Rebased

Added soft commits
Some small refactoring
2026-01-04 00:02:55 +00:00 · 2020-03-03 10:03:01 +09:00 · 2020-03-03 10:01:05 +09:00 · 2020-03-03 10:01:05 +09:00 · 2020-03-03 10:01:05 +09:00 · 2020-03-03 10:01:05 +09:00
57 changed files with 1026 additions and 941 deletions
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,6 +1,6 @@
-Tantivy 0.13.0
-======================
- Bugfix in `FuzzyTermQuery` not matching terms by prefix when it should (@Peachball)
+Tantivy 0.12.1
+=====================
+- By default IndexReader are in `Manual` mode.

 Tantivy 0.12.0
 ======================
--- a/Cargo.toml
+++ b/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 = {path="../tantivy-fst", version="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"
@@ -38,7 +40,7 @@ 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"
--- a/README.md
+++ b/README.md
@@ -40,7 +40,7 @@ performance for different type of queries / collection.
 # 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)
--- a/appveyor.yml
+++ b/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
--- a/examples/pre_tokenized_text.rs
+++ b/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;

--- a/query-grammar/Cargo.toml
+++ b/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"]
--- a/query-grammar/src/query_grammar.rs
+++ b/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\"}");
--- a/query-grammar/src/user_input_ast.rs
+++ b/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),
        }
--- a/src/collector/top_score_collector.rs
+++ b/src/collector/top_score_collector.rs
@@ -450,6 +450,7 @@ mod tests {
    use crate::Index;
    use crate::IndexWriter;
    use crate::Score;
+    use itertools::Itertools;

    fn make_index() -> Index {
        let mut schema_builder = Schema::builder();
@@ -523,8 +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);
--- a/src/common/mod.rs
+++ b/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)));
-    }
 }
--- a/src/core/index.rs
+++ b/src/core/index.rs
@@ -282,7 +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."
@@ -334,8 +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)
    }

@@ -347,6 +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
@@ -432,7 +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));
@@ -469,7 +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")]
@@ -493,7 +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]
@@ -535,12 +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(());
--- a/src/core/index_meta.rs
+++ b/src/core/index_meta.rs
@@ -4,7 +4,6 @@ use crate::schema::Schema;
 use crate::Opstamp;
 use census::{Inventory, TrackedObject};
 use serde;
-use serde::{Deserialize, Serialize};
 use serde_json;
 use std::collections::HashSet;
 use std::fmt;
--- a/src/core/segment.rs
+++ b/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)
    }
 }
--- a/src/core/segment_id.rs
+++ b/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)]
--- a/src/core/segment_reader.rs
+++ b/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,
--- a/src/directory/mmap_directory.rs
+++ b/src/directory/mmap_directory.rs
@@ -22,7 +22,6 @@ use crate::directory::WatchHandle;
 use crate::directory::{TerminatingWrite, WritePtr};
 use atomicwrites;
 use memmap::Mmap;
-use serde::{Deserialize, Serialize};
 use std::collections::HashMap;
 use std::convert::From;
 use std::fmt;
--- a/src/directory/ram_directory.rs
+++ b/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);
-    }
-}
--- a/src/fastfield/multivalued/writer.rs
+++ b/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 {
--- a/src/indexer/index_writer.rs
+++ b/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::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(
@@ -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)
@@ -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);
    }
 }
--- a/src/indexer/merger.rs
+++ b/src/indexer/merger.rs
@@ -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,
--- a/src/indexer/mod.rs
+++ b/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;

