term_freq in TermFrequencyRecorder untested

PR to demonstrate #2285

.github/workflows/coverage.yml

@@ -3,8 +3,6 @@ name: Coverage
 on:
   push:
     branches: [main]
-  pull_request:
-    branches: [main]
 
 # Ensures that we cancel running jobs for the same PR / same workflow.
 concurrency:

.github/workflows/test.yml

@@ -39,6 +39,13 @@ jobs:
 
     - name: Check Formatting
       run: cargo +nightly fmt --all -- --check
+    
+    - name: Check Stable Compilation
+      run: cargo build --all-features
+
+    
+    - name: Check Bench Compilation
+      run: cargo +nightly bench --no-run --profile=dev --all-features
 
     - uses: actions-rs/clippy-check@v1
       with:

CHANGELOG.md

@@ -1,3 +1,9 @@
+Tantivy 0.21.1
+================================
+#### Bugfixes
+- Range queries on fast fields with less values on that field than documents had an invalid end condition, leading to missing results. [#2226](https://github.com/quickwit-oss/tantivy/issues/2226)(@appaquet @PSeitz)
+- Increase the minimum memory budget from 3MB to 15MB to avoid single doc segments (API fix). [#2176](https://github.com/quickwit-oss/tantivy/issues/2176)(@PSeitz)
+
 Tantivy 0.21
 ================================
 #### Bugfixes

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "tantivy"
-version = "0.21.0"
+version = "0.22.0-dev"
 authors = ["Paul Masurel <paul.masurel@gmail.com>"]
 license = "MIT"
 categories = ["database-implementations", "data-structures"]
@@ -22,36 +22,34 @@ crc32fast = "1.3.2"
 once_cell = "1.10.0"
 regex = { version = "1.5.5", default-features = false, features = ["std", "unicode"] }
 aho-corasick = "1.0"
-tantivy-fst = "0.4.0"
-memmap2 = { version = "0.7.1", optional = true }
+tantivy-fst = "0.5"
+memmap2 = { version = "0.9.0", optional = true }
 lz4_flex = { version = "0.11", default-features = false, optional = true }
-zstd = { version = "0.12", optional = true, default-features = false }
+zstd = { version = "0.13", optional = true, default-features = false }
 tempfile = { version = "3.3.0", optional = true }
 log = "0.4.16"
 serde = { version = "1.0.136", features = ["derive"] }
 serde_json = "1.0.79"
 num_cpus = "1.13.1"
-fs4 = { version = "0.6.3", optional = true }
+fs4 = { version = "0.7.0", optional = true }
 levenshtein_automata = "0.2.1"
 uuid = { version = "1.0.0", features = ["v4", "serde"] }
 crossbeam-channel = "0.5.4"
 rust-stemmers = "1.2.0"
 downcast-rs = "1.2.0"
-bitpacking = { version = "0.8.4", default-features = false, features = ["bitpacker4x"] }
+bitpacking = { version = "0.9.2", default-features = false, features = ["bitpacker4x"] }
 census = "0.4.0"
 rustc-hash = "1.1.0"
 thiserror = "1.0.30"
 htmlescape = "0.3.1"
 fail = { version = "0.5.0", optional = true }
-murmurhash32 = "0.3.0"
 time = { version = "0.3.10", features = ["serde-well-known"] }
 smallvec = "1.8.0"
 rayon = "1.5.2"
-lru = "0.11.0"
+lru = "0.12.0"
 fastdivide = "0.4.0"
-itertools = "0.11.0"
+itertools = "0.12.0"
 measure_time = "0.8.2"
-async-trait = "0.1.53"
 arc-swap = "1.5.0"
 
 columnar = { version= "0.2", path="./columnar", package ="tantivy-columnar" }
@@ -63,6 +61,7 @@ common = { version= "0.6", path = "./common/", package = "tantivy-common" }
 tokenizer-api = { version= "0.2", path="./tokenizer-api", package="tantivy-tokenizer-api" }
 sketches-ddsketch = { version = "0.2.1", features = ["use_serde"] }
 futures-util = { version = "0.3.28", optional = true }
+fnv = "1.0.7"
 
 [target.'cfg(windows)'.dependencies]
 winapi = "0.3.9"
@@ -74,15 +73,13 @@ matches = "0.1.9"
 pretty_assertions = "1.2.1"
 proptest = "1.0.0"
 test-log = "0.2.10"
-env_logger = "0.10.0"
 futures = "0.3.21"
 paste = "1.0.11"
 more-asserts = "0.3.1"
 rand_distr = "0.4.3"
 
 [target.'cfg(not(windows))'.dev-dependencies]
-criterion = "0.5"
-pprof = { git = "https://github.com/PSeitz/pprof-rs/", rev = "53af24b", features = ["flamegraph", "criterion"] } # temp fork that works with criterion 0.5
+criterion = { version = "0.5", default-features = false }
 
 [dev-dependencies.fail]
 version = "0.5.0"
@@ -115,6 +112,11 @@ unstable = [] # useful for benches.
 
 quickwit = ["sstable", "futures-util"]
 
+# Compares only the hash of a string when indexing data. 
+# Increases indexing speed, but may lead to extremely rare missing terms, when there's a hash collision.
+# Uses 64bit ahash.
+compare_hash_only = ["stacker/compare_hash_only"]
+
 [workspace]
 members = ["query-grammar", "bitpacker", "common", "ownedbytes", "stacker", "sstable", "tokenizer-api", "columnar"]
 

benches/index-bench.rs

@@ -1,14 +1,99 @@
-use criterion::{criterion_group, criterion_main, Criterion, Throughput};
-use pprof::criterion::{Output, PProfProfiler};
-use tantivy::schema::{FAST, INDEXED, STORED, STRING, TEXT};
-use tantivy::Index;
+use criterion::{criterion_group, criterion_main, BatchSize, Bencher, Criterion, Throughput};
+use tantivy::schema::{TantivyDocument, FAST, INDEXED, STORED, STRING, TEXT};
+use tantivy::{tokenizer, Index, IndexWriter};
 
 const HDFS_LOGS: &str = include_str!("hdfs.json");
 const GH_LOGS: &str = include_str!("gh.json");
 const WIKI: &str = include_str!("wiki.json");
 
-fn get_lines(input: &str) -> Vec<&str> {
-    input.trim().split('\n').collect()
+fn benchmark(
+    b: &mut Bencher,
+    input: &str,
+    schema: tantivy::schema::Schema,
+    commit: bool,
+    parse_json: bool,
+    is_dynamic: bool,
+) {
+    if is_dynamic {
+        benchmark_dynamic_json(b, input, schema, commit, parse_json)
+    } else {
+        _benchmark(b, input, schema, commit, parse_json, |schema, doc_json| {
+            TantivyDocument::parse_json(&schema, doc_json).unwrap()
+        })
+    }
+}
+
+fn get_index(schema: tantivy::schema::Schema) -> Index {
+    let mut index = Index::create_in_ram(schema.clone());
+    let ff_tokenizer_manager = tokenizer::TokenizerManager::default();
+    ff_tokenizer_manager.register(
+        "raw",
+        tokenizer::TextAnalyzer::builder(tokenizer::RawTokenizer::default())
+            .filter(tokenizer::RemoveLongFilter::limit(255))
+            .build(),
+    );
+    index.set_fast_field_tokenizers(ff_tokenizer_manager.clone());
+    index
+}
+
+fn _benchmark(
+    b: &mut Bencher,
+    input: &str,
+    schema: tantivy::schema::Schema,
+    commit: bool,
+    include_json_parsing: bool,
+    create_doc: impl Fn(&tantivy::schema::Schema, &str) -> TantivyDocument,
+) {
+    if include_json_parsing {
+        let lines: Vec<&str> = input.trim().split('\n').collect();
+        b.iter(|| {
+            let index = get_index(schema.clone());
+            let mut index_writer: IndexWriter =
+                index.writer_with_num_threads(1, 100_000_000).unwrap();
+            for doc_json in &lines {
+                let doc = create_doc(&schema, doc_json);
+                index_writer.add_document(doc).unwrap();
+            }
+            if commit {
+                index_writer.commit().unwrap();
+            }
+        })
+    } else {
+        let docs: Vec<_> = input
+            .trim()
+            .split('\n')
+            .map(|doc_json| create_doc(&schema, doc_json))
+            .collect();
+        b.iter_batched(
+            || docs.clone(),
+            |docs| {
+                let index = get_index(schema.clone());
+                let mut index_writer: IndexWriter =
+                    index.writer_with_num_threads(1, 100_000_000).unwrap();
+                for doc in docs {
+                    index_writer.add_document(doc).unwrap();
+                }
+                if commit {
+                    index_writer.commit().unwrap();
+                }
+            },
+            BatchSize::SmallInput,
+        )
+    }
+}
+fn benchmark_dynamic_json(
+    b: &mut Bencher,
+    input: &str,
+    schema: tantivy::schema::Schema,
+    commit: bool,
+    parse_json: bool,
+) {
+    let json_field = schema.get_field("json").unwrap();
+    _benchmark(b, input, schema, commit, parse_json, |_schema, doc_json| {
+        let json_val: serde_json::Map<String, serde_json::Value> =
+            serde_json::from_str(doc_json).unwrap();
+        tantivy::doc!(json_field=>json_val)
+    })
 }
 
 pub fn hdfs_index_benchmark(c: &mut Criterion) {
@@ -19,7 +104,14 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
         schema_builder.add_text_field("severity", STRING);
         schema_builder.build()
     };
-    let schema_with_store = {
+    let schema_only_fast = {
+        let mut schema_builder = tantivy::schema::SchemaBuilder::new();
+        schema_builder.add_u64_field("timestamp", FAST);
+        schema_builder.add_text_field("body", FAST);
+        schema_builder.add_text_field("severity", FAST);
+        schema_builder.build()
+    };
+    let _schema_with_store = {
         let mut schema_builder = tantivy::schema::SchemaBuilder::new();
         schema_builder.add_u64_field("timestamp", INDEXED | STORED);
         schema_builder.add_text_field("body", TEXT | STORED);
@@ -28,74 +120,39 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
     };
     let dynamic_schema = {
         let mut schema_builder = tantivy::schema::SchemaBuilder::new();
-        schema_builder.add_json_field("json", TEXT);
+        schema_builder.add_json_field("json", TEXT | FAST);
         schema_builder.build()
     };
 
     let mut group = c.benchmark_group("index-hdfs");
     group.throughput(Throughput::Bytes(HDFS_LOGS.len() as u64));
     group.sample_size(20);
-    group.bench_function("index-hdfs-no-commit", |b| {
-        let lines = get_lines(HDFS_LOGS);
-        b.iter(|| {
-            let index = Index::create_in_ram(schema.clone());
-            let index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
-            for doc_json in &lines {
-                let doc = schema.parse_document(doc_json).unwrap();
-                index_writer.add_document(doc).unwrap();
+
+    let benches = [
+        ("only-indexed-".to_string(), schema, false),
+        //("stored-".to_string(), _schema_with_store, false),
+        ("only-fast-".to_string(), schema_only_fast, false),
+        ("dynamic-".to_string(), dynamic_schema, true),
+    ];
+
+    for (prefix, schema, is_dynamic) in benches {
+        for commit in [false, true] {
+            let suffix = if commit { "with-commit" } else { "no-commit" };
+            for parse_json in [false] {
+                // for parse_json in [false, true] {
+                let suffix = if parse_json {
+                    format!("{}-with-json-parsing", suffix)
+                } else {
+                    format!("{}", suffix)
+                };
+
+                let bench_name = format!("{}{}", prefix, suffix);
+                group.bench_function(bench_name, |b| {
+                    benchmark(b, HDFS_LOGS, schema.clone(), commit, parse_json, is_dynamic)
+                });
             }
-        })
-    });
-    group.bench_function("index-hdfs-with-commit", |b| {
-        let lines = get_lines(HDFS_LOGS);
-        b.iter(|| {
-            let index = Index::create_in_ram(schema.clone());
-            let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
-            for doc_json in &lines {
-                let doc = schema.parse_document(doc_json).unwrap();
-                index_writer.add_document(doc).unwrap();
-            }
-            index_writer.commit().unwrap();
-        })
-    });
-    group.bench_function("index-hdfs-no-commit-with-docstore", |b| {
-        let lines = get_lines(HDFS_LOGS);
-        b.iter(|| {
-            let index = Index::create_in_ram(schema_with_store.clone());
-            let index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
-            for doc_json in &lines {
-                let doc = schema.parse_document(doc_json).unwrap();
-                index_writer.add_document(doc).unwrap();
-            }
-        })
-    });
-    group.bench_function("index-hdfs-with-commit-with-docstore", |b| {
-        let lines = get_lines(HDFS_LOGS);
-        b.iter(|| {
-            let index = Index::create_in_ram(schema_with_store.clone());
-            let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
-            for doc_json in &lines {
-                let doc = schema.parse_document(doc_json).unwrap();
-                index_writer.add_document(doc).unwrap();
-            }
-            index_writer.commit().unwrap();
-        })
-    });
-    group.bench_function("index-hdfs-no-commit-json-without-docstore", |b| {
-        let lines = get_lines(HDFS_LOGS);
-        b.iter(|| {
-            let index = Index::create_in_ram(dynamic_schema.clone());
-            let json_field = dynamic_schema.get_field("json").unwrap();
-            let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
-            for doc_json in &lines {
-                let json_val: serde_json::Map<String, serde_json::Value> =
-                    serde_json::from_str(doc_json).unwrap();
-                let doc = tantivy::doc!(json_field=>json_val);
-                index_writer.add_document(doc).unwrap();
-            }
-            index_writer.commit().unwrap();
-        })
-    });
+        }
+    }
 }
 
 pub fn gh_index_benchmark(c: &mut Criterion) {
@@ -104,38 +161,24 @@ pub fn gh_index_benchmark(c: &mut Criterion) {
         schema_builder.add_json_field("json", TEXT | FAST);
         schema_builder.build()
     };
+    let dynamic_schema_fast = {
+        let mut schema_builder = tantivy::schema::SchemaBuilder::new();
+        schema_builder.add_json_field("json", FAST);
+        schema_builder.build()
+    };
 
     let mut group = c.benchmark_group("index-gh");
     group.throughput(Throughput::Bytes(GH_LOGS.len() as u64));
 
     group.bench_function("index-gh-no-commit", |b| {
-        let lines = get_lines(GH_LOGS);
-        b.iter(|| {
-            let json_field = dynamic_schema.get_field("json").unwrap();
-            let index = Index::create_in_ram(dynamic_schema.clone());
-            let index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
-            for doc_json in &lines {
-                let json_val: serde_json::Map<String, serde_json::Value> =
-                    serde_json::from_str(doc_json).unwrap();
-                let doc = tantivy::doc!(json_field=>json_val);
-                index_writer.add_document(doc).unwrap();
-            }
-        })
+        benchmark_dynamic_json(b, GH_LOGS, dynamic_schema.clone(), false, false)
     });
-    group.bench_function("index-gh-with-commit", |b| {
-        let lines = get_lines(GH_LOGS);
-        b.iter(|| {
-            let json_field = dynamic_schema.get_field("json").unwrap();
-            let index = Index::create_in_ram(dynamic_schema.clone());
-            let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
-            for doc_json in &lines {
-                let json_val: serde_json::Map<String, serde_json::Value> =
-                    serde_json::from_str(doc_json).unwrap();
-                let doc = tantivy::doc!(json_field=>json_val);
-                index_writer.add_document(doc).unwrap();
-            }
-            index_writer.commit().unwrap();
-        })
+    group.bench_function("index-gh-fast", |b| {
+        benchmark_dynamic_json(b, GH_LOGS, dynamic_schema_fast.clone(), false, false)
+    });
+
+    group.bench_function("index-gh-fast-with-commit", |b| {
+        benchmark_dynamic_json(b, GH_LOGS, dynamic_schema_fast.clone(), true, false)
     });
 }
 
@@ -150,33 +193,10 @@ pub fn wiki_index_benchmark(c: &mut Criterion) {
     group.throughput(Throughput::Bytes(WIKI.len() as u64));
 
     group.bench_function("index-wiki-no-commit", |b| {
-        let lines = get_lines(WIKI);
-        b.iter(|| {
-            let json_field = dynamic_schema.get_field("json").unwrap();
-            let index = Index::create_in_ram(dynamic_schema.clone());
-            let index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
-            for doc_json in &lines {
-                let json_val: serde_json::Map<String, serde_json::Value> =
-                    serde_json::from_str(doc_json).unwrap();
-                let doc = tantivy::doc!(json_field=>json_val);
-                index_writer.add_document(doc).unwrap();
-            }
-        })
+        benchmark_dynamic_json(b, WIKI, dynamic_schema.clone(), false, false)
     });
     group.bench_function("index-wiki-with-commit", |b| {
-        let lines = get_lines(WIKI);
-        b.iter(|| {
-            let json_field = dynamic_schema.get_field("json").unwrap();
-            let index = Index::create_in_ram(dynamic_schema.clone());
-            let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
-            for doc_json in &lines {
-                let json_val: serde_json::Map<String, serde_json::Value> =
-                    serde_json::from_str(doc_json).unwrap();
-                let doc = tantivy::doc!(json_field=>json_val);
-                index_writer.add_document(doc).unwrap();
-            }
-            index_writer.commit().unwrap();
-        })
+        benchmark_dynamic_json(b, WIKI, dynamic_schema.clone(), true, false)
     });
 }
 
@@ -187,12 +207,12 @@ criterion_group! {
 }
 criterion_group! {
     name = gh_benches;
-    config = Criterion::default().with_profiler(PProfProfiler::new(100, Output::Flamegraph(None)));
+    config = Criterion::default();
     targets = gh_index_benchmark
 }
 criterion_group! {
     name = wiki_benches;
-    config = Criterion::default().with_profiler(PProfProfiler::new(100, Output::Flamegraph(None)));
+    config = Criterion::default();
     targets = wiki_index_benchmark
 }
 criterion_main!(benches, gh_benches, wiki_benches);

bitpacker/Cargo.toml

@@ -15,7 +15,7 @@ homepage = "https://github.com/quickwit-oss/tantivy"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-bitpacking = {version="0.8", default-features=false, features = ["bitpacker1x"]}
+bitpacking = { version = "0.9.2", default-features = false, features = ["bitpacker1x"] }
 
 [dev-dependencies]
 rand = "0.8"

bitpacker/src/bitpacker.rs

@@ -367,7 +367,7 @@ mod test {
             let mut output: Vec<u32> = Vec::new();
             for len in [0, 1, 2, 32, 33, 34, 64] {
                 for start_idx in 0u32..32u32 {
-                    output.resize(len as usize, 0);
+                    output.resize(len, 0);
                     bitunpacker.get_batch_u32s(start_idx, &buffer, &mut output);
                     for i in 0..len {
                         let expected = (start_idx + i as u32) & mask;

columnar/Cargo.toml

@@ -9,8 +9,7 @@ description = "column oriented storage for tantivy"
 categories = ["database-implementations", "data-structures", "compression"]
 
 [dependencies]
-itertools = "0.11.0"
-fnv = "1.0.7"
+itertools = "0.12.0"
 fastdivide = "0.4.0"
 
 stacker = { version= "0.2", path = "../stacker", package="tantivy-stacker"}

columnar/columnar-cli/Cargo.toml

@@ -8,7 +8,6 @@ license = "MIT"
 columnar = {path="../", package="tantivy-columnar"}
 serde_json = "1"
 serde_json_borrow = {git="https://github.com/PSeitz/serde_json_borrow/"}
-serde = "1"
 
 [workspace]
 members = []

columnar/src/column_index/mod.rs

@@ -1,3 +1,8 @@
+//! # `column_index`
+//!
+//! `column_index` provides rank and select operations to associate positions when not all
+//! documents have exactly one element.
+
 mod merge;
 mod multivalued_index;
 mod optional_index;
@@ -41,10 +46,10 @@ impl ColumnIndex {
     pub fn is_multivalue(&self) -> bool {
         matches!(self, ColumnIndex::Multivalued(_))
     }
-    // Returns the cardinality of the column index.
-    //
-    // By convention, if the column contains no docs, we consider that it is
-    // full.
+    /// Returns the cardinality of the column index.
+    ///
+    /// By convention, if the column contains no docs, we consider that it is
+    /// full.
     #[inline]
     pub fn get_cardinality(&self) -> Cardinality {
         match self {

columnar/src/column_index/optional_index/tests.rs

@@ -215,12 +215,12 @@ mod bench {
         let vals: Vec<RowId> = (0..TOTAL_NUM_VALUES)
             .map(|_| rng.gen_bool(fill_ratio))
             .enumerate()
-            .filter(|(pos, val)| *val)
+            .filter(|(_pos, val)| *val)
             .map(|(pos, _)| pos as RowId)
             .collect();
         serialize_optional_index(&&vals[..], TOTAL_NUM_VALUES, &mut out).unwrap();
-        let codec = open_optional_index(OwnedBytes::new(out)).unwrap();
-        codec
+
+        open_optional_index(OwnedBytes::new(out)).unwrap()
     }
 
     fn random_range_iterator(
@@ -242,7 +242,7 @@ mod bench {
     }
 
     fn n_percent_step_iterator(percent: f32, num_values: u32) -> impl Iterator<Item = u32> {
-        let ratio = percent as f32 / 100.0;
+        let ratio = percent / 100.0;
         let step_size = (1f32 / ratio) as u32;
         let deviation = step_size - 1;
         random_range_iterator(0, num_values, step_size, deviation)

columnar/src/column_index/serialize.rs

@@ -30,6 +30,7 @@ impl<'a> SerializableColumnIndex<'a> {
     }
 }
 
+/// Serialize a column index.
 pub fn serialize_column_index(
     column_index: SerializableColumnIndex,
     output: &mut impl Write,
@@ -51,6 +52,7 @@ pub fn serialize_column_index(
     Ok(column_index_num_bytes)
 }
 
+/// Open a serialized column index.
 pub fn open_column_index(mut bytes: OwnedBytes) -> io::Result<ColumnIndex> {
     if bytes.is_empty() {
         return Err(io::Error::new(

columnar/src/columnar/writer/column_writers.rs

@@ -269,7 +269,8 @@ impl StrOrBytesColumnWriter {
         dictionaries: &mut [DictionaryBuilder],
         arena: &mut MemoryArena,
     ) {
-        let unordered_id = dictionaries[self.dictionary_id as usize].get_or_allocate_id(bytes);
+        let unordered_id =
+            dictionaries[self.dictionary_id as usize].get_or_allocate_id(bytes, arena);
         self.column_writer.record(doc, unordered_id, arena);
     }
 

columnar/src/columnar/writer/mod.rs

@@ -338,7 +338,7 @@ impl ColumnarWriter {
         let mut columns: Vec<(&[u8], ColumnType, Addr)> = self
             .numerical_field_hash_map
             .iter()
-            .map(|(column_name, addr, _)| {
+            .map(|(column_name, addr)| {
                 let numerical_column_writer: NumericalColumnWriter =
                     self.numerical_field_hash_map.read(addr);
                 let column_type = numerical_column_writer.numerical_type().into();
@@ -348,27 +348,27 @@ impl ColumnarWriter {
         columns.extend(
             self.bytes_field_hash_map
                 .iter()
-                .map(|(term, addr, _)| (term, ColumnType::Bytes, addr)),
+                .map(|(term, addr)| (term, ColumnType::Bytes, addr)),
         );
         columns.extend(
             self.str_field_hash_map
                 .iter()
-                .map(|(column_name, addr, _)| (column_name, ColumnType::Str, addr)),
+                .map(|(column_name, addr)| (column_name, ColumnType::Str, addr)),
         );
         columns.extend(
             self.bool_field_hash_map
                 .iter()
-                .map(|(column_name, addr, _)| (column_name, ColumnType::Bool, addr)),
+                .map(|(column_name, addr)| (column_name, ColumnType::Bool, addr)),
         );
         columns.extend(
             self.ip_addr_field_hash_map
                 .iter()
-                .map(|(column_name, addr, _)| (column_name, ColumnType::IpAddr, addr)),
+                .map(|(column_name, addr)| (column_name, ColumnType::IpAddr, addr)),
         );
         columns.extend(
             self.datetime_field_hash_map
                 .iter()
-                .map(|(column_name, addr, _)| (column_name, ColumnType::DateTime, addr)),
+                .map(|(column_name, addr)| (column_name, ColumnType::DateTime, addr)),
         );
         columns.sort_unstable_by_key(|(column_name, col_type, _)| (*column_name, *col_type));
 
@@ -437,6 +437,7 @@ impl ColumnarWriter {
                             &mut symbol_byte_buffer,
                         ),
                         buffers,
+                        &self.arena,
                         &mut column_serializer,
                     )?;
                     column_serializer.finalize()?;
@@ -490,6 +491,7 @@ impl ColumnarWriter {
 
 // Serialize [Dictionary, Column, dictionary num bytes U32::LE]
 // Column: [Column Index, Column Values, column index num bytes U32::LE]
+#[allow(clippy::too_many_arguments)]
 fn serialize_bytes_or_str_column(
     cardinality: Cardinality,
     num_docs: RowId,
@@ -497,6 +499,7 @@ fn serialize_bytes_or_str_column(
     dictionary_builder: &DictionaryBuilder,
     operation_it: impl Iterator<Item = ColumnOperation<UnorderedId>>,
     buffers: &mut SpareBuffers,
+    arena: &MemoryArena,
     wrt: impl io::Write,
 ) -> io::Result<()> {
     let SpareBuffers {
@@ -505,7 +508,8 @@ fn serialize_bytes_or_str_column(
         ..
     } = buffers;
     let mut counting_writer = CountingWriter::wrap(wrt);
-    let term_id_mapping: TermIdMapping = dictionary_builder.serialize(&mut counting_writer)?;
+    let term_id_mapping: TermIdMapping =
+        dictionary_builder.serialize(arena, &mut counting_writer)?;
     let dictionary_num_bytes: u32 = counting_writer.written_bytes() as u32;
     let mut wrt = counting_writer.finish();
     let operation_iterator = operation_it.map(|symbol: ColumnOperation<UnorderedId>| {

columnar/src/dictionary.rs

@@ -1,7 +1,7 @@
 use std::io;
 
-use fnv::FnvHashMap;
 use sstable::SSTable;
+use stacker::{MemoryArena, SharedArenaHashMap};
 
 pub(crate) struct TermIdMapping {
     unordered_to_ord: Vec<OrderedId>,
@@ -31,29 +31,38 @@ pub struct OrderedId(pub u32);
 /// mapping.
 #[derive(Default)]
 pub(crate) struct DictionaryBuilder {
-    dict: FnvHashMap<Vec<u8>, UnorderedId>,
-    memory_consumption: usize,
+    dict: SharedArenaHashMap,
 }
 
 impl DictionaryBuilder {
     /// Get or allocate an unordered id.
     /// (This ID is simply an auto-incremented id.)
-    pub fn get_or_allocate_id(&mut self, term: &[u8]) -> UnorderedId {
-        if let Some(term_id) = self.dict.get(term) {
-            return *term_id;
-        }
-        let new_id = UnorderedId(self.dict.len() as u32);
-        self.dict.insert(term.to_vec(), new_id);
-        self.memory_consumption += term.len();
-        self.memory_consumption += 40; // Term Metadata + HashMap overhead
-        new_id
+    pub fn get_or_allocate_id(&mut self, term: &[u8], arena: &mut MemoryArena) -> UnorderedId {
+        let next_id = self.dict.len() as u32;
+        let unordered_id = self
+            .dict
+            .mutate_or_create(term, arena, |unordered_id: Option<u32>| {
+                if let Some(unordered_id) = unordered_id {
+                    unordered_id
+                } else {
+                    next_id
+                }
+            });
+        UnorderedId(unordered_id)
     }
 
     /// Serialize the dictionary into an fst, and returns the
     /// `UnorderedId -> TermOrdinal` map.
-    pub fn serialize<'a, W: io::Write + 'a>(&self, wrt: &mut W) -> io::Result<TermIdMapping> {
-        let mut terms: Vec<(&[u8], UnorderedId)> =
-            self.dict.iter().map(|(k, v)| (k.as_slice(), *v)).collect();
+    pub fn serialize<'a, W: io::Write + 'a>(
+        &self,
+        arena: &MemoryArena,
+        wrt: &mut W,
+    ) -> io::Result<TermIdMapping> {
+        let mut terms: Vec<(&[u8], UnorderedId)> = self
+            .dict
+            .iter(arena)
+            .map(|(k, v)| (k, arena.read(v)))
+            .collect();
         terms.sort_unstable_by_key(|(key, _)| *key);
         // TODO Remove the allocation.
         let mut unordered_to_ord: Vec<OrderedId> = vec![OrderedId(0u32); terms.len()];
@@ -68,7 +77,7 @@ impl DictionaryBuilder {
     }
 
     pub(crate) fn mem_usage(&self) -> usize {
-        self.memory_consumption
+        self.dict.mem_usage()
     }
 }
 
@@ -78,12 +87,13 @@ mod tests {
 
     #[test]
     fn test_dictionary_builder() {
+        let mut arena = MemoryArena::default();
         let mut dictionary_builder = DictionaryBuilder::default();
-        let hello_uid = dictionary_builder.get_or_allocate_id(b"hello");
-        let happy_uid = dictionary_builder.get_or_allocate_id(b"happy");
-        let tax_uid = dictionary_builder.get_or_allocate_id(b"tax");
+        let hello_uid = dictionary_builder.get_or_allocate_id(b"hello", &mut arena);
+        let happy_uid = dictionary_builder.get_or_allocate_id(b"happy", &mut arena);
+        let tax_uid = dictionary_builder.get_or_allocate_id(b"tax", &mut arena);
         let mut buffer = Vec::new();
-        let id_mapping = dictionary_builder.serialize(&mut buffer).unwrap();
+        let id_mapping = dictionary_builder.serialize(&arena, &mut buffer).unwrap();
         assert_eq!(id_mapping.to_ord(hello_uid), OrderedId(1));
         assert_eq!(id_mapping.to_ord(happy_uid), OrderedId(0));
         assert_eq!(id_mapping.to_ord(tax_uid), OrderedId(2));

columnar/src/lib.rs

@@ -1,3 +1,22 @@
+//! # Tantivy-Columnar
+//!
+//! `tantivy-columnar`provides a columnar storage for tantivy.
+//! The crate allows for efficient read operations on specific columns rather than entire records.
+//!
+//! ## Overview
+//!
+//! - **columnar**: Reading, writing, and merging multiple columns:
+//!   - **[ColumnarWriter]**: Makes it possible to create a new columnar.
+//!   - **[ColumnarReader]**: The ColumnarReader makes it possible to access a set of columns
+//!     associated to field names.
+//!   - **[merge_columnar]**: Contains the functionalities to merge multiple ColumnarReader or
+//!     segments into a single one.
+//!
+//! - **column**: A single column, which contains
+//!     - [column_index]: Resolves the rows for a document id. Manages the cardinality of the
+//!       column.
+//!     - [column_values]: Stores the values of a column in a dense format.
+
 #![cfg_attr(all(feature = "unstable", test), feature(test))]
 
 #[cfg(test)]
@@ -12,7 +31,7 @@ use std::io;
 
 mod block_accessor;
 mod column;
-mod column_index;
+pub mod column_index;
 pub mod column_values;
 mod columnar;
 mod dictionary;

columnar/src/tests.rs

@@ -26,7 +26,7 @@ fn test_dataframe_writer_str() {
     assert_eq!(columnar.num_columns(), 1);
     let cols: Vec<DynamicColumnHandle> = columnar.read_columns("my_string").unwrap();
     assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 87);
+    assert_eq!(cols[0].num_bytes(), 73);
 }
 
 #[test]
@@ -40,7 +40,7 @@ fn test_dataframe_writer_bytes() {
     assert_eq!(columnar.num_columns(), 1);
     let cols: Vec<DynamicColumnHandle> = columnar.read_columns("my_string").unwrap();
     assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 87);
+    assert_eq!(cols[0].num_bytes(), 73);
 }
 
 #[test]
@@ -330,9 +330,9 @@ fn bytes_strategy() -> impl Strategy<Value = &'static [u8]> {
 // A random column value
 fn column_value_strategy() -> impl Strategy<Value = ColumnValue> {
     prop_oneof![
-        10 => string_strategy().prop_map(|s| ColumnValue::Str(s)),
-        1 => bytes_strategy().prop_map(|b| ColumnValue::Bytes(b)),
-        40 => num_strategy().prop_map(|n| ColumnValue::Numerical(n)),
+        10 => string_strategy().prop_map(ColumnValue::Str),
+        1 => bytes_strategy().prop_map(ColumnValue::Bytes),
+        40 => num_strategy().prop_map(ColumnValue::Numerical),
         1 => (1u16..3u16).prop_map(|ip_addr_byte| ColumnValue::IpAddr(Ipv6Addr::new(
             127,
             0,
@@ -343,7 +343,7 @@ fn column_value_strategy() -> impl Strategy<Value = ColumnValue> {
             0,
             ip_addr_byte
         ))),
-        1 => any::<bool>().prop_map(|b| ColumnValue::Bool(b)),
+        1 => any::<bool>().prop_map(ColumnValue::Bool),
         1 => (0_679_723_993i64..1_679_723_995i64)
             .prop_map(|val| { ColumnValue::DateTime(DateTime::from_timestamp_secs(val)) })
     ]
@@ -419,8 +419,8 @@ fn build_columnar_with_mapping(
     columnar_writer
         .serialize(num_docs, old_to_new_row_ids_opt, &mut buffer)
         .unwrap();
-    let columnar_reader = ColumnarReader::open(buffer).unwrap();
-    columnar_reader
+
+    ColumnarReader::open(buffer).unwrap()
 }
 
 fn build_columnar(docs: &[Vec<(&'static str, ColumnValue)>]) -> ColumnarReader {
@@ -746,7 +746,7 @@ proptest! {
         let stack_merge_order = StackMergeOrder::stack(&columnar_readers_arr[..]).into();
         crate::merge_columnar(&columnar_readers_arr[..], &[], stack_merge_order, &mut output).unwrap();
         let merged_columnar = ColumnarReader::open(output).unwrap();
-        let concat_rows: Vec<Vec<(&'static str, ColumnValue)>> = columnar_docs.iter().cloned().flatten().collect();
+        let concat_rows: Vec<Vec<(&'static str, ColumnValue)>> = columnar_docs.iter().flatten().cloned().collect();
         let expected_merged_columnar = build_columnar(&concat_rows[..]);
         assert_columnar_eq_strict(&merged_columnar, &expected_merged_columnar);
     }
@@ -772,7 +772,7 @@ fn test_columnar_merging_empty_columnar() {
     .unwrap();
     let merged_columnar = ColumnarReader::open(output).unwrap();
     let concat_rows: Vec<Vec<(&'static str, ColumnValue)>> =
-        columnar_docs.iter().cloned().flatten().collect();
+        columnar_docs.iter().flatten().cloned().collect();
     let expected_merged_columnar = build_columnar(&concat_rows[..]);
     assert_columnar_eq_strict(&merged_columnar, &expected_merged_columnar);
 }
@@ -809,7 +809,7 @@ fn test_columnar_merging_number_columns() {
     .unwrap();
     let merged_columnar = ColumnarReader::open(output).unwrap();
     let concat_rows: Vec<Vec<(&'static str, ColumnValue)>> =
-        columnar_docs.iter().cloned().flatten().collect();
+        columnar_docs.iter().flatten().cloned().collect();
     let expected_merged_columnar = build_columnar(&concat_rows[..]);
     assert_columnar_eq_strict(&merged_columnar, &expected_merged_columnar);
 }

common/src/datetime.rs

@@ -1,11 +1,14 @@
 #![allow(deprecated)]
 
 use std::fmt;
+use std::io::{Read, Write};
 
 use serde::{Deserialize, Serialize};
 use time::format_description::well_known::Rfc3339;
 use time::{OffsetDateTime, PrimitiveDateTime, UtcOffset};
 
+use crate::BinarySerializable;
+
 /// Precision with which datetimes are truncated when stored in fast fields. This setting is only
 /// relevant for fast fields. In the docstore, datetimes are always saved with nanosecond precision.
 #[derive(
@@ -164,3 +167,15 @@ impl fmt::Debug for DateTime {
         f.write_str(&utc_rfc3339)
     }
 }
+
+impl BinarySerializable for DateTime {
+    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> std::io::Result<()> {
+        let timestamp_micros = self.into_timestamp_micros();
+        <i64 as BinarySerializable>::serialize(&timestamp_micros, writer)
+    }
+
+    fn deserialize<R: Read>(reader: &mut R) -> std::io::Result<Self> {
+        let timestamp_micros = <i64 as BinarySerializable>::deserialize(reader)?;
+        Ok(Self::from_timestamp_micros(timestamp_micros))
+    }
+}

common/src/json_path_writer.rs

@@ -0,0 +1,112 @@
+use crate::replace_in_place;
+
+/// Separates the different segments of a json path.
+pub const JSON_PATH_SEGMENT_SEP: u8 = 1u8;
+pub const JSON_PATH_SEGMENT_SEP_STR: &str =
+    unsafe { std::str::from_utf8_unchecked(&[JSON_PATH_SEGMENT_SEP]) };
+
+/// Create a new JsonPathWriter, that creates flattened json paths for tantivy.
+#[derive(Clone, Debug, Default)]
+pub struct JsonPathWriter {
+    path: String,
+    indices: Vec<usize>,
+    expand_dots: bool,
+}
+
+impl JsonPathWriter {
+    pub fn new() -> Self {
+        JsonPathWriter {
+            path: String::new(),
+            indices: Vec::new(),
+            expand_dots: false,
+        }
+    }
+
+    /// When expand_dots is enabled, json object like
+    /// `{"k8s.node.id": 5}` is processed as if it was
+    /// `{"k8s": {"node": {"id": 5}}}`.
+    /// This option has the merit of allowing users to
+    /// write queries  like `k8s.node.id:5`.
+    /// On the other, enabling that feature can lead to
+    /// ambiguity.
+    #[inline]
+    pub fn set_expand_dots(&mut self, expand_dots: bool) {
+        self.expand_dots = expand_dots;
+    }
+
+    /// Push a new segment to the path.
+    #[inline]
+    pub fn push(&mut self, segment: &str) {
+        let len_path = self.path.len();
+        self.indices.push(len_path);
+        if !self.path.is_empty() {
+            self.path.push_str(JSON_PATH_SEGMENT_SEP_STR);
+        }
+        self.path.push_str(segment);
+        if self.expand_dots {
+            // This might include the separation byte, which is ok because it is not a dot.
+            let appended_segment = &mut self.path[len_path..];
+            // The unsafe below is safe as long as b'.' and JSON_PATH_SEGMENT_SEP are
+            // valid single byte ut8 strings.
+            // By utf-8 design, they cannot be part of another codepoint.
+            unsafe {
+                replace_in_place(b'.', JSON_PATH_SEGMENT_SEP, appended_segment.as_bytes_mut())
+            };
+        }
+    }
+
+    /// Remove the last segment. Does nothing if the path is empty.
+    #[inline]
+    pub fn pop(&mut self) {
+        if let Some(last_idx) = self.indices.pop() {
+            self.path.truncate(last_idx);
+        }
+    }
+
+    /// Clear the path.
+    #[inline]
+    pub fn clear(&mut self) {
+        self.path.clear();
+        self.indices.clear();
+    }
+
+    /// Get the current path.
+    #[inline]
+    pub fn as_str(&self) -> &str {
+        &self.path
+    }
+}
+
+impl From<JsonPathWriter> for String {
+    #[inline]
+    fn from(value: JsonPathWriter) -> Self {
+        value.path
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn json_path_writer_test() {
+        let mut writer = JsonPathWriter::new();
+
+        writer.push("root");
+        assert_eq!(writer.as_str(), "root");
+
+        writer.push("child");
+        assert_eq!(writer.as_str(), "root\u{1}child");
+
+        writer.pop();
+        assert_eq!(writer.as_str(), "root");
+
+        writer.push("k8s.node.id");
+        assert_eq!(writer.as_str(), "root\u{1}k8s.node.id");
+
+        writer.set_expand_dots(true);
+        writer.pop();
+        writer.push("k8s.node.id");
+        assert_eq!(writer.as_str(), "root\u{1}k8s\u{1}node\u{1}id");
+    }
+}

common/src/lib.rs

@@ -9,6 +9,7 @@ mod byte_count;
 mod datetime;
 pub mod file_slice;
 mod group_by;
+mod json_path_writer;
 mod serialize;
 mod vint;
 mod writer;
@@ -18,6 +19,7 @@ pub use byte_count::ByteCount;
 pub use datetime::DatePrecision;
 pub use datetime::{DateTime, DateTimePrecision};
 pub use group_by::GroupByIteratorExtended;
+pub use json_path_writer::JsonPathWriter;
 pub use ownedbytes::{OwnedBytes, StableDeref};
 pub use serialize::{BinarySerializable, DeserializeFrom, FixedSize};
 pub use vint::{
@@ -116,6 +118,7 @@ pub fn u64_to_f64(val: u64) -> f64 {
 ///
 /// This function assumes that the needle is rarely contained in the bytes string
 /// and offers a fast path if the needle is not present.
+#[inline]
 pub fn replace_in_place(needle: u8, replacement: u8, bytes: &mut [u8]) {
     if !bytes.contains(&needle) {
         return;

common/src/serialize.rs

@@ -1,3 +1,4 @@
+use std::borrow::Cow;
 use std::io::{Read, Write};
 use std::{fmt, io};
 
