remove dynamic dispatch in collect_segment

* add date_histogram

* add return result

.github/workflows/coverage.yml

@@ -2,9 +2,9 @@ name: Coverage
 
 on:
   push:
-    branches: [ main ]
+    branches: [main]
   pull_request:
-    branches: [ main ]
+    branches: [main]
 
 jobs:
   coverage:
@@ -16,7 +16,7 @@ jobs:
       - uses: Swatinem/rust-cache@v2
       - uses: taiki-e/install-action@cargo-llvm-cov
       - name: Generate code coverage
-        run: cargo +nightly llvm-cov --all-features --workspace --lcov --output-path lcov.info
+        run: cargo +nightly llvm-cov --all-features --workspace --doctests --lcov --output-path lcov.info
       - name: Upload coverage to Codecov
         uses: codecov/codecov-action@v3
         continue-on-error: true

Cargo.toml

@@ -23,7 +23,7 @@ regex = { version = "1.5.5", default-features = false, features = ["std", "unico
 aho-corasick = "0.7"
 tantivy-fst = "0.4.0"
 memmap2 = { version = "0.5.3", optional = true }
-lz4_flex = { version = "0.9.2", default-features = false, features = ["checked-decode"], optional = true }
+lz4_flex = { version = "0.10", default-features = false, features = ["checked-decode"], optional = true }
 brotli = { version = "3.3.4", optional = true }
 zstd = { version = "0.12", optional = true, default-features = false }
 snap = { version = "1.0.5", optional = true }
@@ -58,7 +58,7 @@ arc-swap = "1.5.0"
 columnar = { version="0.1", path="./columnar", package ="tantivy-columnar" }
 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" }
+query-grammar = { version= "0.19.0", path="./query-grammar", package = "tantivy-query-grammar" }
 tantivy-bitpacker = { version= "0.3", path="./bitpacker" }
 common = { version= "0.5", path = "./common/", package = "tantivy-common" }
 tokenizer-api = { version="0.1", path="./tokenizer-api", package="tantivy-tokenizer-api" }
@@ -77,6 +77,7 @@ test-log = "0.2.10"
 env_logger = "0.10.0"
 pprof = { version = "0.11.0", features = ["flamegraph", "criterion"] }
 futures = "0.3.21"
+paste = "1.0.11"
 
 [dev-dependencies.fail]
 version = "0.5.0"

ci/before_deploy.ps1

@@ -1,23 +0,0 @@
-# This script takes care of packaging the build artifacts that will go in the
-# release zipfile
-
-$SRC_DIR = $PWD.Path
-$STAGE = [System.Guid]::NewGuid().ToString()
-
-Set-Location $ENV:Temp
-New-Item -Type Directory -Name $STAGE
-Set-Location $STAGE
-
-$ZIP = "$SRC_DIR\$($Env:CRATE_NAME)-$($Env:APPVEYOR_REPO_TAG_NAME)-$($Env:TARGET).zip"
-
-# TODO Update this to package the right artifacts
-Copy-Item "$SRC_DIR\target\$($Env:TARGET)\release\hello.exe" '.\'
-
-7z a "$ZIP" *
-
-Push-AppveyorArtifact "$ZIP"
-
-Remove-Item *.* -Force
-Set-Location ..
-Remove-Item $STAGE
-Set-Location $SRC_DIR

ci/before_deploy.sh

@@ -1,33 +0,0 @@
-# This script takes care of building your crate and packaging it for release
-
-set -ex
-
-main() {
-    local src=$(pwd) \
-          stage=
-
-    case $TRAVIS_OS_NAME in
-        linux)
-            stage=$(mktemp -d)
-            ;;
-        osx)
-            stage=$(mktemp -d -t tmp)
-            ;;
-    esac
-
-    test -f Cargo.lock || cargo generate-lockfile
-
-    # TODO Update this to build the artifacts that matter to you
-    cross rustc --bin hello --target $TARGET --release -- -C lto
-
-    # TODO Update this to package the right artifacts
-    cp target/$TARGET/release/hello $stage/
-
-    cd $stage
-    tar czf $src/$CRATE_NAME-$TRAVIS_TAG-$TARGET.tar.gz *
-    cd $src
-
-    rm -rf $stage
-}
-
-main

ci/install.sh

@@ -1,47 +0,0 @@
-set -ex
-
-main() {
-    local target=
-    if [ $TRAVIS_OS_NAME = linux ]; then
-        target=x86_64-unknown-linux-musl
-        sort=sort
-    else
-        target=x86_64-apple-darwin
-        sort=gsort  # for `sort --sort-version`, from brew's coreutils.
-    fi
-
-    # Builds for iOS are done on OSX, but require the specific target to be
-    # installed.
-    case $TARGET in
-        aarch64-apple-ios)
-            rustup target install aarch64-apple-ios
-            ;;
-        armv7-apple-ios)
-            rustup target install armv7-apple-ios
-            ;;
-        armv7s-apple-ios)
-            rustup target install armv7s-apple-ios
-            ;;
-        i386-apple-ios)
-            rustup target install i386-apple-ios
-            ;;
-        x86_64-apple-ios)
-            rustup target install x86_64-apple-ios
-            ;;
-    esac
-
-    # This fetches latest stable release
-    local tag=$(git ls-remote --tags --refs --exit-code https://github.com/japaric/cross \
-                       | cut -d/ -f3 \
-                       | grep -E '^v[0.1.0-9.]+$' \
-                       | $sort --version-sort \
-                       | tail -n1)
-    curl -LSfs https://japaric.github.io/trust/install.sh | \
-        sh -s -- \
-           --force \
-           --git japaric/cross \
-           --tag $tag \
-           --target $target
-}
-
-main

ci/script.sh

@@ -1,30 +0,0 @@
-#!/usr/bin/env bash
-
-# This script takes care of testing your crate
-
-set -ex
-
-main() {
-    if [ ! -z $CODECOV ]; then
-        echo "Codecov"
-        cargo build --verbose && cargo coverage --verbose --all && bash <(curl -s https://codecov.io/bash) -s target/kcov
-    else
-        echo "Build"
-        cross build --target $TARGET
-        if [ ! -z $DISABLE_TESTS ]; then
-            return
-        fi
-        echo "Test"
-        cross test --target $TARGET --no-default-features --features mmap
-        cross test --target $TARGET --no-default-features --features mmap query-grammar
-    fi
-    for example in $(ls examples/*.rs)
-    do
-        cargo run --example  $(basename $example .rs)
-    done
-}
-
-# we don't run the "test phase" when doing deploys
-if [ -z $TRAVIS_TAG ]; then
-    main
-fi

columnar/Cargo.toml

@@ -17,6 +17,7 @@ stacker = { path = "../stacker", package="tantivy-stacker"}
 sstable = { path = "../sstable", package = "tantivy-sstable" }
 common = { path = "../common", package = "tantivy-common" }
 tantivy-bitpacker = { version= "0.3", path = "../bitpacker/" }
+serde = "1.0.152"
 
 [dev-dependencies]
 proptest = "1"

columnar/benches/bench_u128.rs

@@ -58,7 +58,7 @@ fn bench_intfastfield_getrange_u128_50percent_hit(b: &mut Bencher) {
 
     b.iter(|| {
         let mut positions = Vec::new();
-        column.get_docids_for_value_range(
+        column.get_row_ids_for_value_range(
             *FIFTY_PERCENT_RANGE.start() as u128..=*FIFTY_PERCENT_RANGE.end() as u128,
             0..data.len() as u32,
             &mut positions,
@@ -74,7 +74,7 @@ fn bench_intfastfield_getrange_u128_single_hit(b: &mut Bencher) {
 
     b.iter(|| {
         let mut positions = Vec::new();
-        column.get_docids_for_value_range(
+        column.get_row_ids_for_value_range(
             *SINGLE_ITEM_RANGE.start() as u128..=*SINGLE_ITEM_RANGE.end() as u128,
             0..data.len() as u32,
             &mut positions,
@@ -90,7 +90,7 @@ fn bench_intfastfield_getrange_u128_hit_all(b: &mut Bencher) {
 
     b.iter(|| {
         let mut positions = Vec::new();
-        column.get_docids_for_value_range(0..=u128::MAX, 0..data.len() as u32, &mut positions);
+        column.get_row_ids_for_value_range(0..=u128::MAX, 0..data.len() as u32, &mut positions);
         positions
     });
 }

columnar/benches/bench_u64.rs

@@ -89,7 +89,7 @@ fn bench_intfastfield_getrange_u64_50percent_hit(b: &mut Bencher) {
     let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&data, CodecType::Bitpacked);
     b.iter(|| {
         let mut positions = Vec::new();
-        column.get_docids_for_value_range(
+        column.get_row_ids_for_value_range(
             FIFTY_PERCENT_RANGE,
             0..data.len() as u32,
             &mut positions,
@@ -106,7 +106,7 @@ fn bench_intfastfield_getrange_u64_1percent_hit(b: &mut Bencher) {
 
     b.iter(|| {
         let mut positions = Vec::new();
-        column.get_docids_for_value_range(
+        column.get_row_ids_for_value_range(
             ONE_PERCENT_ITEM_RANGE,
             0..data.len() as u32,
             &mut positions,
@@ -123,7 +123,7 @@ fn bench_intfastfield_getrange_u64_single_hit(b: &mut Bencher) {
 
     b.iter(|| {
         let mut positions = Vec::new();
-        column.get_docids_for_value_range(SINGLE_ITEM_RANGE, 0..data.len() as u32, &mut positions);
+        column.get_row_ids_for_value_range(SINGLE_ITEM_RANGE, 0..data.len() as u32, &mut positions);
         positions
     });
 }
@@ -136,7 +136,7 @@ fn bench_intfastfield_getrange_u64_hit_all(b: &mut Bencher) {
 
     b.iter(|| {
         let mut positions = Vec::new();
-        column.get_docids_for_value_range(0..=u64::MAX, 0..data.len() as u32, &mut positions);
+        column.get_row_ids_for_value_range(0..=u64::MAX, 0..data.len() as u32, &mut positions);
         positions
     });
 }

columnar/columnar-cli/Cargo.toml

@@ -0,0 +1,17 @@
+[package]
+name = "tantivy-columnar-cli"
+version = "0.1.0"
+edition = "2021"
+license = "MIT"
+
+[dependencies]
+columnar = {path="../", package="tantivy-columnar"}
+serde_json = "1"
+serde_json_borrow = {git="https://github.com/PSeitz/serde_json_borrow/"}
+serde = "1"
+
+[workspace]
+members = []
+
+[profile.release]
+debug = true