--- a/src/indexer/prepared_commit.rs
+++ b/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)
    }
 }
--- a/src/indexer/segment_entry.rs
+++ b/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
+        )
    }
 }
--- a/src/indexer/segment_manager.rs
+++ b/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)
    }
--- a/src/indexer/segment_register.rs
+++ b/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]);
--- a/src/indexer/segment_updater.rs
+++ b/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;
@@ -117,8 +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();
@@ -134,11 +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 {
@@ -162,12 +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)
@@ -234,6 +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);
@@ -262,8 +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)
    }
@@ -331,12 +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(())
@@ -474,17 +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(),
--- a/src/indexer/segment_writer.rs
+++ b/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,
        })
--- a/src/lib.rs
+++ b/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.
@@ -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;
@@ -269,144 +272,6 @@ impl DocAddress {
    }
 }

-
-
-mod sse2 {
-    use crate::postings::compression::{AlignedBuffer, COMPRESSION_BLOCK_SIZE};
-    use std::arch::x86_64::__m128i as DataType;
-    use std::arch::x86_64::_mm_add_epi32 as op_add;
-    use std::arch::x86_64::_mm_cmplt_epi32 as op_lt;
-    use std::arch::x86_64::_mm_load_si128 as op_load;
-    // requires 128-bits alignment
-    use std::arch::x86_64::_mm_set1_epi32 as set1;
-    use std::arch::x86_64::_mm_setzero_si128 as set0;
-    use std::arch::x86_64::_mm_sub_epi32 as op_sub;
-    use std::arch::x86_64::{_mm_cvtsi128_si32, _mm_shuffle_epi32};
-
-    const MASK1: i32 = 78;
-    const MASK2: i32 = 177;
-
-    /// Performs an exhaustive linear search over the
-    ///
-    /// There is no early exit here. We simply count the
-    /// number of elements that are `< target`.
-    pub unsafe fn linear_search_sse2_128(arr: &AlignedBuffer, target: u32) -> usize {
-        let ptr = arr as *const AlignedBuffer as *const DataType;
-        let vkey = set1(target as i32);
-        let mut cnt = set0();
-        // We work over 4 `__m128i` at a time.
-        // A single `__m128i` actual contains 4 `u32`.
-        for i in 0..(COMPRESSION_BLOCK_SIZE as isize) / (4 * 4) {
-            let cmp1 = op_lt(op_load(ptr.offset(i * 4)), vkey);
-            let cmp2 = op_lt(op_load(ptr.offset(i * 4 + 1)), vkey);
-            let cmp3 = op_lt(op_load(ptr.offset(i * 4 + 2)), vkey);
-            let cmp4 = op_lt(op_load(ptr.offset(i * 4 + 3)), vkey);
-            let sum = op_add(op_add(cmp1, cmp2), op_add(cmp3, cmp4));
-            cnt = op_sub(cnt, sum);
-        }
-        cnt = op_add(cnt, _mm_shuffle_epi32(cnt, MASK1));
-        cnt = op_add(cnt, _mm_shuffle_epi32(cnt, MASK2));
-        _mm_cvtsi128_si32(cnt) as usize
-    }
-}
-
-
-const MAX_DOC: u32 = 2_000_000_000u32;
-
-use sse2::linear_search_sse2_128;
-use crate::postings::BlockSegmentPostings;
-use crate::schema::IndexRecordOption;
-
-struct DocCursor {
-    block_segment_postings: BlockSegmentPostings,
-    cursor: usize,
-}
-
-impl DocCursor {
-    fn new(mut block_segment_postings: BlockSegmentPostings) -> DocCursor {
-        block_segment_postings.advance();
-        DocCursor {
-            block_segment_postings,
-            cursor: 0,
-        }
-    }
-    fn doc(&self,) -> DocId {
-        self.block_segment_postings.doc(self.cursor)
-    }
-
-    fn len(&self,) -> usize {
-        self.block_segment_postings.doc_freq()
-    }
-
-    fn advance(&mut self) {
-        if self.cursor == 127 {
-            self.block_segment_postings.advance();
-            self.cursor = 0;
-        } else {
-            self.cursor += 1;
-        }
-    }
-
-    fn skip_to(&mut self, target: DocId) {
-        if self.block_segment_postings.skip_reader.doc() >= target {
-            unsafe {
-                let mut ptr = self.block_segment_postings.docs_aligned_b().0.get_unchecked(self.cursor) as *const u32;
-                while *ptr < target {
-                    self.cursor += 1;
-                    ptr = ptr.offset(1);
-                }
-            }
-            return;
-        }
-        self.block_segment_postings.skip_to_b(target);
-        let block= self.block_segment_postings.docs_aligned_b();
-        self.cursor = unsafe { linear_search_sse2_128(&block, target) };
-
-    }
-}
-
-pub fn intersection(idx: &InvertedIndexReader, terms: &[Term]) -> crate::Result<u32> {
-    let mut posting_lists: Vec<DocCursor> = Vec::with_capacity(terms.len());
-    for term in terms {
-        if let Some(mut block_postings) = idx.read_block_postings(term, IndexRecordOption::Basic) {
-            posting_lists.push(DocCursor::new(block_postings ));
-        } else {
-            return Ok(0);
-        }
-    }
-    posting_lists.sort_by_key(|posting_list| posting_list.len());
-    Ok({ intersection_idx(&mut posting_lists[..]) })
-}
-
-fn intersection_idx(posting_lists: &mut [DocCursor]) -> DocId {
-    let mut candidate = posting_lists[0].doc();
-    let mut i = 1;
-
-    let mut count = 0u32;
-
-    let num_posting_lists = posting_lists.len();
-    loop {
-        while i < num_posting_lists {
-            posting_lists[i].skip_to(candidate);
-            if posting_lists[i].doc() != candidate {
-                candidate = posting_lists[i].doc();
-                i = 0;
-                continue;
-            }
-            i += 1;
-        }
-        count += 1;
-        if candidate == MAX_DOC {
-            break;
-        }
-        posting_lists[0].advance();
-        candidate = posting_lists[0].doc();
-        i = 1;
-    }
-    count - 1u32
-}
-
-
 /// `DocAddress` contains all the necessary information
 /// to identify a document given a `Searcher` object.
 ///
--- a/src/postings/block_search.rs
+++ b/src/postings/block_search.rs
@@ -25,7 +25,7 @@ mod sse2 {
    ///
    /// There is no early exit here. We simply count the
    /// number of elements that are `< target`.
-    pub fn linear_search_sse2_128(arr: &AlignedBuffer, target: u32) -> usize {
+    pub(crate) fn linear_search_sse2_128(arr: &AlignedBuffer, target: u32) -> usize {
        unsafe {
            let ptr = arr as *const AlignedBuffer as *const DataType;
            let vkey = set1(target as i32);
@@ -63,8 +63,6 @@ mod sse2 {
    }
 }