@@ -249,6 +250,43 @@ impl BinarySerializable for String {
     }
 }
 
+impl<'a> BinarySerializable for Cow<'a, str> {
+    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
+        let data: &[u8] = self.as_bytes();
+        VInt(data.len() as u64).serialize(writer)?;
+        writer.write_all(data)
+    }
+
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Cow<'a, str>> {
+        let string_length = VInt::deserialize(reader)?.val() as usize;
+        let mut result = String::with_capacity(string_length);
+        reader
+            .take(string_length as u64)
+            .read_to_string(&mut result)?;
+        Ok(Cow::Owned(result))
+    }
+}
+
+impl<'a> BinarySerializable for Cow<'a, [u8]> {
+    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
+        VInt(self.len() as u64).serialize(writer)?;
+        for it in self.iter() {
+            it.serialize(writer)?;
+        }
+        Ok(())
+    }
+
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Cow<'a, [u8]>> {
+        let num_items = VInt::deserialize(reader)?.val();
+        let mut items: Vec<u8> = Vec::with_capacity(num_items as usize);
+        for _ in 0..num_items {
+            let item = u8::deserialize(reader)?;
+            items.push(item);
+        }
+        Ok(Cow::Owned(items))
+    }
+}
+
 #[cfg(test)]
 pub mod test {
 

examples/aggregation.rs

@@ -12,7 +12,7 @@ use tantivy::aggregation::agg_result::AggregationResults;
 use tantivy::aggregation::AggregationCollector;
 use tantivy::query::AllQuery;
 use tantivy::schema::{self, IndexRecordOption, Schema, TextFieldIndexing, FAST};
-use tantivy::Index;
+use tantivy::{Index, IndexWriter, TantivyDocument};
 
 fn main() -> tantivy::Result<()> {
     // # Create Schema
@@ -132,10 +132,10 @@ fn main() -> tantivy::Result<()> {
 
     let stream = Deserializer::from_str(data).into_iter::<Value>();
 
-    let mut index_writer = index.writer(50_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
     let mut num_indexed = 0;
     for value in stream {
-        let doc = schema.parse_document(&serde_json::to_string(&value.unwrap())?)?;
+        let doc = TantivyDocument::parse_json(&schema, &serde_json::to_string(&value.unwrap())?)?;
         index_writer.add_document(doc)?;
         num_indexed += 1;
         if num_indexed > 4 {

examples/basic_search.rs

@@ -15,7 +15,7 @@
 use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
 use tantivy::schema::*;
-use tantivy::{doc, Index, ReloadPolicy};
+use tantivy::{doc, Index, IndexWriter, ReloadPolicy};
 use tempfile::TempDir;
 
 fn main() -> tantivy::Result<()> {
@@ -75,7 +75,7 @@ fn main() -> tantivy::Result<()> {
     // Here we give tantivy a budget of `50MB`.
     // Using a bigger memory_arena for the indexer may increase
     // throughput, but 50 MB is already plenty.
-    let mut index_writer = index.writer(50_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
 
     // Let's index our documents!
     // We first need a handle on the title and the body field.
@@ -87,7 +87,7 @@ fn main() -> tantivy::Result<()> {
     let title = schema.get_field("title").unwrap();
     let body = schema.get_field("body").unwrap();
 
-    let mut old_man_doc = Document::default();
+    let mut old_man_doc = TantivyDocument::default();
     old_man_doc.add_text(title, "The Old Man and the Sea");
     old_man_doc.add_text(
         body,
@@ -164,7 +164,7 @@ fn main() -> tantivy::Result<()> {
     // will reload the index automatically after each commit.
     let reader = index
         .reader_builder()
-        .reload_policy(ReloadPolicy::OnCommit)
+        .reload_policy(ReloadPolicy::OnCommitWithDelay)
         .try_into()?;
 
     // We now need to acquire a searcher.
@@ -217,8 +217,8 @@ fn main() -> tantivy::Result<()> {
     // the document returned will only contain
     // a title.
     for (_score, doc_address) in top_docs {
-        let retrieved_doc = searcher.doc(doc_address)?;
-        println!("{}", schema.to_json(&retrieved_doc));
+        let retrieved_doc: TantivyDocument = searcher.doc(doc_address)?;
+        println!("{}", retrieved_doc.to_json(&schema));
     }
 
     // We can also get an explanation to understand

examples/custom_collector.rs

@@ -13,7 +13,7 @@ use columnar::Column;
 use tantivy::collector::{Collector, SegmentCollector};
 use tantivy::query::QueryParser;
 use tantivy::schema::{Schema, FAST, INDEXED, TEXT};
-use tantivy::{doc, Index, Score, SegmentReader};
+use tantivy::{doc, Index, IndexWriter, Score, SegmentReader};
 
 #[derive(Default)]
 struct Stats {
@@ -142,7 +142,7 @@ fn main() -> tantivy::Result<()> {
     // this example.
     let index = Index::create_in_ram(schema);
 
-    let mut index_writer = index.writer(50_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
     index_writer.add_document(doc!(
         product_name => "Super Broom 2000",
         product_description => "While it is ok for short distance travel, this broom \

examples/custom_tokenizer.rs

@@ -6,7 +6,7 @@ use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
 use tantivy::schema::*;
 use tantivy::tokenizer::NgramTokenizer;
-use tantivy::{doc, Index};
+use tantivy::{doc, Index, IndexWriter};
 
 fn main() -> tantivy::Result<()> {
     // # Defining the schema
@@ -62,7 +62,7 @@ fn main() -> tantivy::Result<()> {
     //
     // Here we use a buffer of 50MB per thread. Using a bigger
     // memory arena for the indexer can increase its throughput.
-    let mut index_writer = index.writer(50_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
     index_writer.add_document(doc!(
     title => "The Old Man and the Sea",
     body => "He was an old man who fished alone in a skiff in the Gulf Stream and \
@@ -103,8 +103,8 @@ fn main() -> tantivy::Result<()> {
     let top_docs = searcher.search(&query, &TopDocs::with_limit(10))?;
 
     for (_, doc_address) in top_docs {
-        let retrieved_doc = searcher.doc(doc_address)?;
-        println!("{}", schema.to_json(&retrieved_doc));
+        let retrieved_doc: TantivyDocument = searcher.doc(doc_address)?;
+        println!("{}", retrieved_doc.to_json(&schema));
     }
 
     Ok(())

examples/date_time_field.rs

@@ -4,8 +4,8 @@
 
 use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
-use tantivy::schema::{DateOptions, Schema, Value, INDEXED, STORED, STRING};
-use tantivy::Index;
+use tantivy::schema::{DateOptions, Document, OwnedValue, Schema, INDEXED, STORED, STRING};
+use tantivy::{Index, IndexWriter, TantivyDocument};
 
 fn main() -> tantivy::Result<()> {
     // # Defining the schema
@@ -22,16 +22,18 @@ fn main() -> tantivy::Result<()> {
     // # Indexing documents
     let index = Index::create_in_ram(schema.clone());
 
-    let mut index_writer = index.writer(50_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
     // The dates are passed as string in the RFC3339 format
-    let doc = schema.parse_document(
+    let doc = TantivyDocument::parse_json(
+        &schema,
         r#"{
         "occurred_at": "2022-06-22T12:53:50.53Z",
         "event": "pull-request"
     }"#,
     )?;
     index_writer.add_document(doc)?;
-    let doc = schema.parse_document(
+    let doc = TantivyDocument::parse_json(
+        &schema,
         r#"{
         "occurred_at": "2022-06-22T13:00:00.22Z",
         "event": "comment"
@@ -58,13 +60,13 @@ fn main() -> tantivy::Result<()> {
         let count_docs = searcher.search(&*query, &TopDocs::with_limit(4))?;
         assert_eq!(count_docs.len(), 1);
         for (_score, doc_address) in count_docs {
-            let retrieved_doc = searcher.doc(doc_address)?;
+            let retrieved_doc = searcher.doc::<TantivyDocument>(doc_address)?;
             assert!(matches!(
                 retrieved_doc.get_first(occurred_at),
-                Some(Value::Date(_))
+                Some(OwnedValue::Date(_))
             ));
             assert_eq!(
-                schema.to_json(&retrieved_doc),
+                retrieved_doc.to_json(&schema),
                 r#"{"event":["comment"],"occurred_at":["2022-06-22T13:00:00.22Z"]}"#
             );
         }

examples/deleting_updating_documents.rs

@@ -11,7 +11,7 @@
 use tantivy::collector::TopDocs;
 use tantivy::query::TermQuery;
 use tantivy::schema::*;
-use tantivy::{doc, Index, IndexReader};
+use tantivy::{doc, Index, IndexReader, IndexWriter};
 
 // A simple helper function to fetch a single document
 // given its id from our index.
@@ -19,7 +19,7 @@ use tantivy::{doc, Index, IndexReader};
 fn extract_doc_given_isbn(
     reader: &IndexReader,
     isbn_term: &Term,
-) -> tantivy::Result<Option<Document>> {
+) -> tantivy::Result<Option<TantivyDocument>> {
     let searcher = reader.searcher();
 
     // This is the simplest query you can think of.
@@ -69,10 +69,10 @@ fn main() -> tantivy::Result<()> {
 
     let index = Index::create_in_ram(schema.clone());
 
-    let mut index_writer = index.writer(50_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
 
     // Let's add a couple of documents, for the sake of the example.
-    let mut old_man_doc = Document::default();
+    let mut old_man_doc = TantivyDocument::default();
     old_man_doc.add_text(title, "The Old Man and the Sea");
     index_writer.add_document(doc!(
         isbn => "978-0099908401",
@@ -94,7 +94,7 @@ fn main() -> tantivy::Result<()> {
     // Oops our frankenstein doc seems misspelled
     let frankenstein_doc_misspelled = extract_doc_given_isbn(&reader, &frankenstein_isbn)?.unwrap();
     assert_eq!(
-        schema.to_json(&frankenstein_doc_misspelled),
+        frankenstein_doc_misspelled.to_json(&schema),
         r#"{"isbn":["978-9176370711"],"title":["Frankentein"]}"#,
     );
 
@@ -136,7 +136,7 @@ fn main() -> tantivy::Result<()> {
     // No more typo!
     let frankenstein_new_doc = extract_doc_given_isbn(&reader, &frankenstein_isbn)?.unwrap();
     assert_eq!(
-        schema.to_json(&frankenstein_new_doc),
+        frankenstein_new_doc.to_json(&schema),
         r#"{"isbn":["978-9176370711"],"title":["Frankenstein"]}"#,
     );
 

examples/faceted_search.rs

@@ -17,7 +17,7 @@
 use tantivy::collector::FacetCollector;
 use tantivy::query::{AllQuery, TermQuery};
 use tantivy::schema::*;
-use tantivy::{doc, Index};
+use tantivy::{doc, Index, IndexWriter};
 
 fn main() -> tantivy::Result<()> {
     // Let's create a temporary directory for the sake of this example
@@ -30,7 +30,7 @@ fn main() -> tantivy::Result<()> {
     let schema = schema_builder.build();
     let index = Index::create_in_ram(schema);
 
-    let mut index_writer = index.writer(30_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(30_000_000)?;
 
     // For convenience, tantivy also comes with a macro to
     // reduce the boilerplate above.

examples/faceted_search_with_tweaked_score.rs

@@ -12,7 +12,7 @@ use std::collections::HashSet;
 use tantivy::collector::TopDocs;
 use tantivy::query::BooleanQuery;
 use tantivy::schema::*;
-use tantivy::{doc, DocId, Index, Score, SegmentReader};
+use tantivy::{doc, DocId, Index, IndexWriter, Score, SegmentReader};
 
 fn main() -> tantivy::Result<()> {
     let mut schema_builder = Schema::builder();
@@ -23,7 +23,7 @@ fn main() -> tantivy::Result<()> {
     let schema = schema_builder.build();
     let index = Index::create_in_ram(schema);
 
-    let mut index_writer = index.writer(30_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(30_000_000)?;
 
     index_writer.add_document(doc!(
         title => "Fried egg",
@@ -91,11 +91,10 @@ fn main() -> tantivy::Result<()> {
             .iter()
             .map(|(_, doc_id)| {
                 searcher
-                    .doc(*doc_id)
+                    .doc::<TantivyDocument>(*doc_id)
                     .unwrap()
                     .get_first(title)
-                    .unwrap()
-                    .as_text()
+                    .and_then(|v| v.as_str())
                     .unwrap()
                     .to_owned()
             })

examples/fuzzy_search.rs

@@ -14,7 +14,7 @@
 use tantivy::collector::{Count, TopDocs};
 use tantivy::query::FuzzyTermQuery;
 use tantivy::schema::*;
-use tantivy::{doc, Index, ReloadPolicy};
+use tantivy::{doc, Index, IndexWriter, ReloadPolicy};
 use tempfile::TempDir;
 
 fn main() -> tantivy::Result<()> {
@@ -66,7 +66,7 @@ fn main() -> tantivy::Result<()> {
     // Here we give tantivy a budget of `50MB`.
     // Using a bigger memory_arena for the indexer may increase
     // throughput, but 50 MB is already plenty.
-    let mut index_writer = index.writer(50_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
 
     // Let's index our documents!
     // We first need a handle on the title and the body field.
@@ -123,7 +123,7 @@ fn main() -> tantivy::Result<()> {
     // will reload the index automatically after each commit.
     let reader = index
         .reader_builder()
-        .reload_policy(ReloadPolicy::OnCommit)
+        .reload_policy(ReloadPolicy::OnCommitWithDelay)
         .try_into()?;
 
     // We now need to acquire a searcher.
@@ -151,10 +151,10 @@ fn main() -> tantivy::Result<()> {
         assert_eq!(count, 3);
         assert_eq!(top_docs.len(), 3);
         for (score, doc_address) in top_docs {
-            let retrieved_doc = searcher.doc(doc_address)?;
             // Note that the score is not lower for the fuzzy hit.
             // There's an issue open for that: https://github.com/quickwit-oss/tantivy/issues/563
-            println!("score {score:?} doc {}", schema.to_json(&retrieved_doc));
+            let retrieved_doc: TantivyDocument = searcher.doc(doc_address)?;
+            println!("score {score:?} doc {}", retrieved_doc.to_json(&schema));
             // score 1.0 doc {"title":["The Diary of Muadib"]}
             //
             // score 1.0 doc {"title":["The Diary of a Young Girl"]}

examples/index_with_json.rs

@@ -21,7 +21,7 @@ fn main() -> tantivy::Result<()> {
     }"#;
 
     // We can parse our document
-    let _mice_and_men_doc = schema.parse_document(mice_and_men_doc_json)?;
+    let _mice_and_men_doc = TantivyDocument::parse_json(&schema, mice_and_men_doc_json)?;
 
     // Multi-valued field are allowed, they are
     // expressed in JSON by an array.
@@ -30,7 +30,7 @@ fn main() -> tantivy::Result<()> {
        "title": ["Frankenstein", "The Modern Prometheus"],
        "year": 1818
     }"#;
-    let _frankenstein_doc = schema.parse_document(frankenstein_json)?;
+    let _frankenstein_doc = TantivyDocument::parse_json(&schema, frankenstein_json)?;
 
     // Note that the schema is saved in your index directory.
     //

examples/integer_range_search.rs

@@ -5,7 +5,7 @@
 use tantivy::collector::Count;
 use tantivy::query::RangeQuery;
 use tantivy::schema::{Schema, INDEXED};
-use tantivy::{doc, Index, Result};
+use tantivy::{doc, Index, IndexWriter, Result};
 
 fn main() -> Result<()> {
     // For the sake of simplicity, this schema will only have 1 field
@@ -17,7 +17,7 @@ fn main() -> Result<()> {
     let index = Index::create_in_ram(schema);
     let reader = index.reader()?;
     {
-        let mut index_writer = index.writer_with_num_threads(1, 6_000_000)?;
+        let mut index_writer: IndexWriter = index.writer_with_num_threads(1, 6_000_000)?;
         for year in 1950u64..2019u64 {
             index_writer.add_document(doc!(year_field => year))?;
         }

examples/ip_field.rs

@@ -6,7 +6,7 @@
 use tantivy::collector::{Count, TopDocs};
 use tantivy::query::QueryParser;
 use tantivy::schema::{Schema, FAST, INDEXED, STORED, STRING};
-use tantivy::Index;
+use tantivy::{Index, IndexWriter, TantivyDocument};
 
 fn main() -> tantivy::Result<()> {
     // # Defining the schema
@@ -22,20 +22,22 @@ fn main() -> tantivy::Result<()> {
     // # Indexing documents
     let index = Index::create_in_ram(schema.clone());
 
-    let mut index_writer = index.writer(50_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
 
     // ### IPv4
     // Adding documents that contain an IPv4 address. Notice that the IP addresses are passed as
     // `String`. Since the field is of type ip, we parse the IP address from the string and store it
     // internally as IPv6.
-    let doc = schema.parse_document(
+    let doc = TantivyDocument::parse_json(
+        &schema,
         r#"{
             "ip": "192.168.0.33",
             "event_type": "login"
         }"#,
     )?;
     index_writer.add_document(doc)?;
-    let doc = schema.parse_document(
+    let doc = TantivyDocument::parse_json(
+        &schema,
         r#"{
             "ip": "192.168.0.80",
             "event_type": "checkout"
@@ -44,7 +46,8 @@ fn main() -> tantivy::Result<()> {
     index_writer.add_document(doc)?;
     // ### IPv6
     // Adding a document that contains an IPv6 address.
-    let doc = schema.parse_document(
+    let doc = TantivyDocument::parse_json(
+        &schema,
         r#"{
             "ip": "2001:0db8:85a3:0000:0000:8a2e:0370:7334",
             "event_type": "checkout"

examples/iterating_docs_and_positions.rs

@@ -10,7 +10,7 @@
 // ---
 // Importing tantivy...
 use tantivy::schema::*;
-use tantivy::{doc, DocSet, Index, Postings, TERMINATED};
+use tantivy::{doc, DocSet, Index, IndexWriter, Postings, TERMINATED};
 
 fn main() -> tantivy::Result<()> {
     // We first create a schema for the sake of the
@@ -24,7 +24,7 @@ fn main() -> tantivy::Result<()> {
 
     let index = Index::create_in_ram(schema);
 
-    let mut index_writer = index.writer_with_num_threads(1, 50_000_000)?;
+    let mut index_writer: IndexWriter = index.writer_with_num_threads(1, 50_000_000)?;
     index_writer.add_document(doc!(title => "The Old Man and the Sea"))?;
     index_writer.add_document(doc!(title => "Of Mice and Men"))?;
     index_writer.add_document(doc!(title => "The modern Promotheus"))?;

examples/json_field.rs

@@ -7,7 +7,7 @@
 use tantivy::collector::{Count, TopDocs};
 use tantivy::query::QueryParser;
 use tantivy::schema::{Schema, FAST, STORED, STRING, TEXT};
-use tantivy::Index;
+use tantivy::{Index, IndexWriter, TantivyDocument};
 
 fn main() -> tantivy::Result<()> {
     // # Defining the schema
@@ -20,8 +20,9 @@ fn main() -> tantivy::Result<()> {
     // # Indexing documents
     let index = Index::create_in_ram(schema.clone());
 
-    let mut index_writer = index.writer(50_000_000)?;
-    let doc = schema.parse_document(
+    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
+    let doc = TantivyDocument::parse_json(
+        &schema,
         r#"{
         "timestamp": "2022-02-22T23:20:50.53Z",
         "event_type": "click",
@@ -33,7 +34,8 @@ fn main() -> tantivy::Result<()> {
     }"#,
     )?;
     index_writer.add_document(doc)?;
-    let doc = schema.parse_document(
+    let doc = TantivyDocument::parse_json(
+        &schema,
         r#"{
         "timestamp": "2022-02-22T23:20:51.53Z",
         "event_type": "click",

examples/phrase_prefix_search.rs

@@ -1,7 +1,7 @@
 use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
 use tantivy::schema::*;
-use tantivy::{doc, Index, ReloadPolicy, Result};
+use tantivy::{doc, Index, IndexWriter, ReloadPolicy, Result};
 use tempfile::TempDir;
 
 fn main() -> Result<()> {
@@ -17,7 +17,7 @@ fn main() -> Result<()> {
 
     let index = Index::create_in_dir(&index_path, schema)?;
 
-    let mut index_writer = index.writer(50_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
 
     index_writer.add_document(doc!(
     title => "The Old Man and the Sea",
@@ -51,7 +51,7 @@ fn main() -> Result<()> {
 
     let reader = index
         .reader_builder()
-        .reload_policy(ReloadPolicy::OnCommit)
+        .reload_policy(ReloadPolicy::OnCommitWithDelay)
         .try_into()?;
 
     let searcher = reader.searcher();
@@ -67,8 +67,12 @@ fn main() -> Result<()> {
     let mut titles = top_docs
         .into_iter()
         .map(|(_score, doc_address)| {
-            let doc = searcher.doc(doc_address)?;
-            let title = doc.get_first(title).unwrap().as_text().unwrap().to_owned();
+            let doc = searcher.doc::<TantivyDocument>(doc_address)?;
+            let title = doc
+                .get_first(title)
+                .and_then(|v| v.as_str())
+                .unwrap()
+                .to_owned();
             Ok(title)
         })
         .collect::<Result<Vec<_>>>()?;

examples/pre_tokenized_text.rs

@@ -13,7 +13,7 @@ use tantivy::collector::{Count, TopDocs};
 use tantivy::query::TermQuery;
 use tantivy::schema::*;
 use tantivy::tokenizer::{PreTokenizedString, SimpleTokenizer, Token, TokenStream, Tokenizer};
-use tantivy::{doc, Index, ReloadPolicy};
+use tantivy::{doc, Index, IndexWriter, ReloadPolicy};
 use tempfile::TempDir;
 
 fn pre_tokenize_text(text: &str) -> Vec<Token> {
@@ -38,7 +38,7 @@ fn main() -> tantivy::Result<()> {
 
     let index = Index::create_in_dir(&index_path, schema.clone())?;
 
-    let mut index_writer = index.writer(50_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
 
     // We can create a document manually, by setting the fields
     // one by one in a Document object.
@@ -83,7 +83,7 @@ fn main() -> tantivy::Result<()> {
         }]
     }"#;
 
-    let short_man_doc = schema.parse_document(short_man_json)?;
+    let short_man_doc = TantivyDocument::parse_json(&schema, short_man_json)?;
 
     index_writer.add_document(short_man_doc)?;
 
@@ -94,7 +94,7 @@ fn main() -> tantivy::Result<()> {
 
     let reader = index
         .reader_builder()
-        .reload_policy(ReloadPolicy::OnCommit)
+        .reload_policy(ReloadPolicy::OnCommitWithDelay)
         .try_into()?;
 
     let searcher = reader.searcher();
@@ -115,8 +115,8 @@ fn main() -> tantivy::Result<()> {
     // Note that the tokens are not stored along with the original text
     // in the document store
     for (_score, doc_address) in top_docs {
-        let retrieved_doc = searcher.doc(doc_address)?;
-        println!("Document: {}", schema.to_json(&retrieved_doc));
+        let retrieved_doc: TantivyDocument = searcher.doc(doc_address)?;
+        println!("{}", retrieved_doc.to_json(&schema));
     }
 
     // In contrary to the previous query, when we search for the "man" term we

examples/snippet.rs

@@ -10,7 +10,8 @@
 use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
 use tantivy::schema::*;
-use tantivy::{doc, Index, Snippet, SnippetGenerator};
+use tantivy::snippet::{Snippet, SnippetGenerator};
+use tantivy::{doc, Index, IndexWriter};
 use tempfile::TempDir;
 
 fn main() -> tantivy::Result<()> {
@@ -27,7 +28,7 @@ fn main() -> tantivy::Result<()> {
     // # Indexing documents
     let index = Index::create_in_dir(&index_path, schema)?;
 
-    let mut index_writer = index.writer(50_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
 
     // we'll only need one doc for this example.
     index_writer.add_document(doc!(
@@ -54,13 +55,10 @@ fn main() -> tantivy::Result<()> {
     let snippet_generator = SnippetGenerator::create(&searcher, &*query, body)?;
 
     for (score, doc_address) in top_docs {
-        let doc = searcher.doc(doc_address)?;
+        let doc = searcher.doc::<TantivyDocument>(doc_address)?;
         let snippet = snippet_generator.snippet_from_doc(&doc);
         println!("Document score {score}:");
-        println!(
-            "title: {}",
-            doc.get_first(title).unwrap().as_text().unwrap()
-        );
+        println!("title: {}", doc.get_first(title).unwrap().as_str().unwrap());
         println!("snippet: {}", snippet.to_html());
         println!("custom highlighting: {}", highlight(snippet));
     }

examples/stop_words.rs

@@ -15,7 +15,7 @@ use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
 use tantivy::schema::*;
 use tantivy::tokenizer::*;
-use tantivy::{doc, Index};
+use tantivy::{doc, Index, IndexWriter};
 
 fn main() -> tantivy::Result<()> {
     // this example assumes you understand the content in `basic_search`
@@ -60,7 +60,7 @@ fn main() -> tantivy::Result<()> {
 
     index.tokenizers().register("stoppy", tokenizer);
 
-    let mut index_writer = index.writer(50_000_000)?;
+    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
 
     let title = schema.get_field("title").unwrap();
     let body = schema.get_field("body").unwrap();
@@ -105,9 +105,9 @@ fn main() -> tantivy::Result<()> {
     let top_docs = searcher.search(&query, &TopDocs::with_limit(10))?;
 
     for (score, doc_address) in top_docs {
-        let retrieved_doc = searcher.doc(doc_address)?;
+        let retrieved_doc: TantivyDocument = searcher.doc(doc_address)?;
         println!("\n==\nDocument score {score}:");
-        println!("{}", schema.to_json(&retrieved_doc));
+        println!("{}", retrieved_doc.to_json(&schema));
     }
 
     Ok(())

examples/warmer.rs

@@ -6,8 +6,8 @@ use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
 use tantivy::schema::{Schema, FAST, TEXT};
 use tantivy::{
-    doc, DocAddress, DocId, Index, Opstamp, Searcher, SearcherGeneration, SegmentId, SegmentReader,
-    Warmer,
+    doc, DocAddress, DocId, Index, IndexWriter, Opstamp, Searcher, SearcherGeneration, SegmentId,
+    SegmentReader, Warmer,
 };
 
 // This example shows how warmers can be used to
@@ -143,7 +143,7 @@ fn main() -> tantivy::Result<()> {
     const SNEAKERS: ProductId = 23222;
 
     let index = Index::create_in_ram(schema);
-    let mut writer = index.writer_with_num_threads(1, 15_000_000)?;
+    let mut writer: IndexWriter = index.writer_with_num_threads(1, 15_000_000)?;
     writer.add_document(doc!(product_id=>OLIVE_OIL, text=>"cooking olive oil from greece"))?;
     writer.add_document(doc!(product_id=>GLOVES, text=>"kitchen gloves, perfect for cooking"))?;
     writer.add_document(doc!(product_id=>SNEAKERS, text=>"uber sweet sneakers"))?;

query-grammar/src/query_grammar.rs

@@ -185,7 +185,7 @@ fn term_or_phrase(inp: &str) -> IResult<&str, UserInputLeaf> {
 fn term_or_phrase_infallible(inp: &str) -> JResult<&str, Option<UserInputLeaf>> {
     map(
         // ~* for slop/prefix, ) inside group or ast tree, ^ if boost
-        tuple_infallible((simple_term_infallible("*)^"), slop_or_prefix_val)),
+        tuple_infallible((simple_term_infallible(")^"), slop_or_prefix_val)),
         |((delimiter_phrase, (slop, prefix)), errors)| {
             let leaf = if let Some((delimiter, phrase)) = delimiter_phrase {
                 Some(
@@ -1113,6 +1113,9 @@ mod test {
         test_parse_query_to_ast_helper("'www-form-encoded'", "'www-form-encoded'");
         test_parse_query_to_ast_helper("www-form-encoded", "www-form-encoded");
         test_parse_query_to_ast_helper("www-form-encoded", "www-form-encoded");
+        test_parse_query_to_ast_helper("mr james bo?d", "(*mr *james *bo?d)");
+        test_parse_query_to_ast_helper("mr james bo*", "(*mr *james *bo*)");
+        test_parse_query_to_ast_helper("mr james b*d", "(*mr *james *b*d)");
     }
 
     #[test]

src/aggregation/agg_bench.rs

@@ -48,7 +48,7 @@ mod bench {
         let score_field_f64 = schema_builder.add_f64_field("score_f64", score_fieldtype.clone());
         let score_field_i64 = schema_builder.add_i64_field("score_i64", score_fieldtype);
         let index = Index::create_from_tempdir(schema_builder.build())?;
-        let few_terms_data = vec!["INFO", "ERROR", "WARN", "DEBUG"];
+        let few_terms_data = ["INFO", "ERROR", "WARN", "DEBUG"];
 
         let lg_norm = rand_distr::LogNormal::new(2.996f64, 0.979f64).unwrap();
 
@@ -85,7 +85,7 @@ mod bench {
             if cardinality == Cardinality::Sparse {
                 doc_with_value /= 20;
             }
-            let val_max = 1_000_000.0;
+            let _val_max = 1_000_000.0;
             for _ in 0..doc_with_value {
                 let val: f64 = rng.gen_range(0.0..1_000_000.0);
                 let json = if rng.gen_bool(0.1) {

src/aggregation/agg_limits.rs

@@ -73,9 +73,9 @@ impl AggregationLimits {
     /// Create a new ResourceLimitGuard, that will release the memory when dropped.
     pub fn new_guard(&self) -> ResourceLimitGuard {
         ResourceLimitGuard {
-            /// The counter which is shared between the aggregations for one request.
+            // The counter which is shared between the aggregations for one request.
             memory_consumption: Arc::clone(&self.memory_consumption),
-            /// The memory_limit in bytes
+            // The memory_limit in bytes
             memory_limit: self.memory_limit,
             allocated_with_the_guard: 0,
         }

src/aggregation/agg_tests.rs

@@ -9,7 +9,7 @@ use crate::aggregation::tests::{get_test_index_2_segments, get_test_index_from_v
 use crate::aggregation::DistributedAggregationCollector;
 use crate::query::{AllQuery, TermQuery};
 use crate::schema::{IndexRecordOption, Schema, FAST};
-use crate::{Index, Term};
+use crate::{Index, IndexWriter, Term};
 
 fn get_avg_req(field_name: &str) -> Aggregation {
     serde_json::from_value(json!({
@@ -586,7 +586,7 @@ fn test_aggregation_on_json_object() {
     let json = schema_builder.add_json_field("json", FAST);
     let schema = schema_builder.build();
     let index = Index::create_in_ram(schema);
-    let mut index_writer = index.writer_for_tests().unwrap();
+    let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
     index_writer
         .add_document(doc!(json => json!({"color": "red"})))
         .unwrap();
@@ -624,13 +624,72 @@ fn test_aggregation_on_json_object() {
     );
 }
 
+#[test]
+fn test_aggregation_on_nested_json_object() {
+    let mut schema_builder = Schema::builder();
+    let json = schema_builder.add_json_field("json.blub", FAST);
+    let schema = schema_builder.build();
+    let index = Index::create_in_ram(schema);
+    let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
+    index_writer
+        .add_document(doc!(json => json!({"color.dot": "red", "color": {"nested":"red"} })))
+        .unwrap();
+    index_writer
+        .add_document(doc!(json => json!({"color.dot": "blue", "color": {"nested":"blue"} })))
+        .unwrap();
+    index_writer.commit().unwrap();
+    let reader = index.reader().unwrap();
+    let searcher = reader.searcher();
+
+    let agg: Aggregations = serde_json::from_value(json!({
+        "jsonagg1": {
+            "terms": {
+                "field": "json\\.blub.color\\.dot",
+            }
+        },
+        "jsonagg2": {
+            "terms": {
+                "field": "json\\.blub.color.nested",
+            }
+        }
+
+    }))
+    .unwrap();
+
+    let aggregation_collector = get_collector(agg);
+    let aggregation_results = searcher.search(&AllQuery, &aggregation_collector).unwrap();
+    let aggregation_res_json = serde_json::to_value(aggregation_results).unwrap();
+    assert_eq!(
+        &aggregation_res_json,
+        &serde_json::json!({
+            "jsonagg1": {
+                "buckets": [
+                    {"doc_count": 1, "key": "blue"},
+                    {"doc_count": 1, "key": "red"}
+                ],
+                "doc_count_error_upper_bound": 0,
+                "sum_other_doc_count": 0
+            },
+            "jsonagg2": {
+                "buckets": [
+                    {"doc_count": 1, "key": "blue"},
+                    {"doc_count": 1, "key": "red"}
+                ],
+                "doc_count_error_upper_bound": 0,
+                "sum_other_doc_count": 0
+            }
+
+        })
+    );
+}
+
 #[test]
 fn test_aggregation_on_json_object_empty_columns() {
     let mut schema_builder = Schema::builder();
     let json = schema_builder.add_json_field("json", FAST);
     let schema = schema_builder.build();
     let index = Index::create_in_ram(schema);
-    let mut index_writer = index.writer_for_tests().unwrap();
+    let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
     // => Empty column when accessing color
     index_writer
         .add_document(doc!(json => json!({"price": 10.0})))
@@ -748,7 +807,7 @@ fn test_aggregation_on_json_object_mixed_types() {
     let json = schema_builder.add_json_field("json", FAST);
     let schema = schema_builder.build();
     let index = Index::create_in_ram(schema);
-    let mut index_writer = index.writer_for_tests().unwrap();
+    let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
     // => Segment with all values numeric
     index_writer
         .add_document(doc!(json => json!({"mixed_type": 10.0})))

src/aggregation/bucket/histogram/date_histogram.rs

@@ -252,7 +252,7 @@ pub mod tests {
     use crate::aggregation::tests::exec_request;
     use crate::indexer::NoMergePolicy;
     use crate::schema::{Schema, FAST, STRING};
-    use crate::Index;
+    use crate::{Index, IndexWriter, TantivyDocument};
 
     #[test]
     fn test_parse_into_millisecs() {
@@ -316,7 +316,7 @@ pub mod tests {
             index_writer.set_merge_policy(Box::new(NoMergePolicy));
             for values in segment_and_docs {
                 for doc_str in values {
-                    let doc = schema.parse_document(doc_str)?;
+                    let doc = TantivyDocument::parse_json(&schema, doc_str)?;
                     index_writer.add_document(doc)?;
                 }
                 // writing the segment
@@ -328,7 +328,7 @@ pub mod tests {
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
             if segment_ids.len() > 1 {
-                let mut index_writer = index.writer_for_tests()?;
+                let mut index_writer: IndexWriter = index.writer_for_tests()?;
                 index_writer.merge(&segment_ids).wait()?;
                 index_writer.wait_merging_threads()?;
             }

src/aggregation/bucket/term_agg.rs

@@ -601,7 +601,7 @@ mod tests {
     use crate::aggregation::AggregationLimits;
     use crate::indexer::NoMergePolicy;
     use crate::schema::{Schema, FAST, STRING};
-    use crate::Index;
+    use crate::{Index, IndexWriter};
 
     #[test]
     fn terms_aggregation_test_single_segment() -> crate::Result<()> {
@@ -1473,7 +1473,7 @@ mod tests {
         let json = schema_builder.add_json_field("json", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         // => Segment with empty json
         index_writer.add_document(doc!()).unwrap();
         index_writer.commit().unwrap();

src/aggregation/bucket/term_missing_agg.rs

@@ -117,7 +117,7 @@ mod tests {
     use crate::aggregation::agg_req::Aggregations;
     use crate::aggregation::tests::exec_request_with_query;
     use crate::schema::{Schema, FAST};
-    use crate::Index;
+    use crate::{Index, IndexWriter};
 
     #[test]
     fn terms_aggregation_missing_mixed_type_mult_seg_sub_agg() -> crate::Result<()> {
@@ -126,7 +126,7 @@ mod tests {
         let score = schema_builder.add_f64_field("score", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         // => Segment with all values numeric
         index_writer
             .add_document(doc!(score => 1.0, json => json!({"mixed_type": 10.0})))
@@ -186,7 +186,7 @@ mod tests {
         let score = schema_builder.add_f64_field("score", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         // => Segment with all values numeric
         index_writer.add_document(doc!(score => 1.0, json => json!({"mixed_type": 10.0})))?;
         index_writer.add_document(doc!(score => 5.0))?;
@@ -231,7 +231,7 @@ mod tests {
         let score = schema_builder.add_f64_field("score", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
 
         index_writer.add_document(doc!(score => 5.0))?;
         index_writer.commit().unwrap();
@@ -278,7 +278,7 @@ mod tests {
         let score = schema_builder.add_f64_field("score", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
 
         index_writer.add_document(doc!(score => 5.0))?;
         index_writer.add_document(doc!(score => 5.0))?;
@@ -323,7 +323,7 @@ mod tests {
         let json = schema_builder.add_json_field("json", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         // => Segment with all values numeric
         index_writer
             .add_document(doc!(json => json!({"mixed_type": 10.0})))
@@ -385,7 +385,7 @@ mod tests {
         let json = schema_builder.add_json_field("json", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         // => Segment with all values numeric
         index_writer
             .add_document(doc!(json => json!({"mixed_type": 10.0})))
@@ -427,7 +427,7 @@ mod tests {
         let json = schema_builder.add_json_field("json", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         // => Segment with all values numeric
         index_writer
             .add_document(doc!(json => json!({"mixed_type": 10.0})))

src/aggregation/metric/max.rs

@@ -71,7 +71,7 @@ mod tests {
     use crate::aggregation::agg_req::Aggregations;
     use crate::aggregation::tests::exec_request_with_query;
     use crate::schema::{Schema, FAST};
-    use crate::Index;
+    use crate::{Index, IndexWriter};
 
     #[test]
     fn test_max_agg_with_missing() -> crate::Result<()> {
@@ -79,7 +79,7 @@ mod tests {
         let json = schema_builder.add_json_field("json", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         // => Segment with empty json
         index_writer.add_document(doc!()).unwrap();
         index_writer.commit().unwrap();

src/aggregation/metric/mod.rs

@@ -88,7 +88,7 @@ mod tests {
     use crate::aggregation::AggregationCollector;
     use crate::query::AllQuery;
     use crate::schema::{NumericOptions, Schema};
-    use crate::Index;
+    use crate::{Index, IndexWriter};
 
     #[test]
     fn test_metric_aggregations() {
@@ -96,7 +96,7 @@ mod tests {
         let field_options = NumericOptions::default().set_fast();
         let field = schema_builder.add_f64_field("price", field_options);
         let index = Index::create_in_ram(schema_builder.build());
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
 
         for i in 0..3 {
             index_writer

src/aggregation/metric/stats.rs

@@ -300,7 +300,7 @@ mod tests {
     use crate::aggregation::AggregationCollector;
     use crate::query::{AllQuery, TermQuery};
     use crate::schema::{IndexRecordOption, Schema, FAST};
-    use crate::{Index, Term};
+    use crate::{Index, IndexWriter, Term};
 
     #[test]
     fn test_aggregation_stats_empty_index() -> crate::Result<()> {
@@ -494,7 +494,7 @@ mod tests {
         let json = schema_builder.add_json_field("json", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         // => Segment with empty json
         index_writer.add_document(doc!()).unwrap();
         index_writer.commit().unwrap();
@@ -541,7 +541,7 @@ mod tests {
         let json = schema_builder.add_json_field("json", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         // => Segment with empty json
         index_writer.add_document(doc!()).unwrap();
         index_writer.commit().unwrap();

src/aggregation/mod.rs

@@ -319,7 +319,7 @@ mod tests {
     use crate::indexer::NoMergePolicy;
     use crate::query::{AllQuery, TermQuery};
     use crate::schema::{IndexRecordOption, Schema, TextFieldIndexing, FAST, STRING};
-    use crate::{Index, Term};
+    use crate::{Index, IndexWriter, Term};
 
     pub fn get_test_index_with_num_docs(
         merge_segments: bool,
@@ -451,7 +451,7 @@ mod tests {
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
             if segment_ids.len() > 1 {
-                let mut index_writer = index.writer_for_tests()?;
+                let mut index_writer: IndexWriter = index.writer_for_tests()?;
                 index_writer.merge(&segment_ids).wait()?;
                 index_writer.wait_merging_threads()?;
             }
@@ -565,7 +565,7 @@ mod tests {
             let segment_ids = index
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             index_writer.merge(&segment_ids).wait()?;
             index_writer.wait_merging_threads()?;
         }

src/collector/facet_collector.rs

@@ -495,8 +495,8 @@ mod tests {
     use crate::collector::Count;
     use crate::core::Index;
     use crate::query::{AllQuery, QueryParser, TermQuery};
-    use crate::schema::{Document, Facet, FacetOptions, IndexRecordOption, Schema};
-    use crate::Term;
+    use crate::schema::{Facet, FacetOptions, IndexRecordOption, Schema, TantivyDocument};
+    use crate::{IndexWriter, Term};
 
     fn test_collapse_mapping_aux(
         facet_terms: &[&str],
@@ -559,7 +559,7 @@ mod tests {
         let facet_field = schema_builder.add_facet_field("facet", FacetOptions::default());
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(facet_field=>Facet::from("/facet/a")))
             .unwrap();
@@ -588,7 +588,7 @@ mod tests {
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
 