columnar/columnar-cli/src/main.rs

@@ -0,0 +1,134 @@
+use columnar::ColumnarWriter;
+use columnar::NumericalValue;
+use serde_json_borrow;
+use std::fs::File;
+use std::io;
+use std::io::BufRead;
+use std::io::BufReader;
+use std::time::Instant;
+
+#[derive(Default)]
+struct JsonStack {
+    path: String,
+    stack: Vec<usize>,
+}
+
+impl JsonStack {
+    fn push(&mut self, seg: &str) {
+        let len = self.path.len();
+        self.stack.push(len);
+        self.path.push('.');
+        self.path.push_str(seg);
+    }
+
+    fn pop(&mut self) {
+        if let Some(len) = self.stack.pop() {
+            self.path.truncate(len);
+        }
+    }
+
+    fn path(&self) -> &str {
+        &self.path[1..]
+    }
+}
+
+fn append_json_to_columnar(
+    doc: u32,
+    json_value: &serde_json_borrow::Value,
+    columnar: &mut ColumnarWriter,
+    stack: &mut JsonStack,
+) -> usize {
+    let mut count = 0;
+    match json_value {
+        serde_json_borrow::Value::Null => {}
+        serde_json_borrow::Value::Bool(val) => {
+            columnar.record_numerical(
+                doc,
+                stack.path(),
+                NumericalValue::from(if *val { 1u64 } else { 0u64 }),
+            );
+            count += 1;
+        }
+        serde_json_borrow::Value::Number(num) => {
+            let numerical_value: NumericalValue = if let Some(num_i64) = num.as_i64() {
+                num_i64.into()
+            } else if let Some(num_u64) = num.as_u64() {
+                num_u64.into()
+            } else if let Some(num_f64) = num.as_f64() {
+                num_f64.into()
+            } else {
+                panic!();
+            };
+            count += 1;
+            columnar.record_numerical(
+                doc,
+                stack.path(),
+                numerical_value,
+            );
+        }
+        serde_json_borrow::Value::Str(msg) => {
+            columnar.record_str(
+                doc,
+                stack.path(),
+                msg,
+            );
+            count += 1;
+        },
+        serde_json_borrow::Value::Array(vals) => {
+            for val in vals {
+                count += append_json_to_columnar(doc, val, columnar, stack);
+            }
+        },
+        serde_json_borrow::Value::Object(json_map) => {
+            for (child_key, child_val) in json_map {
+                stack.push(child_key);
+                count += append_json_to_columnar(doc, child_val, columnar, stack);
+                stack.pop();
+            }
+        },
+    }
+    count
+}
+
+fn main() -> io::Result<()> {
+    let file = File::open("gh_small.json")?;
+    let mut reader = BufReader::new(file);
+    let mut line = String::with_capacity(100);
+    let mut columnar = columnar::ColumnarWriter::default();
+    let mut doc = 0;
+    let start = Instant::now();
+    let mut stack = JsonStack::default();
+    let mut total_count = 0;
+
+    let start_build = Instant::now();
+    loop {
+        line.clear();
+        let len = reader.read_line(&mut line)?;
+        if len == 0 {
+            break;
+        }
+        let Ok(json_value) = serde_json::from_str::<serde_json_borrow::Value>(&line) else { continue; };
+        total_count += append_json_to_columnar(doc, &json_value, &mut columnar, &mut stack);
+        doc += 1;
+    }
+    println!("Build in {:?}", start_build.elapsed());
+
+    println!("value count {total_count}");
+
+    let mut buffer = Vec::new();
+    let start_serialize = Instant::now();
+    columnar.serialize(doc, None, &mut buffer)?;
+    println!("Serialized in {:?}", start_serialize.elapsed());
+    println!("num docs: {doc}, {:?}", start.elapsed());
+    println!("buffer len {} MB", buffer.len() / 1_000_000);
+    let columnar = columnar::ColumnarReader::open(buffer)?;
+    for (column_name, dynamic_column) in columnar.list_columns()? {
+        let num_bytes = dynamic_column.num_bytes();
+        let typ = dynamic_column.column_type();
+        if num_bytes > 1_000_000 {
+            println!("{column_name} {typ:?}  {} KB", num_bytes / 1_000);
+        }
+    }
+    println!("{} columns", columnar.num_columns());
+    Ok(())
+}

columnar/src/TODO.md

@@ -1,7 +1,6 @@
 # zero to one
 
 * revisit line codec
-* removal of all rows of a column in the schema due to deletes
 * add columns from schema on merge
 * Plugging JSON
 * replug examples

columnar/src/column/dictionary_encoded.rs

@@ -32,11 +32,11 @@ impl BytesColumn {
 
     /// Returns the number of rows in the column.
     pub fn num_rows(&self) -> RowId {
-        self.term_ord_column.num_rows()
+        self.term_ord_column.num_docs()
     }
 
     pub fn term_ords(&self, row_id: RowId) -> impl Iterator<Item = u64> + '_ {
-        self.term_ord_column.values(row_id)
+        self.term_ord_column.values_for_doc(row_id)
     }
 
     /// Returns the column of ordinals

columnar/src/column/mod.rs

@@ -3,7 +3,7 @@ mod serialize;
 
 use std::fmt::Debug;
 use std::io::Write;
-use std::ops::Deref;
+use std::ops::{Deref, Range, RangeInclusive};
 use std::sync::Arc;
 
 use common::BinarySerializable;
@@ -38,18 +38,20 @@ impl<T: MonotonicallyMappableToU64> Column<T> {
 }
 
 impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
+    #[inline]
     pub fn get_cardinality(&self) -> Cardinality {
         self.idx.get_cardinality()
     }
 
-    pub fn num_rows(&self) -> RowId {
+    pub fn num_docs(&self) -> RowId {
         match &self.idx {
-            ColumnIndex::Full => self.values.num_vals() as u32,
-            ColumnIndex::Optional(optional_index) => optional_index.num_rows(),
+            ColumnIndex::Empty { num_docs } => *num_docs,
+            ColumnIndex::Full => self.values.num_vals(),
+            ColumnIndex::Optional(optional_index) => optional_index.num_docs(),
             ColumnIndex::Multivalued(col_index) => {
                 // The multivalued index contains all value start row_id,
                 // and one extra value at the end with the overall number of rows.
-                col_index.num_rows()
+                col_index.num_docs()
             }
         }
     }
@@ -63,21 +65,40 @@ impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
     }
 
     pub fn first(&self, row_id: RowId) -> Option<T> {
-        self.values(row_id).next()
+        self.values_for_doc(row_id).next()
     }
 
-    pub fn values(&self, row_id: RowId) -> impl Iterator<Item = T> + '_ {
+    pub fn values_for_doc(&self, row_id: RowId) -> impl Iterator<Item = T> + '_ {
         self.value_row_ids(row_id)
             .map(|value_row_id: RowId| self.values.get_val(value_row_id))
     }
 
+    /// Get the docids of values which are in the provided value range.
+    #[inline]
+    pub fn get_docids_for_value_range(
+        &self,
+        value_range: RangeInclusive<T>,
+        selected_docid_range: Range<u32>,
+        doc_ids: &mut Vec<u32>,
+    ) {
+        // convert passed docid range to row id range
+        let rowid_range = self.idx.docid_range_to_rowids(selected_docid_range.clone());
+
+        // Load rows
+        self.values
+            .get_row_ids_for_value_range(value_range, rowid_range, doc_ids);
+        // Convert rows to docids
+        self.idx
+            .select_batch_in_place(selected_docid_range.start, doc_ids);
+    }
+
     /// Fils the output vector with the (possibly multiple values that are associated_with
     /// `row_id`.
     ///
     /// This method clears the `output` vector.
     pub fn fill_vals(&self, row_id: RowId, output: &mut Vec<T>) {
         output.clear();
-        output.extend(self.values(row_id));
+        output.extend(self.values_for_doc(row_id));
     }
 
     pub fn first_or_default_col(self, default_value: T) -> Arc<dyn ColumnValues<T>> {
@@ -131,9 +152,10 @@ impl<T: PartialOrd + Debug + Send + Sync + Copy + 'static> ColumnValues<T>
 
     fn num_vals(&self) -> u32 {
         match &self.column.idx {
+            ColumnIndex::Empty { .. } => 0u32,
             ColumnIndex::Full => self.column.values.num_vals(),
-            ColumnIndex::Optional(optional_idx) => optional_idx.num_rows(),
-            ColumnIndex::Multivalued(multivalue_idx) => multivalue_idx.num_rows(),
+            ColumnIndex::Optional(optional_idx) => optional_idx.num_docs(),
+            ColumnIndex::Multivalued(multivalue_idx) => multivalue_idx.num_docs(),
         }
     }
 }

columnar/src/column/serialize.rs

@@ -7,9 +7,10 @@ use sstable::Dictionary;
 
 use crate::column::{BytesColumn, Column};
 use crate::column_index::{serialize_column_index, SerializableColumnIndex};
-use crate::column_values::serialize::serialize_column_values_u128;
-use crate::column_values::u64_based::{serialize_u64_based_column_values, CodecType};
-use crate::column_values::{MonotonicallyMappableToU128, MonotonicallyMappableToU64};
+use crate::column_values::{
+    load_u64_based_column_values, serialize_column_values_u128, serialize_u64_based_column_values,
+    CodecType, MonotonicallyMappableToU128, MonotonicallyMappableToU64,
+};
 use crate::iterable::Iterable;
 use crate::StrColumn;
 
@@ -49,8 +50,7 @@ pub fn open_column_u64<T: MonotonicallyMappableToU64>(bytes: OwnedBytes) -> io::
     );
     let (column_index_data, column_values_data) = body.split(column_index_num_bytes as usize);
     let column_index = crate::column_index::open_column_index(column_index_data)?;
-    let column_values =
-        crate::column_values::u64_based::load_u64_based_column_values(column_values_data)?;
+    let column_values = load_u64_based_column_values(column_values_data)?;
     Ok(Column {
         idx: column_index,
         values: column_values,

columnar/src/column_index/merge/shuffled.rs

@@ -91,13 +91,10 @@ fn iter_num_values<'a>(
                     return 0u32;
                 };
         match column_index {
+            ColumnIndex::Empty { .. } => 0u32,
             ColumnIndex::Full => 1,
             ColumnIndex::Optional(optional_index) => {
-                if optional_index.contains(row_addr.row_id) {
-                    1u32
-                } else {
-                    0u32
-                }
+                u32::from(optional_index.contains(row_addr.row_id))
             }
             ColumnIndex::Multivalued(multivalued_index) => {
                 multivalued_index.range(row_addr.row_id).len() as u32

columnar/src/column_index/merge/stacked.rs

@@ -55,7 +55,7 @@ impl<'a> Iterable<RowId> for StackedOptionalIndex<'a> {
                         Some(ColumnIndex::Multivalued(_)) => {
                             panic!("No multivalued index is allowed when stacking column index");
                         }
-                        None => Box::new(std::iter::empty()),
+                        None | Some(ColumnIndex::Empty { .. }) => Box::new(std::iter::empty()),
                     };
                     rows_it
                 }),
@@ -74,7 +74,9 @@ fn convert_column_opt_to_multivalued_index<'a>(
     num_rows: RowId,
 ) -> Box<dyn Iterator<Item = RowId> + 'a> {
     match column_index_opt {
-        None => Box::new(iter::repeat(0u32).take(num_rows as usize + 1)),
+        None | Some(ColumnIndex::Empty { .. }) => {
+            Box::new(iter::repeat(0u32).take(num_rows as usize + 1))
+        }
         Some(ColumnIndex::Full) => Box::new(0..num_rows + 1),
         Some(ColumnIndex::Optional(optional_index)) => {
             Box::new(

columnar/src/column_index/mod.rs

@@ -10,10 +10,13 @@ pub use optional_index::{OptionalIndex, Set};
 pub use serialize::{open_column_index, serialize_column_index, SerializableColumnIndex};
 
 use crate::column_index::multivalued_index::MultiValueIndex;
-use crate::{Cardinality, RowId};
+use crate::{Cardinality, DocId, RowId};
 
