add truncation comment

Fixes #1783

CHANGELOG.md

@@ -2,22 +2,21 @@ Tantivy 0.19
 ================================
 #### Bugfixes
 - Fix missing fieldnorms for u64, i64, f64, bool, bytes and date [#1620](https://github.com/quickwit-oss/tantivy/pull/1620) (@PSeitz)
-- Fix interpolation overflow in linear interpolation fastfield codec [#1480](https://github.com/quickwit-oss/tantivy/pull/1480 (@PSeitz @fulmicoton)
+- Fix interpolation overflow in linear interpolation fastfield codec [#1480](https://github.com/quickwit-oss/tantivy/pull/1480) (@PSeitz @fulmicoton)
 
 #### Features/Improvements
 - Add support for `IN` in queryparser , e.g. `field: IN [val1 val2 val3]` [#1683](https://github.com/quickwit-oss/tantivy/pull/1683) (@trinity-1686a)
 - Skip score calculation, when no scoring is required [#1646](https://github.com/quickwit-oss/tantivy/pull/1646) (@PSeitz)
 - Limit fast fields to u32 (`get_val(u32)`) [#1644](https://github.com/quickwit-oss/tantivy/pull/1644) (@PSeitz)
-- Updated [Date Field Type](https://github.com/quickwit-oss/tantivy/pull/1396)
-  The `DateTime` type has been updated to hold timestamps with microseconds precision.
-  `DateOptions` and `DatePrecision` have been added to configure Date fields. The precision is used to hint on fast values compression. Otherwise, seconds precision is used everywhere else (i.e terms, indexing). (@evanxg852000)
+- The `DateTime` type has been updated to hold timestamps with microseconds precision.
+  `DateOptions` and `DatePrecision` have been added to configure Date fields. The precision is used to hint on fast values compression. Otherwise, seconds precision is used everywhere else (i.e terms, indexing) [#1396](https://github.com/quickwit-oss/tantivy/pull/1396) (@evanxg852000)
 - Add IP address field type [#1553](https://github.com/quickwit-oss/tantivy/pull/1553) (@PSeitz)
 - Add boolean field type [#1382](https://github.com/quickwit-oss/tantivy/pull/1382) (@boraarslan)
 - Remove Searcher pool and make `Searcher` cloneable. (@PSeitz)
-- Validate settings on create [#1570](https://github.com/quickwit-oss/tantivy/pull/1570 (@PSeitz)
+- Validate settings on create [#1570](https://github.com/quickwit-oss/tantivy/pull/1570) (@PSeitz)
 - Detect and apply gcd on fastfield codecs [#1418](https://github.com/quickwit-oss/tantivy/pull/1418) (@PSeitz)
 - Doc store
-  - use separate thread to compress block store [#1389](https://github.com/quickwit-oss/tantivy/pull/1389) [#1510](https://github.com/quickwit-oss/tantivy/pull/1510 (@PSeitz @fulmicoton)
+  - use separate thread to compress block store [#1389](https://github.com/quickwit-oss/tantivy/pull/1389) [#1510](https://github.com/quickwit-oss/tantivy/pull/1510) (@PSeitz @fulmicoton)
   - Expose doc store cache size [#1403](https://github.com/quickwit-oss/tantivy/pull/1403) (@PSeitz)
   - Enable compression levels for doc store [#1378](https://github.com/quickwit-oss/tantivy/pull/1378) (@PSeitz)
   - Make block size configurable [#1374](https://github.com/quickwit-oss/tantivy/pull/1374) (@kryesh)

Cargo.toml

@@ -15,7 +15,7 @@ rust-version = "1.62"
 
 [dependencies]
 oneshot = "0.1.5"
-base64 = "0.13.0"
+base64 = "0.21.0"
 byteorder = "1.4.3"
 crc32fast = "1.3.2"
 once_cell = "1.10.0"
@@ -36,7 +36,6 @@ fs2 = { version = "0.4.3", optional = true }
 levenshtein_automata = "0.2.1"
 uuid = { version = "1.0.0", features = ["v4", "serde"] }
 crossbeam-channel = "0.5.4"
-stable_deref_trait = "1.2.0"
 rust-stemmers = "1.2.0"
 downcast-rs = "1.2.0"
 bitpacking = { version = "0.8.4", default-features = false, features = ["bitpacker4x"] }
@@ -49,19 +48,20 @@ murmurhash32 = "0.2.0"
 time = { version = "0.3.10", features = ["serde-well-known"] }
 smallvec = "1.8.0"
 rayon = "1.5.2"
-lru = "0.7.5"
+lru = "0.9.0"
 fastdivide = "0.4.0"
 itertools = "0.10.3"
 measure_time = "0.8.2"
-ciborium = { version = "0.2", optional = true}
 async-trait = "0.1.53"
 arc-swap = "1.5.0"
 
+sstable = { version="0.1", path="./sstable", package ="tantivy-sstable", optional = true }
+stacker = { version="0.1", path="./stacker", package ="tantivy-stacker" }
 tantivy-query-grammar = { version= "0.19.0", path="./query-grammar" }
 tantivy-bitpacker = 		{ version= "0.3", path="./bitpacker" }
-common = 								{ version= "0.4", path = "./common/", package = "tantivy-common" }
+common = 								{ version= "0.5", path = "./common/", package = "tantivy-common" }
 fastfield_codecs = 			{ version= "0.3", path="./fastfield_codecs", default-features = false }
-ownedbytes = 						{ version= "0.4", path="./ownedbytes" }
+tokenizer-api = { version="0.1", path="./tokenizer-api", package="tantivy-tokenizer-api" }
 
 [target.'cfg(windows)'.dependencies]
 winapi = "0.3.9"
@@ -104,10 +104,10 @@ zstd-compression = ["zstd"]
 failpoints = ["fail/failpoints"]
 unstable = [] # useful for benches.
 
-quickwit = ["ciborium"]
+quickwit = ["sstable"]
 
 [workspace]
-members = ["query-grammar", "bitpacker", "common", "fastfield_codecs", "ownedbytes"]
+members = ["query-grammar", "bitpacker", "common", "fastfield_codecs", "ownedbytes", "stacker", "sstable", "columnar", "tokenizer-api"]
 
 # Following the "fail" crate best practises, we isolate
 # tests that define specific behavior in fail check points

README.md

@@ -29,7 +29,7 @@ Your mileage WILL vary depending on the nature of queries and their load.
 # Features
 
 - Full-text search
-- Configurable tokenizer (stemming available for 17 Latin languages with third party support for Chinese ([tantivy-jieba](https://crates.io/crates/tantivy-jieba) and [cang-jie](https://crates.io/crates/cang-jie)), Japanese ([lindera](https://github.com/lindera-morphology/lindera-tantivy), [Vaporetto](https://crates.io/crates/vaporetto_tantivy), and [tantivy-tokenizer-tiny-segmenter](https://crates.io/crates/tantivy-tokenizer-tiny-segmenter)) and Korean ([lindera](https://github.com/lindera-morphology/lindera-tantivy) + [lindera-ko-dic-builder](https://github.com/lindera-morphology/lindera-ko-dic-builder))
+- Configurable tokenizer (stemming available for 17 Latin languages) with third party support for Chinese ([tantivy-jieba](https://crates.io/crates/tantivy-jieba) and [cang-jie](https://crates.io/crates/cang-jie)), Japanese ([lindera](https://github.com/lindera-morphology/lindera-tantivy), [Vaporetto](https://crates.io/crates/vaporetto_tantivy), and [tantivy-tokenizer-tiny-segmenter](https://crates.io/crates/tantivy-tokenizer-tiny-segmenter)) and Korean ([lindera](https://github.com/lindera-morphology/lindera-tantivy) + [lindera-ko-dic-builder](https://github.com/lindera-morphology/lindera-ko-dic-builder))
 - Fast (check out the :racehorse: :sparkles: [benchmark](https://tantivy-search.github.io/bench/) :sparkles: :racehorse:)
 - Tiny startup time (<10ms), perfect for command-line tools
 - BM25 scoring (the same as Lucene)
@@ -42,12 +42,12 @@ Your mileage WILL vary depending on the nature of queries and their load.
 - Single valued and multivalued u64, i64, and f64 fast fields (equivalent of doc values in Lucene)
 - `&[u8]` fast fields
 - Text, i64, u64, f64, dates, and hierarchical facet fields
-- LZ4 compressed document store
+- Compressed document store (LZ4, Zstd, None, Brotli, Snap)
 - Range queries
 - Faceted search
 - Configurable indexing (optional term frequency and position indexing)
 - JSON Field
-- Aggregation Collector: range buckets, average, and stats metrics
+- Aggregation Collector: histogram, range buckets, average, and stats metrics
 - LogMergePolicy with deletes
 - Searcher Warmer API
 - Cheesy logo with a horse
@@ -81,6 +81,10 @@ There are many ways to support this project.
 
 We use the GitHub Pull Request workflow: reference a GitHub ticket and/or include a comprehensive commit message when opening a PR.
 
+## Tokenizer
+
+When implementing a tokenizer for tantivy depend on the `tantivy-tokenizer-api` crate.
+
 ## Minimum supported Rust version
 
 Tantivy currently requires at least Rust 1.62 or later to compile.

bitpacker/src/bitpacker.rs

@@ -25,15 +25,14 @@ impl BitPacker {
         num_bits: u8,
         output: &mut TWrite,
     ) -> io::Result<()> {
-        let val_u64 = val as u64;
         let num_bits = num_bits as usize;
         if self.mini_buffer_written + num_bits > 64 {
-            self.mini_buffer |= val_u64.wrapping_shl(self.mini_buffer_written as u32);
+            self.mini_buffer |= val.wrapping_shl(self.mini_buffer_written as u32);
             output.write_all(self.mini_buffer.to_le_bytes().as_ref())?;
-            self.mini_buffer = val_u64.wrapping_shr((64 - self.mini_buffer_written) as u32);
+            self.mini_buffer = val.wrapping_shr((64 - self.mini_buffer_written) as u32);
             self.mini_buffer_written = self.mini_buffer_written + num_bits - 64;
         } else {
-            self.mini_buffer |= val_u64 << self.mini_buffer_written;
+            self.mini_buffer |= val << self.mini_buffer_written;
             self.mini_buffer_written += num_bits;
             if self.mini_buffer_written == 64 {
                 output.write_all(self.mini_buffer.to_le_bytes().as_ref())?;
@@ -92,17 +91,15 @@ impl BitUnpacker {
             return 0u64;
         }
         let addr_in_bits = idx * self.num_bits as u32;
-        let addr = addr_in_bits >> 3;
+        let addr = (addr_in_bits >> 3) as usize;
         let bit_shift = addr_in_bits & 7;
         debug_assert!(
-            addr + 8 <= data.len() as u32,
+            addr + 8 <= data.len(),
             "The fast field field should have been padded with 7 bytes."
         );
-        let bytes: [u8; 8] = (&data[(addr as usize)..(addr as usize) + 8])
-            .try_into()
-            .unwrap();
+        let bytes: [u8; 8] = (&data[addr..addr + 8]).try_into().unwrap();
         let val_unshifted_unmasked: u64 = u64::from_le_bytes(bytes);
-        let val_shifted = (val_unshifted_unmasked >> bit_shift) as u64;
+        let val_shifted = val_unshifted_unmasked >> bit_shift;
         val_shifted & self.mask
     }
 }

bitpacker/src/blocked_bitpacker.rs

@@ -84,7 +84,7 @@ impl BlockedBitpacker {
     #[inline]
     pub fn add(&mut self, val: u64) {
         self.buffer.push(val);
-        if self.buffer.len() == BLOCK_SIZE as usize {
+        if self.buffer.len() == BLOCK_SIZE {
             self.flush();
         }
     }
@@ -126,8 +126,8 @@ impl BlockedBitpacker {
     }
     #[inline]
     pub fn get(&self, idx: usize) -> u64 {
-        let metadata_pos = idx / BLOCK_SIZE as usize;
-        let pos_in_block = idx % BLOCK_SIZE as usize;
+        let metadata_pos = idx / BLOCK_SIZE;
+        let pos_in_block = idx % BLOCK_SIZE;
         if let Some(metadata) = self.offset_and_bits.get(metadata_pos) {
             let unpacked = BitUnpacker::new(metadata.num_bits()).get(
                 pos_in_block as u32,

bitpacker/src/lib.rs

@@ -1,6 +1,8 @@
 mod bitpacker;
 mod blocked_bitpacker;
 
+use std::cmp::Ordering;
+
 pub use crate::bitpacker::{BitPacker, BitUnpacker};
 pub use crate::blocked_bitpacker::BlockedBitpacker;
 
@@ -37,44 +39,104 @@ pub fn compute_num_bits(n: u64) -> u8 {
     }
 }
 
+/// Computes the (min, max) of an iterator of `PartialOrd` values.
+///
+/// For values implementing `Ord` (in a way consistent to their `PartialOrd` impl),
+/// this function behaves as expected.
+///
+/// For values with partial ordering, the behavior is non-trivial and may
+/// depends on the order of the values.
+/// For floats however, it simply returns the same results as if NaN were
+/// skipped.
 pub fn minmax<I, T>(mut vals: I) -> Option<(T, T)>
 where
     I: Iterator<Item = T>,
-    T: Copy + Ord,
+    T: Copy + PartialOrd,
 {
-    if let Some(first_el) = vals.next() {
-        return Some(vals.fold((first_el, first_el), |(min_val, max_val), el| {
-            (min_val.min(el), max_val.max(el))
-        }));
+    let first_el = vals.find(|val| {
+        // We use this to make sure we skip all NaN values when
+        // working with a float type.
+        val.partial_cmp(val) == Some(Ordering::Equal)
+    })?;
+    let mut min_so_far: T = first_el;
+    let mut max_so_far: T = first_el;
+    for val in vals {
+        if val.partial_cmp(&min_so_far) == Some(Ordering::Less) {
+            min_so_far = val;
+        }
+        if val.partial_cmp(&max_so_far) == Some(Ordering::Greater) {
+            max_so_far = val;
+        }
     }
-    None
+    Some((min_so_far, max_so_far))
 }
 
-#[test]
-fn test_compute_num_bits() {
-    assert_eq!(compute_num_bits(1), 1u8);
-    assert_eq!(compute_num_bits(0), 0u8);
-    assert_eq!(compute_num_bits(2), 2u8);
-    assert_eq!(compute_num_bits(3), 2u8);
-    assert_eq!(compute_num_bits(4), 3u8);
-    assert_eq!(compute_num_bits(255), 8u8);
-    assert_eq!(compute_num_bits(256), 9u8);
-    assert_eq!(compute_num_bits(5_000_000_000), 33u8);
-}
+#[cfg(test)]
+mod tests {
+    use super::*;
 
-#[test]
-fn test_minmax_empty() {
-    let vals: Vec<u32> = vec![];
-    assert_eq!(minmax(vals.into_iter()), None);
-}
+    #[test]
+    fn test_compute_num_bits() {
+        assert_eq!(compute_num_bits(1), 1u8);
+        assert_eq!(compute_num_bits(0), 0u8);
+        assert_eq!(compute_num_bits(2), 2u8);
+        assert_eq!(compute_num_bits(3), 2u8);
+        assert_eq!(compute_num_bits(4), 3u8);
+        assert_eq!(compute_num_bits(255), 8u8);
+        assert_eq!(compute_num_bits(256), 9u8);
+        assert_eq!(compute_num_bits(5_000_000_000), 33u8);
+    }
 
-#[test]
-fn test_minmax_one() {
-    assert_eq!(minmax(vec![1].into_iter()), Some((1, 1)));
-}
+    #[test]
+    fn test_minmax_empty() {
+        let vals: Vec<u32> = vec![];
+        assert_eq!(minmax(vals.into_iter()), None);
+    }
 
-#[test]
-fn test_minmax_two() {
-    assert_eq!(minmax(vec![1, 2].into_iter()), Some((1, 2)));
-    assert_eq!(minmax(vec![2, 1].into_iter()), Some((1, 2)));
+    #[test]
+    fn test_minmax_one() {
+        assert_eq!(minmax(vec![1].into_iter()), Some((1, 1)));
+    }
+
+    #[test]
+    fn test_minmax_two() {
+        assert_eq!(minmax(vec![1, 2].into_iter()), Some((1, 2)));
+        assert_eq!(minmax(vec![2, 1].into_iter()), Some((1, 2)));
+    }
+
+    #[test]
+    fn test_minmax_nan() {
+        assert_eq!(
+            minmax(vec![f64::NAN, 1f64, 2f64].into_iter()),
+            Some((1f64, 2f64))
+        );
+        assert_eq!(
+            minmax(vec![2f64, f64::NAN, 1f64].into_iter()),
+            Some((1f64, 2f64))
+        );
+        assert_eq!(
+            minmax(vec![2f64, 1f64, f64::NAN].into_iter()),
+            Some((1f64, 2f64))
+        );
+    }
+
+    #[test]
+    fn test_minmax_inf() {
+        assert_eq!(
+            minmax(vec![f64::INFINITY, 1f64, 2f64].into_iter()),
+            Some((1f64, f64::INFINITY))
+        );
+        assert_eq!(
+            minmax(vec![-f64::INFINITY, 1f64, 2f64].into_iter()),
+            Some((-f64::INFINITY, 2f64))
+        );
+        assert_eq!(
+            minmax(vec![2f64, f64::INFINITY, 1f64].into_iter()),
+            Some((1f64, f64::INFINITY))
+        );
+        assert_eq!(
+            minmax(vec![2f64, 1f64, -f64::INFINITY].into_iter()),
+            Some((-f64::INFINITY, 2f64))
+        );
+    }
 }

columnar/Cargo.toml

@@ -0,0 +1,18 @@
+[package]
+name = "tantivy-columnar"
+version = "0.1.0"
+edition = "2021"
+license = "MIT"
+
+[dependencies]
+stacker = { path = "../stacker", package="tantivy-stacker"}
+serde_json = "1"
+thiserror = "1"
+fnv = "1"
+sstable = { path = "../sstable", package = "tantivy-sstable" }
+common = { path = "../common", package = "tantivy-common" }
+fastfield_codecs = { path = "../fastfield_codecs"}
+itertools = "0.10"
+
+[dev-dependencies]
+proptest = "1"

columnar/README.md

@@ -0,0 +1,67 @@
+# Columnar format
+
+This crate describes columnar format used in tantivy.
+
+## Goals
+
+This format is special in the following way.
+- it needs to be compact
+- it does not required to be loaded in memory.
+- it is designed to fit well with quickwit's strange constraint:
+we need to be able to load columns rapidly.
+- columns of several types can be associated with the same column name.
+- it needs to support columns with different types `(str, u64, i64, f64)`
+and different cardinality `(required, optional, multivalued)`.
+- columns, once loaded, offer cheap random access.
+
+# Coercion rules
+
+Users can create a columnar by inserting rows to a `ColumnarWriter`,
+and serializing it into a `Write` object.
+Nothing prevents a user from recording values with different type to the same `column_name`.
+
+In that case, `tantivy-columnar`'s behavior is as follows:
+- JsonValues are grouped into 3 types (String, Number, bool).
+Values that corresponds to different groups are mapped to different columns. For instance, String values are treated independently
+from Number or boolean values. `tantivy-columnar` will simply emit several columns associated to a given column_name.
+- Only one column for a given json value type is emitted.  If number values with different number types are recorded (e.g. u64, i64, f64),
+`tantivy-columnar` will pick the first type that can represents the set of appended value, with the following prioriy order (`i64`, `u64`, `f64`).
+`i64` is picked over `u64` as it is likely to  yield less change of types. Most use cases strictly requiring `u64` show the
+restriction on 50% of the values (e.g. a 64-bit hash). On the other hand, a lot of use cases can show rare negative value.
+
+# Columnar format
+
+This columnar format may have more than one column (with different types) associated to the same `column_name` (see [Coercion rules](#coercion-rules) above).
+The `(column_name, columne_type)` couple however uniquely identifies a column.
+That couple is serialized as a column `column_key`.  The format of that key is:
+`[column_name][ZERO_BYTE][column_type_header: u8]`
+
+```
+COLUMNAR:=
+    [COLUMNAR_DATA]
+    [COLUMNAR_KEY_TO_DATA_INDEX]
+    [COLUMNAR_FOOTER];
+
+
+# Columns are sorted by their column key.
+COLUMNAR_DATA:=
+    [COLUMN_DATA]+;
+
+COLUMNAR_FOOTER := [RANGE_SSTABLE_BYTES_LEN: 8 bytes little endian]
+
+```
+
+The columnar file starts by the actual column data, concatenated one after the other,
+sorted by column key.
+
+A sstable associates
+`(column name, column_cardinality, column_type) to range of bytes.
+
+Column name may not contain the zero byte `\0`.
+
+Listing all columns associated to `column_name` can therefore
+be done by listing all keys prefixed by
+`[column_name][ZERO_BYTE]`
+
+The associated range of bytes refer to a range of bytes
+

columnar/src/column_type_header.rs

@@ -0,0 +1,201 @@
+use crate::utils::{place_bits, select_bits};
+use crate::value::NumericalType;
+use crate::InvalidData;
+
+/// Enum describing the number of values that can exist per document
+/// (or per row if you will).
+///
+/// The cardinality must fit on 2 bits.
+#[derive(Clone, Copy, Hash, Default, Debug, PartialEq, Eq, PartialOrd, Ord)]
+#[repr(u8)]
+pub enum Cardinality {
+    /// All documents contain exactly one value.
+    /// Required is the default for auto-detecting the Cardinality, since it is the most strict.
+    #[default]
+    Required = 0,
+    /// All documents contain at most one value.
+    Optional = 1,
+    /// All documents may contain any number of values.
+    Multivalued = 2,
+}
+
+impl Cardinality {
+    pub(crate) fn to_code(self) -> u8 {
+        self as u8
+    }
+
+    pub(crate) fn try_from_code(code: u8) -> Result<Cardinality, InvalidData> {
+        match code {
+            0 => Ok(Cardinality::Required),
+            1 => Ok(Cardinality::Optional),
+            2 => Ok(Cardinality::Multivalued),
+            _ => Err(InvalidData),
+        }
+    }
+}
+
+/// The column type represents the column type and can fit on 6-bits.
+///
+/// - bits[0..3]: Column category type.
+/// - bits[3..6]: Numerical type if necessary.
+#[derive(Hash, Eq, PartialEq, Debug, Clone, Copy)]
+pub enum ColumnType {
+    Bytes,
+    Numerical(NumericalType),
+    Bool,
+}
+
+impl ColumnType {
+    /// Encoded over 6 bits.
+    pub(crate) fn to_code(self) -> u8 {
+        let column_type_category;
+        let numerical_type_code: u8;
+        match self {
+            ColumnType::Bytes => {
+                column_type_category = ColumnTypeCategory::Str;
+                numerical_type_code = 0u8;
+            }
+            ColumnType::Numerical(numerical_type) => {
+                column_type_category = ColumnTypeCategory::Numerical;
+                numerical_type_code = numerical_type.to_code();
+            }
+            ColumnType::Bool => {
+                column_type_category = ColumnTypeCategory::Bool;
+                numerical_type_code = 0u8;
+            }
+        }
+        place_bits::<0, 3>(column_type_category.to_code()) | place_bits::<3, 6>(numerical_type_code)
+    }
+
+    pub(crate) fn try_from_code(code: u8) -> Result<ColumnType, InvalidData> {
+        if select_bits::<6, 8>(code) != 0u8 {
+            return Err(InvalidData);
+        }
+        let column_type_category_code = select_bits::<0, 3>(code);
+        let numerical_type_code = select_bits::<3, 6>(code);
+        let column_type_category = ColumnTypeCategory::try_from_code(column_type_category_code)?;
+        match column_type_category {
+            ColumnTypeCategory::Bool => {
+                if numerical_type_code != 0u8 {
+                    return Err(InvalidData);
+                }
+                Ok(ColumnType::Bool)
+            }
+            ColumnTypeCategory::Str => {
+                if numerical_type_code != 0u8 {
+                    return Err(InvalidData);
+                }
+                Ok(ColumnType::Bytes)
+            }
+            ColumnTypeCategory::Numerical => {
+                let numerical_type = NumericalType::try_from_code(numerical_type_code)?;
+                Ok(ColumnType::Numerical(numerical_type))
+            }
+        }
+    }
+}
+
+/// Column types are grouped into different categories that
+/// corresponds to the different types of `JsonValue` types.
+///
+/// The columnar writer will apply coercion rules to make sure that
+/// at most one column exist per `ColumnTypeCategory`.
+///
+/// See also [README.md].
+#[derive(Copy, Clone, Ord, PartialOrd, Eq, PartialEq, Debug)]
+#[repr(u8)]
+pub(crate) enum ColumnTypeCategory {
+    Bool = 0u8,
+    Str = 1u8,
+    Numerical = 2u8,
+}
+
+impl ColumnTypeCategory {
+    pub fn to_code(self) -> u8 {
+        self as u8
+    }
+
+    pub fn try_from_code(code: u8) -> Result<Self, InvalidData> {
+        match code {
+            0u8 => Ok(Self::Bool),
+            1u8 => Ok(Self::Str),
+            2u8 => Ok(Self::Numerical),
+            _ => Err(InvalidData),
+        }
+    }
+}
+
+/// Represents the type and cardinality of a column.
+/// This is encoded over one-byte and added to a column key in the
+/// columnar sstable.
+///
+/// - [0..6] bits: encodes the column type
+/// - [6..8] bits: encodes the cardinality
+#[derive(Eq, Hash, PartialEq, Debug, Copy, Clone)]
+pub struct ColumnTypeAndCardinality {
+    pub typ: ColumnType,
+    pub cardinality: Cardinality,
+}
+
+impl ColumnTypeAndCardinality {
+    pub fn to_code(self) -> u8 {
+        place_bits::<0, 6>(self.typ.to_code()) | place_bits::<6, 8>(self.cardinality.to_code())
+    }
+
+    pub fn try_from_code(code: u8) -> Result<ColumnTypeAndCardinality, InvalidData> {
+        let typ_code = select_bits::<0, 6>(code);
+        let cardinality_code = select_bits::<6, 8>(code);
+        let cardinality = Cardinality::try_from_code(cardinality_code)?;
+        let typ = ColumnType::try_from_code(typ_code)?;
+        assert_eq!(typ.to_code(), typ_code);
+        Ok(ColumnTypeAndCardinality { cardinality, typ })
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::collections::HashSet;
+
+    use super::ColumnTypeAndCardinality;
+    use crate::column_type_header::{Cardinality, ColumnType};
+
+    #[test]
+    fn test_column_type_header_to_code() {
+        let mut column_type_header_set: HashSet<ColumnTypeAndCardinality> = HashSet::new();
+        for code in u8::MIN..=u8::MAX {
+            if let Ok(column_type_header) = ColumnTypeAndCardinality::try_from_code(code) {
+                assert_eq!(column_type_header.to_code(), code);
+                assert!(column_type_header_set.insert(column_type_header));
+            }
+        }
+        assert_eq!(
+            column_type_header_set.len(),
+            3 /* cardinality */ *
+            (1 + 1 + 3) // column_types (str, bool, numerical x 3)
+        );
+    }
+
+    #[test]
+    fn test_column_type_to_code() {
+        let mut column_type_set: HashSet<ColumnType> = HashSet::new();
+        for code in u8::MIN..=u8::MAX {
+            if let Ok(column_type) = ColumnType::try_from_code(code) {
+                assert_eq!(column_type.to_code(), code);
+                assert!(column_type_set.insert(column_type));
+            }
+        }
+        assert_eq!(column_type_set.len(), 2 + 3);
+    }
+
+    #[test]
+    fn test_cardinality_to_code() {
+        let mut num_cardinality = 0;
+        for code in u8::MIN..=u8::MAX {
+            if let Ok(cardinality) = Cardinality::try_from_code(code) {
+                assert_eq!(cardinality.to_code(), code);
+                num_cardinality += 1;
+            }
+        }
+        assert_eq!(num_cardinality, 3);
+    }
+}

columnar/src/dictionary.rs

@@ -0,0 +1,84 @@
+use std::io;
+
+use fnv::FnvHashMap;
+use sstable::SSTable;
+
+pub(crate) struct TermIdMapping {
+    unordered_to_ord: Vec<OrderedId>,
+}
+
+impl TermIdMapping {
+    pub fn to_ord(&self, unordered: UnorderedId) -> OrderedId {
+        self.unordered_to_ord[unordered.0 as usize]
+    }
+}
+
+/// When we add values, we cannot know their ordered id yet.
+/// For this reason, we temporarily assign them a `UnorderedId`
+/// that will be mapped to an `OrderedId` upon serialization.
+#[derive(Clone, Copy, Debug, Hash, PartialEq, Eq)]
+pub struct UnorderedId(pub u32);
+
+#[derive(Clone, Copy, Hash, PartialEq, Eq, Debug)]
+pub struct OrderedId(pub u32);
+
+/// `DictionaryBuilder` for dictionary encoding.
+///
+/// It stores the different terms encounterred and assigns them a temporary value
+/// we call unordered id.
+///
+/// Upon serialization, we will sort the ids and hence build a `UnorderedId -> Term ordinal`
+/// mapping.
+#[derive(Default)]
+pub(crate) struct DictionaryBuilder {
+    dict: FnvHashMap<Vec<u8>, UnorderedId>,
+}
+
+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);
+        new_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();
+        terms.sort_unstable_by_key(|(key, _)| *key);
+        // TODO Remove the allocation.
+        let mut unordered_to_ord: Vec<OrderedId> = vec![OrderedId(0u32); terms.len()];
+        let mut sstable_builder = sstable::VoidSSTable::writer(wrt);
+        for (ord, (key, unordered_id)) in terms.into_iter().enumerate() {
+            let ordered_id = OrderedId(ord as u32);
+            sstable_builder.insert(key, &())?;
+            unordered_to_ord[unordered_id.0 as usize] = ordered_id;
+        }
+        sstable_builder.finish()?;
+        Ok(TermIdMapping { unordered_to_ord })
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_dictionary_builder() {
+        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 mut buffer = Vec::new();
+        let id_mapping = dictionary_builder.serialize(&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