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         let num_facets: usize = 3 * 4 * 5;
         let facets: Vec<Facet> = (0..num_facets)
             .map(|mut n| {
@@ -601,7 +601,7 @@ mod tests {
             })
             .collect();
         for i in 0..num_facets * 10 {
-            let mut doc = Document::new();
+            let mut doc = TantivyDocument::new();
             doc.add_facet(facet_field, facets[i % num_facets].clone());
             index_writer.add_document(doc).unwrap();
         }
@@ -732,24 +732,25 @@ mod tests {
         let index = Index::create_in_ram(schema);
 
         let uniform = Uniform::new_inclusive(1, 100_000);
-        let mut docs: Vec<Document> = vec![("a", 10), ("b", 100), ("c", 7), ("d", 12), ("e", 21)]
-            .into_iter()
-            .flat_map(|(c, count)| {
-                let facet = Facet::from(&format!("/facet/{}", c));
-                let doc = doc!(facet_field => facet);
-                iter::repeat(doc).take(count)
-            })
-            .map(|mut doc| {
-                doc.add_facet(
-                    facet_field,
-                    &format!("/facet/{}", thread_rng().sample(uniform)),
-                );
-                doc
-            })
-            .collect();
+        let mut docs: Vec<TantivyDocument> =
+            vec![("a", 10), ("b", 100), ("c", 7), ("d", 12), ("e", 21)]
+                .into_iter()
+                .flat_map(|(c, count)| {
+                    let facet = Facet::from(&format!("/facet/{}", c));
+                    let doc = doc!(facet_field => facet);
+                    iter::repeat(doc).take(count)
+                })
+                .map(|mut doc| {
+                    doc.add_facet(
+                        facet_field,
+                        &format!("/facet/{}", thread_rng().sample(uniform)),
+                    );
+                    doc
+                })
+                .collect();
         docs[..].shuffle(&mut thread_rng());
 
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         for doc in docs {
             index_writer.add_document(doc).unwrap();
         }
@@ -780,7 +781,7 @@ mod tests {
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
 
-        let docs: Vec<Document> = vec![("b", 2), ("a", 2), ("c", 4)]
+        let docs: Vec<TantivyDocument> = vec![("b", 2), ("a", 2), ("c", 4)]
             .into_iter()
             .flat_map(|(c, count)| {
                 let facet = Facet::from(&format!("/facet/{}", c));
@@ -828,7 +829,7 @@ mod bench {
     use crate::collector::FacetCollector;
     use crate::query::AllQuery;
     use crate::schema::{Facet, Schema, INDEXED};
-    use crate::Index;
+    use crate::{Index, IndexWriter};
 
     #[bench]
     fn bench_facet_collector(b: &mut Bencher) {
@@ -847,7 +848,7 @@ mod bench {
         // 40425 docs
         docs[..].shuffle(&mut thread_rng());
 
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         for doc in docs {
             index_writer.add_document(doc).unwrap();
         }

src/collector/filter_collector_wrapper.rs

@@ -12,8 +12,7 @@ use std::marker::PhantomData;
 use columnar::{BytesColumn, Column, DynamicColumn, HasAssociatedColumnType};
 
 use crate::collector::{Collector, SegmentCollector};
-use crate::schema::Field;
-use crate::{DocId, Score, SegmentReader, TantivyError};
+use crate::{DocId, Score, SegmentReader};
 
 /// The `FilterCollector` filters docs using a fast field value and a predicate.
 ///
@@ -50,13 +49,13 @@ use crate::{DocId, Score, SegmentReader, TantivyError};
 ///
 /// let query_parser = QueryParser::for_index(&index, vec![title]);
 /// let query = query_parser.parse_query("diary")?;
-/// let no_filter_collector = FilterCollector::new(price, |value: u64| value > 20_120u64, TopDocs::with_limit(2));
+/// let no_filter_collector = FilterCollector::new("price".to_string(), |value: u64| value > 20_120u64, TopDocs::with_limit(2));
 /// let top_docs = searcher.search(&query, &no_filter_collector)?;
 ///
 /// assert_eq!(top_docs.len(), 1);
 /// assert_eq!(top_docs[0].1, DocAddress::new(0, 1));
 ///
-/// let filter_all_collector: FilterCollector<_, _, u64> = FilterCollector::new(price, |value| value < 5u64, TopDocs::with_limit(2));
+/// let filter_all_collector: FilterCollector<_, _, u64> = FilterCollector::new("price".to_string(), |value| value < 5u64, TopDocs::with_limit(2));
 /// let filtered_top_docs = searcher.search(&query, &filter_all_collector)?;
 ///
 /// assert_eq!(filtered_top_docs.len(), 0);
@@ -70,7 +69,7 @@ use crate::{DocId, Score, SegmentReader, TantivyError};
 pub struct FilterCollector<TCollector, TPredicate, TPredicateValue>
 where TPredicate: 'static + Clone
 {
-    field: Field,
+    field: String,
     collector: TCollector,
     predicate: TPredicate,
     t_predicate_value: PhantomData<TPredicateValue>,
@@ -83,7 +82,7 @@ where
     TPredicate: Fn(TPredicateValue) -> bool + Send + Sync + Clone,
 {
     /// Create a new `FilterCollector`.
-    pub fn new(field: Field, predicate: TPredicate, collector: TCollector) -> Self {
+    pub fn new(field: String, predicate: TPredicate, collector: TCollector) -> Self {
         Self {
             field,
             predicate,
@@ -110,18 +109,7 @@ where
         segment_local_id: u32,
         segment_reader: &SegmentReader,
     ) -> crate::Result<Self::Child> {
-        let schema = segment_reader.schema();
-        let field_entry = schema.get_field_entry(self.field);
-        if !field_entry.is_fast() {
-            return Err(TantivyError::SchemaError(format!(
-                "Field {:?} is not a fast field.",
-                field_entry.name()
-            )));
-        }
-
-        let column_opt = segment_reader
-            .fast_fields()
-            .column_opt(field_entry.name())?;
+        let column_opt = segment_reader.fast_fields().column_opt(&self.field)?;
 
         let segment_collector = self
             .collector
@@ -229,7 +217,7 @@ where
 ///
 /// let query_parser = QueryParser::for_index(&index, vec![title]);
 /// let query = query_parser.parse_query("diary")?;
-/// let filter_collector = BytesFilterCollector::new(barcode, |bytes: &[u8]| bytes.starts_with(b"01"), TopDocs::with_limit(2));
+/// let filter_collector = BytesFilterCollector::new("barcode".to_string(), |bytes: &[u8]| bytes.starts_with(b"01"), TopDocs::with_limit(2));
 /// let top_docs = searcher.search(&query, &filter_collector)?;
 ///
 /// assert_eq!(top_docs.len(), 1);
@@ -240,7 +228,7 @@ where
 pub struct BytesFilterCollector<TCollector, TPredicate>
 where TPredicate: 'static + Clone
 {
-    field: Field,
+    field: String,
     collector: TCollector,
     predicate: TPredicate,
 }
@@ -251,7 +239,7 @@ where
     TPredicate: Fn(&[u8]) -> bool + Send + Sync + Clone,
 {
     /// Create a new `BytesFilterCollector`.
-    pub fn new(field: Field, predicate: TPredicate, collector: TCollector) -> Self {
+    pub fn new(field: String, predicate: TPredicate, collector: TCollector) -> Self {
         Self {
             field,
             predicate,
@@ -274,10 +262,7 @@ where
         segment_local_id: u32,
         segment_reader: &SegmentReader,
     ) -> crate::Result<Self::Child> {
-        let schema = segment_reader.schema();
-        let field_name = schema.get_field_name(self.field);
-
-        let column_opt = segment_reader.fast_fields().bytes(field_name)?;
+        let column_opt = segment_reader.fast_fields().bytes(&self.field)?;
 
         let segment_collector = self
             .collector

src/collector/mod.rs

@@ -97,7 +97,7 @@ pub use self::multi_collector::{FruitHandle, MultiCollector, MultiFruit};
 mod top_collector;
 
 mod top_score_collector;
-pub use self::top_score_collector::TopDocs;
+pub use self::top_score_collector::{TopDocs, TopNComputer};
 
 mod custom_score_top_collector;
 pub use self::custom_score_top_collector::{CustomScorer, CustomSegmentScorer};

src/collector/tests.rs

@@ -7,7 +7,9 @@ use crate::query::{AllQuery, QueryParser};
 use crate::schema::{Schema, FAST, TEXT};
 use crate::time::format_description::well_known::Rfc3339;
 use crate::time::OffsetDateTime;
-use crate::{doc, DateTime, DocAddress, DocId, Document, Index, Score, Searcher, SegmentOrdinal};
+use crate::{
+    doc, DateTime, DocAddress, DocId, Index, Score, Searcher, SegmentOrdinal, TantivyDocument,
+};
 
 pub const TEST_COLLECTOR_WITH_SCORE: TestCollector = TestCollector {
     compute_score: true,
@@ -40,7 +42,7 @@ pub fn test_filter_collector() -> crate::Result<()> {
     let query_parser = QueryParser::for_index(&index, vec![title]);
     let query = query_parser.parse_query("diary")?;
     let filter_some_collector = FilterCollector::new(
-        price,
+        "price".to_string(),
         &|value: u64| value > 20_120u64,
         TopDocs::with_limit(2),
     );
@@ -49,8 +51,11 @@ pub fn test_filter_collector() -> crate::Result<()> {
     assert_eq!(top_docs.len(), 1);
     assert_eq!(top_docs[0].1, DocAddress::new(0, 1));
 
-    let filter_all_collector: FilterCollector<_, _, u64> =
-        FilterCollector::new(price, &|value| value < 5u64, TopDocs::with_limit(2));
+    let filter_all_collector: FilterCollector<_, _, u64> = FilterCollector::new(
+        "price".to_string(),
+        &|value| value < 5u64,
+        TopDocs::with_limit(2),
+    );
     let filtered_top_docs = searcher.search(&query, &filter_all_collector).unwrap();
 
     assert_eq!(filtered_top_docs.len(), 0);
@@ -61,7 +66,8 @@ pub fn test_filter_collector() -> crate::Result<()> {
             > 0
     }
 
-    let filter_dates_collector = FilterCollector::new(date, &date_filter, TopDocs::with_limit(5));
+    let filter_dates_collector =
+        FilterCollector::new("date".to_string(), &date_filter, TopDocs::with_limit(5));
     let filtered_date_docs = searcher.search(&query, &filter_dates_collector)?;
 
     assert_eq!(filtered_date_docs.len(), 2);
@@ -280,8 +286,8 @@ fn make_test_searcher() -> crate::Result<Searcher> {
     let schema = Schema::builder().build();
     let index = Index::create_in_ram(schema);
     let mut index_writer = index.writer_for_tests()?;
-    index_writer.add_document(Document::default())?;
-    index_writer.add_document(Document::default())?;
+    index_writer.add_document(TantivyDocument::default())?;
+    index_writer.add_document(TantivyDocument::default())?;
     index_writer.commit()?;
     Ok(index.reader()?.searcher())
 }

src/collector/top_collector.rs

@@ -1,7 +1,7 @@
 use std::cmp::Ordering;
-use std::collections::BinaryHeap;
 use std::marker::PhantomData;
 
+use super::top_score_collector::TopNComputer;
 use crate::{DocAddress, DocId, SegmentOrdinal, SegmentReader};
 
 /// Contains a feature (field, score, etc.) of a document along with the document address.
@@ -20,6 +20,14 @@ pub(crate) struct ComparableDoc<T, D> {
     pub feature: T,
     pub doc: D,
 }
+impl<T: std::fmt::Debug, D: std::fmt::Debug> std::fmt::Debug for ComparableDoc<T, D> {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("ComparableDoc")
+            .field("feature", &self.feature)
+            .field("doc", &self.doc)
+            .finish()
+    }
+}
 
 impl<T: PartialOrd, D: PartialOrd> PartialOrd for ComparableDoc<T, D> {
     fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
@@ -91,18 +99,13 @@ where T: PartialOrd + Clone
         if self.limit == 0 {
             return Ok(Vec::new());
         }
-        let mut top_collector = BinaryHeap::new();
+        let mut top_collector = TopNComputer::new(self.limit + self.offset);
         for child_fruit in children {
             for (feature, doc) in child_fruit {
-                if top_collector.len() < (self.limit + self.offset) {
-                    top_collector.push(ComparableDoc { feature, doc });
-                } else if let Some(mut head) = top_collector.peek_mut() {
-                    if head.feature < feature {
-                        *head = ComparableDoc { feature, doc };
-                    }
-                }
+                top_collector.push(ComparableDoc { feature, doc });
             }
         }
+
         Ok(top_collector
             .into_sorted_vec()
             .into_iter()
@@ -111,7 +114,7 @@ where T: PartialOrd + Clone
             .collect())
     }
 
-    pub(crate) fn for_segment<F: PartialOrd>(
+    pub(crate) fn for_segment<F: PartialOrd + Clone>(
         &self,
         segment_id: SegmentOrdinal,
         _: &SegmentReader,
@@ -136,20 +139,18 @@ where T: PartialOrd + Clone
 /// The Top Collector keeps track of the K documents
 /// sorted by type `T`.
 ///
-/// The implementation is based on a `BinaryHeap`.
+/// The implementation is based on a repeatedly truncating on the median after K * 2 documents
 /// The theoretical complexity for collecting the top `K` out of `n` documents
-/// is `O(n log K)`.
+/// is `O(n + K)`.
 pub(crate) struct TopSegmentCollector<T> {
-    limit: usize,
-    heap: BinaryHeap<ComparableDoc<T, DocId>>,
+    topn_computer: TopNComputer<T, DocId>,
     segment_ord: u32,
 }
 
-impl<T: PartialOrd> TopSegmentCollector<T> {
+impl<T: PartialOrd + Clone> TopSegmentCollector<T> {
     fn new(segment_ord: SegmentOrdinal, limit: usize) -> TopSegmentCollector<T> {
         TopSegmentCollector {
-            limit,
-            heap: BinaryHeap::with_capacity(limit),
+            topn_computer: TopNComputer::new(limit),
             segment_ord,
         }
     }
@@ -158,7 +159,7 @@ impl<T: PartialOrd> TopSegmentCollector<T> {
 impl<T: PartialOrd + Clone> TopSegmentCollector<T> {
     pub fn harvest(self) -> Vec<(T, DocAddress)> {
         let segment_ord = self.segment_ord;
-        self.heap
+        self.topn_computer
             .into_sorted_vec()
             .into_iter()
             .map(|comparable_doc| {
@@ -173,33 +174,13 @@ impl<T: PartialOrd + Clone> TopSegmentCollector<T> {
             .collect()
     }
 
-    /// Return true if more documents have been collected than the limit.
-    #[inline]
-    pub(crate) fn at_capacity(&self) -> bool {
-        self.heap.len() >= self.limit
-    }
-
     /// Collects a document scored by the given feature
     ///
     /// It collects documents until it has reached the max capacity. Once it reaches capacity, it
     /// will compare the lowest scoring item with the given one and keep whichever is greater.
     #[inline]
     pub fn collect(&mut self, doc: DocId, feature: T) {
-        if self.at_capacity() {
-            // It's ok to unwrap as long as a limit of 0 is forbidden.
-            if let Some(limit_feature) = self.heap.peek().map(|head| head.feature.clone()) {
-                if limit_feature < feature {
-                    if let Some(mut head) = self.heap.peek_mut() {
-                        head.feature = feature;
-                        head.doc = doc;
-                    }
-                }
-            }
-        } else {
-            // we have not reached capacity yet, so we can just push the
-            // element.
-            self.heap.push(ComparableDoc { feature, doc });
-        }
+        self.topn_computer.push(ComparableDoc { feature, doc });
     }
 }
 

src/collector/top_score_collector.rs

@@ -1,4 +1,3 @@
-use std::collections::BinaryHeap;
 use std::fmt;
 use std::marker::PhantomData;
 use std::sync::Arc;
@@ -86,12 +85,15 @@ where
 /// The `TopDocs` collector keeps track of the top `K` documents
 /// sorted by their score.
 ///
-/// The implementation is based on a `BinaryHeap`.
-/// The theoretical complexity for collecting the top `K` out of `n` documents
-/// is `O(n log K)`.
+/// The implementation is based on a repeatedly truncating on the median after K * 2 documents
+/// with pattern defeating QuickSort.
+/// The theoretical complexity for collecting the top `K` out of `N` documents
+/// is `O(N + K)`.
 ///
-/// This collector guarantees a stable sorting in case of a tie on the
-/// document score. As such, it is suitable to implement pagination.
+/// This collector does not guarantee a stable sorting in case of a tie on the
+/// document score, for stable sorting `PartialOrd` needs to resolve on other fields
+/// like docid in case of score equality.
+/// Only then, it is suitable for pagination.
 ///
 /// ```rust
 /// use tantivy::collector::TopDocs;
@@ -661,50 +663,35 @@ impl Collector for TopDocs {
         reader: &SegmentReader,
     ) -> crate::Result<<Self::Child as SegmentCollector>::Fruit> {
         let heap_len = self.0.limit + self.0.offset;
-        let mut heap: BinaryHeap<ComparableDoc<Score, DocId>> = BinaryHeap::with_capacity(heap_len);
+        let mut top_n = TopNComputer::new(heap_len);
 
         if let Some(alive_bitset) = reader.alive_bitset() {
             let mut threshold = Score::MIN;
-            weight.for_each_pruning(threshold, reader, &mut |doc, score| {
+            top_n.threshold = Some(threshold);
+            weight.for_each_pruning(Score::MIN, reader, &mut |doc, score| {
                 if alive_bitset.is_deleted(doc) {
                     return threshold;
                 }
-                let heap_item = ComparableDoc {
+                let doc = ComparableDoc {
                     feature: score,
                     doc,
                 };
-                if heap.len() < heap_len {
-                    heap.push(heap_item);
-                    if heap.len() == heap_len {
-                        threshold = heap.peek().map(|el| el.feature).unwrap_or(Score::MIN);
-                    }
-                    return threshold;
-                }
-                *heap.peek_mut().unwrap() = heap_item;
-                threshold = heap.peek().map(|el| el.feature).unwrap_or(Score::MIN);
+                top_n.push(doc);
+                threshold = top_n.threshold.unwrap_or(Score::MIN);
                 threshold
             })?;
         } else {
             weight.for_each_pruning(Score::MIN, reader, &mut |doc, score| {
-                let heap_item = ComparableDoc {
+                let doc = ComparableDoc {
                     feature: score,
                     doc,
                 };
-                if heap.len() < heap_len {
-                    heap.push(heap_item);
-                    // TODO the threshold is suboptimal for heap.len == heap_len
-                    if heap.len() == heap_len {
-                        return heap.peek().map(|el| el.feature).unwrap_or(Score::MIN);
-                    } else {
-                        return Score::MIN;
-                    }
-                }
-                *heap.peek_mut().unwrap() = heap_item;
-                heap.peek().map(|el| el.feature).unwrap_or(Score::MIN)
+                top_n.push(doc);
+                top_n.threshold.unwrap_or(Score::MIN)
             })?;
         }
 
-        let fruit = heap
+        let fruit = top_n
             .into_sorted_vec()
             .into_iter()
             .map(|cid| {
@@ -736,9 +723,81 @@ impl SegmentCollector for TopScoreSegmentCollector {
     }
 }
 
+/// Fast TopN Computation
+///
+/// For TopN == 0, it will be relative expensive.
+pub struct TopNComputer<Score, DocId> {
+    buffer: Vec<ComparableDoc<Score, DocId>>,
+    top_n: usize,
+    pub(crate) threshold: Option<Score>,
+}
+
+impl<Score, DocId> TopNComputer<Score, DocId>
+where
+    Score: PartialOrd + Clone,
+    DocId: Ord + Clone,
+{
+    /// Create a new `TopNComputer`.
+    /// Internally it will allocate a buffer of size `2 * top_n`.
+    pub fn new(top_n: usize) -> Self {
+        let vec_cap = top_n.max(1) * 2;
+        TopNComputer {
+            buffer: Vec::with_capacity(vec_cap),
+            top_n,
+            threshold: None,
+        }
+    }
+
+    #[inline]
+    pub(crate) fn push(&mut self, doc: ComparableDoc<Score, DocId>) {
+        if let Some(last_median) = self.threshold.clone() {
+            if doc.feature < last_median {
+                return;
+            }
+        }
+        if self.buffer.len() == self.buffer.capacity() {
+            let median = self.truncate_top_n();
+            self.threshold = Some(median);
+        }
+
+        // This is faster since it avoids the buffer resizing to be inlined from vec.push()
+        // (this is in the hot path)
+        // TODO: Replace with `push_within_capacity` when it's stabilized
+        let uninit = self.buffer.spare_capacity_mut();
+        // This cannot panic, because we truncate_median will at least remove one element, since
+        // the min capacity is 2.
+        uninit[0].write(doc);
+        // This is safe because it would panic in the line above
+        unsafe {
+            self.buffer.set_len(self.buffer.len() + 1);
+        }
+    }
+
+    #[inline(never)]
+    fn truncate_top_n(&mut self) -> Score {
+        // Use select_nth_unstable to find the top nth score
+        let (_, median_el, _) = self.buffer.select_nth_unstable(self.top_n);
+
+        let median_score = median_el.feature.clone();
+        // Remove all elements below the top_n
+        self.buffer.truncate(self.top_n);
+
+        median_score
+    }
+
+    pub(crate) fn into_sorted_vec(mut self) -> Vec<ComparableDoc<Score, DocId>> {
+        if self.buffer.len() > self.top_n {
+            self.truncate_top_n();
+        }
+        self.buffer.sort_unstable();
+        self.buffer
+    }
+}
+
 #[cfg(test)]
 mod tests {
-    use super::TopDocs;
+    use super::{TopDocs, TopNComputer};
+    use crate::collector::top_collector::ComparableDoc;
     use crate::collector::Collector;
     use crate::query::{AllQuery, Query, QueryParser};
     use crate::schema::{Field, Schema, FAST, STORED, TEXT};
@@ -767,6 +826,78 @@ mod tests {
         }
     }
 
+    #[test]
+    fn test_empty_topn_computer() {
+        let mut computer: TopNComputer<u32, u32> = TopNComputer::new(0);
+
+        computer.push(ComparableDoc {
+            feature: 1u32,
+            doc: 1u32,
+        });
+        computer.push(ComparableDoc {
+            feature: 1u32,
+            doc: 2u32,
+        });
+        computer.push(ComparableDoc {
+            feature: 1u32,
+            doc: 3u32,
+        });
+        assert!(computer.into_sorted_vec().is_empty());
+    }
+    #[test]
+    fn test_topn_computer() {
+        let mut computer: TopNComputer<u32, u32> = TopNComputer::new(2);
+
+        computer.push(ComparableDoc {
+            feature: 1u32,
+            doc: 1u32,
+        });
+        computer.push(ComparableDoc {
+            feature: 2u32,
+            doc: 2u32,
+        });
+        computer.push(ComparableDoc {
+            feature: 3u32,
+            doc: 3u32,
+        });
+        computer.push(ComparableDoc {
+            feature: 2u32,
+            doc: 4u32,
+        });
+        computer.push(ComparableDoc {
+            feature: 1u32,
+            doc: 5u32,
+        });
+        assert_eq!(
+            computer.into_sorted_vec(),
+            &[
+                ComparableDoc {
+                    feature: 3u32,
+                    doc: 3u32,
+                },
+                ComparableDoc {
+                    feature: 2u32,
+                    doc: 2u32,
+                }
+            ]
+        );
+    }
+
+    #[test]
+    fn test_topn_computer_no_panic() {
+        for top_n in 0..10 {
+            let mut computer: TopNComputer<u32, u32> = TopNComputer::new(top_n);
+
+            for _ in 0..1 + top_n * 2 {
+                computer.push(ComparableDoc {
+                    feature: 1u32,
+                    doc: 1u32,
+                });
+            }
+            let _vals = computer.into_sorted_vec();
+        }
+    }
+
     #[test]
     fn test_top_collector_not_at_capacity_without_offset() -> crate::Result<()> {
         let index = make_index()?;
@@ -852,20 +983,25 @@ mod tests {
         // using AllQuery to get a constant score
         let searcher = index.reader().unwrap().searcher();
 
+        let page_0 = searcher.search(&AllQuery, &TopDocs::with_limit(1)).unwrap();
+
         let page_1 = searcher.search(&AllQuery, &TopDocs::with_limit(2)).unwrap();
 
         let page_2 = searcher.search(&AllQuery, &TopDocs::with_limit(3)).unwrap();
 
         // precondition for the test to be meaningful: we did get documents
         // with the same score
+        assert!(page_0.iter().all(|result| result.0 == page_1[0].0));
         assert!(page_1.iter().all(|result| result.0 == page_1[0].0));
         assert!(page_2.iter().all(|result| result.0 == page_2[0].0));
 
         // sanity check since we're relying on make_index()
+        assert_eq!(page_0.len(), 1);
         assert_eq!(page_1.len(), 2);
         assert_eq!(page_2.len(), 3);
 
         assert_eq!(page_1, &page_2[..page_1.len()]);
+        assert_eq!(page_0, &page_2[..page_0.len()]);
     }
 
     #[test]

src/core/index.rs

@@ -18,10 +18,12 @@ use crate::directory::{Directory, ManagedDirectory, RamDirectory, INDEX_WRITER_L
 use crate::error::{DataCorruption, TantivyError};
 use crate::indexer::index_writer::{MAX_NUM_THREAD, MEMORY_BUDGET_NUM_BYTES_MIN};
 use crate::indexer::segment_updater::save_metas;
+use crate::indexer::IndexWriter;
 use crate::reader::{IndexReader, IndexReaderBuilder};
+use crate::schema::document::Document;
 use crate::schema::{Field, FieldType, Schema};
 use crate::tokenizer::{TextAnalyzer, TokenizerManager};
-use crate::IndexWriter;
+use crate::{merge_field_meta_data, FieldMetadata, SegmentReader};
 
 fn load_metas(
     directory: &dyn Directory,
@@ -184,11 +186,11 @@ impl IndexBuilder {
     ///
     /// It expects an originally empty directory, and will not run any GC operation.
     #[doc(hidden)]
-    pub fn single_segment_index_writer(
+    pub fn single_segment_index_writer<D: Document>(
         self,
         dir: impl Into<Box<dyn Directory>>,
         mem_budget: usize,
-    ) -> crate::Result<SingleSegmentIndexWriter> {
+    ) -> crate::Result<SingleSegmentIndexWriter<D>> {
         let index = self.create(dir)?;
         let index_simple_writer = SingleSegmentIndexWriter::new(index, mem_budget)?;
         Ok(index_simple_writer)
@@ -488,6 +490,28 @@ impl Index {
         self.inventory.all()
     }
 
+    /// Returns the list of fields that have been indexed in the Index.
+    /// The field list includes the field defined in the schema as well as the fields
+    /// that have been indexed as a part of a JSON field.
+    /// The returned field name is the full field name, including the name of the JSON field.
+    ///
+    /// The returned field names can be used in queries.
+    ///
+    /// Notice: If your data contains JSON fields this is **very expensive**, as it requires
+    /// browsing through the inverted index term dictionary and the columnar field dictionary.
+    ///
+    /// Disclaimer: Some fields may not be listed here. For instance, if the schema contains a json
+    /// field that is not indexed nor a fast field but is stored, it is possible for the field
+    /// to not be listed.
+    pub fn fields_metadata(&self) -> crate::Result<Vec<FieldMetadata>> {
+        let segments = self.searchable_segments()?;
+        let fields_metadata: Vec<Vec<FieldMetadata>> = segments
+            .into_iter()
+            .map(|segment| SegmentReader::open(&segment)?.fields_metadata())
+            .collect::<Result<_, _>>()?;
+        Ok(merge_field_meta_data(fields_metadata, &self.schema()))
+    }
+
     /// Creates a new segment_meta (Advanced user only).
     ///
     /// As long as the `SegmentMeta` lives, the files associated with the
@@ -531,11 +555,11 @@ impl Index {
     /// If the lockfile already exists, returns `Error::DirectoryLockBusy` or an `Error::IoError`.
     /// If the memory arena per thread is too small or too big, returns
     /// `TantivyError::InvalidArgument`
-    pub fn writer_with_num_threads(
+    pub fn writer_with_num_threads<D: Document>(
         &self,
         num_threads: usize,
         overall_memory_budget_in_bytes: usize,
-    ) -> crate::Result<IndexWriter> {
+    ) -> crate::Result<IndexWriter<D>> {
         let directory_lock = self
             .directory
             .acquire_lock(&INDEX_WRITER_LOCK)
@@ -564,8 +588,8 @@ impl Index {
     /// That index writer only simply has a single thread and a memory budget of 15 MB.
     /// Using a single thread gives us a deterministic allocation of DocId.
     #[cfg(test)]
-    pub fn writer_for_tests(&self) -> crate::Result<IndexWriter> {
-        self.writer_with_num_threads(1, 15_000_000)
+    pub fn writer_for_tests<D: Document>(&self) -> crate::Result<IndexWriter<D>> {
+        self.writer_with_num_threads(1, MEMORY_BUDGET_NUM_BYTES_MIN)
     }
 
     /// Creates a multithreaded writer
@@ -579,7 +603,10 @@ impl Index {
     /// If the lockfile already exists, returns `Error::FileAlreadyExists`.
     /// If the memory arena per thread is too small or too big, returns
     /// `TantivyError::InvalidArgument`
-    pub fn writer(&self, memory_budget_in_bytes: usize) -> crate::Result<IndexWriter> {
+    pub fn writer<D: Document>(
+        &self,
+        memory_budget_in_bytes: usize,
+    ) -> crate::Result<IndexWriter<D>> {
         let mut num_threads = std::cmp::min(num_cpus::get(), MAX_NUM_THREAD);
         let memory_budget_num_bytes_per_thread = memory_budget_in_bytes / num_threads;
         if memory_budget_num_bytes_per_thread < MEMORY_BUDGET_NUM_BYTES_MIN {

src/core/inverted_index_reader.rs

@@ -1,11 +1,12 @@
 use std::io;
 
 use common::BinarySerializable;
+use fnv::FnvHashSet;
 
 use crate::directory::FileSlice;
 use crate::positions::PositionReader;
 use crate::postings::{BlockSegmentPostings, SegmentPostings, TermInfo};
-use crate::schema::{IndexRecordOption, Term};
+use crate::schema::{IndexRecordOption, Term, Type, JSON_END_OF_PATH};
 use crate::termdict::TermDictionary;
 
 /// The inverted index reader is in charge of accessing
@@ -69,6 +70,28 @@ impl InvertedIndexReader {
         &self.termdict
     }
 
+    /// Return the fields and types encoded in the dictionary in lexicographic oder.
+    /// Only valid on JSON fields.
+    ///
+    /// Notice: This requires a full scan and therefore **very expensive**.
+    /// TODO: Move to sstable to use the index.
+    pub fn list_encoded_fields(&self) -> io::Result<Vec<(String, Type)>> {
+        let mut stream = self.termdict.stream()?;
+        let mut fields = Vec::new();
+        let mut fields_set = FnvHashSet::default();
+        while let Some((term, _term_info)) = stream.next() {
+            if let Some(index) = term.iter().position(|&byte| byte == JSON_END_OF_PATH) {
+                if !fields_set.contains(&term[..index + 2]) {
+                    fields_set.insert(term[..index + 2].to_vec());
+                    let typ = Type::from_code(term[index + 1]).unwrap();
+                    fields.push((String::from_utf8_lossy(&term[..index]).to_string(), typ));
+                }
+            }
+        }
+
+        Ok(fields)
+    }
+
     /// Resets the block segment to another position of the postings
     /// file.
     ///

src/core/json_utils.rs

@@ -1,11 +1,11 @@
 use columnar::MonotonicallyMappableToU64;
-use common::replace_in_place;
-use murmurhash32::murmurhash2;
+use common::{replace_in_place, JsonPathWriter};
 use rustc_hash::FxHashMap;
 
 use crate::fastfield::FastValue;
 use crate::postings::{IndexingContext, IndexingPosition, PostingsWriter};
-use crate::schema::term::{JSON_PATH_SEGMENT_SEP, JSON_PATH_SEGMENT_SEP_STR};
+use crate::schema::document::{ReferenceValue, ReferenceValueLeaf, Value};
+use crate::schema::term::JSON_PATH_SEGMENT_SEP;
 use crate::schema::{Field, Type, DATE_TIME_PRECISION_INDEXED};
 use crate::time::format_description::well_known::Rfc3339;
 use crate::time::{OffsetDateTime, UtcOffset};
@@ -57,31 +57,41 @@ struct IndexingPositionsPerPath {
 }
 
 impl IndexingPositionsPerPath {
-    fn get_position(&mut self, term: &Term) -> &mut IndexingPosition {
-        self.positions_per_path
-            .entry(murmurhash2(term.serialized_term()))
-            .or_default()
+    fn get_position_from_id(&mut self, id: u32) -> &mut IndexingPosition {
+        self.positions_per_path.entry(id).or_default()
     }
 }
 
-pub(crate) fn index_json_values<'a>(
+/// Convert JSON_PATH_SEGMENT_SEP to a dot.
+pub fn json_path_sep_to_dot(path: &mut str) {
+    // This is safe since we are replacing a ASCII character by another ASCII character.
+    unsafe {
+        replace_in_place(JSON_PATH_SEGMENT_SEP, b'.', path.as_bytes_mut());
+    }
+}
+
+#[allow(clippy::too_many_arguments)]
+pub(crate) fn index_json_values<'a, V: Value<'a>>(
     doc: DocId,
-    json_values: impl Iterator<Item = crate::Result<&'a serde_json::Map<String, serde_json::Value>>>,
+    json_visitors: impl Iterator<Item = crate::Result<V::ObjectIter>>,
     text_analyzer: &mut TextAnalyzer,
     expand_dots_enabled: bool,
     term_buffer: &mut Term,
     postings_writer: &mut dyn PostingsWriter,
+    json_path_writer: &mut JsonPathWriter,
     ctx: &mut IndexingContext,
 ) -> crate::Result<()> {
-    let mut json_term_writer = JsonTermWriter::wrap(term_buffer, expand_dots_enabled);
+    json_path_writer.clear();
+    json_path_writer.set_expand_dots(expand_dots_enabled);
     let mut positions_per_path: IndexingPositionsPerPath = Default::default();
-    for json_value_res in json_values {
-        let json_value = json_value_res?;
-        index_json_object(
+    for json_visitor_res in json_visitors {
+        let json_visitor = json_visitor_res?;
+        index_json_object::<V>(
             doc,
-            json_value,
+            json_visitor,
             text_analyzer,
-            &mut json_term_writer,
+            term_buffer,
+            json_path_writer,
             postings_writer,
             ctx,
             &mut positions_per_path,
@@ -90,93 +100,154 @@ pub(crate) fn index_json_values<'a>(
     Ok(())
 }
 
-fn index_json_object(
+#[allow(clippy::too_many_arguments)]
+fn index_json_object<'a, V: Value<'a>>(
     doc: DocId,
-    json_value: &serde_json::Map<String, serde_json::Value>,
+    json_visitor: V::ObjectIter,
     text_analyzer: &mut TextAnalyzer,
-    json_term_writer: &mut JsonTermWriter,
+    term_buffer: &mut Term,
+    json_path_writer: &mut JsonPathWriter,
     postings_writer: &mut dyn PostingsWriter,
     ctx: &mut IndexingContext,
     positions_per_path: &mut IndexingPositionsPerPath,
 ) {
-    for (json_path_segment, json_value) in json_value {
-        json_term_writer.push_path_segment(json_path_segment);
+    for (json_path_segment, json_value_visitor) in json_visitor {
+        json_path_writer.push(json_path_segment);
         index_json_value(
             doc,
-            json_value,
+            json_value_visitor,
             text_analyzer,
-            json_term_writer,
+            term_buffer,
+            json_path_writer,
             postings_writer,
             ctx,
             positions_per_path,
         );
-        json_term_writer.pop_path_segment();
+        json_path_writer.pop();
     }
 }
 
-fn index_json_value(
+#[allow(clippy::too_many_arguments)]
+fn index_json_value<'a, V: Value<'a>>(
     doc: DocId,
-    json_value: &serde_json::Value,
+    json_value: V,
     text_analyzer: &mut TextAnalyzer,
-    json_term_writer: &mut JsonTermWriter,
+    term_buffer: &mut Term,
+    json_path_writer: &mut JsonPathWriter,
     postings_writer: &mut dyn PostingsWriter,
     ctx: &mut IndexingContext,
     positions_per_path: &mut IndexingPositionsPerPath,
 ) {
-    match json_value {
-        serde_json::Value::Null => {}
-        serde_json::Value::Bool(val_bool) => {
-            json_term_writer.set_fast_value(*val_bool);
-            postings_writer.subscribe(doc, 0u32, json_term_writer.term(), ctx);
-        }
-        serde_json::Value::Number(number) => {
-            if let Some(number_i64) = number.as_i64() {
-                json_term_writer.set_fast_value(number_i64);
-            } else if let Some(number_u64) = number.as_u64() {
-                json_term_writer.set_fast_value(number_u64);
-            } else if let Some(number_f64) = number.as_f64() {
-                json_term_writer.set_fast_value(number_f64);
-            }
-            postings_writer.subscribe(doc, 0u32, json_term_writer.term(), ctx);
-        }
-        serde_json::Value::String(text) => match infer_type_from_str(text) {
-            TextOrDateTime::Text(text) => {
-                let mut token_stream = text_analyzer.token_stream(text);
-                // TODO make sure the chain position works out.
-                json_term_writer.close_path_and_set_type(Type::Str);
-                let indexing_position = positions_per_path.get_position(json_term_writer.term());
+    let set_path_id = |term_buffer: &mut Term, unordered_id: u32| {
+        term_buffer.truncate_value_bytes(0);
+        term_buffer.append_bytes(&unordered_id.to_be_bytes());
+    };
+    let set_type = |term_buffer: &mut Term, typ: Type| {
+        term_buffer.append_bytes(&[typ.to_code()]);
+    };
+
+    match json_value.as_value() {
+        ReferenceValue::Leaf(leaf) => match leaf {
+            ReferenceValueLeaf::Null => {}
+            ReferenceValueLeaf::Str(val) => {
+                let mut token_stream = text_analyzer.token_stream(val);
+                let unordered_id = ctx
+                    .path_to_unordered_id
+                    .get_or_allocate_unordered_id(json_path_writer.as_str());
+
+                // TODO: make sure the chain position works out.
+                set_path_id(term_buffer, unordered_id);
+                set_type(term_buffer, Type::Str);
+                let indexing_position = positions_per_path.get_position_from_id(unordered_id);
                 postings_writer.index_text(
                     doc,
                     &mut *token_stream,
-                    json_term_writer.term_buffer,
+                    term_buffer,
                     ctx,
                     indexing_position,
                 );
             }
-            TextOrDateTime::DateTime(dt) => {
-                json_term_writer.set_fast_value(DateTime::from_utc(dt));
-                postings_writer.subscribe(doc, 0u32, json_term_writer.term(), ctx);
+            ReferenceValueLeaf::U64(val) => {
+                set_path_id(
+                    term_buffer,
+                    ctx.path_to_unordered_id
+                        .get_or_allocate_unordered_id(json_path_writer.as_str()),
+                );
+                term_buffer.append_type_and_fast_value(val);
+                postings_writer.subscribe(doc, 0u32, term_buffer, ctx);
+            }
+            ReferenceValueLeaf::I64(val) => {
+                set_path_id(
+                    term_buffer,
+                    ctx.path_to_unordered_id
+                        .get_or_allocate_unordered_id(json_path_writer.as_str()),
+                );
+                term_buffer.append_type_and_fast_value(val);
+                postings_writer.subscribe(doc, 0u32, term_buffer, ctx);
+            }
+            ReferenceValueLeaf::F64(val) => {
+                set_path_id(
+                    term_buffer,
+                    ctx.path_to_unordered_id
+                        .get_or_allocate_unordered_id(json_path_writer.as_str()),
+                );
+                term_buffer.append_type_and_fast_value(val);
+                postings_writer.subscribe(doc, 0u32, term_buffer, ctx);
+            }
+            ReferenceValueLeaf::Bool(val) => {
+                set_path_id(
+                    term_buffer,
+                    ctx.path_to_unordered_id
+                        .get_or_allocate_unordered_id(json_path_writer.as_str()),
+                );
+                term_buffer.append_type_and_fast_value(val);
+                postings_writer.subscribe(doc, 0u32, term_buffer, ctx);
+            }
+            ReferenceValueLeaf::Date(val) => {
+                set_path_id(
+                    term_buffer,
+                    ctx.path_to_unordered_id
+                        .get_or_allocate_unordered_id(json_path_writer.as_str()),
+                );
+                term_buffer.append_type_and_fast_value(val);
+                postings_writer.subscribe(doc, 0u32, term_buffer, ctx);
+            }
+            ReferenceValueLeaf::PreTokStr(_) => {
+                unimplemented!(
+                    "Pre-tokenized string support in dynamic fields is not yet implemented"
+                )
+            }
+            ReferenceValueLeaf::Bytes(_) => {
+                unimplemented!("Bytes support in dynamic fields is not yet implemented")
+            }
+            ReferenceValueLeaf::Facet(_) => {
+                unimplemented!("Facet support in dynamic fields is not yet implemented")
+            }
+            ReferenceValueLeaf::IpAddr(_) => {
+                unimplemented!("IP address support in dynamic fields is not yet implemented")
             }
         },
-        serde_json::Value::Array(arr) => {
-            for val in arr {
+        ReferenceValue::Array(elements) => {
+            for val in elements {
                 index_json_value(
                     doc,
                     val,
                     text_analyzer,
-                    json_term_writer,
+                    term_buffer,
+                    json_path_writer,
                     postings_writer,
                     ctx,
                     positions_per_path,
                 );
             }
         }
-        serde_json::Value::Object(map) => {
-            index_json_object(
+        ReferenceValue::Object(object) => {
+            index_json_object::<V>(
                 doc,
-                map,
+                object,
                 text_analyzer,
-                json_term_writer,
+                term_buffer,
+                json_path_writer,
                 postings_writer,
                 ctx,
                 positions_per_path,
@@ -185,21 +256,6 @@ fn index_json_value(
     }
 }
 