 #[derive(Clone)]
 pub enum ColumnIndex {
+    Empty {
+        num_docs: u32,
+    },
     Full,
     Optional(OptionalIndex),
     /// In addition, at index num_rows, an extra value is added
@@ -34,8 +37,10 @@ impl From<MultiValueIndex> for ColumnIndex {
 }
 
 impl ColumnIndex {
+    #[inline]
     pub fn get_cardinality(&self) -> Cardinality {
         match self {
+            ColumnIndex::Empty { .. } => Cardinality::Optional,
             ColumnIndex::Full => Cardinality::Full,
             ColumnIndex::Optional(_) => Cardinality::Optional,
             ColumnIndex::Multivalued(_) => Cardinality::Multivalued,
@@ -43,32 +48,58 @@ impl ColumnIndex {
     }
 
     /// Returns true if and only if there are at least one value associated to the row.
-    pub fn has_value(&self, row_id: RowId) -> bool {
+    pub fn has_value(&self, doc_id: DocId) -> bool {
         match self {
+            ColumnIndex::Empty { .. } => false,
             ColumnIndex::Full => true,
-            ColumnIndex::Optional(optional_index) => optional_index.contains(row_id),
+            ColumnIndex::Optional(optional_index) => optional_index.contains(doc_id),
             ColumnIndex::Multivalued(multivalued_index) => {
-                multivalued_index.range(row_id).len() > 0
+                !multivalued_index.range(doc_id).is_empty()
             }
         }
     }
 
-    pub fn value_row_ids(&self, row_id: RowId) -> Range<RowId> {
+    pub fn value_row_ids(&self, doc_id: DocId) -> Range<RowId> {
         match self {
-            ColumnIndex::Full => row_id..row_id + 1,
+            ColumnIndex::Empty { .. } => 0..0,
+            ColumnIndex::Full => doc_id..doc_id + 1,
             ColumnIndex::Optional(optional_index) => {
-                if let Some(val) = optional_index.rank_if_exists(row_id) {
+                if let Some(val) = optional_index.rank_if_exists(doc_id) {
                     val..val + 1
                 } else {
                     0..0
                 }
             }
-            ColumnIndex::Multivalued(multivalued_index) => multivalued_index.range(row_id),
+            ColumnIndex::Multivalued(multivalued_index) => multivalued_index.range(doc_id),
         }
     }
 
-    pub fn select_batch_in_place(&self, rank_ids: &mut Vec<RowId>) {
+    pub fn docid_range_to_rowids(&self, doc_id: Range<DocId>) -> Range<RowId> {
         match self {
+            ColumnIndex::Empty { .. } => 0..0,
+            ColumnIndex::Full => doc_id,
+            ColumnIndex::Optional(optional_index) => {
+                let row_start = optional_index.rank(doc_id.start);
+                let row_end = optional_index.rank(doc_id.end);
+                row_start..row_end
+            }
+            ColumnIndex::Multivalued(multivalued_index) => {
+                let end_docid = doc_id.end.min(multivalued_index.num_docs() - 1) + 1;
+                let start_docid = doc_id.start.min(end_docid);
+
+                let row_start = multivalued_index.start_index_column.get_val(start_docid);
+                let row_end = multivalued_index.start_index_column.get_val(end_docid);
+
+                row_start..row_end
+            }
+        }
+    }
+
+    pub fn select_batch_in_place(&self, doc_id_start: DocId, rank_ids: &mut Vec<RowId>) {
+        match self {
+            ColumnIndex::Empty { .. } => {
+                rank_ids.clear();
+            }
             ColumnIndex::Full => {
                 // No need to do anything:
                 // value_idx and row_idx are the same.
@@ -77,8 +108,7 @@ impl ColumnIndex {
                 optional_index.select_batch(&mut rank_ids[..]);
             }
             ColumnIndex::Multivalued(multivalued_index) => {
-                // TODO important: avoid using 0u32, and restart from the beginning all of the time.
-                multivalued_index.select_batch_in_place(0u32, rank_ids)
+                multivalued_index.select_batch_in_place(doc_id_start, rank_ids)
             }
         }
     }

columnar/src/column_index/multivalued_index.rs

@@ -5,16 +5,17 @@ use std::sync::Arc;
 
 use common::OwnedBytes;
 
-use crate::column_values::u64_based::CodecType;
-use crate::column_values::ColumnValues;
+use crate::column_values::{
+    load_u64_based_column_values, serialize_u64_based_column_values, CodecType, ColumnValues,
+};
 use crate::iterable::Iterable;
-use crate::RowId;
+use crate::{DocId, RowId};
 
 pub fn serialize_multivalued_index(
     multivalued_index: &dyn Iterable<RowId>,
     output: &mut impl Write,
 ) -> io::Result<()> {
-    crate::column_values::u64_based::serialize_u64_based_column_values(
+    serialize_u64_based_column_values(
         multivalued_index,
         &[CodecType::Bitpacked, CodecType::Linear],
         output,
@@ -23,8 +24,7 @@ pub fn serialize_multivalued_index(
 }
 
 pub fn open_multivalued_index(bytes: OwnedBytes) -> io::Result<MultiValueIndex> {
-    let start_index_column: Arc<dyn ColumnValues<RowId>> =
-        crate::column_values::u64_based::load_u64_based_column_values(bytes)?;
+    let start_index_column: Arc<dyn ColumnValues<RowId>> = load_u64_based_column_values(bytes)?;
     Ok(MultiValueIndex { start_index_column })
 }
 
@@ -52,20 +52,20 @@ impl MultiValueIndex {
     /// Returns `[start, end)`, such that the values associated with
     /// the given document are `start..end`.
     #[inline]
-    pub(crate) fn range(&self, row_id: RowId) -> Range<RowId> {
-        let start = self.start_index_column.get_val(row_id);
-        let end = self.start_index_column.get_val(row_id + 1);
+    pub(crate) fn range(&self, doc_id: DocId) -> Range<RowId> {
+        let start = self.start_index_column.get_val(doc_id);
+        let end = self.start_index_column.get_val(doc_id + 1);
         start..end
     }
 
     /// Returns the number of documents in the index.
     #[inline]
-    pub fn num_rows(&self) -> u32 {
+    pub fn num_docs(&self) -> u32 {
         self.start_index_column.num_vals() - 1
     }
 
     /// Converts a list of ranks (row ids of values) in a 1:n index to the corresponding list of
-    /// row_ids. Positions are converted inplace to docids.
+    /// docids. Positions are converted inplace to docids.
     ///
     /// Since there is no index for value pos -> docid, but docid -> value pos range, we scan the
     /// index.
@@ -76,20 +76,20 @@ 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 select_batch_in_place(&self, row_start: RowId, ranks: &mut Vec<u32>) {
+    pub(crate) fn select_batch_in_place(&self, docid_start: DocId, ranks: &mut Vec<u32>) {
         if ranks.is_empty() {
             return;
         }
-        let mut cur_doc = row_start;
+        let mut cur_doc = docid_start;
         let mut last_doc = None;
 
-        assert!(self.start_index_column.get_val(row_start) as u32 <= ranks[0]);
+        assert!(self.start_index_column.get_val(docid_start) <= ranks[0]);
 
         let mut write_doc_pos = 0;
         for i in 0..ranks.len() {
             let pos = ranks[i];
             loop {
-                let end = self.start_index_column.get_val(cur_doc + 1) as u32;
+                let end = self.start_index_column.get_val(cur_doc + 1);
                 if end > pos {
                     ranks[write_doc_pos] = cur_doc;
                     write_doc_pos += if last_doc == Some(cur_doc) { 0 } else { 1 };
@@ -127,7 +127,7 @@ mod tests {
         let offsets: Vec<RowId> = vec![0, 10, 12, 15, 22, 23]; // docid values are [0..10, 10..12, 12..15, etc.]
         let column: Arc<dyn ColumnValues<RowId>> = Arc::new(IterColumn::from(offsets.into_iter()));
         let index = MultiValueIndex::from(column);
-        assert_eq!(index.num_rows(), 5);
+        assert_eq!(index.num_docs(), 5);
         let positions = &[10u32, 11, 15, 20, 21, 22];
         assert_eq!(index_to_pos_helper(&index, 0..5, positions), vec![1, 3, 4]);
         assert_eq!(index_to_pos_helper(&index, 1..5, positions), vec![1, 3, 4]);

columnar/src/column_index/optional_index/mod.rs

@@ -11,7 +11,7 @@ use set_block::{
 };
 
 use crate::iterable::Iterable;
-use crate::{InvalidData, RowId};
+use crate::{DocId, InvalidData, RowId};
 
 /// The threshold for for number of elements after which we switch to dense block encoding.
 ///
@@ -177,11 +177,11 @@ impl Set<RowId> for OptionalIndex {
     }
 
     #[inline]
-    fn rank(&self, row_id: RowId) -> RowId {
+    fn rank(&self, doc_id: DocId) -> RowId {
         let RowAddr {
             block_id,
             in_block_row_id,
-        } = row_addr_from_row_id(row_id);
+        } = row_addr_from_row_id(doc_id);
         let block_meta = self.block_metas[block_id as usize];
         let block = self.block(block_meta);
         let block_offset_row_id = match block {
@@ -192,11 +192,11 @@ impl Set<RowId> for OptionalIndex {
     }
 
     #[inline]
-    fn rank_if_exists(&self, row_id: RowId) -> Option<RowId> {
+    fn rank_if_exists(&self, doc_id: DocId) -> Option<RowId> {
         let RowAddr {
             block_id,
             in_block_row_id,
-        } = row_addr_from_row_id(row_id);
+        } = row_addr_from_row_id(doc_id);
         let block_meta = self.block_metas[block_id as usize];
         let block = self.block(block_meta);
         let block_offset_row_id = match block {
@@ -220,7 +220,7 @@ impl Set<RowId> for OptionalIndex {
         block_doc_idx_start + in_block_rank as u32
     }
 
-    fn select_cursor<'b>(&'b self) -> OptionalIndexSelectCursor<'b> {
+    fn select_cursor(&self) -> OptionalIndexSelectCursor<'_> {
         OptionalIndexSelectCursor {
             current_block_cursor: BlockSelectCursor::Sparse(
                 SparseBlockCodec::open(b"").select_cursor(),
@@ -247,7 +247,7 @@ impl OptionalIndex {
         open_optional_index(bytes).unwrap()
     }
 
-    pub fn num_rows(&self) -> RowId {
+    pub fn num_docs(&self) -> RowId {
         self.num_rows
     }
 
@@ -255,7 +255,7 @@ impl OptionalIndex {
         self.num_non_null_rows
     }
 
-    pub fn iter_rows<'a>(&'a self) -> impl Iterator<Item = RowId> + 'a {
+    pub fn iter_rows(&self) -> impl Iterator<Item = RowId> + '_ {
         // TODO optimize
         let mut select_batch = self.select_cursor();
         (0..self.num_non_null_rows).map(move |rank| select_batch.select(rank))
@@ -268,7 +268,7 @@ impl OptionalIndex {
     }
 
     #[inline]
-    fn block<'a>(&'a self, block_meta: BlockMeta) -> Block<'a> {
+    fn block(&self, block_meta: BlockMeta) -> Block<'_> {
         let BlockMeta {
             start_byte_offset,
             block_variant,
@@ -351,7 +351,7 @@ fn serialize_optional_index_block(block_els: &[u16], out: &mut impl io::Write) -
     Ok(())
 }
 
-pub fn serialize_optional_index<'a, W: io::Write>(
+pub fn serialize_optional_index<W: io::Write>(
     non_null_rows: &dyn Iterable<RowId>,
     num_rows: RowId,
     output: &mut W,
@@ -427,7 +427,7 @@ impl SerializedBlockMeta {
     }
 