@@ -0,0 +1,89 @@
+mod column_type_header;
+mod dictionary;
+mod reader;
+pub(crate) mod utils;
+mod value;
+mod writer;
+
+pub use column_type_header::Cardinality;
+pub use reader::ColumnarReader;
+pub use value::{NumericalType, NumericalValue};
+pub use writer::ColumnarWriter;
+
+pub type DocId = u32;
+
+#[derive(Copy, Clone, Debug)]
+pub struct InvalidData;
+
+#[cfg(test)]
+mod tests {
+    use std::ops::Range;
+
+    use common::file_slice::FileSlice;
+
+    use crate::column_type_header::{ColumnType, ColumnTypeAndCardinality};
+    use crate::reader::ColumnarReader;
+    use crate::value::NumericalValue;
+    use crate::{Cardinality, ColumnarWriter};
+
+    #[test]
+    fn test_dataframe_writer_bytes() {
+        let mut dataframe_writer = ColumnarWriter::default();
+        dataframe_writer.record_str(1u32, "my_string", "hello");
+        dataframe_writer.record_str(3u32, "my_string", "helloeee");
+        let mut buffer: Vec<u8> = Vec::new();
+        dataframe_writer.serialize(5, &mut buffer).unwrap();
+        let columnar_fileslice = FileSlice::from(buffer);
+        let columnar = ColumnarReader::open(columnar_fileslice).unwrap();
+        assert_eq!(columnar.num_columns(), 1);
+        let cols: Vec<(ColumnTypeAndCardinality, Range<u64>)> =
+            columnar.read_columns("my_string").unwrap();
+        assert_eq!(cols.len(), 1);
+        assert_eq!(cols[0].1, 0..158);
+    }
+
+    #[test]
+    fn test_dataframe_writer_bool() {
+        let mut dataframe_writer = ColumnarWriter::default();
+        dataframe_writer.record_bool(1u32, "bool.value", false);
+        let mut buffer: Vec<u8> = Vec::new();
+        dataframe_writer.serialize(5, &mut buffer).unwrap();
+        let columnar_fileslice = FileSlice::from(buffer);
+        let columnar = ColumnarReader::open(columnar_fileslice).unwrap();
+        assert_eq!(columnar.num_columns(), 1);
+        let cols: Vec<(ColumnTypeAndCardinality, Range<u64>)> =
+            columnar.read_columns("bool.value").unwrap();
+        assert_eq!(cols.len(), 1);
+        assert_eq!(
+            cols[0].0,
+            ColumnTypeAndCardinality {
+                cardinality: Cardinality::Optional,
+                typ: ColumnType::Bool
+            }
+        );
+        assert_eq!(cols[0].1, 0..21);
+    }
+
+    #[test]
+    fn test_dataframe_writer_numerical() {
+        let mut dataframe_writer = ColumnarWriter::default();
+        dataframe_writer.record_numerical(1u32, "srical.value", NumericalValue::U64(12u64));
+        dataframe_writer.record_numerical(2u32, "srical.value", NumericalValue::U64(13u64));
+        dataframe_writer.record_numerical(4u32, "srical.value", NumericalValue::U64(15u64));
+        let mut buffer: Vec<u8> = Vec::new();
+        dataframe_writer.serialize(5, &mut buffer).unwrap();
+        let columnar_fileslice = FileSlice::from(buffer);
+        let columnar = ColumnarReader::open(columnar_fileslice).unwrap();
+        assert_eq!(columnar.num_columns(), 1);
+        let cols: Vec<(ColumnTypeAndCardinality, Range<u64>)> =
+            columnar.read_columns("srical.value").unwrap();
+        assert_eq!(cols.len(), 1);
+        // Right now this 31 bytes are spent as follows
+        //
+        // - header 14 bytes
+        // - vals  8 //< due to padding? could have been 1byte?.
+        // - null footer 6 bytes
+        // - version footer 3 bytes // Should be file-wide
+        assert_eq!(cols[0].1, 0..31);
+    }
+}

columnar/src/reader/mod.rs

@@ -0,0 +1,110 @@
+use std::ops::Range;
+use std::{io, mem};
+
+use common::file_slice::FileSlice;
+use common::BinarySerializable;
+use sstable::{Dictionary, RangeSSTable};
+
+use crate::column_type_header::ColumnTypeAndCardinality;
+
+fn io_invalid_data(msg: String) -> io::Error {
+    io::Error::new(io::ErrorKind::InvalidData, msg)
+}
+
+/// The ColumnarReader makes it possible to access a set of columns
+/// associated to field names.
+pub struct ColumnarReader {
+    column_dictionary: Dictionary<RangeSSTable>,
+    column_data: FileSlice,
+}
+
+impl ColumnarReader {
+    /// Opens a new Columnar file.
+    pub fn open<F>(file_slice: F) -> io::Result<ColumnarReader>
+    where FileSlice: From<F> {
+        Self::open_inner(file_slice.into())
+    }
+
+    fn open_inner(file_slice: FileSlice) -> io::Result<ColumnarReader> {
+        let (file_slice_without_sstable_len, sstable_len_bytes) =
+            file_slice.split_from_end(mem::size_of::<u64>());
+        let mut sstable_len_bytes = sstable_len_bytes.read_bytes()?;
+        let sstable_len = u64::deserialize(&mut sstable_len_bytes)?;
+        let (column_data, sstable) =
+            file_slice_without_sstable_len.split_from_end(sstable_len as usize);
+        let column_dictionary = Dictionary::open(sstable)?;
+        Ok(ColumnarReader {
+            column_dictionary,
+            column_data,
+        })
+    }
+
+    // TODO fix ugly API
+    pub fn list_columns(
+        &self,
+    ) -> io::Result<Vec<(String, ColumnTypeAndCardinality, Range<u64>, u64)>> {
+        let mut stream = self.column_dictionary.stream()?;
+        let mut results = Vec::new();
+        while stream.advance() {
+            let key_bytes: &[u8] = stream.key();
+            let column_code: u8 = key_bytes.last().cloned().unwrap();
+            let column_type_and_cardinality = ColumnTypeAndCardinality::try_from_code(column_code)
+                .map_err(|_| io_invalid_data(format!("Unknown column code `{column_code}`")))?;
+            let range = stream.value().clone();
+            let column_name = String::from_utf8_lossy(&key_bytes[..key_bytes.len() - 1]);
+            let range_len = range.end - range.start;
+            results.push((
+                column_name.to_string(),
+                column_type_and_cardinality,
+                range,
+                range_len,
+            ));
+        }
+        Ok(results)
+    }
+
+    /// Get all columns for the given column name.
+    ///
+    /// There can be more than one column associated to a given column name, provided they have
+    /// different types.
+    // TODO fix ugly API
+    pub fn read_columns(
+        &self,
+        column_name: &str,
+    ) -> io::Result<Vec<(ColumnTypeAndCardinality, Range<u64>)>> {
+        // Each column is a associated to a given `column_key`,
+        // that starts by `column_name\0column_header`.
+        //
+        // Listing the columns associated to the given column name is therefore equivalent to
+        // listing `column_key` with the prefix `column_name\0`.
+        //
+        // This is in turn equivalent to searching for the range
+        // `[column_name,\0`..column_name\1)`.
+        let mut start_key = column_name.to_string();
+        start_key.push('\0');
+        let mut end_key = column_name.to_string();
+        end_key.push(1u8 as char);
+        let mut stream = self
+            .column_dictionary
+            .range()
+            .ge(start_key.as_bytes())
+            .lt(end_key.as_bytes())
+            .into_stream()?;
+        let mut results = Vec::new();
+        while stream.advance() {
+            let key_bytes: &[u8] = stream.key();
+            assert!(key_bytes.starts_with(start_key.as_bytes()));
+            let column_code: u8 = key_bytes.last().cloned().unwrap();
+            let column_type_and_cardinality = ColumnTypeAndCardinality::try_from_code(column_code)
+                .map_err(|_| io_invalid_data(format!("Unknown column code `{column_code}`")))?;
+            let range = stream.value().clone();
+            results.push((column_type_and_cardinality, range));
+        }
+        Ok(results)
+    }
+
+    /// Return the number of columns in the columnar.
+    pub fn num_columns(&self) -> usize {
+        self.column_dictionary.num_terms()
+    }
+}

columnar/src/utils.rs

@@ -0,0 +1,76 @@
+const fn compute_mask(num_bits: u8) -> u8 {
+    if num_bits == 8 {
+        u8::MAX
+    } else {
+        (1u8 << num_bits) - 1
+    }
+}
+
+#[inline(always)]
+#[must_use]
+pub(crate) fn select_bits<const START: u8, const END: u8>(code: u8) -> u8 {
+    assert!(START <= END);
+    assert!(END <= 8);
+    let num_bits: u8 = END - START;
+    let mask: u8 = compute_mask(num_bits);
+    (code >> START) & mask
+}
+
+#[inline(always)]
+#[must_use]
+pub(crate) fn place_bits<const START: u8, const END: u8>(code: u8) -> u8 {
+    assert!(START <= END);
+    assert!(END <= 8);
+    let num_bits: u8 = END - START;
+    let mask: u8 = compute_mask(num_bits);
+    assert!(code <= mask);
+    code << START
+}
+
+/// Pop-front one bytes from a slice of bytes.
+#[inline(always)]
+pub fn pop_first_byte(bytes: &mut &[u8]) -> Option<u8> {
+    if bytes.is_empty() {
+        return None;
+    }
+    let first_byte = bytes[0];
+    *bytes = &bytes[1..];
+    Some(first_byte)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_select_bits() {
+        assert_eq!(255u8, select_bits::<0, 8>(255u8));
+        assert_eq!(0u8, select_bits::<0, 0>(255u8));
+        assert_eq!(8u8, select_bits::<0, 4>(8u8));
+        assert_eq!(4u8, select_bits::<1, 4>(8u8));
+        assert_eq!(0u8, select_bits::<1, 3>(8u8));
+    }
+
+    #[test]
+    fn test_place_bits() {
+        assert_eq!(255u8, place_bits::<0, 8>(255u8));
+        assert_eq!(4u8, place_bits::<2, 3>(1u8));
+        assert_eq!(0u8, place_bits::<2, 2>(0u8));
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_place_bits_overflows() {
+        let _ = place_bits::<1, 4>(8u8);
+    }
+
+    #[test]
+    fn test_pop_first_byte() {
+        let mut cursor: &[u8] = &b"abcd"[..];
+        assert_eq!(pop_first_byte(&mut cursor), Some(b'a'));
+        assert_eq!(pop_first_byte(&mut cursor), Some(b'b'));
+        assert_eq!(pop_first_byte(&mut cursor), Some(b'c'));
+        assert_eq!(pop_first_byte(&mut cursor), Some(b'd'));
+        assert_eq!(pop_first_byte(&mut cursor), None);
+    }
+}

columnar/src/value.rs

@@ -0,0 +1,124 @@
+use crate::InvalidData;
+
+#[derive(Copy, Clone, Debug, PartialEq)]
+pub enum NumericalValue {
+    I64(i64),
+    U64(u64),
+    F64(f64),
+}
+
+impl From<u64> for NumericalValue {
+    fn from(val: u64) -> NumericalValue {
+        NumericalValue::U64(val)
+    }
+}
+
+impl From<i64> for NumericalValue {
+    fn from(val: i64) -> Self {
+        NumericalValue::I64(val)
+    }
+}
+
+impl From<f64> for NumericalValue {
+    fn from(val: f64) -> Self {
+        NumericalValue::F64(val)
+    }
+}
+
+impl NumericalValue {
+    pub fn numerical_type(&self) -> NumericalType {
+        match self {
+            NumericalValue::F64(_) => NumericalType::F64,
+            NumericalValue::I64(_) => NumericalType::I64,
+            NumericalValue::U64(_) => NumericalType::U64,
+        }
+    }
+}
+
+impl Eq for NumericalValue {}
+
+#[derive(Clone, Copy, Debug, Default, Hash, Eq, PartialEq)]
+#[repr(u8)]
+pub enum NumericalType {
+    #[default]
+    I64 = 0,
+    U64 = 1,
+    F64 = 2,
+}
+
+impl NumericalType {
+    pub fn to_code(self) -> u8 {
+        self as u8
+    }
+
+    pub fn try_from_code(code: u8) -> Result<NumericalType, InvalidData> {
+        match code {
+            0 => Ok(NumericalType::I64),
+            1 => Ok(NumericalType::U64),
+            2 => Ok(NumericalType::F64),
+            _ => Err(InvalidData),
+        }
+    }
+}
+
+/// We voluntarily avoid using `Into` here to keep this
+/// implementation quirk as private as possible.
+///
+/// # Panics
+/// This coercion trait actually panics if it is used
+/// to convert a loose types to a stricter type.
+///
+/// The level is strictness is somewhat arbitrary.
+/// - i64
+/// - u64
+/// - f64.
+pub(crate) trait Coerce {
+    fn coerce(numerical_value: NumericalValue) -> Self;
+}
+
+impl Coerce for i64 {
+    fn coerce(value: NumericalValue) -> Self {
+        match value {
+            NumericalValue::I64(val) => val,
+            NumericalValue::U64(val) => val as i64,
+            NumericalValue::F64(_) => unreachable!(),
+        }
+    }
+}
+
+impl Coerce for u64 {
+    fn coerce(value: NumericalValue) -> Self {
+        match value {
+            NumericalValue::I64(val) => val as u64,
+            NumericalValue::U64(val) => val,
+            NumericalValue::F64(_) => unreachable!(),
+        }
+    }
+}
+
+impl Coerce for f64 {
+    fn coerce(value: NumericalValue) -> Self {
+        match value {
+            NumericalValue::I64(val) => val as f64,
+            NumericalValue::U64(val) => val as f64,
+            NumericalValue::F64(val) => val,
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::NumericalType;
+
+    #[test]
+    fn test_numerical_type_code() {
+        let mut num_numerical_type = 0;
+        for code in u8::MIN..=u8::MAX {
+            if let Ok(numerical_type) = NumericalType::try_from_code(code) {
+                assert_eq!(numerical_type.to_code(), code);
+                num_numerical_type += 1;
+            }
+        }
+        assert_eq!(num_numerical_type, 3);
+    }
+}

columnar/src/writer/column_operation.rs

@@ -0,0 +1,346 @@
+use crate::dictionary::UnorderedId;
+use crate::utils::{place_bits, pop_first_byte, select_bits};
+use crate::value::NumericalValue;
+use crate::{DocId, InvalidData, NumericalType};
+
+/// When we build a columnar dataframe, we first just group
+/// all mutations per column, and appends them in append-only buffer
+/// in the stacker.
+///
+/// These ColumnOperation<T> are therefore serialize/deserialized
+/// in memory.
+///
+/// We represents all of these operations as `ColumnOperation`.
+#[derive(Eq, PartialEq, Debug, Clone, Copy)]
+pub(super) enum ColumnOperation<T> {
+    NewDoc(DocId),
+    Value(T),
+}
+
+#[derive(Copy, Clone, Eq, PartialEq, Debug)]
+struct ColumnOperationMetadata {
+    op_type: ColumnOperationType,
+    len: u8,
+}
+
+impl ColumnOperationMetadata {
+    fn to_code(self) -> u8 {
+        place_bits::<0, 4>(self.len) | place_bits::<4, 8>(self.op_type.to_code())
+    }
+
+    fn try_from_code(code: u8) -> Result<Self, InvalidData> {
+        let len = select_bits::<0, 4>(code);
+        let typ_code = select_bits::<4, 8>(code);
+        let column_type = ColumnOperationType::try_from_code(typ_code)?;
+        Ok(ColumnOperationMetadata {
+            op_type: column_type,
+            len,
+        })
+    }
+}
+
+#[derive(Copy, Clone, Eq, PartialEq, Debug)]
+#[repr(u8)]
+enum ColumnOperationType {
+    NewDoc = 0u8,
+    AddValue = 1u8,
+}
+
+impl ColumnOperationType {
+    pub fn to_code(self) -> u8 {
+        self as u8
+    }
+
+    pub fn try_from_code(code: u8) -> Result<Self, InvalidData> {
+        match code {
+            0 => Ok(Self::NewDoc),
+            1 => Ok(Self::AddValue),
+            _ => Err(InvalidData),
+        }
+    }
+}
+
+impl<V: SymbolValue> ColumnOperation<V> {
+    pub(super) fn serialize(self) -> impl AsRef<[u8]> {
+        let mut minibuf = MiniBuffer::default();
+        let column_op_metadata = match self {
+            ColumnOperation::NewDoc(new_doc) => {
+                let symbol_len = new_doc.serialize(&mut minibuf.bytes[1..]);
+                ColumnOperationMetadata {
+                    op_type: ColumnOperationType::NewDoc,
+                    len: symbol_len,
+                }
+            }
+            ColumnOperation::Value(val) => {
+                let symbol_len = val.serialize(&mut minibuf.bytes[1..]);
+                ColumnOperationMetadata {
+                    op_type: ColumnOperationType::AddValue,
+                    len: symbol_len,
+                }
+            }
+        };
+        minibuf.bytes[0] = column_op_metadata.to_code();
+        // +1 for the metadata
+        minibuf.len = 1 + column_op_metadata.len;
+        minibuf
+    }
+
+    /// Deserialize a colummn operation.
+    /// Returns None if the buffer is empty.
+    ///
+    /// Panics if the payload is invalid:
+    /// this deserialize method is meant to target in memory.
+    pub(super) fn deserialize(bytes: &mut &[u8]) -> Option<Self> {
+        let column_op_metadata_byte = pop_first_byte(bytes)?;
+        let column_op_metadata = ColumnOperationMetadata::try_from_code(column_op_metadata_byte)
+            .expect("Invalid op metadata byte");
+        let symbol_bytes: &[u8];
+        (symbol_bytes, *bytes) = bytes.split_at(column_op_metadata.len as usize);
+        match column_op_metadata.op_type {
+            ColumnOperationType::NewDoc => {
+                let new_doc = u32::deserialize(symbol_bytes);
+                Some(ColumnOperation::NewDoc(new_doc))
+            }
+            ColumnOperationType::AddValue => {
+                let value = V::deserialize(symbol_bytes);
+                Some(ColumnOperation::Value(value))
+            }
+        }
+    }
+}
+
+impl<T> From<T> for ColumnOperation<T> {
+    fn from(value: T) -> Self {
+        ColumnOperation::Value(value)
+    }
+}
+
+// Serialization trait very local to the writer.
+// As we write fast fields, we accumulate them in "in memory".
+// In order to limit memory usage, and in order
+// to benefit from the stacker, we do this by serialization our data
+// as "Symbols".
+#[allow(clippy::from_over_into)]
+pub(super) trait SymbolValue: Clone + Copy {
+    // Serializes the symbol into the given buffer.
+    // Returns the number of bytes written into the buffer.
+    /// # Panics
+    /// May not exceed 9bytes
+    fn serialize(self, buffer: &mut [u8]) -> u8;
+    // Panics if invalid
+    fn deserialize(bytes: &[u8]) -> Self;
+}
+
+impl SymbolValue for bool {
+    fn serialize(self, buffer: &mut [u8]) -> u8 {
+        buffer[0] = u8::from(self);
+        1u8
+    }
+
+    fn deserialize(bytes: &[u8]) -> Self {
+        bytes[0] == 1u8
+    }
+}
+
+#[derive(Default)]
+struct MiniBuffer {
+    pub bytes: [u8; 10],
+    pub len: u8,
+}
+
+impl AsRef<[u8]> for MiniBuffer {
+    fn as_ref(&self) -> &[u8] {
+        &self.bytes[..self.len as usize]
+    }
+}
+
+impl SymbolValue for NumericalValue {
+    fn deserialize(mut bytes: &[u8]) -> Self {
+        let type_code = pop_first_byte(&mut bytes).unwrap();
+        let symbol_type = NumericalType::try_from_code(type_code).unwrap();
+        let mut octet: [u8; 8] = [0u8; 8];
+        octet[..bytes.len()].copy_from_slice(bytes);
+        match symbol_type {
+            NumericalType::U64 => {
+                let val: u64 = u64::from_le_bytes(octet);
+                NumericalValue::U64(val)
+            }
+            NumericalType::I64 => {
+                let encoded: u64 = u64::from_le_bytes(octet);
+                let val: i64 = decode_zig_zag(encoded);
+                NumericalValue::I64(val)
+            }
+            NumericalType::F64 => {
+                debug_assert_eq!(bytes.len(), 8);
+                let val: f64 = f64::from_le_bytes(octet);
+                NumericalValue::F64(val)
+            }
+        }
+    }
+
+    /// F64: Serialize with a fixed size of 9 bytes
+    /// U64: Serialize without leading zeroes
+    /// I64: ZigZag encoded and serialize without leading zeroes
+    fn serialize(self, output: &mut [u8]) -> u8 {
+        match self {
+            NumericalValue::F64(val) => {
+                output[0] = NumericalType::F64 as u8;
+                output[1..9].copy_from_slice(&val.to_le_bytes());
+                9u8
+            }
+            NumericalValue::U64(val) => {
+                let len = compute_num_bytes_for_u64(val) as u8;
+                output[0] = NumericalType::U64 as u8;
+                output[1..9].copy_from_slice(&val.to_le_bytes());
+                len + 1u8
+            }
+            NumericalValue::I64(val) => {
+                let zig_zag_encoded = encode_zig_zag(val);
+                let len = compute_num_bytes_for_u64(zig_zag_encoded) as u8;
+                output[0] = NumericalType::I64 as u8;
+                output[1..9].copy_from_slice(&zig_zag_encoded.to_le_bytes());
+                len + 1u8
+            }
+        }
+    }
+}
+
+impl SymbolValue for u32 {
+    fn serialize(self, output: &mut [u8]) -> u8 {
+        let len = compute_num_bytes_for_u64(self as u64);
+        output[0..4].copy_from_slice(&self.to_le_bytes());
+        len as u8
+    }
+
+    fn deserialize(bytes: &[u8]) -> Self {
+        let mut quartet: [u8; 4] = [0u8; 4];
+        quartet[..bytes.len()].copy_from_slice(bytes);
+        u32::from_le_bytes(quartet)
+    }
+}
+
+impl SymbolValue for UnorderedId {
+    fn serialize(self, output: &mut [u8]) -> u8 {
+        self.0.serialize(output)
+    }
+
+    fn deserialize(bytes: &[u8]) -> Self {
+        UnorderedId(u32::deserialize(bytes))
+    }
+}
+
+fn compute_num_bytes_for_u64(val: u64) -> usize {
+    let msb = (64u32 - val.leading_zeros()) as usize;
+    (msb + 7) / 8
+}
+
+fn encode_zig_zag(n: i64) -> u64 {
+    ((n << 1) ^ (n >> 63)) as u64
+}
+
+fn decode_zig_zag(n: u64) -> i64 {
+    ((n >> 1) as i64) ^ (-((n & 1) as i64))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[track_caller]
+    fn test_zig_zag_aux(val: i64) {
+        let encoded = super::encode_zig_zag(val);
+        assert_eq!(decode_zig_zag(encoded), val);
+        if let Some(abs_val) = val.checked_abs() {
+            let abs_val = abs_val as u64;
+            assert!(encoded <= abs_val * 2);
+        }
+    }
+
+    #[test]
+    fn test_zig_zag() {
+        assert_eq!(encode_zig_zag(0i64), 0u64);
+        assert_eq!(encode_zig_zag(-1i64), 1u64);
+        assert_eq!(encode_zig_zag(1i64), 2u64);
+        test_zig_zag_aux(0i64);
+        test_zig_zag_aux(i64::MIN);
+        test_zig_zag_aux(i64::MAX);
+    }
+
+    use proptest::prelude::any;
+    use proptest::proptest;
+
+    proptest! {
+        #[test]
+        fn test_proptest_zig_zag(val in any::<i64>()) {
+            test_zig_zag_aux(val);
+        }
+    }
+
+    #[test]
+    fn test_column_op_metadata_byte_serialization() {
+        for len in 0..=15 {
+            for op_type in [ColumnOperationType::AddValue, ColumnOperationType::NewDoc] {
+                let column_op_metadata = ColumnOperationMetadata { op_type, len };
+                let column_op_metadata_code = column_op_metadata.to_code();
+                let serdeser_metadata =
+                    ColumnOperationMetadata::try_from_code(column_op_metadata_code).unwrap();
+                assert_eq!(column_op_metadata, serdeser_metadata);
+            }
+        }
+    }
+
+    #[track_caller]
+    fn ser_deser_symbol(column_op: ColumnOperation<NumericalValue>) {
+        let buf = column_op.serialize();
+        let mut buffer = buf.as_ref().to_vec();
+        buffer.extend_from_slice(b"234234");
+        let mut bytes = &buffer[..];
+        let serdeser_symbol = ColumnOperation::deserialize(&mut bytes).unwrap();
+        assert_eq!(bytes.len() + buf.as_ref().len() as usize, buffer.len());
+        assert_eq!(column_op, serdeser_symbol);
+    }
+
+    #[test]
+    fn test_compute_num_bytes_for_u64() {
+        assert_eq!(compute_num_bytes_for_u64(0), 0);
+        assert_eq!(compute_num_bytes_for_u64(1), 1);
+        assert_eq!(compute_num_bytes_for_u64(255), 1);
+        assert_eq!(compute_num_bytes_for_u64(256), 2);
+        assert_eq!(compute_num_bytes_for_u64((1 << 16) - 1), 2);
+        assert_eq!(compute_num_bytes_for_u64(1 << 16), 3);
+    }
+
+    #[test]
+    fn test_symbol_serialization() {
+        ser_deser_symbol(ColumnOperation::NewDoc(0));
+        ser_deser_symbol(ColumnOperation::NewDoc(3));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::I64(0i64)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::I64(1i64)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::U64(257u64)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::I64(-257i64)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::I64(i64::MIN)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::U64(0u64)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::U64(u64::MIN)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::U64(u64::MAX)));
+    }
+
+    fn test_column_operation_unordered_aux(val: u32, expected_len: usize) {
+        let column_op = ColumnOperation::Value(UnorderedId(val));
+        let minibuf = column_op.serialize();
+        assert_eq!(minibuf.as_ref().len() as usize, expected_len);
+        let mut buf = minibuf.as_ref().to_vec();
+        buf.extend_from_slice(&[2, 2, 2, 2, 2, 2]);
+        let mut cursor = &buf[..];
+        let column_op_serdeser: ColumnOperation<UnorderedId> =
+            ColumnOperation::deserialize(&mut cursor).unwrap();
+        assert_eq!(column_op_serdeser, ColumnOperation::Value(UnorderedId(val)));
+        assert_eq!(cursor.len() + expected_len, buf.len());
+    }
+
+    #[test]
+    fn test_column_operation_unordered() {
+        test_column_operation_unordered_aux(300u32, 3);
+        test_column_operation_unordered_aux(1u32, 2);
+        test_column_operation_unordered_aux(0u32, 1);
+    }
+}

columnar/src/writer/column_writers.rs

@@ -0,0 +1,265 @@
+use std::cmp::Ordering;
+
+use stacker::{ExpUnrolledLinkedList, MemoryArena};
+
+use crate::dictionary::{DictionaryBuilder, UnorderedId};
+use crate::writer::column_operation::{ColumnOperation, SymbolValue};
+use crate::{Cardinality, DocId, NumericalType, NumericalValue};
+
+#[derive(Copy, Clone, Debug, Eq, PartialEq)]
+#[repr(u8)]
+enum DocumentStep {
+    Same = 0,
+    Next = 1,
+    Skipped = 2,
+}
+
+#[inline(always)]
+fn delta_with_last_doc(last_doc_opt: Option<u32>, doc: u32) -> DocumentStep {
+    let expected_next_doc = last_doc_opt.map(|last_doc| last_doc + 1).unwrap_or(0u32);
+    match doc.cmp(&expected_next_doc) {
+        Ordering::Less => DocumentStep::Same,
+        Ordering::Equal => DocumentStep::Next,
+        Ordering::Greater => DocumentStep::Skipped,
+    }
+}
+
+#[derive(Copy, Clone, Default)]
+pub struct ColumnWriter {
+    // Detected cardinality of the column so far.
+    cardinality: Cardinality,
+    // Last document inserted.
+    // None if no doc has been added yet.
+    last_doc_opt: Option<u32>,
+    // Buffer containing the serialized values.
+    values: ExpUnrolledLinkedList,
+}
+
+impl ColumnWriter {
+    /// Returns an iterator over the Symbol that have been recorded
+    /// for the given column.
+    pub(super) fn operation_iterator<'a, V: SymbolValue>(
+        &self,
+        arena: &MemoryArena,
+        buffer: &'a mut Vec<u8>,
+    ) -> impl Iterator<Item = ColumnOperation<V>> + 'a {
+        buffer.clear();
+        self.values.read_to_end(arena, buffer);
+        let mut cursor: &[u8] = &buffer[..];
+        std::iter::from_fn(move || ColumnOperation::deserialize(&mut cursor))
+    }
+
+    /// Records a change of the document being recorded.
+    ///
+    /// This function will also update the cardinality of the column
+    /// if necessary.
+    pub(super) fn record<S: SymbolValue>(&mut self, doc: DocId, value: S, arena: &mut MemoryArena) {
+        // Difference between `doc` and the last doc.
+        match delta_with_last_doc(self.last_doc_opt, doc) {
+            DocumentStep::Same => {
+                // This is the last encounterred document.
+                self.cardinality = Cardinality::Multivalued;
+            }
+            DocumentStep::Next => {
+                self.last_doc_opt = Some(doc);
+                self.write_symbol::<S>(ColumnOperation::NewDoc(doc), arena);
+            }
+            DocumentStep::Skipped => {
+                self.cardinality = self.cardinality.max(Cardinality::Optional);
+                self.last_doc_opt = Some(doc);
+                self.write_symbol::<S>(ColumnOperation::NewDoc(doc), arena);
+            }
+        }
+        self.write_symbol(ColumnOperation::Value(value), arena);
+    }
+
+    // Get the cardinality.
+    // The overall number of docs in the column is necessary to
+    // deal with the case where the all docs contain 1 value, except some documents
+    // at the end of the column.
+    pub(crate) fn get_cardinality(&self, num_docs: DocId) -> Cardinality {
+        match delta_with_last_doc(self.last_doc_opt, num_docs) {
+            DocumentStep::Same | DocumentStep::Next => self.cardinality,
+            DocumentStep::Skipped => self.cardinality.max(Cardinality::Optional),
+        }
+    }
+
+    /// Appends a new symbol to the `ColumnWriter`.
+    fn write_symbol<V: SymbolValue>(
+        &mut self,
+        column_operation: ColumnOperation<V>,
+        arena: &mut MemoryArena,
+    ) {
+        self.values
+            .writer(arena)
+            .extend_from_slice(column_operation.serialize().as_ref());
+    }
+}
+
+#[derive(Clone, Copy, Default)]
+pub(crate) struct NumericalColumnWriter {
+    compatible_numerical_types: CompatibleNumericalTypes,
+    column_writer: ColumnWriter,
+}
+
+/// State used to store what types are still acceptable
+/// after having seen a set of numerical values.
+#[derive(Clone, Copy)]
+struct CompatibleNumericalTypes {
+    all_values_within_i64_range: bool,
+    all_values_within_u64_range: bool,
+    // f64 is always acceptable.
+}
+
+impl Default for CompatibleNumericalTypes {
+    fn default() -> CompatibleNumericalTypes {
+        CompatibleNumericalTypes {
+            all_values_within_i64_range: true,
+            all_values_within_u64_range: true,
+        }
+    }
+}
+
+impl CompatibleNumericalTypes {
+    fn accept_value(&mut self, numerical_value: NumericalValue) {
+        match numerical_value {
+            NumericalValue::I64(val_i64) => {
+                let value_within_u64_range = val_i64 >= 0i64;
+                self.all_values_within_u64_range &= value_within_u64_range;
+            }
+            NumericalValue::U64(val_u64) => {
+                let value_within_i64_range = val_u64 < i64::MAX as u64;
+                self.all_values_within_i64_range &= value_within_i64_range;
+            }
+            NumericalValue::F64(_) => {
+                self.all_values_within_i64_range = false;
+                self.all_values_within_u64_range = false;
+            }
+        }
+    }
+
+    pub fn to_numerical_type(self) -> NumericalType {
+        if self.all_values_within_i64_range {
+            NumericalType::I64
+        } else if self.all_values_within_u64_range {
+            NumericalType::U64
+        } else {
+            NumericalType::F64
+        }
+    }
+}
+
+impl NumericalColumnWriter {
+    pub fn column_type_and_cardinality(&self, num_docs: DocId) -> (NumericalType, Cardinality) {
+        let numerical_type = self.compatible_numerical_types.to_numerical_type();
+        let cardinality = self.column_writer.get_cardinality(num_docs);
+        (numerical_type, cardinality)
+    }
+
+    pub fn record_numerical_value(
+        &mut self,
+        doc: DocId,
+        value: NumericalValue,
+        arena: &mut MemoryArena,
+    ) {
+        self.compatible_numerical_types.accept_value(value);
+        self.column_writer.record(doc, value, arena);
+    }
+
+    pub(super) fn operation_iterator<'a>(
+        self,
+        arena: &MemoryArena,
+        buffer: &'a mut Vec<u8>,
+    ) -> impl Iterator<Item = ColumnOperation<NumericalValue>> + 'a {
+        self.column_writer.operation_iterator(arena, buffer)
+    }
+}
+
+#[derive(Copy, Clone, Default)]
+pub(crate) struct StrColumnWriter {
+    pub(crate) dictionary_id: u32,
+    pub(crate) column_writer: ColumnWriter,
+}
+
+impl StrColumnWriter {
+    pub(crate) fn with_dictionary_id(dictionary_id: u32) -> StrColumnWriter {
+        StrColumnWriter {
+            dictionary_id,
+            column_writer: Default::default(),
+        }
+    }
+
+    pub(crate) fn record_bytes(
+        &mut self,
+        doc: DocId,
+        bytes: &[u8],
+        dictionaries: &mut [DictionaryBuilder],
+        arena: &mut MemoryArena,
+    ) {
+        let unordered_id = dictionaries[self.dictionary_id as usize].get_or_allocate_id(bytes);
+        self.column_writer.record(doc, unordered_id, arena);
+    }
+
+    pub(super) fn operation_iterator<'a>(
+        &self,
+        arena: &MemoryArena,
+        byte_buffer: &'a mut Vec<u8>,
+    ) -> impl Iterator<Item = ColumnOperation<UnorderedId>> + 'a {
+        self.column_writer.operation_iterator(arena, byte_buffer)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_delta_with_last_doc() {
+        assert_eq!(delta_with_last_doc(None, 0u32), DocumentStep::Next);
+        assert_eq!(delta_with_last_doc(None, 1u32), DocumentStep::Skipped);
+        assert_eq!(delta_with_last_doc(None, 2u32), DocumentStep::Skipped);
+        assert_eq!(delta_with_last_doc(Some(0u32), 0u32), DocumentStep::Same);
+        assert_eq!(delta_with_last_doc(Some(1u32), 1u32), DocumentStep::Same);
+        assert_eq!(delta_with_last_doc(Some(1u32), 2u32), DocumentStep::Next);
+        assert_eq!(delta_with_last_doc(Some(1u32), 3u32), DocumentStep::Skipped);
+        assert_eq!(delta_with_last_doc(Some(1u32), 4u32), DocumentStep::Skipped);
+    }
+
+    #[track_caller]
+    fn test_column_writer_coercion_iter_aux(
+        values: impl Iterator<Item = NumericalValue>,
+        expected_numerical_type: NumericalType,
+    ) {
+        let mut compatible_numerical_types = CompatibleNumericalTypes::default();
+        for value in values {
+            compatible_numerical_types.accept_value(value);
+        }
+        assert_eq!(
+            compatible_numerical_types.to_numerical_type(),
+            expected_numerical_type
+        );
+    }
+
+    #[track_caller]
+    fn test_column_writer_coercion_aux(
+        values: &[NumericalValue],
+        expected_numerical_type: NumericalType,
+    ) {
+        test_column_writer_coercion_iter_aux(values.iter().copied(), expected_numerical_type);
+        test_column_writer_coercion_iter_aux(values.iter().rev().copied(), expected_numerical_type);
+    }
+
+    #[test]
+    fn test_column_writer_coercion() {
+        test_column_writer_coercion_aux(&[], NumericalType::I64);
+        test_column_writer_coercion_aux(&[1i64.into()], NumericalType::I64);
+        test_column_writer_coercion_aux(&[1u64.into()], NumericalType::I64);
+        // We don't detect exact integer at the moment. We could!
+        test_column_writer_coercion_aux(&[1f64.into()], NumericalType::F64);
+        test_column_writer_coercion_aux(&[u64::MAX.into()], NumericalType::U64);
+        test_column_writer_coercion_aux(&[(i64::MAX as u64).into()], NumericalType::U64);
+        test_column_writer_coercion_aux(&[(1u64 << 63).into()], NumericalType::U64);
+        test_column_writer_coercion_aux(&[1i64.into(), 1u64.into()], NumericalType::I64);
+        test_column_writer_coercion_aux(&[u64::MAX.into(), (-1i64).into()], NumericalType::F64);
+    }
+}

