_document(doc);
                 }
                 assert!(index_writer.commit().is_ok());
@@ -375,10 +392,11 @@ pub mod tests {
                     .inverted_index(term_2.field())
                     .read_postings(&term_2, IndexRecordOption::Basic)
                     .unwrap();
-                assert_eq!(segment_postings.seek(i), i);
+
+                assert_eq!(segment_postings.skip_next(i), SkipResult::Reached);
                 assert_eq!(segment_postings.doc(), i);
 
-                assert_eq!(segment_postings.seek(j), j);
+                assert_eq!(segment_postings.skip_next(j), SkipResult::Reached);
                 assert_eq!(segment_postings.doc(), j);
             }
         }
@@ -390,16 +408,17 @@ pub mod tests {
                 .unwrap();
 
             // check that `skip_next` advances the iterator
+            assert!(segment_postings.advance());
             assert_eq!(segment_postings.doc(), 0);
 
-            assert_eq!(segment_postings.seek(1), 1);
+            assert_eq!(segment_postings.skip_next(1), SkipResult::Reached);
             assert_eq!(segment_postings.doc(), 1);
 
-            assert_eq!(segment_postings.seek(1), 1);
-            assert_eq!(segment_postings.doc(), 1);
+            assert_eq!(segment_postings.skip_next(1), SkipResult::OverStep);
+            assert_eq!(segment_postings.doc(), 2);
 
             // check that going beyond the end is handled
-            assert_eq!(segment_postings.seek(num_docs), TERMINATED);
+            assert_eq!(segment_postings.skip_next(num_docs), SkipResult::End);
         }
 
         // check that filtering works
@@ -410,7 +429,7 @@ pub mod tests {
                 .unwrap();
 
             for i in 0..num_docs / 2 {
-                assert_eq!(segment_postings.seek(i * 2), i * 2);
+                assert_eq!(segment_postings.skip_next(i * 2), SkipResult::Reached);
                 assert_eq!(segment_postings.doc(), i * 2);
             }
 
@@ -420,7 +439,7 @@ pub mod tests {
                 .unwrap();
 
             for i in 0..num_docs / 2 - 1 {
-                assert!(segment_postings.seek(i * 2 + 1) > (i * 1) * 2);
+                assert_eq!(segment_postings.skip_next(i * 2 + 1), SkipResult::OverStep);
                 assert_eq!(segment_postings.doc(), (i + 1) * 2);
             }
         }
@@ -432,7 +451,6 @@ pub mod tests {
             assert!(index_writer.commit().is_ok());
         }
         let searcher = index.reader().unwrap().searcher();
-        assert_eq!(searcher.segment_readers().len(), 1);
         let segment_reader = searcher.segment_reader(0);
 
         // make sure seeking still works
@@ -443,11 +461,11 @@ pub mod tests {
                 .unwrap();
 
             if i % 2 == 0 {
-                assert_eq!(segment_postings.seek(i), i);
+                assert_eq!(segment_postings.skip_next(i), SkipResult::Reached);
                 assert_eq!(segment_postings.doc(), i);
                 assert!(segment_reader.is_deleted(i));
             } else {
-                assert_eq!(segment_postings.seek(i), i);
+                assert_eq!(segment_postings.skip_next(i), SkipResult::Reached);
                 assert_eq!(segment_postings.doc(), i);
             }
         }
@@ -462,16 +480,12 @@ pub mod tests {
             let mut last = 2; // start from 5 to avoid seeking to 3 twice
             let mut cur = 3;
             loop {
-                let seek = segment_postings.seek(cur);
-                if seek == TERMINATED {
-                    break;
-                }
-                assert_eq!(seek, segment_postings.doc());
-                if seek == cur {
-                    assert_eq!(segment_postings.doc(), cur);
-                } else {
-                    assert_eq!(segment_postings.doc(), cur + 1);
+                match segment_postings.skip_next(cur) {
+                    SkipResult::End => break,
+                    SkipResult::Reached => assert_eq!(segment_postings.doc(), cur),
+                    SkipResult::OverStep => assert_eq!(segment_postings.doc(), cur + 1),
                 }
+
                 let next = cur + last;
                 last = cur;
                 cur = next;
@@ -557,7 +571,7 @@ pub mod tests {
     }
 
     impl<TDocSet: DocSet> DocSet for UnoptimizedDocSet<TDocSet> {
-        fn advance(&mut self) -> DocId {
+        fn advance(&mut self) -> bool {
             self.0.advance()
         }
 
@@ -583,22 +597,30 @@ pub mod tests {
         for target in targets {
             let mut postings_opt = postings_factory();
             let mut postings_unopt = UnoptimizedDocSet::wrap(postings_factory());
-            let skip_result_opt = postings_opt.seek(target);
-            let skip_result_unopt = postings_unopt.seek(target);
+            let skip_result_opt = postings_opt.skip_next(target);
+            let skip_result_unopt = postings_unopt.skip_next(target);
             assert_eq!(
                 skip_result_unopt, skip_result_opt,
                 "Failed while skipping to {}",
                 target
             );
-            assert!(skip_result_opt >= target);
-            assert_eq!(skip_result_opt, postings_opt.doc());
-            if skip_result_opt == TERMINATED {
-                return;
+            match skip_result_opt {
+                SkipResult::Reached => assert_eq!(postings_opt.doc(), target),
+                SkipResult::OverStep => assert!(postings_opt.doc() > target),
+                SkipResult::End => {
+                    return;
+                }
             }
-            while postings_opt.doc() != TERMINATED {
-                assert_eq!(postings_opt.doc(), postings_unopt.doc());
-                assert_eq!(postings_opt.advance(), postings_unopt.advance());
+            while postings_opt.advance() {
+                assert!(postings_unopt.advance());
+                assert_eq!(
+                    postings_opt.doc(),
+                    postings_unopt.doc(),
+                    "Failed while skipping to {}",
+                    target
+                );
             }
+            assert!(!postings_unopt.advance());
         }
     }
 }
@@ -607,7 +629,7 @@ pub mod tests {
 mod bench {
 
     use super::tests::*;
-    use crate::docset::TERMINATED;
+    use crate::docset::SkipResult;
     use crate::query::Intersection;
     use crate::schema::IndexRecordOption;
     use crate::tests;
@@ -625,7 +647,7 @@ mod bench {
                 .inverted_index(TERM_A.field())
                 .read_postings(&*TERM_A, IndexRecordOption::Basic)
                 .unwrap();
-            while segment_postings.advance() != TERMINATED {}
+            while segment_postings.advance() {}
         });
     }
 
@@ -657,7 +679,7 @@ mod bench {
                 segment_postings_c,
                 segment_postings_d,
             ]);
-            while intersection.advance() != TERMINATED {}
+            while intersection.advance() {}
         });
     }
 
@@ -673,10 +695,11 @@ mod bench {
             .unwrap();
 
         let mut existing_docs = Vec::new();
+        segment_postings.advance();
         for doc in &docs {
             if *doc >= segment_postings.doc() {
                 existing_docs.push(*doc);
-                if segment_postings.seek(*doc) == TERMINATED {
+                if segment_postings.skip_next(*doc) == SkipResult::End {
                     break;
                 }
             }
@@ -688,7 +711,7 @@ mod bench {
                 .read_postings(&*TERM_A, IndexRecordOption::Basic)
                 .unwrap();
             for doc in &existing_docs {
-                if segment_postings.seek(*doc) == TERMINATED {
+                if segment_postings.skip_next(*doc) == SkipResult::End {
                     break;
                 }
             }
@@ -727,9 +750,8 @@ mod bench {
                 .read_postings(&*TERM_A, IndexRecordOption::Basic)
                 .unwrap();
             let mut s = 0u32;
-            while segment_postings.doc() != TERMINATED {
+            while segment_postings.advance() {
                 s += (segment_postings.doc() & n) % 1024;
-                segment_postings.advance()
             }
             s
         });

src/postings/segment_postings.rs

@@ -1,19 +1,56 @@
+use crate::common::BitSet;
 use crate::common::HasLen;
-
-use crate::docset::DocSet;
+use crate::common::{BinarySerializable, VInt};
+use crate::docset::{DocSet, SkipResult};
 use crate::positions::PositionReader;
-
-use crate::postings::compression::COMPRESSION_BLOCK_SIZE;
+use crate::postings::compression::{compressed_block_size, AlignedBuffer};
+use crate::postings::compression::{BlockDecoder, VIntDecoder, COMPRESSION_BLOCK_SIZE};
 use crate::postings::serializer::PostingsSerializer;
 use crate::postings::BlockSearcher;
-
+use crate::postings::FreqReadingOption;
 use crate::postings::Postings;
-
+use crate::postings::SkipReader;
+use crate::postings::USE_SKIP_INFO_LIMIT;
 use crate::schema::IndexRecordOption;
 use crate::DocId;
+use owned_read::OwnedRead;
+use std::cmp::Ordering;
+use tantivy_fst::Streamer;
 
-use crate::directory::ReadOnlySource;
-use crate::postings::BlockSegmentPostings;
+struct PositionComputer {
+    // store the amount of position int
+    // before reading positions.
+    //
+    // if none, position are already loaded in
+    // the positions vec.
+    position_to_skip: usize,
+    position_reader: PositionReader,
+}
+
+impl PositionComputer {
+    pub fn new(position_reader: PositionReader) -> PositionComputer {
+        PositionComputer {
+            position_to_skip: 0,
+            position_reader,
+        }
+    }
+
+    pub fn add_skip(&mut self, num_skip: usize) {
+        self.position_to_skip += num_skip;
+    }
+
+    // Positions can only be read once.
+    pub fn positions_with_offset(&mut self, offset: u32, output: &mut [u32]) {
+        self.position_reader.skip(self.position_to_skip);
+        self.position_to_skip = 0;
+        self.position_reader.read(output);
+        let mut cum = offset;
+        for output_mut in output.iter_mut() {
+            cum += *output_mut;
+            *output_mut = cum;
+        }
+    }
+}
 
 /// `SegmentPostings` represents the inverted list or postings associated to
 /// a term in a `Segment`.
@@ -23,17 +60,18 @@ use crate::postings::BlockSegmentPostings;
 pub struct SegmentPostings {
     block_cursor: BlockSegmentPostings,
     cur: usize,
-    position_reader: Option<PositionReader>,
+    position_computer: Option<PositionComputer>,
     block_searcher: BlockSearcher,
 }
 
 impl SegmentPostings {
     /// Returns an empty segment postings object
     pub fn empty() -> Self {
+        let empty_block_cursor = BlockSegmentPostings::empty();
         SegmentPostings {
-            block_cursor: BlockSegmentPostings::empty(),
-            cur: 0,
-            position_reader: None,
+            block_cursor: empty_block_cursor,
+            cur: COMPRESSION_BLOCK_SIZE,
+            position_computer: None,
             block_searcher: BlockSearcher::default(),
         }
     }
@@ -59,13 +97,15 @@ impl SegmentPostings {
         }
         let block_segment_postings = BlockSegmentPostings::from_data(
             docs.len() as u32,
-            ReadOnlySource::from(buffer),
+            OwnedRead::new(buffer),
             IndexRecordOption::Basic,
             IndexRecordOption::Basic,
         );
         SegmentPostings::from_block_postings(block_segment_postings, None)
     }
+}
 
+impl SegmentPostings {
     /// Reads a Segment postings from an &[u8]
     ///
     /// * `len` - number of document in the posting lists.
@@ -74,12 +114,12 @@ impl SegmentPostings {
     ///   frequencies and/or positions
     pub(crate) fn from_block_postings(
         segment_block_postings: BlockSegmentPostings,
-        position_reader: Option<PositionReader>,
+        positions_stream_opt: Option<PositionReader>,
     ) -> SegmentPostings {
         SegmentPostings {
             block_cursor: segment_block_postings,
-            cur: 0, // cursor within the block
-            position_reader,
+            cur: COMPRESSION_BLOCK_SIZE, // cursor within the block
+            position_computer: positions_stream_opt.map(PositionComputer::new),
             block_searcher: BlockSearcher::default(),
         }
     }
@@ -89,52 +129,134 @@ impl DocSet for SegmentPostings {
     // goes to the next element.
     // next needs to be called a first time to point to the correct element.
     #[inline]
-    fn advance(&mut self) -> DocId {
-        if self.cur == COMPRESSION_BLOCK_SIZE - 1 {
-            self.cur = 0;
-            self.block_cursor.advance();
-        } else {
-            self.cur += 1;
+    fn advance(&mut self) -> bool {
+        if self.position_computer.is_some() && self.cur < COMPRESSION_BLOCK_SIZE {
+            let term_freq = self.term_freq() as usize;
+            if let Some(position_computer) = self.position_computer.as_mut() {
+                position_computer.add_skip(term_freq);
+            }
         }
-        self.doc()
+        self.cur += 1;
+        if self.cur >= self.block_cursor.block_len() {
+            self.cur = 0;
+            if !self.block_cursor.advance() {
+                self.cur = COMPRESSION_BLOCK_SIZE;
+                return false;
+            }
+        }
+        true
     }
 
-    fn seek(&mut self, target: DocId) -> DocId {
-        if self.doc() == target {
-            return target;
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        if !self.advance() {
+            return SkipResult::End;
+        }
+        match self.doc().cmp(&target) {
+            Ordering::Equal => {
+                return SkipResult::Reached;
+            }
+            Ordering::Greater => {
+                return SkipResult::OverStep;
+            }
+            _ => {
+                // ...
+            }
         }
-        self.block_cursor.seek(target);
 
-        // At this point we are on the block, that might contain our document.
-        let output = self.block_cursor.docs_aligned();
+        // In the following, thanks to the call to advance above,
+        // we know that the position is not loaded and we need
+        // to skip every doc_freq we cross.
 
-        self.cur = self.block_searcher.search_in_block(&output, target);
+        // skip blocks until one that might contain the target
+        // check if we need to go to the next block
+        let mut sum_freqs_skipped: u32 = 0;
+        if !self
+            .block_cursor
+            .docs()
+            .last()
+            .map(|doc| *doc >= target)
+            .unwrap_or(false)
+        // there should always be at least a document in the block
+        // since advance returned.
+        {
+            // we are not in the right block.
+            //
+            // First compute all of the freqs skipped from the current block.
+            if self.position_computer.is_some() {
+                sum_freqs_skipped = self.block_cursor.freqs()[self.cur..].iter().sum();
+                match self.block_cursor.skip_to(target) {
+                    BlockSegmentPostingsSkipResult::Success(block_skip_freqs) => {
+                        sum_freqs_skipped += block_skip_freqs;
+                    }
+                    BlockSegmentPostingsSkipResult::Terminated => {
+                        return SkipResult::End;
+                    }
+                }
+            } else if self.block_cursor.skip_to(target)
+                == BlockSegmentPostingsSkipResult::Terminated
+            {
+                // no positions needed. no need to sum freqs.
+                return SkipResult::End;
+            }
+            self.cur = 0;
+        }
 
-        // The last block is not full and padded with the value TERMINATED,
-        // so that we are guaranteed to have at least doc in the block (a real one or the padding)
-        // that is greater or equal to the target.
-        debug_assert!(self.cur < COMPRESSION_BLOCK_SIZE);
+        let cur = self.cur;
+
+        // we're in the right block now, start with an exponential search
+        let (output, len) = self.block_cursor.docs_aligned();
+        let new_cur = self
+            .block_searcher
+            .search_in_block(&output, len, cur, target);
+        if let Some(position_computer) = self.position_computer.as_mut() {
+            sum_freqs_skipped += self.block_cursor.freqs()[cur..new_cur].iter().sum::<u32>();
+            position_computer.add_skip(sum_freqs_skipped as usize);
+        }
+        self.cur = new_cur;
 
         // `doc` is now the first element >= `target`
-
-        // If all docs are smaller than target the current block should be incomplemented and padded
-        // with the value `TERMINATED`.
-        //
-        // After the search, the cursor should point to the first value of TERMINATED.
-        let doc = output.0[self.cur];
+        let doc = output.0[new_cur];
         debug_assert!(doc >= target);
-        doc
+        if doc == target {
+            SkipResult::Reached
+        } else {
+            SkipResult::OverStep
+        }
     }
 
     /// Return the current document's `DocId`.
+    ///
+    /// # Panics
+    ///
+    /// Will panics if called without having called advance before.
     #[inline]
     fn doc(&self) -> DocId {
-        self.block_cursor.doc(self.cur)
+        let docs = self.block_cursor.docs();
+        debug_assert!(
+            self.cur < docs.len(),
+            "Have you forgotten to call `.advance()` at least once before calling `.doc()`                                      ."
+        );
+        docs[self.cur]
     }
 
     fn size_hint(&self) -> u32 {
         self.len() as u32
     }
+
+    fn append_to_bitset(&mut self, bitset: &mut BitSet) {
+        // finish the current block
+        if self.advance() {
+            for &doc in &self.block_cursor.docs()[self.cur..] {
+                bitset.insert(doc);
+            }
+            // ... iterate through the remaining blocks.
+            while self.block_cursor.advance() {
+                for &doc in self.block_cursor.docs() {
+                    bitset.insert(doc);
+                }
+            }
+        }
+    }
 }
 
 impl HasLen for SegmentPostings {
@@ -168,52 +290,515 @@ impl Postings for SegmentPostings {
 
     fn positions_with_offset(&mut self, offset: u32, output: &mut Vec<u32>) {
         let term_freq = self.term_freq() as usize;
-        if let Some(position_reader) = self.position_reader.as_mut() {
-            let read_offset = self.block_cursor.position_offset()
-                + (self.block_cursor.freqs()[..self.cur]
-                    .iter()
-                    .cloned()
-                    .sum::<u32>() as u64);
+        if let Some(position_comp) = self.position_computer.as_mut() {
             output.resize(term_freq, 0u32);
-            position_reader.read(read_offset, &mut output[..]);
-            let mut cum = offset;
-            for output_mut in output.iter_mut() {
-                cum += *output_mut;
-                *output_mut = cum;
-            }
+            position_comp.positions_with_offset(offset, &mut output[..]);
         } else {
             output.clear();
         }
     }
 }
 