-enum TextOrDateTime<'a> {
-    Text(&'a str),
-    DateTime(OffsetDateTime),
-}
-
-fn infer_type_from_str(text: &str) -> TextOrDateTime {
-    match OffsetDateTime::parse(text, &Rfc3339) {
-        Ok(dt) => {
-            let dt_utc = dt.to_offset(UtcOffset::UTC);
-            TextOrDateTime::DateTime(dt_utc)
-        }
-        Err(_) => TextOrDateTime::Text(text),
-    }
-}
-
 // Tries to infer a JSON type from a string.
 pub fn convert_to_fast_value_and_get_term(
     json_term_writer: &mut JsonTermWriter,
@@ -272,7 +328,7 @@ pub struct JsonTermWriter<'a> {
 /// In other words,
 /// - `k8s.node` ends up as `["k8s", "node"]`.
 /// - `k8s\.node` ends up as `["k8s.node"]`.
-fn split_json_path(json_path: &str) -> Vec<String> {
+pub fn split_json_path(json_path: &str) -> Vec<String> {
     let mut escaped_state: bool = false;
     let mut json_path_segments = Vec::new();
     let mut buffer = String::new();
@@ -312,17 +368,13 @@ pub(crate) fn encode_column_name(
     json_path: &str,
     expand_dots_enabled: bool,
 ) -> String {
-    let mut column_key: String = String::with_capacity(field_name.len() + json_path.len() + 1);
-    column_key.push_str(field_name);
-    for mut segment in split_json_path(json_path) {
-        column_key.push_str(JSON_PATH_SEGMENT_SEP_STR);
-        if expand_dots_enabled {
-            // We need to replace `.` by JSON_PATH_SEGMENT_SEP.
-            unsafe { replace_in_place(b'.', JSON_PATH_SEGMENT_SEP, segment.as_bytes_mut()) };
-        }
-        column_key.push_str(&segment);
+    let mut path = JsonPathWriter::default();
+    path.push(field_name);
+    path.set_expand_dots(expand_dots_enabled);
+    for segment in split_json_path(json_path) {
+        path.push(&segment);
     }
-    column_key
+    path.into()
 }
 
 impl<'a> JsonTermWriter<'a> {
@@ -362,6 +414,7 @@ impl<'a> JsonTermWriter<'a> {
         self.term_buffer.append_bytes(&[typ.to_code()]);
     }
 
+    // TODO: Remove this function and use JsonPathWriter instead.
     pub fn push_path_segment(&mut self, segment: &str) {
         // the path stack should never be empty.
         self.trim_to_end_of_path();

src/core/mod.rs

@@ -25,7 +25,7 @@ pub use self::searcher::{Searcher, SearcherGeneration};
 pub use self::segment::Segment;
 pub use self::segment_component::SegmentComponent;
 pub use self::segment_id::SegmentId;
-pub use self::segment_reader::SegmentReader;
+pub use self::segment_reader::{merge_field_meta_data, FieldMetadata, SegmentReader};
 pub use self::single_segment_index_writer::SingleSegmentIndexWriter;
 
 /// The meta file contains all the information about the list of segments and the schema

src/core/searcher.rs

@@ -5,7 +5,8 @@ use std::{fmt, io};
 use crate::collector::Collector;
 use crate::core::{Executor, SegmentReader};
 use crate::query::{Bm25StatisticsProvider, EnableScoring, Query};
-use crate::schema::{Document, Schema, Term};
+use crate::schema::document::DocumentDeserialize;
+use crate::schema::{Schema, Term};
 use crate::space_usage::SearcherSpaceUsage;
 use crate::store::{CacheStats, StoreReader};
 use crate::{DocAddress, Index, Opstamp, SegmentId, TrackedObject};
@@ -83,7 +84,7 @@ impl Searcher {
     ///
     /// The searcher uses the segment ordinal to route the
     /// request to the right `Segment`.
-    pub fn doc(&self, doc_address: DocAddress) -> crate::Result<Document> {
+    pub fn doc<D: DocumentDeserialize>(&self, doc_address: DocAddress) -> crate::Result<D> {
         let store_reader = &self.inner.store_readers[doc_address.segment_ord as usize];
         store_reader.get(doc_address.doc_id)
     }
@@ -103,7 +104,10 @@ impl Searcher {
 
     /// Fetches a document in an asynchronous manner.
     #[cfg(feature = "quickwit")]
-    pub async fn doc_async(&self, doc_address: DocAddress) -> crate::Result<Document> {
+    pub async fn doc_async<D: DocumentDeserialize>(
+        &self,
+        doc_address: DocAddress,
+    ) -> crate::Result<D> {
         let store_reader = &self.inner.store_readers[doc_address.segment_ord as usize];
         store_reader.get_async(doc_address.doc_id).await
     }

src/core/segment_reader.rs

@@ -1,12 +1,17 @@
 use std::collections::HashMap;
+use std::ops::BitOrAssign;
 use std::sync::{Arc, RwLock};
 use std::{fmt, io};
 
+use fnv::FnvHashMap;
+use itertools::Itertools;
+
 use crate::core::{InvertedIndexReader, Segment, SegmentComponent, SegmentId};
 use crate::directory::{CompositeFile, FileSlice};
 use crate::error::DataCorruption;
 use crate::fastfield::{intersect_alive_bitsets, AliveBitSet, FacetReader, FastFieldReaders};
 use crate::fieldnorm::{FieldNormReader, FieldNormReaders};
+use crate::json_utils::json_path_sep_to_dot;
 use crate::schema::{Field, IndexRecordOption, Schema, Type};
 use crate::space_usage::SegmentSpaceUsage;
 use crate::store::StoreReader;
@@ -280,6 +285,103 @@ impl SegmentReader {
         Ok(inv_idx_reader)
     }
 
+    /// Returns the list of fields that have been indexed in the segment.
+    /// The field list includes the field defined in the schema as well as the fields
+    /// that have been indexed as a part of a JSON field.
+    /// The returned field name is the full field name, including the name of the JSON field.
+    ///
+    /// The returned field names can be used in queries.
+    ///
+    /// Notice: If your data contains JSON fields this is **very expensive**, as it requires
+    /// browsing through the inverted index term dictionary and the columnar field dictionary.
+    ///
+    /// Disclaimer: Some fields may not be listed here. For instance, if the schema contains a json
+    /// field that is not indexed nor a fast field but is stored, it is possible for the field
+    /// to not be listed.
+    pub fn fields_metadata(&self) -> crate::Result<Vec<FieldMetadata>> {
+        let mut indexed_fields: Vec<FieldMetadata> = Vec::new();
+        let mut map_to_canonical = FnvHashMap::default();
+        for (field, field_entry) in self.schema().fields() {
+            let field_name = field_entry.name().to_string();
+            let is_indexed = field_entry.is_indexed();
+
+            if is_indexed {
+                let is_json = field_entry.field_type().value_type() == Type::Json;
+                if is_json {
+                    let inv_index = self.inverted_index(field)?;
+                    let encoded_fields_in_index = inv_index.list_encoded_fields()?;
+                    let mut build_path = |field_name: &str, mut json_path: String| {
+                        // In this case we need to map the potential fast field to the field name
+                        // accepted by the query parser.
+                        let create_canonical =
+                            !field_entry.is_expand_dots_enabled() && json_path.contains('.');
+                        if create_canonical {
+                            // Without expand dots enabled dots need to be escaped.
+                            let escaped_json_path = json_path.replace('.', "\\.");
+                            let full_path = format!("{}.{}", field_name, escaped_json_path);
+                            let full_path_unescaped = format!("{}.{}", field_name, &json_path);
+                            map_to_canonical.insert(full_path_unescaped, full_path.to_string());
+                            full_path
+                        } else {
+                            // With expand dots enabled, we can use '.' instead of '\u{1}'.
+                            json_path_sep_to_dot(&mut json_path);
+                            format!("{}.{}", field_name, json_path)
+                        }
+                    };
+                    indexed_fields.extend(
+                        encoded_fields_in_index
+                            .into_iter()
+                            .map(|(name, typ)| (build_path(&field_name, name), typ))
+                            .map(|(field_name, typ)| FieldMetadata {
+                                indexed: true,
+                                stored: false,
+                                field_name,
+                                fast: false,
+                                typ,
+                            }),
+                    );
+                } else {
+                    indexed_fields.push(FieldMetadata {
+                        indexed: true,
+                        stored: false,
+                        field_name: field_name.to_string(),
+                        fast: false,
+                        typ: field_entry.field_type().value_type(),
+                    });
+                }
+            }
+        }
+        let mut fast_fields: Vec<FieldMetadata> = self
+            .fast_fields()
+            .columnar()
+            .iter_columns()?
+            .map(|(mut field_name, handle)| {
+                json_path_sep_to_dot(&mut field_name);
+                // map to canonical path, to avoid similar but different entries.
+                // Eventually we should just accept '.' seperated for all cases.
+                let field_name = map_to_canonical
+                    .get(&field_name)
+                    .unwrap_or(&field_name)
+                    .to_string();
+                FieldMetadata {
+                    indexed: false,
+                    stored: false,
+                    field_name,
+                    fast: true,
+                    typ: Type::from(handle.column_type()),
+                }
+            })
+            .collect();
+        // Since the type is encoded differently in the fast field and in the inverted index,
+        // the order of the fields is not guaranteed to be the same. Therefore, we sort the fields.
+        // If we are sure that the order is the same, we can remove this sort.
+        indexed_fields.sort_unstable();
+        fast_fields.sort_unstable();
+        let merged = merge_field_meta_data(vec![indexed_fields, fast_fields], &self.schema);
+
+        Ok(merged)
+    }
+
     /// Returns the segment id
     pub fn segment_id(&self) -> SegmentId {
         self.segment_id
@@ -330,6 +432,65 @@ impl SegmentReader {
     }
 }
 
+#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord)]
+/// FieldMetadata
+pub struct FieldMetadata {
+    /// The field name
+    // Notice: Don't reorder the declaration of 1.field_name 2.typ, as it is used for ordering by
+    // field_name then typ.
+    pub field_name: String,
+    /// The field type
+    // Notice: Don't reorder the declaration of 1.field_name 2.typ, as it is used for ordering by
+    // field_name then typ.
+    pub typ: Type,
+    /// Is the field indexed for search
+    pub indexed: bool,
+    /// Is the field stored in the doc store
+    pub stored: bool,
+    /// Is the field stored in the columnar storage
+    pub fast: bool,
+}
+impl BitOrAssign for FieldMetadata {
+    fn bitor_assign(&mut self, rhs: Self) {
+        assert!(self.field_name == rhs.field_name);
+        assert!(self.typ == rhs.typ);
+        self.indexed |= rhs.indexed;
+        self.stored |= rhs.stored;
+        self.fast |= rhs.fast;
+    }
+}
+
+// Maybe too slow for the high cardinality case
+fn is_field_stored(field_name: &str, schema: &Schema) -> bool {
+    schema
+        .find_field(field_name)
+        .map(|(field, _path)| schema.get_field_entry(field).is_stored())
+        .unwrap_or(false)
+}
+
+/// Helper to merge the field metadata from multiple segments.
+pub fn merge_field_meta_data(
+    field_metadatas: Vec<Vec<FieldMetadata>>,
+    schema: &Schema,
+) -> Vec<FieldMetadata> {
+    let mut merged_field_metadata = Vec::new();
+    for (_key, mut group) in &field_metadatas
+        .into_iter()
+        .kmerge_by(|left, right| left < right)
+        // TODO: Remove allocation
+        .group_by(|el| (el.field_name.to_string(), el.typ))
+    {
+        let mut merged: FieldMetadata = group.next().unwrap();
+        for el in group {
+            merged |= el;
+        }
+        // Currently is_field_stored is maybe too slow for the high cardinality case
+        merged.stored = is_field_stored(&merged.field_name, schema);
+        merged_field_metadata.push(merged);
+    }
+    merged_field_metadata
+}
+
 fn intersect_alive_bitset(
     left_opt: Option<AliveBitSet>,
     right_opt: Option<AliveBitSet>,
@@ -353,9 +514,127 @@ impl fmt::Debug for SegmentReader {
 
 #[cfg(test)]
 mod test {
+    use super::*;
     use crate::core::Index;
-    use crate::schema::{Schema, Term, STORED, TEXT};
-    use crate::DocId;
+    use crate::schema::{Schema, SchemaBuilder, Term, STORED, TEXT};
+    use crate::{DocId, FieldMetadata, IndexWriter};
+
+    #[test]
+    fn test_merge_field_meta_data_same() {
+        let schema = SchemaBuilder::new().build();
+        let field_metadata1 = FieldMetadata {
+            field_name: "a".to_string(),
+            typ: crate::schema::Type::Str,
+            indexed: true,
+            stored: false,
+            fast: true,
+        };
+        let field_metadata2 = FieldMetadata {
+            field_name: "a".to_string(),
+            typ: crate::schema::Type::Str,
+            indexed: true,
+            stored: false,
+            fast: true,
+        };
+        let res = merge_field_meta_data(
+            vec![vec![field_metadata1.clone()], vec![field_metadata2]],
+            &schema,
+        );
+        assert_eq!(res, vec![field_metadata1]);
+    }
+    #[test]
+    fn test_merge_field_meta_data_different() {
+        let schema = SchemaBuilder::new().build();
+        let field_metadata1 = FieldMetadata {
+            field_name: "a".to_string(),
+            typ: crate::schema::Type::Str,
+            indexed: false,
+            stored: false,
+            fast: true,
+        };
+        let field_metadata2 = FieldMetadata {
+            field_name: "b".to_string(),
+            typ: crate::schema::Type::Str,
+            indexed: false,
+            stored: false,
+            fast: true,
+        };
+        let field_metadata3 = FieldMetadata {
+            field_name: "a".to_string(),
+            typ: crate::schema::Type::Str,
+            indexed: true,
+            stored: false,
+            fast: false,
+        };
+        let res = merge_field_meta_data(
+            vec![
+                vec![field_metadata1.clone(), field_metadata2.clone()],
+                vec![field_metadata3],
+            ],
+            &schema,
+        );
+        let field_metadata_expected1 = FieldMetadata {
+            field_name: "a".to_string(),
+            typ: crate::schema::Type::Str,
+            indexed: true,
+            stored: false,
+            fast: true,
+        };
+        assert_eq!(res, vec![field_metadata_expected1, field_metadata2.clone()]);
+    }
+    #[test]
+    fn test_merge_field_meta_data_merge() {
+        use pretty_assertions::assert_eq;
+        let get_meta_data = |name: &str, typ: Type| FieldMetadata {
+            field_name: name.to_string(),
+            typ,
+            indexed: false,
+            stored: false,
+            fast: true,
+        };
+        let schema = SchemaBuilder::new().build();
+        let mut metas = vec![get_meta_data("d", Type::Str), get_meta_data("e", Type::U64)];
+        metas.sort();
+        let res = merge_field_meta_data(vec![vec![get_meta_data("e", Type::Str)], metas], &schema);
+        assert_eq!(
+            res,
+            vec![
+                get_meta_data("d", Type::Str),
+                get_meta_data("e", Type::Str),
+                get_meta_data("e", Type::U64),
+            ]
+        );
+    }
+    #[test]
+    fn test_merge_field_meta_data_bitxor() {
+        let field_metadata1 = FieldMetadata {
+            field_name: "a".to_string(),
+            typ: crate::schema::Type::Str,
+            indexed: false,
+            stored: false,
+            fast: true,
+        };
+        let field_metadata2 = FieldMetadata {
+            field_name: "a".to_string(),
+            typ: crate::schema::Type::Str,
+            indexed: true,
+            stored: false,
+            fast: false,
+        };
+        let field_metadata_expected = FieldMetadata {
+            field_name: "a".to_string(),
+            typ: crate::schema::Type::Str,
+            indexed: true,
+            stored: false,
+            fast: true,
+        };
+        let mut res1 = field_metadata1.clone();
+        res1 |= field_metadata2.clone();
+        let mut res2 = field_metadata2.clone();
+        res2 |= field_metadata1;
+        assert_eq!(res1, field_metadata_expected);
+        assert_eq!(res2, field_metadata_expected);
+    }
 
     #[test]
     fn test_num_alive() -> crate::Result<()> {
@@ -366,7 +645,7 @@ mod test {
         let name = schema.get_field("name").unwrap();
 
         {
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             index_writer.add_document(doc!(name => "tantivy"))?;
             index_writer.add_document(doc!(name => "horse"))?;
             index_writer.add_document(doc!(name => "jockey"))?;
@@ -392,7 +671,7 @@ mod test {
         let name = schema.get_field("name").unwrap();
 
         {
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             index_writer.add_document(doc!(name => "tantivy"))?;
             index_writer.add_document(doc!(name => "horse"))?;
             index_writer.add_document(doc!(name => "jockey"))?;
@@ -402,7 +681,7 @@ mod test {
         }
 
         {
-            let mut index_writer2 = index.writer(50_000_000)?;
+            let mut index_writer2: IndexWriter = index.writer(50_000_000)?;
             index_writer2.delete_term(Term::from_field_text(name, "horse"));
             index_writer2.delete_term(Term::from_field_text(name, "cap"));
 

src/core/single_segment_index_writer.rs

@@ -1,16 +1,20 @@
+use std::marker::PhantomData;
+
 use crate::indexer::operation::AddOperation;
 use crate::indexer::segment_updater::save_metas;
 use crate::indexer::SegmentWriter;
-use crate::{Directory, Document, Index, IndexMeta, Opstamp, Segment};
+use crate::schema::document::Document;
+use crate::{Directory, Index, IndexMeta, Opstamp, Segment, TantivyDocument};
 
 #[doc(hidden)]
-pub struct SingleSegmentIndexWriter {
+pub struct SingleSegmentIndexWriter<D: Document = TantivyDocument> {
     segment_writer: SegmentWriter,
     segment: Segment,
     opstamp: Opstamp,
+    _phantom: PhantomData<D>,
 }
 
-impl SingleSegmentIndexWriter {
+impl<D: Document> SingleSegmentIndexWriter<D> {
     pub fn new(index: Index, mem_budget: usize) -> crate::Result<Self> {
         let segment = index.new_segment();
         let segment_writer = SegmentWriter::for_segment(mem_budget, segment.clone())?;
@@ -18,6 +22,7 @@ impl SingleSegmentIndexWriter {
             segment_writer,
             segment,
             opstamp: 0,
+            _phantom: PhantomData,
         })
     }
 
@@ -25,7 +30,7 @@ impl SingleSegmentIndexWriter {
         self.segment_writer.mem_usage()
     }
 
-    pub fn add_document(&mut self, document: Document) -> crate::Result<()> {
+    pub fn add_document(&mut self, document: D) -> crate::Result<()> {
         let opstamp = self.opstamp;
         self.opstamp += 1;
         self.segment_writer

src/core/tests.rs

@@ -1,12 +1,13 @@
 use crate::collector::Count;
 use crate::directory::{RamDirectory, WatchCallback};
-use crate::indexer::NoMergePolicy;
+use crate::indexer::{LogMergePolicy, NoMergePolicy};
+use crate::json_utils::JsonTermWriter;
 use crate::query::TermQuery;
-use crate::schema::{Field, IndexRecordOption, Schema, INDEXED, STRING, TEXT};
+use crate::schema::{Field, IndexRecordOption, Schema, Type, INDEXED, STRING, TEXT};
 use crate::tokenizer::TokenizerManager;
 use crate::{
-    Directory, Document, Index, IndexBuilder, IndexReader, IndexSettings, ReloadPolicy, SegmentId,
-    Term,
+    Directory, DocSet, Index, IndexBuilder, IndexReader, IndexSettings, IndexWriter, Postings,
+    ReloadPolicy, SegmentId, TantivyDocument, Term,
 };
 
 #[test]
@@ -121,7 +122,7 @@ fn test_index_on_commit_reload_policy() -> crate::Result<()> {
     let index = Index::create_in_ram(schema);
     let reader = index
         .reader_builder()
-        .reload_policy(ReloadPolicy::OnCommit)
+        .reload_policy(ReloadPolicy::OnCommitWithDelay)
         .try_into()
         .unwrap();
     assert_eq!(reader.searcher().num_docs(), 0);
@@ -147,7 +148,7 @@ mod mmap_specific {
         let index = Index::create_in_dir(tempdir_path, schema).unwrap();
         let reader = index
             .reader_builder()
-            .reload_policy(ReloadPolicy::OnCommit)
+            .reload_policy(ReloadPolicy::OnCommitWithDelay)
             .try_into()
             .unwrap();
         assert_eq!(reader.searcher().num_docs(), 0);
@@ -159,7 +160,7 @@ mod mmap_specific {
         let schema = throw_away_schema();
         let field = schema.get_field("num_likes").unwrap();
         let mut index = Index::create_from_tempdir(schema)?;
-        let mut writer = index.writer_for_tests()?;
+        let mut writer: IndexWriter = index.writer_for_tests()?;
         writer.commit()?;
         let reader = index
             .reader_builder()
@@ -189,7 +190,7 @@ mod mmap_specific {
         let read_index = Index::open_in_dir(&tempdir_path).unwrap();
         let reader = read_index
             .reader_builder()
-            .reload_policy(ReloadPolicy::OnCommit)
+            .reload_policy(ReloadPolicy::OnCommitWithDelay)
             .try_into()
             .unwrap();
         assert_eq!(reader.searcher().num_docs(), 0);
@@ -208,7 +209,7 @@ fn test_index_on_commit_reload_policy_aux(
         .watch(WatchCallback::new(move || {
             let _ = sender.send(());
         }));
-    let mut writer = index.writer_for_tests()?;
+    let mut writer: IndexWriter = index.writer_for_tests()?;
     assert_eq!(reader.searcher().num_docs(), 0);
     writer.add_document(doc!(field=>1u64))?;
     writer.commit().unwrap();
@@ -242,7 +243,7 @@ fn garbage_collect_works_as_intended() -> crate::Result<()> {
     let field = schema.get_field("num_likes").unwrap();
     let index = Index::create(directory.clone(), schema, IndexSettings::default())?;
 
-    let mut writer = index.writer_with_num_threads(1, 32_000_000).unwrap();
+    let mut writer: IndexWriter = index.writer_with_num_threads(1, 32_000_000).unwrap();
     for _seg in 0..8 {
         for i in 0u64..1_000u64 {
             writer.add_document(doc!(field => i))?;
@@ -306,7 +307,7 @@ fn test_merging_segment_update_docfreq() {
     let id_field = schema_builder.add_text_field("id", STRING);
     let schema = schema_builder.build();
     let index = Index::create_in_ram(schema);
-    let mut writer = index.writer_for_tests().unwrap();
+    let mut writer: IndexWriter = index.writer_for_tests().unwrap();
     writer.set_merge_policy(Box::new(NoMergePolicy));
     for _ in 0..5 {
         writer.add_document(doc!(text_field=>"hello")).unwrap();
@@ -317,13 +318,13 @@ fn test_merging_segment_update_docfreq() {
     writer
         .add_document(doc!(text_field=>"hello", id_field=>"TO_BE_DELETED"))
         .unwrap();
-    writer.add_document(Document::default()).unwrap();
+    writer.add_document(TantivyDocument::default()).unwrap();
     writer.commit().unwrap();
     for _ in 0..7 {
         writer.add_document(doc!(text_field=>"hello")).unwrap();
     }
-    writer.add_document(Document::default()).unwrap();
-    writer.add_document(Document::default()).unwrap();
+    writer.add_document(TantivyDocument::default()).unwrap();
+    writer.add_document(TantivyDocument::default()).unwrap();
     writer.delete_term(Term::from_field_text(id_field, "TO_BE_DELETED"));
     writer.commit().unwrap();
 
@@ -344,3 +345,132 @@ fn test_merging_segment_update_docfreq() {
     let term_info = inv_index.get_term_info(&term).unwrap().unwrap();
     assert_eq!(term_info.doc_freq, 12);
 }
+
+// motivated by https://github.com/quickwit-oss/quickwit/issues/4130
+#[test]
+fn test_positions_merge_bug_non_text_json_vint() {
+    let mut schema_builder = Schema::builder();
+    let field = schema_builder.add_json_field("dynamic", TEXT);
+    let schema = schema_builder.build();
+    let index = Index::create_in_ram(schema.clone());
+    let mut writer: IndexWriter = index.writer_for_tests().unwrap();
+    let mut merge_policy = LogMergePolicy::default();
+    merge_policy.set_min_num_segments(2);
+    writer.set_merge_policy(Box::new(merge_policy));
+    // Here a string would work.
+    let doc_json = r#"{"tenant_id":75}"#;
+    let vals = serde_json::from_str(doc_json).unwrap();
+    let mut doc = TantivyDocument::default();
+    doc.add_object(field, vals);
+    writer.add_document(doc.clone()).unwrap();
+    writer.commit().unwrap();
+    writer.add_document(doc.clone()).unwrap();
+    writer.commit().unwrap();
+    writer.wait_merging_threads().unwrap();
+    let reader = index.reader().unwrap();
+    assert_eq!(reader.searcher().segment_readers().len(), 1);
+}
+
+// Same as above but with bitpacked blocks
+#[test]
+fn test_positions_merge_bug_non_text_json_bitpacked_block() {
+    let mut schema_builder = Schema::builder();
+    let field = schema_builder.add_json_field("dynamic", TEXT);
+    let schema = schema_builder.build();
+    let index = Index::create_in_ram(schema.clone());
+    let mut writer: IndexWriter = index.writer_for_tests().unwrap();
+    let mut merge_policy = LogMergePolicy::default();
+    merge_policy.set_min_num_segments(2);
+    writer.set_merge_policy(Box::new(merge_policy));
+    // Here a string would work.
+    let doc_json = r#"{"tenant_id":75}"#;
+    let vals = serde_json::from_str(doc_json).unwrap();
+    let mut doc = TantivyDocument::default();
+    doc.add_object(field, vals);
+    for _ in 0..128 {
+        writer.add_document(doc.clone()).unwrap();
+    }
+    writer.commit().unwrap();
+    writer.add_document(doc.clone()).unwrap();
+    writer.commit().unwrap();
+    writer.wait_merging_threads().unwrap();
+    let reader = index.reader().unwrap();
+    assert_eq!(reader.searcher().segment_readers().len(), 1);
+}
+
+#[test]
+fn test_non_text_json_term_freq() {
+    let mut schema_builder = Schema::builder();
+    let field = schema_builder.add_json_field("dynamic", TEXT);
+    let schema = schema_builder.build();
+    let index = Index::create_in_ram(schema.clone());
+    let mut writer: IndexWriter = index.writer_for_tests().unwrap();
+    // Here a string would work.
+    let doc_json = r#"{"tenant_id":75}"#;
+    let vals = serde_json::from_str(doc_json).unwrap();
+    let mut doc = TantivyDocument::default();
+    doc.add_object(field, vals);
+    writer.add_document(doc.clone()).unwrap();
+    writer.commit().unwrap();
+    let reader = index.reader().unwrap();
+    assert_eq!(reader.searcher().segment_readers().len(), 1);
+    let searcher = reader.searcher();
+    let segment_reader = searcher.segment_reader(0u32);
+    let inv_idx = segment_reader.inverted_index(field).unwrap();
+    let mut term = Term::with_type_and_field(Type::Json, field);
+    let mut json_term_writer = JsonTermWriter::wrap(&mut term, false);
+    json_term_writer.push_path_segment("tenant_id");
+    json_term_writer.close_path_and_set_type(Type::U64);
+    json_term_writer.set_fast_value(75u64);
+    let postings = inv_idx
+        .read_postings(
+            &json_term_writer.term(),
+            IndexRecordOption::WithFreqsAndPositions,
+        )
+        .unwrap()
+        .unwrap();
+    assert_eq!(postings.doc(), 0);
+    assert_eq!(postings.term_freq(), 1u32);
+}
+
+#[test]
+fn test_non_text_json_term_freq_bitpacked() {
+    let mut schema_builder = Schema::builder();
+    let field = schema_builder.add_json_field("dynamic", TEXT);
+    let schema = schema_builder.build();
+    let index = Index::create_in_ram(schema.clone());
+    let mut writer: IndexWriter = index.writer_for_tests().unwrap();
+    // Here a string would work.
+    let doc_json = r#"{"tenant_id":75}"#;
+    let vals = serde_json::from_str(doc_json).unwrap();
+    let mut doc = TantivyDocument::default();
+    doc.add_object(field, vals);
+    let num_docs = 132;
+    for _ in 0..num_docs {
+        writer.add_document(doc.clone()).unwrap();
+    }
+    writer.commit().unwrap();
+    let reader = index.reader().unwrap();
+    assert_eq!(reader.searcher().segment_readers().len(), 1);
+    let searcher = reader.searcher();
+    let segment_reader = searcher.segment_reader(0u32);
+    let inv_idx = segment_reader.inverted_index(field).unwrap();
+    let mut term = Term::with_type_and_field(Type::Json, field);
+    let mut json_term_writer = JsonTermWriter::wrap(&mut term, false);
+    json_term_writer.push_path_segment("tenant_id");
+    json_term_writer.close_path_and_set_type(Type::U64);
+    json_term_writer.set_fast_value(75u64);
+    let mut postings = inv_idx
+        .read_postings(
+            &json_term_writer.term(),
+            IndexRecordOption::WithFreqsAndPositions,
+        )
+        .unwrap()
+        .unwrap();
+    assert_eq!(postings.doc(), 0);
+    assert_eq!(postings.term_freq(), 1u32);
+    for i in 1..num_docs {
+        assert_eq!(postings.advance(), i);
+        assert_eq!(postings.term_freq(), 1u32);
+    }
+}

src/directory/directory.rs

@@ -222,8 +222,8 @@ pub trait Directory: DirectoryClone + fmt::Debug + Send + Sync + 'static {
     /// registered (and whose [`WatchHandle`] is still alive) are triggered.
     ///
     /// Internally, tantivy only uses this API to detect new commits to implement the
-    /// `OnCommit` `ReloadPolicy`. Not implementing watch in a `Directory` only prevents the
-    /// `OnCommit` `ReloadPolicy` to work properly.
+    /// `OnCommitWithDelay` `ReloadPolicy`. Not implementing watch in a `Directory` only prevents
+    /// the `OnCommitWithDelay` `ReloadPolicy` to work properly.
     fn watch(&self, watch_callback: WatchCallback) -> crate::Result<WatchHandle>;
 }
 

src/directory/footer.rs

@@ -7,7 +7,7 @@ use serde::{Deserialize, Serialize};
 
 use crate::directory::error::Incompatibility;
 use crate::directory::{AntiCallToken, FileSlice, TerminatingWrite};
-use crate::{Version, INDEX_FORMAT_VERSION};
+use crate::{Version, INDEX_FORMAT_OLDEST_SUPPORTED_VERSION, INDEX_FORMAT_VERSION};
 
 const FOOTER_MAX_LEN: u32 = 50_000;
 
@@ -102,10 +102,11 @@ impl Footer {
     /// Confirms that the index will be read correctly by this version of tantivy
     /// Has to be called after `extract_footer` to make sure it's not accessing uninitialised memory
     pub fn is_compatible(&self) -> Result<(), Incompatibility> {
+        const SUPPORTED_INDEX_FORMAT_VERSION_RANGE: std::ops::RangeInclusive<u32> =
+            INDEX_FORMAT_OLDEST_SUPPORTED_VERSION..=INDEX_FORMAT_VERSION;
+
         let library_version = crate::version();
-        if self.version.index_format_version < 4
-            || self.version.index_format_version > INDEX_FORMAT_VERSION
-        {
+        if !SUPPORTED_INDEX_FORMAT_VERSION_RANGE.contains(&self.version.index_format_version) {
             return Err(Incompatibility::IndexMismatch {
                 library_version: library_version.clone(),
                 index_version: self.version.clone(),

src/directory/mmap_directory.rs

@@ -1,13 +1,15 @@
 use std::collections::HashMap;
 use std::fmt;
 use std::fs::{self, File, OpenOptions};
-use std::io::{self, BufWriter, Read, Seek, Write};
+use std::io::{self, BufWriter, Read, Write};
 use std::ops::Deref;
 use std::path::{Path, PathBuf};
 use std::sync::{Arc, RwLock, Weak};
 
 use common::StableDeref;
 use fs4::FileExt;
+#[cfg(all(feature = "mmap", unix))]
+pub use memmap2::Advice;
 use memmap2::Mmap;
 use serde::{Deserialize, Serialize};
 use tempfile::TempDir;
@@ -21,8 +23,6 @@ use crate::directory::{
     AntiCallToken, Directory, DirectoryLock, FileHandle, Lock, OwnedBytes, TerminatingWrite,
     WatchCallback, WatchHandle, WritePtr,
 };
-#[cfg(unix)]
-use crate::Advice;
 
 pub type ArcBytes = Arc<dyn Deref<Target = [u8]> + Send + Sync + 'static>;
 pub type WeakArcBytes = Weak<dyn Deref<Target = [u8]> + Send + Sync + 'static>;
@@ -328,12 +328,6 @@ impl Write for SafeFileWriter {
     }
 }
 
-impl Seek for SafeFileWriter {
-    fn seek(&mut self, pos: io::SeekFrom) -> io::Result<u64> {
-        self.0.seek(pos)
-    }
-}
-
 impl TerminatingWrite for SafeFileWriter {
     fn terminate_ref(&mut self, _: AntiCallToken) -> io::Result<()> {
         self.0.flush()?;
@@ -539,7 +533,7 @@ mod tests {
     use super::*;
     use crate::indexer::LogMergePolicy;
     use crate::schema::{Schema, SchemaBuilder, TEXT};
-    use crate::{Index, IndexSettings, ReloadPolicy};
+    use crate::{Index, IndexSettings, IndexWriter, ReloadPolicy};
 
     #[test]
     fn test_open_non_existent_path() {
@@ -651,7 +645,7 @@ mod tests {
             let index =
                 Index::create(mmap_directory.clone(), schema, IndexSettings::default()).unwrap();
 
-            let mut index_writer = index.writer_for_tests().unwrap();
+            let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
             let mut log_merge_policy = LogMergePolicy::default();
             log_merge_policy.set_min_num_segments(3);
             index_writer.set_merge_policy(Box::new(log_merge_policy));

src/directory/mod.rs

@@ -42,6 +42,9 @@ pub struct GarbageCollectionResult {
     pub failed_to_delete_files: Vec<PathBuf>,
 }
 
+#[cfg(all(feature = "mmap", unix))]
+pub use memmap2::Advice;
+
 pub use self::managed_directory::ManagedDirectory;
 #[cfg(feature = "mmap")]
 pub use self::mmap_directory::MmapDirectory;

src/directory/ram_directory.rs

@@ -1,5 +1,5 @@
 use std::collections::HashMap;
-use std::io::{self, BufWriter, Cursor, Seek, SeekFrom, Write};
+use std::io::{self, BufWriter, Cursor, Write};
 use std::path::{Path, PathBuf};
 use std::sync::{Arc, RwLock};
 use std::{fmt, result};
@@ -48,12 +48,6 @@ impl Drop for VecWriter {
     }
 }
 
-impl Seek for VecWriter {
-    fn seek(&mut self, pos: SeekFrom) -> io::Result<u64> {
-        self.data.seek(pos)
-    }
-}
-
 impl Write for VecWriter {
     fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
         self.is_flushed = false;

src/docset.rs

@@ -17,7 +17,7 @@ pub trait DocSet: Send {
     ///
     /// The DocId of the next element is returned.
     /// In other words we should always have :
-    /// ```ignore
+    /// ```compile_fail
     /// let doc = docset.advance();
     /// assert_eq!(doc, docset.doc());
     /// ```

src/error.rs

@@ -11,6 +11,7 @@ use crate::directory::error::{
     Incompatibility, LockError, OpenDirectoryError, OpenReadError, OpenWriteError,
 };
 use crate::fastfield::FastFieldNotAvailableError;
+use crate::schema::document::DeserializeError;
 use crate::{query, schema};
 
 /// Represents a `DataCorruption` error.
@@ -106,6 +107,9 @@ pub enum TantivyError {
     /// e.g. a datastructure is incorrectly inititalized.
     #[error("Internal error: '{0}'")]
     InternalError(String),
+    #[error("Deserialize error: {0}")]
+    /// An error occurred while attempting to deserialize a document.
+    DeserializeError(DeserializeError),
 }
 
 impl From<io::Error> for TantivyError {
@@ -176,3 +180,9 @@ impl From<rayon::ThreadPoolBuildError> for TantivyError {
         TantivyError::SystemError(error.to_string())
     }
 }
+
+impl From<DeserializeError> for TantivyError {
+    fn from(error: DeserializeError) -> TantivyError {
+        TantivyError::DeserializeError(error)
+    }
+}

src/fastfield/facet_reader.rs

@@ -62,8 +62,9 @@ impl FacetReader {
 
 #[cfg(test)]
 mod tests {
-    use crate::schema::{Facet, FacetOptions, SchemaBuilder, Value, STORED};
-    use crate::{DocAddress, Document, Index};
+    use crate::schema::document::Value;
+    use crate::schema::{Facet, FacetOptions, SchemaBuilder, STORED};
+    use crate::{DocAddress, Index, IndexWriter, TantivyDocument};
 
     #[test]
     fn test_facet_only_indexed() {
@@ -71,7 +72,7 @@ mod tests {
         let facet_field = schema_builder.add_facet_field("facet", FacetOptions::default());
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(facet_field=>Facet::from_text("/a/b").unwrap()))
             .unwrap();
@@ -85,8 +86,10 @@ mod tests {
         let mut facet = Facet::default();
         facet_reader.facet_from_ord(0, &mut facet).unwrap();
         assert_eq!(facet.to_path_string(), "/a/b");
-        let doc = searcher.doc(DocAddress::new(0u32, 0u32)).unwrap();
-        let value = doc.get_first(facet_field).and_then(Value::as_facet);
+        let doc = searcher
+            .doc::<TantivyDocument>(DocAddress::new(0u32, 0u32))
+            .unwrap();
+        let value = doc.get_first(facet_field).and_then(|v| v.as_facet());
         assert_eq!(value, None);
     }
 
@@ -96,7 +99,7 @@ mod tests {
         let facet_field = schema_builder.add_facet_field("facet", FacetOptions::default());
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(facet_field=>Facet::from_text("/parent/child1").unwrap()))
             .unwrap();
@@ -142,8 +145,8 @@ mod tests {
         let mut facet_ords = Vec::new();
         facet_ords.extend(facet_reader.facet_ords(0u32));
         assert_eq!(&facet_ords, &[0u64]);
-        let doc = searcher.doc(DocAddress::new(0u32, 0u32))?;
-        let value: Option<&Facet> = doc.get_first(facet_field).and_then(Value::as_facet);
+        let doc = searcher.doc::<TantivyDocument>(DocAddress::new(0u32, 0u32))?;
+        let value: Option<&Facet> = doc.get_first(facet_field).and_then(|v| v.as_facet());
         assert_eq!(value, Facet::from_text("/a/b").ok().as_ref());
         Ok(())
     }
@@ -156,7 +159,7 @@ mod tests {
         let index = Index::create_in_ram(schema);
         let mut index_writer = index.writer_for_tests()?;
         index_writer.add_document(doc!(facet_field=>Facet::from_text("/a/b").unwrap()))?;
-        index_writer.add_document(Document::default())?;
+        index_writer.add_document(TantivyDocument::default())?;
         index_writer.commit()?;
         let searcher = index.reader()?.searcher();
         let facet_reader = searcher.segment_reader(0u32).facet_reader("facet").unwrap();
@@ -176,8 +179,8 @@ mod tests {
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
         let mut index_writer = index.writer_for_tests()?;
-        index_writer.add_document(Document::default())?;
-        index_writer.add_document(Document::default())?;
+        index_writer.add_document(TantivyDocument::default())?;
+        index_writer.add_document(TantivyDocument::default())?;
         index_writer.commit()?;
         let searcher = index.reader()?.searcher();
         let facet_reader = searcher.segment_reader(0u32).facet_reader("facet").unwrap();

src/fastfield/mod.rs

@@ -90,12 +90,12 @@ mod tests {
     use crate::directory::{Directory, RamDirectory, WritePtr};
     use crate::merge_policy::NoMergePolicy;
     use crate::schema::{
-        Document, Facet, FacetOptions, Field, JsonObjectOptions, Schema, SchemaBuilder,
+        Facet, FacetOptions, Field, JsonObjectOptions, Schema, SchemaBuilder, TantivyDocument,
         TextOptions, FAST, INDEXED, STORED, STRING, TEXT,
     };
     use crate::time::OffsetDateTime;
     use crate::tokenizer::{LowerCaser, RawTokenizer, TextAnalyzer, TokenizerManager};
-    use crate::{DateOptions, DateTimePrecision, Index, SegmentId, SegmentReader};
+    use crate::{DateOptions, DateTimePrecision, Index, IndexWriter, SegmentId, SegmentReader};
 
     pub static SCHEMA: Lazy<Schema> = Lazy::new(|| {
         let mut schema_builder = Schema::builder();
@@ -131,7 +131,7 @@ mod tests {
         }
         let file = directory.open_read(path).unwrap();
 