columnar/src/writer/mod.rs

@@ -0,0 +1,516 @@
+mod column_operation;
+mod column_writers;
+mod serializer;
+mod value_index;
+
+use std::io;
+
+use column_operation::ColumnOperation;
+use common::CountingWriter;
+use fastfield_codecs::serialize::ValueIndexInfo;
+use fastfield_codecs::{Column, MonotonicallyMappableToU64, VecColumn};
+use serializer::ColumnarSerializer;
+use stacker::{Addr, ArenaHashMap, MemoryArena};
+
+use crate::column_type_header::{ColumnType, ColumnTypeAndCardinality, ColumnTypeCategory};
+use crate::dictionary::{DictionaryBuilder, TermIdMapping, UnorderedId};
+use crate::value::{Coerce, NumericalType, NumericalValue};
+use crate::writer::column_writers::{ColumnWriter, NumericalColumnWriter, StrColumnWriter};
+use crate::writer::value_index::{IndexBuilder, SpareIndexBuilders};
+use crate::{Cardinality, DocId};
+
+/// This is a set of buffers that are used to temporarily write the values into before passing them
+/// to the fast field codecs.
+#[derive(Default)]
+struct SpareBuffers {
+    value_index_builders: SpareIndexBuilders,
+    i64_values: Vec<i64>,
+    u64_values: Vec<u64>,
+    f64_values: Vec<f64>,
+    bool_values: Vec<bool>,
+}
+
+/// Makes it possible to create a new columnar.
+///
+/// ```rust
+/// use tantivy_columnar::ColumnarWriter;
+///
+/// let mut columnar_writer = ColumnarWriter::default();
+/// columnar_writer.record_str(0u32 /* doc id */, "product_name", "Red backpack");
+/// columnar_writer.record_numerical(0u32 /* doc id */, "price", 10u64);
+/// columnar_writer.record_str(1u32 /* doc id */, "product_name", "Apple");
+/// columnar_writer.record_numerical(0u32 /* doc id */, "price", 10.5f64); //< uh oh we ended up mixing integer and floats.
+/// let mut wrt: Vec<u8> =  Vec::new();
+/// columnar_writer.serialize(2u32, &mut wrt).unwrap();
+/// ```
+pub struct ColumnarWriter {
+    numerical_field_hash_map: ArenaHashMap,
+    bool_field_hash_map: ArenaHashMap,
+    bytes_field_hash_map: ArenaHashMap,
+    arena: MemoryArena,
+    // Dictionaries used to store dictionary-encoded values.
+    dictionaries: Vec<DictionaryBuilder>,
+    buffers: SpareBuffers,
+}
+
+impl Default for ColumnarWriter {
+    fn default() -> Self {
+        ColumnarWriter {
+            numerical_field_hash_map: ArenaHashMap::new(10_000),
+            bool_field_hash_map: ArenaHashMap::new(10_000),
+            bytes_field_hash_map: ArenaHashMap::new(10_000),
+            dictionaries: Vec::new(),
+            arena: MemoryArena::default(),
+            buffers: SpareBuffers::default(),
+        }
+    }
+}
+
+impl ColumnarWriter {
+    pub fn record_numerical<T: Into<NumericalValue> + Copy>(
+        &mut self,
+        doc: DocId,
+        column_name: &str,
+        numerical_value: T,
+    ) {
+        assert!(
+            !column_name.as_bytes().contains(&0u8),
+            "key may not contain the 0 byte"
+        );
+        let (hash_map, arena) = (&mut self.numerical_field_hash_map, &mut self.arena);
+        hash_map.mutate_or_create(
+            column_name.as_bytes(),
+            |column_opt: Option<NumericalColumnWriter>| {
+                let mut column: NumericalColumnWriter = column_opt.unwrap_or_default();
+                column.record_numerical_value(doc, numerical_value.into(), arena);
+                column
+            },
+        );
+    }
+
+    pub fn record_bool(&mut self, doc: DocId, column_name: &str, val: bool) {
+        assert!(
+            !column_name.as_bytes().contains(&0u8),
+            "key may not contain the 0 byte"
+        );
+        let (hash_map, arena) = (&mut self.bool_field_hash_map, &mut self.arena);
+        hash_map.mutate_or_create(
+            column_name.as_bytes(),
+            |column_opt: Option<ColumnWriter>| {
+                let mut column: ColumnWriter = column_opt.unwrap_or_default();
+                column.record(doc, val, arena);
+                column
+            },
+        );
+    }
+
+    pub fn record_str(&mut self, doc: DocId, column_name: &str, value: &str) {
+        assert!(
+            !column_name.as_bytes().contains(&0u8),
+            "key may not contain the 0 byte"
+        );
+        let (hash_map, arena, dictionaries) = (
+            &mut self.bytes_field_hash_map,
+            &mut self.arena,
+            &mut self.dictionaries,
+        );
+        hash_map.mutate_or_create(
+            column_name.as_bytes(),
+            |column_opt: Option<StrColumnWriter>| {
+                let mut column: StrColumnWriter = column_opt.unwrap_or_else(|| {
+                    // Each column has its own dictionary
+                    let dictionary_id = dictionaries.len() as u32;
+                    dictionaries.push(DictionaryBuilder::default());
+                    StrColumnWriter::with_dictionary_id(dictionary_id)
+                });
+                column.record_bytes(doc, value.as_bytes(), dictionaries, arena);
+                column
+            },
+        );
+    }
+
+    pub fn serialize(&mut self, num_docs: DocId, wrt: &mut dyn io::Write) -> io::Result<()> {
+        let mut serializer = ColumnarSerializer::new(wrt);
+        let mut field_columns: Vec<(&[u8], ColumnTypeCategory, Addr)> = self
+            .numerical_field_hash_map
+            .iter()
+            .map(|(term, addr, _)| (term, ColumnTypeCategory::Numerical, addr))
+            .collect();
+        field_columns.extend(
+            self.bytes_field_hash_map
+                .iter()
+                .map(|(term, addr, _)| (term, ColumnTypeCategory::Str, addr)),
+        );
+        field_columns.extend(
+            self.bool_field_hash_map
+                .iter()
+                .map(|(term, addr, _)| (term, ColumnTypeCategory::Bool, addr)),
+        );
+        field_columns.sort_unstable_by_key(|(column_name, col_type, _)| (*column_name, *col_type));
+        let (arena, buffers, dictionaries) = (&self.arena, &mut self.buffers, &self.dictionaries);
+        let mut symbol_byte_buffer: Vec<u8> = Vec::new();
+        for (column_name, bytes_or_numerical, addr) in field_columns {
+            match bytes_or_numerical {
+                ColumnTypeCategory::Bool => {
+                    let column_writer: ColumnWriter = self.bool_field_hash_map.read(addr);
+                    let cardinality = column_writer.get_cardinality(num_docs);
+                    let column_type_and_cardinality = ColumnTypeAndCardinality {
+                        cardinality,
+                        typ: ColumnType::Bool,
+                    };
+                    let mut column_serializer =
+                        serializer.serialize_column(column_name, column_type_and_cardinality);
+                    serialize_bool_column(
+                        cardinality,
+                        num_docs,
+                        column_writer.operation_iterator(arena, &mut symbol_byte_buffer),
+                        buffers,
+                        &mut column_serializer,
+                    )?;
+                }
+                ColumnTypeCategory::Str => {
+                    let str_column_writer: StrColumnWriter = self.bytes_field_hash_map.read(addr);
+                    let dictionary_builder =
+                        &dictionaries[str_column_writer.dictionary_id as usize];
+                    let cardinality = str_column_writer.column_writer.get_cardinality(num_docs);
+                    let column_type_and_cardinality = ColumnTypeAndCardinality {
+                        cardinality,
+                        typ: ColumnType::Bytes,
+                    };
+                    let mut column_serializer =
+                        serializer.serialize_column(column_name, column_type_and_cardinality);
+                    serialize_bytes_column(
+                        cardinality,
+                        num_docs,
+                        dictionary_builder,
+                        str_column_writer.operation_iterator(arena, &mut symbol_byte_buffer),
+                        buffers,
+                        &mut column_serializer,
+                    )?;
+                }
+                ColumnTypeCategory::Numerical => {
+                    let numerical_column_writer: NumericalColumnWriter =
+                        self.numerical_field_hash_map.read(addr);
+                    let (numerical_type, cardinality) =
+                        numerical_column_writer.column_type_and_cardinality(num_docs);
+                    let column_type_and_cardinality = ColumnTypeAndCardinality {
+                        cardinality,
+                        typ: ColumnType::Numerical(numerical_type),
+                    };
+                    let mut column_serializer =
+                        serializer.serialize_column(column_name, column_type_and_cardinality);
+                    serialize_numerical_column(
+                        cardinality,
+                        num_docs,
+                        numerical_type,
+                        numerical_column_writer.operation_iterator(arena, &mut symbol_byte_buffer),
+                        buffers,
+                        &mut column_serializer,
+                    )?;
+                }
+            };
+        }
+        serializer.finalize()?;
+        Ok(())
+    }
+}
+
+fn serialize_bytes_column(
+    cardinality: Cardinality,
+    num_docs: DocId,
+    dictionary_builder: &DictionaryBuilder,
+    operation_it: impl Iterator<Item = ColumnOperation<UnorderedId>>,
+    buffers: &mut SpareBuffers,
+    wrt: impl io::Write,
+) -> io::Result<()> {
+    let SpareBuffers {
+        value_index_builders,
+        u64_values,
+        ..
+    } = buffers;
+    let mut counting_writer = CountingWriter::wrap(wrt);
+    let term_id_mapping: TermIdMapping = dictionary_builder.serialize(&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>| {
+        // We map unordered ids to ordered ids.
+        match symbol {
+            ColumnOperation::Value(unordered_id) => {
+                let ordered_id = term_id_mapping.to_ord(unordered_id);
+                ColumnOperation::Value(ordered_id.0 as u64)
+            }
+            ColumnOperation::NewDoc(doc) => ColumnOperation::NewDoc(doc),
+        }
+    });
+    serialize_column(
+        operation_iterator,
+        cardinality,
+        num_docs,
+        value_index_builders,
+        u64_values,
+        &mut wrt,
+    )?;
+    wrt.write_all(&dictionary_num_bytes.to_le_bytes()[..])?;
+    Ok(())
+}
+
+fn serialize_numerical_column(
+    cardinality: Cardinality,
+    num_docs: DocId,
+    numerical_type: NumericalType,
+    op_iterator: impl Iterator<Item = ColumnOperation<NumericalValue>>,
+    buffers: &mut SpareBuffers,
+    wrt: &mut impl io::Write,
+) -> io::Result<()> {
+    let SpareBuffers {
+        value_index_builders,
+        u64_values,
+        i64_values,
+        f64_values,
+        ..
+    } = buffers;
+    match numerical_type {
+        NumericalType::I64 => {
+            serialize_column(
+                coerce_numerical_symbol::<i64>(op_iterator),
+                cardinality,
+                num_docs,
+                value_index_builders,
+                i64_values,
+                wrt,
+            )?;
+        }
+        NumericalType::U64 => {
+            serialize_column(
+                coerce_numerical_symbol::<u64>(op_iterator),
+                cardinality,
+                num_docs,
+                value_index_builders,
+                u64_values,
+                wrt,
+            )?;
+        }
+        NumericalType::F64 => {
+            serialize_column(
+                coerce_numerical_symbol::<f64>(op_iterator),
+                cardinality,
+                num_docs,
+                value_index_builders,
+                f64_values,
+                wrt,
+            )?;
+        }
+    };
+    Ok(())
+}
+
+fn serialize_bool_column(
+    cardinality: Cardinality,
+    num_docs: DocId,
+    column_operations_it: impl Iterator<Item = ColumnOperation<bool>>,
+    buffers: &mut SpareBuffers,
+    wrt: &mut impl io::Write,
+) -> io::Result<()> {
+    let SpareBuffers {
+        value_index_builders,
+        bool_values,
+        ..
+    } = buffers;
+    serialize_column(
+        column_operations_it,
+        cardinality,
+        num_docs,
+        value_index_builders,
+        bool_values,
+        wrt,
+    )?;
+    Ok(())
+}
+
+fn serialize_column<
+    T: Copy + Default + std::fmt::Debug + Send + Sync + MonotonicallyMappableToU64 + PartialOrd,
+>(
+    op_iterator: impl Iterator<Item = ColumnOperation<T>>,
+    cardinality: Cardinality,
+    num_docs: DocId,
+    value_index_builders: &mut SpareIndexBuilders,
+    values: &mut Vec<T>,
+    mut wrt: impl io::Write,
+) -> io::Result<()>
+where
+    for<'a> VecColumn<'a, T>: Column<T>,
+{
+    values.clear();
+    match cardinality {
+        Cardinality::Required => {
+            consume_operation_iterator(
+                op_iterator,
+                value_index_builders.borrow_required_index_builder(),
+                values,
+            );
+            fastfield_codecs::serialize(
+                VecColumn::from(&values[..]),
+                &mut wrt,
+                &fastfield_codecs::ALL_CODEC_TYPES[..],
+            )?;
+        }
+        Cardinality::Optional => {
+            let optional_index_builder = value_index_builders.borrow_optional_index_builder();
+            consume_operation_iterator(op_iterator, optional_index_builder, values);
+            let optional_index = optional_index_builder.finish(num_docs);
+            fastfield_codecs::serialize::serialize_new(
+                ValueIndexInfo::SingleValue(Box::new(optional_index)),
+                VecColumn::from(&values[..]),
+                &mut wrt,
+                &fastfield_codecs::ALL_CODEC_TYPES[..],
+            )?;
+        }
+        Cardinality::Multivalued => {
+            let multivalued_index_builder = value_index_builders.borrow_multivalued_index_builder();
+            consume_operation_iterator(op_iterator, multivalued_index_builder, values);
+            let multivalued_index = multivalued_index_builder.finish(num_docs);
+            fastfield_codecs::serialize::serialize_new(
+                ValueIndexInfo::MultiValue(Box::new(multivalued_index)),
+                VecColumn::from(&values[..]),
+                &mut wrt,
+                &fastfield_codecs::ALL_CODEC_TYPES[..],
+            )?;
+        }
+    }
+    Ok(())
+}
+
+fn coerce_numerical_symbol<T>(
+    operation_iterator: impl Iterator<Item = ColumnOperation<NumericalValue>>,
+) -> impl Iterator<Item = ColumnOperation<T>>
+where T: Coerce {
+    operation_iterator.map(|symbol| match symbol {
+        ColumnOperation::NewDoc(doc) => ColumnOperation::NewDoc(doc),
+        ColumnOperation::Value(numerical_value) => {
+            ColumnOperation::Value(Coerce::coerce(numerical_value))
+        }
+    })
+}
+
+fn consume_operation_iterator<T: std::fmt::Debug, TIndexBuilder: IndexBuilder>(
+    operation_iterator: impl Iterator<Item = ColumnOperation<T>>,
+    index_builder: &mut TIndexBuilder,
+    values: &mut Vec<T>,
+) {
+    for symbol in operation_iterator {
+        match symbol {
+            ColumnOperation::NewDoc(doc) => {
+                index_builder.record_doc(doc);
+            }
+            ColumnOperation::Value(value) => {
+                index_builder.record_value();
+                values.push(value);
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use column_operation::ColumnOperation;
+    use stacker::MemoryArena;
+
+    use super::*;
+    use crate::value::NumericalValue;
+    use crate::Cardinality;
+
+    #[test]
+    fn test_column_writer_required_simple() {
+        let mut arena = MemoryArena::default();
+        let mut column_writer = super::ColumnWriter::default();
+        column_writer.record(0u32, NumericalValue::from(14i64), &mut arena);
+        column_writer.record(1u32, NumericalValue::from(15i64), &mut arena);
+        column_writer.record(2u32, NumericalValue::from(-16i64), &mut arena);
+        assert_eq!(column_writer.get_cardinality(3), Cardinality::Required);
+        let mut buffer = Vec::new();
+        let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
+            .operation_iterator(&mut arena, &mut buffer)
+            .collect();
+        assert_eq!(symbols.len(), 6);
+        assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));
+        assert!(matches!(
+            symbols[1],
+            ColumnOperation::Value(NumericalValue::I64(14i64))
+        ));
+        assert!(matches!(symbols[2], ColumnOperation::NewDoc(1u32)));
+        assert!(matches!(
+            symbols[3],
+            ColumnOperation::Value(NumericalValue::I64(15i64))
+        ));
+        assert!(matches!(symbols[4], ColumnOperation::NewDoc(2u32)));
+        assert!(matches!(
+            symbols[5],
+            ColumnOperation::Value(NumericalValue::I64(-16i64))
+        ));
+    }
+
+    #[test]
+    fn test_column_writer_optional_cardinality_missing_first() {
+        let mut arena = MemoryArena::default();
+        let mut column_writer = super::ColumnWriter::default();
+        column_writer.record(1u32, NumericalValue::from(15i64), &mut arena);
+        column_writer.record(2u32, NumericalValue::from(-16i64), &mut arena);
+        assert_eq!(column_writer.get_cardinality(3), Cardinality::Optional);
+        let mut buffer = Vec::new();
+        let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
+            .operation_iterator(&mut arena, &mut buffer)
+            .collect();
+        assert_eq!(symbols.len(), 4);
+        assert!(matches!(symbols[0], ColumnOperation::NewDoc(1u32)));
+        assert!(matches!(
+            symbols[1],
+            ColumnOperation::Value(NumericalValue::I64(15i64))
+        ));
+        assert!(matches!(symbols[2], ColumnOperation::NewDoc(2u32)));
+        assert!(matches!(
+            symbols[3],
+            ColumnOperation::Value(NumericalValue::I64(-16i64))
+        ));
+    }
+
+    #[test]
+    fn test_column_writer_optional_cardinality_missing_last() {
+        let mut arena = MemoryArena::default();
+        let mut column_writer = super::ColumnWriter::default();
+        column_writer.record(0u32, NumericalValue::from(15i64), &mut arena);
+        assert_eq!(column_writer.get_cardinality(2), Cardinality::Optional);
+        let mut buffer = Vec::new();
+        let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
+            .operation_iterator(&mut arena, &mut buffer)
+            .collect();
+        assert_eq!(symbols.len(), 2);
+        assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));
+        assert!(matches!(
+            symbols[1],
+            ColumnOperation::Value(NumericalValue::I64(15i64))
+        ));
+    }
+
+    #[test]
+    fn test_column_writer_multivalued() {
+        let mut arena = MemoryArena::default();
+        let mut column_writer = super::ColumnWriter::default();
+        column_writer.record(0u32, NumericalValue::from(16i64), &mut arena);
+        column_writer.record(0u32, NumericalValue::from(17i64), &mut arena);
+        assert_eq!(column_writer.get_cardinality(1), Cardinality::Multivalued);
+        let mut buffer = Vec::new();
+        let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
+            .operation_iterator(&mut arena, &mut buffer)
+            .collect();
+        assert_eq!(symbols.len(), 3);
+        assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));
+        assert!(matches!(
+            symbols[1],
+            ColumnOperation::Value(NumericalValue::I64(16i64))
+        ));
+        assert!(matches!(
+            symbols[2],
+            ColumnOperation::Value(NumericalValue::I64(17i64))
+        ));
+    }
+}

columnar/src/writer/serializer.rs

@@ -0,0 +1,116 @@
+use std::io;
+use std::io::Write;
+
+use common::CountingWriter;
+use sstable::value::RangeValueWriter;
+use sstable::RangeSSTable;
+
+use crate::column_type_header::ColumnTypeAndCardinality;
+
+pub struct ColumnarSerializer<W: io::Write> {
+    wrt: CountingWriter<W>,
+    sstable_range: sstable::Writer<Vec<u8>, RangeValueWriter>,
+    prepare_key_buffer: Vec<u8>,
+}
+
+/// Returns a key consisting of the concatenation of the key and the column_type_and_cardinality
+/// code.
+fn prepare_key(
+    key: &[u8],
+    column_type_cardinality: ColumnTypeAndCardinality,
+    buffer: &mut Vec<u8>,
+) {
+    buffer.clear();
+    buffer.extend_from_slice(key);
+    buffer.push(0u8);
+    buffer.push(column_type_cardinality.to_code());
+}
+
+impl<W: io::Write> ColumnarSerializer<W> {
+    pub(crate) fn new(wrt: W) -> ColumnarSerializer<W> {
+        let sstable_range: sstable::Writer<Vec<u8>, RangeValueWriter> =
+            sstable::Dictionary::<RangeSSTable>::builder(Vec::with_capacity(100_000)).unwrap();
+        ColumnarSerializer {
+            wrt: CountingWriter::wrap(wrt),
+            sstable_range,
+            prepare_key_buffer: Vec::new(),
+        }
+    }
+
+    pub fn serialize_column<'a>(
+        &'a mut self,
+        column_name: &[u8],
+        column_type_cardinality: ColumnTypeAndCardinality,
+    ) -> impl io::Write + 'a {
+        let start_offset = self.wrt.written_bytes();
+        prepare_key(
+            column_name,
+            column_type_cardinality,
+            &mut self.prepare_key_buffer,
+        );
+        ColumnSerializer {
+            columnar_serializer: self,
+            start_offset,
+        }
+    }
+
+    pub(crate) fn finalize(mut self) -> io::Result<()> {
+        let sstable_bytes: Vec<u8> = self.sstable_range.finish()?;
+        let sstable_num_bytes: u64 = sstable_bytes.len() as u64;
+        self.wrt.write_all(&sstable_bytes)?;
+        self.wrt.write_all(&sstable_num_bytes.to_le_bytes()[..])?;
+        Ok(())
+    }
+}
+
+struct ColumnSerializer<'a, W: io::Write> {
+    columnar_serializer: &'a mut ColumnarSerializer<W>,
+    start_offset: u64,
+}
+
+impl<'a, W: io::Write> Drop for ColumnSerializer<'a, W> {
+    fn drop(&mut self) {
+        let end_offset: u64 = self.columnar_serializer.wrt.written_bytes();
+        let byte_range = self.start_offset..end_offset;
+        self.columnar_serializer.sstable_range.insert_cannot_fail(
+            &self.columnar_serializer.prepare_key_buffer[..],
+            &byte_range,
+        );
+        self.columnar_serializer.prepare_key_buffer.clear();
+    }
+}
+
+impl<'a, W: io::Write> io::Write for ColumnSerializer<'a, W> {
+    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
+        self.columnar_serializer.wrt.write(buf)
+    }
+
+    fn flush(&mut self) -> io::Result<()> {
+        self.columnar_serializer.wrt.flush()
+    }
+
+    fn write_all(&mut self, buf: &[u8]) -> io::Result<()> {
+        self.columnar_serializer.wrt.write_all(buf)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::column_type_header::ColumnType;
+    use crate::Cardinality;
+
+    #[test]
+    fn test_prepare_key_bytes() {
+        let mut buffer: Vec<u8> = b"somegarbage".to_vec();
+        let column_type_and_cardinality = ColumnTypeAndCardinality {
+            typ: ColumnType::Bytes,
+            cardinality: Cardinality::Optional,
+        };
+        prepare_key(b"root\0child", column_type_and_cardinality, &mut buffer);
+        assert_eq!(buffer.len(), 12);
+        assert_eq!(&buffer[..10], b"root\0child");
+        assert_eq!(buffer[10], 0u8);
+        assert_eq!(buffer[11], column_type_and_cardinality.to_code());
+    }
+}

columnar/src/writer/value_index.rs

@@ -0,0 +1,220 @@
+use fastfield_codecs::serialize::{MultiValueIndexInfo, SingleValueIndexInfo};
+
+use crate::DocId;
+
+/// The `IndexBuilder` interprets a sequence of
+/// calls of the form:
+/// (record_doc,record_value+)*
+/// and can then serialize the results into an index to associate docids with their value[s].
+///
+/// It has different implementation depending on whether the
+/// cardinality is required, optional, or multivalued.
+pub(crate) trait IndexBuilder {
+    fn record_doc(&mut self, doc: DocId);
+    #[inline]
+    fn record_value(&mut self) {}
+}
+
+/// The RequiredIndexBuilder does nothing.
+#[derive(Default)]
+pub struct RequiredIndexBuilder;
+
+impl IndexBuilder for RequiredIndexBuilder {
+    #[inline(always)]
+    fn record_doc(&mut self, _doc: DocId) {}
+}
+
+#[derive(Default)]
+pub struct OptionalIndexBuilder {
+    docs: Vec<DocId>,
+}
+
+struct SingleValueArrayIndex<'a> {
+    // DocIds with a value. DocIds are strictly increasing
+    docs: &'a [DocId],
+    num_docs: DocId,
+}
+
+impl<'a> SingleValueIndexInfo for SingleValueArrayIndex<'a> {
+    fn num_vals(&self) -> u32 {
+        self.num_docs as u32
+    }
+
+    fn num_non_nulls(&self) -> u32 {
+        self.docs.len() as u32
+    }
+
+    fn iter(&self) -> Box<dyn Iterator<Item = u32> + '_> {
+        Box::new(self.docs.iter().copied())
+    }
+}
+
+impl OptionalIndexBuilder {
+    pub fn finish(&mut self, num_docs: DocId) -> impl SingleValueIndexInfo + '_ {
+        debug_assert!(self
+            .docs
+            .last()
+            .copied()
+            .map(|last_doc| last_doc < num_docs)
+            .unwrap_or(true));
+        SingleValueArrayIndex {
+            docs: &self.docs[..],
+            num_docs,
+        }
+    }
+
+    fn reset(&mut self) {
+        self.docs.clear();
+    }
+}
+
+impl IndexBuilder for OptionalIndexBuilder {
+    #[inline(always)]
+    fn record_doc(&mut self, doc: DocId) {
+        debug_assert!(self
+            .docs
+            .last()
+            .copied()
+            .map(|prev_doc| doc > prev_doc)
+            .unwrap_or(true));
+        self.docs.push(doc);
+    }
+}
+
+#[derive(Default)]
+pub struct MultivaluedIndexBuilder {
+    // TODO should we switch to `start_offset`?
+    // contains the num values so far for each `DocId`.
+    end_offsets: Vec<DocId>,
+    total_num_vals_seen: u32,
+}
+
+pub struct MultivaluedValueArrayIndex<'a> {
+    end_offsets: &'a [DocId],
+}
+
+impl<'a> MultiValueIndexInfo for MultivaluedValueArrayIndex<'a> {
+    fn num_docs(&self) -> u32 {
+        self.end_offsets.len() as u32
+    }
+
+    fn num_vals(&self) -> u32 {
+        self.end_offsets.last().copied().unwrap_or(0u32)
+    }
+
+    fn iter(&self) -> Box<dyn Iterator<Item = u32> + '_> {
+        if self.end_offsets.is_empty() {
+            return Box::new(std::iter::empty());
+        }
+        let n = self.end_offsets.len();
+        Box::new(std::iter::once(0u32).chain(self.end_offsets[..n - 1].iter().copied()))
+    }
+}
+
+impl MultivaluedIndexBuilder {
+    pub fn finish(&mut self, num_docs: DocId) -> impl MultiValueIndexInfo + '_ {
+        self.end_offsets
+            .resize(num_docs as usize, self.total_num_vals_seen);
+        MultivaluedValueArrayIndex {
+            end_offsets: &self.end_offsets[..],
+        }
+    }
+
+    fn reset(&mut self) {
+        self.end_offsets.clear();
+        self.total_num_vals_seen = 0;
+    }
+}
+
+impl IndexBuilder for MultivaluedIndexBuilder {
+    fn record_doc(&mut self, doc: DocId) {
+        self.end_offsets
+            .resize(doc as usize, self.total_num_vals_seen);
+    }
+
+    fn record_value(&mut self) {
+        self.total_num_vals_seen += 1;
+    }
+}
+
+/// The `SpareIndexBuilders` is there to avoid allocating a
+/// new index builder for every single column.
+#[derive(Default)]
+pub struct SpareIndexBuilders {
+    required_index_builder: RequiredIndexBuilder,
+    optional_index_builder: OptionalIndexBuilder,
+    multivalued_index_builder: MultivaluedIndexBuilder,
+}
+
+impl SpareIndexBuilders {
+    pub fn borrow_required_index_builder(&mut self) -> &mut RequiredIndexBuilder {
+        &mut self.required_index_builder
+    }
+
+    pub fn borrow_optional_index_builder(&mut self) -> &mut OptionalIndexBuilder {
+        self.optional_index_builder.reset();
+        &mut self.optional_index_builder
+    }
+
+    pub fn borrow_multivalued_index_builder(&mut self) -> &mut MultivaluedIndexBuilder {
+        self.multivalued_index_builder.reset();
+        &mut self.multivalued_index_builder
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_optional_value_index_builder() {
+        let mut opt_value_index_builder = OptionalIndexBuilder::default();
+        opt_value_index_builder.record_doc(0u32);
+        opt_value_index_builder.record_value();
+        assert_eq!(
+            &opt_value_index_builder
+                .finish(1u32)
+                .iter()
+                .collect::<Vec<u32>>(),
+            &[0]
+        );
+        opt_value_index_builder.reset();
+        opt_value_index_builder.record_doc(1u32);
+        opt_value_index_builder.record_value();
+        assert_eq!(
+            &opt_value_index_builder
+                .finish(2u32)
+                .iter()
+                .collect::<Vec<u32>>(),
+            &[1]
+        );
+    }
+
+    #[test]
+    fn test_multivalued_value_index_builder() {
+        let mut multivalued_value_index_builder = MultivaluedIndexBuilder::default();
+        multivalued_value_index_builder.record_doc(1u32);
+        multivalued_value_index_builder.record_value();
+        multivalued_value_index_builder.record_value();
+        multivalued_value_index_builder.record_doc(2u32);
+        multivalued_value_index_builder.record_value();
+        assert_eq!(
+            multivalued_value_index_builder
+                .finish(4u32)
+                .iter()
+                .collect::<Vec<u32>>(),
+            vec![0, 0, 2, 3]
+        );
+        multivalued_value_index_builder.reset();
+        multivalued_value_index_builder.record_doc(2u32);
+        multivalued_value_index_builder.record_value();
+        multivalued_value_index_builder.record_value();
+        assert_eq!(
+            multivalued_value_index_builder
+                .finish(4u32)
+                .iter()
+                .collect::<Vec<u32>>(),
+            vec![0, 0, 0, 2]
+        );
+    }
+}

common/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "tantivy-common"
-version = "0.4.0"
+version = "0.5.0"
 authors = ["Paul Masurel <paul@quickwit.io>", "Pascal Seitz <pascal@quickwit.io>"]
 license = "MIT"
 edition = "2021"
@@ -14,7 +14,8 @@ repository = "https://github.com/quickwit-oss/tantivy"
 
 [dependencies]
 byteorder = "1.4.3"
-ownedbytes = { version= "0.4", path="../ownedbytes" }
+ownedbytes = { version= "0.5", path="../ownedbytes" }
+async-trait = "0.1"
 
 [dev-dependencies]
 proptest = "1.0.0"

common/src/bitset.rs

@@ -151,7 +151,7 @@ impl TinySet {
         if self.is_empty() {
             None
         } else {
-            let lowest = self.0.trailing_zeros() as u32;
+            let lowest = self.0.trailing_zeros();
             self.0 ^= TinySet::singleton(lowest).0;
             Some(lowest)
         }
@@ -421,7 +421,7 @@ mod tests {
             bitset.serialize(&mut out).unwrap();
 
             let bitset = ReadOnlyBitSet::open(OwnedBytes::new(out));
-            assert_eq!(bitset.len() as usize, i as usize);
+            assert_eq!(bitset.len(), i as usize);
         }
     }
 
@@ -432,7 +432,7 @@ mod tests {
         bitset.serialize(&mut out).unwrap();
 
         let bitset = ReadOnlyBitSet::open(OwnedBytes::new(out));
-        assert_eq!(bitset.len() as usize, 64);
+        assert_eq!(bitset.len(), 64);
     }
 
     #[test]

common/src/file_slice.rs

@@ -1,19 +1,18 @@
-use std::ops::{Deref, Range};
+use std::ops::{Deref, Range, RangeBounds};
 use std::sync::Arc;
 use std::{fmt, io};
 
 use async_trait::async_trait;
-use common::HasLen;
-use stable_deref_trait::StableDeref;
+use ownedbytes::{OwnedBytes, StableDeref};
 
-use crate::directory::OwnedBytes;
+use crate::HasLen;
 
 /// Objects that represents files sections in tantivy.
 ///
 /// By contract, whatever happens to the directory file, as long as a FileHandle
 /// is alive, the data associated with it cannot be altered or destroyed.
 ///
-/// The underlying behavior is therefore specific to the [`Directory`](crate::Directory) that
+/// The underlying behavior is therefore specific to the `Directory` that
 /// created it. Despite its name, a [`FileSlice`] may or may not directly map to an actual file
 /// on the filesystem.
 