+/// `BlockSegmentPostings` is a cursor iterating over blocks
+/// of documents.
+///
+/// # Warning
+///
+/// While it is useful for some very specific high-performance
+/// use cases, you should prefer using `SegmentPostings` for most usage.
+pub struct BlockSegmentPostings {
+    doc_decoder: BlockDecoder,
+    freq_decoder: BlockDecoder,
+    freq_reading_option: FreqReadingOption,
+
+    doc_freq: usize,
+    doc_offset: DocId,
+
+    num_vint_docs: usize,
+
+    remaining_data: OwnedRead,
+    skip_reader: SkipReader,
+}
+
+fn split_into_skips_and_postings(
+    doc_freq: u32,
+    mut data: OwnedRead,
+) -> (Option<OwnedRead>, OwnedRead) {
+    if doc_freq >= USE_SKIP_INFO_LIMIT {
+        let skip_len = VInt::deserialize(&mut data).expect("Data corrupted").0 as usize;
+        let mut postings_data = data.clone();
+        postings_data.advance(skip_len);
+        data.clip(skip_len);
+        (Some(data), postings_data)
+    } else {
+        (None, data)
+    }
+}
+
+#[derive(Debug, Eq, PartialEq)]
+pub enum BlockSegmentPostingsSkipResult {
+    Terminated,
+    Success(u32), //< number of term freqs to skip
+}
+
+impl BlockSegmentPostings {
+    pub(crate) fn from_data(
+        doc_freq: u32,
+        data: OwnedRead,
+        record_option: IndexRecordOption,
+        requested_option: IndexRecordOption,
+    ) -> BlockSegmentPostings {
+        let freq_reading_option = match (record_option, requested_option) {
+            (IndexRecordOption::Basic, _) => FreqReadingOption::NoFreq,
+            (_, IndexRecordOption::Basic) => FreqReadingOption::SkipFreq,
+            (_, _) => FreqReadingOption::ReadFreq,
+        };
+
+        let (skip_data_opt, postings_data) = split_into_skips_and_postings(doc_freq, data);
+        let skip_reader = match skip_data_opt {
+            Some(skip_data) => SkipReader::new(skip_data, record_option),
+            None => SkipReader::new(OwnedRead::new(&[][..]), record_option),
+        };
+        let doc_freq = doc_freq as usize;
+        let num_vint_docs = doc_freq % COMPRESSION_BLOCK_SIZE;
+        BlockSegmentPostings {
+            num_vint_docs,
+            doc_decoder: BlockDecoder::new(),
+            freq_decoder: BlockDecoder::with_val(1),
+            freq_reading_option,
+            doc_offset: 0,
+            doc_freq,
+            remaining_data: postings_data,
+            skip_reader,
+        }
+    }
+
+    // Resets the block segment postings on another position
+    // in the postings file.
+    //
+    // This is useful for enumerating through a list of terms,
+    // and consuming the associated posting lists while avoiding
+    // reallocating a `BlockSegmentPostings`.
+    //
+    // # Warning
+    //
+    // This does not reset the positions list.
+    pub(crate) fn reset(&mut self, doc_freq: u32, postings_data: OwnedRead) {
+        let (skip_data_opt, postings_data) = split_into_skips_and_postings(doc_freq, postings_data);
+        let num_vint_docs = (doc_freq as usize) & (COMPRESSION_BLOCK_SIZE - 1);
+        self.num_vint_docs = num_vint_docs;
+        self.remaining_data = postings_data;
+        if let Some(skip_data) = skip_data_opt {
+            self.skip_reader.reset(skip_data);
+        } else {
+            self.skip_reader.reset(OwnedRead::new(&[][..]))
+        }
+        self.doc_offset = 0;
+        self.doc_freq = doc_freq as usize;
+    }
+
+    /// Returns the document frequency associated to this block postings.
+    ///
+    /// This `doc_freq` is simply the sum of the length of all of the blocks
+    /// length, and it does not take in account deleted documents.
+    pub fn doc_freq(&self) -> usize {
+        self.doc_freq
+    }
+
+    /// Returns the array of docs in the current block.
+    ///
+    /// Before the first call to `.advance()`, the block
+    /// returned by `.docs()` is empty.
+    #[inline]
+    pub fn docs(&self) -> &[DocId] {
+        self.doc_decoder.output_array()
+    }
+
+    pub(crate) fn docs_aligned(&self) -> (&AlignedBuffer, usize) {
+        self.doc_decoder.output_aligned()
+    }
+
+    /// Return the document at index `idx` of the block.
+    #[inline]
+    pub fn doc(&self, idx: usize) -> u32 {
+        self.doc_decoder.output(idx)
+    }
+
+    /// Return the array of `term freq` in the block.
+    #[inline]
+    pub fn freqs(&self) -> &[u32] {
+        self.freq_decoder.output_array()
+    }
+
+    /// Return the frequency at index `idx` of the block.
+    #[inline]
+    pub fn freq(&self, idx: usize) -> u32 {
+        self.freq_decoder.output(idx)
+    }
+
+    /// Returns the length of the current block.
+    ///
+    /// All blocks have a length of `NUM_DOCS_PER_BLOCK`,
+    /// except the last block that may have a length
+    /// of any number between 1 and `NUM_DOCS_PER_BLOCK - 1`
+    #[inline]
+    fn block_len(&self) -> usize {
+        self.doc_decoder.output_len
+    }
+
+    /// position on a block that may contains `doc_id`.
+    /// Always advance the current block.
+    ///
+    /// Returns true if a block that has an element greater or equal to the target is found.
+    /// Returning true does not guarantee that the smallest element of the block is smaller
+    /// than the target. It only guarantees that the last element is greater or equal.
+    ///
+    /// Returns false iff all of the document remaining are smaller than
+    /// `doc_id`. In that case, all of these document are consumed.
+    ///
+    pub fn skip_to(&mut self, target_doc: DocId) -> BlockSegmentPostingsSkipResult {
+        let mut skip_freqs = 0u32;
+        while self.skip_reader.advance() {
+            if self.skip_reader.doc() >= target_doc {
+                // the last document of the current block is larger
+                // than the target.
+                //
+                // We found our block!
+                let num_bits = self.skip_reader.doc_num_bits();
+                let num_consumed_bytes = self.doc_decoder.uncompress_block_sorted(
+                    self.remaining_data.as_ref(),
+                    self.doc_offset,
+                    num_bits,
+                );
+                self.remaining_data.advance(num_consumed_bytes);
+                let tf_num_bits = self.skip_reader.tf_num_bits();
+                match self.freq_reading_option {
+                    FreqReadingOption::NoFreq => {}
+                    FreqReadingOption::SkipFreq => {
+                        let num_bytes_to_skip = compressed_block_size(tf_num_bits);
+                        self.remaining_data.advance(num_bytes_to_skip);
+                    }
+                    FreqReadingOption::ReadFreq => {
+                        let num_consumed_bytes = self
+                            .freq_decoder
+                            .uncompress_block_unsorted(self.remaining_data.as_ref(), tf_num_bits);
+                        self.remaining_data.advance(num_consumed_bytes);
+                    }
+                }
+                self.doc_offset = self.skip_reader.doc();
+                return BlockSegmentPostingsSkipResult::Success(skip_freqs);
+            } else {
+                skip_freqs += self.skip_reader.tf_sum();
+                let advance_len = self.skip_reader.total_block_len();
+                self.doc_offset = self.skip_reader.doc();
+                self.remaining_data.advance(advance_len);
+            }
+        }
+
+        // we are now on the last, incomplete, variable encoded block.
+        if self.num_vint_docs > 0 {
+            let num_compressed_bytes = self.doc_decoder.uncompress_vint_sorted(
+                self.remaining_data.as_ref(),
+                self.doc_offset,
+                self.num_vint_docs,
+            );
+            self.remaining_data.advance(num_compressed_bytes);
+            match self.freq_reading_option {
+                FreqReadingOption::NoFreq | FreqReadingOption::SkipFreq => {}
+                FreqReadingOption::ReadFreq => {
+                    self.freq_decoder
+                        .uncompress_vint_unsorted(self.remaining_data.as_ref(), self.num_vint_docs);
+                }
+            }
+            self.num_vint_docs = 0;
+            return self
+                .docs()
+                .last()
+                .map(|last_doc| {
+                    if *last_doc >= target_doc {
+                        BlockSegmentPostingsSkipResult::Success(skip_freqs)
+                    } else {
+                        BlockSegmentPostingsSkipResult::Terminated
+                    }
+                })
+                .unwrap_or(BlockSegmentPostingsSkipResult::Terminated);
+        }
+        BlockSegmentPostingsSkipResult::Terminated
+    }
+
+    /// Advance to the next block.
+    ///
+    /// Returns false iff there was no remaining blocks.
+    pub fn advance(&mut self) -> bool {
+        if self.skip_reader.advance() {
+            let num_bits = self.skip_reader.doc_num_bits();
+            let num_consumed_bytes = self.doc_decoder.uncompress_block_sorted(
+                self.remaining_data.as_ref(),
+                self.doc_offset,
+                num_bits,
+            );
+            self.remaining_data.advance(num_consumed_bytes);
+            let tf_num_bits = self.skip_reader.tf_num_bits();
+            match self.freq_reading_option {
+                FreqReadingOption::NoFreq => {}
+                FreqReadingOption::SkipFreq => {
+                    let num_bytes_to_skip = compressed_block_size(tf_num_bits);
+                    self.remaining_data.advance(num_bytes_to_skip);
+                }
+                FreqReadingOption::ReadFreq => {
+                    let num_consumed_bytes = self
+                        .freq_decoder
+                        .uncompress_block_unsorted(self.remaining_data.as_ref(), tf_num_bits);
+                    self.remaining_data.advance(num_consumed_bytes);
+                }
+            }
+            // it will be used as the next offset.
+            self.doc_offset = self.doc_decoder.output(COMPRESSION_BLOCK_SIZE - 1);
+            true
+        } else if self.num_vint_docs > 0 {
+            let num_compressed_bytes = self.doc_decoder.uncompress_vint_sorted(
+                self.remaining_data.as_ref(),
+                self.doc_offset,
+                self.num_vint_docs,
+            );
+            self.remaining_data.advance(num_compressed_bytes);
+            match self.freq_reading_option {
+                FreqReadingOption::NoFreq | FreqReadingOption::SkipFreq => {}
+                FreqReadingOption::ReadFreq => {
+                    self.freq_decoder
+                        .uncompress_vint_unsorted(self.remaining_data.as_ref(), self.num_vint_docs);
+                }
+            }
+            self.num_vint_docs = 0;
+            true
+        } else {
+            false
+        }
+    }
+
+    /// Returns an empty segment postings object
+    pub fn empty() -> BlockSegmentPostings {
+        BlockSegmentPostings {
+            num_vint_docs: 0,
+
+            doc_decoder: BlockDecoder::new(),
+            freq_decoder: BlockDecoder::with_val(1),
+            freq_reading_option: FreqReadingOption::NoFreq,
+
+            doc_offset: 0,
+            doc_freq: 0,
+
+            remaining_data: OwnedRead::new(vec![]),
+            skip_reader: SkipReader::new(OwnedRead::new(vec![]), IndexRecordOption::Basic),
+        }
+    }
+}
+
+impl<'b> Streamer<'b> for BlockSegmentPostings {
+    type Item = &'b [DocId];
+
+    fn next(&'b mut self) -> Option<&'b [DocId]> {
+        if self.advance() {
+            Some(self.docs())
+        } else {
+            None
+        }
+    }
+}
+
 #[cfg(test)]
 mod tests {
-
+    use super::BlockSegmentPostings;
+    use super::BlockSegmentPostingsSkipResult;
     use super::SegmentPostings;
     use crate::common::HasLen;
-
-    use crate::docset::{DocSet, TERMINATED};
+    use crate::core::Index;
+    use crate::docset::DocSet;
     use crate::postings::postings::Postings;
+    use crate::schema::IndexRecordOption;
+    use crate::schema::Schema;
+    use crate::schema::Term;
+    use crate::schema::INDEXED;
+    use crate::DocId;
+    use crate::SkipResult;
+    use tantivy_fst::Streamer;
 
     #[test]
     fn test_empty_segment_postings() {
         let mut postings = SegmentPostings::empty();
-        assert_eq!(postings.advance(), TERMINATED);
-        assert_eq!(postings.advance(), TERMINATED);
+        assert!(!postings.advance());
+        assert!(!postings.advance());
         assert_eq!(postings.len(), 0);
     }
 
     #[test]
-    fn test_empty_postings_doc_returns_terminated() {
-        let mut postings = SegmentPostings::empty();
-        assert_eq!(postings.doc(), TERMINATED);
-        assert_eq!(postings.advance(), TERMINATED);
+    #[should_panic(expected = "Have you forgotten to call `.advance()`")]
+    fn test_panic_if_doc_called_before_advance() {
+        SegmentPostings::empty().doc();
     }
 
     #[test]
-    fn test_empty_postings_doc_term_freq_returns_0() {
-        let postings = SegmentPostings::empty();
-        assert_eq!(postings.term_freq(), 1);
+    #[should_panic(expected = "Have you forgotten to call `.advance()`")]
+    fn test_panic_if_freq_called_before_advance() {
+        SegmentPostings::empty().term_freq();
+    }
+
+    #[test]
+    fn test_empty_block_segment_postings() {
+        let mut postings = BlockSegmentPostings::empty();
+        assert!(!postings.advance());
+        assert_eq!(postings.doc_freq(), 0);
+    }
+
+    #[test]
+    fn test_block_segment_postings() {
+        let mut block_segments = build_block_postings(&(0..100_000).collect::<Vec<u32>>());
+        let mut offset: u32 = 0u32;
+        // checking that the block before calling advance is empty
+        assert!(block_segments.docs().is_empty());
+        // checking that the `doc_freq` is correct
+        assert_eq!(block_segments.doc_freq(), 100_000);
+        while let Some(block) = block_segments.next() {
+            for (i, doc) in block.iter().cloned().enumerate() {
+                assert_eq!(offset + (i as u32), doc);
+            }
+            offset += block.len() as u32;
+        }
+    }
+
+    #[test]
+    fn test_skip_right_at_new_block() {
+        let mut doc_ids = (0..128).collect::<Vec<u32>>();
+        doc_ids.push(129);
+        doc_ids.push(130);
+        {
+            let block_segments = build_block_postings(&doc_ids);
+            let mut docset = SegmentPostings::from_block_postings(block_segments, None);
+            assert_eq!(docset.skip_next(128), SkipResult::OverStep);
+            assert_eq!(docset.doc(), 129);
+            assert!(docset.advance());
+            assert_eq!(docset.doc(), 130);
+            assert!(!docset.advance());
+        }
+        {
+            let block_segments = build_block_postings(&doc_ids);
+            let mut docset = SegmentPostings::from_block_postings(block_segments, None);
+            assert_eq!(docset.skip_next(129), SkipResult::Reached);
+            assert_eq!(docset.doc(), 129);
+            assert!(docset.advance());
+            assert_eq!(docset.doc(), 130);
+            assert!(!docset.advance());
+        }
+        {
+            let block_segments = build_block_postings(&doc_ids);
+            let mut docset = SegmentPostings::from_block_postings(block_segments, None);
+            assert_eq!(docset.skip_next(131), SkipResult::End);
+        }
+    }
+
+    fn build_block_postings(docs: &[DocId]) -> BlockSegmentPostings {
+        let mut schema_builder = Schema::builder();
+        let int_field = schema_builder.add_u64_field("id", INDEXED);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        let mut last_doc = 0u32;
+        for &doc in docs {
+            for _ in last_doc..doc {
+                index_writer.add_document(doc!(int_field=>1u64));
+            }
+            index_writer.add_document(doc!(int_field=>0u64));
+            last_doc = doc + 1;
+        }
+        index_writer.commit().unwrap();
+        let searcher = index.reader().unwrap().searcher();
+        let segment_reader = searcher.segment_reader(0);
+        let inverted_index = segment_reader.inverted_index(int_field);
+        let term = Term::from_field_u64(int_field, 0u64);
+        let term_info = inverted_index.get_term_info(&term).unwrap();
+        inverted_index.read_block_postings_from_terminfo(&term_info, IndexRecordOption::Basic)
+    }
+
+    #[test]
+    fn test_block_segment_postings_skip() {
+        for i in 0..4 {
+            let mut block_postings = build_block_postings(&[3]);
+            assert_eq!(
+                block_postings.skip_to(i),
+                BlockSegmentPostingsSkipResult::Success(0u32)
+            );
+            assert_eq!(
+                block_postings.skip_to(i),
+                BlockSegmentPostingsSkipResult::Terminated
+            );
+        }
+        let mut block_postings = build_block_postings(&[3]);
+        assert_eq!(
+            block_postings.skip_to(4u32),
+            BlockSegmentPostingsSkipResult::Terminated
+        );
+    }
+
+    #[test]
+    fn test_block_segment_postings_skip2() {
+        let mut docs = vec![0];
+        for i in 0..1300 {
+            docs.push((i * i / 100) + i);
+        }
+        let mut block_postings = build_block_postings(&docs[..]);
+        for i in vec![0, 424, 10000] {
+            assert_eq!(
+                block_postings.skip_to(i),
+                BlockSegmentPostingsSkipResult::Success(0u32)
+            );
+            let docs = block_postings.docs();
+            assert!(docs[0] <= i);
+            assert!(docs.last().cloned().unwrap_or(0u32) >= i);
+        }
+        assert_eq!(
+            block_postings.skip_to(100_000),
+            BlockSegmentPostingsSkipResult::Terminated
+        );
+        assert_eq!(
+            block_postings.skip_to(101_000),
+            BlockSegmentPostingsSkipResult::Terminated
+        );
+    }
+
+    #[test]
+    fn test_reset_block_segment_postings() {
+        let mut schema_builder = Schema::builder();
+        let int_field = schema_builder.add_u64_field("id", INDEXED);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        // create two postings list, one containg even number,
+        // the other containing odd numbers.
+        for i in 0..6 {
+            let doc = doc!(int_field=> (i % 2) as u64);
+            index_writer.add_document(doc);
+        }
+        index_writer.commit().unwrap();
+        let searcher = index.reader().unwrap().searcher();
+        let segment_reader = searcher.segment_reader(0);
+
+        let mut block_segments;
+        {
+            let term = Term::from_field_u64(int_field, 0u64);
+            let inverted_index = segment_reader.inverted_index(int_field);
+            let term_info = inverted_index.get_term_info(&term).unwrap();
+            block_segments = inverted_index
+                .read_block_postings_from_terminfo(&term_info, IndexRecordOption::Basic);
+        }
+        assert!(block_segments.advance());
+        assert_eq!(block_segments.docs(), &[0, 2, 4]);
+        {
+            let term = Term::from_field_u64(int_field, 1u64);
+            let inverted_index = segment_reader.inverted_index(int_field);
+            let term_info = inverted_index.get_term_info(&term).unwrap();
+            inverted_index.reset_block_postings_from_terminfo(&term_info, &mut block_segments);
+        }
+        assert!(block_segments.advance());
+        assert_eq!(block_segments.docs(), &[1, 3, 5]);
     }
 }