-        assert_eq!(file.len(), 93);
+        assert_eq!(file.len(), 80);
         let fast_field_readers = FastFieldReaders::open(file, SCHEMA.clone()).unwrap();
         let column = fast_field_readers
             .u64("field")
@@ -181,7 +181,7 @@ mod tests {
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 121);
+        assert_eq!(file.len(), 108);
         let fast_field_readers = FastFieldReaders::open(file, SCHEMA.clone()).unwrap();
         let col = fast_field_readers
             .u64("field")
@@ -214,7 +214,7 @@ mod tests {
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 94);
+        assert_eq!(file.len(), 81);
         let fast_field_readers = FastFieldReaders::open(file, SCHEMA.clone()).unwrap();
         let fast_field_reader = fast_field_readers
             .u64("field")
@@ -246,7 +246,7 @@ mod tests {
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 4489);
+        assert_eq!(file.len(), 4476);
         {
             let fast_field_readers = FastFieldReaders::open(file, SCHEMA.clone()).unwrap();
             let col = fast_field_readers
@@ -271,7 +271,7 @@ mod tests {
             let mut write: WritePtr = directory.open_write(Path::new("test")).unwrap();
             let mut fast_field_writers = FastFieldsWriter::from_schema(&schema).unwrap();
             for i in -100i64..10_000i64 {
-                let mut doc = Document::default();
+                let mut doc = TantivyDocument::default();
                 doc.add_i64(i64_field, i);
                 fast_field_writers.add_document(&doc).unwrap();
             }
@@ -279,7 +279,7 @@ mod tests {
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 265);
+        assert_eq!(file.len(), 252);
 
         {
             let fast_field_readers = FastFieldReaders::open(file, schema).unwrap();
@@ -312,7 +312,7 @@ mod tests {
         {
             let mut write: WritePtr = directory.open_write(Path::new("test")).unwrap();
             let mut fast_field_writers = FastFieldsWriter::from_schema(&schema).unwrap();
-            let doc = Document::default();
+            let doc = TantivyDocument::default();
             fast_field_writers.add_document(&doc).unwrap();
             fast_field_writers.serialize(&mut write, None).unwrap();
             write.terminate().unwrap();
@@ -345,7 +345,7 @@ mod tests {
         {
             let mut write: WritePtr = directory.open_write(Path::new("test")).unwrap();
             let mut fast_field_writers = FastFieldsWriter::from_schema(&schema).unwrap();
-            let doc = Document::default();
+            let doc = TantivyDocument::default();
             fast_field_writers.add_document(&doc).unwrap();
             fast_field_writers.serialize(&mut write, None).unwrap();
             write.terminate().unwrap();
@@ -416,7 +416,7 @@ mod tests {
         let date_field = schema_builder.add_date_field("date", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer.set_merge_policy(Box::new(NoMergePolicy));
         index_writer
             .add_document(doc!(date_field => DateTime::from_utc(OffsetDateTime::now_utc())))
@@ -452,7 +452,7 @@ mod tests {
 
         {
             // first segment
-            let mut index_writer = index.writer_for_tests().unwrap();
+            let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
             index_writer.set_merge_policy(Box::new(NoMergePolicy));
             index_writer
                 .add_document(doc!(
@@ -506,7 +506,7 @@ mod tests {
 
         {
             // second segment
-            let mut index_writer = index.writer_for_tests().unwrap();
+            let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
 
             index_writer
                 .add_document(doc!(
@@ -537,7 +537,7 @@ mod tests {
         // Merging the segments
         {
             let segment_ids = index.searchable_segment_ids().unwrap();
-            let mut index_writer = index.writer_for_tests().unwrap();
+            let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
             index_writer.merge(&segment_ids).wait().unwrap();
             index_writer.wait_merging_threads().unwrap();
         }
@@ -662,7 +662,7 @@ mod tests {
         // Merging the segments
         {
             let segment_ids = index.searchable_segment_ids()?;
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             index_writer.merge(&segment_ids).wait()?;
             index_writer.wait_merging_threads()?;
         }
@@ -773,7 +773,7 @@ mod tests {
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 102);
+        assert_eq!(file.len(), 84);
         let fast_field_readers = FastFieldReaders::open(file, schema).unwrap();
         let bool_col = fast_field_readers.bool("field_bool").unwrap();
         assert_eq!(bool_col.first(0), Some(true));
@@ -805,7 +805,7 @@ mod tests {
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 114);
+        assert_eq!(file.len(), 96);
         let readers = FastFieldReaders::open(file, schema).unwrap();
         let bool_col = readers.bool("field_bool").unwrap();
         for i in 0..25 {
@@ -824,13 +824,13 @@ mod tests {
         {
             let mut write: WritePtr = directory.open_write(path).unwrap();
             let mut fast_field_writers = FastFieldsWriter::from_schema(&schema).unwrap();
-            let doc = Document::default();
+            let doc = TantivyDocument::default();
             fast_field_writers.add_document(&doc).unwrap();
             fast_field_writers.serialize(&mut write, None).unwrap();
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 104);
+        assert_eq!(file.len(), 86);
         let fastfield_readers = FastFieldReaders::open(file, schema).unwrap();
         let col = fastfield_readers.bool("field_bool").unwrap();
         assert_eq!(col.first(0), None);
@@ -846,7 +846,7 @@ mod tests {
         assert_eq!(col.get_val(0), true);
     }
 
-    fn get_index(docs: &[crate::Document], schema: &Schema) -> crate::Result<RamDirectory> {
+    fn get_index(docs: &[crate::TantivyDocument], schema: &Schema) -> crate::Result<RamDirectory> {
         let directory: RamDirectory = RamDirectory::create();
         {
             let mut write: WritePtr = directory.open_write(Path::new("test")).unwrap();
@@ -888,7 +888,7 @@ mod tests {
         let field = schema_builder.add_date_field("field", date_options);
         let schema = schema_builder.build();
 
-        let docs: Vec<Document> = times.iter().map(|time| doc!(field=>*time)).collect();
+        let docs: Vec<TantivyDocument> = times.iter().map(|time| doc!(field=>*time)).collect();
 
         let directory = get_index(&docs[..], &schema).unwrap();
         let path = Path::new("test");
@@ -962,11 +962,15 @@ mod tests {
         let ip_field = schema_builder.add_u64_field("ip", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         let ip_addr = Ipv6Addr::new(1, 2, 3, 4, 5, 1, 2, 3);
-        index_writer.add_document(Document::default()).unwrap();
+        index_writer
+            .add_document(TantivyDocument::default())
+            .unwrap();
         index_writer.add_document(doc!(ip_field=>ip_addr)).unwrap();
-        index_writer.add_document(Document::default()).unwrap();
+        index_writer
+            .add_document(TantivyDocument::default())
+            .unwrap();
         index_writer.commit().unwrap();
         let searcher = index.reader().unwrap().searcher();
         let fastfields = searcher.segment_reader(0u32).fast_fields();
@@ -1086,7 +1090,7 @@ mod tests {
         let json = schema_builder.add_json_field("json", json_option);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(json => json!({"attr.age": 32})))
             .unwrap();
@@ -1112,7 +1116,7 @@ mod tests {
         let json = schema_builder.add_json_field("json", json_option);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(json => json!({"age": 32})))
             .unwrap();
@@ -1139,7 +1143,7 @@ mod tests {
         let json = schema_builder.add_json_field("json", json_option);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(json => json!({"attr.age": 32})))
             .unwrap();
@@ -1162,7 +1166,7 @@ mod tests {
         let field_with_dot = schema_builder.add_i64_field("field.with.dot", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(field_with_dot => 32i64))
             .unwrap();
@@ -1184,7 +1188,7 @@ mod tests {
         let shadowing_json_field = schema_builder.add_json_field("jsonfield.attr", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(json_field=> json!({"attr": {"age": 32}}), shadowing_json_field=>json!({"age": 33})))
             .unwrap();
@@ -1215,7 +1219,7 @@ mod tests {
 
         let mut index = Index::create_in_ram(schema);
         index.set_fast_field_tokenizers(ff_tokenizer_manager);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(text_field => "Test1 test2"))
             .unwrap();
@@ -1244,7 +1248,7 @@ mod tests {
         let log_field = schema_builder.add_text_field("log_level", text_fieldtype);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(log_field => "info"))
             .unwrap();
@@ -1277,18 +1281,25 @@ mod tests {
         let shadowing_json_field = schema_builder.add_json_field("jsonfield.attr", json_option);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(json_field=> json!({"attr.age": 32}), shadowing_json_field=>json!({"age": 33})))
             .unwrap();
         index_writer.commit().unwrap();
         let searcher = index.reader().unwrap().searcher();
         let fast_field_reader = searcher.segment_reader(0u32).fast_fields();
+        // Supported for now, maybe dropped in the future.
         let column = fast_field_reader
             .column_opt::<i64>("jsonfield.attr.age")
             .unwrap()
             .unwrap();
         let vals: Vec<i64> = column.values_for_doc(0u32).collect();
         assert_eq!(&vals, &[33]);
+        let column = fast_field_reader
+            .column_opt::<i64>("jsonfield\\.attr.age")
+            .unwrap()
+            .unwrap();
+        let vals: Vec<i64> = column.values_for_doc(0u32).collect();
+        assert_eq!(&vals, &[33]);
     }
 }

src/fastfield/readers.rs

@@ -357,7 +357,7 @@ mod tests {
     use columnar::ColumnType;
 
     use crate::schema::{JsonObjectOptions, Schema, FAST};
-    use crate::{Document, Index};
+    use crate::{Index, IndexWriter, TantivyDocument};
 
     #[test]
     fn test_fast_field_reader_resolve_with_dynamic_internal() {
@@ -373,8 +373,10 @@ mod tests {
         let dynamic_field = schema_builder.add_json_field("_dyna", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
-        index_writer.add_document(Document::default()).unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
+        index_writer
+            .add_document(TantivyDocument::default())
+            .unwrap();
         index_writer.commit().unwrap();
         let reader = index.reader().unwrap();
         let searcher = reader.searcher();
@@ -443,7 +445,7 @@ mod tests {
         let json = schema_builder.add_json_field("json", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(id=> 1u64, json => json!({"foo": 42})))
             .unwrap();

src/fastfield/writer.rs

@@ -1,12 +1,12 @@
 use std::io;
 
 use columnar::{ColumnarWriter, NumericalValue};
-use common::replace_in_place;
+use common::JsonPathWriter;
 use tokenizer_api::Token;
 
 use crate::indexer::doc_id_mapping::DocIdMapping;
-use crate::schema::term::{JSON_PATH_SEGMENT_SEP, JSON_PATH_SEGMENT_SEP_STR};
-use crate::schema::{value_type_to_column_type, Document, FieldType, Schema, Type, Value};
+use crate::schema::document::{Document, ReferenceValue, ReferenceValueLeaf, Value};
+use crate::schema::{value_type_to_column_type, Field, FieldType, Schema, Type};
 use crate::tokenizer::{TextAnalyzer, TokenizerManager};
 use crate::{DateTimePrecision, DocId, TantivyError};
 
@@ -23,7 +23,7 @@ pub struct FastFieldsWriter {
     expand_dots: Vec<bool>,
     num_docs: DocId,
     // Buffer that we recycle to avoid allocation.
-    json_path_buffer: String,
+    json_path_buffer: JsonPathWriter,
 }
 
 impl FastFieldsWriter {
@@ -97,7 +97,7 @@ impl FastFieldsWriter {
             num_docs: 0u32,
             date_precisions,
             expand_dots,
-            json_path_buffer: String::new(),
+            json_path_buffer: JsonPathWriter::default(),
         })
     }
 
@@ -117,114 +117,121 @@ impl FastFieldsWriter {
     }
 
     /// Indexes all of the fastfields of a new document.
-    pub fn add_document(&mut self, doc: &Document) -> crate::Result<()> {
+    pub fn add_document<D: Document>(&mut self, doc: &D) -> crate::Result<()> {
         let doc_id = self.num_docs;
-        for field_value in doc.field_values() {
-            if let Some(field_name) =
-                &self.fast_field_names[field_value.field().field_id() as usize]
-            {
-                match &field_value.value {
-                    Value::U64(u64_val) => {
-                        self.columnar_writer.record_numerical(
-                            doc_id,
-                            field_name.as_str(),
-                            NumericalValue::from(*u64_val),
-                        );
-                    }
-                    Value::I64(i64_val) => {
-                        self.columnar_writer.record_numerical(
-                            doc_id,
-                            field_name.as_str(),
-                            NumericalValue::from(*i64_val),
-                        );
-                    }
-                    Value::F64(f64_val) => {
-                        self.columnar_writer.record_numerical(
-                            doc_id,
-                            field_name.as_str(),
-                            NumericalValue::from(*f64_val),
-                        );
-                    }
-                    Value::Str(text_val) => {
-                        if let Some(tokenizer) =
-                            &mut self.per_field_tokenizer[field_value.field().field_id() as usize]
-                        {
-                            let mut token_stream = tokenizer.token_stream(text_val);
-                            token_stream.process(&mut |token: &Token| {
-                                self.columnar_writer.record_str(
-                                    doc_id,
-                                    field_name.as_str(),
-                                    &token.text,
-                                );
-                            })
-                        } else {
-                            self.columnar_writer
-                                .record_str(doc_id, field_name.as_str(), text_val);
-                        }
-                    }
-                    Value::Bytes(bytes_val) => {
-                        self.columnar_writer
-                            .record_bytes(doc_id, field_name.as_str(), bytes_val);
-                    }
-                    Value::PreTokStr(pre_tok) => {
-                        for token in &pre_tok.tokens {
-                            self.columnar_writer.record_str(
-                                doc_id,
-                                field_name.as_str(),
-                                &token.text,
-                            );
-                        }
-                    }
-                    Value::Bool(bool_val) => {
-                        self.columnar_writer
-                            .record_bool(doc_id, field_name.as_str(), *bool_val);
-                    }
-                    Value::Date(datetime) => {
-                        let date_precision =
-                            self.date_precisions[field_value.field().field_id() as usize];
-                        let truncated_datetime = datetime.truncate(date_precision);
-                        self.columnar_writer.record_datetime(
-                            doc_id,
-                            field_name.as_str(),
-                            truncated_datetime,
-                        );
-                    }
-                    Value::Facet(facet) => {
-                        self.columnar_writer.record_str(
-                            doc_id,
-                            field_name.as_str(),
-                            facet.encoded_str(),
-                        );
-                    }
-                    Value::JsonObject(json_obj) => {
-                        let expand_dots = self.expand_dots[field_value.field().field_id() as usize];
-                        self.json_path_buffer.clear();
-                        self.json_path_buffer.push_str(field_name);
+        for (field, value) in doc.iter_fields_and_values() {
+            let value_access = value as D::Value<'_>;
 
-                        let text_analyzer =
-                            &mut self.per_field_tokenizer[field_value.field().field_id() as usize];
-
-                        record_json_obj_to_columnar_writer(
-                            doc_id,
-                            json_obj,
-                            expand_dots,
-                            JSON_DEPTH_LIMIT,
-                            &mut self.json_path_buffer,
-                            &mut self.columnar_writer,
-                            text_analyzer,
-                        );
-                    }
-                    Value::IpAddr(ip_addr) => {
-                        self.columnar_writer
-                            .record_ip_addr(doc_id, field_name.as_str(), *ip_addr);
-                    }
-                }
-            }
+            self.add_doc_value(doc_id, field, value_access)?;
         }
         self.num_docs += 1;
         Ok(())
     }
 
+    fn add_doc_value<'a, V: Value<'a>>(
+        &mut self,
+        doc_id: DocId,
+        field: Field,
+        value: V,
+    ) -> crate::Result<()> {
+        let field_name = match &self.fast_field_names[field.field_id() as usize] {
+            None => return Ok(()),
+            Some(name) => name,
+        };
+
+        match value.as_value() {
+            ReferenceValue::Leaf(leaf) => match leaf {
+                ReferenceValueLeaf::Null => {}
+                ReferenceValueLeaf::Str(val) => {
+                    if let Some(tokenizer) =
+                        &mut self.per_field_tokenizer[field.field_id() as usize]
+                    {
+                        let mut token_stream = tokenizer.token_stream(val);
+                        token_stream.process(&mut |token: &Token| {
+                            self.columnar_writer
+                                .record_str(doc_id, field_name, &token.text);
+                        })
+                    } else {
+                        self.columnar_writer.record_str(doc_id, field_name, val);
+                    }
+                }
+                ReferenceValueLeaf::U64(val) => {
+                    self.columnar_writer.record_numerical(
+                        doc_id,
+                        field_name,
+                        NumericalValue::from(val),
+                    );
+                }
+                ReferenceValueLeaf::I64(val) => {
+                    self.columnar_writer.record_numerical(
+                        doc_id,
+                        field_name,
+                        NumericalValue::from(val),
+                    );
+                }
+                ReferenceValueLeaf::F64(val) => {
+                    self.columnar_writer.record_numerical(
+                        doc_id,
+                        field_name,
+                        NumericalValue::from(val),
+                    );
+                }
+                ReferenceValueLeaf::Date(val) => {
+                    let date_precision = self.date_precisions[field.field_id() as usize];
+                    let truncated_datetime = val.truncate(date_precision);
+                    self.columnar_writer
+                        .record_datetime(doc_id, field_name, truncated_datetime);
+                }
+                ReferenceValueLeaf::Facet(val) => {
+                    self.columnar_writer
+                        .record_str(doc_id, field_name, val.encoded_str());
+                }
+                ReferenceValueLeaf::Bytes(val) => {
+                    self.columnar_writer.record_bytes(doc_id, field_name, val);
+                }
+                ReferenceValueLeaf::IpAddr(val) => {
+                    self.columnar_writer.record_ip_addr(doc_id, field_name, val);
+                }
+                ReferenceValueLeaf::Bool(val) => {
+                    self.columnar_writer.record_bool(doc_id, field_name, val);
+                }
+                ReferenceValueLeaf::PreTokStr(val) => {
+                    for token in &val.tokens {
+                        self.columnar_writer
+                            .record_str(doc_id, field_name, &token.text);
+                    }
+                }
+            },
+            ReferenceValue::Array(val) => {
+                // TODO: Check this is the correct behaviour we want.
+                for value in val {
+                    self.add_doc_value(doc_id, field, value)?;
+                }
+            }
+            ReferenceValue::Object(val) => {
+                let expand_dots = self.expand_dots[field.field_id() as usize];
+                self.json_path_buffer.clear();
+                // First field should not be expanded.
+                self.json_path_buffer.set_expand_dots(false);
+                self.json_path_buffer.push(field_name);
+                self.json_path_buffer.set_expand_dots(expand_dots);
+
+                let text_analyzer = &mut self.per_field_tokenizer[field.field_id() as usize];
+
+                record_json_obj_to_columnar_writer::<V>(
+                    doc_id,
+                    val,
+                    JSON_DEPTH_LIMIT,
+                    &mut self.json_path_buffer,
+                    &mut self.columnar_writer,
+                    text_analyzer,
+                );
+            }
+        }
+
+        Ok(())
+    }
+
     /// Serializes all of the `FastFieldWriter`s by pushing them in
     /// order to the fast field serializer.
     pub fn serialize(
@@ -241,66 +248,33 @@ impl FastFieldsWriter {
     }
 }
 
-#[inline]
-fn columnar_numerical_value(json_number: &serde_json::Number) -> Option<NumericalValue> {
-    if let Some(num_i64) = json_number.as_i64() {
-        return Some(num_i64.into());
-    }
-    if let Some(num_u64) = json_number.as_u64() {
-        return Some(num_u64.into());
-    }
-    if let Some(num_f64) = json_number.as_f64() {
-        return Some(num_f64.into());
-    }
-    // This can happen with arbitrary precision.... but we do not handle it.
-    None
-}
-
-fn record_json_obj_to_columnar_writer(
+fn record_json_obj_to_columnar_writer<'a, V: Value<'a>>(
     doc: DocId,
-    json_obj: &serde_json::Map<String, serde_json::Value>,
-    expand_dots: bool,
+    json_visitor: V::ObjectIter,
     remaining_depth_limit: usize,
-    json_path_buffer: &mut String,
+    json_path_buffer: &mut JsonPathWriter,
     columnar_writer: &mut columnar::ColumnarWriter,
     tokenizer: &mut Option<TextAnalyzer>,
 ) {
-    for (key, child) in json_obj {
-        let len_path = json_path_buffer.len();
-        if !json_path_buffer.is_empty() {
-            json_path_buffer.push_str(JSON_PATH_SEGMENT_SEP_STR);
-        }
-        json_path_buffer.push_str(key);
-        if expand_dots {
-            // This might include the separation byte, which is ok because it is not a dot.
-            let appended_segment = &mut json_path_buffer[len_path..];
-            // The unsafe below is safe as long as b'.' and JSON_PATH_SEGMENT_SEP are
-            // valid single byte ut8 strings.
-            // By utf-8 design, they cannot be part of another codepoint.
-            replace_in_place(b'.', JSON_PATH_SEGMENT_SEP, unsafe {
-                appended_segment.as_bytes_mut()
-            });
-        }
+    for (key, child) in json_visitor {
+        json_path_buffer.push(key);
         record_json_value_to_columnar_writer(
             doc,
             child,
-            expand_dots,
             remaining_depth_limit,
             json_path_buffer,
             columnar_writer,
             tokenizer,
         );
-        // popping our sub path.
-        json_path_buffer.truncate(len_path);
+        json_path_buffer.pop();
     }
 }
 
-fn record_json_value_to_columnar_writer(
+fn record_json_value_to_columnar_writer<'a, V: Value<'a>>(
     doc: DocId,
-    json_val: &serde_json::Value,
-    expand_dots: bool,
+    json_val: V,
     mut remaining_depth_limit: usize,
-    json_path_writer: &mut String,
+    json_path_writer: &mut JsonPathWriter,
     columnar_writer: &mut columnar::ColumnarWriter,
     tokenizer: &mut Option<TextAnalyzer>,
 ) {
@@ -308,34 +282,69 @@ fn record_json_value_to_columnar_writer(
         return;
     }
     remaining_depth_limit -= 1;
-    match json_val {
-        serde_json::Value::Null => {
-            // TODO handle null
-        }
-        serde_json::Value::Bool(bool_val) => {
-            columnar_writer.record_bool(doc, json_path_writer, *bool_val);
-        }
-        serde_json::Value::Number(json_number) => {
-            if let Some(numerical_value) = columnar_numerical_value(json_number) {
-                columnar_writer.record_numerical(doc, json_path_writer.as_str(), numerical_value);
+
+    match json_val.as_value() {
+        ReferenceValue::Leaf(leaf) => match leaf {
+            ReferenceValueLeaf::Null => {} // TODO: Handle null
+            ReferenceValueLeaf::Str(val) => {
+                if let Some(text_analyzer) = tokenizer.as_mut() {
+                    let mut token_stream = text_analyzer.token_stream(val);
+                    token_stream.process(&mut |token| {
+                        columnar_writer.record_str(doc, json_path_writer.as_str(), &token.text);
+                    })
+                } else {
+                    columnar_writer.record_str(doc, json_path_writer.as_str(), val);
+                }
             }
-        }
-        serde_json::Value::String(text) => {
-            if let Some(text_analyzer) = tokenizer.as_mut() {
-                let mut token_stream = text_analyzer.token_stream(text);
-                token_stream.process(&mut |token| {
-                    columnar_writer.record_str(doc, json_path_writer.as_str(), &token.text);
-                })
-            } else {
-                columnar_writer.record_str(doc, json_path_writer.as_str(), text);
+            ReferenceValueLeaf::U64(val) => {
+                columnar_writer.record_numerical(
+                    doc,
+                    json_path_writer.as_str(),
+                    NumericalValue::from(val),
+                );
             }
-        }
-        serde_json::Value::Array(arr) => {
-            for el in arr {
+            ReferenceValueLeaf::I64(val) => {
+                columnar_writer.record_numerical(
+                    doc,
+                    json_path_writer.as_str(),
+                    NumericalValue::from(val),
+                );
+            }
+            ReferenceValueLeaf::F64(val) => {
+                columnar_writer.record_numerical(
+                    doc,
+                    json_path_writer.as_str(),
+                    NumericalValue::from(val),
+                );
+            }
+            ReferenceValueLeaf::Bool(val) => {
+                columnar_writer.record_bool(doc, json_path_writer.as_str(), val);
+            }
+            ReferenceValueLeaf::Date(val) => {
+                columnar_writer.record_datetime(doc, json_path_writer.as_str(), val);
+            }
+            ReferenceValueLeaf::Facet(_) => {
+                unimplemented!("Facet support in dynamic fields is not yet implemented")
+            }
+            ReferenceValueLeaf::Bytes(_) => {
+                // TODO: This can be re added once it is added to the JSON Utils section as well.
+                // columnar_writer.record_bytes(doc, json_path_writer.as_str(), val);
+                unimplemented!("Bytes support in dynamic fields is not yet implemented")
+            }
+            ReferenceValueLeaf::IpAddr(_) => {
+                unimplemented!("IP address support in dynamic fields is not yet implemented")
+            }
+            ReferenceValueLeaf::PreTokStr(_) => {
+                unimplemented!(
+                    "Pre-tokenized string support in dynamic fields is not yet implemented"
+                )
+            }
+        },
+        ReferenceValue::Array(elements) => {
+            for el in elements {
                 record_json_value_to_columnar_writer(
                     doc,
                     el,
-                    expand_dots,
                     remaining_depth_limit,
                     json_path_writer,
                     columnar_writer,
@@ -343,11 +352,10 @@ fn record_json_value_to_columnar_writer(
                 );
             }
         }
-        serde_json::Value::Object(json_obj) => {
-            record_json_obj_to_columnar_writer(
+        ReferenceValue::Object(object) => {
+            record_json_obj_to_columnar_writer::<V>(
                 doc,
-                json_obj,
-                expand_dots,
+                object,
                 remaining_depth_limit,
                 json_path_writer,
                 columnar_writer,
@@ -360,6 +368,7 @@ fn record_json_value_to_columnar_writer(
 #[cfg(test)]
 mod tests {
     use columnar::{Column, ColumnarReader, ColumnarWriter, StrColumn};
+    use common::JsonPathWriter;
 
     use super::record_json_value_to_columnar_writer;
     use crate::fastfield::writer::JSON_DEPTH_LIMIT;
@@ -370,12 +379,12 @@ mod tests {
         expand_dots: bool,
     ) -> ColumnarReader {
         let mut columnar_writer = ColumnarWriter::default();
-        let mut json_path = String::new();
+        let mut json_path = JsonPathWriter::default();
+        json_path.set_expand_dots(expand_dots);
         for (doc, json_doc) in json_docs.iter().enumerate() {
             record_json_value_to_columnar_writer(
                 doc as u32,
                 json_doc,
-                expand_dots,
                 JSON_DEPTH_LIMIT,
                 &mut json_path,
                 &mut columnar_writer,

src/functional_test.rs

@@ -4,7 +4,7 @@ use rand::{thread_rng, Rng};
 
 use crate::indexer::index_writer::MEMORY_BUDGET_NUM_BYTES_MIN;
 use crate::schema::*;
-use crate::{doc, schema, Index, IndexSettings, IndexSortByField, Order, Searcher};
+use crate::{doc, schema, Index, IndexSettings, IndexSortByField, IndexWriter, Order, Searcher};
 
 fn check_index_content(searcher: &Searcher, vals: &[u64]) -> crate::Result<()> {
     assert!(searcher.segment_readers().len() < 20);
@@ -12,7 +12,7 @@ fn check_index_content(searcher: &Searcher, vals: &[u64]) -> crate::Result<()> {
     for segment_reader in searcher.segment_readers() {
         let store_reader = segment_reader.get_store_reader(1)?;
         for doc_id in 0..segment_reader.max_doc() {
-            let _doc = store_reader.get(doc_id)?;
+            let _doc: TantivyDocument = store_reader.get(doc_id)?;
         }
     }
     Ok(())
@@ -31,7 +31,8 @@ fn test_functional_store() -> crate::Result<()> {
 
     let mut rng = thread_rng();
 
-    let mut index_writer = index.writer_with_num_threads(3, MEMORY_BUDGET_NUM_BYTES_MIN)?;
+    let mut index_writer: IndexWriter =
+        index.writer_with_num_threads(3, 3 * MEMORY_BUDGET_NUM_BYTES_MIN)?;
 
     let mut doc_set: Vec<u64> = Vec::new();
 
@@ -91,7 +92,8 @@ fn test_functional_indexing_sorted() -> crate::Result<()> {
 
     let mut rng = thread_rng();
 
-    let mut index_writer = index.writer_with_num_threads(3, 120_000_000)?;
+    let mut index_writer: IndexWriter =
+        index.writer_with_num_threads(3, 3 * MEMORY_BUDGET_NUM_BYTES_MIN)?;
 
     let mut committed_docs: HashSet<u64> = HashSet::new();
     let mut uncommitted_docs: HashSet<u64> = HashSet::new();
@@ -114,7 +116,7 @@ fn test_functional_indexing_sorted() -> crate::Result<()> {
             index_writer.delete_term(doc_id_term);
         } else {
             uncommitted_docs.insert(random_val);
-            let mut doc = Document::new();
+            let mut doc = TantivyDocument::new();
             doc.add_u64(id_field, random_val);
             for i in 1u64..10u64 {
                 doc.add_u64(multiples_field, random_val * i);
@@ -166,7 +168,8 @@ fn test_functional_indexing_unsorted() -> crate::Result<()> {
 
     let mut rng = thread_rng();
 
-    let mut index_writer = index.writer_with_num_threads(3, 120_000_000)?;
+    let mut index_writer: IndexWriter =
+        index.writer_with_num_threads(3, 3 * MEMORY_BUDGET_NUM_BYTES_MIN)?;
 
     let mut committed_docs: HashSet<u64> = HashSet::new();
     let mut uncommitted_docs: HashSet<u64> = HashSet::new();
@@ -189,7 +192,7 @@ fn test_functional_indexing_unsorted() -> crate::Result<()> {
             index_writer.delete_term(doc_id_term);
         } else {
             uncommitted_docs.insert(random_val);
-            let mut doc = Document::new();
+            let mut doc = TantivyDocument::new();
             doc.add_u64(id_field, random_val);
             for i in 1u64..10u64 {
                 doc.add_u64(multiples_field, random_val * i);

src/indexer/doc_id_mapping.rs

@@ -158,6 +158,7 @@ mod tests_indexsorting {
     use crate::indexer::doc_id_mapping::DocIdMapping;
     use crate::indexer::NoMergePolicy;
     use crate::query::QueryParser;
+    use crate::schema::document::Value;
     use crate::schema::{Schema, *};
     use crate::{DocAddress, Index, IndexSettings, IndexSortByField, Order};
 
@@ -308,16 +309,16 @@ mod tests_indexsorting {
         {
             assert_eq!(
                 searcher
-                    .doc(DocAddress::new(0, 0))?
+                    .doc::<TantivyDocument>(DocAddress::new(0, 0))?
                     .get_first(my_string_field),
                 None
             );
             assert_eq!(
                 searcher
-                    .doc(DocAddress::new(0, 3))?
+                    .doc::<TantivyDocument>(DocAddress::new(0, 3))?
                     .get_first(my_string_field)
                     .unwrap()
-                    .as_text(),
+                    .as_str(),
                 Some("blublub")
             );
         }
@@ -337,13 +338,13 @@ mod tests_indexsorting {
         {
             assert_eq!(
                 searcher
-                    .doc(DocAddress::new(0, 0))?
+                    .doc::<TantivyDocument>(DocAddress::new(0, 0))?
                     .get_first(my_string_field)
                     .unwrap()
-                    .as_text(),
+                    .as_str(),
                 Some("blublub")
             );
-            let doc = searcher.doc(DocAddress::new(0, 4))?;
+            let doc = searcher.doc::<TantivyDocument>(DocAddress::new(0, 4))?;
             assert_eq!(doc.get_first(my_string_field), None);
         }
         // sort by field desc
@@ -360,9 +361,9 @@ mod tests_indexsorting {
         let my_string_field = index.schema().get_field("string_field").unwrap();
         let searcher = index.reader()?.searcher();
         {
-            let doc = searcher.doc(DocAddress::new(0, 4))?;
+            let doc = searcher.doc::<TantivyDocument>(DocAddress::new(0, 4))?;
             assert_eq!(
-                doc.get_first(my_string_field).unwrap().as_text(),
+                doc.get_first(my_string_field).unwrap().as_str(),
                 Some("blublub")
             );
         }

src/indexer/index_writer.rs

@@ -20,7 +20,8 @@ use crate::indexer::operation::DeleteOperation;
 use crate::indexer::stamper::Stamper;
 use crate::indexer::{MergePolicy, SegmentEntry, SegmentWriter};
 use crate::query::{EnableScoring, Query, TermQuery};
-use crate::schema::{Document, IndexRecordOption, Term};
+use crate::schema::document::Document;
+use crate::schema::{IndexRecordOption, TantivyDocument, Term};
 use crate::{FutureResult, Opstamp};
 
 // Size of the margin for the `memory_arena`. A segment is closed when the remaining memory
@@ -50,7 +51,7 @@ fn error_in_index_worker_thread(context: &str) -> TantivyError {
 /// indexing queue.
 /// Each indexing thread builds its own independent [`Segment`], via
 /// a `SegmentWriter` object.
-pub struct IndexWriter {
+pub struct IndexWriter<D: Document = TantivyDocument> {
     // the lock is just used to bind the
     // lifetime of the lock with that of the IndexWriter.
     _directory_lock: Option<DirectoryLock>,
@@ -62,8 +63,8 @@ pub struct IndexWriter {
 
     workers_join_handle: Vec<JoinHandle<crate::Result<()>>>,
 
-    index_writer_status: IndexWriterStatus,
-    operation_sender: AddBatchSender,
+    index_writer_status: IndexWriterStatus<D>,
+    operation_sender: AddBatchSender<D>,
 
     segment_updater: SegmentUpdater,
 
@@ -164,10 +165,10 @@ pub(crate) fn advance_deletes(
     Ok(())
 }
 
-fn index_documents(
+fn index_documents<D: Document>(
     memory_budget: usize,
     segment: Segment,
-    grouped_document_iterator: &mut dyn Iterator<Item = AddBatch>,
+    grouped_document_iterator: &mut dyn Iterator<Item = AddBatch<D>>,
     segment_updater: &SegmentUpdater,
     mut delete_cursor: DeleteCursor,
 ) -> crate::Result<()> {
@@ -247,7 +248,7 @@ fn apply_deletes(
     })
 }
 
-impl IndexWriter {
+impl<D: Document> IndexWriter<D> {
     /// Create a new index writer. Attempts to acquire a lockfile.
     ///
     /// The lockfile should be deleted on drop, but it is possible
@@ -267,7 +268,7 @@ impl IndexWriter {
         num_threads: usize,
         memory_budget_in_bytes_per_thread: usize,
         directory_lock: DirectoryLock,
-    ) -> crate::Result<IndexWriter> {
+    ) -> crate::Result<Self> {
         if memory_budget_in_bytes_per_thread < MEMORY_BUDGET_NUM_BYTES_MIN {
             let err_msg = format!(
                 "The memory arena in bytes per thread needs to be at least \
@@ -281,7 +282,7 @@ impl IndexWriter {
             );
             return Err(TantivyError::InvalidArgument(err_msg));
         }
-        let (document_sender, document_receiver): (AddBatchSender, AddBatchReceiver) =
+        let (document_sender, document_receiver) =
             crossbeam_channel::bounded(PIPELINE_MAX_SIZE_IN_DOCS);
 
         let delete_queue = DeleteQueue::new();
@@ -293,7 +294,7 @@ impl IndexWriter {
         let segment_updater =
             SegmentUpdater::create(index.clone(), stamper.clone(), &delete_queue.cursor())?;
 
-        let mut index_writer = IndexWriter {
+        let mut index_writer = Self {
             _directory_lock: Some(directory_lock),
 
             memory_budget_in_bytes_per_thread,
@@ -375,7 +376,7 @@ impl IndexWriter {
         self.index.new_segment()
     }
 
-    fn operation_receiver(&self) -> crate::Result<AddBatchReceiver> {
+    fn operation_receiver(&self) -> crate::Result<AddBatchReceiver<D>> {
         self.index_writer_status
             .operation_receiver()
             .ok_or_else(|| {
@@ -525,7 +526,7 @@ impl IndexWriter {
     ///
     /// Returns the former segment_ready channel.
     fn recreate_document_channel(&mut self) {
-        let (document_sender, document_receiver): (AddBatchSender, AddBatchReceiver) =
+        let (document_sender, document_receiver) =
             crossbeam_channel::bounded(PIPELINE_MAX_SIZE_IN_DOCS);
         self.operation_sender = document_sender;
         self.index_writer_status = IndexWriterStatus::from(document_receiver);
@@ -552,7 +553,7 @@ impl IndexWriter {
             .take()
             .expect("The IndexWriter does not have any lock. This is a bug, please report.");
 
-        let new_index_writer: IndexWriter = IndexWriter::new(
+        let new_index_writer = IndexWriter::new(
             &self.index,
             self.num_threads,
             self.memory_budget_in_bytes_per_thread,
@@ -598,7 +599,7 @@ impl IndexWriter {
     /// It is also possible to add a payload to the `commit`
     /// using this API.
     /// See [`PreparedCommit::set_payload()`].
-    pub fn prepare_commit(&mut self) -> crate::Result<PreparedCommit> {
+    pub fn prepare_commit(&mut self) -> crate::Result<PreparedCommit<D>> {
         // Here, because we join all of the worker threads,
         // all of the segment update for this commit have been
         // sent.
@@ -707,7 +708,7 @@ impl IndexWriter {
     /// The opstamp is an increasing `u64` that can
     /// be used by the client to align commits with its own
     /// document queue.
-    pub fn add_document(&self, document: Document) -> crate::Result<Opstamp> {
+    pub fn add_document(&self, document: D) -> crate::Result<Opstamp> {
         let opstamp = self.stamper.stamp();
         self.send_add_documents_batch(smallvec![AddOperation { opstamp, document }])?;
         Ok(opstamp)
@@ -744,7 +745,7 @@ impl IndexWriter {
     /// visible to readers only after calling `commit()`.
     pub fn run<I>(&self, user_operations: I) -> crate::Result<Opstamp>
     where
-        I: IntoIterator<Item = UserOperation>,
+        I: IntoIterator<Item = UserOperation<D>>,
         I::IntoIter: ExactSizeIterator,
     {
         let user_operations_it = user_operations.into_iter();
@@ -778,7 +779,7 @@ impl IndexWriter {
         Ok(batch_opstamp)
     }
 
-    fn send_add_documents_batch(&self, add_ops: AddBatch) -> crate::Result<()> {
+    fn send_add_documents_batch(&self, add_ops: AddBatch<D>) -> crate::Result<()> {
         if self.index_writer_status.is_alive() && self.operation_sender.send(add_ops).is_ok() {
             Ok(())
         } else {
@@ -787,7 +788,7 @@ impl IndexWriter {
     }
 }
 
-impl Drop for IndexWriter {
+impl<D: Document> Drop for IndexWriter<D> {
     fn drop(&mut self) {
         self.segment_updater.kill();
         self.drop_sender();
@@ -814,13 +815,15 @@ mod tests {
     use crate::indexer::index_writer::MEMORY_BUDGET_NUM_BYTES_MIN;
     use crate::indexer::NoMergePolicy;
     use crate::query::{BooleanQuery, Occur, Query, QueryParser, TermQuery};
+    use crate::schema::document::Value;
     use crate::schema::{
         self, Facet, FacetOptions, IndexRecordOption, IpAddrOptions, NumericOptions, Schema,
         TextFieldIndexing, TextOptions, FAST, INDEXED, STORED, STRING, TEXT,
     };
     use crate::store::DOCSTORE_CACHE_CAPACITY;
     use crate::{
-        DateTime, DocAddress, Index, IndexSettings, IndexSortByField, Order, ReloadPolicy, Term,
+        DateTime, DocAddress, Index, IndexSettings, IndexSortByField, IndexWriter, Order,
+        ReloadPolicy, TantivyDocument, Term,
     };
 
     const LOREM: &str = "Doc Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do \
@@ -852,7 +855,7 @@ mod tests {
         let text_field = schema_builder.add_text_field("text", TEXT);
         let index = Index::create_in_ram(schema_builder.build());
 