@@ -24,13 +23,12 @@ pub trait FileHandle: 'static + Send + Sync + HasLen + fmt::Debug {
     /// This method may panic if the range requested is invalid.
     fn read_bytes(&self, range: Range<usize>) -> io::Result<OwnedBytes>;
 
-    #[cfg(feature = "quickwit")]
     #[doc(hidden)]
-    async fn read_bytes_async(
-        &self,
-        _byte_range: Range<usize>,
-    ) -> crate::AsyncIoResult<OwnedBytes> {
-        Err(crate::error::AsyncIoError::AsyncUnsupported)
+    async fn read_bytes_async(&self, _byte_range: Range<usize>) -> io::Result<OwnedBytes> {
+        Err(io::Error::new(
+            io::ErrorKind::Unsupported,
+            "Async read is not supported.",
+        ))
     }
 }
 
@@ -41,8 +39,7 @@ impl FileHandle for &'static [u8] {
         Ok(OwnedBytes::new(bytes))
     }
 
-    #[cfg(feature = "quickwit")]
-    async fn read_bytes_async(&self, byte_range: Range<usize>) -> crate::AsyncIoResult<OwnedBytes> {
+    async fn read_bytes_async(&self, byte_range: Range<usize>) -> io::Result<OwnedBytes> {
         Ok(self.read_bytes(byte_range)?)
     }
 }
@@ -70,6 +67,34 @@ impl fmt::Debug for FileSlice {
     }
 }
 
+/// Takes a range, a `RangeBounds` object, and returns
+/// a `Range` that corresponds to the relative application of the
+/// `RangeBounds` object to the original `Range`.
+///
+/// For instance, combine_ranges(`[2..11)`, `[5..7]`) returns `[7..10]`
+/// as it reads, what is the sub-range that starts at the 5 element of
+/// `[2..11)` and ends at the 9th element included.
+///
+/// This function panics, if the result would suggest something outside
+/// of the bounds of the original range.
+fn combine_ranges<R: RangeBounds<usize>>(orig_range: Range<usize>, rel_range: R) -> Range<usize> {
+    let start: usize = orig_range.start
+        + match rel_range.start_bound().cloned() {
+            std::ops::Bound::Included(rel_start) => rel_start,
+            std::ops::Bound::Excluded(rel_start) => rel_start + 1,
+            std::ops::Bound::Unbounded => 0,
+        };
+    assert!(start <= orig_range.end);
+    let end: usize = match rel_range.end_bound().cloned() {
+        std::ops::Bound::Included(rel_end) => orig_range.start + rel_end + 1,
+        std::ops::Bound::Excluded(rel_end) => orig_range.start + rel_end,
+        std::ops::Bound::Unbounded => orig_range.end,
+    };
+    assert!(end >= start);
+    assert!(end <= orig_range.end);
+    start..end
+}
+
 impl FileSlice {
     /// Wraps a FileHandle.
     pub fn new(file_handle: Arc<dyn FileHandle>) -> Self {
@@ -93,11 +118,11 @@ impl FileSlice {
     ///
     /// Panics if `byte_range.end` exceeds the filesize.
     #[must_use]
-    pub fn slice(&self, byte_range: Range<usize>) -> FileSlice {
-        assert!(byte_range.end <= self.len());
+    #[inline]
+    pub fn slice<R: RangeBounds<usize>>(&self, byte_range: R) -> FileSlice {
         FileSlice {
             data: self.data.clone(),
-            range: self.range.start + byte_range.start..self.range.start + byte_range.end,
+            range: combine_ranges(self.range.clone(), byte_range),
         }
     }
 
@@ -117,9 +142,8 @@ impl FileSlice {
         self.data.read_bytes(self.range.clone())
     }
 
-    #[cfg(feature = "quickwit")]
     #[doc(hidden)]
-    pub async fn read_bytes_async(&self) -> crate::AsyncIoResult<OwnedBytes> {
+    pub async fn read_bytes_async(&self) -> io::Result<OwnedBytes> {
         self.data.read_bytes_async(self.range.clone()).await
     }
 
@@ -137,12 +161,8 @@ impl FileSlice {
             .read_bytes(self.range.start + range.start..self.range.start + range.end)
     }
 
-    #[cfg(feature = "quickwit")]
     #[doc(hidden)]
-    pub async fn read_bytes_slice_async(
-        &self,
-        byte_range: Range<usize>,
-    ) -> crate::AsyncIoResult<OwnedBytes> {
+    pub async fn read_bytes_slice_async(&self, byte_range: Range<usize>) -> io::Result<OwnedBytes> {
         assert!(
             self.range.start + byte_range.end <= self.range.end,
             "`to` exceeds the fileslice length"
@@ -204,8 +224,7 @@ impl FileHandle for FileSlice {
         self.read_bytes_slice(range)
     }
 
-    #[cfg(feature = "quickwit")]
-    async fn read_bytes_async(&self, byte_range: Range<usize>) -> crate::AsyncIoResult<OwnedBytes> {
+    async fn read_bytes_async(&self, byte_range: Range<usize>) -> io::Result<OwnedBytes> {
         self.read_bytes_slice_async(byte_range).await
     }
 }
@@ -222,21 +241,20 @@ impl FileHandle for OwnedBytes {
         Ok(self.slice(range))
     }
 
-    #[cfg(feature = "quickwit")]
-    async fn read_bytes_async(&self, range: Range<usize>) -> crate::AsyncIoResult<OwnedBytes> {
-        let bytes = self.read_bytes(range)?;
-        Ok(bytes)
+    async fn read_bytes_async(&self, range: Range<usize>) -> io::Result<OwnedBytes> {
+        self.read_bytes(range)
     }
 }
 
 #[cfg(test)]
 mod tests {
     use std::io;
+    use std::ops::Bound;
     use std::sync::Arc;
 
-    use common::HasLen;
-
     use super::{FileHandle, FileSlice};
+    use crate::file_slice::combine_ranges;
+    use crate::HasLen;
 
     #[test]
     fn test_file_slice() -> io::Result<()> {
@@ -307,4 +325,23 @@ mod tests {
             b"bcd"
         );
     }
+
+    #[test]
+    fn test_combine_range() {
+        assert_eq!(combine_ranges(1..3, 0..1), 1..2);
+        assert_eq!(combine_ranges(1..3, 1..), 2..3);
+        assert_eq!(combine_ranges(1..4, ..2), 1..3);
+        assert_eq!(combine_ranges(3..10, 2..5), 5..8);
+        assert_eq!(combine_ranges(2..11, 5..=7), 7..10);
+        assert_eq!(
+            combine_ranges(2..11, (Bound::Excluded(5), Bound::Unbounded)),
+            8..11
+        );
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_combine_range_panics() {
+        let _ = combine_ranges(3..5, 1..4);
+    }
 }

common/src/group_by.rs

@@ -0,0 +1,166 @@
+use std::cell::RefCell;
+use std::iter::Peekable;
+use std::rc::Rc;
+
+pub trait GroupByIteratorExtended: Iterator {
+    /// Return an `Iterator` that groups iterator elements. Consecutive elements that map to the
+    /// same key are assigned to the same group.
+    ///
+    /// The returned Iterator item is `(K, impl Iterator)`, where Iterator are the items of the
+    /// group.
+    ///
+    /// ```
+    /// use tantivy_common::GroupByIteratorExtended;
+    ///
+    /// // group data into blocks of larger than zero or not.
+    /// let data: Vec<i32> = vec![1, 3, -2, -2, 1, 0, 1, 2];
+    /// // groups:               |---->|------>|--------->|
+    ///
+    /// let mut data_grouped = Vec::new();
+    /// // Note: group is an iterator
+    /// for (key, group) in data.into_iter().group_by(|val| *val >= 0) {
+    ///     data_grouped.push((key, group.collect()));
+    /// }
+    /// assert_eq!(data_grouped, vec![(true, vec![1, 3]), (false, vec![-2, -2]), (true, vec![1, 0, 1, 2])]);
+    /// ```
+    fn group_by<K, F>(self, key: F) -> GroupByIterator<Self, F, K>
+    where
+        Self: Sized,
+        F: FnMut(&Self::Item) -> K,
+        K: PartialEq + Copy,
+        Self::Item: Copy,
+    {
+        GroupByIterator::new(self, key)
+    }
+}
+impl<I: Iterator> GroupByIteratorExtended for I {}
+
+pub struct GroupByIterator<I, F, K: Copy>
+where
+    I: Iterator,
+    F: FnMut(&I::Item) -> K,
+{
+    // I really would like to avoid the Rc<RefCell>, but the Iterator is shared between
+    // `GroupByIterator` and `GroupIter`. In practice they are used consecutive and
+    // `GroupByIter` is finished before calling next on `GroupByIterator`. I'm not sure there
+    // is a solution with lifetimes for that, because we would need to enforce it in the usage
+    // somehow.
+    //
+    // One potential solution would be to replace the iterator approach with something similar.
+    inner: Rc<RefCell<GroupByShared<I, F, K>>>,
+}
+
+struct GroupByShared<I, F, K: Copy>
+where
+    I: Iterator,
+    F: FnMut(&I::Item) -> K,
+{
+    iter: Peekable<I>,
+    group_by_fn: F,
+}
+
+impl<I, F, K> GroupByIterator<I, F, K>
+where
+    I: Iterator,
+    F: FnMut(&I::Item) -> K,
+    K: Copy,
+{
+    fn new(inner: I, group_by_fn: F) -> Self {
+        let inner = GroupByShared {
+            iter: inner.peekable(),
+            group_by_fn,
+        };
+
+        Self {
+            inner: Rc::new(RefCell::new(inner)),
+        }
+    }
+}
+
+impl<I, F, K> Iterator for GroupByIterator<I, F, K>
+where
+    I: Iterator,
+    I::Item: Copy,
+    F: FnMut(&I::Item) -> K,
+    K: Copy,
+{
+    type Item = (K, GroupIterator<I, F, K>);
+
+    fn next(&mut self) -> Option<Self::Item> {
+        let mut inner = self.inner.borrow_mut();
+        let value = *inner.iter.peek()?;
+        let key = (inner.group_by_fn)(&value);
+
+        let inner = self.inner.clone();
+
+        let group_iter = GroupIterator {
+            inner,
+            group_key: key,
+        };
+        Some((key, group_iter))
+    }
+}
+
+pub struct GroupIterator<I, F, K: Copy>
+where
+    I: Iterator,
+    F: FnMut(&I::Item) -> K,
+{
+    inner: Rc<RefCell<GroupByShared<I, F, K>>>,
+    group_key: K,
+}
+
+impl<I, F, K: PartialEq + Copy> Iterator for GroupIterator<I, F, K>
+where
+    I: Iterator,
+    I::Item: Copy,
+    F: FnMut(&I::Item) -> K,
+{
+    type Item = I::Item;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        let mut inner = self.inner.borrow_mut();
+        // peek if next value is in group
+        let peek_val = *inner.iter.peek()?;
+        if (inner.group_by_fn)(&peek_val) == self.group_key {
+            inner.iter.next()
+        } else {
+            None
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn group_by_collect<I: Iterator<Item = u32>>(iter: I) -> Vec<(I::Item, Vec<I::Item>)> {
+        iter.group_by(|val| val / 10)
+            .map(|(el, iter)| (el, iter.collect::<Vec<_>>()))
+            .collect::<Vec<_>>()
+    }
+
+    #[test]
+    fn group_by_two_groups() {
+        let vals = vec![1u32, 4, 15];
+        let grouped_vals = group_by_collect(vals.into_iter());
+        assert_eq!(grouped_vals, vec![(0, vec![1, 4]), (1, vec![15])]);
+    }
+
+    #[test]
+    fn group_by_test_empty() {
+        let vals = vec![];
+        let grouped_vals = group_by_collect(vals.into_iter());
+        assert_eq!(grouped_vals, vec![]);
+    }
+
+    #[test]
+    fn group_by_three_groups() {
+        let vals = vec![1u32, 4, 15, 1];
+        let grouped_vals = group_by_collect(vals.into_iter());
+        assert_eq!(
+            grouped_vals,
+            vec![(0, vec![1, 4]), (1, vec![15]), (0, vec![1])]
+        );
+    }
+}

common/src/lib.rs

@@ -5,11 +5,14 @@ use std::ops::Deref;
 pub use byteorder::LittleEndian as Endianness;
 
 mod bitset;
+pub mod file_slice;
+mod group_by;
 mod serialize;
 mod vint;
 mod writer;
-
 pub use bitset::*;
+pub use group_by::GroupByIteratorExtended;
+pub use ownedbytes::{OwnedBytes, StableDeref};
 pub use serialize::{BinarySerializable, DeserializeFrom, FixedSize};
 pub use vint::{
     deserialize_vint_u128, read_u32_vint, read_u32_vint_no_advance, serialize_vint_u128,

examples/faceted_search.rs

@@ -1,15 +1,17 @@
-// # Basic Example
+// # Faceted Search
 //
-// This example covers the basic functionalities of
+// This example covers the faceted search functionalities of
 // tantivy.
 //
 // We will :
-// - define our schema
-// = create an index in a directory
-// - index few documents in our index
-// - search for the best document matchings "sea whale"
-// - retrieve the best document original content.
-
+// - define a text field "name" in our schema
+// - define a facet field "classification" in our schema
+// - create an index in memory
+// - index few documents with respective facets in our index
+// - search and count the number of documents that the classifications start the facet "/Felidae"
+// - Search the facet "/Felidae/Pantherinae" and count the number of documents that the
+//   classifications include the facet.
+//
 // ---
 // Importing tantivy...
 use tantivy::collector::FacetCollector;
@@ -21,7 +23,7 @@ fn main() -> tantivy::Result<()> {
     // Let's create a temporary directory for the sake of this example
     let mut schema_builder = Schema::builder();
 
-    let name = schema_builder.add_text_field("felin_name", TEXT | STORED);
+    let name = schema_builder.add_text_field("name", TEXT | STORED);
     // this is our faceted field: its scientific classification
     let classification = schema_builder.add_facet_field("classification", FacetOptions::default());
 

fastfield_codecs/Cargo.toml

@@ -12,10 +12,9 @@ repository = "https://github.com/quickwit-oss/tantivy"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-common = { version = "0.4", path = "../common/", package = "tantivy-common" }
+common = { version = "0.5", path = "../common/", package = "tantivy-common" }
 tantivy-bitpacker = { version= "0.3", path = "../bitpacker/" }
-ownedbytes = { version = "0.4.0", path = "../ownedbytes" }
-prettytable-rs = {version="0.9.0", optional= true}
+prettytable-rs = {version="0.10.0", optional= true}
 rand = {version="0.8.3", optional= true}
 fastdivide = "0.4"
 log = "0.4"

fastfield_codecs/benches/bench.rs

@@ -4,11 +4,11 @@ extern crate test;
 
 #[cfg(test)]
 mod tests {
-    use std::iter;
+    use std::ops::RangeInclusive;
     use std::sync::Arc;
 
+    use common::OwnedBytes;
     use fastfield_codecs::*;
-    use ownedbytes::OwnedBytes;
     use rand::prelude::*;
     use test::Bencher;
 
@@ -71,27 +71,24 @@ mod tests {
         });
     }
 
-    fn get_exp_data() -> Vec<u64> {
+    const FIFTY_PERCENT_RANGE: RangeInclusive<u64> = 1..=50;
+    const SINGLE_ITEM: u64 = 90;
+    const SINGLE_ITEM_RANGE: RangeInclusive<u64> = 90..=90;
+    const ONE_PERCENT_ITEM_RANGE: RangeInclusive<u64> = 49..=49;
+    fn get_data_50percent_item() -> Vec<u128> {
+        let mut rng = StdRng::from_seed([1u8; 32]);
+
         let mut data = vec![];
-        for i in 0..100 {
-            let num = i * i;
-            data.extend(iter::repeat(i as u64).take(num));
+        for _ in 0..300_000 {
+            let val = rng.gen_range(1..=100);
+            data.push(val);
         }
-        data.shuffle(&mut StdRng::from_seed([1u8; 32]));
+        data.push(SINGLE_ITEM);
 
-        // lengt = 328350
+        data.shuffle(&mut rng);
+        let data = data.iter().map(|el| *el as u128).collect::<Vec<_>>();
         data
     }
-
-    fn get_data_50percent_item() -> (u128, u128, Vec<u128>) {
-        let mut permutation = get_exp_data();
-        let major_item = 20;
-        let minor_item = 10;
-        permutation.extend(iter::repeat(major_item).take(permutation.len()));
-        permutation.shuffle(&mut StdRng::from_seed([1u8; 32]));
-        let permutation = permutation.iter().map(|el| *el as u128).collect::<Vec<_>>();
-        (major_item as u128, minor_item as u128, permutation)
-    }
     fn get_u128_column_random() -> Arc<dyn Column<u128>> {
         let permutation = generate_random();
         let permutation = permutation.iter().map(|el| *el as u128).collect::<Vec<_>>();
@@ -106,15 +103,82 @@ mod tests {
         open_u128::<u128>(out).unwrap()
     }
 
+    // U64 RANGE START
+    #[bench]
+    fn bench_intfastfield_getrange_u64_50percent_hit(b: &mut Bencher) {
+        let data = get_data_50percent_item();
+        let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
+        let column: Arc<dyn Column<u64>> = serialize_and_load(&data);
+
+        b.iter(|| {
+            let mut positions = Vec::new();
+            column.get_docids_for_value_range(
+                FIFTY_PERCENT_RANGE,
+                0..data.len() as u32,
+                &mut positions,
+            );
+            positions
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_getrange_u64_1percent_hit(b: &mut Bencher) {
+        let data = get_data_50percent_item();
+        let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
+        let column: Arc<dyn Column<u64>> = serialize_and_load(&data);
+
+        b.iter(|| {
+            let mut positions = Vec::new();
+            column.get_docids_for_value_range(
+                ONE_PERCENT_ITEM_RANGE,
+                0..data.len() as u32,
+                &mut positions,
+            );
+            positions
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_getrange_u64_single_hit(b: &mut Bencher) {
+        let data = get_data_50percent_item();
+        let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
+        let column: Arc<dyn Column<u64>> = serialize_and_load(&data);
+
+        b.iter(|| {
+            let mut positions = Vec::new();
+            column.get_docids_for_value_range(
+                SINGLE_ITEM_RANGE,
+                0..data.len() as u32,
+                &mut positions,
+            );
+            positions
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_getrange_u64_hit_all(b: &mut Bencher) {
+        let data = get_data_50percent_item();
+        let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
+        let column: Arc<dyn Column<u64>> = serialize_and_load(&data);
+
+        b.iter(|| {
+            let mut positions = Vec::new();
+            column.get_docids_for_value_range(0..=u64::MAX, 0..data.len() as u32, &mut positions);
+            positions
+        });
+    }
+    // U64 RANGE END
+
+    // U128 RANGE START
     #[bench]
     fn bench_intfastfield_getrange_u128_50percent_hit(b: &mut Bencher) {
-        let (major_item, _minor_item, data) = get_data_50percent_item();
+        let data = get_data_50percent_item();
         let column = get_u128_column_from_data(&data);
 
         b.iter(|| {
             let mut positions = Vec::new();
             column.get_docids_for_value_range(
-                major_item..=major_item,
+                *FIFTY_PERCENT_RANGE.start() as u128..=*FIFTY_PERCENT_RANGE.end() as u128,
                 0..data.len() as u32,
                 &mut positions,
             );
@@ -124,13 +188,13 @@ mod tests {
 
     #[bench]
     fn bench_intfastfield_getrange_u128_single_hit(b: &mut Bencher) {
-        let (_major_item, minor_item, data) = get_data_50percent_item();
+        let data = get_data_50percent_item();
         let column = get_u128_column_from_data(&data);
 
         b.iter(|| {
             let mut positions = Vec::new();
             column.get_docids_for_value_range(
-                minor_item..=minor_item,
+                *SINGLE_ITEM_RANGE.start() as u128..=*SINGLE_ITEM_RANGE.end() as u128,
                 0..data.len() as u32,
                 &mut positions,
             );
@@ -140,7 +204,7 @@ mod tests {
 
     #[bench]
     fn bench_intfastfield_getrange_u128_hit_all(b: &mut Bencher) {
-        let (_major_item, _minor_item, data) = get_data_50percent_item();
+        let data = get_data_50percent_item();
         let column = get_u128_column_from_data(&data);
 
         b.iter(|| {
@@ -149,6 +213,7 @@ mod tests {
             positions
         });
     }
+    // U128 RANGE END
 
     #[bench]
     fn bench_intfastfield_scan_all_fflookup_u128(b: &mut Bencher) {

fastfield_codecs/src/bitpacked.rs

@@ -1,6 +1,6 @@
 use std::io::{self, Write};
 
-use ownedbytes::OwnedBytes;
+use common::OwnedBytes;
 use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
 
 use crate::serialize::NormalizedHeader;

fastfield_codecs/src/blockwise_linear.rs

@@ -1,8 +1,7 @@
 use std::sync::Arc;
 use std::{io, iter};
 
-use common::{BinarySerializable, CountingWriter, DeserializeFrom};
-use ownedbytes::OwnedBytes;
+use common::{BinarySerializable, CountingWriter, DeserializeFrom, OwnedBytes};
 use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
 
 use crate::line::Line;
@@ -47,7 +46,7 @@ impl FastFieldCodec for BlockwiseLinearCodec {
     type Reader = BlockwiseLinearReader;
 
     fn open_from_bytes(
-        bytes: ownedbytes::OwnedBytes,
+        bytes: common::OwnedBytes,
         normalized_header: NormalizedHeader,
     ) -> io::Result<Self::Reader> {
         let footer_len: u32 = (&bytes[bytes.len() - 4..]).deserialize()?;
@@ -75,7 +74,7 @@ impl FastFieldCodec for BlockwiseLinearCodec {
         if column.num_vals() < 10 * CHUNK_SIZE as u32 {
             return None;
         }
-        let mut first_chunk: Vec<u64> = column.iter().take(CHUNK_SIZE as usize).collect();
+        let mut first_chunk: Vec<u64> = column.iter().take(CHUNK_SIZE).collect();
         let line = Line::train(&VecColumn::from(&first_chunk));
         for (i, buffer_val) in first_chunk.iter_mut().enumerate() {
             let interpolated_val = line.eval(i as u32);
@@ -171,15 +170,18 @@ impl Column for BlockwiseLinearReader {
         interpoled_val.wrapping_add(bitpacked_diff)
     }
 
+    #[inline(always)]
     fn min_value(&self) -> u64 {
         // The BlockwiseLinearReader assumes a normalized vector.
         0u64
     }
 
+    #[inline(always)]
     fn max_value(&self) -> u64 {
         self.normalized_header.max_value
     }
 
+    #[inline(always)]
     fn num_vals(&self) -> u32 {
         self.normalized_header.num_vals
     }

fastfield_codecs/src/column.rs

@@ -1,3 +1,4 @@
+use std::fmt::{self, Debug};
 use std::marker::PhantomData;
 use std::ops::{Range, RangeInclusive};
 
@@ -6,7 +7,7 @@ use tantivy_bitpacker::minmax;
 use crate::monotonic_mapping::StrictlyMonotonicFn;
 
 /// `Column` provides columnar access on a field.
-pub trait Column<T: PartialOrd = u64>: Send + Sync {
+pub trait Column<T: PartialOrd + Debug = u64>: Send + Sync {
     /// Return the value associated with the given idx.
     ///
     /// This accessor should return as fast as possible.
@@ -34,6 +35,10 @@ pub trait Column<T: PartialOrd = u64>: Send + Sync {
     /// Get the positions of values which are in the provided value range.
     ///
     /// Note that position == docid for single value fast fields
+    ///
+    /// # Truncation
+    /// `DateTime` has a truncation setting. This function should get passed the truncated values
+    /// to avoid unexpected results.
     #[inline]
     fn get_docids_for_value_range(
         &self,
@@ -83,7 +88,7 @@ pub struct VecColumn<'a, T = u64> {
     max_value: T,
 }
 
-impl<'a, C: Column<T>, T: Copy + PartialOrd> Column<T> for &'a C {
+impl<'a, C: Column<T>, T: Copy + PartialOrd + fmt::Debug> Column<T> for &'a C {
     fn get_val(&self, idx: u32) -> T {
         (*self).get_val(idx)
     }
@@ -109,7 +114,7 @@ impl<'a, C: Column<T>, T: Copy + PartialOrd> Column<T> for &'a C {
     }
 }
 
-impl<'a, T: Copy + PartialOrd + Send + Sync> Column<T> for VecColumn<'a, T> {
+impl<'a, T: Copy + PartialOrd + Send + Sync + Debug> Column<T> for VecColumn<'a, T> {
     fn get_val(&self, position: u32) -> T {
         self.values[position as usize]
     }
@@ -135,8 +140,9 @@ impl<'a, T: Copy + PartialOrd + Send + Sync> Column<T> for VecColumn<'a, T> {
     }
 }
 
-impl<'a, T: Copy + Ord + Default, V> From<&'a V> for VecColumn<'a, T>
-where V: AsRef<[T]> + ?Sized
+impl<'a, T: Copy + PartialOrd + Default, V> From<&'a V> for VecColumn<'a, T>
+where
+    V: AsRef<[T]> + ?Sized,
 {
     fn from(values: &'a V) -> Self {
         let values = values.as_ref();
@@ -177,8 +183,8 @@ pub fn monotonic_map_column<C, T, Input, Output>(
 where
     C: Column<Input>,
     T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
-    Input: PartialOrd + Send + Sync + Clone,
-    Output: PartialOrd + Send + Sync + Clone,
+    Input: PartialOrd + Send + Sync + Copy + Debug,
+    Output: PartialOrd + Send + Sync + Copy + Debug,
 {
     MonotonicMappingColumn {
         from_column,
@@ -191,8 +197,8 @@ impl<C, T, Input, Output> Column<Output> for MonotonicMappingColumn<C, T, Input>
 where
     C: Column<Input>,
     T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
-    Input: PartialOrd + Send + Sync + Clone,
-    Output: PartialOrd + Send + Sync + Clone,
+    Input: PartialOrd + Send + Sync + Copy + Debug,
+    Output: PartialOrd + Send + Sync + Copy + Debug,
 {
     #[inline]
     fn get_val(&self, idx: u32) -> Output {
@@ -228,12 +234,15 @@ where
         doc_id_range: Range<u32>,
         positions: &mut Vec<u32>,
     ) {
-        self.from_column.get_docids_for_value_range(
-            self.monotonic_mapping.inverse(range.start().clone())
-                ..=self.monotonic_mapping.inverse(range.end().clone()),
-            doc_id_range,
-            positions,
-        )
+        if range.start() > &self.max_value() || range.end() < &self.min_value() {
+            return;
+        }
+        let range = self.monotonic_mapping.inverse_coerce(range);
+        if range.start() > range.end() {
+            return;
+        }
+        self.from_column
+            .get_docids_for_value_range(range, doc_id_range, positions)
     }
 
     // We voluntarily do not implement get_range as it yields a regression,
@@ -244,7 +253,8 @@ where
 pub struct IterColumn<T>(T);
 
 impl<T> From<T> for IterColumn<T>
-where T: Iterator + Clone + ExactSizeIterator
+where
+    T: Iterator + Clone + ExactSizeIterator,
 {
     fn from(iter: T) -> Self {
         IterColumn(iter)
@@ -254,7 +264,7 @@ where T: Iterator + Clone + ExactSizeIterator
 impl<T> Column<T::Item> for IterColumn<T>
 where
     T: Iterator + Clone + ExactSizeIterator + Send + Sync,
-    T::Item: PartialOrd,
+    T::Item: PartialOrd + fmt::Debug,
 {
     fn get_val(&self, idx: u32) -> T::Item {
         self.0.clone().nth(idx as usize).unwrap()

fastfield_codecs/src/compact_space/build_compact_space.rs

@@ -208,7 +208,7 @@ impl CompactSpaceBuilder {
             };
             let covered_range_len = range_mapping.range_length();
             ranges_mapping.push(range_mapping);
-            compact_start += covered_range_len as u64;
+            compact_start += covered_range_len;
         }
         // println!("num ranges {}", ranges_mapping.len());
         CompactSpace { ranges_mapping }

fastfield_codecs/src/compact_space/mod.rs

@@ -17,8 +17,7 @@ use std::{
     ops::{Range, RangeInclusive},
 };
 
-use common::{BinarySerializable, CountingWriter, VInt, VIntU128};
-use ownedbytes::OwnedBytes;
+use common::{BinarySerializable, CountingWriter, OwnedBytes, VInt, VIntU128};
 use tantivy_bitpacker::{self, BitPacker, BitUnpacker};
 
 use crate::compact_space::build_compact_space::get_compact_space;
@@ -97,7 +96,7 @@ impl BinarySerializable for CompactSpace {
             };
             let range_length = range_mapping.range_length();
             ranges_mapping.push(range_mapping);
-            compact_start += range_length as u64;
+            compact_start += range_length;
         }
 
         Ok(Self { ranges_mapping })
@@ -407,10 +406,10 @@ impl CompactSpaceDecompressor {
             let idx2 = idx + 1;
             let idx3 = idx + 2;
             let idx4 = idx + 3;
-            let val1 = get_val(idx1 as u32);
-            let val2 = get_val(idx2 as u32);
-            let val3 = get_val(idx3 as u32);
-            let val4 = get_val(idx4 as u32);
+            let val1 = get_val(idx1);
+            let val2 = get_val(idx2);
+            let val3 = get_val(idx3);
+            let val4 = get_val(idx4);
             push_if_in_range(idx1, val1);
             push_if_in_range(idx2, val2);
             push_if_in_range(idx3, val3);
@@ -419,14 +418,13 @@ impl CompactSpaceDecompressor {
 
         // handle rest
         for idx in cutoff..position_range.end {
-            push_if_in_range(idx, get_val(idx as u32));
+            push_if_in_range(idx, get_val(idx));
         }
     }
 
     #[inline]
     fn iter_compact(&self) -> impl Iterator<Item = u64> + '_ {
-        (0..self.params.num_vals)
-            .map(move |idx| self.params.bit_unpacker.get(idx, &self.data) as u64)
+        (0..self.params.num_vals).map(move |idx| self.params.bit_unpacker.get(idx, &self.data))
     }
 
     #[inline]
@@ -455,6 +453,8 @@ impl CompactSpaceDecompressor {
 #[cfg(test)]
 mod tests {
 
+    use std::fmt;
+
     use super::*;
     use crate::format_version::read_format_version;
     use crate::null_index_footer::read_null_index_footer;
@@ -569,7 +569,7 @@ mod tests {
         let decomp = CompactSpaceDecompressor::open(data).unwrap();
         let complete_range = 0..vals.len() as u32;
         for (pos, val) in vals.iter().enumerate() {
-            let val = *val as u128;
+            let val = *val;
             let pos = pos as u32;
             let mut positions = Vec::new();
             decomp.get_positions_for_value_range(val..=val, pos..pos + 1, &mut positions);
@@ -666,7 +666,7 @@ mod tests {
             get_positions_for_value_range_helper(
                 &decomp,
                 4_000_211_221u128..=5_000_000_000u128,
-                complete_range.clone()
+                complete_range
             ),
             vec![6, 7]
         );
@@ -703,12 +703,12 @@ mod tests {
             vec![0]
         );
         assert_eq!(
-            get_positions_for_value_range_helper(&decomp, 0..=105, complete_range.clone()),
+            get_positions_for_value_range_helper(&decomp, 0..=105, complete_range),
             vec![0]
         );
     }
 
-    fn get_positions_for_value_range_helper<C: Column<T> + ?Sized, T: PartialOrd>(
+    fn get_positions_for_value_range_helper<C: Column<T> + ?Sized, T: PartialOrd + fmt::Debug>(
         column: &C,
         value_range: RangeInclusive<T>,
         doc_id_range: Range<u32>,
@@ -756,11 +756,7 @@ mod tests {
         );
 
         assert_eq!(
-            get_positions_for_value_range_helper(
-                &*decomp,
-                1_000_000..=1_000_000,
-                complete_range.clone()
-            ),
+            get_positions_for_value_range_helper(&*decomp, 1_000_000..=1_000_000, complete_range),
             vec![11]
         );
     }

fastfield_codecs/src/format_version.rs

@@ -1,7 +1,6 @@
 use std::io;
 
-use common::BinarySerializable;
-use ownedbytes::OwnedBytes;
+use common::{BinarySerializable, OwnedBytes};
 
 const MAGIC_NUMBER: u16 = 4335u16;
 const FASTFIELD_FORMAT_VERSION: u8 = 1;

fastfield_codecs/src/gcd.rs

@@ -45,7 +45,7 @@ mod tests {
     use std::io;
     use std::num::NonZeroU64;
 
-    use ownedbytes::OwnedBytes;
+    use common::OwnedBytes;
 
     use crate::gcd::{compute_gcd, find_gcd};
     use crate::{FastFieldCodecType, VecColumn};

fastfield_codecs/src/lib.rs

@@ -14,11 +14,11 @@ extern crate more_asserts;
 #[cfg(all(test, feature = "unstable"))]
 extern crate test;
 
-use std::io;
 use std::io::Write;
 use std::sync::Arc;
+use std::{fmt, io};
 
-use common::BinarySerializable;
+use common::{BinarySerializable, OwnedBytes};
 use compact_space::CompactSpaceDecompressor;
 use format_version::read_format_version;
 use monotonic_mapping::{
@@ -26,7 +26,6 @@ use monotonic_mapping::{
     StrictlyMonotonicMappingToInternalBaseval, StrictlyMonotonicMappingToInternalGCDBaseval,
 };
 use null_index_footer::read_null_index_footer;
-use ownedbytes::OwnedBytes;
 use serialize::{Header, U128Header};
 
 mod bitpacked;
@@ -37,11 +36,13 @@ mod line;
 mod linear;
 mod monotonic_mapping;
 mod monotonic_mapping_u128;
+#[allow(dead_code)]
+mod null_index;
 mod null_index_footer;
 
 mod column;
 mod gcd;
-mod serialize;
+pub mod serialize;
 
 use self::bitpacked::BitpackedCodec;
 use self::blockwise_linear::BlockwiseLinearCodec;
@@ -132,7 +133,7 @@ impl U128FastFieldCodecType {
 }
 
 /// Returns the correct codec reader wrapped in the `Arc` for the data.
-pub fn open_u128<Item: MonotonicallyMappableToU128>(
+pub fn open_u128<Item: MonotonicallyMappableToU128 + fmt::Debug>(
     bytes: OwnedBytes,
 ) -> io::Result<Arc<dyn Column<Item>>> {
     let (bytes, _format_version) = read_format_version(bytes)?;
@@ -146,7 +147,9 @@ pub fn open_u128<Item: MonotonicallyMappableToU128>(
 }
 
 /// Returns the correct codec reader wrapped in the `Arc` for the data.
-pub fn open<T: MonotonicallyMappableToU64>(bytes: OwnedBytes) -> io::Result<Arc<dyn Column<T>>> {
+pub fn open<T: MonotonicallyMappableToU64 + fmt::Debug>(
+    bytes: OwnedBytes,
+) -> io::Result<Arc<dyn Column<T>>> {
     let (bytes, _format_version) = read_format_version(bytes)?;
     let (mut bytes, _null_index_footer) = read_null_index_footer(bytes)?;
     let header = Header::deserialize(&mut bytes)?;
@@ -159,7 +162,7 @@ pub fn open<T: MonotonicallyMappableToU64>(bytes: OwnedBytes) -> io::Result<Arc<
     }
 }
 
-fn open_specific_codec<C: FastFieldCodec, Item: MonotonicallyMappableToU64>(
+fn open_specific_codec<C: FastFieldCodec, Item: MonotonicallyMappableToU64 + fmt::Debug>(
     bytes: OwnedBytes,
     header: &Header,
 ) -> io::Result<Arc<dyn Column<Item>>> {
@@ -320,6 +323,9 @@ mod tests {
     pub fn get_codec_test_datasets() -> Vec<(Vec<u64>, &'static str)> {
         let mut data_and_names = vec![];
 
+        let data = vec![10];
+        data_and_names.push((data, "minimal test"));
+
         let data = (10..=10_000_u64).collect::<Vec<_>>();
         data_and_names.push((data, "simple monotonically increasing"));
 
@@ -327,6 +333,9 @@ mod tests {
             vec![5, 6, 7, 8, 9, 10, 99, 100],
             "offset in linear interpol",
         ));
+
+        data_and_names.push((vec![3, 18446744073709551613, 5], "docid range regression"));
+
         data_and_names.push((vec![5, 50, 3, 13, 1, 1000, 35], "rand small"));
         data_and_names.push((vec![10], "single value"));
 
@@ -434,7 +443,7 @@ mod tests {
 mod bench {
     use std::sync::Arc;
 
-    use ownedbytes::OwnedBytes;
+    use common::OwnedBytes;
     use rand::rngs::StdRng;
     use rand::{Rng, SeedableRng};
     use test::{self, Bencher};

fastfield_codecs/src/linear.rs

@@ -1,7 +1,6 @@
 use std::io::{self, Write};
 
-use common::BinarySerializable;
-use ownedbytes::OwnedBytes;
+use common::{BinarySerializable, OwnedBytes};
 use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
 
 use crate::line::Line;
@@ -25,13 +24,13 @@ impl Column for LinearReader {
         interpoled_val.wrapping_add(bitpacked_diff)
     }
 
-    #[inline]
+    #[inline(always)]
     fn min_value(&self) -> u64 {
         // The LinearReader assumes a normalized vector.
         0u64
     }
 
-    #[inline]
+    #[inline(always)]
     fn max_value(&self) -> u64 {
         self.header.max_value
     }

fastfield_codecs/src/main.rs

@@ -6,10 +6,10 @@ use std::io::BufRead;
 use std::net::{IpAddr, Ipv6Addr};
 use std::str::FromStr;
 
+use common::OwnedBytes;
 use fastfield_codecs::{open_u128, serialize_u128, Column, FastFieldCodecType, VecColumn};
 use itertools::Itertools;
 use measure_time::print_time;
-use ownedbytes::OwnedBytes;
 use prettytable::{Cell, Row, Table};
 
 fn print_set_stats(ip_addrs: &[u128]) {

fastfield_codecs/src/monotonic_mapping.rs

@@ -1,4 +1,6 @@
+use std::fmt;
 use std::marker::PhantomData;
+use std::ops::RangeInclusive;
 
 use fastdivide::DividerU64;
 