     #[inline]
-    fn to_bytes(&self) -> [u8; SERIALIZED_BLOCK_META_NUM_BYTES] {
+    fn to_bytes(self) -> [u8; SERIALIZED_BLOCK_META_NUM_BYTES] {
         assert!(self.num_non_null_rows > 0);
         let mut bytes = [0u8; SERIALIZED_BLOCK_META_NUM_BYTES];
         bytes[0..2].copy_from_slice(&self.block_id.to_le_bytes());
@@ -440,7 +440,7 @@ impl SerializedBlockMeta {
 
 #[inline]
 fn is_sparse(num_rows_in_block: u32) -> bool {
-    num_rows_in_block < DENSE_BLOCK_THRESHOLD as u32
+    num_rows_in_block < DENSE_BLOCK_THRESHOLD
 }
 
 fn deserialize_optional_index_block_metadatas(
@@ -448,7 +448,7 @@ fn deserialize_optional_index_block_metadatas(
     num_rows: u32,
 ) -> (Box<[BlockMeta]>, u32) {
     let num_blocks = data.len() / SERIALIZED_BLOCK_META_NUM_BYTES;
-    let mut block_metas = Vec::with_capacity(num_blocks as usize + 1);
+    let mut block_metas = Vec::with_capacity(num_blocks + 1);
     let mut start_byte_offset = 0;
     let mut non_null_rows_before_block = 0;
     for block_meta_bytes in data.chunks_exact(SERIALIZED_BLOCK_META_NUM_BYTES) {
@@ -479,7 +479,7 @@ fn deserialize_optional_index_block_metadatas(
             block_variant,
         });
         start_byte_offset += block_variant.num_bytes_in_block();
-        non_null_rows_before_block += num_non_null_rows as u32;
+        non_null_rows_before_block += num_non_null_rows;
     }
     block_metas.resize(
         ((num_rows + BLOCK_SIZE - 1) / BLOCK_SIZE) as usize,
@@ -501,7 +501,7 @@ pub fn open_optional_index(bytes: OwnedBytes) -> io::Result<OptionalIndex> {
         num_non_empty_block_bytes as usize * SERIALIZED_BLOCK_META_NUM_BYTES;
     let (block_data, block_metas) = bytes.rsplit(block_metas_num_bytes);
     let (block_metas, num_non_null_rows) =
-        deserialize_optional_index_block_metadatas(block_metas.as_slice(), num_rows).into();
+        deserialize_optional_index_block_metadatas(block_metas.as_slice(), num_rows);
     let optional_index = OptionalIndex {
         num_rows,
         num_non_null_rows,

columnar/src/column_index/optional_index/set.rs

@@ -10,7 +10,7 @@ pub trait SetCodec {
     ///
     /// May panic if the elements are not sorted.
     fn serialize(els: impl Iterator<Item = Self::Item>, wrt: impl io::Write) -> io::Result<()>;
-    fn open<'a>(data: &'a [u8]) -> Self::Reader<'a>;
+    fn open(data: &[u8]) -> Self::Reader<'_>;
 }
 
 /// Stateful object that makes it possible to compute several select in a row,
@@ -43,5 +43,5 @@ pub trait Set<T> {
     fn select(&self, rank: T) -> T;
 
     /// Creates a brand new select cursor.
-    fn select_cursor<'b>(&'b self) -> Self::SelectCursor<'b>;
+    fn select_cursor(&self) -> Self::SelectCursor<'_>;
 }

columnar/src/column_index/optional_index/set_block/dense.rs

@@ -32,7 +32,7 @@ pub const MINI_BLOCK_NUM_BYTES: usize = MINI_BLOCK_BITVEC_NUM_BYTES + MINI_BLOCK
 
 /// Number of bytes in a dense block.
 pub const DENSE_BLOCK_NUM_BYTES: u32 =
-    (ELEMENTS_PER_BLOCK as u32 / ELEMENTS_PER_MINI_BLOCK as u32) * MINI_BLOCK_NUM_BYTES as u32;
+    (ELEMENTS_PER_BLOCK / ELEMENTS_PER_MINI_BLOCK as u32) * MINI_BLOCK_NUM_BYTES as u32;
 
 pub struct DenseBlockCodec;
 
@@ -45,7 +45,7 @@ impl SetCodec for DenseBlockCodec {
     }
 
     #[inline]
-    fn open<'a>(data: &'a [u8]) -> Self::Reader<'a> {
+    fn open(data: &[u8]) -> Self::Reader<'_> {
         assert_eq!(data.len(), DENSE_BLOCK_NUM_BYTES as usize);
         DenseBlock(data)
     }
@@ -94,7 +94,7 @@ impl DenseMiniBlock {
         Self { bitvec, rank }
     }
 
-    fn to_bytes(&self) -> [u8; MINI_BLOCK_NUM_BYTES] {
+    fn to_bytes(self) -> [u8; MINI_BLOCK_NUM_BYTES] {
         let mut bytes = [0u8; MINI_BLOCK_NUM_BYTES];
         bytes[..MINI_BLOCK_BITVEC_NUM_BYTES].copy_from_slice(&self.bitvec.to_le_bytes());
         bytes[MINI_BLOCK_BITVEC_NUM_BYTES..].copy_from_slice(&self.rank.to_le_bytes());
@@ -166,7 +166,7 @@ impl<'a> Set<u16> for DenseBlock<'a> {
     }
 
     #[inline(always)]
-    fn select_cursor<'b>(&'b self) -> Self::SelectCursor<'b> {
+    fn select_cursor(&self) -> Self::SelectCursor<'_> {
         DenseBlockSelectCursor {
             block_id: 0,
             dense_block: *self,
@@ -229,7 +229,7 @@ pub fn serialize_dense_codec(
         while block_id > current_block_id {
             let dense_mini_block = DenseMiniBlock {
                 bitvec: block,
-                rank: non_null_rows_before as u16,
+                rank: non_null_rows_before,
             };
             output.write_all(&dense_mini_block.to_bytes())?;
             non_null_rows_before += block.count_ones() as u16;

columnar/src/column_index/optional_index/set_block/sparse.rs

@@ -16,7 +16,7 @@ impl SetCodec for SparseBlockCodec {
         Ok(())
     }
 
-    fn open<'a>(data: &'a [u8]) -> Self::Reader<'a> {
+    fn open(data: &[u8]) -> Self::Reader<'_> {
         SparseBlock(data)
     }
 }
@@ -56,7 +56,7 @@ impl<'a> Set<u16> for SparseBlock<'a> {
     }
 
     #[inline(always)]
-    fn select_cursor<'b>(&'b self) -> Self::SelectCursor<'b> {
+    fn select_cursor(&self) -> Self::SelectCursor<'_> {
         *self
     }
 }

columnar/src/column_index/optional_index/tests.rs

@@ -37,7 +37,7 @@ proptest! {
 fn test_with_random_sets_simple() {
     let vals = 10..BLOCK_SIZE * 2;
     let mut out: Vec<u8> = Vec::new();
-    serialize_optional_index(&vals.clone(), 100, &mut out).unwrap();
+    serialize_optional_index(&vals, 100, &mut out).unwrap();
     let null_index = open_optional_index(OwnedBytes::new(out)).unwrap();
     let ranks: Vec<u32> = (65_472u32..65_473u32).collect();
     let els: Vec<u32> = ranks.iter().copied().map(|rank| rank + 10).collect();
@@ -142,7 +142,7 @@ fn test_optional_index_large() {
 
 fn test_optional_index_iter_aux(row_ids: &[RowId], num_rows: RowId) {
     let optional_index = OptionalIndex::for_test(num_rows, row_ids);
-    assert_eq!(optional_index.num_rows(), num_rows);
+    assert_eq!(optional_index.num_docs(), num_rows);
     assert!(optional_index.iter_rows().eq(row_ids.iter().copied()));
 }
 
@@ -154,7 +154,7 @@ fn test_optional_index_iter_empty() {
 fn test_optional_index_rank_aux(row_ids: &[RowId]) {
     let num_rows = row_ids.last().copied().unwrap_or(0u32) + 1;
     let null_index = OptionalIndex::for_test(num_rows, row_ids);
-    assert_eq!(null_index.num_rows(), num_rows);
+    assert_eq!(null_index.num_docs(), num_rows);
     for (row_id, row_val) in row_ids.iter().copied().enumerate() {
         assert_eq!(null_index.rank(row_val), row_id as u32);
         assert_eq!(null_index.rank_if_exists(row_val), Some(row_id as u32));
@@ -196,7 +196,7 @@ fn test_optional_index_for_tests() {
     assert!(optional_index.contains(1));
     assert!(optional_index.contains(2));
     assert!(!optional_index.contains(3));
-    assert_eq!(optional_index.num_rows(), 4);
+    assert_eq!(optional_index.num_docs(), 4);
 }
 
 #[cfg(all(test, feature = "unstable"))]
@@ -212,10 +212,13 @@ mod bench {
     fn gen_bools(fill_ratio: f64) -> OptionalIndex {
         let mut out = Vec::new();
         let mut rng: StdRng = StdRng::from_seed([1u8; 32]);
-        let vals: Vec<bool> = (0..TOTAL_NUM_VALUES)
+        let vals: Vec<RowId> = (0..TOTAL_NUM_VALUES)
             .map(|_| rng.gen_bool(fill_ratio))
+            .enumerate()
+            .filter(|(pos, val)| *val)
+            .map(|(pos, _)| pos as RowId)
             .collect();
-        serialize_optional_index(&&vals[..], &mut out).unwrap();
+        serialize_optional_index(&&vals[..], TOTAL_NUM_VALUES, &mut out).unwrap();
         let codec = open_optional_index(OwnedBytes::new(out)).unwrap();
         codec
     }