src/postings/serializer.rs

@@ -6,6 +6,7 @@ use crate::directory::WritePtr;
 use crate::positions::PositionSerializer;
 use crate::postings::compression::{BlockEncoder, VIntEncoder, COMPRESSION_BLOCK_SIZE};
 use crate::postings::skip::SkipSerializer;
+use crate::postings::USE_SKIP_INFO_LIMIT;
 use crate::schema::Schema;
 use crate::schema::{Field, FieldEntry, FieldType};
 use crate::termdict::{TermDictionaryBuilder, TermOrdinal};
@@ -390,7 +391,7 @@ impl<W: Write> PostingsSerializer<W> {
             }
             self.block.clear();
         }
-        if doc_freq >= COMPRESSION_BLOCK_SIZE as u32 {
+        if doc_freq >= USE_SKIP_INFO_LIMIT {
             let skip_data = self.skip_write.data();
             VInt(skip_data.len() as u64).serialize(&mut self.output_write)?;
             self.output_write.write_all(skip_data)?;

src/postings/skip.rs

@@ -1,8 +1,7 @@
 use crate::common::BinarySerializable;
-use crate::directory::ReadOnlySource;
-use crate::postings::compression::{compressed_block_size, COMPRESSION_BLOCK_SIZE};
+use crate::postings::compression::COMPRESSION_BLOCK_SIZE;
 use crate::schema::IndexRecordOption;
-use crate::{DocId, TERMINATED};
+use crate::DocId;
 use owned_read::OwnedRead;
 
 pub struct SkipSerializer {
@@ -51,143 +50,80 @@ impl SkipSerializer {
 }
 
 pub(crate) struct SkipReader {
-    last_doc_in_block: DocId,
-    pub(crate) last_doc_in_previous_block: DocId,
+    doc: DocId,
     owned_read: OwnedRead,
+    doc_num_bits: u8,
+    tf_num_bits: u8,
+    tf_sum: u32,
     skip_info: IndexRecordOption,
-    byte_offset: usize,
-    remaining_docs: u32, // number of docs remaining, including the
-    // documents in the current block.
-    block_info: BlockInfo,
-
-    position_offset: u64,
-}
-
-#[derive(Clone, Eq, PartialEq, Copy, Debug)]
-pub(crate) enum BlockInfo {
-    BitPacked {
-        doc_num_bits: u8,
-        tf_num_bits: u8,
-        tf_sum: u32,
-    },
-    VInt(u32),
-}
-
-impl Default for BlockInfo {
-    fn default() -> Self {
-        BlockInfo::VInt(0)
-    }
 }
 
 impl SkipReader {
-    pub fn new(data: ReadOnlySource, doc_freq: u32, skip_info: IndexRecordOption) -> SkipReader {
+    pub fn new(data: OwnedRead, skip_info: IndexRecordOption) -> SkipReader {
         SkipReader {
-            last_doc_in_block: 0u32,
-            last_doc_in_previous_block: 0u32,
-            owned_read: OwnedRead::new(data),
+            doc: 0u32,
+            owned_read: data,
             skip_info,
-            block_info: BlockInfo::default(),
-            byte_offset: 0,
-            remaining_docs: doc_freq,
-            position_offset: 0u64,
+            doc_num_bits: 0u8,
+            tf_num_bits: 0u8,
+            tf_sum: 0u32,
         }
     }
 
-    pub fn reset(&mut self, data: ReadOnlySource, doc_freq: u32) {
-        self.last_doc_in_block = 0u32;
-        self.last_doc_in_previous_block = 0u32;
-        self.owned_read = OwnedRead::new(data);
-        self.block_info = BlockInfo::default();
-        self.byte_offset = 0;
-        self.remaining_docs = doc_freq;
+    pub fn reset(&mut self, data: OwnedRead) {
+        self.doc = 0u32;
+        self.owned_read = data;
+        self.doc_num_bits = 0u8;
+        self.tf_num_bits = 0u8;
+        self.tf_sum = 0u32;
     }
 
-    #[cfg(test)]
-    #[inline(always)]
-    pub(crate) fn last_doc_in_block(&self) -> DocId {
-        self.last_doc_in_block
+    pub fn total_block_len(&self) -> usize {
+        (self.doc_num_bits + self.tf_num_bits) as usize * COMPRESSION_BLOCK_SIZE / 8
     }
 
-    pub fn position_offset(&self) -> u64 {
-        self.position_offset
+    pub fn doc(&self) -> DocId {
+        self.doc
     }
 
-    pub fn byte_offset(&self) -> usize {
-        self.byte_offset
+    pub fn doc_num_bits(&self) -> u8 {
+        self.doc_num_bits
     }
 
-    fn read_block_info(&mut self) {
-        let doc_delta = u32::deserialize(&mut self.owned_read).expect("Skip data corrupted");
-        self.last_doc_in_block += doc_delta as DocId;
-        let doc_num_bits = self.owned_read.get(0);
-        match self.skip_info {
-            IndexRecordOption::Basic => {
-                self.owned_read.advance(1);
-                self.block_info = BlockInfo::BitPacked {
-                    doc_num_bits,
-                    tf_num_bits: 0,
-                    tf_sum: 0,
-                };
-            }
-            IndexRecordOption::WithFreqs => {
-                let tf_num_bits = self.owned_read.get(1);
-                self.block_info = BlockInfo::BitPacked {
-                    doc_num_bits,
-                    tf_num_bits,
-                    tf_sum: 0,
-                };
-                self.owned_read.advance(2);
-            }
-            IndexRecordOption::WithFreqsAndPositions => {
-                let tf_num_bits = self.owned_read.get(1);
-                self.owned_read.advance(2);
-                let tf_sum = u32::deserialize(&mut self.owned_read).expect("Failed reading tf_sum");
-                self.block_info = BlockInfo::BitPacked {
-                    doc_num_bits,
-                    tf_num_bits,
-                    tf_sum,
-                };
-            }
-        }
-    }
-
-    pub fn block_info(&self) -> BlockInfo {
-        self.block_info
-    }
-
-    /// Advance the skip reader to the block that may contain the target.
+    /// Number of bits used to encode term frequencies
     ///
-    /// If the target is larger than all documents, the skip_reader
-    /// then advance to the last Variable In block.
-    pub fn seek(&mut self, target: DocId) {
-        while self.last_doc_in_block < target {
-            self.advance();
-        }
+    /// 0 if term frequencies are not enabled.
+    pub fn tf_num_bits(&self) -> u8 {
+        self.tf_num_bits
+    }
+
+    pub fn tf_sum(&self) -> u32 {
+        self.tf_sum
     }
 
     pub fn advance(&mut self) -> bool {
-        match self.block_info {
-            BlockInfo::BitPacked {
-                doc_num_bits,
-                tf_num_bits,
-                tf_sum,
-            } => {
-                self.remaining_docs -= COMPRESSION_BLOCK_SIZE as u32;
-                self.byte_offset += compressed_block_size(doc_num_bits + tf_num_bits);
-                self.position_offset += tf_sum as u64;
-            }
-            BlockInfo::VInt(num_vint_docs) => {
-                self.remaining_docs -= num_vint_docs;
-            }
-        }
-        self.last_doc_in_previous_block = self.last_doc_in_block;
-        if self.remaining_docs >= COMPRESSION_BLOCK_SIZE as u32 {
-            self.read_block_info();
-            true
+        if self.owned_read.as_ref().is_empty() {
+            false
         } else {
-            self.last_doc_in_block = TERMINATED;
-            self.block_info = BlockInfo::VInt(self.remaining_docs);
-            self.remaining_docs > 0
+            let doc_delta = u32::deserialize(&mut self.owned_read).expect("Skip data corrupted");
+            self.doc += doc_delta as DocId;
+            self.doc_num_bits = self.owned_read.get(0);
+            match self.skip_info {
+                IndexRecordOption::Basic => {
+                    self.owned_read.advance(1);
+                }
+                IndexRecordOption::WithFreqs => {
+                    self.tf_num_bits = self.owned_read.get(1);
+                    self.owned_read.advance(2);
+                }
+                IndexRecordOption::WithFreqsAndPositions => {
+                    self.tf_num_bits = self.owned_read.get(1);
+                    self.owned_read.advance(2);
+                    self.tf_sum =
+                        u32::deserialize(&mut self.owned_read).expect("Failed reading tf_sum");
+                }
+            }
+            true
         }
     }
 }
@@ -195,11 +131,9 @@ impl SkipReader {
 #[cfg(test)]
 mod tests {
 
-    use super::BlockInfo;
     use super::IndexRecordOption;
     use super::{SkipReader, SkipSerializer};
-    use crate::directory::ReadOnlySource;
-    use crate::postings::compression::COMPRESSION_BLOCK_SIZE;
+    use owned_read::OwnedRead;
 
     #[test]
     fn test_skip_with_freq() {
@@ -211,34 +145,15 @@ mod tests {
             skip_serializer.write_term_freq(2u8);
             skip_serializer.data().to_owned()
         };
-        let doc_freq = 3u32 + (COMPRESSION_BLOCK_SIZE * 2) as u32;
-        let mut skip_reader = SkipReader::new(
-            ReadOnlySource::new(buf),
-            doc_freq,
-            IndexRecordOption::WithFreqs,
-        );
+        let mut skip_reader = SkipReader::new(OwnedRead::new(buf), IndexRecordOption::WithFreqs);
         assert!(skip_reader.advance());
-        assert_eq!(skip_reader.last_doc_in_block(), 1u32);
-        assert_eq!(
-            skip_reader.block_info(),
-            BlockInfo::BitPacked {
-                doc_num_bits: 2u8,
-                tf_num_bits: 3u8,
-                tf_sum: 0
-            }
-        );
+        assert_eq!(skip_reader.doc(), 1u32);
+        assert_eq!(skip_reader.doc_num_bits(), 2u8);
+        assert_eq!(skip_reader.tf_num_bits(), 3u8);
         assert!(skip_reader.advance());
-        assert_eq!(skip_reader.last_doc_in_block(), 5u32);
-        assert_eq!(
-            skip_reader.block_info(),
-            BlockInfo::BitPacked {
-                doc_num_bits: 5u8,
-                tf_num_bits: 2u8,
-                tf_sum: 0
-            }
-        );
-        assert!(skip_reader.advance());
-        assert_eq!(skip_reader.block_info(), BlockInfo::VInt(3u32));
+        assert_eq!(skip_reader.doc(), 5u32);
+        assert_eq!(skip_reader.doc_num_bits(), 5u8);
+        assert_eq!(skip_reader.tf_num_bits(), 2u8);
         assert!(!skip_reader.advance());
     }
 
@@ -250,60 +165,13 @@ mod tests {
             skip_serializer.write_doc(5u32, 5u8);
             skip_serializer.data().to_owned()
         };
-        let doc_freq = 3u32 + (COMPRESSION_BLOCK_SIZE * 2) as u32;
-        let mut skip_reader = SkipReader::new(
-            ReadOnlySource::from(buf),
-            doc_freq,
-            IndexRecordOption::Basic,
-        );
+        let mut skip_reader = SkipReader::new(OwnedRead::new(buf), IndexRecordOption::Basic);
         assert!(skip_reader.advance());
-        assert_eq!(skip_reader.last_doc_in_block(), 1u32);
-        assert_eq!(
-            skip_reader.block_info(),
-            BlockInfo::BitPacked {
-                doc_num_bits: 2u8,
-                tf_num_bits: 0,
-                tf_sum: 0u32
-            }
-        );
+        assert_eq!(skip_reader.doc(), 1u32);
+        assert_eq!(skip_reader.doc_num_bits(), 2u8);
         assert!(skip_reader.advance());
-        assert_eq!(skip_reader.last_doc_in_block(), 5u32);
-        assert_eq!(
-            skip_reader.block_info(),
-            BlockInfo::BitPacked {
-                doc_num_bits: 5u8,
-                tf_num_bits: 0,
-                tf_sum: 0u32
-            }
-        );
-        assert!(skip_reader.advance());
-        assert_eq!(skip_reader.block_info(), BlockInfo::VInt(3u32));
-        assert!(!skip_reader.advance());
-    }
-
-    #[test]
-    fn test_skip_multiple_of_block_size() {
-        let buf = {
-            let mut skip_serializer = SkipSerializer::new();
-            skip_serializer.write_doc(1u32, 2u8);
-            skip_serializer.data().to_owned()
-        };
-        let doc_freq = COMPRESSION_BLOCK_SIZE as u32;
-        let mut skip_reader = SkipReader::new(
-            ReadOnlySource::from(buf),
-            doc_freq,
-            IndexRecordOption::Basic,
-        );
-        assert!(skip_reader.advance());
-        assert_eq!(skip_reader.last_doc_in_block(), 1u32);
-        assert_eq!(
-            skip_reader.block_info(),
-            BlockInfo::BitPacked {
-                doc_num_bits: 2u8,
-                tf_num_bits: 0,
-                tf_sum: 0u32
-            }
-        );
+        assert_eq!(skip_reader.doc(), 5u32);
+        assert_eq!(skip_reader.doc_num_bits(), 5u8);
         assert!(!skip_reader.advance());
     }
 }

src/postings/stacker/term_hashmap.rs

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

src/query/all_query.rs

@@ -1,6 +1,6 @@
 use crate::core::Searcher;
 use crate::core::SegmentReader;
-use crate::docset::{DocSet, TERMINATED};
+use crate::docset::DocSet;
 use crate::query::boost_query::BoostScorer;
 use crate::query::explanation::does_not_match;
 use crate::query::{Explanation, Query, Scorer, Weight};
@@ -25,6 +25,7 @@ pub struct AllWeight;
 impl Weight for AllWeight {
     fn scorer(&self, reader: &SegmentReader, boost: f32) -> crate::Result<Box<dyn Scorer>> {
         let all_scorer = AllScorer {
+            state: State::NotStarted,
             doc: 0u32,
             max_doc: reader.max_doc(),
         };
@@ -39,20 +40,39 @@ impl Weight for AllWeight {
     }
 }
 