-pub use self::sse2::linear_search_sse2_128;
-
 /// This `linear search` browser exhaustively through the array.
 /// but the early exit is very difficult to predict.
 ///
--- a/src/postings/compression/mod.rs
+++ b/src/postings/compression/mod.rs
@@ -46,7 +46,7 @@ impl BlockEncoder {
 /// We ensure that the OutputBuffer is align on 128 bits
 /// in order to run SSE2 linear search on it.
 #[repr(align(128))]
-pub struct AlignedBuffer(pub [u32; COMPRESSION_BLOCK_SIZE]);
+pub(crate) struct AlignedBuffer(pub [u32; COMPRESSION_BLOCK_SIZE]);

 pub struct BlockDecoder {
    bitpacker: BitPacker4x,
@@ -67,18 +67,6 @@ impl BlockDecoder {
        }
    }

-    pub fn uncompress_vint_sorted_b<'a>(
-        &mut self,
-        compressed_data: &'a [u8],
-        offset: u32,
-        num_els: usize,
-    ) {
-        if num_els > 0 {
-            vint::uncompress_sorted(compressed_data, &mut self.output.0[..num_els], offset);
-        }
-        self.output.0[num_els..].iter_mut().for_each(|val| *val = 2_000_000_000u32);
-    }
-
    pub fn uncompress_block_sorted(
        &mut self,
        compressed_data: &[u8],
@@ -106,11 +94,6 @@ impl BlockDecoder {
        (&self.output, self.output_len)
    }