columnar/src/column_values/bench.rs

@@ -0,0 +1,135 @@
+use std::sync::Arc;
+
+use common::OwnedBytes;
+use rand::rngs::StdRng;
+use rand::{Rng, SeedableRng};
+use test::{self, Bencher};
+
+use super::*;
+use crate::column_values::u64_based::*;
+
+fn get_data() -> Vec<u64> {
+    let mut rng = StdRng::seed_from_u64(2u64);
+    let mut data: Vec<_> = (100..55000_u64)
+        .map(|num| num + rng.gen::<u8>() as u64)
+        .collect();
+    data.push(99_000);
+    data.insert(1000, 2000);
+    data.insert(2000, 100);
+    data.insert(3000, 4100);
+    data.insert(4000, 100);
+    data.insert(5000, 800);
+    data
+}
+
+fn compute_stats(vals: impl Iterator<Item = u64>) -> ColumnStats {
+    let mut stats_collector = StatsCollector::default();
+    for val in vals {
+        stats_collector.collect(val);
+    }
+    stats_collector.stats()
+}
+
+#[inline(never)]
+fn value_iter() -> impl Iterator<Item = u64> {
+    0..20_000
+}
+fn get_reader_for_bench<Codec: ColumnCodec>(data: &[u64]) -> Codec::ColumnValues {
+    let mut bytes = Vec::new();
+    let stats = compute_stats(data.iter().cloned());
+    let mut codec_serializer = Codec::estimator();
+    for val in data {
+        codec_serializer.collect(*val);
+    }
+    codec_serializer.serialize(&stats, Box::new(data.iter().copied()).as_mut(), &mut bytes);
+
+    Codec::load(OwnedBytes::new(bytes)).unwrap()
+}
+fn bench_get<Codec: ColumnCodec>(b: &mut Bencher, data: &[u64]) {
+    let col = get_reader_for_bench::<Codec>(data);
+    b.iter(|| {
+        let mut sum = 0u64;
+        for pos in value_iter() {
+            let val = col.get_val(pos as u32);
+            sum = sum.wrapping_add(val);
+        }
+        sum
+    });
+}
+
+#[inline(never)]
+fn bench_get_dynamic_helper(b: &mut Bencher, col: Arc<dyn ColumnValues>) {
+    b.iter(|| {
+        let mut sum = 0u64;
+        for pos in value_iter() {
+            let val = col.get_val(pos as u32);
+            sum = sum.wrapping_add(val);
+        }
+        sum
+    });
+}
+
+fn bench_get_dynamic<Codec: ColumnCodec>(b: &mut Bencher, data: &[u64]) {
+    let col = Arc::new(get_reader_for_bench::<Codec>(data));
+    bench_get_dynamic_helper(b, col);
+}
+fn bench_create<Codec: ColumnCodec>(b: &mut Bencher, data: &[u64]) {
+    let stats = compute_stats(data.iter().cloned());
+
+    let mut bytes = Vec::new();
+    b.iter(|| {
+        bytes.clear();
+        let mut codec_serializer = Codec::estimator();
+        for val in data.iter().take(1024) {
+            codec_serializer.collect(*val);
+        }
+
+        codec_serializer.serialize(&stats, Box::new(data.iter().copied()).as_mut(), &mut bytes)
+    });
+}
+
+#[bench]
+fn bench_fastfield_bitpack_create(b: &mut Bencher) {
+    let data: Vec<_> = get_data();
+    bench_create::<BitpackedCodec>(b, &data);
+}
+#[bench]
+fn bench_fastfield_linearinterpol_create(b: &mut Bencher) {
+    let data: Vec<_> = get_data();
+    bench_create::<LinearCodec>(b, &data);
+}
+#[bench]
+fn bench_fastfield_multilinearinterpol_create(b: &mut Bencher) {
+    let data: Vec<_> = get_data();
+    bench_create::<BlockwiseLinearCodec>(b, &data);
+}
+#[bench]
+fn bench_fastfield_bitpack_get(b: &mut Bencher) {
+    let data: Vec<_> = get_data();
+    bench_get::<BitpackedCodec>(b, &data);
+}
+#[bench]
+fn bench_fastfield_bitpack_get_dynamic(b: &mut Bencher) {
+    let data: Vec<_> = get_data();
+    bench_get_dynamic::<BitpackedCodec>(b, &data);
+}
+#[bench]
+fn bench_fastfield_linearinterpol_get(b: &mut Bencher) {
+    let data: Vec<_> = get_data();
+    bench_get::<LinearCodec>(b, &data);
+}
+#[bench]
+fn bench_fastfield_linearinterpol_get_dynamic(b: &mut Bencher) {
+    let data: Vec<_> = get_data();
+    bench_get_dynamic::<LinearCodec>(b, &data);
+}
+#[bench]
+fn bench_fastfield_multilinearinterpol_get(b: &mut Bencher) {
+    let data: Vec<_> = get_data();
+    bench_get::<BlockwiseLinearCodec>(b, &data);
+}
+#[bench]
+fn bench_fastfield_multilinearinterpol_get_dynamic(b: &mut Bencher) {
+    let data: Vec<_> = get_data();
+    bench_get_dynamic::<BlockwiseLinearCodec>(b, &data);
+}

columnar/src/column_values/column.rs

@@ -1,383 +0,0 @@
-use std::fmt::Debug;
-use std::marker::PhantomData;
-use std::ops::{Range, RangeInclusive};
-use std::sync::Arc;
-
-use tantivy_bitpacker::minmax;
-
-use crate::column_values::monotonic_mapping::StrictlyMonotonicFn;
-
-/// `ColumnValues` provides access to a dense field column.
-///
-/// `Column` are just a wrapper over `ColumnValues` and a `ColumnIndex`.
-pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
-    /// Return the value associated with the given idx.
-    ///
-    /// This accessor should return as fast as possible.
-    ///
-    /// # Panics
-    ///
-    /// May panic if `idx` is greater than the column length.
-    fn get_val(&self, idx: u32) -> T;
-
-    /// Fills an output buffer with the fast field values
-    /// associated with the `DocId` going from
-    /// `start` to `start + output.len()`.
-    ///
-    /// # Panics
-    ///
-    /// Must panic if `start + output.len()` is greater than
-    /// the segment's `maxdoc`.
-    #[inline(always)]
-    fn get_range(&self, start: u64, output: &mut [T]) {
-        for (out, idx) in output.iter_mut().zip(start..) {
-            *out = self.get_val(idx as u32);
-        }
-    }
-
-    /// Get the positions of values which are in the provided value range.
-    ///
-    /// Note that position == docid for single value fast fields
-    #[inline(always)]
-    fn get_docids_for_value_range(
-        &self,
-        value_range: RangeInclusive<T>,
-        doc_id_range: Range<u32>,
-        positions: &mut Vec<u32>,
-    ) {
-        let doc_id_range = doc_id_range.start..doc_id_range.end.min(self.num_vals());
-        for idx in doc_id_range.start..doc_id_range.end {
-            let val = self.get_val(idx);
-            if value_range.contains(&val) {
-                positions.push(idx);
-            }
-        }
-    }
-
-    /// Returns the minimum value for this fast field.
-    ///
-    /// This min_value may not be exact.
-    /// For instance, the min value does not take in account of possible
-    /// deleted document. All values are however guaranteed to be higher than
-    /// `.min_value()`.
-    fn min_value(&self) -> T;
-
-    /// Returns the maximum value for this fast field.
-    ///
-    /// This max_value may not be exact.
-    /// For instance, the max value does not take in account of possible
-    /// deleted document. All values are however guaranteed to be higher than
-    /// `.max_value()`.
-    fn max_value(&self) -> T;
-
-    /// The number of values in the column.
-    fn num_vals(&self) -> u32;
-
-    /// Returns a iterator over the data
-    fn iter<'a>(&'a self) -> Box<dyn Iterator<Item = T> + 'a> {
-        Box::new((0..self.num_vals()).map(|idx| self.get_val(idx)))
-    }
-}
-
-impl<T: Copy + PartialOrd + Debug> ColumnValues<T> for Arc<dyn ColumnValues<T>> {
-    #[inline(always)]
-    fn get_val(&self, idx: u32) -> T {
-        self.as_ref().get_val(idx)
-    }
-
-    #[inline(always)]
-    fn min_value(&self) -> T {
-        self.as_ref().min_value()
-    }
-
-    #[inline(always)]
-    fn max_value(&self) -> T {
-        self.as_ref().max_value()
-    }
-
-    #[inline(always)]
-    fn num_vals(&self) -> u32 {
-        self.as_ref().num_vals()
-    }
-
-    #[inline(always)]
-    fn iter<'b>(&'b self) -> Box<dyn Iterator<Item = T> + 'b> {
-        self.as_ref().iter()
-    }
-
-    #[inline(always)]
-    fn get_range(&self, start: u64, output: &mut [T]) {
-        self.as_ref().get_range(start, output)
-    }
-}
-
-impl<'a, C: ColumnValues<T> + ?Sized, T: Copy + PartialOrd + Debug> ColumnValues<T> for &'a C {
-    fn get_val(&self, idx: u32) -> T {
-        (*self).get_val(idx)
-    }
-
-    fn min_value(&self) -> T {
-        (*self).min_value()
-    }
-
-    fn max_value(&self) -> T {
-        (*self).max_value()
-    }
-
-    fn num_vals(&self) -> u32 {
-        (*self).num_vals()
-    }
-
-    fn iter<'b>(&'b self) -> Box<dyn Iterator<Item = T> + 'b> {
-        (*self).iter()
-    }
-
-    fn get_range(&self, start: u64, output: &mut [T]) {
-        (*self).get_range(start, output)
-    }
-}
-
-/// VecColumn provides `Column` over a slice.
-pub struct VecColumn<'a, T = u64> {
-    pub(crate) values: &'a [T],
-    pub(crate) min_value: T,
-    pub(crate) max_value: T,
-}
-
-impl<'a, T: Copy + PartialOrd + Send + Sync + Debug> ColumnValues<T> for VecColumn<'a, T> {
-    fn get_val(&self, position: u32) -> T {
-        self.values[position as usize]
-    }
-
-    fn iter(&self) -> Box<dyn Iterator<Item = T> + '_> {
-        Box::new(self.values.iter().copied())
-    }
-
-    fn min_value(&self) -> T {
-        self.min_value
-    }
-
-    fn max_value(&self) -> T {
-        self.max_value
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.values.len() as u32
-    }
-
-    fn get_range(&self, start: u64, output: &mut [T]) {
-        output.copy_from_slice(&self.values[start as usize..][..output.len()])
-    }
-}
-
-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();
-        let (min_value, max_value) = minmax(values.iter().copied()).unwrap_or_default();
-        Self {
-            values,
-            min_value,
-            max_value,
-        }
-    }
-}
-
-struct MonotonicMappingColumn<C, T, Input> {
-    from_column: C,
-    monotonic_mapping: T,
-    _phantom: PhantomData<Input>,
-}
-
-/// Creates a view of a column transformed by a strictly monotonic mapping. See
-/// [`StrictlyMonotonicFn`].
-///
-/// E.g. apply a gcd monotonic_mapping([100, 200, 300]) == [1, 2, 3]
-/// monotonic_mapping.mapping() is expected to be injective, and we should always have
-/// monotonic_mapping.inverse(monotonic_mapping.mapping(el)) == el
-///
-/// The inverse of the mapping is required for:
-/// `fn get_positions_for_value_range(&self, range: RangeInclusive<T>) -> Vec<u64> `
-/// The user provides the original value range and we need to monotonic map them in the same way the
-/// serialization does before calling the underlying column.
-///
-/// Note that when opening a codec, the monotonic_mapping should be the inverse of the mapping
-/// during serialization. And therefore the monotonic_mapping_inv when opening is the same as
-/// monotonic_mapping during serialization.
-pub fn monotonic_map_column<C, T, Input, Output>(
-    from_column: C,
-    monotonic_mapping: T,
-) -> impl ColumnValues<Output>
-where
-    C: ColumnValues<Input>,
-    T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
-    Input: PartialOrd + Debug + Send + Sync + Clone,
-    Output: PartialOrd + Debug + Send + Sync + Clone,
-{
-    MonotonicMappingColumn {
-        from_column,
-        monotonic_mapping,
-        _phantom: PhantomData,
-    }
-}
-
-impl<C, T, Input, Output> ColumnValues<Output> for MonotonicMappingColumn<C, T, Input>
-where
-    C: ColumnValues<Input>,
-    T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
-    Input: PartialOrd + Send + Debug + Sync + Clone,
-    Output: PartialOrd + Send + Debug + Sync + Clone,
-{
-    #[inline]
-    fn get_val(&self, idx: u32) -> Output {
-        let from_val = self.from_column.get_val(idx);
-        self.monotonic_mapping.mapping(from_val)
-    }
-
-    fn min_value(&self) -> Output {
-        let from_min_value = self.from_column.min_value();
-        self.monotonic_mapping.mapping(from_min_value)
-    }
-
-    fn max_value(&self) -> Output {
-        let from_max_value = self.from_column.max_value();
-        self.monotonic_mapping.mapping(from_max_value)
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.from_column.num_vals()
-    }
-
-    fn iter(&self) -> Box<dyn Iterator<Item = Output> + '_> {
-        Box::new(
-            self.from_column
-                .iter()
-                .map(|el| self.monotonic_mapping.mapping(el)),
-        )
-    }
-
-    fn get_docids_for_value_range(
-        &self,
-        range: RangeInclusive<Output>,
-        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,
-        )
-    }
-
-    // We voluntarily do not implement get_range as it yields a regression,
-    // and we do not have any specialized implementation anyway.
-}
-
-/// Wraps an iterator into a `Column`.
-pub struct IterColumn<T>(T);
-
-impl<T> From<T> for IterColumn<T>
-where T: Iterator + Clone + ExactSizeIterator
-{
-    fn from(iter: T) -> Self {
-        IterColumn(iter)
-    }
-}
-
-impl<T> ColumnValues<T::Item> for IterColumn<T>
-where
-    T: Iterator + Clone + ExactSizeIterator + Send + Sync,
-    T::Item: PartialOrd + Debug,
-{
-    fn get_val(&self, idx: u32) -> T::Item {
-        self.0.clone().nth(idx as usize).unwrap()
-    }
-
-    fn min_value(&self) -> T::Item {
-        self.0.clone().next().unwrap()
-    }
-
-    fn max_value(&self) -> T::Item {
-        self.0.clone().last().unwrap()
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.0.len() as u32
-    }
-
-    fn iter(&self) -> Box<dyn Iterator<Item = T::Item> + '_> {
-        Box::new(self.0.clone())
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::column_values::monotonic_mapping::{
-        StrictlyMonotonicMappingInverter, StrictlyMonotonicMappingToInternalBaseval,
-        StrictlyMonotonicMappingToInternalGCDBaseval,
-    };
-
-    #[test]
-    fn test_monotonic_mapping() {
-        let vals = &[3u64, 5u64][..];
-        let col = VecColumn::from(vals);
-        let mapped = monotonic_map_column(col, StrictlyMonotonicMappingToInternalBaseval::new(2));
-        assert_eq!(mapped.min_value(), 1u64);
-        assert_eq!(mapped.max_value(), 3u64);
-        assert_eq!(mapped.num_vals(), 2);
-        assert_eq!(mapped.num_vals(), 2);
-        assert_eq!(mapped.get_val(0), 1);
-        assert_eq!(mapped.get_val(1), 3);
-    }
-
-    #[test]
-    fn test_range_as_col() {
-        let col = IterColumn::from(10..100);
-        assert_eq!(col.num_vals(), 90);
-        assert_eq!(col.max_value(), 99);
-    }
-
-    #[test]
-    fn test_monotonic_mapping_iter() {
-        let vals: Vec<u64> = (10..110u64).map(|el| el * 10).collect();
-        let col = VecColumn::from(&vals);
-        let mapped = monotonic_map_column(
-            col,
-            StrictlyMonotonicMappingInverter::from(
-                StrictlyMonotonicMappingToInternalGCDBaseval::new(10, 100),
-            ),
-        );
-        let val_i64s: Vec<u64> = mapped.iter().collect();
-        for i in 0..100 {
-            assert_eq!(val_i64s[i as usize], mapped.get_val(i));
-        }
-    }
-
-    #[test]
-    fn test_monotonic_mapping_get_range() {
-        let vals: Vec<u64> = (0..100u64).map(|el| el * 10).collect();
-        let col = VecColumn::from(&vals);
-        let mapped = monotonic_map_column(
-            col,
-            StrictlyMonotonicMappingInverter::from(
-                StrictlyMonotonicMappingToInternalGCDBaseval::new(10, 0),
-            ),
-        );
-
-        assert_eq!(mapped.min_value(), 0u64);
-        assert_eq!(mapped.max_value(), 9900u64);
-        assert_eq!(mapped.num_vals(), 100);
-        let val_u64s: Vec<u64> = mapped.iter().collect();
-        assert_eq!(val_u64s.len(), 100);
-        for i in 0..100 {
-            assert_eq!(val_u64s[i as usize], mapped.get_val(i));
-            assert_eq!(val_u64s[i as usize], vals[i as usize] * 10);
-        }
-        let mut buf = [0u64; 20];
-        mapped.get_range(7, &mut buf[..]);
-        assert_eq!(&val_u64s[7..][..20], &buf);
-    }
-}