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(text_field => "hello1"))
             .unwrap();
@@ -905,7 +908,7 @@ mod tests {
             .reload_policy(ReloadPolicy::Manual)
             .try_into()
             .unwrap();
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         let a_term = Term::from_field_text(text_field, "a");
         let b_term = Term::from_field_text(text_field, "b");
         let operations = vec![
@@ -943,7 +946,7 @@ mod tests {
     fn test_empty_operations_group() {
         let schema_builder = schema::Schema::builder();
         let index = Index::create_in_ram(schema_builder.build());
-        let index_writer = index.writer_for_tests().unwrap();
+        let index_writer: IndexWriter = index.writer_for_tests().unwrap();
         let operations1 = vec![];
         let batch_opstamp1 = index_writer.run(operations1).unwrap();
         assert_eq!(batch_opstamp1, 0u64);
@@ -956,8 +959,8 @@ mod tests {
     fn test_lockfile_stops_duplicates() {
         let schema_builder = schema::Schema::builder();
         let index = Index::create_in_ram(schema_builder.build());
-        let _index_writer = index.writer_for_tests().unwrap();
-        match index.writer_for_tests() {
+        let _index_writer: IndexWriter = index.writer_for_tests().unwrap();
+        match index.writer_for_tests::<TantivyDocument>() {
             Err(TantivyError::LockFailure(LockError::LockBusy, _)) => {}
             _ => panic!("Expected a `LockFailure` error"),
         }
@@ -967,8 +970,8 @@ mod tests {
     fn test_lockfile_already_exists_error_msg() {
         let schema_builder = schema::Schema::builder();
         let index = Index::create_in_ram(schema_builder.build());
-        let _index_writer = index.writer_for_tests().unwrap();
-        match index.writer_for_tests() {
+        let _index_writer: IndexWriter = index.writer_for_tests().unwrap();
+        match index.writer_for_tests::<TantivyDocument>() {
             Err(err) => {
                 let err_msg = err.to_string();
                 assert!(err_msg.contains("already an `IndexWriter`"));
@@ -981,7 +984,7 @@ mod tests {
     fn test_set_merge_policy() {
         let schema_builder = schema::Schema::builder();
         let index = Index::create_in_ram(schema_builder.build());
-        let index_writer = index.writer_for_tests().unwrap();
+        let index_writer: IndexWriter = index.writer_for_tests().unwrap();
         assert_eq!(
             format!("{:?}", index_writer.get_merge_policy()),
             "LogMergePolicy { min_num_segments: 8, max_docs_before_merge: 10000000, \
@@ -1000,11 +1003,11 @@ mod tests {
         let schema_builder = schema::Schema::builder();
         let index = Index::create_in_ram(schema_builder.build());
         {
-            let _index_writer = index.writer_for_tests().unwrap();
+            let _index_writer: IndexWriter = index.writer_for_tests().unwrap();
             // the lock should be released when the
             // index_writer leaves the scope.
         }
-        let _index_writer_two = index.writer_for_tests().unwrap();
+        let _index_writer_two: IndexWriter = index.writer_for_tests().unwrap();
     }
 
     #[test]
@@ -1056,7 +1059,7 @@ mod tests {
             reader.searcher().doc_freq(&term_a).unwrap()
         };
         // writing the segment
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer.add_document(doc!(text_field=>"a"))?;
         index_writer.commit()?;
         //  this should create 1 segment
@@ -1096,7 +1099,7 @@ mod tests {
             reader.searcher().doc_freq(&term_a).unwrap()
         };
         // writing the segment
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer.add_document(doc!(text_field=>"a"))?;
         index_writer.commit()?;
         index_writer.add_document(doc!(text_field=>"a"))?;
@@ -1382,7 +1385,7 @@ mod tests {
     fn test_delete_all_documents_empty_index() {
         let schema_builder = schema::Schema::builder();
         let index = Index::create_in_ram(schema_builder.build());
-        let mut index_writer = index
+        let mut index_writer: IndexWriter = index
             .writer_with_num_threads(4, MEMORY_BUDGET_NUM_BYTES_MIN * 4)
             .unwrap();
         let clear = index_writer.delete_all_documents();
@@ -1395,7 +1398,7 @@ mod tests {
     fn test_delete_all_documents_index_twice() {
         let schema_builder = schema::Schema::builder();
         let index = Index::create_in_ram(schema_builder.build());
-        let mut index_writer = index
+        let mut index_writer: IndexWriter = index
             .writer_with_num_threads(4, MEMORY_BUDGET_NUM_BYTES_MIN * 4)
             .unwrap();
         let clear = index_writer.delete_all_documents();
@@ -1415,7 +1418,7 @@ mod tests {
         let schema = schema_builder.build();
 
         let index = Index::builder().schema(schema).create_in_ram().unwrap();
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer
             .add_document(doc!(text_field => "one"))
             .unwrap();
@@ -1703,7 +1706,8 @@ mod tests {
 
         let old_reader = index.reader()?;
 
-        let id_exists = |id| id % 3 != 0; // 0 does not exist
+        // Every 3rd doc has only id field
+        let id_is_full_doc = |id| id % 3 != 0;
 
         let multi_text_field_text1 = "test1 test2 test3 test1 test2 test3";
         // rotate left
@@ -1719,7 +1723,7 @@ mod tests {
                     let facet = Facet::from(&("/cola/".to_string() + &id.to_string()));
                     let ip = ip_from_id(id);
 
-                    if !id_exists(id) {
+                    if !id_is_full_doc(id) {
                         // every 3rd doc has no ip field
                         index_writer.add_document(doc!(
                             id_field=>id,
@@ -1777,7 +1781,7 @@ mod tests {
         let num_segments_before_merge = searcher.segment_readers().len();
         if force_end_merge {
             index_writer.wait_merging_threads()?;
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             let segment_ids = index
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
@@ -1839,7 +1843,7 @@ mod tests {
 
         let num_docs_with_values = expected_ids_and_num_occurrences
             .iter()
-            .filter(|(id, _id_occurrences)| id_exists(**id))
+            .filter(|(id, _id_occurrences)| id_is_full_doc(**id))
             .map(|(_, id_occurrences)| *id_occurrences as usize)
             .sum::<usize>();
 
@@ -1863,7 +1867,7 @@ mod tests {
         if force_end_merge && num_segments_before_merge > 1 && num_segments_after_merge == 1 {
             let mut expected_multi_ips: Vec<_> = id_list
                 .iter()
-                .filter(|id| id_exists(**id))
+                .filter(|id| id_is_full_doc(**id))
                 .flat_map(|id| vec![ip_from_id(*id), ip_from_id(*id)])
                 .collect();
             assert_eq!(num_ips, expected_multi_ips.len() as u32);
@@ -1901,7 +1905,7 @@ mod tests {
         let expected_ips = expected_ids_and_num_occurrences
             .keys()
             .flat_map(|id| {
-                if !id_exists(*id) {
+                if !id_is_full_doc(*id) {
                     None
                 } else {
                     Some(Ipv6Addr::from_u128(*id as u128))
@@ -1913,7 +1917,7 @@ mod tests {
         let expected_ips = expected_ids_and_num_occurrences
             .keys()
             .filter_map(|id| {
-                if !id_exists(*id) {
+                if !id_is_full_doc(*id) {
                     None
                 } else {
                     Some(Ipv6Addr::from_u128(*id as u128))
@@ -1948,7 +1952,7 @@ mod tests {
                 let id = id_reader.first(doc).unwrap();
 
                 let vals: Vec<u64> = ff_reader.values_for_doc(doc).collect();
-                if id_exists(id) {
+                if id_is_full_doc(id) {
                     assert_eq!(vals.len(), 2);
                     assert_eq!(vals[0], vals[1]);
                     assert!(expected_ids_and_num_occurrences.contains_key(&vals[0]));
@@ -1958,7 +1962,7 @@ mod tests {
                 }
 
                 let bool_vals: Vec<bool> = bool_ff_reader.values_for_doc(doc).collect();
-                if id_exists(id) {
+                if id_is_full_doc(id) {
                     assert_eq!(bool_vals.len(), 2);
                     assert_ne!(bool_vals[0], bool_vals[1]);
                 } else {
@@ -1973,23 +1977,23 @@ mod tests {
                 .get_store_reader(DOCSTORE_CACHE_CAPACITY)
                 .unwrap();
             // test store iterator
-            for doc in store_reader.iter(segment_reader.alive_bitset()) {
+            for doc in store_reader.iter::<TantivyDocument>(segment_reader.alive_bitset()) {
                 let id = doc.unwrap().get_first(id_field).unwrap().as_u64().unwrap();
                 assert!(expected_ids_and_num_occurrences.contains_key(&id));
             }
             // test store random access
             for doc_id in segment_reader.doc_ids_alive() {
                 let id = store_reader
-                    .get(doc_id)
+                    .get::<TantivyDocument>(doc_id)
                     .unwrap()
                     .get_first(id_field)
                     .unwrap()
                     .as_u64()
                     .unwrap();
                 assert!(expected_ids_and_num_occurrences.contains_key(&id));
-                if id_exists(id) {
+                if id_is_full_doc(id) {
                     let id2 = store_reader
-                        .get(doc_id)
+                        .get::<TantivyDocument>(doc_id)
                         .unwrap()
                         .get_first(multi_numbers)
                         .unwrap()
@@ -1997,13 +2001,13 @@ mod tests {
                         .unwrap();
                     assert_eq!(id, id2);
                     let bool = store_reader
-                        .get(doc_id)
+                        .get::<TantivyDocument>(doc_id)
                         .unwrap()
                         .get_first(bool_field)
                         .unwrap()
                         .as_bool()
                         .unwrap();
-                    let doc = store_reader.get(doc_id).unwrap();
+                    let doc = store_reader.get::<TantivyDocument>(doc_id).unwrap();
                     let mut bool2 = doc.get_all(multi_bools);
                     assert_eq!(bool, bool2.next().unwrap().as_bool().unwrap());
                     assert_ne!(bool, bool2.next().unwrap().as_bool().unwrap());
@@ -2034,7 +2038,7 @@ mod tests {
             let (existing_id, count) = (*id, *count);
             let get_num_hits = |field| do_search(&existing_id.to_string(), field).len() as u64;
             assert_eq!(get_num_hits(id_field), count);
-            if !id_exists(existing_id) {
+            if !id_is_full_doc(existing_id) {
                 continue;
             }
             assert_eq!(get_num_hits(text_field), count);
@@ -2084,7 +2088,7 @@ mod tests {
         //
         for (existing_id, count) in &expected_ids_and_num_occurrences {
             let (existing_id, count) = (*existing_id, *count);
-            if !id_exists(existing_id) {
+            if !id_is_full_doc(existing_id) {
                 continue;
             }
             let do_search_ip_field = |term: &str| do_search(term, ip_field).len() as u64;
@@ -2101,34 +2105,84 @@ mod tests {
             }
         }
 
-        // assert data is like expected
+        // Range query
         //
-        for (existing_id, count) in expected_ids_and_num_occurrences.iter().take(10) {
-            let (existing_id, count) = (*existing_id, *count);
-            if !id_exists(existing_id) {
-                continue;
-            }
-            let gen_query_inclusive = |field: &str, from: Ipv6Addr, to: Ipv6Addr| {
-                format!("{}:[{} TO {}]", field, &from.to_string(), &to.to_string())
+        // Take half as sample
+        let mut sample: Vec<_> = expected_ids_and_num_occurrences.iter().collect();
+        sample.sort_by_key(|(k, _num_occurences)| *k);
+        // sample.truncate(sample.len() / 2);
+        if !sample.is_empty() {
+            let (left_sample, right_sample) = sample.split_at(sample.len() / 2);
+
+            let expected_count = |sample: &[(&u64, &u64)]| {
+                sample
+                    .iter()
+                    .filter(|(id, _)| id_is_full_doc(**id))
+                    .map(|(_id, num_occurences)| **num_occurences)
+                    .sum::<u64>()
             };
-            let ip = ip_from_id(existing_id);
+            fn gen_query_inclusive<T1: ToString, T2: ToString>(
+                field: &str,
+                from: T1,
+                to: T2,
+            ) -> String {
+                format!("{}:[{} TO {}]", field, &from.to_string(), &to.to_string())
+            }
 
-            let do_search_ip_field = |term: &str| do_search(term, ip_field).len() as u64;
-            // Range query on single value field
-            let query = gen_query_inclusive("ip", ip, ip);
-            assert_eq!(do_search_ip_field(&query), count);
+            // Query first half
+            if !left_sample.is_empty() {
+                let expected_count = expected_count(left_sample);
 
-            // Range query on multi value field
-            let query = gen_query_inclusive("ips", ip, ip);
+                let start_range = *left_sample[0].0;
+                let end_range = *left_sample.last().unwrap().0;
+                let query = gen_query_inclusive("id_opt", start_range, end_range);
+                assert_eq!(do_search(&query, id_opt_field).len() as u64, expected_count);
 
-            assert_eq!(do_search_ip_field(&query), count);
+                // Range query on ip field
+                let ip1 = ip_from_id(start_range);
+                let ip2 = ip_from_id(end_range);
+                let do_search_ip_field = |term: &str| do_search(term, ip_field).len() as u64;
+                let query = gen_query_inclusive("ip", ip1, ip2);
+                assert_eq!(do_search_ip_field(&query), expected_count);
+                let query = gen_query_inclusive("ip", "*", ip2);
+                assert_eq!(do_search_ip_field(&query), expected_count);
+                // Range query on multi value field
+                let query = gen_query_inclusive("ips", ip1, ip2);
+                assert_eq!(do_search_ip_field(&query), expected_count);
+                let query = gen_query_inclusive("ips", "*", ip2);
+                assert_eq!(do_search_ip_field(&query), expected_count);
+            }
+            // Query second half
+            if !right_sample.is_empty() {
+                let expected_count = expected_count(right_sample);
+                let start_range = *right_sample[0].0;
+                let end_range = *right_sample.last().unwrap().0;
+                // Range query on id opt field
+                let query =
+                    gen_query_inclusive("id_opt", start_range.to_string(), end_range.to_string());
+                assert_eq!(do_search(&query, id_opt_field).len() as u64, expected_count);
+
+                // Range query on ip field
+                let ip1 = ip_from_id(start_range);
+                let ip2 = ip_from_id(end_range);
+                let do_search_ip_field = |term: &str| do_search(term, ip_field).len() as u64;
+                let query = gen_query_inclusive("ip", ip1, ip2);
+                assert_eq!(do_search_ip_field(&query), expected_count);
+                let query = gen_query_inclusive("ip", ip1, "*");
+                assert_eq!(do_search_ip_field(&query), expected_count);
+                // Range query on multi value field
+                let query = gen_query_inclusive("ips", ip1, ip2);
+                assert_eq!(do_search_ip_field(&query), expected_count);
+                let query = gen_query_inclusive("ips", ip1, "*");
+                assert_eq!(do_search_ip_field(&query), expected_count);
+            }
         }
 
         // ip range query on fast field
         //
         for (existing_id, count) in expected_ids_and_num_occurrences.iter().take(10) {
             let (existing_id, count) = (*existing_id, *count);
-            if !id_exists(existing_id) {
+            if !id_is_full_doc(existing_id) {
                 continue;
             }
             let gen_query_inclusive = |field: &str, from: Ipv6Addr, to: Ipv6Addr| {
@@ -2156,7 +2210,7 @@ mod tests {
                 .first_or_default_col(9999);
             for doc_id in segment_reader.doc_ids_alive() {
                 let id = ff_reader.get_val(doc_id);
-                if !id_exists(id) {
+                if !id_is_full_doc(id) {
                     continue;
                 }
                 let facet_ords: Vec<u64> = facet_reader.facet_ords(doc_id).collect();
@@ -2194,6 +2248,12 @@ mod tests {
         Ok(index)
     }
 
+    #[test]
+    fn test_fast_field_range() {
+        let ops: Vec<_> = (0..1000).map(|id| IndexingOp::AddDoc { id }).collect();
+        assert!(test_operation_strategy(&ops, false, true).is_ok());
+    }
+
     #[test]
     fn test_sort_index_on_opt_field_regression() {
         assert!(test_operation_strategy(
@@ -2543,7 +2603,7 @@ mod tests {
         // Merge
         {
             assert!(index_writer.wait_merging_threads().is_ok());
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             let segment_ids = index
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
@@ -2585,7 +2645,7 @@ mod tests {
         // Merge
         {
             assert!(index_writer.wait_merging_threads().is_ok());
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             let segment_ids = index
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");

src/indexer/index_writer_status.rs

@@ -2,13 +2,15 @@ use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::{Arc, RwLock};
 
 use super::AddBatchReceiver;
+use crate::schema::document::Document;
+use crate::TantivyDocument;
 
 #[derive(Clone)]
-pub(crate) struct IndexWriterStatus {
-    inner: Arc<Inner>,
+pub(crate) struct IndexWriterStatus<D: Document = TantivyDocument> {
+    inner: Arc<Inner<D>>,
 }
 
-impl IndexWriterStatus {
+impl<D: Document> IndexWriterStatus<D> {
     /// Returns true iff the index writer is alive.
     pub fn is_alive(&self) -> bool {
         self.inner.as_ref().is_alive()
@@ -16,7 +18,7 @@ impl IndexWriterStatus {
 
     /// Returns a copy of the operation receiver.
     /// If the index writer was killed, returns `None`.
-    pub fn operation_receiver(&self) -> Option<AddBatchReceiver> {
+    pub fn operation_receiver(&self) -> Option<AddBatchReceiver<D>> {
         let rlock = self
             .inner
             .receive_channel
@@ -27,19 +29,19 @@ impl IndexWriterStatus {
 
     /// Create an index writer bomb.
     /// If dropped, the index writer status will be killed.
-    pub(crate) fn create_bomb(&self) -> IndexWriterBomb {
+    pub(crate) fn create_bomb(&self) -> IndexWriterBomb<D> {
         IndexWriterBomb {
             inner: Some(self.inner.clone()),
         }
     }
 }
 
-struct Inner {
+struct Inner<D: Document> {
     is_alive: AtomicBool,
-    receive_channel: RwLock<Option<AddBatchReceiver>>,
+    receive_channel: RwLock<Option<AddBatchReceiver<D>>>,
 }
 
-impl Inner {
+impl<D: Document> Inner<D> {
     fn is_alive(&self) -> bool {
         self.is_alive.load(Ordering::Relaxed)
     }
@@ -53,8 +55,8 @@ impl Inner {
     }
 }
 
-impl From<AddBatchReceiver> for IndexWriterStatus {
-    fn from(receiver: AddBatchReceiver) -> Self {
+impl<D: Document> From<AddBatchReceiver<D>> for IndexWriterStatus<D> {
+    fn from(receiver: AddBatchReceiver<D>) -> Self {
         IndexWriterStatus {
             inner: Arc::new(Inner {
                 is_alive: AtomicBool::new(true),
@@ -66,11 +68,11 @@ impl From<AddBatchReceiver> for IndexWriterStatus {
 
 /// If dropped, the index writer will be killed.
 /// To prevent this, clients can call `.defuse()`.
-pub(crate) struct IndexWriterBomb {
-    inner: Option<Arc<Inner>>,
+pub(crate) struct IndexWriterBomb<D: Document> {
+    inner: Option<Arc<Inner<D>>>,
 }
 
-impl IndexWriterBomb {
+impl<D: Document> IndexWriterBomb<D> {
     /// Defuses the bomb.
     ///
     /// This is the only way to drop the bomb without killing
@@ -80,7 +82,7 @@ impl IndexWriterBomb {
     }
 }
 
-impl Drop for IndexWriterBomb {
+impl<D: Document> Drop for IndexWriterBomb<D> {
     fn drop(&mut self) {
         if let Some(inner) = self.inner.take() {
             inner.kill();

src/indexer/merge_operation.rs

@@ -63,10 +63,13 @@ impl MergeOperation {
         }
     }
 
+    /// Returns the opstamp up to which we want to consume the delete queue and reflect their
+    /// deletes.
     pub fn target_opstamp(&self) -> Opstamp {
         self.inner.target_opstamp
     }
 
+    /// Returns the list of segment to be merged.
     pub fn segment_ids(&self) -> &[SegmentId] {
         &self.inner.segment_ids[..]
     }

src/indexer/merger.rs

@@ -552,7 +552,41 @@ impl IndexMerger {
                 continue;
             }
 
-            field_serializer.new_term(term_bytes, total_doc_freq)?;
+            // This should never happen as we early exited for total_doc_freq == 0.
+            assert!(!segment_postings_containing_the_term.is_empty());
+
+            let has_term_freq = {
+                let has_term_freq = !segment_postings_containing_the_term[0]
+                    .1
+                    .block_cursor
+                    .freqs()
+                    .is_empty();
+                for (_, postings) in &segment_postings_containing_the_term[1..] {
+                    // This may look at a strange way to test whether we have term freq or not.
+                    // With JSON object, the schema is not sufficient to know whether a term
+                    // has its term frequency encoded or not:
+                    // strings may have term frequencies, while number terms never have one.
+                    //
+                    // Ideally, we should have burnt one bit of two in the `TermInfo`.
+                    // However, we preferred not changing the codec too much and detect this
+                    // instead by
+                    // - looking at the size of the skip data for bitpacked blocks
+                    // - observing the absence of remaining data after reading the docs for vint
+                    // blocks.
+                    //
+                    // Overall the reliable way to know if we have actual frequencies loaded or not
+                    // is to check whether the actual decoded array is empty or not.
+                    if has_term_freq != !postings.block_cursor.freqs().is_empty() {
+                        return Err(DataCorruption::comment_only(
+                            "Term freqs are inconsistent across segments",
+                        )
+                        .into());
+                    }
+                }
+                has_term_freq
+            };
+
+            field_serializer.new_term(term_bytes, total_doc_freq, has_term_freq)?;
 
             // We can now serialize this postings, by pushing each document to the
             // postings serializer.
@@ -567,8 +601,13 @@ impl IndexMerger {
                     if let Some(remapped_doc_id) = old_to_new_doc_id[doc as usize] {
                         // we make sure to only write the term if
                         // there is at least one document.
-                        let term_freq = segment_postings.term_freq();
-                        segment_postings.positions(&mut positions_buffer);
+                        let term_freq = if has_term_freq {
+                            segment_postings.positions(&mut positions_buffer);
+                            segment_postings.term_freq()
+                        } else {
+                            0u32
+                        };
+
                         // if doc_id_mapping exists, the doc_ids are reordered, they are
                         // not just stacked. The field serializer expects monotonically increasing
                         // doc_ids, so we collect and sort them first, before writing.
@@ -753,9 +792,10 @@ mod tests {
     use crate::collector::{Count, FacetCollector};
     use crate::core::Index;
     use crate::query::{AllQuery, BooleanQuery, EnableScoring, Scorer, TermQuery};
+    use crate::schema::document::Value;
     use crate::schema::{
-        Document, Facet, FacetOptions, IndexRecordOption, NumericOptions, Term, TextFieldIndexing,
-        INDEXED, TEXT,
+        Facet, FacetOptions, IndexRecordOption, NumericOptions, TantivyDocument, Term,
+        TextFieldIndexing, INDEXED, TEXT,
     };
     use crate::time::OffsetDateTime;
     use crate::{
@@ -817,7 +857,7 @@ mod tests {
             let segment_ids = index
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             index_writer.merge(&segment_ids).wait()?;
             index_writer.wait_merging_threads()?;
         }
@@ -866,30 +906,24 @@ mod tests {
                 );
             }
             {
-                let doc = searcher.doc(DocAddress::new(0, 0))?;
-                assert_eq!(doc.get_first(text_field).unwrap().as_text(), Some("af b"));
+                let doc = searcher.doc::<TantivyDocument>(DocAddress::new(0, 0))?;
+                assert_eq!(doc.get_first(text_field).unwrap().as_str(), Some("af b"));
             }
             {
-                let doc = searcher.doc(DocAddress::new(0, 1))?;
-                assert_eq!(doc.get_first(text_field).unwrap().as_text(), Some("a b c"));
+                let doc = searcher.doc::<TantivyDocument>(DocAddress::new(0, 1))?;
+                assert_eq!(doc.get_first(text_field).unwrap().as_str(), Some("a b c"));
             }
             {
-                let doc = searcher.doc(DocAddress::new(0, 2))?;
-                assert_eq!(
-                    doc.get_first(text_field).unwrap().as_text(),
-                    Some("a b c d")
-                );
+                let doc = searcher.doc::<TantivyDocument>(DocAddress::new(0, 2))?;
+                assert_eq!(doc.get_first(text_field).unwrap().as_str(), Some("a b c d"));
             }
             {
-                let doc = searcher.doc(DocAddress::new(0, 3))?;
-                assert_eq!(doc.get_first(text_field).unwrap().as_text(), Some("af b"));
+                let doc = searcher.doc::<TantivyDocument>(DocAddress::new(0, 3))?;
+                assert_eq!(doc.get_first(text_field).unwrap().as_str(), Some("af b"));
             }
             {
-                let doc = searcher.doc(DocAddress::new(0, 4))?;
-                assert_eq!(
-                    doc.get_first(text_field).unwrap().as_text(),
-                    Some("a b c g")
-                );
+                let doc = searcher.doc::<TantivyDocument>(DocAddress::new(0, 4))?;
+                assert_eq!(doc.get_first(text_field).unwrap().as_str(), Some("a b c g"));
             }
 
             {
@@ -1300,10 +1334,10 @@ mod tests {
         let reader = index.reader().unwrap();
         let mut int_val = 0;
         {
-            let mut index_writer = index.writer_for_tests().unwrap();
+            let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
             let index_doc =
                 |index_writer: &mut IndexWriter, doc_facets: &[&str], int_val: &mut u64| {
-                    let mut doc = Document::default();
+                    let mut doc = TantivyDocument::default();
                     for facet in doc_facets {
                         doc.add_facet(facet_field, Facet::from(facet));
                     }
@@ -1384,7 +1418,7 @@ mod tests {
             let segment_ids = index
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
-            let mut index_writer = index.writer_for_tests().unwrap();
+            let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
             index_writer
                 .merge(&segment_ids)
                 .wait()
@@ -1406,7 +1440,7 @@ mod tests {
 
         // Deleting one term
         {
-            let mut index_writer = index.writer_for_tests().unwrap();
+            let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
             let facet = Facet::from_path(vec!["top", "a", "firstdoc"]);
             let facet_term = Term::from_facet(facet_field, &facet);
             index_writer.delete_term(facet_term);
@@ -1431,7 +1465,7 @@ mod tests {
         let mut schema_builder = schema::Schema::builder();
         let int_field = schema_builder.add_u64_field("intvals", INDEXED);
         let index = Index::create_in_ram(schema_builder.build());
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer.add_document(doc!(int_field => 1u64))?;
         index_writer.commit().expect("commit failed");
         index_writer.add_document(doc!(int_field => 1u64))?;
@@ -1460,7 +1494,7 @@ mod tests {
         let reader = index.reader()?;
         {
             let mut index_writer = index.writer_for_tests()?;
-            let mut doc = Document::default();
+            let mut doc = TantivyDocument::default();
             doc.add_u64(int_field, 1);
             index_writer.add_document(doc.clone())?;
             index_writer.commit()?;
@@ -1503,7 +1537,7 @@ mod tests {
         {
             let mut index_writer = index.writer_for_tests()?;
             let index_doc = |index_writer: &mut IndexWriter, int_vals: &[u64]| {
-                let mut doc = Document::default();
+                let mut doc = TantivyDocument::default();
                 for &val in int_vals {
                     doc.add_u64(int_field, val);
                 }
@@ -1566,7 +1600,7 @@ mod tests {
         // Merging the segments
         {
             let segment_ids = index.searchable_segment_ids()?;
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             index_writer.merge(&segment_ids).wait()?;
             index_writer.wait_merging_threads()?;
         }
@@ -1613,7 +1647,7 @@ mod tests {
         writer.set_merge_policy(Box::new(policy));
 
         for i in 0..100 {
-            let mut doc = Document::new();
+            let mut doc = TantivyDocument::new();
             doc.add_f64(field, 42.0);
             doc.add_f64(multi_field, 0.24);
             doc.add_f64(multi_field, 0.27);

src/indexer/merger_sorted_index_test.rs

@@ -4,11 +4,15 @@ mod tests {
     use crate::core::Index;
     use crate::fastfield::AliveBitSet;
     use crate::query::QueryParser;
+    use crate::schema::document::Value;
     use crate::schema::{
         self, BytesOptions, Facet, FacetOptions, IndexRecordOption, NumericOptions,
         TextFieldIndexing, TextOptions,
     };
-    use crate::{DocAddress, DocSet, IndexSettings, IndexSortByField, Order, Postings, Term};
+    use crate::{
+        DocAddress, DocSet, IndexSettings, IndexSortByField, IndexWriter, Order, Postings,
+        TantivyDocument, Term,
+    };
 
     fn create_test_index_posting_list_issue(index_settings: Option<IndexSettings>) -> Index {
         let mut schema_builder = schema::Schema::builder();
@@ -26,7 +30,7 @@ mod tests {
         let index = index_builder.create_in_ram().unwrap();
 
         {
-            let mut index_writer = index.writer_for_tests().unwrap();
+            let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
             index_writer
                 .add_document(doc!(int_field=>3_u64, facet_field=> Facet::from("/crime")))
                 .unwrap();
@@ -45,7 +49,7 @@ mod tests {
             let segment_ids = index
                 .searchable_segment_ids()
                 .expect("Searchable segments failed.");
-            let mut index_writer = index.writer_for_tests().unwrap();
+            let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
             assert!(index_writer.merge(&segment_ids).wait().is_ok());
             assert!(index_writer.wait_merging_threads().is_ok());
         }
@@ -133,7 +137,7 @@ mod tests {
         // Merging the segments
         {
             let segment_ids = index.searchable_segment_ids()?;
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             index_writer.merge(&segment_ids).wait()?;
             index_writer.wait_merging_threads()?;
         }
@@ -272,12 +276,16 @@ mod tests {
             } else {
                 2
             };
-            let doc = searcher.doc(DocAddress::new(0, blubber_pos)).unwrap();
+            let doc = searcher
+                .doc::<TantivyDocument>(DocAddress::new(0, blubber_pos))
+                .unwrap();
             assert_eq!(
-                doc.get_first(my_text_field).unwrap().as_text(),
+                doc.get_first(my_text_field).unwrap().as_str(),
                 Some("blubber")
             );
-            let doc = searcher.doc(DocAddress::new(0, 0)).unwrap();
+            let doc = searcher
+                .doc::<TantivyDocument>(DocAddress::new(0, 0))
+                .unwrap();
             assert_eq!(doc.get_first(int_field).unwrap().as_u64(), Some(1000));
         }
     }
@@ -494,7 +502,7 @@ mod bench_sorted_index_merge {
         let index = index_builder.create_in_ram().unwrap();
 
         {
-            let mut index_writer = index.writer_for_tests().unwrap();
+            let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
             let index_doc = |index_writer: &mut IndexWriter, val: u64| {
                 index_writer.add_document(doc!(int_field=>val)).unwrap();
             };
@@ -535,7 +543,7 @@ mod bench_sorted_index_merge {
     //);
     //(doc_addr.doc_id, reader, u64_reader)
     //});
-    //// add values in order of the new doc_ids
+    /// add values in order of the new doc_ids
     // let mut val = 0;
     // for (doc_id, _reader, field_reader) in sorted_doc_ids {
     // val = field_reader.get_val(doc_id);

src/indexer/mod.rs

@@ -1,23 +1,30 @@
-pub mod delete_queue;
+//! Indexing and merging data.
+//!
+//! Contains code to create and merge segments.
+//! `IndexWriter` is the main entry point for that, which created from
+//! [`Index::writer`](crate::Index::writer).
 
-pub mod doc_id_mapping;
+pub(crate) mod delete_queue;
+pub(crate) mod path_to_unordered_id;
+
+pub(crate) mod doc_id_mapping;
 mod doc_opstamp_mapping;
 mod flat_map_with_buffer;
-pub mod index_writer;
-mod index_writer_status;
+pub(crate) mod index_writer;
+pub(crate) mod index_writer_status;
 mod log_merge_policy;
 mod merge_operation;
-pub mod merge_policy;
-pub mod merger;
+pub(crate) mod merge_policy;
+pub(crate) mod merger;
 mod merger_sorted_index_test;
-pub mod operation;
-pub mod prepared_commit;
+pub(crate) mod operation;
+pub(crate) mod prepared_commit;
 mod segment_entry;
 mod segment_manager;
 mod segment_register;
-pub mod segment_serializer;
-pub mod segment_updater;
-mod segment_writer;
+pub(crate) mod segment_serializer;
+pub(crate) mod segment_updater;
+pub(crate) mod segment_writer;
 mod stamper;
 
 use crossbeam_channel as channel;
@@ -27,10 +34,10 @@ pub use self::index_writer::IndexWriter;
 pub use self::log_merge_policy::LogMergePolicy;
 pub use self::merge_operation::MergeOperation;
 pub use self::merge_policy::{MergeCandidate, MergePolicy, NoMergePolicy};
+pub use self::operation::UserOperation;
 pub use self::prepared_commit::PreparedCommit;
 pub use self::segment_entry::SegmentEntry;
-pub use self::segment_manager::SegmentManager;
-pub use self::segment_serializer::SegmentSerializer;
+pub(crate) use self::segment_serializer::SegmentSerializer;
 pub use self::segment_updater::{merge_filtered_segments, merge_indices};
 pub use self::segment_writer::SegmentWriter;
 use crate::indexer::operation::AddOperation;
@@ -44,25 +51,28 @@ pub type DefaultMergePolicy = LogMergePolicy;
 // - all docs in the operation will happen on the same segment and continuous doc_ids.
 // - all operations in the group are committed at the same time, making the group
 // atomic.
-type AddBatch = SmallVec<[AddOperation; 4]>;
-type AddBatchSender = channel::Sender<AddBatch>;
-type AddBatchReceiver = channel::Receiver<AddBatch>;
+type AddBatch<D> = SmallVec<[AddOperation<D>; 4]>;
+type AddBatchSender<D> = channel::Sender<AddBatch<D>>;
+type AddBatchReceiver<D> = channel::Receiver<AddBatch<D>>;
 
 #[cfg(feature = "mmap")]
 #[cfg(test)]
 mod tests_mmap {
 
-    use crate::collector::Count;
-    use crate::query::QueryParser;
-    use crate::schema::{JsonObjectOptions, Schema, TEXT};
-    use crate::{Index, Term};
+    use crate::aggregation::agg_req::Aggregations;
+    use crate::aggregation::agg_result::AggregationResults;
+    use crate::aggregation::AggregationCollector;
+    use crate::collector::{Count, TopDocs};
+    use crate::query::{AllQuery, QueryParser};
+    use crate::schema::{JsonObjectOptions, Schema, Type, FAST, INDEXED, STORED, TEXT};
+    use crate::{FieldMetadata, Index, IndexWriter, Term};
 
     #[test]
     fn test_advance_delete_bug() -> crate::Result<()> {
         let mut schema_builder = Schema::builder();
         let text_field = schema_builder.add_text_field("text", TEXT);
         let index = Index::create_from_tempdir(schema_builder.build())?;
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         // there must be one deleted document in the segment
         index_writer.add_document(doc!(text_field=>"b"))?;
         index_writer.delete_term(Term::from_field_text(text_field, "b"));
@@ -79,7 +89,7 @@ mod tests_mmap {
         let mut schema_builder = Schema::builder();
         let json_field = schema_builder.add_json_field("json", TEXT);
         let index = Index::create_in_ram(schema_builder.build());
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         let json = serde_json::json!({"k8s.container.name": "prometheus", "val": "hello"});
         index_writer.add_document(doc!(json_field=>json)).unwrap();
         index_writer.commit().unwrap();
@@ -103,6 +113,35 @@ mod tests_mmap {
         }
     }
 
+    #[test]
+    fn test_json_field_number() {
+        // this test was added specifically to reach some cases related to using json fields, with
+        // frequency enabled, to store integers, with enough documents containing a single integer
+        // that the posting list can be bitpacked.
+        let mut schema_builder = Schema::builder();
+
+        let json_field = schema_builder.add_json_field("json", TEXT);
+        let index = Index::create_in_ram(schema_builder.build());
+        let mut index_writer = index.writer_for_tests().unwrap();
+        for _ in 0..256 {
+            let json = serde_json::json!({"somekey": 1u64, "otherkey": -2i64});
+            index_writer.add_document(doc!(json_field=>json)).unwrap();
+
+            let json = serde_json::json!({"somekey": "1str", "otherkey": "2str"});
+            index_writer.add_document(doc!(json_field=>json)).unwrap();
+        }
+        index_writer.commit().unwrap();
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+        assert_eq!(searcher.num_docs(), 512);
+        let parse_query = QueryParser::for_index(&index, Vec::new());
+        {
+            let query = parse_query.parse_query(r"json.somekey:1").unwrap();
+            let num_docs = searcher.search(&query, &Count).unwrap();
+            assert_eq!(num_docs, 256);
+        }
+    }
+
     #[test]
     fn test_json_field_expand_dots_enabled_dot_escape_not_required() {
         let mut schema_builder = Schema::builder();
@@ -110,7 +149,7 @@ mod tests_mmap {
             JsonObjectOptions::from(TEXT).set_expand_dots_enabled();
         let json_field = schema_builder.add_json_field("json", json_options);
         let index = Index::create_in_ram(schema_builder.build());
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         let json = serde_json::json!({"k8s.container.name": "prometheus", "val": "hello"});
         index_writer.add_document(doc!(json_field=>json)).unwrap();
         index_writer.commit().unwrap();
@@ -133,4 +172,275 @@ mod tests_mmap {
             assert_eq!(num_docs, 1);
         }
     }
+
+    #[test]
+    fn test_json_field_list_fields() {
+        let mut schema_builder = Schema::builder();
+        let json_options: JsonObjectOptions = JsonObjectOptions::from(TEXT);
+        let json_field = schema_builder.add_json_field("json", json_options);
+        let index = Index::create_in_ram(schema_builder.build());
+        let mut index_writer = index.writer_for_tests().unwrap();
+        let json = serde_json::json!({"k8s.container.name": "prometheus", "val": "hello", "sub": {"a": 1, "b": 2}});
+        index_writer.add_document(doc!(json_field=>json)).unwrap();
+        let json = serde_json::json!({"k8s.container.name": "prometheus", "val": "hello", "suber": {"a": 1, "b": 2}});
+        index_writer.add_document(doc!(json_field=>json)).unwrap();
+        let json = serde_json::json!({"k8s.container.name": "prometheus", "val": "hello", "suber": {"a": "mixed", "b": 2}});
+        index_writer.add_document(doc!(json_field=>json)).unwrap();
+        index_writer.commit().unwrap();
+        let reader = index.reader().unwrap();
+
+        let searcher = reader.searcher();
+        assert_eq!(searcher.num_docs(), 3);
+
+        let reader = &searcher.segment_readers()[0];
+        let inverted_index = reader.inverted_index(json_field).unwrap();
+        assert_eq!(
+            inverted_index.list_encoded_fields().unwrap(),
+            [
+                ("k8s.container.name".to_string(), Type::Str),
+                ("sub\u{1}a".to_string(), Type::I64),
+                ("sub\u{1}b".to_string(), Type::I64),
+                ("suber\u{1}a".to_string(), Type::I64),
+                ("suber\u{1}a".to_string(), Type::Str),
+                ("suber\u{1}b".to_string(), Type::I64),
+                ("val".to_string(), Type::Str),
+            ]
+        );
+    }
+
+    #[test]
+    fn test_json_fields_metadata_expanded_dots_one_segment() {
+        test_json_fields_metadata(true, true);
+    }
+    #[test]
+    fn test_json_fields_metadata_expanded_dots_multi_segment() {
+        test_json_fields_metadata(true, false);
+    }
+    #[test]
+    fn test_json_fields_metadata_no_expanded_dots_one_segment() {
+        test_json_fields_metadata(false, true);
+    }
+    #[test]
+    fn test_json_fields_metadata_no_expanded_dots_multi_segment() {
+        test_json_fields_metadata(false, false);
+    }
+
+    fn test_json_fields_metadata(expanded_dots: bool, one_segment: bool) {
+        use pretty_assertions::assert_eq;
+        let mut schema_builder = Schema::builder();
+        let json_options: JsonObjectOptions =
+            JsonObjectOptions::from(TEXT).set_fast(None).set_stored();
+        let json_options = if expanded_dots {
+            json_options.set_expand_dots_enabled()
+        } else {
+            json_options
+        };
+        schema_builder.add_json_field("json.confusing", json_options.clone());
+        let json_field = schema_builder.add_json_field("json.shadow", json_options.clone());
+        let json_field2 = schema_builder.add_json_field("json", json_options.clone());
+        schema_builder.add_json_field("empty_json", json_options);
+        let number_field = schema_builder.add_u64_field("numbers", FAST);
+        schema_builder.add_u64_field("empty", FAST | INDEXED | STORED);
+        let index = Index::create_in_ram(schema_builder.build());
+        let mut index_writer = index.writer_for_tests().unwrap();
+        let json =
+            serde_json::json!({"k8s.container.name": "a", "val": "a", "sub": {"a": 1, "b": 1}});
+        index_writer.add_document(doc!(json_field=>json)).unwrap();
+        let json =
+            serde_json::json!({"k8s.container.name": "a", "val": "a", "suber": {"a": 1, "b": 1}});
+        if !one_segment {
+            index_writer.commit().unwrap();
+        }
+        index_writer.add_document(doc!(json_field=>json)).unwrap();
+        let json = serde_json::json!({"k8s.container.name": "a", "k8s.container.name": "a", "val": "a", "suber": {"a": "a", "b": 1}});
+        index_writer
+            .add_document(doc!(number_field => 50u64, json_field=>json, json_field2=>json!({"shadow": {"val": "a"}})))
+            .unwrap();
+        index_writer.commit().unwrap();
+        let reader = index.reader().unwrap();
+
+        let searcher = reader.searcher();
+        assert_eq!(searcher.num_docs(), 3);
+
+        let fields_metadata = index.fields_metadata().unwrap();
+        assert_eq!(
+            fields_metadata,
+            [
+                FieldMetadata {
+                    field_name: "empty".to_string(),
+                    indexed: true,
+                    stored: true,
+                    fast: true,
+                    typ: Type::U64
+                },
+                FieldMetadata {
+                    field_name: if expanded_dots {
+                        "json.shadow.k8s.container.name".to_string()
+                    } else {
+                        "json.shadow.k8s\\.container\\.name".to_string()
+                    },
+                    indexed: true,
+                    stored: true,
+                    fast: true,
+                    typ: Type::Str
+                },
+                FieldMetadata {
+                    field_name: "json.shadow.sub.a".to_string(),
+                    indexed: true,
+                    stored: true,
+                    fast: true,
+                    typ: Type::I64
+                },
+                FieldMetadata {
+                    field_name: "json.shadow.sub.b".to_string(),
+                    indexed: true,
+                    stored: true,
+                    fast: true,
+                    typ: Type::I64
+                },
+                FieldMetadata {
+                    field_name: "json.shadow.suber.a".to_string(),
+                    indexed: true,
+                    stored: true,
+                    fast: true,
+                    typ: Type::I64
+                },
+                FieldMetadata {
+                    field_name: "json.shadow.suber.a".to_string(),
+                    indexed: true,
+                    stored: true,
+                    fast: true,
+                    typ: Type::Str
+                },
+                FieldMetadata {
+                    field_name: "json.shadow.suber.b".to_string(),
+                    indexed: true,
+                    stored: true,
+                    fast: true,
+                    typ: Type::I64
+                },
+                FieldMetadata {
+                    field_name: "json.shadow.val".to_string(),
+                    indexed: true,
+                    stored: true,
+                    fast: true,
+                    typ: Type::Str
+                },
+                FieldMetadata {
+                    field_name: "numbers".to_string(),
+                    indexed: false,
+                    stored: false,
+                    fast: true,
+                    typ: Type::U64
+                }
+            ]
+        );
+        let query_parser = QueryParser::for_index(&index, vec![]);
+        // Test if returned field name can be queried
+        for indexed_field in fields_metadata.iter().filter(|meta| meta.indexed) {
+            let val = if indexed_field.typ == Type::Str {
+                "a"
+            } else {
+                "1"
+            };
+            let query_str = &format!("{}:{}", indexed_field.field_name, val);
+            let query = query_parser.parse_query(query_str).unwrap();
+            let count_docs = searcher.search(&*query, &TopDocs::with_limit(2)).unwrap();
+            if indexed_field.field_name.contains("empty") || indexed_field.typ == Type::Json {
+                assert_eq!(count_docs.len(), 0);
+            } else {
+                assert!(!count_docs.is_empty(), "{}", indexed_field.field_name);
+            }
+        }
+        // Test if returned field name can be used for aggregation
+        for fast_field in fields_metadata.iter().filter(|meta| meta.fast) {
+            let agg_req_str = json!(
+            {
+              "termagg": {
+                "terms": {
+                  "field": fast_field.field_name,
+                }
+              }
+            });
+
+            let agg_req: Aggregations = serde_json::from_value(agg_req_str).unwrap();
+            let collector = AggregationCollector::from_aggs(agg_req, Default::default());
+            let agg_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
+            let res = serde_json::to_value(agg_res).unwrap();
+            if !fast_field.field_name.contains("empty") && fast_field.typ != Type::Json {
+                assert!(
+                    !res["termagg"]["buckets"].as_array().unwrap().is_empty(),
+                    "{}",
+                    fast_field.field_name
+                );
+            }
+        }
+    }
+
+    #[test]
+    fn test_json_field_shadowing_field_name_bug() {
+        /// This test is only there to display a bug on addressing a field if it gets shadowed
+        /// The issues only occurs if the field name that shadows contains a dot.
+        ///
+        /// Happens independently of the `expand_dots` option. Since that option does not
+        /// affect the field name itself.
+        use pretty_assertions::assert_eq;
+        let mut schema_builder = Schema::builder();
+        let json_options: JsonObjectOptions =
+            JsonObjectOptions::from(TEXT).set_fast(None).set_stored();
+        // let json_options = json_options.set_expand_dots_enabled();
+        let json_field_shadow = schema_builder.add_json_field("json.shadow", json_options.clone());
+        let json_field = schema_builder.add_json_field("json", json_options.clone());
+        let index = Index::create_in_ram(schema_builder.build());
+        let mut index_writer = index.writer_for_tests().unwrap();
+        index_writer
+            .add_document(
+                doc!(json_field_shadow=>json!({"val": "b"}), json_field=>json!({"shadow": {"val": "a"}})),
+            )
+            .unwrap();
+        index_writer.commit().unwrap();
+        let reader = index.reader().unwrap();
+
+        let searcher = reader.searcher();
+
+        let fields_and_vals = vec![
+            // Only way to address or it gets shadowed by `json.shadow` field
+            ("json.shadow\u{1}val".to_string(), "a"), // Succeeds
+            //("json.shadow.val".to_string(), "a"),   // Fails
+            ("json.shadow.val".to_string(), "b"), // Succeeds
+        ];
+
+        let query_parser = QueryParser::for_index(&index, vec![]);
+        // Test if field name can be queried
+        for (indexed_field, val) in fields_and_vals.iter() {
+            let query_str = &format!("{}:{}", indexed_field, val);
+            let query = query_parser.parse_query(query_str).unwrap();
+            let count_docs = searcher.search(&*query, &TopDocs::with_limit(2)).unwrap();
+            assert!(!count_docs.is_empty(), "{}:{}", indexed_field, val);
+        }
+        // Test if field name can be used for aggregation
+        for (field_name, val) in fields_and_vals.iter() {
+            let agg_req_str = json!(
+            {
+              "termagg": {
+                "terms": {
+                  "field": field_name,
+                }
+              }
+            });
+
+            let agg_req: Aggregations = serde_json::from_value(agg_req_str).unwrap();
+            let collector = AggregationCollector::from_aggs(agg_req, Default::default());
+            let agg_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
+            let res = serde_json::to_value(agg_res).unwrap();
+            assert_eq!(
+                res["termagg"]["buckets"].as_array().unwrap()[0]["key"]
+                    .as_str()
+                    .unwrap(),
+                *val,
+                "{}",
+                field_name
+            );
+        }
+    }
 }

src/indexer/operation.rs

@@ -1,5 +1,6 @@
 use crate::query::Weight;
-use crate::schema::{Document, Term};
+use crate::schema::document::Document;
+use crate::schema::{TantivyDocument, Term};
 use crate::Opstamp;
 