@@ -6,7 +8,9 @@ use crate::MonotonicallyMappableToU128;
 
 /// Monotonic maps a value to u64 value space.
 /// Monotonic mapping enables `PartialOrd` on u64 space without conversion to original space.
-pub trait MonotonicallyMappableToU64: 'static + PartialOrd + Copy + Send + Sync {
+pub trait MonotonicallyMappableToU64:
+    'static + PartialOrd + Copy + Send + Sync + fmt::Debug
+{
     /// Converts a value to u64.
     ///
     /// Internally all fast field values are encoded as u64.
@@ -29,11 +33,29 @@ pub trait MonotonicallyMappableToU64: 'static + PartialOrd + Copy + Send + Sync
 /// mapping from their range to their domain. The `inverse` method is required when opening a codec,
 /// so a value can be converted back to its original domain (e.g. ip address or f64) from its
 /// internal representation.
-pub trait StrictlyMonotonicFn<External, Internal> {
+pub trait StrictlyMonotonicFn<External: Copy, Internal: Copy> {
     /// Strictly monotonically maps the value from External to Internal.
     fn mapping(&self, inp: External) -> Internal;
     /// Inverse of `mapping`. Maps the value from Internal to External.
     fn inverse(&self, out: Internal) -> External;
+
+    /// Maps a user provded value from External to Internal.
+    /// It may be necessary to coerce the value if it is outside the value space.
+    /// In that case it tries to find the next greater value in the value space.
+    ///
+    /// Returns a bool to mark if a value was outside the value space and had to be coerced _up_.
+    /// With that information we can detect if two values in a range both map outside the same value
+    /// space.
+    ///
+    /// coerce_up means the next valid upper value in the value space will be chosen if the value
+    /// has to be coerced.
+    fn mapping_coerce(&self, inp: RangeInclusive<External>) -> RangeInclusive<Internal> {
+        self.mapping(*inp.start())..=self.mapping(*inp.end())
+    }
+    /// Inverse of `mapping_coerce`.
+    fn inverse_coerce(&self, out: RangeInclusive<Internal>) -> RangeInclusive<External> {
+        self.inverse(*out.start())..=self.inverse(*out.end())
+    }
 }
 
 /// Inverts a strictly monotonic mapping from `StrictlyMonotonicFn<A, B>` to
@@ -54,15 +76,29 @@ impl<T> From<T> for StrictlyMonotonicMappingInverter<T> {
 }
 
 impl<From, To, T> StrictlyMonotonicFn<To, From> for StrictlyMonotonicMappingInverter<T>
-where T: StrictlyMonotonicFn<From, To>
+where
+    T: StrictlyMonotonicFn<From, To>,
+    From: Copy,
+    To: Copy,
 {
+    #[inline(always)]
     fn mapping(&self, val: To) -> From {
         self.orig_mapping.inverse(val)
     }
 
+    #[inline(always)]
     fn inverse(&self, val: From) -> To {
         self.orig_mapping.mapping(val)
     }
+
+    #[inline]
+    fn mapping_coerce(&self, inp: RangeInclusive<To>) -> RangeInclusive<From> {
+        self.orig_mapping.inverse_coerce(inp)
+    }
+    #[inline]
+    fn inverse_coerce(&self, out: RangeInclusive<From>) -> RangeInclusive<To> {
+        self.orig_mapping.mapping_coerce(out)
+    }
 }
 
 /// Applies the strictly monotonic mapping from `T` without any additional changes.
@@ -82,10 +118,12 @@ impl<External: MonotonicallyMappableToU128, T: MonotonicallyMappableToU128>
     StrictlyMonotonicFn<External, u128> for StrictlyMonotonicMappingToInternal<T>
 where T: MonotonicallyMappableToU128
 {
+    #[inline(always)]
     fn mapping(&self, inp: External) -> u128 {
         External::to_u128(inp)
     }
 
+    #[inline(always)]
     fn inverse(&self, out: u128) -> External {
         External::from_u128(out)
     }
@@ -95,10 +133,12 @@ impl<External: MonotonicallyMappableToU64, T: MonotonicallyMappableToU64>
     StrictlyMonotonicFn<External, u64> for StrictlyMonotonicMappingToInternal<T>
 where T: MonotonicallyMappableToU64
 {
+    #[inline(always)]
     fn mapping(&self, inp: External) -> u64 {
         External::to_u64(inp)
     }
 
+    #[inline(always)]
     fn inverse(&self, out: u64) -> External {
         External::from_u64(out)
     }
@@ -126,14 +166,41 @@ impl StrictlyMonotonicMappingToInternalGCDBaseval {
 impl<External: MonotonicallyMappableToU64> StrictlyMonotonicFn<External, u64>
     for StrictlyMonotonicMappingToInternalGCDBaseval
 {
+    #[inline(always)]
     fn mapping(&self, inp: External) -> u64 {
         self.gcd_divider
             .divide(External::to_u64(inp) - self.min_value)
     }
 
+    #[inline(always)]
     fn inverse(&self, out: u64) -> External {
         External::from_u64(self.min_value + out * self.gcd)
     }
+
+    #[inline]
+    #[allow(clippy::reversed_empty_ranges)]
+    fn mapping_coerce(&self, inp: RangeInclusive<External>) -> RangeInclusive<u64> {
+        let end = External::to_u64(*inp.end());
+        if end < self.min_value || inp.end() < inp.start() {
+            return 1..=0;
+        }
+        let map_coerce = |mut inp, coerce_up| {
+            let inp_lower_bound = self.inverse(0);
+            if inp < inp_lower_bound {
+                inp = inp_lower_bound;
+            }
+            let val = External::to_u64(inp);
+            let need_coercion = coerce_up && (val - self.min_value) % self.gcd != 0;
+            let mut mapped_val = self.mapping(inp);
+            if need_coercion {
+                mapped_val += 1;
+            }
+            mapped_val
+        };
+        let start = map_coerce(*inp.start(), true);
+        let end = map_coerce(*inp.end(), false);
+        start..=end
+    }
 }
 
 /// Strictly monotonic mapping with a base value.
@@ -141,6 +208,7 @@ pub(crate) struct StrictlyMonotonicMappingToInternalBaseval {
     min_value: u64,
 }
 impl StrictlyMonotonicMappingToInternalBaseval {
+    #[inline(always)]
     pub(crate) fn new(min_value: u64) -> Self {
         Self { min_value }
     }
@@ -149,20 +217,35 @@ impl StrictlyMonotonicMappingToInternalBaseval {
 impl<External: MonotonicallyMappableToU64> StrictlyMonotonicFn<External, u64>
     for StrictlyMonotonicMappingToInternalBaseval
 {
+    #[inline]
+    #[allow(clippy::reversed_empty_ranges)]
+    fn mapping_coerce(&self, inp: RangeInclusive<External>) -> RangeInclusive<u64> {
+        if External::to_u64(*inp.end()) < self.min_value {
+            return 1..=0;
+        }
+        let start = self.mapping(External::to_u64(*inp.start()).max(self.min_value));
+        let end = self.mapping(External::to_u64(*inp.end()));
+        start..=end
+    }
+
+    #[inline(always)]
     fn mapping(&self, val: External) -> u64 {
         External::to_u64(val) - self.min_value
     }
 
+    #[inline(always)]
     fn inverse(&self, val: u64) -> External {
         External::from_u64(self.min_value + val)
     }
 }
 
 impl MonotonicallyMappableToU64 for u64 {
+    #[inline(always)]
     fn to_u64(self) -> u64 {
         self
     }
 
+    #[inline(always)]
     fn from_u64(val: u64) -> Self {
         val
     }
@@ -192,11 +275,15 @@ impl MonotonicallyMappableToU64 for bool {
     }
 }
 
+// TODO remove me.
+// Tantivy should refuse NaN values and work with NotNaN internally.
 impl MonotonicallyMappableToU64 for f64 {
+    #[inline(always)]
     fn to_u64(self) -> u64 {
         common::f64_to_u64(self)
     }
 
+    #[inline(always)]
     fn from_u64(val: u64) -> Self {
         common::u64_to_f64(val)
     }
@@ -224,7 +311,7 @@ mod tests {
         test_round_trip::<_, _, u64>(&mapping, 100u64);
     }
 
-    fn test_round_trip<T: StrictlyMonotonicFn<K, L>, K: std::fmt::Debug + Eq + Copy, L>(
+    fn test_round_trip<T: StrictlyMonotonicFn<K, L>, K: std::fmt::Debug + Eq + Copy, L: Copy>(
         mapping: &T,
         test_val: K,
     ) {

fastfield_codecs/src/monotonic_mapping_u128.rs

@@ -1,8 +1,11 @@
+use std::fmt;
 use std::net::Ipv6Addr;
 
 /// Montonic maps a value to u128 value space
 /// Monotonic mapping enables `PartialOrd` on u128 space without conversion to original space.
-pub trait MonotonicallyMappableToU128: 'static + PartialOrd + Copy + Send + Sync {
+pub trait MonotonicallyMappableToU128:
+    'static + PartialOrd + Copy + Send + Sync + fmt::Debug
+{
     /// Converts a value to u128.
     ///
     /// Internally all fast field values are encoded as u64.

fastfield_codecs/src/null_index/dense.rs

@@ -0,0 +1,500 @@
+use std::convert::TryInto;
+use std::io::{self, Write};
+
+use common::{BinarySerializable, OwnedBytes};
+use itertools::Itertools;
+
+use super::{get_bit_at, set_bit_at};
+
+/// For the `DenseCodec`, `data` which contains the encoded blocks.
+/// Each block consists of [u8; 12]. The first 8 bytes is a bitvec for 64 elements.
+/// The last 4 bytes are the offset, the number of set bits so far.
+///
+/// When translating the original index to a dense index, the correct block can be computed
+/// directly `orig_idx/64`. Inside the block the position is `orig_idx%64`.
+///
+/// When translating a dense index to the original index, we can use the offset to find the correct
+/// block. Direct computation is not possible, but we can employ a linear or binary search.
+#[derive(Clone)]
+pub struct DenseCodec {
+    // data consists of blocks of 64 bits.
+    //
+    // The format is &[(u64, u32)]
+    // u64 is the bitvec
+    // u32 is the offset of the block, the number of set bits so far.
+    //
+    // At the end one block is appended, to store the number of values in the index in offset.
+    data: OwnedBytes,
+}
+const ELEMENTS_PER_BLOCK: u32 = 64;
+const BLOCK_BITVEC_SIZE: usize = 8;
+const BLOCK_OFFSET_SIZE: usize = 4;
+const SERIALIZED_BLOCK_SIZE: usize = BLOCK_BITVEC_SIZE + BLOCK_OFFSET_SIZE;
+
+/// Interpreting the bitvec as a list of 64 bits from the low weight to the
+/// high weight.
+///
+/// This function returns the number of bits set to 1 within
+/// `[0..pos_in_vec)`.
+#[inline]
+fn count_ones(bitvec: u64, pos_in_bitvec: u32) -> u32 {
+    let mask = (1u64 << pos_in_bitvec) - 1;
+    let masked_bitvec = bitvec & mask;
+    masked_bitvec.count_ones()
+}
+
+#[derive(Clone, Copy)]
+struct DenseIndexBlock {
+    bitvec: u64,
+    offset: u32,
+}
+
+impl From<[u8; SERIALIZED_BLOCK_SIZE]> for DenseIndexBlock {
+    fn from(data: [u8; SERIALIZED_BLOCK_SIZE]) -> Self {
+        let bitvec = u64::from_le_bytes(data[..BLOCK_BITVEC_SIZE].try_into().unwrap());
+        let offset = u32::from_le_bytes(data[BLOCK_BITVEC_SIZE..].try_into().unwrap());
+        Self { bitvec, offset }
+    }
+}
+
+impl DenseCodec {
+    /// Open the DenseCodec from OwnedBytes
+    pub fn open(data: OwnedBytes) -> Self {
+        Self { data }
+    }
+    #[inline]
+    /// Check if value at position is not null.
+    pub fn exists(&self, idx: u32) -> bool {
+        let block_pos = idx / ELEMENTS_PER_BLOCK;
+        let bitvec = self.dense_index_block(block_pos).bitvec;
+        let pos_in_bitvec = idx % ELEMENTS_PER_BLOCK;
+        get_bit_at(bitvec, pos_in_bitvec)
+    }
+    #[inline]
+    fn dense_index_block(&self, block_pos: u32) -> DenseIndexBlock {
+        dense_index_block(&self.data, block_pos)
+    }
+
+    /// Return the number of non-null values in an index
+    pub fn num_non_nulls(&self) -> u32 {
+        let last_block = (self.data.len() / SERIALIZED_BLOCK_SIZE) - 1;
+        self.dense_index_block(last_block as u32).offset
+    }
+
+    #[inline]
+    /// Translate from the original index to the codec index.
+    pub fn translate_to_codec_idx(&self, idx: u32) -> Option<u32> {
+        let block_pos = idx / ELEMENTS_PER_BLOCK;
+        let index_block = self.dense_index_block(block_pos);
+        let pos_in_block_bit_vec = idx % ELEMENTS_PER_BLOCK;
+        let ones_in_block = count_ones(index_block.bitvec, pos_in_block_bit_vec);
+        if get_bit_at(index_block.bitvec, pos_in_block_bit_vec) {
+            Some(index_block.offset + ones_in_block)
+        } else {
+            None
+        }
+    }
+
+    /// Translate positions from the codec index to the original index.
+    ///
+    /// # Panics
+    ///
+    /// May panic if any `idx` is greater than the max codec index.
+    pub fn translate_codec_idx_to_original_idx<'a>(
+        &'a self,
+        iter: impl Iterator<Item = u32> + 'a,
+    ) -> impl Iterator<Item = u32> + 'a {
+        let mut block_pos = 0u32;
+        iter.map(move |dense_idx| {
+            // update block_pos to limit search scope
+            block_pos = find_block(dense_idx, block_pos, &self.data);
+            let index_block = self.dense_index_block(block_pos);
+
+            // The next offset is higher than dense_idx and therefore:
+            // dense_idx <= offset + num_set_bits in block
+            let mut num_set_bits = 0;
+            for idx_in_bitvec in 0..ELEMENTS_PER_BLOCK {
+                if get_bit_at(index_block.bitvec, idx_in_bitvec) {
+                    num_set_bits += 1;
+                }
+                if num_set_bits == (dense_idx - index_block.offset + 1) {
+                    let orig_idx = block_pos * ELEMENTS_PER_BLOCK + idx_in_bitvec;
+                    return orig_idx;
+                }
+            }
+            panic!("Internal Error: Offset calculation in dense idx seems to be wrong.");
+        })
+    }
+}
+
+#[inline]
+fn dense_index_block(data: &[u8], block_pos: u32) -> DenseIndexBlock {
+    let data_start_pos = block_pos as usize * SERIALIZED_BLOCK_SIZE;
+    let block_data: [u8; SERIALIZED_BLOCK_SIZE] = data[data_start_pos..][..SERIALIZED_BLOCK_SIZE]
+        .try_into()
+        .unwrap();
+    block_data.into()
+}
+
+#[inline]
+/// Finds the block position containing the dense_idx.
+///
+/// # Correctness
+/// dense_idx needs to be smaller than the number of values in the index
+///
+/// The last offset number is equal to the number of values in the index.
+fn find_block(dense_idx: u32, mut block_pos: u32, data: &[u8]) -> u32 {
+    loop {
+        let offset = dense_index_block(data, block_pos).offset;
+        if offset > dense_idx {
+            return block_pos - 1;
+        }
+        block_pos += 1;
+    }
+}
+
+/// Iterator over all values, true if set, otherwise false
+pub fn serialize_dense_codec(
+    iter: impl Iterator<Item = bool>,
+    mut out: impl Write,
+) -> io::Result<()> {
+    let mut offset: u32 = 0;
+
+    for chunk in &iter.chunks(ELEMENTS_PER_BLOCK as usize) {
+        let mut block: u64 = 0;
+        for (pos, is_bit_set) in chunk.enumerate() {
+            if is_bit_set {
+                set_bit_at(&mut block, pos as u64);
+            }
+        }
+
+        block.serialize(&mut out)?;
+        offset.serialize(&mut out)?;
+
+        offset += block.count_ones();
+    }
+    // Add sentinal block for the offset
+    let block: u64 = 0;
+    block.serialize(&mut out)?;
+    offset.serialize(&mut out)?;
+
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use proptest::prelude::{any, prop, *};
+    use proptest::strategy::Strategy;
+    use proptest::{prop_oneof, proptest};
+
+    use super::*;
+
+    fn random_bitvec() -> BoxedStrategy<Vec<bool>> {
+        prop_oneof![
+            1 => prop::collection::vec(proptest::bool::weighted(1.0), 0..100),
+            1 => prop::collection::vec(proptest::bool::weighted(1.0), 0..64),
+            1 => prop::collection::vec(proptest::bool::weighted(0.0), 0..100),
+            1 => prop::collection::vec(proptest::bool::weighted(0.0), 0..64),
+            8 => vec![any::<bool>()],
+            2 => prop::collection::vec(any::<bool>(), 0..50),
+        ]
+        .boxed()
+    }
+
+    proptest! {
+        #![proptest_config(ProptestConfig::with_cases(500))]
+        #[test]
+        fn test_with_random_bitvecs(bitvec1 in random_bitvec(), bitvec2 in random_bitvec(), bitvec3 in random_bitvec()) {
+            let mut bitvec = Vec::new();
+            bitvec.extend_from_slice(&bitvec1);
+            bitvec.extend_from_slice(&bitvec2);
+            bitvec.extend_from_slice(&bitvec3);
+            test_null_index(bitvec);
+        }
+    }
+
+    #[test]
+    fn dense_codec_test_one_block_false() {
+        let mut iter = vec![false; 64];
+        iter.push(true);
+        test_null_index(iter);
+    }
+
+    fn test_null_index(data: Vec<bool>) {
+        let mut out = vec![];
+
+        serialize_dense_codec(data.iter().cloned(), &mut out).unwrap();
+        let null_index = DenseCodec::open(OwnedBytes::new(out));
+
+        let orig_idx_with_value: Vec<u32> = data
+            .iter()
+            .enumerate()
+            .filter(|(_pos, val)| **val)
+            .map(|(pos, _val)| pos as u32)
+            .collect();
+
+        assert_eq!(
+            null_index
+                .translate_codec_idx_to_original_idx(0..orig_idx_with_value.len() as u32)
+                .collect_vec(),
+            orig_idx_with_value
+        );
+
+        for (dense_idx, orig_idx) in orig_idx_with_value.iter().enumerate() {
+            assert_eq!(
+                null_index.translate_to_codec_idx(*orig_idx),
+                Some(dense_idx as u32)
+            );
+        }
+
+        for (pos, value) in data.iter().enumerate() {
+            assert_eq!(null_index.exists(pos as u32), *value);
+        }
+    }
+
+    #[test]
+    fn dense_codec_test_translation() {
+        let mut out = vec![];
+
+        let iter = ([true, false, true, false]).iter().cloned();
+        serialize_dense_codec(iter, &mut out).unwrap();
+        let null_index = DenseCodec::open(OwnedBytes::new(out));
+
+        assert_eq!(
+            null_index
+                .translate_codec_idx_to_original_idx(0..2)
+                .collect_vec(),
+            vec![0, 2]
+        );
+    }
+
+    #[test]
+    fn dense_codec_translate() {
+        let mut out = vec![];
+
+        let iter = ([true, false, true, false]).iter().cloned();
+        serialize_dense_codec(iter, &mut out).unwrap();
+        let null_index = DenseCodec::open(OwnedBytes::new(out));
+        assert_eq!(null_index.translate_to_codec_idx(0), Some(0));
+        assert_eq!(null_index.translate_to_codec_idx(2), Some(1));
+    }
+
+    #[test]
+    fn dense_codec_test_small() {
+        let mut out = vec![];
+
+        let iter = ([true, false, true, false]).iter().cloned();
+        serialize_dense_codec(iter, &mut out).unwrap();
+        let null_index = DenseCodec::open(OwnedBytes::new(out));
+        assert!(null_index.exists(0));
+        assert!(!null_index.exists(1));
+        assert!(null_index.exists(2));
+        assert!(!null_index.exists(3));
+    }
+
+    #[test]
+    fn dense_codec_test_large() {
+        let mut docs = vec![];
+        docs.extend((0..1000).map(|_idx| false));
+        docs.extend((0..=1000).map(|_idx| true));
+
+        let iter = docs.iter().cloned();
+        let mut out = vec![];
+        serialize_dense_codec(iter, &mut out).unwrap();
+        let null_index = DenseCodec::open(OwnedBytes::new(out));
+        assert!(!null_index.exists(0));
+        assert!(!null_index.exists(100));
+        assert!(!null_index.exists(999));
+        assert!(null_index.exists(1000));
+        assert!(null_index.exists(1999));
+        assert!(null_index.exists(2000));
+        assert!(!null_index.exists(2001));
+    }
+
+    #[test]
+    fn test_count_ones() {
+        let mut block = 0;
+        set_bit_at(&mut block, 0);
+        set_bit_at(&mut block, 2);
+
+        assert_eq!(count_ones(block, 0), 0);
+        assert_eq!(count_ones(block, 1), 1);
+        assert_eq!(count_ones(block, 2), 1);
+        assert_eq!(count_ones(block, 3), 2);
+    }
+}
+
+#[cfg(all(test, feature = "unstable"))]
+mod bench {
+
+    use rand::rngs::StdRng;
+    use rand::{Rng, SeedableRng};
+    use test::Bencher;
+
+    use super::*;
+
+    const TOTAL_NUM_VALUES: u32 = 1_000_000;
+    fn gen_bools(fill_ratio: f64) -> DenseCodec {
+        let mut out = Vec::new();
+        let mut rng: StdRng = StdRng::from_seed([1u8; 32]);
+        let bools: Vec<_> = (0..TOTAL_NUM_VALUES)
+            .map(|_| rng.gen_bool(fill_ratio))
+            .collect();
+        serialize_dense_codec(bools.into_iter(), &mut out).unwrap();
+
+        let codec = DenseCodec::open(OwnedBytes::new(out));
+        codec
+    }
+
+    fn random_range_iterator(
+        start: u32,
+        end: u32,
+        avg_step_size: u32,
+        avg_deviation: u32,
+    ) -> impl Iterator<Item = u32> {
+        let mut rng: StdRng = StdRng::from_seed([1u8; 32]);
+        let mut current = start;
+        std::iter::from_fn(move || {
+            current += rng.gen_range(avg_step_size - avg_deviation..=avg_step_size + avg_deviation);
+            if current >= end {
+                None
+            } else {
+                Some(current)
+            }
+        })
+    }
+
+    fn n_percent_step_iterator(percent: f32, num_values: u32) -> impl Iterator<Item = u32> {
+        let ratio = percent as f32 / 100.0;
+        let step_size = (1f32 / ratio) as u32;
+        let deviation = step_size - 1;
+        random_range_iterator(0, num_values, step_size, deviation)
+    }
+
+    fn walk_over_data(codec: &DenseCodec, avg_step_size: u32) -> Option<u32> {
+        walk_over_data_from_positions(
+            codec,
+            random_range_iterator(0, TOTAL_NUM_VALUES, avg_step_size, 0),
+        )
+    }
+
+    fn walk_over_data_from_positions(
+        codec: &DenseCodec,
+        positions: impl Iterator<Item = u32>,
+    ) -> Option<u32> {
+        let mut dense_idx: Option<u32> = None;
+        for idx in positions {
+            dense_idx = dense_idx.or(codec.translate_to_codec_idx(idx));
+        }
+        dense_idx
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_1percent_filled_10percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.01f64);
+        bench.iter(|| walk_over_data(&codec, 100));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_5percent_filled_10percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.05f64);
+        bench.iter(|| walk_over_data(&codec, 100));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_5percent_filled_1percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.05f64);
+        bench.iter(|| walk_over_data(&codec, 1000));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_full_scan_1percent_filled(bench: &mut Bencher) {
+        let codec = gen_bools(0.01f64);
+        bench.iter(|| walk_over_data_from_positions(&codec, 0..TOTAL_NUM_VALUES));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_full_scan_10percent_filled(bench: &mut Bencher) {
+        let codec = gen_bools(0.1f64);
+        bench.iter(|| walk_over_data_from_positions(&codec, 0..TOTAL_NUM_VALUES));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_full_scan_90percent_filled(bench: &mut Bencher) {
+        let codec = gen_bools(0.9f64);
+        bench.iter(|| walk_over_data_from_positions(&codec, 0..TOTAL_NUM_VALUES));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_10percent_filled_1percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.1f64);
+        bench.iter(|| walk_over_data(&codec, 100));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_50percent_filled_1percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.5f64);
+        bench.iter(|| walk_over_data(&codec, 100));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_90percent_filled_1percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.9f64);
+        bench.iter(|| walk_over_data(&codec, 100));
+    }
+
+    #[bench]
+    fn bench_translate_codec_to_orig_1percent_filled_0comma005percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.01f64);
+        let num_non_nulls = codec.num_non_nulls();
+        bench.iter(|| {
+            codec
+                .translate_codec_idx_to_original_idx(n_percent_step_iterator(0.005, num_non_nulls))
+                .last()
+        });
+    }
+
+    #[bench]
+    fn bench_translate_codec_to_orig_1percent_filled_10percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.01f64);
+        let num_non_nulls = codec.num_non_nulls();
+        bench.iter(|| {
+            codec
+                .translate_codec_idx_to_original_idx(n_percent_step_iterator(10.0, num_non_nulls))
+                .last()
+        });
+    }
+
+    #[bench]
+    fn bench_translate_codec_to_orig_1percent_filled_full_scan(bench: &mut Bencher) {
+        let codec = gen_bools(0.01f64);
+        let num_vals = codec.num_non_nulls();
+        bench.iter(|| {
+            codec
+                .translate_codec_idx_to_original_idx(0..num_vals)
+                .last()
+        });
+    }
+
+    #[bench]
+    fn bench_translate_codec_to_orig_90percent_filled_0comma005percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.90f64);
+        let num_non_nulls = codec.num_non_nulls();
+        bench.iter(|| {
+            codec
+                .translate_codec_idx_to_original_idx(n_percent_step_iterator(0.005, num_non_nulls))
+                .last()
+        });
+    }
+
+    #[bench]
+    fn bench_translate_codec_to_orig_90percent_filled_full_scan(bench: &mut Bencher) {
+        let codec = gen_bools(0.9f64);
+        let num_vals = codec.num_non_nulls();
+        bench.iter(|| {
+            codec
+                .translate_codec_idx_to_original_idx(0..num_vals)
+                .last()
+        });
+    }
+}

fastfield_codecs/src/null_index/mod.rs

@@ -0,0 +1,14 @@
+pub use dense::{serialize_dense_codec, DenseCodec};
+
+mod dense;
+mod sparse;
+
+#[inline]
+fn get_bit_at(input: u64, n: u32) -> bool {
+    input & (1 << n) != 0
+}
+
+#[inline]
+fn set_bit_at(input: &mut u64, n: u64) {
+    *input |= 1 << n;
+}