columnar/src/column_values/merge.rs

@@ -0,0 +1,41 @@
+use std::fmt::Debug;
+use std::sync::Arc;
+
+use crate::iterable::Iterable;
+use crate::{ColumnIndex, ColumnValues, MergeRowOrder};
+
+pub(crate) struct MergedColumnValues<'a, T> {
+    pub(crate) column_indexes: &'a [Option<ColumnIndex>],
+    pub(crate) column_values: &'a [Option<Arc<dyn ColumnValues<T>>>],
+    pub(crate) merge_row_order: &'a MergeRowOrder,
+}
+
+impl<'a, T: Copy + PartialOrd + Debug> Iterable<T> for MergedColumnValues<'a, T> {
+    fn boxed_iter(&self) -> Box<dyn Iterator<Item = T> + '_> {
+        match self.merge_row_order {
+            MergeRowOrder::Stack(_) => Box::new(
+                self.column_values
+                    .iter()
+                    .flatten()
+                    .flat_map(|column_value| column_value.iter()),
+            ),
+            MergeRowOrder::Shuffled(shuffle_merge_order) => Box::new(
+                shuffle_merge_order
+                    .iter_new_to_old_row_addrs()
+                    .flat_map(|row_addr| {
+                        let column_index =
+                            self.column_indexes[row_addr.segment_ord as usize].as_ref()?;
+                        let column_values =
+                            self.column_values[row_addr.segment_ord as usize].as_ref()?;
+                        let value_range = column_index.value_row_ids(row_addr.row_id);
+                        Some((value_range, column_values))
+                    })
+                    .flat_map(|(value_range, column_values)| {
+                        value_range
+                            .into_iter()
+                            .map(|val| column_values.get_val(val))
+                    }),
+            ),
+        }
+    }
+}

columnar/src/column_values/mod.rs

@@ -1,5 +1,4 @@
 #![warn(missing_docs)]
-#![cfg_attr(all(feature = "unstable", test), feature(test))]
 
 //! # `fastfield_codecs`
 //!
@@ -8,248 +7,214 @@
 //! - Monotonically map values to u64/u128
 
 use std::fmt::Debug;
-use std::io;
-use std::io::Write;
+use std::ops::{Range, RangeInclusive};
 use std::sync::Arc;
 
-use common::{BinarySerializable, OwnedBytes};
-use compact_space::CompactSpaceDecompressor;
 pub use monotonic_mapping::{MonotonicallyMappableToU64, StrictlyMonotonicFn};
-use monotonic_mapping::{StrictlyMonotonicMappingInverter, StrictlyMonotonicMappingToInternal};
 pub use monotonic_mapping_u128::MonotonicallyMappableToU128;
-use serialize::U128Header;
 
-mod compact_space;
+mod merge;
 pub(crate) mod monotonic_mapping;
 pub(crate) mod monotonic_mapping_u128;
 mod stats;
-pub(crate) mod u64_based;
+mod u128_based;
+mod u64_based;
+mod vec_column;
 
-mod column;
-pub mod serialize;
+mod monotonic_column;
 
-pub use serialize::serialize_column_values_u128;
-pub use stats::Stats;
+pub(crate) use merge::MergedColumnValues;
+pub use stats::ColumnStats;
+pub use u128_based::{open_u128_mapped, serialize_column_values_u128};
 pub use u64_based::{
     load_u64_based_column_values, serialize_and_load_u64_based_column_values,
     serialize_u64_based_column_values, CodecType, ALL_U64_CODEC_TYPES,
 };
+pub use vec_column::VecColumn;
 
-pub use self::column::{monotonic_map_column, ColumnValues, IterColumn, VecColumn};
-use crate::iterable::Iterable;
-use crate::{ColumnIndex, MergeRowOrder};
+pub use self::monotonic_column::monotonic_map_column;
+use crate::RowId;
 
-pub(crate) struct MergedColumnValues<'a, T> {
-    pub(crate) column_indexes: &'a [Option<ColumnIndex>],
-    pub(crate) column_values: &'a [Option<Arc<dyn ColumnValues<T>>>],
-    pub(crate) merge_row_order: &'a MergeRowOrder,
-}
+/// `ColumnValues` provides access to a dense field column.
+///
+/// `Column` are just a wrapper over `ColumnValues` and a `ColumnIndex`.
+///
+/// Any methods with a default and specialized implementation need to be called in the
+/// wrappers that implement the trait: Arc and MonotonicMappingColumn
+pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
+    /// Return the value associated with the given idx.
+    ///
+    /// This accessor should return as fast as possible.
+    ///
+    /// # Panics
+    ///
+    /// May panic if `idx` is greater than the column length.
+    fn get_val(&self, idx: u32) -> T;
 
-impl<'a, T: Copy + PartialOrd + Debug> Iterable<T> for MergedColumnValues<'a, T> {
-    fn boxed_iter(&self) -> Box<dyn Iterator<Item = T> + '_> {
-        match self.merge_row_order {
-            MergeRowOrder::Stack(_) => {
-                Box::new(self
-                    .column_values
-                    .iter()
-                    .flatten()
-                    .flat_map(|column_value| column_value.iter()))
-            },
-            MergeRowOrder::Shuffled(shuffle_merge_order) => {
-                Box::new(shuffle_merge_order
-                    .iter_new_to_old_row_addrs()
-                    .flat_map(|row_addr| {
-                        let Some(column_index) = self.column_indexes[row_addr.segment_ord as usize].as_ref() else {
-                            return None;
-                        };
-                        let Some(column_values) = self.column_values[row_addr.segment_ord as usize].as_ref() else {
-                            return None;
-                        };
-                        let value_range = column_index.value_row_ids(row_addr.row_id);
-                        Some((value_range, column_values))
-                    })
-                    .flat_map(|(value_range, column_values)| {
-                        value_range
-                            .into_iter()
-                            .map(|val| column_values.get_val(val))
-                    })
-                )
-            },
+    /// Allows to push down multiple fetch calls, to avoid dynamic dispatch overhead.
+    ///
+    /// idx and output should have the same length
+    ///
+    /// # Panics
+    ///
+    /// May panic if `idx` is greater than the column length.
+    fn get_vals(&self, idx: &[u32], output: &mut [T]) {
+        assert!(idx.len() == output.len());
+        for (out, idx) in output.iter_mut().zip(idx.iter()) {
+            *out = self.get_val(*idx as u32);
         }
     }
-}
 