+enum State {
+    NotStarted,
+    Started,
+    Finished,
+}
+
 /// Scorer associated to the `AllQuery` query.
 pub struct AllScorer {
+    state: State,
     doc: DocId,
     max_doc: DocId,
 }
 
 impl DocSet for AllScorer {
-    fn advance(&mut self) -> DocId {
-        if self.doc + 1 >= self.max_doc {
-            self.doc = TERMINATED;
-            return TERMINATED;
+    fn advance(&mut self) -> bool {
+        match self.state {
+            State::NotStarted => {
+                self.state = State::Started;
+                self.doc = 0;
+            }
+            State::Started => {
+                self.doc += 1u32;
+            }
+            State::Finished => {
+                return false;
+            }
+        }
+        if self.doc < self.max_doc {
+            true
+        } else {
+            self.state = State::Finished;
+            false
         }
-        self.doc += 1;
-        self.doc
     }
 
     fn doc(&self) -> DocId {
@@ -73,7 +93,6 @@ impl Scorer for AllScorer {
 #[cfg(test)]
 mod tests {
     use super::AllQuery;
-    use crate::docset::TERMINATED;
     use crate::query::Query;
     use crate::schema::{Schema, TEXT};
     use crate::Index;
@@ -101,16 +120,18 @@ mod tests {
         {
             let reader = searcher.segment_reader(0);
             let mut scorer = weight.scorer(reader, 1.0f32).unwrap();
+            assert!(scorer.advance());
             assert_eq!(scorer.doc(), 0u32);
-            assert_eq!(scorer.advance(), 1u32);
+            assert!(scorer.advance());
             assert_eq!(scorer.doc(), 1u32);
-            assert_eq!(scorer.advance(), TERMINATED);
+            assert!(!scorer.advance());
         }
         {
             let reader = searcher.segment_reader(1);
             let mut scorer = weight.scorer(reader, 1.0f32).unwrap();
+            assert!(scorer.advance());
             assert_eq!(scorer.doc(), 0u32);
-            assert_eq!(scorer.advance(), TERMINATED);
+            assert!(!scorer.advance());
         }
     }
 
@@ -123,11 +144,13 @@ mod tests {
         let reader = searcher.segment_reader(0);
         {
             let mut scorer = weight.scorer(reader, 2.0f32).unwrap();
+            assert!(scorer.advance());
             assert_eq!(scorer.doc(), 0u32);
             assert_eq!(scorer.score(), 2.0f32);
         }
         {
             let mut scorer = weight.scorer(reader, 1.5f32).unwrap();
+            assert!(scorer.advance());
             assert_eq!(scorer.doc(), 0u32);
             assert_eq!(scorer.score(), 1.5f32);
         }

src/query/automaton_weight.rs

@@ -6,8 +6,8 @@ use crate::query::{Scorer, Weight};
 use crate::schema::{Field, IndexRecordOption};
 use crate::termdict::{TermDictionary, TermStreamer};
 use crate::DocId;
-use crate::Result;
 use crate::TantivyError;
+use crate::{Result, SkipResult};
 use std::sync::Arc;
 use tantivy_fst::Automaton;
 
@@ -51,13 +51,10 @@ where
             let term_info = term_stream.value();
             let mut block_segment_postings = inverted_index
                 .read_block_postings_from_terminfo(term_info, IndexRecordOption::Basic);
-            loop {
+            while block_segment_postings.advance() {
                 for &doc in block_segment_postings.docs() {
                     doc_bitset.insert(doc);
                 }
-                if !block_segment_postings.advance() {
-                    break;
-                }
             }
         }
         let doc_bitset = BitSetDocSet::from(doc_bitset);
@@ -67,7 +64,7 @@ where
 
     fn explain(&self, reader: &SegmentReader, doc: DocId) -> Result<Explanation> {
         let mut scorer = self.scorer(reader, 1.0f32)?;
-        if scorer.seek(doc) == doc {
+        if scorer.skip_next(doc) == SkipResult::Reached {
             Ok(Explanation::new("AutomatonScorer", 1.0f32))
         } else {
             Err(TantivyError::InvalidArgument(
@@ -80,7 +77,6 @@ where
 #[cfg(test)]
 mod tests {
     use super::AutomatonWeight;
-    use crate::docset::TERMINATED;
     use crate::query::Weight;
     use crate::schema::{Schema, STRING};
     use crate::Index;
@@ -145,12 +141,13 @@ mod tests {
         let mut scorer = automaton_weight
             .scorer(searcher.segment_reader(0u32), 1.0f32)
             .unwrap();
+        assert!(scorer.advance());
         assert_eq!(scorer.doc(), 0u32);
         assert_eq!(scorer.score(), 1.0f32);
-        assert_eq!(scorer.advance(), 2u32);
+        assert!(scorer.advance());
         assert_eq!(scorer.doc(), 2u32);
         assert_eq!(scorer.score(), 1.0f32);
-        assert_eq!(scorer.advance(), TERMINATED);
+        assert!(!scorer.advance());
     }
 
     #[test]
@@ -163,6 +160,7 @@ mod tests {
         let mut scorer = automaton_weight
             .scorer(searcher.segment_reader(0u32), 1.32f32)
             .unwrap();
+        assert!(scorer.advance());
         assert_eq!(scorer.doc(), 0u32);
         assert_eq!(scorer.score(), 1.32f32);
     }

src/query/bitset/mod.rs

@@ -1,6 +1,7 @@
 use crate::common::{BitSet, TinySet};
-use crate::docset::{DocSet, TERMINATED};
+use crate::docset::{DocSet, SkipResult};
 use crate::DocId;
+use std::cmp::Ordering;
 
 /// A `BitSetDocSet` makes it possible to iterate through a bitset as if it was a `DocSet`.
 ///
@@ -32,50 +33,75 @@ impl From<BitSet> for BitSetDocSet {
         } else {
             docs.tinyset(0)
         };
-        let mut docset = BitSetDocSet {
+        BitSetDocSet {
             docs,
             cursor_bucket: 0,
             cursor_tinybitset: first_tiny_bitset,
             doc: 0u32,
-        };
-        docset.advance();
-        docset
+        }
     }
 }
 
 impl DocSet for BitSetDocSet {
-    fn advance(&mut self) -> DocId {
+    fn advance(&mut self) -> bool {
         if let Some(lower) = self.cursor_tinybitset.pop_lowest() {
             self.doc = (self.cursor_bucket as u32 * 64u32) | lower;
-            return self.doc;
+            return true;
         }
         if let Some(cursor_bucket) = self.docs.first_non_empty_bucket(self.cursor_bucket + 1) {
             self.go_to_bucket(cursor_bucket);
             let lower = self.cursor_tinybitset.pop_lowest().unwrap();
             self.doc = (cursor_bucket * 64u32) | lower;
-            self.doc
+            true
         } else {
-            self.doc = TERMINATED;
-            TERMINATED
+            false
         }
     }
 
-    fn seek(&mut self, target: DocId) -> DocId {
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        // skip is required to advance.
+        if !self.advance() {
+            return SkipResult::End;
+        }
         let target_bucket = target / 64u32;
 
         // Mask for all of the bits greater or equal
         // to our target document.
-        if target_bucket > self.cursor_bucket {
-            self.go_to_bucket(target_bucket);
-            let greater_filter: TinySet = TinySet::range_greater_or_equal(target);
-            self.cursor_tinybitset = self.cursor_tinybitset.intersect(greater_filter);
-            self.advance();
+        match target_bucket.cmp(&self.cursor_bucket) {
+            Ordering::Greater => {
+                self.go_to_bucket(target_bucket);
+                let greater_filter: TinySet = TinySet::range_greater_or_equal(target);
+                self.cursor_tinybitset = self.cursor_tinybitset.intersect(greater_filter);
+                if !self.advance() {
+                    SkipResult::End
+                } else if self.doc() == target {
+                    SkipResult::Reached
+                } else {
+                    debug_assert!(self.doc() > target);
+                    SkipResult::OverStep
+                }
+            }
+            Ordering::Equal => loop {
+                match self.doc().cmp(&target) {
+                    Ordering::Less => {
+                        if !self.advance() {
+                            return SkipResult::End;
+                        }
+                    }
+                    Ordering::Equal => {
+                        return SkipResult::Reached;
+                    }
+                    Ordering::Greater => {
+                        debug_assert!(self.doc() > target);
+                        return SkipResult::OverStep;
+                    }
+                }
+            },
+            Ordering::Less => {
+                debug_assert!(self.doc() > target);
+                SkipResult::OverStep
+            }
         }
-        let mut doc = self.doc();
-        while doc < target {
-            doc = self.advance();
-        }
-        doc
     }
 
     /// Returns the current document
@@ -96,7 +122,7 @@ impl DocSet for BitSetDocSet {
 mod tests {
     use super::BitSetDocSet;
     use crate::common::BitSet;
-    use crate::docset::{DocSet, TERMINATED};
+    use crate::docset::{DocSet, SkipResult};
     use crate::DocId;
 
     fn create_docbitset(docs: &[DocId], max_doc: DocId) -> BitSetDocSet {
@@ -107,24 +133,19 @@ mod tests {
         BitSetDocSet::from(docset)
     }
 
-    #[test]
-    fn test_empty() {
-        let bitset = BitSet::with_max_value(1000);
-        let mut empty = BitSetDocSet::from(bitset);
-        assert_eq!(empty.advance(), TERMINATED)
-    }
-
     fn test_go_through_sequential(docs: &[DocId]) {
         let mut docset = create_docbitset(docs, 1_000u32);
         for &doc in docs {
+            assert!(docset.advance());
             assert_eq!(doc, docset.doc());
-            docset.advance();
         }
-        assert_eq!(docset.advance(), TERMINATED);
+        assert!(!docset.advance());
+        assert!(!docset.advance());
     }
 
     #[test]
     fn test_docbitset_sequential() {
+        test_go_through_sequential(&[]);
         test_go_through_sequential(&[1, 2, 3]);
         test_go_through_sequential(&[1, 2, 3, 4, 5, 63, 64, 65]);
         test_go_through_sequential(&[63, 64, 65]);
@@ -135,64 +156,64 @@ mod tests {
     fn test_docbitset_skip() {
         {
             let mut docset = create_docbitset(&[1, 5, 6, 7, 5112], 10_000);
-            assert_eq!(docset.seek(7), 7);
+            assert_eq!(docset.skip_next(7), SkipResult::Reached);
             assert_eq!(docset.doc(), 7);
-            assert_eq!(docset.advance(), 5112);
+            assert!(docset.advance(), 7);
             assert_eq!(docset.doc(), 5112);
-            assert_eq!(docset.advance(), TERMINATED);
+            assert!(!docset.advance());
         }
         {
             let mut docset = create_docbitset(&[1, 5, 6, 7, 5112], 10_000);
-            assert_eq!(docset.seek(3), 5);
+            assert_eq!(docset.skip_next(3), SkipResult::OverStep);
             assert_eq!(docset.doc(), 5);
-            assert_eq!(docset.advance(), 6);
+            assert!(docset.advance());
         }
         {
             let mut docset = create_docbitset(&[5112], 10_000);
-            assert_eq!(docset.seek(5112), 5112);
+            assert_eq!(docset.skip_next(5112), SkipResult::Reached);
             assert_eq!(docset.doc(), 5112);
-            assert_eq!(docset.advance(), TERMINATED);
+            assert!(!docset.advance());
         }
         {
             let mut docset = create_docbitset(&[5112], 10_000);
-            assert_eq!(docset.seek(5113), TERMINATED);
-            assert_eq!(docset.advance(), TERMINATED);
+            assert_eq!(docset.skip_next(5113), SkipResult::End);
+            assert!(!docset.advance());
         }
         {
             let mut docset = create_docbitset(&[5112], 10_000);
-            assert_eq!(docset.seek(5111), 5112);
+            assert_eq!(docset.skip_next(5111), SkipResult::OverStep);
             assert_eq!(docset.doc(), 5112);
-            assert_eq!(docset.advance(), TERMINATED);
+            assert!(!docset.advance());
         }
         {
             let mut docset = create_docbitset(&[1, 5, 6, 7, 5112, 5500, 6666], 10_000);
-            assert_eq!(docset.seek(5112), 5112);
+            assert_eq!(docset.skip_next(5112), SkipResult::Reached);
             assert_eq!(docset.doc(), 5112);
-            assert_eq!(docset.advance(), 5500);
+            assert!(docset.advance());
             assert_eq!(docset.doc(), 5500);
-            assert_eq!(docset.advance(), 6666);
+            assert!(docset.advance());
             assert_eq!(docset.doc(), 6666);
-            assert_eq!(docset.advance(), TERMINATED);
+            assert!(!docset.advance());
         }
         {
             let mut docset = create_docbitset(&[1, 5, 6, 7, 5112, 5500, 6666], 10_000);
-            assert_eq!(docset.seek(5111), 5112);
+            assert_eq!(docset.skip_next(5111), SkipResult::OverStep);
             assert_eq!(docset.doc(), 5112);
-            assert_eq!(docset.advance(), 5500);
+            assert!(docset.advance());
             assert_eq!(docset.doc(), 5500);
-            assert_eq!(docset.advance(), 6666);
+            assert!(docset.advance());
             assert_eq!(docset.doc(), 6666);
-            assert_eq!(docset.advance(), TERMINATED);
+            assert!(!docset.advance());
         }
         {
             let mut docset = create_docbitset(&[1, 5, 6, 7, 5112, 5513, 6666], 10_000);
-            assert_eq!(docset.seek(5111), 5112);
+            assert_eq!(docset.skip_next(5111), SkipResult::OverStep);
             assert_eq!(docset.doc(), 5112);
-            assert_eq!(docset.advance(), 5513);
+            assert!(docset.advance());
             assert_eq!(docset.doc(), 5513);
-            assert_eq!(docset.advance(), 6666);
+            assert!(docset.advance());
             assert_eq!(docset.doc(), 6666);
-            assert_eq!(docset.advance(), TERMINATED);
+            assert!(!docset.advance());
         }
     }
 }
@@ -202,7 +223,6 @@ mod bench {
 
     use super::BitSet;
     use super::BitSetDocSet;
-    use crate::docset::TERMINATED;
     use crate::test;
     use crate::tests;
     use crate::DocSet;
@@ -237,7 +257,7 @@ mod bench {
         }
         b.iter(|| {
             let mut docset = BitSetDocSet::from(bitset.clone());
-            while docset.advance() != TERMINATED {}
+            while docset.advance() {}
         });
     }
 }

src/query/boolean_query/boolean_weight.rs

@@ -2,7 +2,6 @@ use crate::core::SegmentReader;
 use crate::query::explanation::does_not_match;
 use crate::query::score_combiner::{DoNothingCombiner, ScoreCombiner, SumWithCoordsCombiner};
 use crate::query::term_query::TermScorer;
-use crate::query::weight::{for_each_pruning_scorer, for_each_scorer};
 use crate::query::EmptyScorer;
 use crate::query::Exclude;
 use crate::query::Occur;
@@ -11,21 +10,16 @@ use crate::query::Scorer;
 use crate::query::Union;
 use crate::query::Weight;
 use crate::query::{intersect_scorers, Explanation};
-use crate::{DocId, Score};
+use crate::{DocId, SkipResult};
 use std::collections::HashMap;
 
-enum SpecializedScorer<TScoreCombiner: ScoreCombiner> {
-    TermUnion(Union<TermScorer, TScoreCombiner>),
-    Other(Box<dyn Scorer>),
-}
-
-fn scorer_union<TScoreCombiner>(scorers: Vec<Box<dyn Scorer>>) -> SpecializedScorer<TScoreCombiner>
+fn scorer_union<TScoreCombiner>(scorers: Vec<Box<dyn Scorer>>) -> Box<dyn Scorer>
 where
     TScoreCombiner: ScoreCombiner,
 {
     assert!(!scorers.is_empty());
     if scorers.len() == 1 {
-        return SpecializedScorer::Other(scorers.into_iter().next().unwrap()); //< we checked the size beforehands
+        return scorers.into_iter().next().unwrap(); //< we checked the size beforehands
     }
 
     {
@@ -35,21 +29,14 @@ where
                 .into_iter()
                 .map(|scorer| *(scorer.downcast::<TermScorer>().map_err(|_| ()).unwrap()))
                 .collect();
-            return SpecializedScorer::TermUnion(Union::<TermScorer, TScoreCombiner>::from(
-                scorers,
-            ));
+            let scorer: Box<dyn Scorer> =
+                Box::new(Union::<TermScorer, TScoreCombiner>::from(scorers));
+            return scorer;
         }
     }
-    SpecializedScorer::Other(Box::new(Union::<_, TScoreCombiner>::from(scorers)))
-}
 
-impl<TScoreCombiner: ScoreCombiner> Into<Box<dyn Scorer>> for SpecializedScorer<TScoreCombiner> {
-    fn into(self) -> Box<dyn Scorer> {
-        match self {
-            Self::TermUnion(union) => Box::new(union),
-            Self::Other(scorer) => scorer,
-        }
-    }
+    let scorer: Box<dyn Scorer> = Box::new(Union::<_, TScoreCombiner>::from(scorers));
+    scorer
 }
 
 pub struct BooleanWeight {
@@ -85,50 +72,41 @@ impl BooleanWeight {
         &self,
         reader: &SegmentReader,
         boost: f32,
-    ) -> crate::Result<SpecializedScorer<TScoreCombiner>> {
+    ) -> crate::Result<Box<dyn Scorer>> {
         let mut per_occur_scorers = self.per_occur_scorers(reader, boost)?;
 