fastfield_codecs/src/null_index/sparse.rs

@@ -0,0 +1,768 @@
+use std::io::{self, Write};
+
+use common::{BitSet, GroupByIteratorExtended, OwnedBytes};
+
+use super::{serialize_dense_codec, DenseCodec};
+
+/// `SparseCodec` is the codec for data, when only few documents have values.
+/// In contrast to `DenseCodec` opening a `SparseCodec` causes runtime data to be produced, for
+/// faster access.
+///
+/// The lower 16 bits of doc ids are stored as u16 while the upper 16 bits are given by the block
+/// id. Each block contains 1<<16 docids.
+///
+/// # Serialized Data Layout
+/// The data starts with the block data. Each block is either dense or sparse encoded, depending on
+/// the number of values in the block. A block is sparse when it contains less than
+/// DENSE_BLOCK_THRESHOLD (6144) values.
+/// [Sparse data block | dense data block, .. #repeat*; Desc: Either a sparse or dense encoded
+/// block]
+/// ### Sparse block data
+/// [u16 LE, .. #repeat*; Desc: Positions with values in a block]
+/// ### Dense block data
+/// [Dense codec for the whole block; Desc: Similar to a bitvec(0..ELEMENTS_PER_BLOCK) + Metadata
+/// for faster lookups. See dense.rs]
+///
+/// The data is followed by block metadata, to know which area of the raw block data belongs to
+/// which block. Only metadata for blocks with elements is recorded to
+/// keep the overhead low for scenarios with many very sparse columns. The block metadata consists
+/// of the block index and the number of values in the block. Since we don't store empty blocks
+/// num_vals is incremented by 1, e.g. 0 means 1 value.
+///
+/// The last u16 is storing the number of metadata blocks.
+/// [u16 LE, .. #repeat*; Desc: Positions with values in a block][(u16 LE, u16 LE), .. #repeat*;
+/// Desc: (Block Id u16, Num Elements u16)][u16 LE; Desc: num blocks with values u16]
+///
+/// # Opening
+/// When opening the data layout, the data is expanded to `Vec<SparseCodecBlockVariant>`, where the
+/// index is the block index. For each block `byte_start` and `offset` is computed.
+pub struct SparseCodec {
+    data: OwnedBytes,
+    blocks: Vec<SparseCodecBlockVariant>,
+}
+
+/// The threshold for for number of elements after which we switch to dense block encoding
+const DENSE_BLOCK_THRESHOLD: u32 = 6144;
+
+const ELEMENTS_PER_BLOCK: u32 = u16::MAX as u32 + 1;
+
+/// 1.5 bit per Element + 12 bytes for the sentinal block
+const NUM_BYTES_DENSE_BLOCK: u32 = (ELEMENTS_PER_BLOCK + ELEMENTS_PER_BLOCK / 2 + 64 + 32) / 8;
+
+#[derive(Clone)]
+enum SparseCodecBlockVariant {
+    Empty { offset: u32 },
+    Dense(DenseBlock),
+    Sparse(SparseBlock),
+}
+
+impl SparseCodecBlockVariant {
+    /// The number of non-null values that preceeded that block.
+    #[inline]
+    fn offset(&self) -> u32 {
+        match self {
+            SparseCodecBlockVariant::Empty { offset } => *offset,
+            SparseCodecBlockVariant::Dense(dense) => dense.offset,
+            SparseCodecBlockVariant::Sparse(sparse) => sparse.offset,
+        }
+    }
+}
+
+/// A block consists of max u16 values
+#[derive(Clone)]
+struct DenseBlock {
+    /// The number of values set before the block
+    offset: u32,
+    /// The data for the dense encoding
+    codec: DenseCodec,
+}
+
+impl DenseBlock {
+    #[inline]
+    pub fn exists(&self, idx: u32) -> bool {
+        self.codec.exists(idx)
+    }
+    #[inline]
+    pub fn translate_to_codec_idx(&self, idx: u32) -> Option<u32> {
+        self.codec.translate_to_codec_idx(idx)
+    }
+    #[inline]
+    pub fn translate_codec_idx_to_original_idx_iter<'a>(
+        &'a self,
+        iter: impl Iterator<Item = u32> + 'a,
+    ) -> impl Iterator<Item = u32> + 'a {
+        self.codec.translate_codec_idx_to_original_idx(iter)
+    }
+    #[inline]
+    pub fn translate_codec_idx_to_original_idx(&self, idx: u32) -> u32 {
+        self.codec
+            .translate_codec_idx_to_original_idx(idx..=idx)
+            .next()
+            .unwrap()
+    }
+}
+
+/// A block consists of max u16 values
+#[derive(Debug, Copy, Clone)]
+struct SparseBlock {
+    /// The number of values in the block
+    num_vals: u32,
+    /// The number of values set before the block
+    offset: u32,
+    /// The start position of the data for the block
+    byte_start: u32,
+}
+
+impl SparseBlock {
+    fn empty_block(offset: u32) -> Self {
+        Self {
+            num_vals: 0,
+            byte_start: 0,
+            offset,
+        }
+    }
+
+    #[inline]
+    fn value_at_idx(&self, data: &[u8], idx: u16) -> u16 {
+        let start_offset: usize = self.byte_start as usize + (idx as u32 as usize * 2);
+        get_u16(data, start_offset)
+    }
+
+    #[inline]
+    #[allow(clippy::comparison_chain)]
+    // Looks for the element in the block. Returns the positions if found.
+    fn binary_search(&self, data: &[u8], target: u16) -> Option<u16> {
+        let mut size = self.num_vals as u16;
+        let mut left = 0;
+        let mut right = size;
+        // TODO try different implem.
+        //  e.g. exponential search into binary search
+        while left < right {
+            let mid = left + size / 2;
+
+            // TODO do boundary check only once, and then use an
+            // unsafe `value_at_idx`
+            let mid_val = self.value_at_idx(data, mid);
+
+            if target > mid_val {
+                left = mid + 1;
+            } else if target < mid_val {
+                right = mid;
+            } else {
+                return Some(mid);
+            }
+
+            size = right - left;
+        }
+        None
+    }
+}
+
+#[inline]
+fn get_u16(data: &[u8], byte_position: usize) -> u16 {
+    let bytes: [u8; 2] = data[byte_position..byte_position + 2].try_into().unwrap();
+    u16::from_le_bytes(bytes)
+}
+
+const SERIALIZED_BLOCK_METADATA_SIZE: usize = 4;
+
+fn deserialize_sparse_codec_block(data: &OwnedBytes) -> Vec<SparseCodecBlockVariant> {
+    // The number of vals so far
+    let mut offset = 0;
+    let mut sparse_codec_blocks = Vec::new();
+    let num_blocks = get_u16(data, data.len() - 2);
+    let block_data_index_start =
+        data.len() - 2 - num_blocks as usize * SERIALIZED_BLOCK_METADATA_SIZE;
+    let mut byte_start = 0;
+    for block_num in 0..num_blocks as usize {
+        let block_data_index = block_data_index_start + SERIALIZED_BLOCK_METADATA_SIZE * block_num;
+        let block_idx = get_u16(data, block_data_index);
+        let num_vals = get_u16(data, block_data_index + 2) as u32 + 1;
+        sparse_codec_blocks.resize(
+            block_idx as usize,
+            SparseCodecBlockVariant::Empty { offset },
+        );
+
+        if is_sparse(num_vals) {
+            let block = SparseBlock {
+                num_vals,
+                offset,
+                byte_start,
+            };
+            sparse_codec_blocks.push(SparseCodecBlockVariant::Sparse(block));
+            byte_start += 2 * num_vals;
+        } else {
+            let block = DenseBlock {
+                offset,
+                codec: DenseCodec::open(data.slice(byte_start as usize..data.len()).clone()),
+            };
+            sparse_codec_blocks.push(SparseCodecBlockVariant::Dense(block));
+            // Dense blocks have a fixed size spanning ELEMENTS_PER_BLOCK.
+            byte_start += NUM_BYTES_DENSE_BLOCK;
+        }
+
+        offset += num_vals;
+    }
+    sparse_codec_blocks.push(SparseCodecBlockVariant::Empty { offset });
+    sparse_codec_blocks
+}
+
+/// Splits a value address into lower and upper 16bits.
+/// The lower 16 bits are the value in the block
+/// The upper 16 bits are the block index
+#[derive(Debug, Clone, Copy)]
+struct ValueAddr {
+    block_idx: u16,
+    value_in_block: u16,
+}
+
+/// Splits a idx into block index and value in the block
+#[inline]
+fn value_addr(idx: u32) -> ValueAddr {
+    /// Static assert number elements per block this method expects
+    #[allow(clippy::assertions_on_constants)]
+    const _: () = assert!(ELEMENTS_PER_BLOCK == (1 << 16));
+
+    let value_in_block = idx as u16;
+    let block_idx = (idx >> 16) as u16;
+    ValueAddr {
+        block_idx,
+        value_in_block,
+    }
+}
+
+impl SparseCodec {
+    /// Open the SparseCodec from OwnedBytes
+    pub fn open(data: OwnedBytes) -> Self {
+        let blocks = deserialize_sparse_codec_block(&data);
+        Self { data, blocks }
+    }
+
+    #[inline]
+    /// Check if value at position is not null.
+    pub fn exists(&self, idx: u32) -> bool {
+        let value_addr = value_addr(idx);
+        // There may be trailing nulls without data, those are not stored as blocks. It would be
+        // possible to create empty blocks, but for that we would need to serialize the number of
+        // values or pass them when opening
+
+        if let Some(block) = self.blocks.get(value_addr.block_idx as usize) {
+            match block {
+                SparseCodecBlockVariant::Empty { offset: _ } => false,
+                SparseCodecBlockVariant::Dense(block) => {
+                    block.exists(value_addr.value_in_block as u32)
+                }
+                SparseCodecBlockVariant::Sparse(block) => block
+                    .binary_search(&self.data, value_addr.value_in_block)
+                    .is_some(),
+            }
+        } else {
+            false
+        }
+    }
+
+    /// Return the number of non-null values in an index
+    pub fn num_non_nulls(&self) -> u32 {
+        self.blocks.last().map(|block| block.offset()).unwrap_or(0)
+    }
+
+    #[inline]
+    /// Translate from the original index to the codec index.
+    pub fn translate_to_codec_idx(&self, idx: u32) -> Option<u32> {
+        let value_addr = value_addr(idx);
+        let block = self.blocks.get(value_addr.block_idx as usize)?;
+
+        match block {
+            SparseCodecBlockVariant::Empty { offset: _ } => None,
+            SparseCodecBlockVariant::Dense(block) => block
+                .translate_to_codec_idx(value_addr.value_in_block as u32)
+                .map(|pos_in_block| pos_in_block + block.offset),
+            SparseCodecBlockVariant::Sparse(block) => {
+                let pos_in_block = block.binary_search(&self.data, value_addr.value_in_block);
+                pos_in_block.map(|pos_in_block: u16| block.offset + pos_in_block as u32)
+            }
+        }
+    }
+
+    #[inline]
+    fn find_block(&self, dense_idx: u32, mut block_pos: u32) -> u32 {
+        loop {
+            let offset = self.blocks[block_pos as usize].offset();
+            if offset > dense_idx {
+                return block_pos - 1;
+            }
+            block_pos += 1;
+        }
+    }
+
+    /// Translate positions from the codec index to the original index.
+    /// Correctness: Provided values must be in increasing values
+    ///
+    /// # Panics
+    ///
+    /// May panic if any `idx` is greater than the max codec index.
+    pub fn translate_codec_idx_to_original_idx<'a>(
+        &'a self,
+        iter: impl Iterator<Item = u32> + 'a,
+    ) -> impl Iterator<Item = u32> + 'a {
+        let mut block_pos = 0u32;
+        iter.group_by(move |codec_idx| {
+            block_pos = self.find_block(*codec_idx, block_pos);
+            block_pos
+        })
+        .flat_map(move |(block_pos, block_iter)| {
+            let block_doc_idx_start = block_pos * ELEMENTS_PER_BLOCK;
+            let block = &self.blocks[block_pos as usize];
+            let offset = block.offset();
+            let indexes_in_block_iter = block_iter.map(move |codec_idx| codec_idx - offset);
+            match block {
+                SparseCodecBlockVariant::Empty { offset: _ } => {
+                    panic!(
+                        "invalid input, cannot translate to original index. associated empty \
+                         block with dense idx. block_pos {}, idx_in_block {:?}",
+                        block_pos,
+                        indexes_in_block_iter.collect::<Vec<_>>()
+                    )
+                }
+                SparseCodecBlockVariant::Dense(dense) => {
+                    Box::new(dense.translate_codec_idx_to_original_idx_iter(indexes_in_block_iter))
+                        as Box<dyn Iterator<Item = u32>>
+                }
+                SparseCodecBlockVariant::Sparse(block) => {
+                    Box::new(indexes_in_block_iter.map(move |idx_in_block| {
+                        block.value_at_idx(&self.data, idx_in_block as u16) as u32
+                    }))
+                }
+            }
+            .map(move |idx| idx + block_doc_idx_start)
+        })
+    }
+}
+
+#[inline]
+fn is_sparse(num_elem_in_block: u32) -> bool {
+    num_elem_in_block < DENSE_BLOCK_THRESHOLD
+}
+
+#[derive(Default)]
+struct BlockDataSerialized {
+    block_idx: u16,
+    num_vals: u32,
+}
+
+/// Iterator over positions of set values.
+pub fn serialize_sparse_codec<W: Write>(
+    mut iter: impl Iterator<Item = u32>,
+    mut out: W,
+) -> io::Result<()> {
+    let mut block_metadata: Vec<BlockDataSerialized> = Vec::new();
+    let mut current_block = Vec::new();
+    // This if-statement for the first element ensures that
+    // `block_metadata` is not empty in the loop below.
+    if let Some(idx) = iter.next() {
+        let value_addr = value_addr(idx);
+        block_metadata.push(BlockDataSerialized {
+            block_idx: value_addr.block_idx,
+            num_vals: 1,
+        });
+        current_block.push(value_addr.value_in_block);
+    }
+    let flush_block = |current_block: &mut Vec<u16>, out: &mut W| -> io::Result<()> {
+        let is_sparse = is_sparse(current_block.len() as u32);
+        if is_sparse {
+            for val_in_block in current_block.iter() {
+                out.write_all(val_in_block.to_le_bytes().as_ref())?;
+            }
+        } else {
+            let mut bitset = BitSet::with_max_value(ELEMENTS_PER_BLOCK + 1);
+            for val_in_block in current_block.iter() {
+                bitset.insert(*val_in_block as u32);
+            }
+
+            let iter = (0..ELEMENTS_PER_BLOCK).map(|idx| bitset.contains(idx));
+            serialize_dense_codec(iter, out)?;
+        }
+        current_block.clear();
+        Ok(())
+    };
+    for idx in iter {
+        let value_addr = value_addr(idx);
+        if block_metadata[block_metadata.len() - 1].block_idx == value_addr.block_idx {
+            let last_idx_metadata = block_metadata.len() - 1;
+            block_metadata[last_idx_metadata].num_vals += 1;
+        } else {
+            // flush prev block
+            flush_block(&mut current_block, &mut out)?;
+
+            block_metadata.push(BlockDataSerialized {
+                block_idx: value_addr.block_idx,
+                num_vals: 1,
+            });
+        }
+        current_block.push(value_addr.value_in_block);
+    }
+    // handle last block
+    flush_block(&mut current_block, &mut out)?;
+
+    for block in &block_metadata {
+        out.write_all(block.block_idx.to_le_bytes().as_ref())?;
+        // We don't store empty blocks, therefore we can subtract 1.
+        // This way we will be able to use u16 when the number of elements is 1 << 16 or u16::MAX+1
+        out.write_all(((block.num_vals - 1) as u16).to_le_bytes().as_ref())?;
+    }
+    out.write_all((block_metadata.len() as u16).to_le_bytes().as_ref())?;
+
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use itertools::Itertools;
+    use proptest::prelude::{any, prop, *};
+    use proptest::strategy::Strategy;
+    use proptest::{prop_oneof, proptest};
+
+    use super::*;
+
+    fn random_bitvec() -> BoxedStrategy<Vec<bool>> {
+        prop_oneof![
+            1 => prop::collection::vec(proptest::bool::weighted(1.0), 0..100),
+            1 => prop::collection::vec(proptest::bool::weighted(0.00), 0..(ELEMENTS_PER_BLOCK as usize * 3)), // empty blocks
+            1 => prop::collection::vec(proptest::bool::weighted(1.00), 0..(ELEMENTS_PER_BLOCK as usize + 10)), // full block
+            1 => prop::collection::vec(proptest::bool::weighted(0.01), 0..100),
+            1 => prop::collection::vec(proptest::bool::weighted(0.01), 0..u16::MAX as usize),
+            8 => vec![any::<bool>()],
+        ]
+        .boxed()
+    }
+
+    proptest! {
+        #![proptest_config(ProptestConfig::with_cases(50))]
+        #[test]
+        fn test_with_random_bitvecs(bitvec1 in random_bitvec(), bitvec2 in random_bitvec(), bitvec3 in random_bitvec()) {
+            let mut bitvec = Vec::new();
+            bitvec.extend_from_slice(&bitvec1);
+            bitvec.extend_from_slice(&bitvec2);
+            bitvec.extend_from_slice(&bitvec3);
+            test_null_index(bitvec);
+        }
+    }
+
+    #[test]
+    fn sparse_codec_test_one_block_false() {
+        let mut iter = vec![false; ELEMENTS_PER_BLOCK as usize];
+        iter.push(true);
+        test_null_index(iter);
+    }
+
+    #[test]
+    fn sparse_codec_test_one_block_true() {
+        let mut iter = vec![true; ELEMENTS_PER_BLOCK as usize];
+        iter.push(true);
+        test_null_index(iter);
+    }
+
+    fn test_null_index(data: Vec<bool>) {
+        let mut out = vec![];
+
+        serialize_sparse_codec(
+            data.iter()
+                .cloned()
+                .enumerate()
+                .filter(|(_pos, val)| *val)
+                .map(|(pos, _val)| pos as u32),
+            &mut out,
+        )
+        .unwrap();
+        let null_index = SparseCodec::open(OwnedBytes::new(out));
+
+        let orig_idx_with_value: Vec<u32> = data
+            .iter()
+            .enumerate()
+            .filter(|(_pos, val)| **val)
+            .map(|(pos, _val)| pos as u32)
+            .collect();
+
+        assert_eq!(
+            null_index
+                .translate_codec_idx_to_original_idx(0..orig_idx_with_value.len() as u32)
+                .collect_vec(),
+            orig_idx_with_value
+        );
+
+        let step_size = (orig_idx_with_value.len() / 100).max(1);
+        for (dense_idx, orig_idx) in orig_idx_with_value.iter().enumerate().step_by(step_size) {
+            assert_eq!(
+                null_index.translate_to_codec_idx(*orig_idx),
+                Some(dense_idx as u32)
+            );
+        }
+
+        // 100 samples
+        let step_size = (data.len() / 100).max(1);
+        for (pos, value) in data.iter().enumerate().step_by(step_size) {
+            assert_eq!(null_index.exists(pos as u32), *value);
+        }
+    }
+
+    #[test]
+    fn sparse_codec_test_translation() {
+        let mut out = vec![];
+
+        let iter = ([true, false, true, false]).iter().cloned();
+        serialize_sparse_codec(
+            iter.enumerate()
+                .filter(|(_pos, val)| *val)
+                .map(|(pos, _val)| pos as u32),
+            &mut out,
+        )
+        .unwrap();
+        let null_index = SparseCodec::open(OwnedBytes::new(out));
+
+        assert_eq!(
+            null_index
+                .translate_codec_idx_to_original_idx(0..2)
+                .collect_vec(),
+            vec![0, 2]
+        );
+    }
+
+    #[test]
+    fn sparse_codec_translate() {
+        let mut out = vec![];
+
+        let iter = ([true, false, true, false]).iter().cloned();
+        serialize_sparse_codec(
+            iter.enumerate()
+                .filter(|(_pos, val)| *val)
+                .map(|(pos, _val)| pos as u32),
+            &mut out,
+        )
+        .unwrap();
+        let null_index = SparseCodec::open(OwnedBytes::new(out));
+        assert_eq!(null_index.translate_to_codec_idx(0), Some(0));
+        assert_eq!(null_index.translate_to_codec_idx(2), Some(1));
+    }
+
+    #[test]
+    fn sparse_codec_test_small() {
+        let mut out = vec![];
+
+        let iter = ([true, false, true, false]).iter().cloned();
+        serialize_sparse_codec(
+            iter.enumerate()
+                .filter(|(_pos, val)| *val)
+                .map(|(pos, _val)| pos as u32),
+            &mut out,
+        )
+        .unwrap();
+        let null_index = SparseCodec::open(OwnedBytes::new(out));
+        assert!(null_index.exists(0));
+        assert!(!null_index.exists(1));
+        assert!(null_index.exists(2));
+        assert!(!null_index.exists(3));
+    }
+
+    #[test]
+    fn sparse_codec_test_large() {
+        let mut docs = vec![];
+        docs.extend((0..ELEMENTS_PER_BLOCK).map(|_idx| false));
+        docs.extend((0..=1).map(|_idx| true));
+
+        let iter = docs.iter().cloned();
+        let mut out = vec![];
+        serialize_sparse_codec(
+            iter.enumerate()
+                .filter(|(_pos, val)| *val)
+                .map(|(pos, _val)| pos as u32),
+            &mut out,
+        )
+        .unwrap();
+        let null_index = SparseCodec::open(OwnedBytes::new(out));
+        assert!(!null_index.exists(0));
+        assert!(!null_index.exists(100));
+        assert!(!null_index.exists(ELEMENTS_PER_BLOCK - 1));
+        assert!(null_index.exists(ELEMENTS_PER_BLOCK));
+        assert!(null_index.exists(ELEMENTS_PER_BLOCK + 1));
+    }
+}
+
+#[cfg(all(test, feature = "unstable"))]
+mod bench {
+
+    use rand::rngs::StdRng;
+    use rand::{Rng, SeedableRng};
+    use test::Bencher;
+
+    use super::*;
+
+    const TOTAL_NUM_VALUES: u32 = 1_000_000;
+    fn gen_bools(fill_ratio: f64) -> SparseCodec {
+        let mut out = Vec::new();
+        let mut rng: StdRng = StdRng::from_seed([1u8; 32]);
+        serialize_sparse_codec(
+            (0..TOTAL_NUM_VALUES)
+                .map(|_| rng.gen_bool(fill_ratio))
+                .enumerate()
+                .filter(|(_pos, val)| *val)
+                .map(|(pos, _val)| pos as u32),
+            &mut out,
+        )
+        .unwrap();
+
+        let codec = SparseCodec::open(OwnedBytes::new(out));
+        codec
+    }
+
+    fn random_range_iterator(
+        start: u32,
+        end: u32,
+        avg_step_size: u32,
+        avg_deviation: u32,
+    ) -> impl Iterator<Item = u32> {
+        let mut rng: StdRng = StdRng::from_seed([1u8; 32]);
+        let mut current = start;
+        std::iter::from_fn(move || {
+            current += rng.gen_range(avg_step_size - avg_deviation..=avg_step_size + avg_deviation);
+            if current >= end {
+                None
+            } else {
+                Some(current)
+            }
+        })
+    }
+
+    fn n_percent_step_iterator(percent: f32, num_values: u32) -> impl Iterator<Item = u32> {
+        let ratio = percent as f32 / 100.0;
+        let step_size = (1f32 / ratio) as u32;
+        let deviation = step_size - 1;
+        random_range_iterator(0, num_values, step_size, deviation)
+    }
+
+    fn walk_over_data(codec: &SparseCodec, avg_step_size: u32) -> Option<u32> {
+        walk_over_data_from_positions(
+            codec,
+            random_range_iterator(0, TOTAL_NUM_VALUES, avg_step_size, 0),
+        )
+    }
+
+    fn walk_over_data_from_positions(
+        codec: &SparseCodec,
+        positions: impl Iterator<Item = u32>,
+    ) -> Option<u32> {
+        let mut dense_idx: Option<u32> = None;
+        for idx in positions {
+            dense_idx = dense_idx.or(codec.translate_to_codec_idx(idx));
+        }
+        dense_idx
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_1percent_filled_10percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.01f64);
+        bench.iter(|| walk_over_data(&codec, 100));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_5percent_filled_10percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.05f64);
+        bench.iter(|| walk_over_data(&codec, 100));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_5percent_filled_1percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.05f64);
+        bench.iter(|| walk_over_data(&codec, 1000));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_full_scan_1percent_filled(bench: &mut Bencher) {
+        let codec = gen_bools(0.01f64);
+        bench.iter(|| walk_over_data_from_positions(&codec, 0..TOTAL_NUM_VALUES));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_full_scan_10percent_filled(bench: &mut Bencher) {
+        let codec = gen_bools(0.1f64);
+        bench.iter(|| walk_over_data_from_positions(&codec, 0..TOTAL_NUM_VALUES));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_full_scan_90percent_filled(bench: &mut Bencher) {
+        let codec = gen_bools(0.9f64);
+        bench.iter(|| walk_over_data_from_positions(&codec, 0..TOTAL_NUM_VALUES));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_10percent_filled_1percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.1f64);
+        bench.iter(|| walk_over_data(&codec, 100));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_50percent_filled_1percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.5f64);
+        bench.iter(|| walk_over_data(&codec, 100));
+    }
+
+    #[bench]
+    fn bench_translate_orig_to_codec_90percent_filled_1percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.9f64);
+        bench.iter(|| walk_over_data(&codec, 100));
+    }
+
+    #[bench]
+    fn bench_translate_codec_to_orig_1percent_filled_0comma005percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.01f64);
+        let num_non_nulls = codec.num_non_nulls();
+        bench.iter(|| {
+            codec
+                .translate_codec_idx_to_original_idx(n_percent_step_iterator(0.005, num_non_nulls))
+                .last()
+        });
+    }
+
+    #[bench]
+    fn bench_translate_codec_to_orig_1percent_filled_10percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.01f64);
+        let num_non_nulls = codec.num_non_nulls();
+        bench.iter(|| {
+            codec
+                .translate_codec_idx_to_original_idx(n_percent_step_iterator(10.0, num_non_nulls))
+                .last()
+        });
+    }
+
+    #[bench]
+    fn bench_translate_codec_to_orig_1percent_filled_full_scan(bench: &mut Bencher) {
+        let codec = gen_bools(0.01f64);
+        let num_vals = codec.num_non_nulls();
+        bench.iter(|| {
+            codec
+                .translate_codec_idx_to_original_idx(0..num_vals)
+                .last()
+        });
+    }
+
+    #[bench]
+    fn bench_translate_codec_to_orig_90percent_filled_0comma005percent_hit(bench: &mut Bencher) {
+        let codec = gen_bools(0.90f64);
+        let num_non_nulls = codec.num_non_nulls();
+        bench.iter(|| {
+            codec
+                .translate_codec_idx_to_original_idx(n_percent_step_iterator(0.005, num_non_nulls))
+                .last()
+        });
+    }
+
+    #[bench]
+    fn bench_translate_codec_to_orig_90percent_filled_full_scan(bench: &mut Bencher) {
+        let codec = gen_bools(0.9f64);
+        let num_vals = codec.num_non_nulls();
+        bench.iter(|| {
+            codec
+                .translate_codec_idx_to_original_idx(0..num_vals)
+                .last()
+        });
+    }
+}

fastfield_codecs/src/null_index_footer.rs

@@ -1,12 +1,12 @@
 use std::io::{self, Write};
 use std::ops::Range;
 
-use common::{BinarySerializable, CountingWriter, VInt};
-use ownedbytes::OwnedBytes;
+use common::{BinarySerializable, CountingWriter, OwnedBytes, VInt};
 
 #[derive(Debug, Clone, Copy, Eq, PartialEq)]
 pub(crate) enum FastFieldCardinality {
     Single = 1,
+    Multi = 2,
 }
 
 impl BinarySerializable for FastFieldCardinality {
@@ -30,6 +30,7 @@ impl FastFieldCardinality {
     pub(crate) fn from_code(code: u8) -> Option<Self> {
         match code {
             1 => Some(Self::Single),
+            2 => Some(Self::Multi),
             _ => None,
         }
     }

fastfield_codecs/src/serialize.rs

@@ -17,13 +17,12 @@
 // You should have received a copy of the GNU Affero General Public License
 // along with this program. If not, see <http://www.gnu.org/licenses/>.
 
-use std::io;
 use std::num::NonZeroU64;
 use std::sync::Arc;
+use std::{fmt, io};
 
-use common::{BinarySerializable, VInt};
+use common::{BinarySerializable, OwnedBytes, VInt};
 use log::warn;
-use ownedbytes::OwnedBytes;
 
 use crate::bitpacked::BitpackedCodec;
 use crate::blockwise_linear::BlockwiseLinearCodec;
@@ -168,7 +167,7 @@ impl BinarySerializable for Header {
 
 /// Return estimated compression for given codec in the value range [0.0..1.0], where 1.0 means no
 /// compression.
-pub fn estimate<T: MonotonicallyMappableToU64>(
+pub fn estimate<T: MonotonicallyMappableToU64 + fmt::Debug>(
     typed_column: impl Column<T>,
     codec_type: FastFieldCodecType,
 ) -> Option<f32> {
@@ -193,6 +192,69 @@ pub fn serialize_u128<F: Fn() -> I, I: Iterator<Item = u128>>(
     iter_gen: F,
     num_vals: u32,
     output: &mut impl io::Write,
+) -> io::Result<()> {
+    serialize_u128_new(ValueIndexInfo::default(), iter_gen, num_vals, output)
+}
+
+#[allow(dead_code)]
+pub enum ValueIndexInfo<'a> {
+    MultiValue(Box<dyn MultiValueIndexInfo + 'a>),
+    SingleValue(Box<dyn SingleValueIndexInfo + 'a>),
+}
+
+// TODO Remove me
+impl Default for ValueIndexInfo<'static> {
+    fn default() -> Self {
+        struct Dummy {}
+        impl SingleValueIndexInfo for Dummy {
+            fn num_vals(&self) -> u32 {
+                todo!()
+            }
+            fn num_non_nulls(&self) -> u32 {
+                todo!()
+            }
+            fn iter(&self) -> Box<dyn Iterator<Item = u32>> {
+                todo!()
+            }
+        }
+
+        Self::SingleValue(Box::new(Dummy {}))
+    }
+}
+
+impl<'a> ValueIndexInfo<'a> {
+    fn get_cardinality(&self) -> FastFieldCardinality {
+        match self {
+            ValueIndexInfo::MultiValue(_) => FastFieldCardinality::Multi,
+            ValueIndexInfo::SingleValue(_) => FastFieldCardinality::Single,
+        }
+    }
+}
+
+pub trait MultiValueIndexInfo {
+    /// The number of docs in the column.
+    fn num_docs(&self) -> u32;
+    /// The number of values in the column.
+    fn num_vals(&self) -> u32;
+    /// Return the start index of the values for each doc
+    fn iter(&self) -> Box<dyn Iterator<Item = u32> + '_>;
+}
+
+pub trait SingleValueIndexInfo {
+    /// The number of values including nulls in the column.
+    fn num_vals(&self) -> u32;
+    /// The number of non-null values in the column.
+    fn num_non_nulls(&self) -> u32;
+    /// Return a iterator of the positions of docs with a value
+    fn iter(&self) -> Box<dyn Iterator<Item = u32> + '_>;
+}
+
+/// Serializes u128 values with the compact space codec.
+pub fn serialize_u128_new<F: Fn() -> I, I: Iterator<Item = u128>>(
+    value_index: ValueIndexInfo,
+    iter_gen: F,
+    num_vals: u32,
+    output: &mut impl io::Write,
 ) -> io::Result<()> {
     let header = U128Header {
         num_vals,
@@ -203,7 +265,7 @@ pub fn serialize_u128<F: Fn() -> I, I: Iterator<Item = u128>>(
     compressor.compress_into(iter_gen(), output).unwrap();
 
     let null_index_footer = NullIndexFooter {
-        cardinality: FastFieldCardinality::Single,
+        cardinality: value_index.get_cardinality(),
         null_index_codec: NullIndexCodec::Full,
         null_index_byte_range: 0..0,
     };
@@ -214,7 +276,17 @@ pub fn serialize_u128<F: Fn() -> I, I: Iterator<Item = u128>>(
 }
 
 /// Serializes the column with the codec with the best estimate on the data.
-pub fn serialize<T: MonotonicallyMappableToU64>(
+pub fn serialize<T: MonotonicallyMappableToU64 + fmt::Debug>(
+    typed_column: impl Column<T>,
+    output: &mut impl io::Write,
+    codecs: &[FastFieldCodecType],
+) -> io::Result<()> {
+    serialize_new(ValueIndexInfo::default(), typed_column, output, codecs)
+}
+
+/// Serializes the column with the codec with the best estimate on the data.
+pub fn serialize_new<T: MonotonicallyMappableToU64 + fmt::Debug>(
+    value_index: ValueIndexInfo,
     typed_column: impl Column<T>,
     output: &mut impl io::Write,
     codecs: &[FastFieldCodecType],
@@ -235,7 +307,7 @@ pub fn serialize<T: MonotonicallyMappableToU64>(
     serialize_given_codec(normalized_column, header.codec_type, output)?;
 
     let null_index_footer = NullIndexFooter {
-        cardinality: FastFieldCardinality::Single,
+        cardinality: value_index.get_cardinality(),
         null_index_codec: NullIndexCodec::Full,
         null_index_byte_range: 0..0,
     };
@@ -294,7 +366,7 @@ fn serialize_given_codec(
 }
 
 /// Helper function to serialize a column (autodetect from all codecs) and then open it
-pub fn serialize_and_load<T: MonotonicallyMappableToU64 + Ord + Default>(
+pub fn serialize_and_load<T: MonotonicallyMappableToU64 + Ord + Default + fmt::Debug>(
     column: &[T],
 ) -> Arc<dyn Column<T>> {
     let mut buffer = Vec::new();

ownedbytes/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 authors = ["Paul Masurel <paul@quickwit.io>", "Pascal Seitz <pascal@quickwit.io>"]
 name = "ownedbytes"
-version = "0.4.0"
+version = "0.5.0"
 edition = "2021"
 description = "Expose data as static slice"
 license = "MIT"

ownedbytes/src/lib.rs

@@ -3,7 +3,7 @@ use std::ops::{Deref, Range};
 use std::sync::Arc;
 use std::{fmt, io, mem};
 
-use stable_deref_trait::StableDeref;
+pub use stable_deref_trait::StableDeref;
 
 /// An OwnedBytes simply wraps an object that owns a slice of data and exposes
 /// this data as a slice.

src/aggregation/bucket/histogram/histogram.rs

@@ -401,7 +401,7 @@ impl SegmentHistogramCollector {
 
             debug_assert_eq!(
                 self.buckets[bucket_pos].key,
-                get_bucket_val(val, self.interval, self.offset) as f64
+                get_bucket_val(val, self.interval, self.offset)
             );
             self.increment_bucket(bucket_pos, doc, &bucket_with_accessor.sub_aggregation)?;
         }
@@ -428,7 +428,7 @@ impl SegmentHistogramCollector {
         if bounds.contains(val) {
             debug_assert_eq!(
                 self.buckets[bucket_pos].key,
-                get_bucket_val(val, self.interval, self.offset) as f64
+                get_bucket_val(val, self.interval, self.offset)
             );
 
             self.increment_bucket(bucket_pos, doc, bucket_with_accessor)?;

src/aggregation/intermediate_agg_result.rs

@@ -282,8 +282,8 @@ impl IntermediateBucketResult {
             IntermediateBucketResult::Range(range_res) => {
                 let mut buckets: Vec<RangeBucketEntry> = range_res
                     .buckets
-                    .into_iter()
-                    .map(|(_, bucket)| {
+                    .into_values()
+                    .map(|bucket| {
                         bucket.into_final_bucket_entry(
                             &req.sub_aggregation,
                             schema,

src/aggregation/mod.rs

@@ -451,9 +451,9 @@ mod tests {
                         text_field_id => term.to_string(),
                         string_field_id => term.to_string(),
                         score_field => i as u64,
-                        score_field_f64 => i as f64,
+                        score_field_f64 => i,
                         score_field_i64 => i as i64,
-                        fraction_field => i as f64/100.0,
+                        fraction_field => i/100.0,
                     ))?;
                 }
                 index_writer.commit()?;
@@ -1208,7 +1208,7 @@ mod tests {
                         text_field_many_terms => many_terms_data.choose(&mut rng).unwrap().to_string(),
                         text_field_few_terms => few_terms_data.choose(&mut rng).unwrap().to_string(),
                         score_field => val as u64,
-                        score_field_f64 => val as f64,
+                        score_field_f64 => val,
                         score_field_i64 => val as i64,
                     ))?;
                 }
@@ -1250,10 +1250,7 @@ mod tests {
                 let collector = AggregationCollector::from_aggs(agg_req_1, None, index.schema());
 
                 let searcher = reader.searcher();
-                let agg_res: AggregationResults =
-                    searcher.search(&term_query, &collector).unwrap().into();
-
-                agg_res
+                searcher.search(&term_query, &collector).unwrap()
             });
         }
 
@@ -1281,10 +1278,7 @@ mod tests {
                 let collector = AggregationCollector::from_aggs(agg_req_1, None, index.schema());
 
                 let searcher = reader.searcher();
-                let agg_res: AggregationResults =
-                    searcher.search(&term_query, &collector).unwrap().into();
-
-                agg_res
+                searcher.search(&term_query, &collector).unwrap()
             });
         }
 
@@ -1312,10 +1306,7 @@ mod tests {
                 let collector = AggregationCollector::from_aggs(agg_req_1, None, index.schema());
 
                 let searcher = reader.searcher();
-                let agg_res: AggregationResults =
-                    searcher.search(&term_query, &collector).unwrap().into();
-
-                agg_res
+                searcher.search(&term_query, &collector).unwrap()
             });
         }
 
@@ -1351,10 +1342,7 @@ mod tests {
                 let collector = AggregationCollector::from_aggs(agg_req_1, None, index.schema());
 
                 let searcher = reader.searcher();
-                let agg_res: AggregationResults =
-                    searcher.search(&term_query, &collector).unwrap().into();
-
-                agg_res
+                searcher.search(&term_query, &collector).unwrap()
             });
         }
 
@@ -1380,10 +1368,7 @@ mod tests {
                 let collector = AggregationCollector::from_aggs(agg_req, None, index.schema());
 
                 let searcher = reader.searcher();
-                let agg_res: AggregationResults =
-                    searcher.search(&AllQuery, &collector).unwrap().into();
-
-                agg_res
+                searcher.search(&AllQuery, &collector).unwrap()
             });
         }
 
@@ -1409,10 +1394,7 @@ mod tests {
                 let collector = AggregationCollector::from_aggs(agg_req, None, index.schema());
 
                 let searcher = reader.searcher();
-                let agg_res: AggregationResults =
-                    searcher.search(&AllQuery, &collector).unwrap().into();
-
-                agg_res
+                searcher.search(&AllQuery, &collector).unwrap()
             });
         }
 
@@ -1446,10 +1428,7 @@ mod tests {
                 let collector = AggregationCollector::from_aggs(agg_req_1, None, index.schema());
 
                 let searcher = reader.searcher();
-                let agg_res: AggregationResults =
-                    searcher.search(&AllQuery, &collector).unwrap().into();
-
-                agg_res
+                searcher.search(&AllQuery, &collector).unwrap()
             });
         }
 
@@ -1481,10 +1460,7 @@ mod tests {
                 let collector = AggregationCollector::from_aggs(agg_req_1, None, index.schema());
 
                 let searcher = reader.searcher();
-                let agg_res: AggregationResults =
-                    searcher.search(&AllQuery, &collector).unwrap().into();
-
-                agg_res
+                searcher.search(&AllQuery, &collector).unwrap()
             });
         }
 
@@ -1520,10 +1496,7 @@ mod tests {
                 let collector = AggregationCollector::from_aggs(agg_req_1, None, index.schema());
 
                 let searcher = reader.searcher();
-                let agg_res: AggregationResults =
-                    searcher.search(&AllQuery, &collector).unwrap().into();
-
-                agg_res
+                searcher.search(&AllQuery, &collector).unwrap()
             });
         }
 
@@ -1550,10 +1523,7 @@ mod tests {
                 let collector = AggregationCollector::from_aggs(agg_req_1, None, index.schema());
 
                 let searcher = reader.searcher();
-                let agg_res: AggregationResults =
-                    searcher.search(&AllQuery, &collector).unwrap().into();
-
-                agg_res
+                searcher.search(&AllQuery, &collector).unwrap()
             });
         }
 