-#[derive(PartialEq, Eq, PartialOrd, Ord, Debug, Clone, Copy)]
-#[repr(u8)]
-/// Available codecs to use to encode the u128 (via [`MonotonicallyMappableToU128`]) converted data.
-pub enum U128FastFieldCodecType {
-    /// This codec takes a large number space (u128) and reduces it to a compact number space, by
-    /// removing the holes.
-    CompactSpace = 1,
-}
-
-impl BinarySerializable for U128FastFieldCodecType {
-    fn serialize<W: Write + ?Sized>(&self, wrt: &mut W) -> io::Result<()> {
-        self.to_code().serialize(wrt)
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let code = u8::deserialize(reader)?;
-        let codec_type: Self = Self::from_code(code)
-            .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidData, "Unknown code `{code}.`"))?;
-        Ok(codec_type)
-    }
-}
-
-impl U128FastFieldCodecType {
-    pub(crate) fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    pub(crate) fn from_code(code: u8) -> Option<Self> {
-        match code {
-            1 => Some(Self::CompactSpace),
-            _ => None,
+    /// Fills an output buffer with the fast field values
+    /// associated with the `DocId` going from
+    /// `start` to `start + output.len()`.
+    ///
+    /// # Panics
+    ///
+    /// Must panic if `start + output.len()` is greater than
+    /// the segment's `maxdoc`.
+    #[inline(always)]
+    fn get_range(&self, start: u64, output: &mut [T]) {
+        for (out, idx) in output.iter_mut().zip(start..) {
+            *out = self.get_val(idx as u32);
         }
     }
+
+    /// Get the row ids of values which are in the provided value range.
+    ///
+    /// Note that position == docid for single value fast fields
+    #[inline(always)]
+    fn get_row_ids_for_value_range(
+        &self,
+        value_range: RangeInclusive<T>,
+        row_id_range: Range<RowId>,
+        row_id_hits: &mut Vec<RowId>,
+    ) {
+        let row_id_range = row_id_range.start..row_id_range.end.min(self.num_vals());
+        for idx in row_id_range.start..row_id_range.end {
+            let val = self.get_val(idx);
+            if value_range.contains(&val) {
+                row_id_hits.push(idx);
+            }
+        }
+    }
+
+    /// Returns the minimum value for this fast field.
+    ///
+    /// This min_value may not be exact.
+    /// For instance, the min value does not take in account of possible
+    /// deleted document. All values are however guaranteed to be higher than
+    /// `.min_value()`.
+    fn min_value(&self) -> T;
+
+    /// Returns the maximum value for this fast field.
+    ///
+    /// This max_value may not be exact.
+    /// For instance, the max value does not take in account of possible
+    /// deleted document. All values are however guaranteed to be higher than
+    /// `.max_value()`.
+    fn max_value(&self) -> T;
+
+    /// The number of values in the column.
+    fn num_vals(&self) -> u32;
+
+    /// Returns a iterator over the data
+    fn iter<'a>(&'a self) -> Box<dyn Iterator<Item = T> + 'a> {
+        Box::new((0..self.num_vals()).map(|idx| self.get_val(idx)))
+    }
 }
 
-/// Returns the correct codec reader wrapped in the `Arc` for the data.
-pub fn open_u128_mapped<T: MonotonicallyMappableToU128 + Debug>(
-    mut bytes: OwnedBytes,
-) -> io::Result<Arc<dyn ColumnValues<T>>> {
-    let header = U128Header::deserialize(&mut bytes)?;
-    assert_eq!(header.codec_type, U128FastFieldCodecType::CompactSpace);
-    let reader = CompactSpaceDecompressor::open(bytes)?;
+impl<T: Copy + PartialOrd + Debug> ColumnValues<T> for Arc<dyn ColumnValues<T>> {
+    #[inline(always)]
+    fn get_val(&self, idx: u32) -> T {
+        self.as_ref().get_val(idx)
+    }
 
-    let inverted: StrictlyMonotonicMappingInverter<StrictlyMonotonicMappingToInternal<T>> =
-        StrictlyMonotonicMappingToInternal::<T>::new().into();
-    Ok(Arc::new(monotonic_map_column(reader, inverted)))
+    #[inline(always)]
+    fn min_value(&self) -> T {
+        self.as_ref().min_value()
+    }
+
+    #[inline(always)]
+    fn max_value(&self) -> T {
+        self.as_ref().max_value()
+    }
+
+    #[inline(always)]
+    fn num_vals(&self) -> u32 {
+        self.as_ref().num_vals()
+    }
+
+    #[inline(always)]
+    fn iter<'b>(&'b self) -> Box<dyn Iterator<Item = T> + 'b> {
+        self.as_ref().iter()
+    }
+
+    #[inline(always)]
+    fn get_range(&self, start: u64, output: &mut [T]) {
+        self.as_ref().get_range(start, output)
+    }
+
+    #[inline(always)]
+    fn get_row_ids_for_value_range(
+        &self,
+        range: RangeInclusive<T>,
+        doc_id_range: Range<u32>,
+        positions: &mut Vec<u32>,
+    ) {
+        self.as_ref()
+            .get_row_ids_for_value_range(range, doc_id_range, positions)
+    }
+}
+
+/// Wraps an cloneable iterator into a `Column`.
+pub struct IterColumn<T>(T);
+
+impl<T> From<T> for IterColumn<T>
+where T: Iterator + Clone + ExactSizeIterator
+{
+    fn from(iter: T) -> Self {
+        IterColumn(iter)
+    }
+}
+
+impl<T> ColumnValues<T::Item> for IterColumn<T>
+where
+    T: Iterator + Clone + ExactSizeIterator + Send + Sync,
+    T::Item: PartialOrd + Debug,
+{
+    fn get_val(&self, idx: u32) -> T::Item {
+        self.0.clone().nth(idx as usize).unwrap()
+    }
+
+    fn min_value(&self) -> T::Item {
+        self.0.clone().next().unwrap()
+    }
+
+    fn max_value(&self) -> T::Item {
+        self.0.clone().last().unwrap()
+    }
+
+    fn num_vals(&self) -> u32 {
+        self.0.len() as u32
+    }
+
+    fn iter(&self) -> Box<dyn Iterator<Item = T::Item> + '_> {
+        Box::new(self.0.clone())
+    }
 }
 
 #[cfg(all(test, feature = "unstable"))]
-mod bench {
-    use std::sync::Arc;
-
-    use common::OwnedBytes;
-    use rand::rngs::StdRng;
-    use rand::{Rng, SeedableRng};
-    use test::{self, Bencher};
+mod bench;
 
+#[cfg(test)]
+mod tests {
     use super::*;
 
-    fn get_data() -> Vec<u64> {
-        let mut rng = StdRng::seed_from_u64(2u64);
-        let mut data: Vec<_> = (100..55000_u64)
-            .map(|num| num + rng.gen::<u8>() as u64)
-            .collect();
-        data.push(99_000);
-        data.insert(1000, 2000);
-        data.insert(2000, 100);
-        data.insert(3000, 4100);
-        data.insert(4000, 100);
-        data.insert(5000, 800);
-        data
-    }
-
-    #[inline(never)]
-    fn value_iter() -> impl Iterator<Item = u64> {
-        0..20_000
-    }
-    fn get_reader_for_bench<Codec: FastFieldCodec>(data: &[u64]) -> Codec::Reader {
-        let mut bytes = Vec::new();
-        let min_value = *data.iter().min().unwrap();
-        let data = data.iter().map(|el| *el - min_value).collect::<Vec<_>>();
-        let col = VecColumn::from(&data);
-        let normalized_header = NormalizedHeader {
-            num_vals: col.num_vals(),
-            max_value: col.max_value(),
-        };
-        Codec::serialize(&VecColumn::from(&data), &mut bytes).unwrap();
-        Codec::open_from_bytes(OwnedBytes::new(bytes), normalized_header).unwrap()
-    }
-    fn bench_get<Codec: FastFieldCodec>(b: &mut Bencher, data: &[u64]) {
-        let col = get_reader_for_bench::<Codec>(data);
-        b.iter(|| {
-            let mut sum = 0u64;
-            for pos in value_iter() {
-                let val = col.get_val(pos as u32);
-                sum = sum.wrapping_add(val);
-            }
-            sum
-        });
-    }
-
-    #[inline(never)]
-    fn bench_get_dynamic_helper(b: &mut Bencher, col: Arc<dyn ColumnValues>) {
-        b.iter(|| {
-            let mut sum = 0u64;
-            for pos in value_iter() {
-                let val = col.get_val(pos as u32);
-                sum = sum.wrapping_add(val);
-            }
-            sum
-        });
-    }
-
-    fn bench_get_dynamic<Codec: FastFieldCodec>(b: &mut Bencher, data: &[u64]) {
-        let col = Arc::new(get_reader_for_bench::<Codec>(data));
-        bench_get_dynamic_helper(b, col);
-    }
-    fn bench_create<Codec: FastFieldCodec>(b: &mut Bencher, data: &[u64]) {
-        let min_value = *data.iter().min().unwrap();
-        let data = data.iter().map(|el| *el - min_value).collect::<Vec<_>>();
-
-        let mut bytes = Vec::new();
-        b.iter(|| {
-            bytes.clear();
-            Codec::serialize(&VecColumn::from(&data), &mut bytes).unwrap();
-        });
-    }
-
-    #[bench]
-    fn bench_fastfield_bitpack_create(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_create::<BitpackedCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_linearinterpol_create(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_create::<LinearCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_multilinearinterpol_create(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_create::<BlockwiseLinearCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_bitpack_get(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_get::<BitpackedCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_bitpack_get_dynamic(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_get_dynamic::<BitpackedCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_linearinterpol_get(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_get::<LinearCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_linearinterpol_get_dynamic(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_get_dynamic::<LinearCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_multilinearinterpol_get(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_get::<BlockwiseLinearCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_multilinearinterpol_get_dynamic(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_get_dynamic::<BlockwiseLinearCodec>(b, &data);
+    #[test]
+    fn test_range_as_col() {
+        let col = IterColumn::from(10..100);
+        assert_eq!(col.num_vals(), 90);
+        assert_eq!(col.max_value(), 99);
     }
 }

columnar/src/column_values/monotonic_column.rs

@@ -0,0 +1,120 @@
+use std::fmt::Debug;
+use std::marker::PhantomData;
+use std::ops::{Range, RangeInclusive};
+
+use crate::column_values::monotonic_mapping::StrictlyMonotonicFn;
+use crate::ColumnValues;
+
+struct MonotonicMappingColumn<C, T, Input> {
+    from_column: C,
+    monotonic_mapping: T,
+    _phantom: PhantomData<Input>,
+}
+
+/// Creates a view of a column transformed by a strictly monotonic mapping. See
+/// [`StrictlyMonotonicFn`].
+///
+/// E.g. apply a gcd monotonic_mapping([100, 200, 300]) == [1, 2, 3]
+/// monotonic_mapping.mapping() is expected to be injective, and we should always have
+/// monotonic_mapping.inverse(monotonic_mapping.mapping(el)) == el
+///
+/// The inverse of the mapping is required for:
+/// `fn get_positions_for_value_range(&self, range: RangeInclusive<T>) -> Vec<u64> `
+/// The user provides the original value range and we need to monotonic map them in the same way the
+/// serialization does before calling the underlying column.
+///
+/// Note that when opening a codec, the monotonic_mapping should be the inverse of the mapping
+/// during serialization. And therefore the monotonic_mapping_inv when opening is the same as
+/// monotonic_mapping during serialization.
+pub fn monotonic_map_column<C, T, Input, Output>(
+    from_column: C,
+    monotonic_mapping: T,
+) -> impl ColumnValues<Output>
+where
+    C: ColumnValues<Input>,
+    T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
+    Input: PartialOrd + Debug + Send + Sync + Clone,
+    Output: PartialOrd + Debug + Send + Sync + Clone,
+{
+    MonotonicMappingColumn {
+        from_column,
+        monotonic_mapping,
+        _phantom: PhantomData,
+    }
+}
+
+impl<C, T, Input, Output> ColumnValues<Output> for MonotonicMappingColumn<C, T, Input>
+where
+    C: ColumnValues<Input>,
+    T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
+    Input: PartialOrd + Send + Debug + Sync + Clone,
+    Output: PartialOrd + Send + Debug + Sync + Clone,
+{
+    #[inline]
+    fn get_val(&self, idx: u32) -> Output {
+        let from_val = self.from_column.get_val(idx);
+        self.monotonic_mapping.mapping(from_val)
+    }
+
+    fn min_value(&self) -> Output {
+        let from_min_value = self.from_column.min_value();
+        self.monotonic_mapping.mapping(from_min_value)
+    }
+
+    fn max_value(&self) -> Output {
+        let from_max_value = self.from_column.max_value();
+        self.monotonic_mapping.mapping(from_max_value)
+    }
+
+    fn num_vals(&self) -> u32 {
+        self.from_column.num_vals()
+    }
+
+    fn iter(&self) -> Box<dyn Iterator<Item = Output> + '_> {
+        Box::new(
+            self.from_column
+                .iter()
+                .map(|el| self.monotonic_mapping.mapping(el)),
+        )
+    }
+
+    fn get_row_ids_for_value_range(
+        &self,
+        range: RangeInclusive<Output>,
+        doc_id_range: Range<u32>,
+        positions: &mut Vec<u32>,
+    ) {
+        self.from_column.get_row_ids_for_value_range(
+            self.monotonic_mapping.inverse(range.start().clone())
+                ..=self.monotonic_mapping.inverse(range.end().clone()),
+            doc_id_range,
+            positions,
+        )
+    }
+
+    // We voluntarily do not implement get_range as it yields a regression,
+    // and we do not have any specialized implementation anyway.
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::column_values::monotonic_mapping::{
+        StrictlyMonotonicMappingInverter, StrictlyMonotonicMappingToInternal,
+    };
+    use crate::column_values::VecColumn;
+
+    #[test]
+    fn test_monotonic_mapping_iter() {
+        let vals: Vec<u64> = (0..100u64).map(|el| el * 10).collect();
+        let col = VecColumn::from(&vals);
+        let mapped = monotonic_map_column(
+            col,
+            StrictlyMonotonicMappingInverter::from(StrictlyMonotonicMappingToInternal::<i64>::new()),
+        );
+        let val_i64s: Vec<u64> = mapped.iter().collect();
+        for i in 0..100 {
+            assert_eq!(val_i64s[i as usize], mapped.get_val(i));
+        }
+    }
+}