-        let should_scorer_opt: Option<SpecializedScorer<TScoreCombiner>> = per_occur_scorers
+        let should_scorer_opt: Option<Box<dyn Scorer>> = per_occur_scorers
             .remove(&Occur::Should)
             .map(scorer_union::<TScoreCombiner>);
 
         let exclude_scorer_opt: Option<Box<dyn Scorer>> = per_occur_scorers
             .remove(&Occur::MustNot)
-            .map(scorer_union::<TScoreCombiner>)
-            .map(Into::into);
+            .map(scorer_union::<TScoreCombiner>);
 
         let must_scorer_opt: Option<Box<dyn Scorer>> = per_occur_scorers
             .remove(&Occur::Must)
             .map(intersect_scorers);
 
-        let positive_scorer: SpecializedScorer<TScoreCombiner> =
-            match (should_scorer_opt, must_scorer_opt) {
-                (Some(should_scorer), Some(must_scorer)) => {
-                    if self.scoring_enabled {
-                        SpecializedScorer::Other(Box::new(RequiredOptionalScorer::<
-                            Box<dyn Scorer>,
-                            Box<dyn Scorer>,
-                            TScoreCombiner,
-                        >::new(
-                            must_scorer, should_scorer.into()
-                        )))
-                    } else {
-                        SpecializedScorer::Other(must_scorer)
-                    }
+        let positive_scorer: Box<dyn Scorer> = match (should_scorer_opt, must_scorer_opt) {
+            (Some(should_scorer), Some(must_scorer)) => {
+                if self.scoring_enabled {
+                    Box::new(RequiredOptionalScorer::<_, _, TScoreCombiner>::new(
+                        must_scorer,
+                        should_scorer,
+                    ))
+                } else {
+                    must_scorer
                 }
-                (None, Some(must_scorer)) => SpecializedScorer::Other(must_scorer),
-                (Some(should_scorer), None) => should_scorer,
-                (None, None) => {
-                    return Ok(SpecializedScorer::Other(Box::new(EmptyScorer)));
-                }
-            };
+            }
+            (None, Some(must_scorer)) => must_scorer,
+            (Some(should_scorer), None) => should_scorer,
+            (None, None) => {
+                return Ok(Box::new(EmptyScorer));
+            }
+        };
 
         if let Some(exclude_scorer) = exclude_scorer_opt {
-            let positive_scorer_boxed: Box<dyn Scorer> = positive_scorer.into();
-            Ok(SpecializedScorer::Other(Box::new(Exclude::new(
-                positive_scorer_boxed,
-                exclude_scorer,
-            ))))
+            Ok(Box::new(Exclude::new(positive_scorer, exclude_scorer)))
         } else {
             Ok(positive_scorer)
         }
@@ -148,16 +126,14 @@ impl Weight for BooleanWeight {
             }
         } else if self.scoring_enabled {
             self.complex_scorer::<SumWithCoordsCombiner>(reader, boost)
-                .map(Into::into)
         } else {
             self.complex_scorer::<DoNothingCombiner>(reader, boost)
-                .map(Into::into)
         }
     }
 
     fn explain(&self, reader: &SegmentReader, doc: DocId) -> crate::Result<Explanation> {
         let mut scorer = self.scorer(reader, 1.0f32)?;
-        if scorer.seek(doc) != doc {
+        if scorer.skip_next(doc) != SkipResult::Reached {
             return Err(does_not_match(doc));
         }
         if !self.scoring_enabled {
@@ -174,51 +150,6 @@ impl Weight for BooleanWeight {
         }
         Ok(explanation)
     }
-
-    fn for_each(
-        &self,
-        reader: &SegmentReader,
-        callback: &mut dyn FnMut(DocId, Score),
-    ) -> crate::Result<()> {
-        let scorer = self.complex_scorer::<SumWithCoordsCombiner>(reader, 1.0f32)?;
-        match scorer {
-            SpecializedScorer::TermUnion(mut union_scorer) => {
-                for_each_scorer(&mut union_scorer, callback);
-            }
-            SpecializedScorer::Other(mut scorer) => {
-                for_each_scorer(scorer.as_mut(), callback);
-            }
-        }
-        Ok(())
-    }
-
-    /// Calls `callback` with all of the `(doc, score)` for which score
-    /// is exceeding a given threshold.
-    ///
-    /// This method is useful for the TopDocs collector.
-    /// For all docsets, the blanket implementation has the benefit
-    /// of prefiltering (doc, score) pairs, avoiding the
-    /// virtual dispatch cost.
-    ///
-    /// More importantly, it makes it possible for scorers to implement
-    /// important optimization (e.g. BlockWAND for union).
-    fn for_each_pruning(
-        &self,
-        threshold: f32,
-        reader: &SegmentReader,
-        callback: &mut dyn FnMut(DocId, Score) -> Score,
-    ) -> crate::Result<()> {
-        let scorer = self.complex_scorer::<SumWithCoordsCombiner>(reader, 1.0f32)?;
-        match scorer {
-            SpecializedScorer::TermUnion(mut union_scorer) => {
-                for_each_pruning_scorer(&mut union_scorer, threshold, callback);
-            }
-            SpecializedScorer::Other(mut scorer) => {
-                for_each_pruning_scorer(scorer.as_mut(), threshold, callback);
-            }
-        }
-        Ok(())
-    }
 }
 
 fn is_positive_occur(occur: Occur) -> bool {

src/query/boolean_query/mod.rs

@@ -31,11 +31,24 @@ mod tests {
             // writing the segment
             let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
             {
-                index_writer.add_document(doc!(text_field => "a 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"));
+                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);
             }
             assert!(index_writer.commit().is_ok());
         }
@@ -207,6 +220,7 @@ mod tests {
             let mut boolean_scorer = boolean_weight
                 .scorer(searcher.segment_reader(0u32), 1.0f32)
                 .unwrap();
+            assert!(boolean_scorer.advance());
             assert_eq!(boolean_scorer.doc(), 0u32);
             assert_nearly_equals(boolean_scorer.score(), 0.84163445f32);
         }
@@ -214,6 +228,7 @@ mod tests {
             let mut boolean_scorer = boolean_weight
                 .scorer(searcher.segment_reader(0u32), 2.0f32)
                 .unwrap();
+            assert!(boolean_scorer.advance());
             assert_eq!(boolean_scorer.doc(), 0u32);
             assert_nearly_equals(boolean_scorer.score(), 1.6832689f32);
         }

src/query/boost_query.rs

@@ -1,7 +1,8 @@
+use crate::common::BitSet;
 use crate::fastfield::DeleteBitSet;
 use crate::query::explanation::does_not_match;
 use crate::query::{Explanation, Query, Scorer, Weight};
-use crate::{DocId, DocSet, Searcher, SegmentReader, Term};
+use crate::{DocId, DocSet, Searcher, SegmentReader, SkipResult, Term};
 use std::collections::BTreeSet;
 use std::fmt;
 
@@ -71,7 +72,7 @@ impl Weight for BoostWeight {
 
     fn explain(&self, reader: &SegmentReader, doc: u32) -> crate::Result<Explanation> {
         let mut scorer = self.scorer(reader, 1.0f32)?;
-        if scorer.seek(doc) != doc {
+        if scorer.skip_next(doc) != SkipResult::Reached {
             return Err(does_not_match(doc));
         }
         let mut explanation =
@@ -98,12 +99,12 @@ impl<S: Scorer> BoostScorer<S> {
 }
 
 impl<S: Scorer> DocSet for BoostScorer<S> {
-    fn advance(&mut self) -> DocId {
+    fn advance(&mut self) -> bool {
         self.underlying.advance()
     }
 
-    fn seek(&mut self, target: DocId) -> DocId {
-        self.underlying.seek(target)
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        self.underlying.skip_next(target)
     }
 
     fn fill_buffer(&mut self, buffer: &mut [DocId]) -> usize {
@@ -118,6 +119,10 @@ impl<S: Scorer> DocSet for BoostScorer<S> {
         self.underlying.size_hint()
     }
 
+    fn append_to_bitset(&mut self, bitset: &mut BitSet) {
+        self.underlying.append_to_bitset(bitset)
+    }
+
     fn count(&mut self, delete_bitset: &DeleteBitSet) -> u32 {
         self.underlying.count(delete_bitset)
     }

src/query/empty_query.rs

@@ -1,5 +1,4 @@
 use super::Scorer;
-use crate::docset::TERMINATED;
 use crate::query::explanation::does_not_match;
 use crate::query::Weight;
 use crate::query::{Explanation, Query};
@@ -49,12 +48,15 @@ impl Weight for EmptyWeight {
 pub struct EmptyScorer;
 
 impl DocSet for EmptyScorer {
-    fn advance(&mut self) -> DocId {
-        TERMINATED
+    fn advance(&mut self) -> bool {
+        false
     }
 
     fn doc(&self) -> DocId {
-        TERMINATED
+        panic!(
+            "You may not call .doc() on a scorer \
+             where the last call to advance() did not return true."
+        );
     }
 
     fn size_hint(&self) -> u32 {
@@ -70,15 +72,18 @@ impl Scorer for EmptyScorer {
 
 #[cfg(test)]
 mod tests {
-    use crate::docset::TERMINATED;
     use crate::query::EmptyScorer;
     use crate::DocSet;
 
     #[test]
     fn test_empty_scorer() {
         let mut empty_scorer = EmptyScorer;
-        assert_eq!(empty_scorer.doc(), TERMINATED);
-        assert_eq!(empty_scorer.advance(), TERMINATED);
-        assert_eq!(empty_scorer.doc(), TERMINATED);
+        assert!(!empty_scorer.advance());
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_empty_scorer_panic_on_doc_call() {
+        EmptyScorer.doc();
     }
 }

src/query/exclude.rs

@@ -1,37 +1,41 @@
-use crate::docset::{DocSet, TERMINATED};
+use crate::docset::{DocSet, SkipResult};
 use crate::query::Scorer;
 use crate::DocId;
 use crate::Score;
 
+#[derive(Clone, Copy, Debug)]
+enum State {
+    ExcludeOne(DocId),
+    Finished,
+}
+
 /// Filters a given `DocSet` by removing the docs from a given `DocSet`.
 ///
 /// The excluding docset has no impact on scoring.
 pub struct Exclude<TDocSet, TDocSetExclude> {
     underlying_docset: TDocSet,
     excluding_docset: TDocSetExclude,
+    excluding_state: State,
 }
 
 impl<TDocSet, TDocSetExclude> Exclude<TDocSet, TDocSetExclude>
 where
-    TDocSet: DocSet,
     TDocSetExclude: DocSet,
 {
     /// Creates a new `ExcludeScorer`
     pub fn new(
-        mut underlying_docset: TDocSet,
+        underlying_docset: TDocSet,
         mut excluding_docset: TDocSetExclude,
     ) -> Exclude<TDocSet, TDocSetExclude> {
-        while underlying_docset.doc() != TERMINATED {
-            let target = underlying_docset.doc();
-            if excluding_docset.seek(target) != target {
-                // this document is not excluded.
-                break;
-            }
-            underlying_docset.advance();
-        }
+        let state = if excluding_docset.advance() {
+            State::ExcludeOne(excluding_docset.doc())
+        } else {
+            State::Finished
+        };
         Exclude {
             underlying_docset,
             excluding_docset,
+            excluding_state: state,
         }
     }
 }
@@ -47,7 +51,28 @@ where
     /// increasing `doc`.
     fn accept(&mut self) -> bool {
         let doc = self.underlying_docset.doc();
-        self.excluding_docset.seek(doc) != doc
+        match self.excluding_state {
+            State::ExcludeOne(excluded_doc) => {
+                if doc == excluded_doc {
+                    return false;
+                }
+                if excluded_doc > doc {
+                    return true;
+                }
+                match self.excluding_docset.skip_next(doc) {
+                    SkipResult::OverStep => {
+                        self.excluding_state = State::ExcludeOne(self.excluding_docset.doc());
+                        true
+                    }
+                    SkipResult::End => {
+                        self.excluding_state = State::Finished;
+                        true
+                    }
+                    SkipResult::Reached => false,
+                }
+            }
+            State::Finished => true,
+        }
     }
 }
 
@@ -56,24 +81,27 @@ where
     TDocSet: DocSet,
     TDocSetExclude: DocSet,
 {
-    fn advance(&mut self) -> DocId {
-        while self.underlying_docset.advance() != TERMINATED {
+    fn advance(&mut self) -> bool {
+        while self.underlying_docset.advance() {
             if self.accept() {
-                return self.doc();
+                return true;
             }
         }
-        TERMINATED
+        false
     }
 
-    fn seek(&mut self, target: DocId) -> DocId {
-        let underlying_seek_result = self.underlying_docset.seek(target);
-        if underlying_seek_result == TERMINATED {
-            return TERMINATED;
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        let underlying_skip_result = self.underlying_docset.skip_next(target);
+        if underlying_skip_result == SkipResult::End {
+            return SkipResult::End;
         }
         if self.accept() {
-            return underlying_seek_result;
+            underlying_skip_result
+        } else if self.advance() {
+            SkipResult::OverStep
+        } else {
+            SkipResult::End
         }
-        self.advance()
     }
 
     fn doc(&self) -> DocId {
@@ -113,9 +141,8 @@ mod tests {
             VecDocSet::from(vec![1, 2, 3, 10, 16, 24]),
         );
         let mut els = vec![];
-        while exclude_scorer.doc() != TERMINATED {
+        while exclude_scorer.advance() {
             els.push(exclude_scorer.doc());
-            exclude_scorer.advance();
         }
         assert_eq!(els, vec![5, 8, 15]);
     }

src/query/explanation.rs

@@ -1,5 +1,4 @@
 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

@@ -2,36 +2,10 @@ use crate::query::{AutomatonWeight, Query, Weight};
 use crate::schema::Term;
 use crate::Searcher;
 use crate::TantivyError::InvalidArgument;
-use levenshtein_automata::{Distance, LevenshteinAutomatonBuilder, DFA};
+use levenshtein_automata::{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
@@ -117,7 +91,7 @@ impl FuzzyTermQuery {
         }
     }
 
-    /// Creates a new Fuzzy Query of the Term prefix
+    /// Creates a new Fuzzy Query that treats transpositions as cost one rather than two
     pub fn new_prefix(term: Term, distance: u8, transposition_cost_one: bool) -> FuzzyTermQuery {
         FuzzyTermQuery {
             term,
@@ -127,20 +101,13 @@ impl FuzzyTermQuery {
         }
     }
 
-    fn specialized_weight(&self) -> crate::Result<AutomatonWeight<DFAWrapper>> {
+    fn specialized_weight(&self) -> crate::Result<AutomatonWeight<DFA>> {
         // 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 = 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),
-                ))
+                let automaton = automaton_builder.build_dfa(self.term.text());
+                Ok(AutomatonWeight::new(self.term.field(), automaton))
             }
             None => Err(InvalidArgument(format!(
                 "Levenshtein distance of {} is not allowed. Choose a value in the {:?} range",
@@ -188,8 +155,6 @@ mod test {
         }
         let reader = index.reader().unwrap();
         let searcher = reader.searcher();
-
-        // passes because Levenshtein distance is 1 (substitute 'o' with 'a')
         {
             let term = Term::from_field_text(country_field, "japon");
 
@@ -201,29 +166,5 @@ mod test {
             let (score, _) = top_docs[0];
             assert_nearly_equals(1f32, score);
         }
-
-        // fails because non-prefix Levenshtein distance is more than 1 (add 'a' and 'n')
-        {
-            let term = Term::from_field_text(country_field, "jap");
-
-            let fuzzy_query = FuzzyTermQuery::new(term, 1, true);
-            let top_docs = searcher
-                .search(&fuzzy_query, &TopDocs::with_limit(2))
-                .unwrap();
-            assert_eq!(top_docs.len(), 0, "Expected no document");
-        }
-
-        // passes because prefix Levenshtein distance is 0
-        {
-            let term = Term::from_field_text(country_field, "jap");
-
-            let fuzzy_query = FuzzyTermQuery::new_prefix(term, 1, true);
-            let top_docs = searcher
-                .search(&fuzzy_query, &TopDocs::with_limit(2))
-                .unwrap();
-            assert_eq!(top_docs.len(), 1, "Expected only 1 document");
-            let (score, _) = top_docs[0];
-            assert_nearly_equals(1f32, score);
-        }
     }
 }

src/query/intersection.rs

@@ -1,4 +1,4 @@
-use crate::docset::{DocSet, TERMINATED};
+use crate::docset::{DocSet, SkipResult};
 use crate::query::term_query::TermScorer;
 use crate::query::EmptyScorer;
 use crate::query::Scorer;
@@ -20,14 +20,12 @@ pub fn intersect_scorers(mut scorers: Vec<Box<dyn Scorer>>) -> Box<dyn Scorer> {
     if scorers.len() == 1 {
         return scorers.pop().unwrap();
     }
-    scorers.sort_by_key(|scorer| scorer.size_hint());
-    let doc = go_to_first_doc(&mut scorers[..]);
-    if doc == TERMINATED {
-        return Box::new(EmptyScorer);
-    }
     // We know that we have at least 2 elements.
-    let left = scorers.remove(0);
-    let right = scorers.remove(0);
+    let num_docsets = scorers.len();
+    scorers.sort_by(|left, right| right.size_hint().cmp(&left.size_hint()));
+    let left = scorers.pop().unwrap();
+    let right = scorers.pop().unwrap();
+    scorers.reverse();
     let all_term_scorers = [&left, &right]
         .iter()
         .all(|&scorer| scorer.is::<TermScorer>());
@@ -36,12 +34,14 @@ pub fn intersect_scorers(mut scorers: Vec<Box<dyn Scorer>>) -> Box<dyn Scorer> {
             left: *(left.downcast::<TermScorer>().map_err(|_| ()).unwrap()),
             right: *(right.downcast::<TermScorer>().map_err(|_| ()).unwrap()),
             others: scorers,
+            num_docsets,
         });
     }
     Box::new(Intersection {
         left,
         right,
         others: scorers,
+        num_docsets,
     })
 }
 