-    #[inline]
-    pub(crate) fn output_aligned_2(&self) -> &AlignedBuffer {
-        &self.output
-    }
-
    #[inline]
    pub fn output(&self, idx: usize) -> u32 {
        self.output.0[idx]
--- a/src/postings/mod.rs
+++ b/src/postings/mod.rs
@@ -3,7 +3,7 @@ Postings module (also called inverted index)
 */

 mod block_search;
-pub mod compression;
+pub(crate) mod compression;
 /// Postings module
 ///
 /// Postings, also called inverted lists, is the key datastructure
@@ -220,7 +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
--- a/src/postings/segment_postings.rs
+++ b/src/postings/segment_postings.rs
@@ -317,7 +317,7 @@ pub struct BlockSegmentPostings {
    num_vint_docs: usize,

    remaining_data: OwnedRead,
-    pub skip_reader: SkipReader,
+    skip_reader: SkipReader,
 }

 fn split_into_skips_and_postings(
@@ -414,12 +414,7 @@ impl BlockSegmentPostings {
        self.doc_decoder.output_array()
    }

-    #[inline(always)]
-    pub fn docs_aligned_b(&self) -> &AlignedBuffer {
-        self.doc_decoder.output_aligned_2()
-    }
-
-    pub fn docs_aligned(&self) -> (&AlignedBuffer, usize) {
+    pub(crate) fn docs_aligned(&self) -> (&AlignedBuffer, usize) {
        self.doc_decoder.output_aligned()
    }

@@ -491,7 +486,6 @@ impl BlockSegmentPostings {
                    }
                }
                self.doc_offset = self.skip_reader.doc();
-                unsafe { core::arch::x86_64::_mm_prefetch(self.remaining_data.as_ref() as *mut i8, _MM_HINT_T1) };
                return BlockSegmentPostingsSkipResult::Success(skip_freqs);
            } else {
                skip_freqs += self.skip_reader.tf_sum();
@@ -532,41 +526,6 @@ impl BlockSegmentPostings {
        BlockSegmentPostingsSkipResult::Terminated
    }

-    /// Ensure we are located on the right block.
-    #[inline(never)]
-    pub fn skip_to_b(&mut self, target_doc: DocId) {
-       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,
-                );
-                let tf_num_bits = self.skip_reader.tf_num_bits();
-                let num_bytes_to_skip = compressed_block_size(tf_num_bits);
-                self.remaining_data.advance(num_consumed_bytes + num_bytes_to_skip);
-                self.doc_offset = self.skip_reader.doc();
-                return;
-            } else {
-                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.
-        self.doc_decoder.uncompress_vint_sorted_b(
-           self.remaining_data.as_ref(),
-           self.doc_offset,
-          self.num_vint_docs
-        );
-    }
-
    /// Advance to the next block.
    ///
    /// Returns false iff there was no remaining blocks.
--- a/src/postings/skip.rs
+++ b/src/postings/skip.rs
@@ -49,7 +49,7 @@ impl SkipSerializer {
    }
 }

-pub struct SkipReader {
+pub(crate) struct SkipReader {
    doc: DocId,
    owned_read: OwnedRead,
    doc_num_bits: u8,
--- a/src/query/boolean_query/mod.rs
+++ b/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());
        }
--- a/src/query/explanation.rs
+++ b/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))
--- a/src/query/fuzzy_query.rs
+++ b/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
@@ -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",
@@ -199,17 +166,5 @@ mod test {
            let (score, _) = top_docs[0];
            assert_nearly_equals(1f32, score);
        }
-
-        {
-            let term = Term::from_field_text(country_field, "jap");
-
-            let fuzzy_query = FuzzyTermQuery::new_prefix(term, 1, true);
-            let top_docs = searcher
-                .search(&fuzzy_query, &TopDocs::with_limit(2))
-                .unwrap();
-            assert_eq!(top_docs.len(), 1, "Expected only 1 document");
-            let (score, _) = top_docs[0];
-            assert_nearly_equals(1f32, score);
-        }
    }
 }
--- a/src/query/mod.rs
+++ b/src/query/mod.rs
@@ -24,7 +24,6 @@ mod term_query;
 mod union;
 mod weight;