 /// Timestamped Delete operation.
@@ -10,16 +11,16 @@ pub struct DeleteOperation {
 
 /// Timestamped Add operation.
 #[derive(Eq, PartialEq, Debug)]
-pub struct AddOperation {
+pub struct AddOperation<D: Document = TantivyDocument> {
     pub opstamp: Opstamp,
-    pub document: Document,
+    pub document: D,
 }
 
 /// UserOperation is an enum type that encapsulates other operation types.
 #[derive(Eq, PartialEq, Debug)]
-pub enum UserOperation {
+pub enum UserOperation<D: Document = TantivyDocument> {
     /// Add operation
-    Add(Document),
+    Add(D),
     /// Delete operation
     Delete(Term),
 }

src/indexer/path_to_unordered_id.rs

@@ -0,0 +1,92 @@
+use fnv::FnvHashMap;
+
+/// `Field` is represented by an unsigned 32-bit integer type.
+/// The schema holds the mapping between field names and `Field` objects.
+#[derive(Copy, Default, Clone, Debug, PartialEq, PartialOrd, Eq, Ord, Hash)]
+pub struct OrderedPathId(u32);
+
+impl OrderedPathId {
+    /// Create a new field object for the given PathId.
+    pub const fn from_ordered_id(field_id: u32) -> OrderedPathId {
+        OrderedPathId(field_id)
+    }
+
+    /// Returns a u32 identifying uniquely a path within a schema.
+    pub const fn path_id(self) -> u32 {
+        self.0
+    }
+}
+impl From<u32> for OrderedPathId {
+    fn from(id: u32) -> Self {
+        Self(id)
+    }
+}
+
+#[derive(Default)]
+pub(crate) struct PathToUnorderedId {
+    map: FnvHashMap<String, u32>,
+}
+
+impl PathToUnorderedId {
+    #[inline]
+    pub(crate) fn get_or_allocate_unordered_id(&mut self, path: &str) -> u32 {
+        if let Some(id) = self.map.get(path) {
+            return *id;
+        }
+        self.insert_new_path(path)
+    }
+    #[cold]
+    fn insert_new_path(&mut self, path: &str) -> u32 {
+        let next_id = self.map.len() as u32;
+        self.map.insert(path.to_string(), next_id);
+        next_id
+    }
+
+    /// Retuns ids which reflect the lexical order of the paths.
+    ///
+    /// The returned vec can be indexed with the unordered id to get the ordered id.
+    pub(crate) fn unordered_id_to_ordered_id(&self) -> Vec<OrderedPathId> {
+        let mut sorted_ids: Vec<(&str, &u32)> =
+            self.map.iter().map(|(k, v)| (k.as_str(), v)).collect();
+        sorted_ids.sort_unstable_by_key(|(path, _)| *path);
+        let mut result = vec![OrderedPathId::default(); sorted_ids.len()];
+        for (ordered, unordered) in sorted_ids.iter().map(|(_k, v)| v).enumerate() {
+            result[**unordered as usize] = OrderedPathId::from_ordered_id(ordered as u32);
+        }
+        result
+    }
+
+    /// Retuns the paths so they can be queried by the ordered id (which is the index).
+    pub(crate) fn ordered_id_to_path(&self) -> Vec<&str> {
+        let mut paths = self.map.keys().map(String::as_str).collect::<Vec<_>>();
+        paths.sort_unstable();
+        paths
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn path_to_unordered_test() {
+        let mut path_to_id = PathToUnorderedId::default();
+        let terms = vec!["b", "a", "b", "c"];
+        let ids = terms
+            .iter()
+            .map(|term| path_to_id.get_or_allocate_unordered_id(term))
+            .collect::<Vec<u32>>();
+        assert_eq!(ids, vec![0, 1, 0, 2]);
+        let ordered_ids = ids
+            .iter()
+            .map(|id| path_to_id.unordered_id_to_ordered_id()[*id as usize])
+            .collect::<Vec<OrderedPathId>>();
+        assert_eq!(ordered_ids, vec![1.into(), 0.into(), 1.into(), 2.into()]);
+        // Fetch terms
+        let terms_fetched = ordered_ids
+            .iter()
+            .map(|id| path_to_id.ordered_id_to_path()[id.path_id() as usize])
+            .collect::<Vec<&str>>();
+        assert_eq!(terms_fetched, terms);
+    }
+}

src/indexer/prepared_commit.rs

@@ -1,16 +1,17 @@
 use super::IndexWriter;
-use crate::{FutureResult, Opstamp};
+use crate::schema::document::Document;
+use crate::{FutureResult, Opstamp, TantivyDocument};
 
 /// A prepared commit
-pub struct PreparedCommit<'a> {
-    index_writer: &'a mut IndexWriter,
+pub struct PreparedCommit<'a, D: Document = TantivyDocument> {
+    index_writer: &'a mut IndexWriter<D>,
     payload: Option<String>,
     opstamp: Opstamp,
 }
 
-impl<'a> PreparedCommit<'a> {
-    pub(crate) fn new(index_writer: &'a mut IndexWriter, opstamp: Opstamp) -> PreparedCommit<'_> {
-        PreparedCommit {
+impl<'a, D: Document> PreparedCommit<'a, D> {
+    pub(crate) fn new(index_writer: &'a mut IndexWriter<D>, opstamp: Opstamp) -> Self {
+        Self {
             index_writer,
             payload: None,
             opstamp,

src/indexer/segment_writer.rs

@@ -1,4 +1,5 @@
 use columnar::MonotonicallyMappableToU64;
+use common::JsonPathWriter;
 use itertools::Itertools;
 use tokenizer_api::BoxTokenStream;
 
@@ -13,10 +14,11 @@ use crate::postings::{
     compute_table_memory_size, serialize_postings, IndexingContext, IndexingPosition,
     PerFieldPostingsWriter, PostingsWriter,
 };
-use crate::schema::{FieldEntry, FieldType, Schema, Term, Value, DATE_TIME_PRECISION_INDEXED};
+use crate::schema::document::{Document, ReferenceValue, Value};
+use crate::schema::{FieldEntry, FieldType, Schema, Term, DATE_TIME_PRECISION_INDEXED};
 use crate::store::{StoreReader, StoreWriter};
 use crate::tokenizer::{FacetTokenizer, PreTokenizedStream, TextAnalyzer, Tokenizer};
-use crate::{DocId, Document, Opstamp, SegmentComponent, TantivyError};
+use crate::{DocId, Opstamp, SegmentComponent, TantivyError};
 
 /// Computes the initial size of the hash table.
 ///
@@ -65,6 +67,7 @@ pub struct SegmentWriter {
     pub(crate) segment_serializer: SegmentSerializer,
     pub(crate) fast_field_writers: FastFieldsWriter,
     pub(crate) fieldnorms_writer: FieldNormsWriter,
+    pub(crate) json_path_writer: JsonPathWriter,
     pub(crate) doc_opstamps: Vec<Opstamp>,
     per_field_text_analyzers: Vec<TextAnalyzer>,
     term_buffer: Term,
@@ -81,10 +84,7 @@ impl SegmentWriter {
     /// the flushing behavior as a memory limit.
     /// - segment: The segment being written
     /// - schema
-    pub fn for_segment(
-        memory_budget_in_bytes: usize,
-        segment: Segment,
-    ) -> crate::Result<SegmentWriter> {
+    pub fn for_segment(memory_budget_in_bytes: usize, segment: Segment) -> crate::Result<Self> {
         let schema = segment.schema();
         let tokenizer_manager = segment.index().tokenizers().clone();
         let tokenizer_manager_fast_field = segment.index().fast_field_tokenizer().clone();
@@ -113,11 +113,12 @@ impl SegmentWriter {
                 })
             })
             .collect::<Result<Vec<_>, _>>()?;
-        Ok(SegmentWriter {
+        Ok(Self {
             max_doc: 0,
             ctx: IndexingContext::new(table_size),
             per_field_postings_writers,
             fieldnorms_writer: FieldNormsWriter::for_schema(&schema),
+            json_path_writer: JsonPathWriter::default(),
             segment_serializer,
             fast_field_writers: FastFieldsWriter::from_schema_and_tokenizer_manager(
                 &schema,
@@ -146,6 +147,7 @@ impl SegmentWriter {
             .map(|sort_by_field| get_doc_id_mapping_from_field(sort_by_field, &self))
             .transpose()?;
         remap_and_write(
+            self.schema,
             &self.per_field_postings_writers,
             self.ctx,
             self.fast_field_writers,
@@ -157,6 +159,8 @@ impl SegmentWriter {
         Ok(doc_opstamps)
     }
 
+    /// Returns an estimation of the current memory usage of the segment writer.
+    /// If the mem usage exceeds the `memory_budget`, the segment be serialized.
     pub fn mem_usage(&self) -> usize {
         self.ctx.mem_usage()
             + self.fieldnorms_writer.mem_usage()
@@ -164,18 +168,21 @@ impl SegmentWriter {
             + self.segment_serializer.mem_usage()
     }
 
-    fn index_document(&mut self, doc: &Document) -> crate::Result<()> {
+    fn index_document<D: Document>(&mut self, doc: &D) -> crate::Result<()> {
         let doc_id = self.max_doc;
+
+        // TODO: Can this be optimised a bit?
         let vals_grouped_by_field = doc
-            .field_values()
-            .iter()
-            .sorted_by_key(|el| el.field())
-            .group_by(|el| el.field());
+            .iter_fields_and_values()
+            .sorted_by_key(|(field, _)| *field)
+            .group_by(|(field, _)| *field);
+
         for (field, field_values) in &vals_grouped_by_field {
-            let values = field_values.map(|field_value| field_value.value());
+            let values = field_values.map(|el| el.1);
+
             let field_entry = self.schema.get_field_entry(field);
             let make_schema_error = || {
-                crate::TantivyError::SchemaError(format!(
+                TantivyError::SchemaError(format!(
                     "Expected a {:?} for field {:?}",
                     field_entry.field_type().value_type(),
                     field_entry.name()
@@ -193,7 +200,10 @@ impl SegmentWriter {
             match field_entry.field_type() {
                 FieldType::Facet(_) => {
                     let mut facet_tokenizer = FacetTokenizer::default(); // this can be global
-                    for value in values {
+                    for value_access in values {
+                        // Used to help with linting and type checking.
+                        let value = value_access as D::Value<'_>;
+
                         let facet = value.as_facet().ok_or_else(make_schema_error)?;
                         let facet_str = facet.encoded_str();
                         let mut facet_tokenizer = facet_tokenizer.token_stream(facet_str);
@@ -209,19 +219,18 @@ impl SegmentWriter {
                 }
                 FieldType::Str(_) => {
                     let mut indexing_position = IndexingPosition::default();
-                    for value in values {
-                        let mut token_stream = match value {
-                            Value::PreTokStr(tok_str) => {
-                                BoxTokenStream::new(PreTokenizedStream::from(tok_str.clone()))
-                            }
-                            Value::Str(ref text) => {
-                                let text_analyzer =
-                                    &mut self.per_field_text_analyzers[field.field_id() as usize];
-                                text_analyzer.token_stream(text)
-                            }
-                            _ => {
-                                continue;
-                            }
+                    for value_access in values {
+                        // Used to help with linting and type checking.
+                        let value = value_access as D::Value<'_>;
+
+                        let mut token_stream = if let Some(text) = value.as_str() {
+                            let text_analyzer =
+                                &mut self.per_field_text_analyzers[field.field_id() as usize];
+                            text_analyzer.token_stream(text)
+                        } else if let Some(tok_str) = value.as_pre_tokenized_text() {
+                            BoxTokenStream::new(PreTokenizedStream::from(tok_str.clone()))
+                        } else {
+                            continue;
                         };
 
                         assert!(term_buffer.is_empty());
@@ -240,7 +249,10 @@ impl SegmentWriter {
                 }
                 FieldType::U64(_) => {
                     let mut num_vals = 0;
-                    for value in values {
+                    for value_access in values {
+                        // Used to help with linting and type checking.
+                        let value = value_access as D::Value<'_>;
+
                         num_vals += 1;
                         let u64_val = value.as_u64().ok_or_else(make_schema_error)?;
                         term_buffer.set_u64(u64_val);
@@ -252,9 +264,13 @@ impl SegmentWriter {
                 }
                 FieldType::Date(_) => {
                     let mut num_vals = 0;
-                    for value in values {
+                    for value_access in values {
+                        // Used to help with linting and type checking.
+                        let value_access = value_access as D::Value<'_>;
+                        let value = value_access.as_value();
+
                         num_vals += 1;
-                        let date_val = value.as_date().ok_or_else(make_schema_error)?;
+                        let date_val = value.as_datetime().ok_or_else(make_schema_error)?;
                         term_buffer
                             .set_u64(date_val.truncate(DATE_TIME_PRECISION_INDEXED).to_u64());
                         postings_writer.subscribe(doc_id, 0u32, term_buffer, ctx);
@@ -265,7 +281,10 @@ impl SegmentWriter {
                 }
                 FieldType::I64(_) => {
                     let mut num_vals = 0;
-                    for value in values {
+                    for value_access in values {
+                        // Used to help with linting and type checking.
+                        let value = value_access as D::Value<'_>;
+
                         num_vals += 1;
                         let i64_val = value.as_i64().ok_or_else(make_schema_error)?;
                         term_buffer.set_i64(i64_val);
@@ -277,7 +296,10 @@ impl SegmentWriter {
                 }
                 FieldType::F64(_) => {
                     let mut num_vals = 0;
-                    for value in values {
+                    for value_access in values {
+                        // Used to help with linting and type checking.
+                        let value = value_access as D::Value<'_>;
+
                         num_vals += 1;
                         let f64_val = value.as_f64().ok_or_else(make_schema_error)?;
                         term_buffer.set_f64(f64_val);
@@ -289,7 +311,10 @@ impl SegmentWriter {
                 }
                 FieldType::Bool(_) => {
                     let mut num_vals = 0;
-                    for value in values {
+                    for value_access in values {
+                        // Used to help with linting and type checking.
+                        let value = value_access as D::Value<'_>;
+
                         num_vals += 1;
                         let bool_val = value.as_bool().ok_or_else(make_schema_error)?;
                         term_buffer.set_bool(bool_val);
@@ -301,7 +326,10 @@ impl SegmentWriter {
                 }
                 FieldType::Bytes(_) => {
                     let mut num_vals = 0;
-                    for value in values {
+                    for value_access in values {
+                        // Used to help with linting and type checking.
+                        let value = value_access as D::Value<'_>;
+
                         num_vals += 1;
                         let bytes = value.as_bytes().ok_or_else(make_schema_error)?;
                         term_buffer.set_bytes(bytes);
@@ -314,21 +342,33 @@ impl SegmentWriter {
                 FieldType::JsonObject(json_options) => {
                     let text_analyzer =
                         &mut self.per_field_text_analyzers[field.field_id() as usize];
-                    let json_values_it =
-                        values.map(|value| value.as_json().ok_or_else(make_schema_error));
-                    index_json_values(
+                    let json_values_it = values.map(|value_access| {
+                        // Used to help with linting and type checking.
+                        let value_access = value_access as D::Value<'_>;
+                        let value = value_access.as_value();
+
+                        match value {
+                            ReferenceValue::Object(object_iter) => Ok(object_iter),
+                            _ => Err(make_schema_error()),
+                        }
+                    });
+                    index_json_values::<D::Value<'_>>(
                         doc_id,
                         json_values_it,
                         text_analyzer,
                         json_options.is_expand_dots_enabled(),
                         term_buffer,
                         postings_writer,
+                        &mut self.json_path_writer,
                         ctx,
                     )?;
                 }
                 FieldType::IpAddr(_) => {
                     let mut num_vals = 0;
-                    for value in values {
+                    for value_access in values {
+                        // Used to help with linting and type checking.
+                        let value = value_access as D::Value<'_>;
+
                         num_vals += 1;
                         let ip_addr = value.as_ip_addr().ok_or_else(make_schema_error)?;
                         term_buffer.set_ip_addr(ip_addr);
@@ -346,7 +386,10 @@ impl SegmentWriter {
     /// Indexes a new document
     ///
     /// As a user, you should rather use `IndexWriter`'s add_document.
-    pub fn add_document(&mut self, add_operation: AddOperation) -> crate::Result<()> {
+    pub fn add_document<D: Document>(
+        &mut self,
+        add_operation: AddOperation<D>,
+    ) -> crate::Result<()> {
         let AddOperation { document, opstamp } = add_operation;
         self.doc_opstamps.push(opstamp);
         self.fast_field_writers.add_document(&document)?;
@@ -384,6 +427,7 @@ impl SegmentWriter {
 ///
 /// `doc_id_map` is used to map to the new doc_id order.
 fn remap_and_write(
+    schema: Schema,
     per_field_postings_writers: &PerFieldPostingsWriter,
     ctx: IndexingContext,
     fast_field_writers: FastFieldsWriter,
@@ -401,6 +445,7 @@ fn remap_and_write(
     let fieldnorm_readers = FieldNormReaders::open(fieldnorm_data)?;
     serialize_postings(
         ctx,
+        schema,
         per_field_postings_writers,
         fieldnorm_readers,
         doc_id_map,
@@ -445,32 +490,37 @@ fn remap_and_write(
 
 #[cfg(test)]
 mod tests {
+    use std::collections::BTreeMap;
     use std::path::{Path, PathBuf};
 
     use tempfile::TempDir;
 
-    use super::compute_initial_table_size;
-    use crate::collector::Count;
+    use crate::collector::{Count, TopDocs};
     use crate::core::json_utils::JsonTermWriter;
     use crate::directory::RamDirectory;
     use crate::postings::TermInfo;
-    use crate::query::PhraseQuery;
+    use crate::query::{PhraseQuery, QueryParser};
+    use crate::schema::document::Value;
     use crate::schema::{
-        IndexRecordOption, Schema, TextFieldIndexing, TextOptions, Type, STORED, STRING, TEXT,
+        Document, IndexRecordOption, Schema, TextFieldIndexing, TextOptions, Type, STORED, STRING,
+        TEXT,
     };
     use crate::store::{Compressor, StoreReader, StoreWriter};
     use crate::time::format_description::well_known::Rfc3339;
     use crate::time::OffsetDateTime;
     use crate::tokenizer::{PreTokenizedString, Token};
     use crate::{
-        DateTime, Directory, DocAddress, DocSet, Document, Index, Postings, Term, TERMINATED,
+        DateTime, Directory, DocAddress, DocSet, Index, IndexWriter, Postings, TantivyDocument,
+        Term, TERMINATED,
     };
 
     #[test]
+    #[cfg(not(feature = "compare_hash_only"))]
     fn test_hashmap_size() {
-        assert_eq!(compute_initial_table_size(100_000).unwrap(), 1 << 11);
-        assert_eq!(compute_initial_table_size(1_000_000).unwrap(), 1 << 14);
-        assert_eq!(compute_initial_table_size(15_000_000).unwrap(), 1 << 18);
+        use super::compute_initial_table_size;
+        assert_eq!(compute_initial_table_size(100_000).unwrap(), 1 << 12);
+        assert_eq!(compute_initial_table_size(1_000_000).unwrap(), 1 << 15);
+        assert_eq!(compute_initial_table_size(15_000_000).unwrap(), 1 << 19);
         assert_eq!(compute_initial_table_size(1_000_000_000).unwrap(), 1 << 19);
         assert_eq!(compute_initial_table_size(4_000_000_000).unwrap(), 1 << 19);
     }
@@ -480,7 +530,7 @@ mod tests {
         let mut schema_builder = Schema::builder();
         let text_field = schema_builder.add_text_field("title", TEXT | STORED);
         let schema = schema_builder.build();
-        let mut doc = Document::default();
+        let mut doc = TantivyDocument::default();
         let pre_tokenized_text = PreTokenizedString {
             text: String::from("A"),
             tokens: vec![Token {
@@ -504,11 +554,48 @@ mod tests {
         store_writer.close().unwrap();
 
         let reader = StoreReader::open(directory.open_read(path).unwrap(), 0).unwrap();
-        let doc = reader.get(0).unwrap();
+        let doc = reader.get::<TantivyDocument>(0).unwrap();
 
         assert_eq!(doc.field_values().len(), 2);
-        assert_eq!(doc.field_values()[0].value().as_text(), Some("A"));
-        assert_eq!(doc.field_values()[1].value().as_text(), Some("title"));
+        assert_eq!(doc.field_values()[0].value().as_str(), Some("A"));
+        assert_eq!(doc.field_values()[1].value().as_str(), Some("title"));
+    }
+    #[test]
+    fn test_simple_json_indexing() {
+        let mut schema_builder = Schema::builder();
+        let json_field = schema_builder.add_json_field("json", STORED | STRING);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema.clone());
+        let mut writer = index.writer_for_tests().unwrap();
+        writer
+            .add_document(doc!(json_field=>json!({"my_field": "b"})))
+            .unwrap();
+        writer
+            .add_document(doc!(json_field=>json!({"my_field": "a"})))
+            .unwrap();
+        writer
+            .add_document(doc!(json_field=>json!({"my_field": "b"})))
+            .unwrap();
+        writer.commit().unwrap();
+
+        let query_parser = QueryParser::for_index(&index, vec![json_field]);
+        let text_query = query_parser.parse_query("my_field:a").unwrap();
+        let score_docs: Vec<(_, DocAddress)> = index
+            .reader()
+            .unwrap()
+            .searcher()
+            .search(&text_query, &TopDocs::with_limit(4))
+            .unwrap();
+        assert_eq!(score_docs.len(), 1);
+
+        let text_query = query_parser.parse_query("my_field:b").unwrap();
+        let score_docs: Vec<(_, DocAddress)> = index
+            .reader()
+            .unwrap()
+            .searcher()
+            .search(&text_query, &TopDocs::with_limit(4))
+            .unwrap();
+        assert_eq!(score_docs.len(), 2);
     }
 
     #[test]
@@ -539,13 +626,13 @@ mod tests {
         let reader = index.reader().unwrap();
         let searcher = reader.searcher();
         let doc = searcher
-            .doc(DocAddress {
+            .doc::<TantivyDocument>(DocAddress {
                 segment_ord: 0u32,
                 doc_id: 0u32,
             })
             .unwrap();
         let serdeser_json_val = serde_json::from_str::<serde_json::Map<String, serde_json::Value>>(
-            &schema.to_json(&doc),
+            &doc.to_json(&schema),
         )
         .unwrap()
         .get("json")
@@ -675,10 +762,10 @@ mod tests {
         let mut schema_builder = Schema::builder();
         let json_field = schema_builder.add_json_field("json", STORED | TEXT);
         let schema = schema_builder.build();
-        let mut doc = Document::default();
-        let json_val: serde_json::Map<String, serde_json::Value> =
+        let mut doc = TantivyDocument::default();
+        let json_val: BTreeMap<String, crate::schema::OwnedValue> =
             serde_json::from_str(r#"{"mykey": "repeated token token"}"#).unwrap();
-        doc.add_json_object(json_field, json_val);
+        doc.add_object(json_field, json_val);
         let index = Index::create_in_ram(schema);
         let mut writer = index.writer_for_tests().unwrap();
         writer.add_document(doc).unwrap();
@@ -802,11 +889,10 @@ mod tests {
         let mut schema_builder = Schema::builder();
         let text = schema_builder.add_text_field("text", TEXT);
         let schema = schema_builder.build();
-        let doc = schema
-            .parse_document(r#"{"text": [ "bbb", "aaa", "", "aaa"]}"#)
+        let doc = TantivyDocument::parse_json(&schema, r#"{"text": [ "bbb", "aaa", "", "aaa"]}"#)
             .unwrap();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer.add_document(doc).unwrap();
         // On debug this did panic on the underflow
         index_writer.commit().unwrap();
@@ -831,7 +917,7 @@ mod tests {
         let mut schema_builder = Schema::builder();
         let text = schema_builder.add_text_field("text", TEXT);
         let schema = schema_builder.build();
-        let mut doc = Document::default();
+        let mut doc = TantivyDocument::default();
         // This is a bit of a contrived example.
         let tokens = PreTokenizedString {
             text: "roller-coaster".to_string(),
@@ -846,7 +932,7 @@ mod tests {
         doc.add_pre_tokenized_text(text, tokens.clone());
         doc.add_pre_tokenized_text(text, tokens);
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer.add_document(doc).unwrap();
         index_writer.commit().unwrap();
         let reader = index.reader().unwrap();
@@ -869,7 +955,7 @@ mod tests {
         let mut schema_builder = Schema::builder();
         let text = schema_builder.add_text_field("text", TEXT);
         let schema = schema_builder.build();
-        let mut doc = Document::default();
+        let mut doc = TantivyDocument::default();
         // This is a bit of a contrived example.
         let tokens = PreTokenizedString {
             text: "contrived-example".to_string(), //< I can't think of a use case where this corner case happens in real life.
@@ -894,7 +980,7 @@ mod tests {
         doc.add_pre_tokenized_text(text, tokens);
         doc.add_text(text, "hello");
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
         index_writer.add_document(doc).unwrap();
         index_writer.commit().unwrap();
         let reader = index.reader().unwrap();
@@ -930,7 +1016,7 @@ mod tests {
         let schema = index.schema();
         let mut index_writer = index.writer(50_000_000).unwrap();
         let title = schema.get_field("title").unwrap();
-        let mut document = Document::default();
+        let mut document = TantivyDocument::default();
         document.add_text(title, "The Old Man and the Sea");
         index_writer.add_document(document).unwrap();
         let error = index_writer.commit().unwrap_err();

src/lib.rs

@@ -21,7 +21,7 @@
 //! # use tantivy::collector::TopDocs;
 //! # use tantivy::query::QueryParser;
 //! # use tantivy::schema::*;
-//! # use tantivy::{doc, DocAddress, Index, Score};
+//! # use tantivy::{doc, DocAddress, Index, IndexWriter, Score};
 //! #
 //! # fn main() {
 //! #     // Let's create a temporary directory for the
@@ -53,7 +53,7 @@
 //!
 //! // Here we use a buffer of 100MB that will be split
 //! // between indexing threads.
-//! let mut index_writer = index.writer(100_000_000)?;
+//! let mut index_writer: IndexWriter = index.writer(100_000_000)?;
 //!
 //! // Let's index one documents!
 //! index_writer.add_document(doc!(
@@ -89,8 +89,8 @@
 //!
 //! for (_score, doc_address) in top_docs {
 //!     // Retrieve the actual content of documents given its `doc_address`.
-//!     let retrieved_doc = searcher.doc(doc_address)?;
-//!     println!("{}", schema.to_json(&retrieved_doc));
+//!     let retrieved_doc = searcher.doc::<TantivyDocument>(doc_address)?;
+//!     println!("{}", retrieved_doc.to_json(&schema));
 //! }
 //!
 //! # Ok(())
@@ -103,7 +103,48 @@
 //! the example code (
 //! [literate programming](https://tantivy-search.github.io/examples/basic_search.html) /
 //! [source code](https://github.com/quickwit-oss/tantivy/blob/main/examples/basic_search.rs))
-
+//!
+//! # Tantivy Architecture Overview
+//!
+//! Tantivy is inspired by Lucene, the Architecture is very similar.
+//!
+//! ## Core Concepts
+//!
+//! - **[Index]**: A collection of segments. The top level entry point for tantivy users to search
+//!   and index data.
+//!
+//! - **[Segment]**: At the heart of Tantivy's indexing structure is the [Segment]. It contains
+//!   documents and indices and is the atomic unit of indexing and search.
+//!
+//! - **[Schema](schema)**: A schema is a set of fields in an index. Each field has a specific data
+//!   type and set of attributes.
+//!
+//! - **[IndexWriter]**: Responsible creating and merging segments. It executes the indexing
+//!   pipeline including tokenization, creating indices, and storing the index in the
+//!   [Directory](directory).
+//!
+//! - **Searching**: [Searcher] searches the segments with anything that implements
+//!   [Query](query::Query) and merges the results. The list of [supported
+//! queries](query::Query#implementors). Custom Queries are supported by implementing the
+//! [Query](query::Query) trait.
+//!
+//! - **[Directory](directory)**: Abstraction over the storage where the index data is stored.
+//!
+//! - **[Tokenizer](tokenizer)**: Breaks down text into individual tokens. Users can implement or
+//!   use provided tokenizers.
+//!
+//! ## Architecture Flow
+//!
+//! 1. **Document Addition**: Users create documents according to the defined schema. The documents
+//!    fields are tokenized, processed, and added to the current segment. See
+//!    [Document](schema::document) for the structure and usage.
+//!
+//! 2. **Segment Creation**: Once the memory limit threshold is reached or a commit is called, the
+//!    segment is written to the Directory. Documents are searchable after `commit`.
+//!
+//! 3. **Merging**: To optimize space and search speed, segments might be merged. This operation is
+//!    performed in the background. Customize the merge behaviour via
+//!    [IndexWriter::set_merge_policy].
 #[cfg_attr(test, macro_use)]
 extern crate serde_json;
 #[macro_use]
@@ -137,7 +178,7 @@ pub use crate::future_result::FutureResult;
 pub type Result<T> = std::result::Result<T, TantivyError>;
 
 mod core;
-mod indexer;
+pub mod indexer;
 
 #[allow(unused_doc_comments)]
 pub mod error;
@@ -161,8 +202,7 @@ pub mod termdict;
 mod reader;
 
 pub use self::reader::{IndexReader, IndexReaderBuilder, ReloadPolicy, Warmer};
-mod snippet;
-pub use self::snippet::{Snippet, SnippetGenerator};
+pub mod snippet;
 
 mod docset;
 use std::fmt;
@@ -173,26 +213,34 @@ use once_cell::sync::Lazy;
 use serde::{Deserialize, Serialize};
 
 pub use self::docset::{DocSet, TERMINATED};
+#[deprecated(
+    since = "0.22.0",
+    note = "Will be removed in tantivy 0.23. Use export from snippet module instead"
+)]
+pub use self::snippet::{Snippet, SnippetGenerator};
 #[doc(hidden)]
 pub use crate::core::json_utils;
 pub use crate::core::{
-    Executor, Index, IndexBuilder, IndexMeta, IndexSettings, IndexSortByField, InvertedIndexReader,
-    Order, Searcher, SearcherGeneration, Segment, SegmentComponent, SegmentId, SegmentMeta,
-    SegmentReader, SingleSegmentIndexWriter,
+    merge_field_meta_data, Executor, FieldMetadata, Index, IndexBuilder, IndexMeta, IndexSettings,
+    IndexSortByField, InvertedIndexReader, Order, Searcher, SearcherGeneration, Segment,
+    SegmentComponent, SegmentId, SegmentMeta, SegmentReader, SingleSegmentIndexWriter,
 };
 pub use crate::directory::Directory;
-pub use crate::indexer::operation::UserOperation;
-pub use crate::indexer::{merge_filtered_segments, merge_indices, IndexWriter, PreparedCommit};
+pub use crate::indexer::IndexWriter;
+#[deprecated(
+    since = "0.22.0",
+    note = "Will be removed in tantivy 0.23. Use export from indexer module instead"
+)]
+pub use crate::indexer::{merge_filtered_segments, merge_indices, PreparedCommit};
 pub use crate::postings::Postings;
 #[allow(deprecated)]
 pub use crate::schema::DatePrecision;
-pub use crate::schema::{DateOptions, DateTimePrecision, Document, Term};
+pub use crate::schema::{DateOptions, DateTimePrecision, Document, TantivyDocument, Term};
 
 /// Index format version.
-const INDEX_FORMAT_VERSION: u32 = 5;
-
-#[cfg(all(feature = "mmap", unix))]
-pub use memmap2::Advice;
+const INDEX_FORMAT_VERSION: u32 = 6;
+/// Oldest index format version this tantivy version can read.
+const INDEX_FORMAT_OLDEST_SUPPORTED_VERSION: u32 = 4;
 
 /// Structure version for the index.
 #[derive(Clone, PartialEq, Eq, Serialize, Deserialize)]
@@ -342,8 +390,9 @@ pub mod tests {
     use crate::docset::{DocSet, TERMINATED};
     use crate::merge_policy::NoMergePolicy;
     use crate::query::BooleanQuery;
+    use crate::schema::document::Value;
     use crate::schema::*;
-    use crate::{DateTime, DocAddress, Index, Postings, ReloadPolicy};
+    use crate::{DateTime, DocAddress, Index, IndexWriter, Postings, ReloadPolicy};
 
     pub fn fixed_size_test<O: BinarySerializable + FixedSize + Default>() {
         let mut buffer = Vec::new();
@@ -414,7 +463,7 @@ pub mod tests {
         let schema = schema_builder.build();
         let index = Index::create_from_tempdir(schema)?;
         // writing the segment
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         {
             let doc = doc!(text_field=>"af b");
             index_writer.add_document(doc)?;
@@ -436,7 +485,7 @@ pub mod tests {
         let mut schema_builder = Schema::builder();
         let text_field = schema_builder.add_text_field("text", TEXT);
         let index = Index::create_in_ram(schema_builder.build());
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         index_writer.add_document(doc!(text_field=>"a b c"))?;
         index_writer.commit()?;
         index_writer.add_document(doc!(text_field=>"a"))?;
@@ -463,7 +512,7 @@ pub mod tests {
         let title_field = schema_builder.add_text_field("title", TEXT);
         let text_field = schema_builder.add_text_field("text", TEXT);
         let index = Index::create_in_ram(schema_builder.build());
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         index_writer.add_document(doc!(text_field=>"a b c"))?;
         index_writer.commit()?;
         let index_reader = index.reader()?;
@@ -485,7 +534,7 @@ pub mod tests {
         let mut schema_builder = Schema::builder();
         let text_field = schema_builder.add_text_field("text", TEXT);
         let index = Index::create_in_ram(schema_builder.build());
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         index_writer.add_document(doc!(text_field=>"a b c"))?;
         index_writer.add_document(doc!())?;
         index_writer.add_document(doc!(text_field=>"a b"))?;
@@ -528,7 +577,7 @@ pub mod tests {
             .unwrap();
         {
             // writing the segment
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             // 0
             index_writer.add_document(doc!(text_field=>"a b"))?;
             // 1
@@ -575,7 +624,7 @@ pub mod tests {
         }
         {
             // writing the segment
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             // 0
             index_writer.add_document(doc!(text_field=>"a b"))?;
             // 1
@@ -612,7 +661,7 @@ pub mod tests {
         }
         {
             // writing the segment
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             index_writer.add_document(doc!(text_field=>"a b"))?;
             index_writer.delete_term(Term::from_field_text(text_field, "c"));
             index_writer.rollback()?;
@@ -662,7 +711,7 @@ pub mod tests {
         let schema = schema_builder.build();
 
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         index_writer.add_document(doc!(field=>1u64))?;
         index_writer.commit()?;
         let reader = index.reader()?;
@@ -685,7 +734,7 @@ pub mod tests {
         let schema = schema_builder.build();
 
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         let negative_val = -1i64;
         index_writer.add_document(doc!(value_field => negative_val))?;
         index_writer.commit()?;
@@ -709,7 +758,7 @@ pub mod tests {
         let schema = schema_builder.build();
 
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         let val = std::f64::consts::PI;
         index_writer.add_document(doc!(value_field => val))?;
         index_writer.commit()?;
@@ -733,7 +782,7 @@ pub mod tests {
         let absent_field = schema_builder.add_text_field("absent_text", TEXT);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         index_writer.add_document(doc!(text_field=>"a"))?;
         assert!(index_writer.commit().is_ok());
         let reader = index.reader()?;
@@ -756,7 +805,7 @@ pub mod tests {
             .try_into()?;
 