@@ -50,34 +50,22 @@ pub struct Intersection<TDocSet: DocSet, TOtherDocSet: DocSet = Box<dyn Scorer>>
     left: TDocSet,
     right: TDocSet,
     others: Vec<TOtherDocSet>,
-}
-
-fn go_to_first_doc<TDocSet: DocSet>(docsets: &mut [TDocSet]) -> DocId {
-    let mut candidate = 0;
-    'outer: loop {
-        for docset in docsets.iter_mut() {
-            let seek_doc = docset.seek(candidate);
-            if seek_doc > candidate {
-                candidate = docset.doc();
-                continue 'outer;
-            }
-        }
-        return candidate;
-    }
+    num_docsets: usize,
 }
 
 impl<TDocSet: DocSet> Intersection<TDocSet, TDocSet> {
     pub(crate) fn new(mut docsets: Vec<TDocSet>) -> Intersection<TDocSet, TDocSet> {
         let num_docsets = docsets.len();
         assert!(num_docsets >= 2);
-        docsets.sort_by_key(|docset| docset.size_hint());
-        go_to_first_doc(&mut docsets);
-        let left = docsets.remove(0);
-        let right = docsets.remove(0);
+        docsets.sort_by(|left, right| right.size_hint().cmp(&left.size_hint()));
+        let left = docsets.pop().unwrap();
+        let right = docsets.pop().unwrap();
+        docsets.reverse();
         Intersection {
             left,
             right,
             others: docsets,
+            num_docsets,
         }
     }
 }
@@ -92,44 +80,128 @@ impl<TDocSet: DocSet> Intersection<TDocSet, TDocSet> {
     }
 }
 
+impl<TDocSet: DocSet, TOtherDocSet: DocSet> Intersection<TDocSet, TOtherDocSet> {
+    pub(crate) fn docset_mut(&mut self, ord: usize) -> &mut dyn DocSet {
+        match ord {
+            0 => &mut self.left,
+            1 => &mut self.right,
+            n => &mut self.others[n - 2],
+        }
+    }
+}
+
 impl<TDocSet: DocSet, TOtherDocSet: DocSet> DocSet for Intersection<TDocSet, TOtherDocSet> {
-    fn advance(&mut self) -> DocId {
+    fn advance(&mut self) -> bool {
         let (left, right) = (&mut self.left, &mut self.right);
-        let mut candidate = left.advance();
+
+        if !left.advance() {
+            return false;
+        }
+
+        let mut candidate = left.doc();
+        let mut other_candidate_ord: usize = usize::max_value();
 
         'outer: loop {
             // In the first part we look for a document in the intersection
             // of the two rarest `DocSet` in the intersection.
-
             loop {
-                let right_doc = right.seek(candidate);
-                candidate = left.seek(right_doc);
-                if candidate == right_doc {
-                    break;
+                match right.skip_next(candidate) {
+                    SkipResult::Reached => {
+                        break;
+                    }
+                    SkipResult::OverStep => {
+                        candidate = right.doc();
+                        other_candidate_ord = usize::max_value();
+                    }
+                    SkipResult::End => {
+                        return false;
+                    }
+                }
+                match left.skip_next(candidate) {
+                    SkipResult::Reached => {
+                        break;
+                    }
+                    SkipResult::OverStep => {
+                        candidate = left.doc();
+                        other_candidate_ord = usize::max_value();
+                    }
+                    SkipResult::End => {
+                        return false;
+                    }
                 }
             }
-
-            debug_assert_eq!(left.doc(), right.doc());
             // test the remaining scorers;
-            for docset in self.others.iter_mut() {
-                let seek_doc = docset.seek(candidate);
-                if seek_doc > candidate {
-                    candidate = left.seek(seek_doc);
-                    continue 'outer;
+            for (ord, docset) in self.others.iter_mut().enumerate() {
+                if ord == other_candidate_ord {
+                    continue;
+                }
+                // `candidate_ord` is already at the
+                // right position.
+                //
+                // Calling `skip_next` would advance this docset
+                // and miss it.
+                match docset.skip_next(candidate) {
+                    SkipResult::Reached => {}
+                    SkipResult::OverStep => {
+                        // this is not in the intersection,
+                        // let's update our candidate.
+                        candidate = docset.doc();
+                        match left.skip_next(candidate) {
+                            SkipResult::Reached => {
+                                other_candidate_ord = ord;
+                            }
+                            SkipResult::OverStep => {
+                                candidate = left.doc();
+                                other_candidate_ord = usize::max_value();
+                            }
+                            SkipResult::End => {
+                                return false;
+                            }
+                        }
+                        continue 'outer;
+                    }
+                    SkipResult::End => {
+                        return false;
+                    }
                 }
             }
-
-            return candidate;
+            return true;
         }
     }
 
-    fn seek(&mut self, target: DocId) -> DocId {
-        self.left.seek(target);
-        let mut docsets: Vec<&mut dyn DocSet> = vec![&mut self.left, &mut self.right];
-        for docset in &mut self.others {
-            docsets.push(docset);
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        // We optimize skipping by skipping every single member
+        // of the intersection to target.
+        let mut current_target: DocId = target;
+        let mut current_ord = self.num_docsets;
+
+        'outer: loop {
+            for ord in 0..self.num_docsets {
+                let docset = self.docset_mut(ord);
+                if ord == current_ord {
+                    continue;
+                }
+                match docset.skip_next(current_target) {
+                    SkipResult::End => {
+                        return SkipResult::End;
+                    }
+                    SkipResult::OverStep => {
+                        // update the target
+                        // for the remaining members of the intersection.
+                        current_target = docset.doc();
+                        current_ord = ord;
+                        continue 'outer;
+                    }
+                    SkipResult::Reached => {}
+                }
+            }
+            if target == current_target {
+                return SkipResult::Reached;
+            } else {
+                assert!(current_target > target);
+                return SkipResult::OverStep;
+            }
         }
-        go_to_first_doc(&mut docsets[..])
     }
 
     fn doc(&self) -> DocId {
@@ -156,7 +228,7 @@ where
 #[cfg(test)]
 mod tests {
     use super::Intersection;
-    use crate::docset::{DocSet, TERMINATED};
+    use crate::docset::{DocSet, SkipResult};
     use crate::postings::tests::test_skip_against_unoptimized;
     use crate::query::VecDocSet;
 
@@ -166,18 +238,20 @@ mod tests {
             let left = VecDocSet::from(vec![1, 3, 9]);
             let right = VecDocSet::from(vec![3, 4, 9, 18]);
             let mut intersection = Intersection::new(vec![left, right]);
+            assert!(intersection.advance());
             assert_eq!(intersection.doc(), 3);
-            assert_eq!(intersection.advance(), 9);
+            assert!(intersection.advance());
             assert_eq!(intersection.doc(), 9);
-            assert_eq!(intersection.advance(), TERMINATED);
+            assert!(!intersection.advance());
         }
         {
             let a = VecDocSet::from(vec![1, 3, 9]);
             let b = VecDocSet::from(vec![3, 4, 9, 18]);
             let c = VecDocSet::from(vec![1, 5, 9, 111]);
             let mut intersection = Intersection::new(vec![a, b, c]);
+            assert!(intersection.advance());
             assert_eq!(intersection.doc(), 9);
-            assert_eq!(intersection.advance(), TERMINATED);
+            assert!(!intersection.advance());
         }
     }
 
@@ -186,8 +260,8 @@ mod tests {
         let left = VecDocSet::from(vec![0]);
         let right = VecDocSet::from(vec![0]);
         let mut intersection = Intersection::new(vec![left, right]);
+        assert!(intersection.advance());
         assert_eq!(intersection.doc(), 0);
-        assert_eq!(intersection.advance(), TERMINATED);
     }
 
     #[test]
@@ -195,7 +269,7 @@ mod tests {
         let left = VecDocSet::from(vec![0, 1, 2, 4]);
         let right = VecDocSet::from(vec![2, 5]);
         let mut intersection = Intersection::new(vec![left, right]);
-        assert_eq!(intersection.seek(2), 2);
+        assert_eq!(intersection.skip_next(2), SkipResult::Reached);
         assert_eq!(intersection.doc(), 2);
     }
 
@@ -238,7 +312,7 @@ mod tests {
         let a = VecDocSet::from(vec![1, 3]);
         let b = VecDocSet::from(vec![1, 4]);
         let c = VecDocSet::from(vec![3, 9]);
-        let intersection = Intersection::new(vec![a, b, c]);
-        assert_eq!(intersection.doc(), TERMINATED);
+        let mut intersection = Intersection::new(vec![a, b, c]);
+        assert!(!intersection.advance());
     }
 }

src/query/mod.rs

@@ -1,4 +1,6 @@
-/*! Query Module */
+/*!
+Query
+*/
 
 mod all_query;
 mod automaton_weight;
@@ -40,8 +42,6 @@ pub use self::boost_query::BoostQuery;
 pub use self::empty_query::{EmptyQuery, EmptyScorer, EmptyWeight};
 pub use self::exclude::Exclude;
 pub use self::explanation::Explanation;
-#[cfg(test)]
-pub(crate) use self::fuzzy_query::DFAWrapper;
 pub use self::fuzzy_query::FuzzyTermQuery;
 pub use self::intersection::intersect_scorers;
 pub use self::phrase_query::PhraseQuery;

src/query/phrase_query/mod.rs

@@ -12,6 +12,7 @@ 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;
@@ -60,8 +61,8 @@ pub mod tests {
                 .map(|docaddr| docaddr.1)
                 .collect::<Vec<_>>()
         };
-        assert_eq!(test_query(vec!["a", "b"]), vec![1, 2, 3, 4]);
         assert_eq!(test_query(vec!["a", "b", "c"]), vec![2, 4]);
+        assert_eq!(test_query(vec!["a", "b"]), vec![1, 2, 3, 4]);
         assert_eq!(test_query(vec!["b", "b"]), vec![0, 1]);
         assert!(test_query(vec!["g", "ewrwer"]).is_empty());
         assert!(test_query(vec!["g", "a"]).is_empty());
@@ -126,16 +127,21 @@ pub mod tests {
             Term::from_field_text(text_field, "a"),
             Term::from_field_text(text_field, "b"),
         ]);
-
-        let search_result = searcher
+        match searcher
             .search(&phrase_query, &TEST_COLLECTOR_WITH_SCORE)
-            .map(|_| ());
-        assert!(matches!(
-            search_result,
-            Err(crate::TantivyError::SchemaError(msg))
-            if msg == "Applied phrase query on field \"text\", which does not have positions \
-            indexed"
-        ));
+            .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");
+            }
+        }
     }
 
     #[test]

src/query/phrase_query/phrase_scorer.rs

@@ -1,4 +1,4 @@
-use crate::docset::{DocSet, TERMINATED};
+use crate::docset::{DocSet, SkipResult};
 use crate::fieldnorm::FieldNormReader;
 use crate::postings::Postings;
 use crate::query::bm25::BM25Weight;
@@ -25,12 +25,12 @@ impl<TPostings: Postings> PostingsWithOffset<TPostings> {
 }
 
 impl<TPostings: Postings> DocSet for PostingsWithOffset<TPostings> {
-    fn advance(&mut self) -> DocId {
+    fn advance(&mut self) -> bool {
         self.postings.advance()
     }
 
-    fn seek(&mut self, target: DocId) -> DocId {
-        self.postings.seek(target)
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        self.postings.skip_next(target)
     }
 
     fn doc(&self) -> DocId {
@@ -149,7 +149,7 @@ impl<TPostings: Postings> PhraseScorer<TPostings> {
                 PostingsWithOffset::new(postings, (max_offset - offset) as u32)
             })
             .collect::<Vec<_>>();
-        let mut scorer = PhraseScorer {
+        PhraseScorer {
             intersection_docset: Intersection::new(postings_with_offsets),
             num_terms: num_docsets,
             left: Vec::with_capacity(100),
@@ -158,11 +158,7 @@ impl<TPostings: Postings> PhraseScorer<TPostings> {
             similarity_weight,
             fieldnorm_reader,
             score_needed,
-        };
-        if scorer.doc() != TERMINATED && !scorer.phrase_match() {
-            scorer.advance();
         }
-        scorer
     }
 
     pub fn phrase_count(&self) -> u32 {
@@ -229,21 +225,31 @@ impl<TPostings: Postings> PhraseScorer<TPostings> {
 }
 
 impl<TPostings: Postings> DocSet for PhraseScorer<TPostings> {
-    fn advance(&mut self) -> DocId {
-        loop {
-            let doc = self.intersection_docset.advance();
-            if doc == TERMINATED || self.phrase_match() {
-                return doc;
+    fn advance(&mut self) -> bool {
+        while self.intersection_docset.advance() {
+            if self.phrase_match() {
+                return true;
             }
         }
+        false
     }
 
-    fn seek(&mut self, target: DocId) -> DocId {
-        let doc = self.intersection_docset.seek(target);
-        if doc == TERMINATED || self.phrase_match() {
-            return doc;
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        if self.intersection_docset.skip_next(target) == SkipResult::End {
+            return SkipResult::End;
+        }
+        if self.phrase_match() {
+            if self.doc() == target {
+                return SkipResult::Reached;
+            } else {
+                return SkipResult::OverStep;
+            }
+        }
+        if self.advance() {
+            SkipResult::OverStep
+        } else {
+            SkipResult::End
         }
-        self.advance()
     }
 
     fn doc(&self) -> DocId {

src/query/phrase_query/phrase_weight.rs

@@ -9,8 +9,8 @@ use crate::query::Weight;
 use crate::query::{EmptyScorer, Explanation};
 use crate::schema::IndexRecordOption;
 use crate::schema::Term;
-use crate::Result;
 use crate::{DocId, DocSet};
+use crate::{Result, SkipResult};
 
 pub struct PhraseWeight {
     phrase_terms: Vec<(usize, Term)>,
@@ -99,7 +99,7 @@ impl Weight for PhraseWeight {
             return Err(does_not_match(doc));
         }
         let mut scorer = scorer_opt.unwrap();
-        if scorer.seek(doc) != doc {
+        if scorer.skip_next(doc) != SkipResult::Reached {
             return Err(does_not_match(doc));
         }
         let fieldnorm_reader = self.fieldnorm_reader(reader);
@@ -114,7 +114,6 @@ impl Weight for PhraseWeight {
 #[cfg(test)]
 mod tests {
     use super::super::tests::create_index;
-    use crate::docset::TERMINATED;
     use crate::query::PhraseQuery;
     use crate::{DocSet, Term};
 
@@ -133,11 +132,12 @@ mod tests {
             .phrase_scorer(searcher.segment_reader(0u32), 1.0f32)
             .unwrap()
             .unwrap();
+        assert!(phrase_scorer.advance());
         assert_eq!(phrase_scorer.doc(), 1);
         assert_eq!(phrase_scorer.phrase_count(), 2);
-        assert_eq!(phrase_scorer.advance(), 2);
+        assert!(phrase_scorer.advance());
         assert_eq!(phrase_scorer.doc(), 2);
         assert_eq!(phrase_scorer.phrase_count(), 1);
-        assert_eq!(phrase_scorer.advance(), TERMINATED);
+        assert!(!phrase_scorer.advance());
     }
 }

src/query/query_parser/query_parser.rs

@@ -113,9 +113,8 @@ fn trim_ast(logical_ast: LogicalAST) -> Option<LogicalAST> {
 /// The language covered by the current parser is extremely simple.
 ///
 /// * simple terms: "e.g.: `Barack Obama` are simply tokenized using
-///   tantivy's [`SimpleTokenizer`](tantivy::tokenizer::SimpleTokenizer), hence
-///   becoming `["barack", "obama"]`. The terms are then searched within
-///   the default terms of the query parser.
+///   tantivy's `StandardTokenizer`, hence becoming `["barack", "obama"]`.
+///   The terms are then searched within the default terms of the query parser.
 ///
 ///   e.g. If `body` and `title` are default fields, our example terms are
 ///   `["title:barack", "body:barack", "title:obama", "body:obama"]`.
@@ -175,16 +174,6 @@ 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
@@ -264,13 +253,8 @@ impl QueryParser {
         &self,
         user_input_ast: UserInputAST,
     ) -> Result<LogicalAST, QueryParserError> {
-        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) {
+        let (occur, ast) = self.compute_logical_ast_with_occur(user_input_ast)?;
+        if occur == Occur::MustNot {
             return Err(QueryParserError::AllButQueryForbidden);
         }
         Ok(ast)
@@ -426,23 +410,31 @@ impl QueryParser {
     fn compute_logical_ast_with_occur(
         &self,
         user_input_ast: UserInputAST,
-    ) -> Result<LogicalAST, QueryParserError> {
+    ) -> Result<(Occur, 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 (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));
+                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));
                 }
-                Ok(LogicalAST::Clause(logical_sub_queries))
+                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))
             }
             UserInputAST::Boost(ast, boost) => {
-                let ast = self.compute_logical_ast_with_occur(*ast)?;
-                Ok(ast.boost(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))
             }
-            UserInputAST::Leaf(leaf) => self.compute_logical_ast_from_leaf(*leaf),
         }
     }
 