@@ -1597,7 +1567,7 @@ mod tests {
                                 ],
                                 ..Default::default()
                             }),
-                            sub_aggregation: sub_agg_req_1.clone(),
+                            sub_aggregation: sub_agg_req_1,
                         }),
                     ),
                 ]
@@ -1607,10 +1577,7 @@ mod tests {
                 let collector = AggregationCollector::from_aggs(agg_req_1, None, index.schema());
 
                 let searcher = reader.searcher();
-                let agg_res: AggregationResults =
-                    searcher.search(&term_query, &collector).unwrap().into();
-
-                agg_res
+                searcher.search(&term_query, &collector).unwrap()
             });
         }
     }

src/aggregation/segment_agg_result.rs

@@ -305,7 +305,7 @@ impl BucketCount {
     }
     pub(crate) fn add_count(&self, count: u32) {
         self.bucket_count
-            .fetch_add(count as u32, std::sync::atomic::Ordering::Relaxed);
+            .fetch_add(count, std::sync::atomic::Ordering::Relaxed);
     }
     pub(crate) fn get_count(&self) -> u32 {
         self.bucket_count.load(std::sync::atomic::Ordering::Relaxed)

src/collector/facet_collector.rs

@@ -357,7 +357,7 @@ impl SegmentCollector for FacetSegmentCollector {
             let mut facet = vec![];
             let facet_ord = self.collapse_facet_ords[collapsed_facet_ord];
             // TODO handle errors.
-            if facet_dict.ord_to_term(facet_ord as u64, &mut facet).is_ok() {
+            if facet_dict.ord_to_term(facet_ord, &mut facet).is_ok() {
                 if let Ok(facet) = Facet::from_encoded(facet) {
                     facet_counts.insert(facet, count);
                 }

src/collector/mod.rs

@@ -170,7 +170,7 @@ pub trait Collector: Sync + Send {
         segment_ord: u32,
         reader: &SegmentReader,
     ) -> crate::Result<<Self::Child as SegmentCollector>::Fruit> {
-        let mut segment_collector = self.for_segment(segment_ord as u32, reader)?;
+        let mut segment_collector = self.for_segment(segment_ord, reader)?;
 
         match (reader.alive_bitset(), self.requires_scoring()) {
             (Some(alive_bitset), true) => {

src/core/index.rs

@@ -813,7 +813,7 @@ mod tests {
             let field = schema.get_field("num_likes").unwrap();
             let tempdir = TempDir::new().unwrap();
             let tempdir_path = PathBuf::from(tempdir.path());
-            let index = Index::create_in_dir(&tempdir_path, schema).unwrap();
+            let index = Index::create_in_dir(tempdir_path, schema).unwrap();
             let reader = index
                 .reader_builder()
                 .reload_policy(ReloadPolicy::OnCommit)

src/core/inverted_index_reader.rs

@@ -200,10 +200,7 @@ impl InvertedIndexReader {
 
 #[cfg(feature = "quickwit")]
 impl InvertedIndexReader {
-    pub(crate) async fn get_term_info_async(
-        &self,
-        term: &Term,
-    ) -> crate::AsyncIoResult<Option<TermInfo>> {
+    pub(crate) async fn get_term_info_async(&self, term: &Term) -> io::Result<Option<TermInfo>> {
         self.termdict.get_async(term.value_bytes()).await
     }
 
@@ -211,12 +208,8 @@ impl InvertedIndexReader {
     /// This method is for an advanced usage only.
     ///
     /// Most users should prefer using [`Self::read_postings()`] instead.
-    pub async fn warm_postings(
-        &self,
-        term: &Term,
-        with_positions: bool,
-    ) -> crate::AsyncIoResult<()> {
-        let term_info_opt = self.get_term_info_async(term).await?;
+    pub async fn warm_postings(&self, term: &Term, with_positions: bool) -> io::Result<()> {
+        let term_info_opt: Option<TermInfo> = self.get_term_info_async(term).await?;
         if let Some(term_info) = term_info_opt {
             self.postings_file_slice
                 .read_bytes_slice_async(term_info.postings_range.clone())
@@ -234,7 +227,7 @@ impl InvertedIndexReader {
     /// This method is for an advanced usage only.
     ///
     /// If you know which terms to pre-load, prefer using [`Self::warm_postings`] instead.
-    pub async fn warm_postings_full(&self, with_positions: bool) -> crate::AsyncIoResult<()> {
+    pub async fn warm_postings_full(&self, with_positions: bool) -> io::Result<()> {
         self.postings_file_slice.read_bytes_async().await?;
         if with_positions {
             self.positions_file_slice.read_bytes_async().await?;
@@ -243,7 +236,7 @@ impl InvertedIndexReader {
     }
 
     /// Returns the number of documents containing the term asynchronously.
-    pub async fn doc_freq_async(&self, term: &Term) -> crate::AsyncIoResult<u32> {
+    pub async fn doc_freq_async(&self, term: &Term) -> io::Result<u32> {
         Ok(self
             .get_term_info_async(term)
             .await?

src/core/searcher.rs

@@ -198,11 +198,10 @@ impl Searcher {
         collector: &C,
         executor: &Executor,
     ) -> crate::Result<C::Fruit> {
-        let scoring_enabled = collector.requires_scoring();
-        let enabled_scoring = if scoring_enabled {
-            EnableScoring::Enabled(self)
+        let enabled_scoring = if collector.requires_scoring() {
+            EnableScoring::enabled_from_searcher(self)
         } else {
-            EnableScoring::Disabled(self.schema())
+            EnableScoring::disabled_from_searcher(self)
         };
         let weight = query.weight(enabled_scoring)?;
         let segment_readers = self.segment_readers();

src/directory/composite_file.rs

@@ -75,7 +75,7 @@ impl<W: TerminatingWrite + Write> CompositeWrite<W> {
 
         let mut prev_offset = 0;
         for (file_addr, offset) in self.offsets {
-            VInt((offset - prev_offset) as u64).serialize(&mut self.write)?;
+            VInt(offset - prev_offset).serialize(&mut self.write)?;
             file_addr.serialize(&mut self.write)?;
             prev_offset = offset;
         }

src/directory/error.rs

@@ -32,7 +32,7 @@ impl LockError {
 /// Error that may occur when opening a directory
 #[derive(Debug, Clone, Error)]
 pub enum OpenDirectoryError {
-    /// The underlying directory does not exists.
+    /// The underlying directory does not exist.
     #[error("Directory does not exist: '{0}'.")]
     DoesNotExist(PathBuf),
     /// The path exists but is not a directory.
@@ -151,8 +151,8 @@ impl fmt::Debug for Incompatibility {
 /// Error that may occur when accessing a file read
 #[derive(Debug, Clone, Error)]
 pub enum OpenReadError {
-    /// The file does not exists.
-    #[error("Files does not exists: {0:?}")]
+    /// The file does not exist.
+    #[error("Files does not exist: {0:?}")]
     FileDoesNotExist(PathBuf),
     /// Any kind of io::Error.
     #[error(
@@ -181,8 +181,8 @@ impl OpenReadError {
 /// Error that may occur when trying to delete a file
 #[derive(Debug, Clone, Error)]
 pub enum DeleteError {
-    /// The file does not exists.
-    #[error("File does not exists: '{0}'.")]
+    /// The file does not exist.
+    #[error("File does not exist: '{0}'.")]
     FileDoesNotExist(PathBuf),
     /// Any kind of IO error that happens when
     /// interacting with the underlying IO device.

src/directory/footer.rs

@@ -38,7 +38,7 @@ impl Footer {
         counting_write.write_all(serde_json::to_string(&self)?.as_ref())?;
         let footer_payload_len = counting_write.written_bytes();
         BinarySerializable::serialize(&(footer_payload_len as u32), write)?;
-        BinarySerializable::serialize(&(FOOTER_MAGIC_NUMBER as u32), write)?;
+        BinarySerializable::serialize(&FOOTER_MAGIC_NUMBER, write)?;
         Ok(())
     }
 
@@ -90,9 +90,10 @@ impl Footer {
             ));
         }
 
-        let footer: Footer = serde_json::from_slice(&file.read_bytes_slice(
-            file.len() - total_footer_size..file.len() - footer_metadata_len as usize,
-        )?)?;
+        let footer: Footer =
+            serde_json::from_slice(&file.read_bytes_slice(
+                file.len() - total_footer_size..file.len() - footer_metadata_len,
+            )?)?;
 
         let body = file.slice_to(file.len() - total_footer_size);
         Ok((footer, body))

src/directory/managed_directory.rs

@@ -388,7 +388,7 @@ mod tests_mmap_specific {
         let tempdir_path = PathBuf::from(tempdir.path());
         let living_files = HashSet::new();
 
-        let mmap_directory = MmapDirectory::open(&tempdir_path).unwrap();
+        let mmap_directory = MmapDirectory::open(tempdir_path).unwrap();
         let mut managed_directory = ManagedDirectory::wrap(Box::new(mmap_directory)).unwrap();
         let mut write = managed_directory.open_write(test_path1).unwrap();
         write.write_all(&[0u8, 1u8]).unwrap();

src/directory/mmap_directory.rs

@@ -6,10 +6,10 @@ use std::path::{Path, PathBuf};
 use std::sync::{Arc, RwLock, Weak};
 use std::{fmt, result};
 
+use common::StableDeref;
 use fs2::FileExt;
 use memmap2::Mmap;
 use serde::{Deserialize, Serialize};
-use stable_deref_trait::StableDeref;
 use tempfile::TempDir;
 
 use crate::core::META_FILEPATH;
@@ -341,7 +341,7 @@ impl Directory for MmapDirectory {
     /// removed before the file is deleted.
     fn delete(&self, path: &Path) -> result::Result<(), DeleteError> {
         let full_path = self.resolve_path(path);
-        fs::remove_file(&full_path).map_err(|e| {
+        fs::remove_file(full_path).map_err(|e| {
             if e.kind() == io::ErrorKind::NotFound {
                 DeleteError::FileDoesNotExist(path.to_owned())
             } else {
@@ -395,7 +395,7 @@ impl Directory for MmapDirectory {
     fn atomic_read(&self, path: &Path) -> Result<Vec<u8>, OpenReadError> {
         let full_path = self.resolve_path(path);
         let mut buffer = Vec::new();
-        match File::open(&full_path) {
+        match File::open(full_path) {
             Ok(mut file) => {
                 file.read_to_end(&mut buffer).map_err(|io_error| {
                     OpenReadError::wrap_io_error(io_error, path.to_path_buf())
@@ -425,7 +425,7 @@ impl Directory for MmapDirectory {
         let file: File = OpenOptions::new()
             .write(true)
             .create(true) //< if the file does not exist yet, create it.
-            .open(&full_path)
+            .open(full_path)
             .map_err(LockError::wrap_io_error)?;
         if lock.is_blocking {
             file.lock_exclusive().map_err(LockError::wrap_io_error)?;

src/directory/mod.rs

@@ -5,7 +5,6 @@ mod mmap_directory;
 
 mod directory;
 mod directory_lock;
-mod file_slice;
 mod file_watcher;
 mod footer;
 mod managed_directory;
@@ -20,13 +19,12 @@ mod composite_file;
 use std::io::BufWriter;
 use std::path::PathBuf;
 
-pub use common::{AntiCallToken, TerminatingWrite};
-pub use ownedbytes::OwnedBytes;
+pub use common::file_slice::{FileHandle, FileSlice};
+pub use common::{AntiCallToken, OwnedBytes, TerminatingWrite};
 
 pub(crate) use self::composite_file::{CompositeFile, CompositeWrite};
 pub use self::directory::{Directory, DirectoryClone, DirectoryLock};
 pub use self::directory_lock::{Lock, INDEX_WRITER_LOCK, META_LOCK};
-pub use self::file_slice::{FileHandle, FileSlice};
 pub use self::ram_directory::RamDirectory;
 pub use self::watch_event_router::{WatchCallback, WatchCallbackList, WatchHandle};
 

src/directory/ram_directory.rs

@@ -232,7 +232,7 @@ impl Directory for RamDirectory {
         let path_buf = PathBuf::from(path);
         self.fs.write().unwrap().write(path_buf, data);
         if path == *META_FILEPATH {
-            let _ = self.fs.write().unwrap().watch_router.broadcast();
+            drop(self.fs.write().unwrap().watch_router.broadcast());
         }
         Ok(())
     }

src/directory/watch_event_router.rs

@@ -168,7 +168,7 @@ mod tests {
         watch_event_router.broadcast().wait().unwrap();
         assert_eq!(2, counter.load(Ordering::SeqCst));
         mem::drop(handle_a);
-        let _ = watch_event_router.broadcast();
+        drop(watch_event_router.broadcast());
         watch_event_router.broadcast().wait().unwrap();
         assert_eq!(2, counter.load(Ordering::SeqCst));
     }

src/error.rs

@@ -104,28 +104,6 @@ pub enum TantivyError {
     InternalError(String),
 }
 
-#[cfg(feature = "quickwit")]
-#[derive(Error, Debug)]
-#[doc(hidden)]
-pub enum AsyncIoError {
-    #[error("io::Error `{0}`")]
-    Io(#[from] io::Error),
-    #[error("Asynchronous API is unsupported by this directory")]
-    AsyncUnsupported,
-}
-
-#[cfg(feature = "quickwit")]
-impl From<AsyncIoError> for TantivyError {
-    fn from(async_io_err: AsyncIoError) -> Self {
-        match async_io_err {
-            AsyncIoError::Io(io_err) => TantivyError::from(io_err),
-            AsyncIoError::AsyncUnsupported => {
-                TantivyError::SystemError(format!("{:?}", async_io_err))
-            }
-        }
-    }
-}
-
 impl From<io::Error> for TantivyError {
     fn from(io_err: io::Error) -> TantivyError {
         TantivyError::IoError(Arc::new(io_err))

src/fastfield/alive_bitset.rs

@@ -1,8 +1,7 @@
 use std::io;
 use std::io::Write;
 
-use common::{intersect_bitsets, BitSet, ReadOnlyBitSet};
-use ownedbytes::OwnedBytes;
+use common::{intersect_bitsets, BitSet, OwnedBytes, ReadOnlyBitSet};
 
 use crate::space_usage::ByteCount;
 use crate::DocId;
@@ -176,7 +175,7 @@ mod bench {
 
     fn get_alive() -> Vec<u32> {
         let mut data = (0..1_000_000_u32).collect::<Vec<u32>>();
-        for _ in 0..(1_000_000) * 1 / 8 {
+        for _ in 0..1_000_000 / 8 {
             remove_rand(&mut data);
         }
         data

src/fastfield/bytes/mod.rs

@@ -96,7 +96,7 @@ mod tests {
         let term = Term::from_field_bytes(field, b"lucene".as_ref());
         let term_query = TermQuery::new(term, IndexRecordOption::Basic);
         let term_weight_err =
-            term_query.specialized_weight(EnableScoring::Disabled(searcher.schema()));
+            term_query.specialized_weight(EnableScoring::disabled_from_schema(searcher.schema()));
         assert!(matches!(
             term_weight_err,
             Err(crate::TantivyError::SchemaError(_))

src/fastfield/facet_reader.rs

@@ -64,9 +64,7 @@ impl FacetReader {
         facet_ord: TermOrdinal,
         output: &mut Facet,
     ) -> crate::Result<()> {
-        let found_term = self
-            .term_dict
-            .ord_to_term(facet_ord as u64, &mut self.buffer)?;
+        let found_term = self.term_dict.ord_to_term(facet_ord, &mut self.buffer)?;
         assert!(found_term, "Term ordinal {} no found.", facet_ord);
         let facet_str = str::from_utf8(&self.buffer[..])
             .map_err(|utf8_err| DataCorruption::comment_only(utf8_err.to_string()))?;

src/fastfield/mod.rs

@@ -12,13 +12,15 @@
 //!
 //!
 //! Fields have to be declared as `FAST` in the schema.
-//! Currently supported fields are: u64, i64, f64, bytes and text.
+//! Currently supported fields are: u64, i64, f64, bytes, ip and text.
 //!
 //! Fast fields are stored in with [different codecs](fastfield_codecs). The best codec is detected
 //! automatically, when serializing.
 //!
 //! Read access performance is comparable to that of an array lookup.
 
+use std::net::Ipv6Addr;
+
 use fastfield_codecs::MonotonicallyMappableToU64;
 
 pub use self::alive_bitset::{intersect_alive_bitsets, write_alive_bitset, AliveBitSet};
@@ -28,7 +30,7 @@ pub use self::facet_reader::FacetReader;
 pub(crate) use self::multivalued::{get_fastfield_codecs_for_multivalue, MultivalueStartIndex};
 pub use self::multivalued::{
     MultiValueIndex, MultiValueU128FastFieldWriter, MultiValuedFastFieldReader,
-    MultiValuedFastFieldWriter, MultiValuedU128FastFieldReader,
+    MultiValuedFastFieldWriter,
 };
 pub(crate) use self::readers::type_and_cardinality;
 pub use self::readers::FastFieldReaders;
@@ -47,6 +49,33 @@ mod readers;
 mod serializer;
 mod writer;
 
+/// Trait for types that provide a zero value.
+///
+/// The resulting value is never used, just as placeholder, e.g. for `vec.resize()`.
+pub trait MakeZero {
+    /// Build a default value. This default value is never used, so the value does not
+    /// really matter.
+    fn make_zero() -> Self;
+}
+
+impl<T: FastValue> MakeZero for T {
+    fn make_zero() -> Self {
+        T::from_u64(0)
+    }
+}
+
+impl MakeZero for u128 {
+    fn make_zero() -> Self {
+        0
+    }
+}
+
+impl MakeZero for Ipv6Addr {
+    fn make_zero() -> Self {
+        Ipv6Addr::from(0u128.to_be_bytes())
+    }
+}
+
 /// Trait for types that are allowed for fast fields:
 /// (u64, i64 and f64, bool, DateTime).
 pub trait FastValue:
@@ -54,12 +83,6 @@ pub trait FastValue:
 {
     /// Returns the `schema::Type` for this FastValue.
     fn to_type() -> Type;
-
-    /// Build a default value. This default value is never used, so the value does not
-    /// really matter.
-    fn make_zero() -> Self {
-        Self::from_u64(0u64)
-    }
 }
 
 impl FastValue for u64 {
@@ -101,12 +124,6 @@ impl FastValue for DateTime {
     fn to_type() -> Type {
         Type::Date
     }
-
-    fn make_zero() -> Self {
-        DateTime {
-            timestamp_micros: 0,
-        }
-    }
 }
 
 fn value_to_u64(value: &Value) -> crate::Result<u64> {
@@ -145,7 +162,7 @@ impl FastFieldType {
 mod tests {
 
     use std::collections::HashMap;
-    use std::ops::Range;
+    use std::ops::{Range, RangeInclusive};
     use std::path::Path;
     use std::sync::Arc;
 
@@ -159,7 +176,9 @@ mod tests {
     use super::*;
     use crate::directory::{CompositeFile, Directory, RamDirectory, WritePtr};
     use crate::merge_policy::NoMergePolicy;
-    use crate::schema::{Cardinality, Document, Field, Schema, SchemaBuilder, FAST, STRING, TEXT};
+    use crate::schema::{
+        Cardinality, Document, Field, Schema, SchemaBuilder, FAST, INDEXED, STRING, TEXT,
+    };
     use crate::time::OffsetDateTime;
     use crate::{DateOptions, DatePrecision, Index, SegmentId, SegmentReader};
 
@@ -473,7 +492,7 @@ mod tests {
             let fast_field_reader = open::<u64>(data)?;
 
             for a in 0..n {
-                assert_eq!(fast_field_reader.get_val(a as u32), permutation[a as usize]);
+                assert_eq!(fast_field_reader.get_val(a as u32), permutation[a]);
             }
         }
         Ok(())
@@ -520,11 +539,6 @@ mod tests {
         Ok(())
     }
 
-    #[test]
-    fn test_default_date() {
-        assert_eq!(0, DateTime::make_zero().into_timestamp_secs());
-    }
-
     fn get_vals_for_docs(ff: &MultiValuedFastFieldReader<u64>, docs: Range<u32>) -> Vec<u64> {
         let mut all = vec![];
 
@@ -969,4 +983,117 @@ mod tests {
         }
         Ok(len)
     }
+
+    #[test]
+    fn test_gcd_bug_regression_1757() {
+        let mut schema_builder = Schema::builder();
+        let num_field = schema_builder.add_u64_field("url_norm_hash", FAST | INDEXED);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        {
+            let mut writer = index.writer_for_tests().unwrap();
+            writer
+                .add_document(doc! {
+                    num_field => 100u64,
+                })
+                .unwrap();
+            writer
+                .add_document(doc! {
+                    num_field => 200u64,
+                })
+                .unwrap();
+            writer
+                .add_document(doc! {
+                    num_field => 300u64,
+                })
+                .unwrap();
+
+            writer.commit().unwrap();
+        }
+
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+        let segment = &searcher.segment_readers()[0];
+        let field = segment.fast_fields().u64(num_field).unwrap();
+
+        let numbers = vec![100, 200, 300];
+        let test_range = |range: RangeInclusive<u64>| {
+            let expexted_count = numbers.iter().filter(|num| range.contains(num)).count();
+            let mut vec = vec![];
+            field.get_docids_for_value_range(range, 0..u32::MAX, &mut vec);
+            assert_eq!(vec.len(), expexted_count);
+        };
+        test_range(50..=50);
+        test_range(150..=150);
+        test_range(350..=350);
+        test_range(100..=250);
+        test_range(101..=200);
+        test_range(101..=199);
+        test_range(100..=300);
+        test_range(100..=299);
+    }
+
+    #[test]
+    fn test_mapping_bug_docids_for_value_range() {
+        let mut schema_builder = Schema::builder();
+        let num_field = schema_builder.add_u64_field("url_norm_hash", FAST | INDEXED);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        {
+            // Values without gcd, but with min_value
+            let mut writer = index.writer_for_tests().unwrap();
+            writer
+                .add_document(doc! {
+                    num_field => 1000u64,
+                })
+                .unwrap();
+            writer
+                .add_document(doc! {
+                    num_field => 1001u64,
+                })
+                .unwrap();
+            writer
+                .add_document(doc! {
+                    num_field => 1003u64,
+                })
+                .unwrap();
+            writer.commit().unwrap();
+        }
+
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+        let segment = &searcher.segment_readers()[0];
+        let field = segment.fast_fields().u64(num_field).unwrap();
+
+        let numbers = vec![1000, 1001, 1003];
+        let test_range = |range: RangeInclusive<u64>| {
+            let expexted_count = numbers.iter().filter(|num| range.contains(num)).count();
+            let mut vec = vec![];
+            field.get_docids_for_value_range(range, 0..u32::MAX, &mut vec);
+            assert_eq!(vec.len(), expexted_count);
+        };
+        let test_range_variant = |start, stop| {
+            let start_range = start..=stop;
+            test_range(start_range);
+            let start_range = start..=(stop - 1);
+            test_range(start_range);
+            let start_range = start..=(stop + 1);
+            test_range(start_range);
+            let start_range = (start - 1)..=stop;
+            test_range(start_range);
+            let start_range = (start - 1)..=(stop - 1);
+            test_range(start_range);
+            let start_range = (start - 1)..=(stop + 1);
+            test_range(start_range);
+            let start_range = (start + 1)..=stop;
+            test_range(start_range);
+            let start_range = (start + 1)..=(stop - 1);
+            test_range(start_range);
+            let start_range = (start + 1)..=(stop + 1);
+            test_range(start_range);
+        };
+        test_range_variant(50, 50);
+        test_range_variant(1000, 1000);
+        test_range_variant(1000, 1002);
+    }
 }

src/fastfield/multivalued/index.rs

@@ -80,6 +80,7 @@ impl MultiValueIndex {
     ///
     /// TODO: Instead of a linear scan we can employ a exponential search into binary search to
     /// match a docid to its value position.
+    #[allow(clippy::bool_to_int_with_if)]
     pub(crate) fn positions_to_docids(&self, doc_id_range: Range<u32>, positions: &mut Vec<u32>) {
         if positions.is_empty() {
             return;

src/fastfield/multivalued/mod.rs

@@ -5,7 +5,7 @@ mod writer;
 use fastfield_codecs::FastFieldCodecType;
 pub use index::MultiValueIndex;
 
-pub use self::reader::{MultiValuedFastFieldReader, MultiValuedU128FastFieldReader};
+pub use self::reader::MultiValuedFastFieldReader;
 pub(crate) use self::writer::MultivalueStartIndex;
 pub use self::writer::{MultiValueU128FastFieldWriter, MultiValuedFastFieldWriter};
 
@@ -525,7 +525,7 @@ mod bench {
             serializer.close().unwrap();
             field
         };
-        let file = directory.open_read(&path).unwrap();
+        let file = directory.open_read(path).unwrap();
         {
             let fast_fields_composite = CompositeFile::open(&file).unwrap();
             let data_idx = fast_fields_composite

src/fastfield/multivalued/reader.rs

@@ -1,107 +1,31 @@
+use core::fmt;
 use std::ops::{Range, RangeInclusive};
 use std::sync::Arc;
 
-use fastfield_codecs::{Column, MonotonicallyMappableToU128};
+use fastfield_codecs::Column;
 
 use super::MultiValueIndex;
-use crate::fastfield::FastValue;
+use crate::fastfield::MakeZero;
 use crate::DocId;
 
-/// Reader for a multivalued `u64` fast field.
+/// Reader for a multivalued fast field.
 ///
-/// The reader is implemented as two `u64` fast field.
+/// The reader is implemented as two fast fields, one u64 fast field for the index and one for the
+/// values.
 ///
-/// The `vals_reader` will access the concatenated list of all
-/// values for all reader.
-/// The `idx_reader` associated, for each document, the index of its first value.
-/// Stores the start position for each document.
+/// The `vals_reader` will access the concatenated list of all values.
+/// The `idx_reader` associates, for each document, the index of its first value.
 #[derive(Clone)]
-pub struct MultiValuedFastFieldReader<Item: FastValue> {
-    idx_reader: MultiValueIndex,
-    vals_reader: Arc<dyn Column<Item>>,
-}
-
-impl<Item: FastValue> MultiValuedFastFieldReader<Item> {
-    pub(crate) fn open(
-        idx_reader: Arc<dyn Column<u64>>,
-        vals_reader: Arc<dyn Column<Item>>,
-    ) -> MultiValuedFastFieldReader<Item> {
-        MultiValuedFastFieldReader {
-            idx_reader: MultiValueIndex::new(idx_reader),
-            vals_reader,
-        }
-    }
-
-    /// Returns the array of values associated with the given `doc`.
-    #[inline]
-    fn get_vals_for_range(&self, range: Range<u32>, vals: &mut Vec<Item>) {
-        let len = (range.end - range.start) as usize;
-        vals.resize(len, Item::make_zero());
-        self.vals_reader
-            .get_range(range.start as u64, &mut vals[..]);
-    }
-
-    /// Returns the array of values associated with the given `doc`.
-    #[inline]
-    pub fn get_vals(&self, doc: DocId, vals: &mut Vec<Item>) {
-        let range = self.idx_reader.range(doc);
-        self.get_vals_for_range(range, vals);
-    }
-
-    /// returns the multivalue index
-    pub fn get_index_reader(&self) -> &MultiValueIndex {
-        &self.idx_reader
-    }
-
-    /// Returns the minimum value for this fast field.
-    ///
-    /// The min value does not take in account of possible
-    /// deleted document, and should be considered as a lower bound
-    /// of the actual minimum value.
-    pub fn min_value(&self) -> Item {
-        self.vals_reader.min_value()
-    }
-
-    /// Returns the maximum value for this fast field.
-    ///
-    /// The max value does not take in account of possible
-    /// deleted document, and should be considered as an upper bound
-    /// of the actual maximum value.
-    pub fn max_value(&self) -> Item {
-        self.vals_reader.max_value()
-    }
-
-    /// Returns the number of values associated with the document `DocId`.
-    #[inline]
-    pub fn num_vals(&self, doc: DocId) -> u32 {
-        self.idx_reader.num_vals_for_doc(doc)
-    }
-
-    /// Returns the overall number of values in this field.
-    #[inline]
-    pub fn total_num_vals(&self) -> u32 {
-        self.idx_reader.total_num_vals()
-    }
-}
-
-/// Reader for a multivalued `u128` fast field.
-///
-/// The reader is implemented as a `u64` fast field for the index and a `u128` fast field.
-///
-/// The `vals_reader` will access the concatenated list of all
-/// values for all reader.
-/// The `idx_reader` associated, for each document, the index of its first value.
-#[derive(Clone)]
-pub struct MultiValuedU128FastFieldReader<T: MonotonicallyMappableToU128> {
+pub struct MultiValuedFastFieldReader<T> {
     idx_reader: MultiValueIndex,
     vals_reader: Arc<dyn Column<T>>,
 }
 
-impl<T: MonotonicallyMappableToU128> MultiValuedU128FastFieldReader<T> {
+impl<T: PartialOrd + MakeZero + Copy + fmt::Debug> MultiValuedFastFieldReader<T> {
     pub(crate) fn open(
         idx_reader: Arc<dyn Column<u64>>,
         vals_reader: Arc<dyn Column<T>>,
-    ) -> MultiValuedU128FastFieldReader<T> {
+    ) -> MultiValuedFastFieldReader<T> {
         Self {
             idx_reader: MultiValueIndex::new(idx_reader),
             vals_reader,
@@ -122,7 +46,7 @@ impl<T: MonotonicallyMappableToU128> MultiValuedU128FastFieldReader<T> {
     #[inline]
     fn get_vals_for_range(&self, range: Range<u32>, vals: &mut Vec<T>) {
         let len = (range.end - range.start) as usize;
-        vals.resize(len, T::from_u128(0));
+        vals.resize(len, T::make_zero());
         self.vals_reader
             .get_range(range.start as u64, &mut vals[..]);
     }
@@ -199,8 +123,131 @@ impl<T: MonotonicallyMappableToU128> MultiValuedU128FastFieldReader<T> {
 #[cfg(test)]
 mod tests {
 
+    use time::{Duration, OffsetDateTime};
+
+    use crate::collector::Count;
     use crate::core::Index;
+    use crate::query::RangeQuery;
     use crate::schema::{Cardinality, Facet, FacetOptions, NumericOptions, Schema};
+    use crate::{DateOptions, DatePrecision, DateTime};
+
+    #[test]
+    fn test_multivalued_date_docids_for_value_range_1() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        let date_field = schema_builder.add_date_field(
+            "multi_date_field",
+            DateOptions::default()
+                .set_fast(Cardinality::MultiValues)
+                .set_indexed()
+                .set_fieldnorm()
+                .set_precision(DatePrecision::Microseconds)
+                .set_stored(),
+        );
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_for_tests()?;
+        let first_time_stamp = OffsetDateTime::now_utc();
+        index_writer.add_document(doc!(
+            date_field => DateTime::from_utc(first_time_stamp),
+            date_field => DateTime::from_utc(first_time_stamp),
+        ))?;
+        // add another second
+        let two_secs_ahead = first_time_stamp + Duration::seconds(2);
+        index_writer.commit()?;
+
+        let reader = index.reader()?;
+        let searcher = reader.searcher();
+        let reader = searcher.segment_reader(0);
+
+        let date_ff_reader = reader.fast_fields().dates(date_field).unwrap();
+        let mut docids = vec![];
+        date_ff_reader.get_docids_for_value_range(
+            DateTime::from_utc(first_time_stamp)..=DateTime::from_utc(two_secs_ahead),
+            0..5,
+            &mut docids,
+        );
+        assert_eq!(docids, vec![0]);
+
+        let count_multiples =
+            |range_query: RangeQuery| searcher.search(&range_query, &Count).unwrap();
+
+        assert_eq!(
+            count_multiples(RangeQuery::new_date(
+                date_field,
+                DateTime::from_utc(first_time_stamp)..DateTime::from_utc(two_secs_ahead)
+            )),
+            1
+        );
+
+        Ok(())
+    }
+
+    #[test]
+    fn test_multivalued_date_docids_for_value_range_2() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        let date_field = schema_builder.add_date_field(
+            "multi_date_field",
+            DateOptions::default()
+                .set_fast(Cardinality::MultiValues)
+                // TODO: Test different precision after fixing https://github.com/quickwit-oss/tantivy/issues/1783
+                .set_precision(DatePrecision::Microseconds)
+                .set_indexed()
+                .set_fieldnorm()
+                .set_stored(),
+        );
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_for_tests()?;
+        let first_time_stamp = OffsetDateTime::now_utc();
+        index_writer.add_document(doc!(
+            date_field => DateTime::from_utc(first_time_stamp),
+            date_field => DateTime::from_utc(first_time_stamp),
+        ))?;
+        index_writer.add_document(doc!())?;
+        // add one second
+        index_writer.add_document(doc!(
+            date_field => DateTime::from_utc(first_time_stamp + Duration::seconds(1)),
+        ))?;
+        // add another second
+        let two_secs_ahead = first_time_stamp + Duration::seconds(2);
+        index_writer.add_document(doc!(
+            date_field => DateTime::from_utc(two_secs_ahead),
+            date_field => DateTime::from_utc(two_secs_ahead),
+            date_field => DateTime::from_utc(two_secs_ahead),
+        ))?;
+        // add three seconds
+        index_writer.add_document(doc!(
+            date_field => DateTime::from_utc(first_time_stamp + Duration::seconds(3)),
+        ))?;
+        index_writer.commit()?;
+
+        let reader = index.reader()?;
+        let searcher = reader.searcher();
+        let reader = searcher.segment_reader(0);
+        assert_eq!(reader.num_docs(), 5);
+
+        let date_ff_reader = reader.fast_fields().dates(date_field).unwrap();
+        let mut docids = vec![];
+        date_ff_reader.get_docids_for_value_range(
+            DateTime::from_utc(first_time_stamp)..=DateTime::from_utc(two_secs_ahead),
+            0..5,
+            &mut docids,
+        );
+        assert_eq!(docids, vec![0, 2, 3]);
+
+        let count_multiples =
+            |range_query: RangeQuery| searcher.search(&range_query, &Count).unwrap();
+
+        assert_eq!(
+            count_multiples(RangeQuery::new_date(
+                date_field,
+                DateTime::from_utc(first_time_stamp)..DateTime::from_utc(two_secs_ahead)
+            )),
+            2
+        );
+
+        Ok(())
+    }
 
     #[test]
     fn test_multifastfield_reader() -> crate::Result<()> {

src/fastfield/multivalued/writer.rs

@@ -264,7 +264,7 @@ fn iter_remapped_multivalue_index<'a, C: Column>(
     std::iter::once(0).chain(doc_id_map.iter_old_doc_ids().map(move |old_doc| {
         let num_vals_for_doc = column.get_val(old_doc + 1) - column.get_val(old_doc);
         offset += num_vals_for_doc;
-        offset as u64
+        offset
     }))
 }
 

src/fastfield/readers.rs

@@ -3,11 +3,9 @@ use std::sync::Arc;
 
 use fastfield_codecs::{open, open_u128, Column};
 
-use super::multivalued::MultiValuedU128FastFieldReader;
+use super::multivalued::MultiValuedFastFieldReader;
 use crate::directory::{CompositeFile, FileSlice};
-use crate::fastfield::{
-    BytesFastFieldReader, FastFieldNotAvailableError, FastValue, MultiValuedFastFieldReader,
-};
+use crate::fastfield::{BytesFastFieldReader, FastFieldNotAvailableError, FastValue};
 use crate::schema::{Cardinality, Field, FieldType, Schema};
 use crate::space_usage::PerFieldSpaceUsage;
 use crate::{DateTime, TantivyError};
@@ -161,20 +159,14 @@ impl FastFieldReaders {
     /// Returns the `ip` fast field reader reader associated to `field`.
     ///
     /// If `field` is not a u128 fast field, this method returns an Error.
-    pub fn ip_addrs(
-        &self,
-        field: Field,
-    ) -> crate::Result<MultiValuedU128FastFieldReader<Ipv6Addr>> {
+    pub fn ip_addrs(&self, field: Field) -> crate::Result<MultiValuedFastFieldReader<Ipv6Addr>> {
         self.check_type(field, FastType::U128, Cardinality::MultiValues)?;
         let idx_reader: Arc<dyn Column<u64>> = self.typed_fast_field_reader(field)?;
 
         let bytes = self.fast_field_data(field, 1)?.read_bytes()?;
         let vals_reader = open_u128::<Ipv6Addr>(bytes)?;
 