-
 #[cfg(test)]
 mod vec_docset;

@@ -44,7 +43,6 @@ pub use self::empty_query::{EmptyQuery, EmptyScorer, EmptyWeight};
 pub use self::exclude::Exclude;
 pub use self::explanation::Explanation;
 pub use self::fuzzy_query::FuzzyTermQuery;
-pub(crate) use self::fuzzy_query::DFAWrapper;
 pub use self::intersection::intersect_scorers;
 pub use self::phrase_query::PhraseQuery;
 pub use self::query::Query;
--- a/src/query/phrase_query/mod.rs
+++ b/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;
@@ -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]
--- a/src/query/query.rs
+++ b/src/query/query.rs
@@ -1,17 +1,11 @@
 use super::Weight;
 use crate::core::searcher::Searcher;
 use crate::query::Explanation;
-use crate::{DocAddress, Score};
+use crate::DocAddress;
 use crate::Term;
 use downcast_rs::impl_downcast;
 use std::collections::BTreeSet;
 use std::fmt;
-use crate::collector::{TopDocs, Collector, SegmentCollector, Count};
-
-pub struct TopKResult {
-    pub count: u64,
-    docs: Vec<(Score, DocAddress)>
-}

 /// The `Query` trait defines a set of documents and a scoring method
 /// for those documents.
@@ -62,27 +56,6 @@ pub trait Query: QueryClone + downcast_rs::Downcast + fmt::Debug {
        weight.explain(reader, doc_address.doc())
    }

-    fn top_k(&self, searcher: &Searcher, num_hits: usize) -> crate::Result<TopKResult> {
-        let top_docs = TopDocs::with_limit(num_hits);
-        let collector = (Count, top_docs);
-        let weight = self.weight(searcher, false)?;
-        let mut count = 0u64;
-        let mut result = 0;
-        let mut child_fruits = Vec::with_capacity(searcher.segment_readers().len());
-        for (segment_ord, reader) in searcher.segment_readers().iter().enumerate() {
-            let mut child_top_k = collector.for_segment(segment_ord as u32, reader)?;
-            let mut scorer = weight.scorer(reader, 1.0f32)?;
-            // handle deletes
-            scorer.top_k(&mut child_top_k);
-            child_fruits.push(child_top_k.harvest());
-        }
-        let (count, docs) = collector.merge_fruits(child_fruits)?;
-        Ok(TopKResult {
-            count: count as u64,
-            docs
-        })
-    }
-
    /// Returns the number of documents matching the query.
    fn count(&self, searcher: &Searcher) -> crate::Result<usize> {
        let weight = self.weight(searcher, false)?;
--- a/src/query/query_parser/query_parser.rs
+++ b/src/query/query_parser/query_parser.rs
@@ -174,16 +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
@@ -263,13 +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)
@@ -425,23 +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),
        }
    }

@@ -789,20 +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(
@@ -822,13 +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
        );
    }

@@ -988,18 +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(
@@ -1019,6 +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]) \
@@ -1042,26 +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,
-            );
-        }
-    }
 }
--- a/src/query/scorer.rs
+++ b/src/query/scorer.rs
@@ -4,8 +4,6 @@ use crate::DocId;
 use crate::Score;
 use downcast_rs::impl_downcast;
 use std::ops::DerefMut;
-use crate::collector::{SegmentCollector, Collector, TopDocs, Count};
-