         // writing the segment
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         index_writer.add_document(doc!(text_field=>"63"))?;
         index_writer.add_document(doc!(text_field=>"70"))?;
         index_writer.add_document(doc!(text_field=>"34"))?;
@@ -781,7 +830,7 @@ pub mod tests {
         let index = Index::create_in_ram(schema);
         {
             // writing the segment
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             index_writer.add_document(doc!(text_field=>"af af af bc bc"))?;
             index_writer.commit()?;
         }
@@ -813,7 +862,7 @@ pub mod tests {
         let index = Index::create_in_ram(schema);
         let reader = index.reader()?;
         // writing the segment
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         index_writer.add_document(doc!(text_field=>"af af af b"))?;
         index_writer.add_document(doc!(text_field=>"a b c"))?;
         index_writer.add_document(doc!(text_field=>"a b c d"))?;
@@ -877,7 +926,7 @@ pub mod tests {
             .try_into()?;
         assert_eq!(reader.searcher().num_docs(), 0u64);
         // writing the segment
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         index_writer.add_document(doc!(text_field=>"af b"))?;
         index_writer.add_document(doc!(text_field=>"a b c"))?;
         index_writer.add_document(doc!(text_field=>"a b c d"))?;
@@ -985,13 +1034,13 @@ pub mod tests {
                             text_field => "some other value",
                             other_text_field => "short");
         assert_eq!(document.len(), 3);
-        let values: Vec<&Value> = document.get_all(text_field).collect();
+        let values: Vec<&OwnedValue> = document.get_all(text_field).collect();
         assert_eq!(values.len(), 2);
-        assert_eq!(values[0].as_text(), Some("tantivy"));
-        assert_eq!(values[1].as_text(), Some("some other value"));
-        let values: Vec<&Value> = document.get_all(other_text_field).collect();
+        assert_eq!(values[0].as_str(), Some("tantivy"));
+        assert_eq!(values[1].as_str(), Some("some other value"));
+        let values: Vec<&OwnedValue> = document.get_all(other_text_field).collect();
         assert_eq!(values.len(), 1);
-        assert_eq!(values[0].as_text(), Some("short"));
+        assert_eq!(values[0].as_str(), Some("short"));
     }
 
     #[test]
@@ -1005,7 +1054,7 @@ pub mod tests {
         let schema = schema_builder.build();
 
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         {
             let document =
                 doc!(fast_field_unsigned => 4u64, fast_field_signed=>4i64, fast_field_float=>4f64);
@@ -1071,7 +1120,7 @@ pub mod tests {
         let index = Index::create_in_ram(schema);
         let index_reader = index.reader()?;
 
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
         index_writer.set_merge_policy(Box::new(NoMergePolicy));
 
         for doc_id in 0u64..DOC_COUNT {
@@ -1124,7 +1173,7 @@ pub mod tests {
         let body = builder.add_text_field("body", TEXT | STORED);
         let schema = builder.build();
         let index = Index::create_in_dir(&index_path, schema)?;
-        let mut writer = index.writer(50_000_000)?;
+        let mut writer: IndexWriter = index.writer(50_000_000)?;
         writer.set_merge_policy(Box::new(NoMergePolicy));
         for _ in 0..5000 {
             writer.add_document(doc!(body => "foo"))?;

src/macros.rs

@@ -45,12 +45,12 @@
 macro_rules! doc(
     () => {
         {
-            ($crate::Document::default())
+            ($crate::TantivyDocument::default())
         }
     }; // avoids a warning due to the useless `mut`.
     ($($field:expr => $value:expr),*) => {
         {
-            let mut document = $crate::Document::default();
+            let mut document = $crate::TantivyDocument::default();
             $(
                 document.add_field_value($field, $value);
             )*

src/positions/reader.rs

@@ -92,7 +92,7 @@ impl PositionReader {
             // that block is bitpacked.
             let bit_width = bit_widths[block_rel_id];
             self.block_decoder
-                .uncompress_block_unsorted(compressed_data, bit_width);
+                .uncompress_block_unsorted(compressed_data, bit_width, false);
         } else {
             // that block is vint encoded.
             self.block_decoder

src/positions/serializer.rs

@@ -62,8 +62,9 @@ impl<W: io::Write> PositionSerializer<W> {
             return;
         }
         if self.block.len() == COMPRESSION_BLOCK_SIZE {
-            let (bit_width, block_encoded): (u8, &[u8]) =
-                self.block_encoder.compress_block_unsorted(&self.block[..]);
+            let (bit_width, block_encoded): (u8, &[u8]) = self
+                .block_encoder
+                .compress_block_unsorted(&self.block[..], false);
             self.bit_widths.push(bit_width);
             self.positions_buffer.extend(block_encoded);
         } else {

src/postings/block_segment_postings.rs

@@ -24,13 +24,13 @@ fn max_score<I: Iterator<Item = Score>>(mut it: I) -> Option<Score> {
 #[derive(Clone)]
 pub struct BlockSegmentPostings {
     pub(crate) doc_decoder: BlockDecoder,
-    loaded_offset: usize,
+    block_loaded: bool,
     freq_decoder: BlockDecoder,
     freq_reading_option: FreqReadingOption,
     block_max_score_cache: Option<Score>,
     doc_freq: u32,
     data: OwnedBytes,
-    pub(crate) skip_reader: SkipReader,
+    skip_reader: SkipReader,
 }
 
 fn decode_bitpacked_block(
@@ -40,10 +40,16 @@ fn decode_bitpacked_block(
     doc_offset: DocId,
     doc_num_bits: u8,
     tf_num_bits: u8,
+    strict_delta: bool,
 ) {
-    let num_consumed_bytes = doc_decoder.uncompress_block_sorted(data, doc_offset, doc_num_bits);
+    let num_consumed_bytes =
+        doc_decoder.uncompress_block_sorted(data, doc_offset, doc_num_bits, strict_delta);
     if let Some(freq_decoder) = freq_decoder_opt {
-        freq_decoder.uncompress_block_unsorted(&data[num_consumed_bytes..], tf_num_bits);
+        freq_decoder.uncompress_block_unsorted(
+            &data[num_consumed_bytes..],
+            tf_num_bits,
+            strict_delta,
+        );
     }
 }
 
@@ -57,11 +63,15 @@ fn decode_vint_block(
     let num_consumed_bytes =
         doc_decoder.uncompress_vint_sorted(data, doc_offset, num_vint_docs, TERMINATED);
     if let Some(freq_decoder) = freq_decoder_opt {
-        freq_decoder.uncompress_vint_unsorted(
-            &data[num_consumed_bytes..],
-            num_vint_docs,
-            TERMINATED,
-        );
+        // if it's a json term with freq, containing less than 256 docs, we can reach here thinking
+        // we have a freq, despite not really having one.
+        if data.len() > num_consumed_bytes {
+            freq_decoder.uncompress_vint_unsorted(
+                &data[num_consumed_bytes..],
+                num_vint_docs,
+                TERMINATED,
+            );
+        }
     }
 }
 
@@ -78,28 +88,46 @@ fn split_into_skips_and_postings(
 }
 
 impl BlockSegmentPostings {
+    /// Opens a `BlockSegmentPostings`.
+    /// `doc_freq` is the number of documents in the posting list.
+    /// `record_option` represents the amount of data available according to the schema.
+    /// `requested_option` is the amount of data requested by the user.
+    /// If for instance, we do not request for term frequencies, this function will not decompress
+    /// term frequency blocks.
     pub(crate) fn open(
         doc_freq: u32,
         data: FileSlice,
-        record_option: IndexRecordOption,
+        mut record_option: IndexRecordOption,
         requested_option: IndexRecordOption,
     ) -> io::Result<BlockSegmentPostings> {
+        let bytes = data.read_bytes()?;
+        let (skip_data_opt, postings_data) = split_into_skips_and_postings(doc_freq, bytes)?;
+        let skip_reader = match skip_data_opt {
+            Some(skip_data) => {
+                let block_count = doc_freq as usize / COMPRESSION_BLOCK_SIZE;
+                // 8 is the minimum size of a block with frequency (can be more if pos are stored
+                // too)
+                if skip_data.len() < 8 * block_count {
+                    // the field might be encoded with frequency, but this term in particular isn't.
+                    // This can happen for JSON field with term frequencies:
+                    // - text terms are encoded with term freqs.
+                    // - numerical terms are encoded without term freqs.
+                    record_option = IndexRecordOption::Basic;
+                }
+                SkipReader::new(skip_data, doc_freq, record_option)
+            }
+            None => SkipReader::new(OwnedBytes::empty(), doc_freq, record_option),
+        };
+
         let freq_reading_option = match (record_option, requested_option) {
             (IndexRecordOption::Basic, _) => FreqReadingOption::NoFreq,
             (_, IndexRecordOption::Basic) => FreqReadingOption::SkipFreq,
             (_, _) => FreqReadingOption::ReadFreq,
         };
 
-        let bytes = data.read_bytes()?;
-        let (skip_data_opt, postings_data) = split_into_skips_and_postings(doc_freq, bytes)?;
-        let skip_reader = match skip_data_opt {
-            Some(skip_data) => SkipReader::new(skip_data, doc_freq, record_option),
-            None => SkipReader::new(OwnedBytes::empty(), doc_freq, record_option),
-        };
-
         let mut block_segment_postings = BlockSegmentPostings {
             doc_decoder: BlockDecoder::with_val(TERMINATED),
-            loaded_offset: usize::MAX,
+            block_loaded: false,
             freq_decoder: BlockDecoder::with_val(1),
             freq_reading_option,
             block_max_score_cache: None,
@@ -169,7 +197,7 @@ impl BlockSegmentPostings {
             split_into_skips_and_postings(doc_freq, postings_data)?;
         self.data = postings_data;
         self.block_max_score_cache = None;
-        self.loaded_offset = usize::MAX;
+        self.block_loaded = false;
         if let Some(skip_data) = skip_data_opt {
             self.skip_reader.reset(skip_data, doc_freq);
         } else {
@@ -265,22 +293,23 @@ impl BlockSegmentPostings {
     pub(crate) fn shallow_seek(&mut self, target_doc: DocId) {
         if self.skip_reader.seek(target_doc) {
             self.block_max_score_cache = None;
+            self.block_loaded = false;
         }
     }
 
     pub(crate) fn block_is_loaded(&self) -> bool {
-        self.loaded_offset == self.skip_reader.byte_offset()
+        self.block_loaded
     }
 
     pub(crate) fn load_block(&mut self) {
         let offset = self.skip_reader.byte_offset();
-        if self.loaded_offset == offset {
+        if self.block_is_loaded() {
             return;
         }
-        self.loaded_offset = offset;
         match self.skip_reader.block_info() {
             BlockInfo::BitPacked {
                 doc_num_bits,
+                strict_delta_encoded,
                 tf_num_bits,
                 ..
             } => {
@@ -295,6 +324,7 @@ impl BlockSegmentPostings {
                     self.skip_reader.last_doc_in_previous_block,
                     doc_num_bits,
                     tf_num_bits,
+                    strict_delta_encoded,
                 );
             }
             BlockInfo::VInt { num_docs } => {
@@ -318,13 +348,13 @@ impl BlockSegmentPostings {
                 );
             }
         }
+        self.block_loaded = true;
     }
 
     /// Advance to the next block.
-    ///
-    /// Returns false if and only if there is no remaining block.
     pub fn advance(&mut self) {
         self.skip_reader.advance();
+        self.block_loaded = false;
         self.block_max_score_cache = None;
         self.load_block();
     }
@@ -333,7 +363,7 @@ impl BlockSegmentPostings {
     pub fn empty() -> BlockSegmentPostings {
         BlockSegmentPostings {
             doc_decoder: BlockDecoder::with_val(TERMINATED),
-            loaded_offset: 0,
+            block_loaded: true,
             freq_decoder: BlockDecoder::with_val(1),
             freq_reading_option: FreqReadingOption::NoFreq,
             block_max_score_cache: None,
@@ -342,6 +372,10 @@ impl BlockSegmentPostings {
             skip_reader: SkipReader::new(OwnedBytes::empty(), 0, IndexRecordOption::Basic),
         }
     }
+
+    pub(crate) fn skip_reader(&self) -> &SkipReader {
+        &self.skip_reader
+    }
 }
 
 #[cfg(test)]

src/postings/compression/mod.rs

@@ -33,14 +33,40 @@ impl BlockEncoder {
     }
 
     pub fn compress_block_sorted(&mut self, block: &[u32], offset: u32) -> (u8, &[u8]) {
-        let num_bits = self.bitpacker.num_bits_sorted(offset, block);
+        // if offset is zero, convert it to None. This is correct as long as we do the same when
+        // decompressing. It's required in case the block starts with an actual zero.
+        let offset = if offset == 0u32 { None } else { Some(offset) };
+
+        let num_bits = self.bitpacker.num_bits_strictly_sorted(offset, block);
         let written_size =
             self.bitpacker
-                .compress_sorted(offset, block, &mut self.output[..], num_bits);
+                .compress_strictly_sorted(offset, block, &mut self.output[..], num_bits);
         (num_bits, &self.output[..written_size])
     }
 
-    pub fn compress_block_unsorted(&mut self, block: &[u32]) -> (u8, &[u8]) {
+    /// Compress a single block of unsorted numbers.
+    ///
+    /// If `minus_one_encoded` is set, each value must be >= 1, and will be encoded in a sligly
+    /// more compact format. This is useful for some values where 0 isn't a correct value, such
+    /// as term frequency, but isn't correct for some usages like position lists, where 0 can
+    /// appear.
+    pub fn compress_block_unsorted(
+        &mut self,
+        block: &[u32],
+        minus_one_encoded: bool,
+    ) -> (u8, &[u8]) {
+        debug_assert!(!minus_one_encoded || !block.contains(&0));
+
+        let mut block_minus_one = [0; COMPRESSION_BLOCK_SIZE];
+        let block = if minus_one_encoded {
+            for (elem_min_one, elem) in block_minus_one.iter_mut().zip(block) {
+                *elem_min_one = elem - 1;
+            }
+            &block_minus_one
+        } else {
+            block
+        };
+
         let num_bits = self.bitpacker.num_bits(block);
         let written_size = self
             .bitpacker
@@ -71,21 +97,55 @@ impl BlockDecoder {
         }
     }
 
+    /// Decompress block of sorted integers.
+    ///
+    /// `strict_delta` depends on what encoding was used. Older version of tantivy never use strict
+    /// deltas, newer versions always use them.
     pub fn uncompress_block_sorted(
         &mut self,
         compressed_data: &[u8],
         offset: u32,
         num_bits: u8,
+        strict_delta: bool,
     ) -> usize {
-        self.output_len = COMPRESSION_BLOCK_SIZE;
-        self.bitpacker
-            .decompress_sorted(offset, compressed_data, &mut self.output, num_bits)
+        if strict_delta {
+            let offset = std::num::NonZeroU32::new(offset).map(std::num::NonZeroU32::get);
+
+            self.output_len = COMPRESSION_BLOCK_SIZE;
+            self.bitpacker.decompress_strictly_sorted(
+                offset,
+                compressed_data,
+                &mut self.output,
+                num_bits,
+            )
+        } else {
+            self.output_len = COMPRESSION_BLOCK_SIZE;
+            self.bitpacker
+                .decompress_sorted(offset, compressed_data, &mut self.output, num_bits)
+        }
     }
 
-    pub fn uncompress_block_unsorted(&mut self, compressed_data: &[u8], num_bits: u8) -> usize {
+    /// Decompress block of unsorted integers.
+    ///
+    /// `minus_one_encoded` depends on what encoding was used. Older version of tantivy never use
+    /// that encoding. Newer version use it for some structures, but not all. See the corresponding
+    /// call to `BlockEncoder::compress_block_unsorted`.
+    pub fn uncompress_block_unsorted(
+        &mut self,
+        compressed_data: &[u8],
+        num_bits: u8,
+        minus_one_encoded: bool,
+    ) -> usize {
         self.output_len = COMPRESSION_BLOCK_SIZE;
-        self.bitpacker
-            .decompress(compressed_data, &mut self.output, num_bits)
+        let res = self
+            .bitpacker
+            .decompress(compressed_data, &mut self.output, num_bits);
+        if minus_one_encoded {
+            for val in &mut self.output {
+                *val += 1;
+            }
+        }
+        res
     }
 
     #[inline]
@@ -218,7 +278,8 @@ pub mod tests {
         let (num_bits, compressed_data) = encoder.compress_block_sorted(&vals, 0);
         let mut decoder = BlockDecoder::default();
         {
-            let consumed_num_bytes = decoder.uncompress_block_sorted(compressed_data, 0, num_bits);
+            let consumed_num_bytes =
+                decoder.uncompress_block_sorted(compressed_data, 0, num_bits, true);
             assert_eq!(consumed_num_bytes, compressed_data.len());
         }
         for i in 0..128 {
@@ -233,7 +294,8 @@ pub mod tests {
         let (num_bits, compressed_data) = encoder.compress_block_sorted(&vals, 10);
         let mut decoder = BlockDecoder::default();
         {
-            let consumed_num_bytes = decoder.uncompress_block_sorted(compressed_data, 10, num_bits);
+            let consumed_num_bytes =
+                decoder.uncompress_block_sorted(compressed_data, 10, num_bits, true);
             assert_eq!(consumed_num_bytes, compressed_data.len());
         }
         for i in 0..128 {
@@ -252,7 +314,8 @@ pub mod tests {
         compressed.push(173u8);
         let mut decoder = BlockDecoder::default();
         {
-            let consumed_num_bytes = decoder.uncompress_block_sorted(&compressed, 10, num_bits);
+            let consumed_num_bytes =
+                decoder.uncompress_block_sorted(&compressed, 10, num_bits, true);
             assert_eq!(consumed_num_bytes, compressed.len() - 1);
             assert_eq!(compressed[consumed_num_bytes], 173u8);
         }
@@ -263,21 +326,25 @@ pub mod tests {
 
     #[test]
     fn test_encode_unsorted_block_with_junk() {
-        let mut compressed: Vec<u8> = Vec::new();
-        let n = 128;
-        let vals: Vec<u32> = (0..n).map(|i| 11u32 + (i as u32) * 7u32 % 12).collect();
-        let mut encoder = BlockEncoder::default();
-        let (num_bits, compressed_data) = encoder.compress_block_unsorted(&vals);
-        compressed.extend_from_slice(compressed_data);
-        compressed.push(173u8);
-        let mut decoder = BlockDecoder::default();
-        {
-            let consumed_num_bytes = decoder.uncompress_block_unsorted(&compressed, num_bits);
-            assert_eq!(consumed_num_bytes + 1, compressed.len());
-            assert_eq!(compressed[consumed_num_bytes], 173u8);
-        }
-        for i in 0..n {
-            assert_eq!(vals[i], decoder.output(i));
+        for minus_one_encode in [false, true] {
+            let mut compressed: Vec<u8> = Vec::new();
+            let n = 128;
+            let vals: Vec<u32> = (0..n).map(|i| 11u32 + (i as u32) * 7u32 % 12).collect();
+            let mut encoder = BlockEncoder::default();
+            let (num_bits, compressed_data) =
+                encoder.compress_block_unsorted(&vals, minus_one_encode);
+            compressed.extend_from_slice(compressed_data);
+            compressed.push(173u8);
+            let mut decoder = BlockDecoder::default();
+            {
+                let consumed_num_bytes =
+                    decoder.uncompress_block_unsorted(&compressed, num_bits, minus_one_encode);
+                assert_eq!(consumed_num_bytes + 1, compressed.len());
+                assert_eq!(compressed[consumed_num_bytes], 173u8);
+            }
+            for i in 0..n {
+                assert_eq!(vals[i], decoder.output(i));
+            }
         }
     }
 
@@ -344,7 +411,7 @@ mod bench {
         let (num_bits, compressed) = encoder.compress_block_sorted(&data, 0u32);
         let mut decoder = BlockDecoder::default();
         b.iter(|| {
-            decoder.uncompress_block_sorted(compressed, 0u32, num_bits);
+            decoder.uncompress_block_sorted(compressed, 0u32, num_bits, true);
         });
     }
 

src/postings/indexing_context.rs

@@ -1,5 +1,7 @@
 use stacker::{ArenaHashMap, MemoryArena};
 
+use crate::indexer::path_to_unordered_id::PathToUnorderedId;
+
 /// IndexingContext contains all of the transient memory arenas
 /// required for building the inverted index.
 pub(crate) struct IndexingContext {
@@ -8,6 +10,7 @@ pub(crate) struct IndexingContext {
     pub term_index: ArenaHashMap,
     /// Arena is a memory arena that stores posting lists / term frequencies / positions.
     pub arena: MemoryArena,
+    pub path_to_unordered_id: PathToUnorderedId,
 }
 
 impl IndexingContext {
@@ -17,6 +20,7 @@ impl IndexingContext {
         IndexingContext {
             arena: MemoryArena::default(),
             term_index,
+            path_to_unordered_id: PathToUnorderedId::default(),
         }
     }
 

src/postings/json_postings_writer.rs

@@ -3,13 +3,18 @@ use std::io;
 use stacker::Addr;
 
 use crate::indexer::doc_id_mapping::DocIdMapping;
+use crate::indexer::path_to_unordered_id::OrderedPathId;
 use crate::postings::postings_writer::SpecializedPostingsWriter;
 use crate::postings::recorder::{BufferLender, DocIdRecorder, Recorder};
 use crate::postings::{FieldSerializer, IndexingContext, IndexingPosition, PostingsWriter};
-use crate::schema::Type;
+use crate::schema::{Field, Type, JSON_END_OF_PATH};
 use crate::tokenizer::TokenStream;
 use crate::{DocId, Term};
 
+/// The `JsonPostingsWriter` is odd in that it relies on a hidden contract:
+///
+/// `subscribe` is called directly to index non-text tokens, while
+/// `index_text` is used to index text.
 #[derive(Default)]
 pub(crate) struct JsonPostingsWriter<Rec: Recorder> {
     str_posting_writer: SpecializedPostingsWriter<Rec>,
@@ -54,18 +59,24 @@ impl<Rec: Recorder> PostingsWriter for JsonPostingsWriter<Rec> {
     /// The actual serialization format is handled by the `PostingsSerializer`.
     fn serialize(
         &self,
-        term_addrs: &[(Term<&[u8]>, Addr)],
+        term_addrs: &[(Field, OrderedPathId, &[u8], Addr)],
+        ordered_id_to_path: &[&str],
         doc_id_map: Option<&DocIdMapping>,
         ctx: &IndexingContext,
         serializer: &mut FieldSerializer,
     ) -> io::Result<()> {
+        let mut term_buffer = Term::with_capacity(48);
         let mut buffer_lender = BufferLender::default();
-        for (term, addr) in term_addrs {
-            if let Some(json_value) = term.value().as_json_value_bytes() {
+        for (_field, path_id, term, addr) in term_addrs {
+            term_buffer.clear_with_field_and_type(Type::Json, Field::from_field_id(0));
+            term_buffer.append_bytes(ordered_id_to_path[path_id.path_id() as usize].as_bytes());
+            term_buffer.append_bytes(&[JSON_END_OF_PATH]);
+            term_buffer.append_bytes(term);
+            if let Some(json_value) = term_buffer.value().as_json_value_bytes() {
                 let typ = json_value.typ();
                 if typ == Type::Str {
                     SpecializedPostingsWriter::<Rec>::serialize_one_term(
-                        term,
+                        term_buffer.serialized_value_bytes(),
                         *addr,
                         doc_id_map,
                         &mut buffer_lender,
@@ -74,7 +85,7 @@ impl<Rec: Recorder> PostingsWriter for JsonPostingsWriter<Rec> {
                     )?;
                 } else {
                     SpecializedPostingsWriter::<DocIdRecorder>::serialize_one_term(
-                        term,
+                        term_buffer.serialized_value_bytes(),
                         *addr,
                         doc_id_map,
                         &mut buffer_lender,

src/postings/mod.rs

@@ -52,7 +52,7 @@ pub mod tests {
         Field, IndexRecordOption, Schema, Term, TextFieldIndexing, TextOptions, INDEXED, TEXT,
     };
     use crate::tokenizer::{SimpleTokenizer, MAX_TOKEN_LEN};
-    use crate::{DocId, HasLen, Score};
+    use crate::{DocId, HasLen, IndexWriter, Score};
 
     #[test]
     pub fn test_position_write() -> crate::Result<()> {
@@ -63,7 +63,7 @@ pub mod tests {
         let mut segment = index.new_segment();
         let mut posting_serializer = InvertedIndexSerializer::open(&mut segment)?;
         let mut field_serializer = posting_serializer.new_field(text_field, 120 * 4, None)?;
-        field_serializer.new_term("abc".as_bytes(), 12u32)?;
+        field_serializer.new_term("abc".as_bytes(), 12u32, true)?;
         for doc_id in 0u32..120u32 {
             let delta_positions = vec![1, 2, 3, 2];
             field_serializer.write_doc(doc_id, 4, &delta_positions);
@@ -432,7 +432,7 @@ pub mod tests {
 
         // delete some of the documents
         {
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             index_writer.delete_term(term_0);
             assert!(index_writer.commit().is_ok());
         }
@@ -483,7 +483,7 @@ pub mod tests {
 
         // delete everything else
         {
-            let mut index_writer = index.writer_for_tests()?;
+            let mut index_writer: IndexWriter = index.writer_for_tests()?;
             index_writer.delete_term(term_1);
             assert!(index_writer.commit().is_ok());
         }
@@ -568,8 +568,8 @@ mod bench {
 
     use crate::docset::TERMINATED;
     use crate::query::Intersection;
-    use crate::schema::{Document, Field, IndexRecordOption, Schema, Term, STRING};
-    use crate::{tests, DocSet, Index};
+    use crate::schema::{Field, IndexRecordOption, Schema, TantivyDocument, Term, STRING};
+    use crate::{tests, DocSet, Index, IndexWriter};
 
     pub static TERM_A: Lazy<Term> = Lazy::new(|| {
         let field = Field::from_field_id(0);
@@ -598,9 +598,9 @@ mod bench {
         let index = Index::create_in_ram(schema);
         let posting_list_size = 1_000_000;
         {
-            let mut index_writer = index.writer_for_tests().unwrap();
+            let mut index_writer: IndexWriter = index.writer_for_tests().unwrap();
             for _ in 0..posting_list_size {
-                let mut doc = Document::default();
+                let mut doc = TantivyDocument::default();
                 if rng.gen_bool(1f64 / 15f64) {
                     doc.add_text(text_field, "a");
                 }

src/postings/postings_writer.rs

@@ -6,20 +6,23 @@ use stacker::Addr;
 
 use crate::fieldnorm::FieldNormReaders;
 use crate::indexer::doc_id_mapping::DocIdMapping;
+use crate::indexer::path_to_unordered_id::OrderedPathId;
 use crate::postings::recorder::{BufferLender, Recorder};
 use crate::postings::{
     FieldSerializer, IndexingContext, InvertedIndexSerializer, PerFieldPostingsWriter,
 };
-use crate::schema::{Field, Term};
+use crate::schema::{Field, Schema, Term, Type};
 use crate::tokenizer::{Token, TokenStream, MAX_TOKEN_LEN};
 use crate::DocId;
 
 const POSITION_GAP: u32 = 1;
 
-fn make_field_partition(term_offsets: &[(Term<&[u8]>, Addr)]) -> Vec<(Field, Range<usize>)> {
+fn make_field_partition(
+    term_offsets: &[(Field, OrderedPathId, &[u8], Addr)],
+) -> Vec<(Field, Range<usize>)> {
     let term_offsets_it = term_offsets
         .iter()
-        .map(|(term, _)| term.field())
+        .map(|(field, _, _, _)| *field)
         .enumerate();
     let mut prev_field_opt = None;
     let mut fields = vec![];
@@ -44,19 +47,36 @@ fn make_field_partition(term_offsets: &[(Term<&[u8]>, Addr)]) -> Vec<(Field, Ran
 /// postings serializer.
 pub(crate) fn serialize_postings(
     ctx: IndexingContext,
+    schema: Schema,
     per_field_postings_writers: &PerFieldPostingsWriter,
     fieldnorm_readers: FieldNormReaders,
     doc_id_map: Option<&DocIdMapping>,
     serializer: &mut InvertedIndexSerializer,
 ) -> crate::Result<()> {
-    let mut term_offsets: Vec<(Term<&[u8]>, Addr)> = Vec::with_capacity(ctx.term_index.len());
-    term_offsets.extend(
-        ctx.term_index
-            .iter()
-            .map(|(bytes, addr, _unordered_id)| (Term::wrap(bytes), addr)),
-    );
-    term_offsets.sort_unstable_by_key(|(k, _)| k.clone());
+    // Replace unordered ids by ordered ids to be able to sort
+    let unordered_id_to_ordered_id: Vec<OrderedPathId> =
+        ctx.path_to_unordered_id.unordered_id_to_ordered_id();
 
+    let mut term_offsets: Vec<(Field, OrderedPathId, &[u8], Addr)> =
+        Vec::with_capacity(ctx.term_index.len());
+    term_offsets.extend(ctx.term_index.iter().map(|(key, addr)| {
+        let field = Term::wrap(key).field();
+        if schema.get_field_entry(field).field_type().value_type() == Type::Json {
+            let byte_range_path = 5..5 + 4;
+            let unordered_id = u32::from_be_bytes(key[byte_range_path.clone()].try_into().unwrap());
+            let path_id = unordered_id_to_ordered_id[unordered_id as usize];
+            (field, path_id, &key[byte_range_path.end..], addr)
+        } else {
+            (field, 0.into(), &key[5..], addr)
+        }
+    }));
+    // Sort by field, path, and term
+    term_offsets.sort_unstable_by(
+        |(field1, path_id1, bytes1, _), (field2, path_id2, bytes2, _)| {
+            (field1, path_id1, bytes1).cmp(&(field2, path_id2, bytes2))
+        },
+    );
+    let ordered_id_to_path = ctx.path_to_unordered_id.ordered_id_to_path();
     let field_offsets = make_field_partition(&term_offsets);
     for (field, byte_offsets) in field_offsets {
         let postings_writer = per_field_postings_writers.get_for_field(field);
@@ -65,12 +85,14 @@ pub(crate) fn serialize_postings(
             serializer.new_field(field, postings_writer.total_num_tokens(), fieldnorm_reader)?;
         postings_writer.serialize(
             &term_offsets[byte_offsets],
+            &ordered_id_to_path,
             doc_id_map,
             &ctx,
             &mut field_serializer,
         )?;
         field_serializer.close()?;
     }
+
     Ok(())
 }
 
@@ -98,7 +120,8 @@ pub(crate) trait PostingsWriter: Send + Sync {
     /// The actual serialization format is handled by the `PostingsSerializer`.
     fn serialize(
         &self,
-        term_addrs: &[(Term<&[u8]>, Addr)],
+        term_addrs: &[(Field, OrderedPathId, &[u8], Addr)],
+        ordered_id_to_path: &[&str],
         doc_id_map: Option<&DocIdMapping>,
         ctx: &IndexingContext,
         serializer: &mut FieldSerializer,
@@ -162,7 +185,7 @@ impl<Rec: Recorder> From<SpecializedPostingsWriter<Rec>> for Box<dyn PostingsWri
 impl<Rec: Recorder> SpecializedPostingsWriter<Rec> {
     #[inline]
     pub(crate) fn serialize_one_term(
-        term: &Term<&[u8]>,
+        term: &[u8],
         addr: Addr,
         doc_id_map: Option<&DocIdMapping>,
         buffer_lender: &mut BufferLender,
@@ -171,7 +194,7 @@ impl<Rec: Recorder> SpecializedPostingsWriter<Rec> {
     ) -> io::Result<()> {
         let recorder: Rec = ctx.term_index.read(addr);
         let term_doc_freq = recorder.term_doc_freq().unwrap_or(0u32);
-        serializer.new_term(term.serialized_value_bytes(), term_doc_freq)?;
+        serializer.new_term(term, term_doc_freq, recorder.has_term_freq())?;
         recorder.serialize(&ctx.arena, doc_id_map, serializer, buffer_lender);
         serializer.close_term()?;
         Ok(())
@@ -204,13 +227,14 @@ impl<Rec: Recorder> PostingsWriter for SpecializedPostingsWriter<Rec> {
 
     fn serialize(
         &self,
-        term_addrs: &[(Term<&[u8]>, Addr)],
+        term_addrs: &[(Field, OrderedPathId, &[u8], Addr)],
+        _ordered_id_to_path: &[&str],
         doc_id_map: Option<&DocIdMapping>,
         ctx: &IndexingContext,
         serializer: &mut FieldSerializer,
     ) -> io::Result<()> {
         let mut buffer_lender = BufferLender::default();
-        for (term, addr) in term_addrs {
+        for (_field, _path_id, term, addr) in term_addrs {
             Self::serialize_one_term(term, *addr, doc_id_map, &mut buffer_lender, ctx, serializer)?;
         }
         Ok(())

src/postings/recorder.rs

@@ -79,24 +79,20 @@ pub(crate) trait Recorder: Copy + Default + Send + Sync + 'static {
     ///
     /// Returns `None` if not available.
     fn term_doc_freq(&self) -> Option<u32>;
+
+    #[inline]
+    fn has_term_freq(&self) -> bool {
+        true
+    }
 }
 
 /// Only records the doc ids
-#[derive(Clone, Copy)]
+#[derive(Clone, Copy, Default)]
 pub struct DocIdRecorder {
     stack: ExpUnrolledLinkedList,
     current_doc: DocId,
 }
 
-impl Default for DocIdRecorder {
-    fn default() -> Self {
-        DocIdRecorder {
-            stack: ExpUnrolledLinkedList::default(),
-            current_doc: u32::MAX,
-        }
-    }
-}
-
 impl Recorder for DocIdRecorder {
     #[inline]
     fn current_doc(&self) -> DocId {
@@ -105,8 +101,9 @@ impl Recorder for DocIdRecorder {
 
     #[inline]
     fn new_doc(&mut self, doc: DocId, arena: &mut MemoryArena) {
+        let delta = doc - self.current_doc;
         self.current_doc = doc;
-        self.stack.writer(arena).write_u32_vint(doc);
+        self.stack.writer(arena).write_u32_vint(delta);
     }
 
     #[inline]
@@ -123,21 +120,20 @@ impl Recorder for DocIdRecorder {
         buffer_lender: &mut BufferLender,
     ) {
         let (buffer, doc_ids) = buffer_lender.lend_all();
-        self.stack.read_to_end(arena, buffer);
         // TODO avoid reading twice.
+        self.stack.read_to_end(arena, buffer);
         if let Some(doc_id_map) = doc_id_map {
-            doc_ids.extend(
-                VInt32Reader::new(&buffer[..])
-                    .map(|old_doc_id| doc_id_map.get_new_doc_id(old_doc_id)),
-            );
+            let iter = get_sum_reader(VInt32Reader::new(&buffer[..]));
+            doc_ids.extend(iter.map(|old_doc_id| doc_id_map.get_new_doc_id(old_doc_id)));
             doc_ids.sort_unstable();
 
             for doc in doc_ids {
                 serializer.write_doc(*doc, 0u32, &[][..]);
             }
         } else {
-            for doc in VInt32Reader::new(&buffer[..]) {
-                serializer.write_doc(doc, 0u32, &[][..]);
+            let iter = get_sum_reader(VInt32Reader::new(&buffer[..]));
+            for doc_id in iter {
+                serializer.write_doc(doc_id, 0u32, &[][..]);
             }
         }
     }
@@ -145,6 +141,19 @@ impl Recorder for DocIdRecorder {
     fn term_doc_freq(&self) -> Option<u32> {
         None
     }
+
+    fn has_term_freq(&self) -> bool {
+        false
+    }
+}
+
+/// Takes an Iterator of delta encoded elements and returns an iterator
+/// that yields the sum of the elements.
+fn get_sum_reader(iter: impl Iterator<Item = u32>) -> impl Iterator<Item = u32> {
+    iter.scan(0, |state, delta| {
+        *state += delta;
+        Some(*state)
+    })
 }
 
 /// Recorder encoding document ids, and term frequencies
@@ -164,9 +173,10 @@ impl Recorder for TermFrequencyRecorder {
 
     #[inline]
     fn new_doc(&mut self, doc: DocId, arena: &mut MemoryArena) {
+        let delta = doc - self.current_doc;
         self.term_doc_freq += 1;
         self.current_doc = doc;
-        self.stack.writer(arena).write_u32_vint(doc);
+        self.stack.writer(arena).write_u32_vint(delta);
     }
 
     #[inline]
@@ -193,19 +203,25 @@ impl Recorder for TermFrequencyRecorder {
         let mut u32_it = VInt32Reader::new(&buffer[..]);
         if let Some(doc_id_map) = doc_id_map {
             let mut doc_id_and_tf = vec![];
-            while let Some(old_doc_id) = u32_it.next() {
+            let mut prev_doc = 0;
+            while let Some(delta_doc_id) = u32_it.next() {
+                let doc_id = prev_doc + delta_doc_id;
+                prev_doc = doc_id;
                 let term_freq = u32_it.next().unwrap_or(self.current_tf);
-                doc_id_and_tf.push((doc_id_map.get_new_doc_id(old_doc_id), term_freq));
+                doc_id_and_tf.push((doc_id_map.get_new_doc_id(doc_id), term_freq));
             }
             doc_id_and_tf.sort_unstable_by_key(|&(doc_id, _)| doc_id);
 
             for (doc_id, tf) in doc_id_and_tf {
-                serializer.write_doc(doc_id, tf, &[][..]);
+                serializer.write_doc(doc_id, 0, &[][..]);
             }
         } else {
-            while let Some(doc) = u32_it.next() {
+            let mut prev_doc = 0;
+            while let Some(delta_doc_id) = u32_it.next() {
+                let doc_id = prev_doc + delta_doc_id;
+                prev_doc = doc_id;
                 let term_freq = u32_it.next().unwrap_or(self.current_tf);
-                serializer.write_doc(doc, term_freq, &[][..]);
+                serializer.write_doc(doc_id, 0, &[][..]);
             }
         }
     }
@@ -216,23 +232,13 @@ impl Recorder for TermFrequencyRecorder {
 }
 
 /// Recorder encoding term frequencies as well as positions.
-#[derive(Clone, Copy)]
+#[derive(Clone, Copy, Default)]
 pub struct TfAndPositionRecorder {
     stack: ExpUnrolledLinkedList,
     current_doc: DocId,
     term_doc_freq: u32,
 }
 
-impl Default for TfAndPositionRecorder {
-    fn default() -> Self {
-        TfAndPositionRecorder {
-            stack: ExpUnrolledLinkedList::default(),
-            current_doc: u32::MAX,
-            term_doc_freq: 0u32,
-        }
-    }
-}
-
 impl Recorder for TfAndPositionRecorder {
     #[inline]
     fn current_doc(&self) -> DocId {
@@ -241,9 +247,10 @@ impl Recorder for TfAndPositionRecorder {
 
     #[inline]
     fn new_doc(&mut self, doc: DocId, arena: &mut MemoryArena) {
+        let delta = doc - self.current_doc;
         self.current_doc = doc;
         self.term_doc_freq += 1u32;
-        self.stack.writer(arena).write_u32_vint(doc);
+        self.stack.writer(arena).write_u32_vint(delta);
     }
 
     #[inline]
@@ -269,7 +276,10 @@ impl Recorder for TfAndPositionRecorder {
         self.stack.read_to_end(arena, buffer_u8);
         let mut u32_it = VInt32Reader::new(&buffer_u8[..]);
         let mut doc_id_and_positions = vec![];
-        while let Some(doc) = u32_it.next() {
+        let mut prev_doc = 0;
+        while let Some(delta_doc_id) = u32_it.next() {
+            let doc_id = prev_doc + delta_doc_id;
+            prev_doc = doc_id;
             let mut prev_position_plus_one = 1u32;
             buffer_positions.clear();
             loop {
@@ -287,9 +297,9 @@ impl Recorder for TfAndPositionRecorder {
             if let Some(doc_id_map) = doc_id_map {
                 // this simple variant to remap may consume to much memory
                 doc_id_and_positions
-                    .push((doc_id_map.get_new_doc_id(doc), buffer_positions.to_vec()));
+                    .push((doc_id_map.get_new_doc_id(doc_id), buffer_positions.to_vec()));
             } else {
-                serializer.write_doc(doc, buffer_positions.len() as u32, buffer_positions);
+                serializer.write_doc(doc_id, buffer_positions.len() as u32, buffer_positions);
             }
         }
         if doc_id_map.is_some() {

src/postings/segment_postings.rs

@@ -71,7 +71,7 @@ impl SegmentPostings {
         {
             let mut postings_serializer =
                 PostingsSerializer::new(&mut buffer, 0.0, IndexRecordOption::Basic, None);
-            postings_serializer.new_term(docs.len() as u32);
+            postings_serializer.new_term(docs.len() as u32, false);
             for &doc in docs {
                 postings_serializer.write_doc(doc, 1u32);
             }
@@ -120,7 +120,7 @@ impl SegmentPostings {
             IndexRecordOption::WithFreqs,
             fieldnorm_reader,
         );
-        postings_serializer.new_term(doc_and_tfs.len() as u32);
+        postings_serializer.new_term(doc_and_tfs.len() as u32, true);
         for &(doc, tf) in doc_and_tfs {
             postings_serializer.write_doc(doc, tf);
         }
@@ -238,14 +238,18 @@ impl Postings for SegmentPostings {
     }
 
     fn positions_with_offset(&mut self, offset: u32, output: &mut Vec<u32>) {
-        let term_freq = self.term_freq() as usize;
+        let term_freq = self.term_freq();
         if let Some(position_reader) = self.position_reader.as_mut() {
+            debug_assert!(
+                !self.block_cursor.freqs().is_empty(),
+                "No positions available"
+            );
             let read_offset = self.block_cursor.position_offset()
                 + (self.block_cursor.freqs()[..self.cur]
                     .iter()
                     .cloned()
                     .sum::<u32>() as u64);
-            output.resize(term_freq, 0u32);
+            output.resize(term_freq as usize, 0u32);
             position_reader.read(read_offset, &mut output[..]);
             let mut cum = offset;
             for output_mut in output.iter_mut() {