@@ -790,20 +782,6 @@ 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(
@@ -823,13 +801,11 @@ mod test {
              Term(field=1,bytes=[116, 105, 116, 105])))",
             false,
         );
-    }
-
-    #[test]
-    fn test_single_negative_term() {
-        assert_matches!(
-            parse_query_to_logical_ast("-title:toto", false),
-            Err(QueryParserError::AllButQueryForbidden)
+        assert_eq!(
+            parse_query_to_logical_ast("-title:toto", false)
+                .err()
+                .unwrap(),
+            QueryParserError::AllButQueryForbidden
         );
     }
 
@@ -989,18 +965,6 @@ 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(
@@ -1020,6 +984,12 @@ 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]) \
@@ -1043,26 +1013,4 @@ mod test {
             false
         );
     }
-
-    #[test]
-    fn test_and_default_regardless_of_default_conjunctive() {
-        for &default_conjunction in &[false, true] {
-            test_parse_query_to_logical_ast_helper(
-                "title:a AND title:b",
-                "(+Term(field=0,bytes=[97]) +Term(field=0,bytes=[98]))",
-                default_conjunction,
-            );
-        }
-    }
-
-    #[test]
-    fn test_or_default_conjunctive() {
-        for &default_conjunction in &[false, true] {
-            test_parse_query_to_logical_ast_helper(
-                "title:a OR title:b",
-                "(Term(field=0,bytes=[97]) Term(field=0,bytes=[98]))",
-                default_conjunction,
-            );
-        }
-    }
 }

src/query/range_query.rs

@@ -10,7 +10,7 @@ use crate::schema::Type;
 use crate::schema::{Field, IndexRecordOption, Term};
 use crate::termdict::{TermDictionary, TermStreamer};
 use crate::DocId;
-use crate::Result;
+use crate::{Result, SkipResult};
 use std::collections::Bound;
 use std::ops::Range;
 
@@ -300,13 +300,10 @@ impl Weight for RangeWeight {
             let term_info = term_range.value();
             let mut block_segment_postings = inverted_index
                 .read_block_postings_from_terminfo(term_info, IndexRecordOption::Basic);
-            loop {
+            while block_segment_postings.advance() {
                 for &doc in block_segment_postings.docs() {
                     doc_bitset.insert(doc);
                 }
-                if !block_segment_postings.advance() {
-                    break;
-                }
             }
         }
         let doc_bitset = BitSetDocSet::from(doc_bitset);
@@ -315,7 +312,7 @@ impl Weight for RangeWeight {
 
     fn explain(&self, reader: &SegmentReader, doc: DocId) -> Result<Explanation> {
         let mut scorer = self.scorer(reader, 1.0f32)?;
-        if scorer.seek(doc) != doc {
+        if scorer.skip_next(doc) != SkipResult::Reached {
             return Err(does_not_match(doc));
         }
         Ok(Explanation::new("RangeQuery", 1.0f32))

src/query/reqopt_scorer.rs

@@ -1,8 +1,9 @@
-use crate::docset::DocSet;
+use crate::docset::{DocSet, SkipResult};
 use crate::query::score_combiner::ScoreCombiner;
 use crate::query::Scorer;
 use crate::DocId;
 use crate::Score;
+use std::cmp::Ordering;
 use std::marker::PhantomData;
 
 /// Given a required scorer and an optional scorer
@@ -16,6 +17,7 @@ pub struct RequiredOptionalScorer<TReqScorer, TOptScorer, TScoreCombiner> {
     req_scorer: TReqScorer,
     opt_scorer: TOptScorer,
     score_cache: Option<Score>,
+    opt_finished: bool,
     _phantom: PhantomData<TScoreCombiner>,
 }
 
@@ -27,12 +29,14 @@ where
     /// Creates a new `RequiredOptionalScorer`.
     pub fn new(
         req_scorer: TReqScorer,
-        opt_scorer: TOptScorer,
+        mut opt_scorer: TOptScorer,
     ) -> RequiredOptionalScorer<TReqScorer, TOptScorer, TScoreCombiner> {
+        let opt_finished = !opt_scorer.advance();
         RequiredOptionalScorer {
             req_scorer,
             opt_scorer,
             score_cache: None,
+            opt_finished,
             _phantom: PhantomData,
         }
     }
@@ -44,7 +48,7 @@ where
     TReqScorer: DocSet,
     TOptScorer: DocSet,
 {
-    fn advance(&mut self) -> DocId {
+    fn advance(&mut self) -> bool {
         self.score_cache = None;
         self.req_scorer.advance()
     }
@@ -72,8 +76,22 @@ where
         let doc = self.doc();
         let mut score_combiner = TScoreCombiner::default();
         score_combiner.update(&mut self.req_scorer);
-        if self.opt_scorer.seek(doc) == doc {
-            score_combiner.update(&mut self.opt_scorer);
+        if !self.opt_finished {
+            match self.opt_scorer.doc().cmp(&doc) {
+                Ordering::Greater => {}
+                Ordering::Equal => {
+                    score_combiner.update(&mut self.opt_scorer);
+                }
+                Ordering::Less => match self.opt_scorer.skip_next(doc) {
+                    SkipResult::Reached => {
+                        score_combiner.update(&mut self.opt_scorer);
+                    }
+                    SkipResult::End => {
+                        self.opt_finished = true;
+                    }
+                    SkipResult::OverStep => {}
+                },
+            }
         }
         let score = score_combiner.score();
         self.score_cache = Some(score);
@@ -84,7 +102,7 @@ where
 #[cfg(test)]
 mod tests {
     use super::RequiredOptionalScorer;
-    use crate::docset::{DocSet, TERMINATED};
+    use crate::docset::DocSet;
     use crate::postings::tests::test_skip_against_unoptimized;
     use crate::query::score_combiner::{DoNothingCombiner, SumCombiner};
     use crate::query::ConstScorer;
@@ -101,9 +119,8 @@ mod tests {
                 ConstScorer::from(VecDocSet::from(vec![])),
             );
         let mut docs = vec![];
-        while reqoptscorer.doc() != TERMINATED {
+        while reqoptscorer.advance() {
             docs.push(reqoptscorer.doc());
-            reqoptscorer.advance();
         }
         assert_eq!(docs, req);
     }
@@ -116,45 +133,46 @@ mod tests {
                 ConstScorer::new(VecDocSet::from(vec![1, 2, 7, 11, 12, 15]), 1.0f32),
             );
         {
+            assert!(reqoptscorer.advance());
             assert_eq!(reqoptscorer.doc(), 1);
             assert_eq!(reqoptscorer.score(), 2f32);
         }
         {
-            assert_eq!(reqoptscorer.advance(), 3);
+            assert!(reqoptscorer.advance());
             assert_eq!(reqoptscorer.doc(), 3);
             assert_eq!(reqoptscorer.score(), 1f32);
         }
         {
-            assert_eq!(reqoptscorer.advance(), 7);
+            assert!(reqoptscorer.advance());
             assert_eq!(reqoptscorer.doc(), 7);
             assert_eq!(reqoptscorer.score(), 2f32);
         }
         {
-            assert_eq!(reqoptscorer.advance(), 8);
+            assert!(reqoptscorer.advance());
             assert_eq!(reqoptscorer.doc(), 8);
             assert_eq!(reqoptscorer.score(), 1f32);
         }
         {
-            assert_eq!(reqoptscorer.advance(), 9);
+            assert!(reqoptscorer.advance());
             assert_eq!(reqoptscorer.doc(), 9);
             assert_eq!(reqoptscorer.score(), 1f32);
         }
         {
-            assert_eq!(reqoptscorer.advance(), 10);
+            assert!(reqoptscorer.advance());
             assert_eq!(reqoptscorer.doc(), 10);
             assert_eq!(reqoptscorer.score(), 1f32);
         }
         {
-            assert_eq!(reqoptscorer.advance(), 13);
+            assert!(reqoptscorer.advance());
             assert_eq!(reqoptscorer.doc(), 13);
             assert_eq!(reqoptscorer.score(), 1f32);
         }
         {
-            assert_eq!(reqoptscorer.advance(), 15);
+            assert!(reqoptscorer.advance());
             assert_eq!(reqoptscorer.doc(), 15);
             assert_eq!(reqoptscorer.score(), 2f32);
         }
-        assert_eq!(reqoptscorer.advance(), TERMINATED);
+        assert!(!reqoptscorer.advance());
     }
 
     #[test]

src/query/scorer.rs

@@ -1,4 +1,5 @@
-use crate::docset::DocSet;
+use crate::common::BitSet;
+use crate::docset::{DocSet, SkipResult};
 use crate::DocId;
 use crate::Score;
 use downcast_rs::impl_downcast;
@@ -12,6 +13,14 @@ pub trait Scorer: downcast_rs::Downcast + DocSet + 'static {
     ///
     /// This method will perform a bit of computation and is not cached.
     fn score(&mut self) -> Score;
+
+    /// Iterates through all of the document matched by the DocSet
+    /// `DocSet` and push the scored documents to the collector.
+    fn for_each(&mut self, callback: &mut dyn FnMut(DocId, Score)) {
+        while self.advance() {
+            callback(self.doc(), self.score());
+        }
+    }
 }
 
 impl_downcast!(Scorer);
@@ -20,6 +29,11 @@ impl Scorer for Box<dyn Scorer> {
     fn score(&mut self) -> Score {
         self.deref_mut().score()
     }
+
+    fn for_each(&mut self, callback: &mut dyn FnMut(DocId, Score)) {
+        let scorer = self.deref_mut();
+        scorer.for_each(callback);
+    }
 }
 
 /// Wraps a `DocSet` and simply returns a constant `Scorer`.
@@ -47,12 +61,12 @@ impl<TDocSet: DocSet> From<TDocSet> for ConstScorer<TDocSet> {
 }
 
 impl<TDocSet: DocSet> DocSet for ConstScorer<TDocSet> {
-    fn advance(&mut self) -> DocId {
+    fn advance(&mut self) -> bool {
         self.docset.advance()
     }
 
-    fn seek(&mut self, target: DocId) -> DocId {
-        self.docset.seek(target)
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        self.docset.skip_next(target)
     }
 
     fn fill_buffer(&mut self, buffer: &mut [DocId]) -> usize {
@@ -66,6 +80,10 @@ impl<TDocSet: DocSet> DocSet for ConstScorer<TDocSet> {
     fn size_hint(&self) -> u32 {
         self.docset.size_hint()
     }
+
+    fn append_to_bitset(&mut self, bitset: &mut BitSet) {
+        self.docset.append_to_bitset(bitset);
+    }
 }
 
 impl<TDocSet: DocSet + 'static> Scorer for ConstScorer<TDocSet> {

src/query/term_query/mod.rs

@@ -26,8 +26,10 @@ 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");
-            index_writer.add_document(doc);
+            {
+                let doc = doc!(text_field => "a");
+                index_writer.add_document(doc);
+            }
             assert!(index_writer.commit().is_ok());
         }
         let searcher = index.reader().unwrap().searcher();
@@ -38,6 +40,7 @@ mod tests {
         let term_weight = term_query.weight(&searcher, true).unwrap();
         let segment_reader = searcher.segment_reader(0);
         let mut term_scorer = term_weight.scorer(segment_reader, 1.0f32).unwrap();
+        assert!(term_scorer.advance());
         assert_eq!(term_scorer.doc(), 0);
         assert_eq!(term_scorer.score(), 0.28768212);
     }
@@ -109,6 +112,7 @@ 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/term_query/term_scorer.rs

@@ -1,4 +1,4 @@
-use crate::docset::DocSet;
+use crate::docset::{DocSet, SkipResult};
 use crate::query::{Explanation, Scorer};
 use crate::DocId;
 use crate::Score;
@@ -45,12 +45,12 @@ impl TermScorer {
 }
 
 impl DocSet for TermScorer {
-    fn advance(&mut self) -> DocId {
+    fn advance(&mut self) -> bool {
         self.postings.advance()
     }
 
-    fn seek(&mut self, target: DocId) -> DocId {
-        self.postings.seek(target)
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        self.postings.skip_next(target)
     }
 
     fn doc(&self) -> DocId {

src/query/term_query/term_weight.rs

@@ -4,13 +4,12 @@ use crate::docset::DocSet;
 use crate::postings::SegmentPostings;
 use crate::query::bm25::BM25Weight;
 use crate::query::explanation::does_not_match;
-use crate::query::weight::{for_each_pruning_scorer, for_each_scorer};
 use crate::query::Weight;
 use crate::query::{Explanation, Scorer};
 use crate::schema::IndexRecordOption;
-use crate::Result;
+use crate::DocId;
 use crate::Term;
-use crate::{DocId, Score};
+use crate::{Result, SkipResult};
 
 pub struct TermWeight {
     term: Term,
@@ -26,7 +25,7 @@ impl Weight for TermWeight {
 
     fn explain(&self, reader: &SegmentReader, doc: DocId) -> Result<Explanation> {
         let mut scorer = self.scorer_specialized(reader, 1.0f32)?;
-        if scorer.seek(doc) != doc {
+        if scorer.skip_next(doc) != SkipResult::Reached {
             return Err(does_not_match(doc));
         }
         Ok(scorer.explain())
@@ -44,39 +43,6 @@ impl Weight for TermWeight {
                 .unwrap_or(0))
         }
     }
-
-    /// Iterates through all of the document matched by the DocSet
-    /// `DocSet` and push the scored documents to the collector.
-    fn for_each(
-        &self,
-        reader: &SegmentReader,
-        callback: &mut dyn FnMut(DocId, Score),
-    ) -> crate::Result<()> {
-        let mut scorer = self.scorer_specialized(reader, 1.0f32)?;
-        for_each_scorer(&mut scorer, callback);
-        Ok(())
-    }
-
-    /// Calls `callback` with all of the `(doc, score)` for which score
-    /// is exceeding a given threshold.
-    ///
-    /// This method is useful for the TopDocs collector.
-    /// For all docsets, the blanket implementation has the benefit
-    /// of prefiltering (doc, score) pairs, avoiding the
-    /// virtual dispatch cost.
-    ///
-    /// More importantly, it makes it possible for scorers to implement
-    /// important optimization (e.g. BlockWAND for union).
-    fn for_each_pruning(
-        &self,
-        threshold: f32,
-        reader: &SegmentReader,
-        callback: &mut dyn FnMut(DocId, Score) -> Score,
-    ) -> crate::Result<()> {
-        let mut scorer = self.scorer(reader, 1.0f32)?;
-        for_each_pruning_scorer(&mut scorer, threshold, callback);
-        Ok(())
-    }
 }
 
 impl TermWeight {

src/query/union.rs

@@ -1,9 +1,10 @@
 use crate::common::TinySet;
-use crate::docset::{DocSet, TERMINATED};
+use crate::docset::{DocSet, SkipResult};
 use crate::query::score_combiner::{DoNothingCombiner, ScoreCombiner};
 use crate::query::Scorer;
 use crate::DocId;
 use crate::Score;
+use std::cmp::Ordering;
 
 const HORIZON_NUM_TINYBITSETS: usize = 64;
 const HORIZON: u32 = 64u32 * HORIZON_NUM_TINYBITSETS as u32;
@@ -46,9 +47,17 @@ where
     fn from(docsets: Vec<TScorer>) -> Union<TScorer, TScoreCombiner> {
         let non_empty_docsets: Vec<TScorer> = docsets
             .into_iter()
-            .filter(|docset| docset.doc() != TERMINATED)
+            .flat_map(
+                |mut docset| {
+                    if docset.advance() {
+                        Some(docset)
+                    } else {
+                        None
+                    }
+                },
+            )
             .collect();
-        let mut union = Union {
+        Union {
             docsets: non_empty_docsets,
             bitsets: Box::new([TinySet::empty(); HORIZON_NUM_TINYBITSETS]),
             scores: Box::new([TScoreCombiner::default(); HORIZON as usize]),
@@ -56,13 +65,7 @@ where
             offset: 0,
             doc: 0,
             score: 0f32,
-        };
-        if union.refill() {
-            union.advance();
-        } else {
-            union.doc = TERMINATED;
         }
-        union
     }
 }
 