 /// Scored set of documents matching a query within a specific segment.
 ///
@@ -23,12 +21,6 @@ pub trait Scorer: downcast_rs::Downcast + DocSet + 'static {
            callback(self.doc(), self.score());
        }
    }
-
-    fn top_k(&mut self, collector: &mut <(Count, TopDocs) as Collector>::Child) {
-        while self.advance() {
-            collector.collect(self.doc(), self.score());
-        }
-    }
 }

 impl_downcast!(Scorer);
--- a/src/query/term_query/mod.rs
+++ b/src/query/term_query/mod.rs
@@ -112,6 +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);
    }

--- a/src/reader/index_writer_reader.rs
+++ b/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,
+        })
+    }
+}
--- a/src/reader/meta_file_reader.rs
+++ b/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()
+    }
+}
--- a/src/reader/mod.rs
+++ b/src/reader/mod.rs
@@ -1,15 +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::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.
 ///
@@ -120,15 +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()))
@@ -154,18 +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()`.
    ///
@@ -176,7 +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
@@ -190,6 +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)
    }
 }
--- a/src/schema/document.rs
+++ b/src/schema/document.rs
@@ -3,9 +3,8 @@ use crate::common::BinarySerializable;
 use crate::common::VInt;
 use crate::tokenizer::PreTokenizedString;
 use crate::DateTime;
-use serde;
+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.
@@ -17,7 +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>,
 }
@@ -132,34 +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
--- a/src/schema/field.rs
+++ b/src/schema/field.rs
@@ -1,14 +1,11 @@
 use crate::common::BinarySerializable;
-use serde;
 use std::io;
 use std::io::Read;
 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 {
--- a/src/schema/field_value.rs
+++ b/src/schema/field_value.rs
@@ -1,13 +1,12 @@
 use crate::common::BinarySerializable;
 use crate::schema::Field;
 use crate::schema::Value;
-use serde;
 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,
--- a/src/schema/index_record_option.rs
+++ b/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.
 ///
--- a/src/schema/int_options.rs
+++ b/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.
--- a/src/schema/named_field_document.rs
+++ b/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
--- a/src/schema/text_options.rs
+++ b/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]
--- a/src/space_usage/mod.rs
+++ b/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
--- a/src/termdict/mod.rs
+++ b/src/termdict/mod.rs
@@ -435,7 +435,6 @@ mod tests {
    #[test]
    fn test_automaton_search() {
        use levenshtein_automata::LevenshteinAutomatonBuilder;
-        use crate::query::DFAWrapper;

        const COUNTRIES: [&'static str; 7] = [
            "San Marino",
@@ -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();

--- a/src/tokenizer/stemmer.rs
+++ b/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)]
--- a/src/tokenizer/tokenized_string.rs
+++ b/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
--- a/src/tokenizer/tokenizer.rs
+++ b/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};
Author	SHA1	Message	Date
Paul Masurel	7428f6ccd3	Rebased	2020-03-03 10:03:01 +09:00
Paul Masurel	8a488b8315	Added soft commits	2020-03-03 10:01:05 +09:00
Paul Masurel	4448854f73	Some small refactoring	2020-03-03 10:01:05 +09:00
Paul Masurel	b449749d63	Added NRTReader	2020-03-03 10:01:05 +09:00
Paul Masurel	2ba6a12ddc	segment registers	2020-03-03 10:01:05 +09:00
Paul Masurel	bbbf95018b	Removed access to index in IndexReader	2020-03-03 10:01:05 +09:00
Paul Masurel	6d76a82fea	minor changes	2020-03-03 10:01:05 +09:00
Paul Masurel	444048d225	Preparing the ground in the IndexReader.	2020-03-03 10:01:05 +09:00
Paul Masurel	4c6bbca661	Added persisted to Segment and SegmentEntry	2020-03-03 10:01:05 +09:00
Paul Masurel	c3d3b3113b	Segment embeds their own directory, schema instead of relying on Index.	2020-03-03 10:01:05 +09:00