-        Ok(MultiValuedU128FastFieldReader::open(
-            idx_reader,
-            vals_reader,
-        ))
+        Ok(MultiValuedFastFieldReader::open(idx_reader, vals_reader))
     }
 
     /// Returns the `u128` fast field reader reader associated to `field`.
@@ -189,17 +181,14 @@ impl FastFieldReaders {
     /// Returns the `u128` multi-valued fast field reader reader associated to `field`.
     ///
     /// If `field` is not a u128 multi-valued fast field, this method returns an Error.
-    pub fn u128s(&self, field: Field) -> crate::Result<MultiValuedU128FastFieldReader<u128>> {
+    pub fn u128s(&self, field: Field) -> crate::Result<MultiValuedFastFieldReader<u128>> {
         self.check_type(field, FastType::U128, Cardinality::MultiValues)?;
         let idx_reader: Arc<dyn Column<u64>> = self.typed_fast_field_reader(field)?;
 
         let bytes = self.fast_field_data(field, 1)?.read_bytes()?;
         let vals_reader = open_u128::<u128>(bytes)?;
 
-        Ok(MultiValuedU128FastFieldReader::open(
-            idx_reader,
-            vals_reader,
-        ))
+        Ok(MultiValuedFastFieldReader::open(idx_reader, vals_reader))
     }
 
     /// Returns the `u64` fast field reader reader associated with `field`, regardless of whether

src/fastfield/serializer/mod.rs

@@ -1,3 +1,4 @@
+use std::fmt;
 use std::io::{self, Write};
 
 pub use fastfield_codecs::Column;
@@ -49,7 +50,7 @@ impl CompositeFastFieldSerializer {
 
     /// Serialize data into a new u64 fast field. The best compression codec will be chosen
     /// automatically.
-    pub fn create_auto_detect_u64_fast_field<T: MonotonicallyMappableToU64>(
+    pub fn create_auto_detect_u64_fast_field<T: MonotonicallyMappableToU64 + fmt::Debug>(
         &mut self,
         field: Field,
         fastfield_accessor: impl Column<T>,
@@ -59,7 +60,9 @@ impl CompositeFastFieldSerializer {
 
     /// Serialize data into a new u64 fast field. The best compression codec will be chosen
     /// automatically.
-    pub fn create_auto_detect_u64_fast_field_with_idx<T: MonotonicallyMappableToU64>(
+    pub fn create_auto_detect_u64_fast_field_with_idx<
+        T: MonotonicallyMappableToU64 + fmt::Debug,
+    >(
         &mut self,
         field: Field,
         fastfield_accessor: impl Column<T>,
@@ -72,7 +75,9 @@ impl CompositeFastFieldSerializer {
 
     /// Serialize data into a new u64 fast field. The best compression codec of the the provided
     /// will be chosen.
-    pub fn create_auto_detect_u64_fast_field_with_idx_and_codecs<T: MonotonicallyMappableToU64>(
+    pub fn create_auto_detect_u64_fast_field_with_idx_and_codecs<
+        T: MonotonicallyMappableToU64 + fmt::Debug,
+    >(
         &mut self,
         field: Field,
         fastfield_accessor: impl Column<T>,

src/fastfield/writer.rs

@@ -360,20 +360,10 @@ impl U128FastFieldWriter {
                     .map(|idx| self.vals[idx as usize])
             };
 
-            serializer.create_u128_fast_field_with_idx(
-                self.field,
-                iter_gen,
-                self.val_count as u32,
-                0,
-            )?;
+            serializer.create_u128_fast_field_with_idx(self.field, iter_gen, self.val_count, 0)?;
         } else {
             let iter_gen = || self.vals.iter().cloned();
-            serializer.create_u128_fast_field_with_idx(
-                self.field,
-                iter_gen,
-                self.val_count as u32,
-                0,
-            )?;
+            serializer.create_u128_fast_field_with_idx(self.field, iter_gen, self.val_count, 0)?;
         }
 
         Ok(())

src/indexer/demuxer.rs

@@ -252,8 +252,8 @@ mod tests {
             &demux_mapping,
             target_settings,
             vec![
-                Box::new(RamDirectory::default()),
-                Box::new(RamDirectory::default()),
+                Box::<RamDirectory>::default(),
+                Box::<RamDirectory>::default(),
             ],
         )?;
 

src/indexer/index_writer.rs

@@ -152,7 +152,7 @@ pub(crate) fn advance_deletes(
     let num_deleted_docs = max_doc - num_alive_docs;
     if num_deleted_docs > num_deleted_docs_before {
         // There are new deletes. We need to write a new delete file.
-        segment = segment.with_delete_meta(num_deleted_docs as u32, target_opstamp);
+        segment = segment.with_delete_meta(num_deleted_docs, target_opstamp);
         let mut alive_doc_file = segment.open_write(SegmentComponent::Delete)?;
         write_alive_bitset(&alive_bitset, &mut alive_doc_file)?;
         alive_doc_file.terminate()?;
@@ -678,7 +678,7 @@ impl IndexWriter {
     /// only after calling `commit()`.
     #[doc(hidden)]
     pub fn delete_query(&self, query: Box<dyn Query>) -> crate::Result<Opstamp> {
-        let weight = query.weight(EnableScoring::Disabled(&self.index.schema()))?;
+        let weight = query.weight(EnableScoring::disabled_from_schema(&self.index.schema()))?;
         let opstamp = self.stamper.stamp();
         let delete_operation = DeleteOperation {
             opstamp,
@@ -759,7 +759,8 @@ impl IndexWriter {
             match user_op {
                 UserOperation::Delete(term) => {
                     let query = TermQuery::new(term, IndexRecordOption::Basic);
-                    let weight = query.weight(EnableScoring::Disabled(&self.index.schema()))?;
+                    let weight =
+                        query.weight(EnableScoring::disabled_from_schema(&self.index.schema()))?;
                     let delete_operation = DeleteOperation {
                         opstamp,
                         target: weight,
@@ -984,7 +985,7 @@ mod tests {
             "LogMergePolicy { min_num_segments: 8, max_docs_before_merge: 10000000, \
              min_layer_size: 10000, level_log_size: 0.75, del_docs_ratio_before_merge: 1.0 }"
         );
-        let merge_policy = Box::new(NoMergePolicy::default());
+        let merge_policy = Box::<NoMergePolicy>::default();
         index_writer.set_merge_policy(merge_policy);
         assert_eq!(
             format!("{:?}", index_writer.get_merge_policy()),
@@ -1813,8 +1814,8 @@ mod tests {
         }
 
         let num_docs_expected = expected_ids_and_num_occurrences
-            .iter()
-            .map(|(_, id_occurrences)| *id_occurrences as usize)
+            .values()
+            .map(|id_occurrences| *id_occurrences as usize)
             .sum::<usize>();
         assert_eq!(searcher.num_docs() as usize, num_docs_expected);
         assert_eq!(old_searcher.num_docs() as usize, num_docs_expected);

src/indexer/json_term_writer.rs

@@ -89,11 +89,11 @@ pub(crate) fn index_json_values<'a>(
     Ok(())
 }
 
-fn index_json_object<'a>(
+fn index_json_object(
     doc: DocId,
     json_value: &serde_json::Map<String, serde_json::Value>,
     text_analyzer: &TextAnalyzer,
-    json_term_writer: &mut JsonTermWriter<'a>,
+    json_term_writer: &mut JsonTermWriter,
     postings_writer: &mut dyn PostingsWriter,
     ctx: &mut IndexingContext,
     positions_per_path: &mut IndexingPositionsPerPath,
@@ -113,11 +113,11 @@ fn index_json_object<'a>(
     }
 }
 
-fn index_json_value<'a>(
+fn index_json_value(
     doc: DocId,
     json_value: &serde_json::Value,
     text_analyzer: &TextAnalyzer,
-    json_term_writer: &mut JsonTermWriter<'a>,
+    json_term_writer: &mut JsonTermWriter,
     postings_writer: &mut dyn PostingsWriter,
     ctx: &mut IndexingContext,
     positions_per_path: &mut IndexingPositionsPerPath,

src/indexer/merger.rs

@@ -13,7 +13,7 @@ use crate::docset::{DocSet, TERMINATED};
 use crate::error::DataCorruption;
 use crate::fastfield::{
     get_fastfield_codecs_for_multivalue, AliveBitSet, Column, CompositeFastFieldSerializer,
-    MultiValueIndex, MultiValuedFastFieldReader, MultiValuedU128FastFieldReader,
+    MultiValueIndex, MultiValuedFastFieldReader,
 };
 use crate::fieldnorm::{FieldNormReader, FieldNormReaders, FieldNormsSerializer, FieldNormsWriter};
 use crate::indexer::doc_id_mapping::{expect_field_id_for_sort_field, SegmentDocIdMapping};
@@ -331,18 +331,18 @@ impl IndexMerger {
         fast_field_serializer: &mut CompositeFastFieldSerializer,
         doc_id_mapping: &SegmentDocIdMapping,
     ) -> crate::Result<()> {
-        let segment_and_ff_readers: Vec<(&SegmentReader, MultiValuedU128FastFieldReader<u128>)> =
-            self.readers
-                .iter()
-                .map(|segment_reader| {
-                    let ff_reader: MultiValuedU128FastFieldReader<u128> =
-                        segment_reader.fast_fields().u128s(field).expect(
-                            "Failed to find index for multivalued field. This is a bug in \
-                             tantivy, please report.",
-                        );
-                    (segment_reader, ff_reader)
-                })
-                .collect::<Vec<_>>();
+        let segment_and_ff_readers: Vec<(&SegmentReader, MultiValuedFastFieldReader<u128>)> = self
+            .readers
+            .iter()
+            .map(|segment_reader| {
+                let ff_reader: MultiValuedFastFieldReader<u128> =
+                    segment_reader.fast_fields().u128s(field).expect(
+                        "Failed to find index for multivalued field. This is a bug in tantivy, \
+                         please report.",
+                    );
+                (segment_reader, ff_reader)
+            })
+            .collect::<Vec<_>>();
 
         Self::write_1_n_fast_field_idx_generic(
             field,
@@ -366,7 +366,7 @@ impl IndexMerger {
                         .map(|doc| reader.num_vals(doc))
                         .sum()
                 } else {
-                    reader.total_num_vals() as u32
+                    reader.total_num_vals()
                 }
             })
             .sum();
@@ -968,7 +968,7 @@ impl IndexMerger {
                     let doc_bytes = doc_bytes_res?;
                     store_writer.store_bytes(&doc_bytes)?;
                 } else {
-                    return Err(DataCorruption::comment_only(&format!(
+                    return Err(DataCorruption::comment_only(format!(
                         "unexpected missing document in docstore on merge, doc address \
                          {old_doc_addr:?}",
                     ))

src/indexer/segment_updater.rs

@@ -577,7 +577,7 @@ impl SegmentUpdater {
         for merge_operation in merge_candidates {
             // If a merge cannot be started this is not a fatal error.
             // We do log a warning in `start_merge`.
-            let _ = self.start_merge(merge_operation);
+            drop(self.start_merge(merge_operation));
         }
     }
 
@@ -866,7 +866,7 @@ mod tests {
         }
 
         assert_eq!(indices.len(), 3);
-        let output_directory: Box<dyn Directory> = Box::new(RamDirectory::default());
+        let output_directory: Box<dyn Directory> = Box::<RamDirectory>::default();
         let index = merge_indices(&indices, output_directory)?;
         assert_eq!(index.schema(), schema);
 

src/indexer/stamper.rs

@@ -16,11 +16,11 @@ mod atomic_impl {
 
     impl AtomicU64Wrapper {
         pub fn new(first_opstamp: Opstamp) -> AtomicU64Wrapper {
-            AtomicU64Wrapper(AtomicU64::new(first_opstamp as u64))
+            AtomicU64Wrapper(AtomicU64::new(first_opstamp))
         }
 
         pub fn fetch_add(&self, val: u64, order: Ordering) -> u64 {
-            self.0.fetch_add(val as u64, order) as u64
+            self.0.fetch_add(val, order)
         }
 
         pub fn revert(&self, val: u64, order: Ordering) -> u64 {
@@ -77,7 +77,7 @@ impl Stamper {
     }
 
     pub fn stamp(&self) -> Opstamp {
-        self.0.fetch_add(1u64, Ordering::SeqCst) as u64
+        self.0.fetch_add(1u64, Ordering::SeqCst)
     }
 
     /// Given a desired count `n`, `stamps` returns an iterator that

src/lib.rs

@@ -1,17 +1,14 @@
 #![doc(html_logo_url = "http://fulmicoton.com/tantivy-logo/tantivy-logo.png")]
 #![cfg_attr(all(feature = "unstable", test), feature(test))]
-#![cfg_attr(
-    feature = "cargo-clippy",
-    allow(
-        clippy::module_inception,
-        clippy::needless_range_loop,
-        clippy::bool_assert_comparison
-    )
-)]
 #![doc(test(attr(allow(unused_variables), deny(warnings))))]
 #![warn(missing_docs)]
-#![allow(clippy::len_without_is_empty)]
-#![allow(clippy::derive_partial_eq_without_eq)]
+#![allow(
+    clippy::len_without_is_empty,
+    clippy::derive_partial_eq_without_eq,
+    clippy::module_inception,
+    clippy::needless_range_loop,
+    clippy::bool_assert_comparison
+)]
 
 //! # `tantivy`
 //!
@@ -144,7 +141,7 @@ use crate::time::{OffsetDateTime, PrimitiveDateTime, UtcOffset};
 /// All constructors and conversions are provided as explicit
 /// functions and not by implementing any `From`/`Into` traits
 /// to prevent unintended usage.
-#[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
+#[derive(Clone, Default, Copy, PartialEq, Eq, PartialOrd, Ord)]
 pub struct DateTime {
     // Timestamp in microseconds.
     pub(crate) timestamp_micros: i64,
@@ -177,7 +174,7 @@ impl DateTime {
     /// The given date/time is converted to UTC and the actual
     /// time zone is discarded.
     pub const fn from_utc(dt: OffsetDateTime) -> Self {
-        let timestamp_micros = dt.unix_timestamp() as i64 * 1_000_000 + dt.microsecond() as i64;
+        let timestamp_micros = dt.unix_timestamp() * 1_000_000 + dt.microsecond() as i64;
         Self { timestamp_micros }
     }
 
@@ -259,10 +256,6 @@ pub use crate::future_result::FutureResult;
 /// and instead, refer to this as `crate::Result<T>`.
 pub type Result<T> = std::result::Result<T, TantivyError>;
 
-/// Result for an Async io operation.
-#[cfg(feature = "quickwit")]
-pub type AsyncIoResult<T> = std::result::Result<T, crate::error::AsyncIoError>;
-
 mod core;
 mod indexer;
 

src/positions/reader.rs

@@ -71,7 +71,7 @@ impl PositionReader {
             .map(|num_bits| num_bits as usize)
             .sum();
         let num_bytes_to_skip = num_bits * COMPRESSION_BLOCK_SIZE / 8;
-        self.bit_widths.advance(num_blocks as usize);
+        self.bit_widths.advance(num_blocks);
         self.positions.advance(num_bytes_to_skip);
         self.anchor_offset += (num_blocks * COMPRESSION_BLOCK_SIZE) as u64;
     }

src/postings/compression/mod.rs

@@ -291,7 +291,7 @@ pub mod tests {
         const PADDING_VALUE: u32 = 234_234_345u32;
         let expected_length = 154;
         let mut encoder = BlockEncoder::new();
-        let input: Vec<u32> = (0u32..123u32).map(|i| 4 + i * 7 / 2).into_iter().collect();
+        let input: Vec<u32> = (0u32..123u32).map(|i| 4 + i * 7 / 2).collect();
         for offset in &[0u32, 1u32, 2u32] {
             let encoded_data = encoder.compress_vint_sorted(&input, *offset);
             assert!(encoded_data.len() <= expected_length);

src/postings/indexing_context.rs

@@ -1,11 +1,11 @@
-use crate::postings::stacker::{MemoryArena, TermHashMap};
+use stacker::{ArenaHashMap, MemoryArena};
 
 /// IndexingContext contains all of the transient memory arenas
 /// required for building the inverted index.
 pub(crate) struct IndexingContext {
     /// The term index is an adhoc hashmap,
     /// itself backed by a dedicated memory arena.
-    pub term_index: TermHashMap,
+    pub term_index: ArenaHashMap,
     /// Arena is a memory arena that stores posting lists / term frequencies / positions.
     pub arena: MemoryArena,
 }
@@ -13,9 +13,9 @@ pub(crate) struct IndexingContext {
 impl IndexingContext {
     /// Create a new IndexingContext given the size of the term hash map.
     pub(crate) fn new(table_size: usize) -> IndexingContext {
-        let term_index = TermHashMap::new(table_size);
+        let term_index = ArenaHashMap::new(table_size);
         IndexingContext {
-            arena: MemoryArena::new(),
+            arena: MemoryArena::default(),
             term_index,
         }
     }

src/postings/json_postings_writer.rs

@@ -1,10 +1,11 @@
 use std::io;
 
+use stacker::Addr;
+
 use crate::fastfield::MultiValuedFastFieldWriter;
 use crate::indexer::doc_id_mapping::DocIdMapping;
 use crate::postings::postings_writer::SpecializedPostingsWriter;
 use crate::postings::recorder::{BufferLender, DocIdRecorder, Recorder};
-use crate::postings::stacker::Addr;
 use crate::postings::{
     FieldSerializer, IndexingContext, IndexingPosition, PostingsWriter, UnorderedTermId,
 };

src/postings/mod.rs

@@ -15,9 +15,10 @@ mod recorder;
 mod segment_postings;
 mod serializer;
 mod skip;
-mod stacker;
 mod term_info;
 
+pub(crate) use stacker::compute_table_size;
+
 pub use self::block_segment_postings::BlockSegmentPostings;
 pub(crate) use self::indexing_context::IndexingContext;
 pub(crate) use self::per_field_postings_writer::PerFieldPostingsWriter;
@@ -26,10 +27,9 @@ pub(crate) use self::postings_writer::{serialize_postings, IndexingPosition, Pos
 pub use self::segment_postings::SegmentPostings;
 pub use self::serializer::{FieldSerializer, InvertedIndexSerializer};
 pub(crate) use self::skip::{BlockInfo, SkipReader};
-pub(crate) use self::stacker::compute_table_size;
 pub use self::term_info::TermInfo;
 
-pub(crate) type UnorderedTermId = u64;
+pub(crate) type UnorderedTermId = stacker::UnorderedId;
 
 #[allow(clippy::enum_variant_names)]
 #[derive(Debug, PartialEq, Clone, Copy, Eq)]
@@ -631,7 +631,7 @@ mod bench {
             let mut segment_postings = segment_reader
                 .inverted_index(TERM_A.field())
                 .unwrap()
-                .read_postings(&*TERM_A, IndexRecordOption::Basic)
+                .read_postings(&TERM_A, IndexRecordOption::Basic)
                 .unwrap()
                 .unwrap();
             while segment_postings.advance() != TERMINATED {}
@@ -647,25 +647,25 @@ mod bench {
             let segment_postings_a = segment_reader
                 .inverted_index(TERM_A.field())
                 .unwrap()
-                .read_postings(&*TERM_A, IndexRecordOption::Basic)
+                .read_postings(&TERM_A, IndexRecordOption::Basic)
                 .unwrap()
                 .unwrap();
             let segment_postings_b = segment_reader
                 .inverted_index(TERM_B.field())
                 .unwrap()
-                .read_postings(&*TERM_B, IndexRecordOption::Basic)
+                .read_postings(&TERM_B, IndexRecordOption::Basic)
                 .unwrap()
                 .unwrap();
             let segment_postings_c = segment_reader
                 .inverted_index(TERM_C.field())
                 .unwrap()
-                .read_postings(&*TERM_C, IndexRecordOption::Basic)
+                .read_postings(&TERM_C, IndexRecordOption::Basic)
                 .unwrap()
                 .unwrap();
             let segment_postings_d = segment_reader
                 .inverted_index(TERM_D.field())
                 .unwrap()
-                .read_postings(&*TERM_D, IndexRecordOption::Basic)
+                .read_postings(&TERM_D, IndexRecordOption::Basic)
                 .unwrap()
                 .unwrap();
             let mut intersection = Intersection::new(vec![
@@ -687,7 +687,7 @@ mod bench {
         let mut segment_postings = segment_reader
             .inverted_index(TERM_A.field())
             .unwrap()
-            .read_postings(&*TERM_A, IndexRecordOption::Basic)
+            .read_postings(&TERM_A, IndexRecordOption::Basic)
             .unwrap()
             .unwrap();
 
@@ -705,7 +705,7 @@ mod bench {
             let mut segment_postings = segment_reader
                 .inverted_index(TERM_A.field())
                 .unwrap()
-                .read_postings(&*TERM_A, IndexRecordOption::Basic)
+                .read_postings(&TERM_A, IndexRecordOption::Basic)
                 .unwrap()
                 .unwrap();
             for doc in &existing_docs {
@@ -746,7 +746,7 @@ mod bench {
             let mut segment_postings = segment_reader
                 .inverted_index(TERM_A.field())
                 .unwrap()
-                .read_postings(&*TERM_A, IndexRecordOption::Basic)
+                .read_postings(&TERM_A, IndexRecordOption::Basic)
                 .unwrap()
                 .unwrap();
             let mut s = 0u32;

src/postings/per_field_postings_writer.rs

@@ -51,7 +51,7 @@ fn posting_writer_from_field_entry(field_entry: &FieldEntry) -> Box<dyn Postings
         | FieldType::Date(_)
         | FieldType::Bytes(_)
         | FieldType::IpAddr(_)
-        | FieldType::Facet(_) => Box::new(SpecializedPostingsWriter::<DocIdRecorder>::default()),
+        | FieldType::Facet(_) => Box::<SpecializedPostingsWriter<DocIdRecorder>>::default(),
         FieldType::JsonObject(ref json_object_options) => {
             if let Some(text_indexing_option) = json_object_options.get_text_indexing_options() {
                 match text_indexing_option.index_option() {

src/postings/postings_writer.rs

@@ -4,8 +4,8 @@ use std::marker::PhantomData;
 use std::ops::Range;
 
 use rustc_hash::FxHashMap;
+use stacker::Addr;
 
-use super::stacker::Addr;
 use crate::fastfield::MultiValuedFastFieldWriter;
 use crate::fieldnorm::FieldNormReaders;
 use crate::indexer::doc_id_mapping::DocIdMapping;
@@ -59,7 +59,11 @@ pub(crate) fn serialize_postings(
 ) -> crate::Result<HashMap<Field, FxHashMap<UnorderedTermId, TermOrdinal>>> {
     let mut term_offsets: Vec<(Term<&[u8]>, Addr, UnorderedTermId)> =
         Vec::with_capacity(ctx.term_index.len());
-    term_offsets.extend(ctx.term_index.iter());
+    term_offsets.extend(
+        ctx.term_index
+            .iter()
+            .map(|(bytes, addr, unordered_id)| (Term::wrap(bytes), addr, unordered_id)),
+    );
     term_offsets.sort_unstable_by_key(|(k, _, _)| k.clone());
     let mut unordered_term_mappings: HashMap<Field, FxHashMap<UnorderedTermId, TermOrdinal>> =
         HashMap::new();

src/postings/recorder.rs

@@ -1,6 +1,6 @@
 use common::read_u32_vint;
+use stacker::{ExpUnrolledLinkedList, MemoryArena};
 
-use super::stacker::{ExpUnrolledLinkedList, MemoryArena};
 use crate::indexer::doc_id_mapping::DocIdMapping;
 use crate::postings::FieldSerializer;
 use crate::DocId;
@@ -91,7 +91,7 @@ pub struct DocIdRecorder {
 impl Default for DocIdRecorder {
     fn default() -> Self {
         DocIdRecorder {
-            stack: ExpUnrolledLinkedList::new(),
+            stack: ExpUnrolledLinkedList::default(),
             current_doc: u32::MAX,
         }
     }
@@ -144,7 +144,7 @@ impl Recorder for DocIdRecorder {
 }
 
 /// Recorder encoding document ids, and term frequencies
-#[derive(Clone, Copy)]
+#[derive(Clone, Copy, Default)]
 pub struct TermFrequencyRecorder {
     stack: ExpUnrolledLinkedList,
     current_doc: DocId,
@@ -152,17 +152,6 @@ pub struct TermFrequencyRecorder {
     term_doc_freq: u32,
 }
 
-impl Default for TermFrequencyRecorder {
-    fn default() -> Self {
-        TermFrequencyRecorder {
-            stack: ExpUnrolledLinkedList::new(),
-            current_doc: 0,
-            current_tf: 0u32,
-            term_doc_freq: 0u32,
-        }
-    }
-}
-
 impl Recorder for TermFrequencyRecorder {
     fn current_doc(&self) -> DocId {
         self.current_doc
@@ -229,7 +218,7 @@ pub struct TfAndPositionRecorder {
 impl Default for TfAndPositionRecorder {
     fn default() -> Self {
         TfAndPositionRecorder {
-            stack: ExpUnrolledLinkedList::new(),
+            stack: ExpUnrolledLinkedList::default(),
             current_doc: u32::MAX,
             term_doc_freq: 0u32,
         }

src/postings/serializer.rs

@@ -465,7 +465,7 @@ impl<W: Write> PostingsSerializer<W> {
     /// When called after writing the postings of a term, this value is used as a
     /// end offset.
     fn written_bytes(&self) -> u64 {
-        self.output_write.written_bytes() as u64
+        self.output_write.written_bytes()
     }
 
     fn clear(&mut self) {

src/postings/stacker/mod.rs

@@ -1,7 +0,0 @@
-mod expull;
-mod memory_arena;
-mod term_hashmap;
-
-pub(crate) use self::expull::ExpUnrolledLinkedList;
-pub(crate) use self::memory_arena::{Addr, MemoryArena};
-pub(crate) use self::term_hashmap::{compute_table_size, TermHashMap};

src/query/all_query.rs

@@ -95,7 +95,7 @@ mod tests {
         let index = create_test_index()?;
         let reader = index.reader()?;
         let searcher = reader.searcher();
-        let weight = AllQuery.weight(EnableScoring::Disabled(&index.schema()))?;
+        let weight = AllQuery.weight(EnableScoring::disabled_from_schema(&index.schema()))?;
         {
             let reader = searcher.segment_reader(0);
             let mut scorer = weight.scorer(reader, 1.0)?;
@@ -118,7 +118,7 @@ mod tests {
         let index = create_test_index()?;
         let reader = index.reader()?;
         let searcher = reader.searcher();
-        let weight = AllQuery.weight(EnableScoring::Disabled(searcher.schema()))?;
+        let weight = AllQuery.weight(EnableScoring::disabled_from_schema(searcher.schema()))?;
         let reader = searcher.segment_reader(0);
         {
             let mut scorer = weight.scorer(reader, 2.0)?;

src/query/bitset/mod.rs

@@ -47,7 +47,7 @@ impl From<BitSet> for BitSetDocSet {
 impl DocSet for BitSetDocSet {
     fn advance(&mut self) -> DocId {
         if let Some(lower) = self.cursor_tinybitset.pop_lowest() {
-            self.doc = (self.cursor_bucket as u32 * 64u32) | lower;
+            self.doc = (self.cursor_bucket * 64u32) | lower;
             return self.doc;
         }
         if let Some(cursor_bucket) = self.docs.first_non_empty_bucket(self.cursor_bucket + 1) {

src/query/boolean_query/boolean_query.rs

@@ -146,7 +146,7 @@ impl Query for BooleanQuery {
         let sub_weights = self
             .subqueries
             .iter()
-            .map(|&(ref occur, ref subquery)| Ok((*occur, subquery.weight(enable_scoring)?)))
+            .map(|(occur, subquery)| Ok((*occur, subquery.weight(enable_scoring)?)))
             .collect::<crate::Result<_>>()?;
         Ok(Box::new(BooleanWeight::new(
             sub_weights,

src/query/boolean_query/boolean_weight.rs

@@ -91,7 +91,7 @@ impl<TScoreCombiner: ScoreCombiner> BooleanWeight<TScoreCombiner> {
         boost: Score,
     ) -> crate::Result<HashMap<Occur, Vec<Box<dyn Scorer>>>> {
         let mut per_occur_scorers: HashMap<Occur, Vec<Box<dyn Scorer>>> = HashMap::new();
-        for &(ref occur, ref subweight) in &self.weights {
+        for (occur, subweight) in &self.weights {
             let sub_scorer: Box<dyn Scorer> = subweight.scorer(reader, boost)?;
             per_occur_scorers
                 .entry(*occur)
@@ -191,7 +191,7 @@ impl<TScoreCombiner: ScoreCombiner + Sync> Weight for BooleanWeight<TScoreCombin
         }
 
         let mut explanation = Explanation::new("BooleanClause. Sum of ...", scorer.score());
-        for &(ref occur, ref subweight) in &self.weights {
+        for (occur, subweight) in &self.weights {
             if is_positive_occur(*occur) {
                 if let Ok(child_explanation) = subweight.explain(reader, doc) {
                     explanation.add_detail(child_explanation);

src/query/boolean_query/mod.rs

@@ -98,7 +98,7 @@ mod tests {
         }
         {
             let query = query_parser.parse_query("+a b")?;
-            let weight = query.weight(EnableScoring::Disabled(searcher.schema()))?;
+            let weight = query.weight(EnableScoring::disabled_from_schema(searcher.schema()))?;
             let scorer = weight.scorer(searcher.segment_reader(0u32), 1.0)?;
             assert!(scorer.is::<TermScorer>());
         }

src/query/disjunction_max_query.rs

@@ -3,7 +3,7 @@ use tantivy_query_grammar::Occur;
 use crate::query::{BooleanWeight, DisjunctionMaxCombiner, EnableScoring, Query, Weight};
 use crate::{Score, Term};
 
-/// The disjunction max query кeturns documents matching one or more wrapped queries,
+/// The disjunction max query returns documents matching one or more wrapped queries,
 /// called query clauses or clauses.
 ///
 /// If a returned document matches multiple query clauses,

src/query/fuzzy_query.rs

@@ -1,8 +1,5 @@
-use std::collections::HashMap;
-use std::ops::Range;
-
 use levenshtein_automata::{Distance, LevenshteinAutomatonBuilder, DFA};
-use once_cell::sync::Lazy;
+use once_cell::sync::OnceCell;
 use tantivy_fst::Automaton;
 
 use crate::query::{AutomatonWeight, EnableScoring, Query, Weight};
@@ -34,22 +31,6 @@ impl Automaton for DfaWrapper {
     }
 }
 
-/// A range of Levenshtein distances that we will build DFAs for our terms
-/// The computation is exponential, so best keep it to low single digits
-const VALID_LEVENSHTEIN_DISTANCE_RANGE: Range<u8> = 0..3;
-
-static LEV_BUILDER: Lazy<HashMap<(u8, bool), LevenshteinAutomatonBuilder>> = Lazy::new(|| {
-    let mut lev_builder_cache = HashMap::new();
-    // TODO make population lazy on a `(distance, val)` basis
-    for distance in VALID_LEVENSHTEIN_DISTANCE_RANGE {
-        for &transposition in &[false, true] {
-            let lev_automaton_builder = LevenshteinAutomatonBuilder::new(distance, transposition);
-            lev_builder_cache.insert((distance, transposition), lev_automaton_builder);
-        }
-    }
-    lev_builder_cache
-});
-
 /// A Fuzzy Query matches all of the documents
 /// containing a specific term that is within
 /// Levenshtein distance
@@ -129,30 +110,39 @@ impl FuzzyTermQuery {
     }
 
     fn specialized_weight(&self) -> crate::Result<AutomatonWeight<DfaWrapper>> {
-        // LEV_BUILDER is a HashMap, whose `get` method returns an Option
-        match LEV_BUILDER.get(&(self.distance, self.transposition_cost_one)) {
-            // Unwrap the option and build the Ok(AutomatonWeight)
-            Some(automaton_builder) => {
-                let term_text = self.term.as_str().ok_or_else(|| {
-                    crate::TantivyError::InvalidArgument(
-                        "The fuzzy term query requires a string term.".to_string(),
-                    )
-                })?;
-                let automaton = if self.prefix {
-                    automaton_builder.build_prefix_dfa(term_text)
-                } else {
-                    automaton_builder.build_dfa(term_text)
-                };
-                Ok(AutomatonWeight::new(
-                    self.term.field(),
-                    DfaWrapper(automaton),
+        static AUTOMATON_BUILDER: [[OnceCell<LevenshteinAutomatonBuilder>; 2]; 3] = [
+            [OnceCell::new(), OnceCell::new()],
+            [OnceCell::new(), OnceCell::new()],
+            [OnceCell::new(), OnceCell::new()],
+        ];
+
+        let automaton_builder = AUTOMATON_BUILDER
+            .get(self.distance as usize)
+            .ok_or_else(|| {
+                InvalidArgument(format!(
+                    "Levenshtein distance of {} is not allowed. Choose a value less than {}",
+                    self.distance,
+                    AUTOMATON_BUILDER.len()
                 ))
-            }
-            None => Err(InvalidArgument(format!(
-                "Levenshtein distance of {} is not allowed. Choose a value in the {:?} range",
-                self.distance, VALID_LEVENSHTEIN_DISTANCE_RANGE
-            ))),
-        }
+            })?
+            .get(self.transposition_cost_one as usize)
+            .unwrap()
+            .get_or_init(|| {
+                LevenshteinAutomatonBuilder::new(self.distance, self.transposition_cost_one)
+            });
+
+        let term_text = self.term.as_str().ok_or_else(|| {
+            InvalidArgument("The fuzzy term query requires a string term.".to_string())
+        })?;
+        let automaton = if self.prefix {
+            automaton_builder.build_prefix_dfa(term_text)
+        } else {
+            automaton_builder.build_dfa(term_text)
+        };
+        Ok(AutomatonWeight::new(
+            self.term.field(),
+            DfaWrapper(automaton),
+        ))
     }
 }
 

src/query/mod.rs

@@ -16,7 +16,6 @@ mod phrase_query;
 mod query;
 mod query_parser;
 mod range_query;
-mod range_query_ip_fastfield;
 mod regex_query;
 mod reqopt_scorer;
 mod scorer;