@@ -83,7 +86,7 @@ fn refill<TScorer: Scorer, TScoreCombiner: ScoreCombiner>(
             let delta = doc - min_doc;
             bitsets[(delta / 64) as usize].insert_mut(delta % 64u32);
             score_combiner[delta as usize].update(scorer);
-            if scorer.advance() == TERMINATED {
+            if !scorer.advance() {
                 // remove the docset, it has been entirely consumed.
                 return true;
             }
@@ -96,7 +99,6 @@ impl<TScorer: Scorer, TScoreCombiner: ScoreCombiner> Union<TScorer, TScoreCombin
         if let Some(min_doc) = self.docsets.iter().map(DocSet::doc).min() {
             self.offset = min_doc;
             self.cursor = 0;
-            self.doc = min_doc;
             refill(
                 &mut self.docsets,
                 &mut *self.bitsets,
@@ -131,23 +133,30 @@ where
     TScorer: Scorer,
     TScoreCombiner: ScoreCombiner,
 {
-    fn advance(&mut self) -> DocId {
+    fn advance(&mut self) -> bool {
         if self.advance_buffered() {
-            return self.doc;
+            return true;
         }
-        if !self.refill() {
-            self.doc = TERMINATED;
-            return TERMINATED;
+        if self.refill() {
+            self.advance();
+            true
+        } else {
+            false
         }
-        if !self.advance_buffered() {
-            return TERMINATED;
-        }
-        self.doc
     }
 
-    fn seek(&mut self, target: DocId) -> DocId {
-        if self.doc >= target {
-            return self.doc;
+    fn skip_next(&mut self, target: DocId) -> SkipResult {
+        if !self.advance() {
+            return SkipResult::End;
+        }
+        match self.doc.cmp(&target) {
+            Ordering::Equal => {
+                return SkipResult::Reached;
+            }
+            Ordering::Greater => {
+                return SkipResult::OverStep;
+            }
+            Ordering::Less => {}
         }
         let gap = target - self.offset;
         if gap < HORIZON {
@@ -165,11 +174,18 @@ where
 
             // Advancing until we reach the end of the bucket
             // or we reach a doc greater or equal to the target.
-            let mut doc = self.doc();
-            while doc < target {
-                doc = self.advance();
+            while self.advance() {
+                match self.doc().cmp(&target) {
+                    Ordering::Equal => {
+                        return SkipResult::Reached;
+                    }
+                    Ordering::Greater => {
+                        return SkipResult::OverStep;
+                    }
+                    Ordering::Less => {}
+                }
             }
-            doc
+            SkipResult::End
         } else {
             // clear the buffered info.
             for obsolete_tinyset in self.bitsets.iter_mut() {
@@ -183,42 +199,45 @@ where
             // advance all docsets to a doc >= to the target.
             #[cfg_attr(feature = "cargo-clippy", allow(clippy::clippy::collapsible_if))]
             unordered_drain_filter(&mut self.docsets, |docset| {
-                docset.seek(target) == TERMINATED
+                if docset.doc() < target {
+                    if docset.skip_next(target) == SkipResult::End {
+                        return true;
+                    }
+                }
+                false
             });
 
             // at this point all of the docsets
             // are positionned on a doc >= to the target.
-            if !self.refill() {
-                self.doc = TERMINATED;
-                return TERMINATED;
+            if self.refill() {
+                self.advance();
+                if self.doc() == target {
+                    SkipResult::Reached
+                } else {
+                    debug_assert!(self.doc() > target);
+                    SkipResult::OverStep
+                }
+            } else {
+                SkipResult::End
             }
-            self.advance()
         }
     }
 
-    // TODO Also implement `count` with deletes efficiently.
+    // TODO implement `count` efficiently.
 
     fn doc(&self) -> DocId {
         self.doc
     }
 
     fn size_hint(&self) -> u32 {
-        self.docsets
-            .iter()
-            .map(|docset| docset.size_hint())
-            .max()
-            .unwrap_or(0u32)
+        0u32
     }
 
     fn count_including_deleted(&mut self) -> u32 {
-        if self.doc == TERMINATED {
-            return 0;
-        }
         let mut count = self.bitsets[self.cursor..HORIZON_NUM_TINYBITSETS]
             .iter()
             .map(|bitset| bitset.len())
-            .sum::<u32>()
-            + 1;
+            .sum::<u32>();
         for bitset in self.bitsets.iter_mut() {
             bitset.clear();
         }
@@ -248,7 +267,7 @@ mod tests {
 
     use super::Union;
     use super::HORIZON;
-    use crate::docset::{DocSet, TERMINATED};
+    use crate::docset::{DocSet, SkipResult};
     use crate::postings::tests::test_skip_against_unoptimized;
     use crate::query::score_combiner::DoNothingCombiner;
     use crate::query::ConstScorer;
@@ -277,12 +296,12 @@ mod tests {
         };
         let mut union: Union<_, DoNothingCombiner> = make_union();
         let mut count = 0;
-        while union.doc() != TERMINATED {
+        while union.advance() {
+            assert!(union_expected.advance());
             assert_eq!(union_expected.doc(), union.doc());
-            assert_eq!(union_expected.advance(), union.advance());
             count += 1;
         }
-        assert_eq!(union_expected.advance(), TERMINATED);
+        assert!(!union_expected.advance());
         assert_eq!(count, make_union().count_including_deleted());
     }
 
@@ -310,7 +329,9 @@ mod tests {
     fn test_aux_union_skip(docs_list: &[Vec<DocId>], skip_targets: Vec<DocId>) {
         let mut btree_set = BTreeSet::new();
         for docs in docs_list {
-            btree_set.extend(docs.iter().cloned());
+            for &doc in docs.iter() {
+                btree_set.insert(doc);
+            }
         }
         let docset_factory = || {
             let res: Box<dyn DocSet> = Box::new(Union::<_, DoNothingCombiner>::from(
@@ -325,10 +346,10 @@ mod tests {
         };
         let mut docset = docset_factory();
         for el in btree_set {
+            assert!(docset.advance());
             assert_eq!(el, docset.doc());
-            docset.advance();
         }
-        assert_eq!(docset.doc(), TERMINATED);
+        assert!(!docset.advance());
         test_skip_against_unoptimized(docset_factory, skip_targets);
     }
 
@@ -351,10 +372,10 @@ mod tests {
             ConstScorer::from(VecDocSet::from(vec![0u32, 5u32])),
             ConstScorer::from(VecDocSet::from(vec![1u32, 4u32])),
         ]);
+        assert!(docset.advance());
         assert_eq!(docset.doc(), 0u32);
-        assert_eq!(docset.seek(0u32), 0u32);
-        assert_eq!(docset.seek(0u32), 0u32);
-        assert_eq!(docset.doc(), 0u32)
+        assert_eq!(docset.skip_next(0u32), SkipResult::OverStep);
+        assert_eq!(docset.doc(), 1u32)
     }
 
     #[test]

src/query/vec_docset.rs

@@ -1,8 +1,9 @@
 #![allow(dead_code)]
 
 use crate::common::HasLen;
-use crate::docset::{DocSet, TERMINATED};
+use crate::docset::DocSet;
 use crate::DocId;
+use std::num::Wrapping;
 
 /// Simulate a `Postings` objects from a `VecPostings`.
 /// `VecPostings` only exist for testing purposes.
@@ -11,30 +12,26 @@ use crate::DocId;
 /// No positions are returned.
 pub struct VecDocSet {
     doc_ids: Vec<DocId>,
-    cursor: usize,
+    cursor: Wrapping<usize>,
 }
 
 impl From<Vec<DocId>> for VecDocSet {
     fn from(doc_ids: Vec<DocId>) -> VecDocSet {
-        VecDocSet { doc_ids, cursor: 0 }
+        VecDocSet {
+            doc_ids,
+            cursor: Wrapping(usize::max_value()),
+        }
     }
 }
 
 impl DocSet for VecDocSet {
-    fn advance(&mut self) -> DocId {
-        self.cursor += 1;
-        if self.cursor >= self.doc_ids.len() {
-            self.cursor = self.doc_ids.len();
-            return TERMINATED;
-        }
-        self.doc()
+    fn advance(&mut self) -> bool {
+        self.cursor += Wrapping(1);
+        self.doc_ids.len() > self.cursor.0
     }
 
     fn doc(&self) -> DocId {
-        if self.cursor == self.doc_ids.len() {
-            return TERMINATED;
-        }
-        self.doc_ids[self.cursor]
+        self.doc_ids[self.cursor.0]
     }
 
     fn size_hint(&self) -> u32 {
@@ -52,21 +49,22 @@ impl HasLen for VecDocSet {
 pub mod tests {
 
     use super::*;
-    use crate::docset::DocSet;
+    use crate::docset::{DocSet, SkipResult};
     use crate::DocId;
 
     #[test]
     pub fn test_vec_postings() {
         let doc_ids: Vec<DocId> = (0u32..1024u32).map(|e| e * 3).collect();
         let mut postings = VecDocSet::from(doc_ids);
+        assert!(postings.advance());
         assert_eq!(postings.doc(), 0u32);
-        assert_eq!(postings.advance(), 3u32);
+        assert!(postings.advance());
         assert_eq!(postings.doc(), 3u32);
-        assert_eq!(postings.seek(14u32), 15u32);
+        assert_eq!(postings.skip_next(14u32), SkipResult::OverStep);
         assert_eq!(postings.doc(), 15u32);
-        assert_eq!(postings.seek(300u32), 300u32);
+        assert_eq!(postings.skip_next(300u32), SkipResult::Reached);
         assert_eq!(postings.doc(), 300u32);
-        assert_eq!(postings.seek(6000u32), TERMINATED);
+        assert_eq!(postings.skip_next(6000u32), SkipResult::End);
     }
 
     #[test]

src/query/weight.rs

@@ -1,45 +1,7 @@
 use super::Scorer;
 use crate::core::SegmentReader;
 use crate::query::Explanation;
-use crate::{DocId, Score, TERMINATED};
-
-/// Iterates through all of the document matched by the DocSet
-/// `DocSet` and push the scored documents to the collector.
-pub(crate) fn for_each_scorer<TScorer: Scorer + ?Sized>(
-    scorer: &mut TScorer,
-    callback: &mut dyn FnMut(DocId, Score),
-) {
-    let mut doc = scorer.doc();
-    while doc != TERMINATED {
-        callback(doc, scorer.score());
-        doc = scorer.advance();
-    }
-}
-
-/// Calls `callback` with all of the `(doc, score)` for which score
-/// is exceeding a given threshold.
-///
-/// This method is useful for the TopDocs collector.
-/// For all docsets, the blanket implementation has the benefit
-/// of prefiltering (doc, score) pairs, avoiding the
-/// virtual dispatch cost.
-///
-/// More importantly, it makes it possible for scorers to implement
-/// important optimization (e.g. BlockWAND for union).
-pub(crate) fn for_each_pruning_scorer<TScorer: Scorer + ?Sized>(
-    scorer: &mut TScorer,
-    mut threshold: f32,
-    callback: &mut dyn FnMut(DocId, Score) -> Score,
-) {
-    let mut doc = scorer.doc();
-    while doc != TERMINATED {
-        let score = scorer.score();
-        if score > threshold {
-            threshold = callback(doc, score);
-        }
-        doc = scorer.advance();
-    }
-}
+use crate::DocId;
 
 /// A Weight is the specialization of a Query
 /// for a given set of segments.
@@ -65,37 +27,4 @@ pub trait Weight: Send + Sync + 'static {
             Ok(scorer.count_including_deleted())
         }
     }
-
-    /// Iterates through all of the document matched by the DocSet
-    /// `DocSet` and push the scored documents to the collector.
-    fn for_each(
-        &self,
-        reader: &SegmentReader,
-        callback: &mut dyn FnMut(DocId, Score),
-    ) -> crate::Result<()> {
-        let mut scorer = self.scorer(reader, 1.0f32)?;
-        for_each_scorer(scorer.as_mut(), callback);
-        Ok(())
-    }
-
-    /// Calls `callback` with all of the `(doc, score)` for which score
-    /// is exceeding a given threshold.
-    ///
-    /// This method is useful for the TopDocs collector.
-    /// For all docsets, the blanket implementation has the benefit
-    /// of prefiltering (doc, score) pairs, avoiding the
-    /// virtual dispatch cost.
-    ///
-    /// More importantly, it makes it possible for scorers to implement
-    /// important optimization (e.g. BlockWAND for union).
-    fn for_each_pruning(
-        &self,
-        threshold: f32,
-        reader: &SegmentReader,
-        callback: &mut dyn FnMut(DocId, Score) -> Score,
-    ) -> crate::Result<()> {
-        let mut scorer = self.scorer(reader, 1.0f32)?;
-        for_each_pruning_scorer(scorer.as_mut(), threshold, callback);
-        Ok(())
-    }
 }

src/reader/index_writer_reader.rs

@@ -0,0 +1,83 @@
+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

@@ -0,0 +1,180 @@
+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

@@ -1,16 +1,32 @@
+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;
-use self::pool::Pool;
-use crate::core::Segment;
-use crate::directory::Directory;
-use crate::directory::WatchHandle;
-use crate::directory::META_LOCK;
-use crate::Index;
+
+
+
+
+
+pub(crate) use crate::reader::index_writer_reader::NRTReader;
+
 use crate::Searcher;
-use crate::SegmentReader;
-use std::convert::TryInto;
-use std::sync::Arc;
+
+/*
+//
+//enum SegmentSource {
+//    FromMetaFile,
+//    FromWriter(Arc<RwLock<SegmentRegisters>>),
+//}
+//
+//impl SegmentSource {
+//    fn from_meta_file() -> SegmentSource {
+//
+//    }
+//
+//}
 
 /// Defines when a new version of the index should be reloaded.
 ///
@@ -61,6 +77,7 @@ 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,
@@ -113,14 +130,6 @@ 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>,
@@ -128,15 +137,24 @@ 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> = {
-            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 segment_readers: Vec<SegmentReader> = self.load_segment_readers()?;
         let schema = self.index.schema();
         let searchers = (0..self.num_searchers)
             .map(|_| Searcher::new(schema.clone(), self.index.clone(), segment_readers.clone()))
@@ -162,18 +180,16 @@ impl InnerIndexReader {
 ///
 /// `Clone` does not clone the different pool of searcher. `IndexReader`
 /// just wraps and `Arc`.
+=======
+>>>>>>> Added NRTReader
+*/
 #[derive(Clone)]
-pub struct IndexReader {
-    inner: Arc<InnerIndexReader>,
-    watch_handle_opt: Option<WatchHandle>,
+pub enum IndexReader {
+    FromMetaFile(MetaFileIndexReader),
+    NRT(NRTReader),
 }
 
 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()`.
     ///
@@ -184,7 +200,10 @@ 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<()> {
-        self.inner.reload()
+        match self {
+            IndexReader::FromMetaFile(meta_file_reader) => meta_file_reader.reload(),
+            IndexReader::NRT(nrt_reader) => nrt_reader.reload(),
+        }
     }
 
     /// Returns a searcher
@@ -198,6 +217,21 @@ 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> {
-        self.inner.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)
     }
 }

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::mem;
 
 /// Documents are really just a list of couple `(field, value)`.
 /// In this list, one field may appear more than once.
-#[derive(Clone, Debug, serde::Serialize, serde::Deserialize, Default)]
+#[derive(Clone, Debug, Serialize, Deserialize, Default)]
 pub struct Document {
     field_values: Vec<FieldValue>,
 }
@@ -131,34 +131,12 @@ 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());
-
-        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
+        field_values
+            .into_iter()
+            .group_by(|field_value| field_value.field())
+            .into_iter()
+            .map(|(key, group)| (key, group.collect()))
+            .collect::<Vec<(Field, Vec<&FieldValue>)>>()
     }
 
     /// Returns all of the `FieldValue`s associated the given field

src/schema/field.rs

@@ -5,20 +5,18 @@ 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, serde::Serialize, serde::Deserialize,
-)]
+#[derive(Copy, Clone, Debug, PartialEq, PartialOrd, Eq, Ord, Hash, Serialize, Deserialize)]
 pub struct Field(u32);
 
 impl Field {
     /// Create a new field object for the given FieldId.
-    pub const fn from_field_id(field_id: u32) -> Field {
+    pub fn from_field_id(field_id: u32) -> Field {
         Field(field_id)
     }
 
     /// Returns a u32 identifying uniquely a field within a schema.
     #[allow(clippy::trivially_copy_pass_by_ref)]
-    pub const fn field_id(&self) -> u32 {
+    pub fn field_id(&self) -> u32 {
         self.0
     }
 }

src/schema/field_value.rs

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

src/schema/index_record_option.rs

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

src/schema/int_options.rs

@@ -1,5 +1,4 @@
 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

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

src/schema/text_options.rs

@@ -1,7 +1,6 @@
 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;
 
@@ -156,17 +155,30 @@ 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);
-        assert!(matches!(field_entry.field_type(),
-                &FieldType::Str(ref text_options)
-                if text_options.get_indexing_options().unwrap().tokenizer() == "default"));
+        {
+            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!("");
+                }
+            }
+        }
     }
 
     #[test]

src/space_usage/mod.rs

@@ -11,7 +11,6 @@ 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

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

src/termdict/mod.rs

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

src/termdict/termdict.rs

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

@@ -1,7 +1,6 @@
 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

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

src/tokenizer/tokenizer.rs

@@ -1,5 +1,4 @@
 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};