columnar/src/column_values/monotonic_mapping.rs

@@ -1,7 +1,7 @@
 use std::fmt::Debug;
 use std::marker::PhantomData;
 
-use fastdivide::DividerU64;
+use common::DateTime;
 
 use super::MonotonicallyMappableToU128;
 use crate::RowId;
@@ -112,65 +112,6 @@ where T: MonotonicallyMappableToU64
     }
 }
 
-/// Mapping dividing by  gcd and a base value.
-///
-/// The function is assumed to be only called on values divided by passed
-/// gcd value. (It is necessary for the function to be monotonic.)
-pub(crate) struct StrictlyMonotonicMappingToInternalGCDBaseval {
-    gcd_divider: DividerU64,
-    gcd: u64,
-    min_value: u64,
-}
-impl StrictlyMonotonicMappingToInternalGCDBaseval {
-    pub(crate) fn new(gcd: u64, min_value: u64) -> Self {
-        let gcd_divider = DividerU64::divide_by(gcd);
-        Self {
-            gcd_divider,
-            gcd,
-            min_value,
-        }
-    }
-}
-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)
-    }
-}
-
-/// Strictly monotonic mapping with a base value.
-pub(crate) struct StrictlyMonotonicMappingToInternalBaseval {
-    min_value: u64,
-}
-impl StrictlyMonotonicMappingToInternalBaseval {
-    #[inline(always)]
-    pub(crate) fn new(min_value: u64) -> Self {
-        Self { min_value }
-    }
-}
-
-impl<External: MonotonicallyMappableToU64> StrictlyMonotonicFn<External, u64>
-    for StrictlyMonotonicMappingToInternalBaseval
-{
-    #[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 {
@@ -195,17 +136,15 @@ impl MonotonicallyMappableToU64 for i64 {
     }
 }
 
-impl MonotonicallyMappableToU64 for crate::DateTime {
+impl MonotonicallyMappableToU64 for DateTime {
     #[inline(always)]
     fn to_u64(self) -> u64 {
-        common::i64_to_u64(self.timestamp_micros)
+        common::i64_to_u64(self.into_timestamp_micros())
     }
 
     #[inline(always)]
     fn from_u64(val: u64) -> Self {
-        crate::DateTime {
-            timestamp_micros: common::u64_to_i64(val),
-        }
+        DateTime::from_timestamp_micros(common::u64_to_i64(val))
     }
 }
 
@@ -261,13 +200,6 @@ mod tests {
         // TODO
         // identity mapping
         // test_round_trip(&StrictlyMonotonicMappingToInternal::<u128>::new(), 100u128);
-
-        // base value to i64 round trip
-        let mapping = StrictlyMonotonicMappingToInternalBaseval::new(100);
-        test_round_trip::<_, _, u64>(&mapping, 100i64);
-        // base value and gcd to u64 round trip
-        let mapping = StrictlyMonotonicMappingToInternalGCDBaseval::new(10, 100);
-        test_round_trip::<_, _, u64>(&mapping, 100u64);
     }
 
     fn test_round_trip<T: StrictlyMonotonicFn<K, L>, K: std::fmt::Debug + Eq + Copy, L>(

columnar/src/column_values/stats.rs

@@ -6,21 +6,28 @@ use common::{BinarySerializable, VInt};
 
 use crate::RowId;
 
+/// Column statistics.
 #[derive(Debug, Clone, Eq, PartialEq)]
-pub struct Stats {
+pub struct ColumnStats {
+    /// GCD of the elements `el - min(column)`.
     pub gcd: NonZeroU64,
+    /// Minimum value of the column.
     pub min_value: u64,
+    /// Maximum value of the column.
     pub max_value: u64,
+    /// Number of rows in the column.
     pub num_rows: RowId,
 }
 
-impl Stats {
+impl ColumnStats {
+    /// Amplitude of value.
+    /// Difference between the maximum and the minimum value.
     pub fn amplitude(&self) -> u64 {
         self.max_value - self.min_value
     }
 }
 
-impl BinarySerializable for Stats {
+impl BinarySerializable for ColumnStats {
     fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
         VInt(self.min_value).serialize(writer)?;
         VInt(self.gcd.get()).serialize(writer)?;
@@ -37,7 +44,7 @@ impl BinarySerializable for Stats {
         let amplitude = VInt::deserialize(reader)?.0 * gcd.get();
         let max_value = min_value + amplitude;
         let num_rows = VInt::deserialize(reader)?.0 as RowId;
-        Ok(Stats {
+        Ok(ColumnStats {
             min_value,
             max_value,
             num_rows,
@@ -52,21 +59,21 @@ mod tests {
 
     use common::BinarySerializable;
 
-    use crate::column_values::Stats;
+    use crate::column_values::ColumnStats;
 
     #[track_caller]
-    fn test_stats_ser_deser_aux(stats: &Stats, num_bytes: usize) {
+    fn test_stats_ser_deser_aux(stats: &ColumnStats, num_bytes: usize) {
         let mut buffer: Vec<u8> = Vec::new();
         stats.serialize(&mut buffer).unwrap();
         assert_eq!(buffer.len(), num_bytes);
-        let deser_stats = Stats::deserialize(&mut &buffer[..]).unwrap();
+        let deser_stats = ColumnStats::deserialize(&mut &buffer[..]).unwrap();
         assert_eq!(stats, &deser_stats);
     }
 
     #[test]
     fn test_stats_serialization() {
         test_stats_ser_deser_aux(
-            &(Stats {
+            &(ColumnStats {
                 gcd: NonZeroU64::new(3).unwrap(),
                 min_value: 1,
                 max_value: 3001,
@@ -75,7 +82,7 @@ mod tests {
             5,
         );
         test_stats_ser_deser_aux(
-            &(Stats {
+            &(ColumnStats {
                 gcd: NonZeroU64::new(1_000).unwrap(),
                 min_value: 1,
                 max_value: 3001,
@@ -84,7 +91,7 @@ mod tests {
             5,
         );
         test_stats_ser_deser_aux(
-            &(Stats {
+            &(ColumnStats {
                 gcd: NonZeroU64::new(1).unwrap(),
                 min_value: 0,
                 max_value: 0,

columnar/src/column_values/u128_based/compact_space/mod.rs

@@ -17,16 +17,16 @@ use std::{
     ops::{Range, RangeInclusive},
 };
 
+mod blank_range;
+mod build_compact_space;
+
+use build_compact_space::get_compact_space;
 use common::{BinarySerializable, CountingWriter, OwnedBytes, VInt, VIntU128};
 use tantivy_bitpacker::{self, BitPacker, BitUnpacker};
 
-use crate::column_values::compact_space::build_compact_space::get_compact_space;
 use crate::column_values::ColumnValues;
 use crate::RowId;
 
-mod blank_range;
-mod build_compact_space;
-
 /// The cost per blank is quite hard actually, since blanks are delta encoded, the actual cost of
 /// blanks depends on the number of blanks.
 ///
@@ -313,7 +313,7 @@ impl ColumnValues<u128> for CompactSpaceDecompressor {
     }
 
     #[inline]
-    fn get_docids_for_value_range(
+    fn get_row_ids_for_value_range(
         &self,
         value_range: RangeInclusive<u128>,
         positions_range: Range<u32>,
@@ -464,7 +464,7 @@ mod tests {
     use itertools::Itertools;
 
     use super::*;
-    use crate::column_values::serialize::U128Header;
+    use crate::column_values::u128_based::U128Header;
     use crate::column_values::{open_u128_mapped, serialize_column_values_u128};
 
     #[test]
@@ -709,7 +709,7 @@ mod tests {
         doc_id_range: Range<u32>,
     ) -> Vec<u32> {
         let mut positions = Vec::new();
-        column.get_docids_for_value_range(value_range, doc_id_range, &mut positions);
+        column.get_row_ids_for_value_range(value_range, doc_id_range, &mut positions);
         positions
     }
 

columnar/src/column_values/u128_based/mod.rs

@@ -1,25 +1,19 @@
 use std::fmt::Debug;
 use std::io;
+use std::io::Write;
+use std::sync::Arc;
 
-use common::{BinarySerializable, VInt};
+mod compact_space;
 
-use crate::column_values::compact_space::CompactSpaceCompressor;
-use crate::column_values::U128FastFieldCodecType;
+use common::{BinarySerializable, OwnedBytes, VInt};
+use compact_space::{CompactSpaceCompressor, CompactSpaceDecompressor};
+
+use crate::column_values::monotonic_map_column;
+use crate::column_values::monotonic_mapping::{
+    StrictlyMonotonicMappingInverter, StrictlyMonotonicMappingToInternal,
+};
 use crate::iterable::Iterable;
-use crate::MonotonicallyMappableToU128;
-
-/// The normalized header gives some parameters after applying the following
-/// normalization of the vector:
-/// `val -> (val - min_value) / gcd`
-///
-/// By design, after normalization, `min_value = 0` and `gcd = 1`.
-#[derive(Debug, Copy, Clone)]
-pub struct NormalizedHeader {
-    /// The number of values in the underlying column.
-    pub num_vals: u32,
-    /// The max value of the underlying column.
-    pub max_value: u64,
-}
+use crate::{ColumnValues, MonotonicallyMappableToU128};
 
 #[derive(Debug, Copy, Clone, PartialEq, Eq)]
 pub(crate) struct U128Header {
@@ -68,6 +62,52 @@ pub fn serialize_column_values_u128<T: MonotonicallyMappableToU128>(
     Ok(())
 }
 
+#[derive(PartialEq, Eq, PartialOrd, Ord, Debug, Clone, Copy)]
+#[repr(u8)]
+/// Available codecs to use to encode the u128 (via [`MonotonicallyMappableToU128`]) converted data.
+pub(crate) enum U128FastFieldCodecType {
+    /// This codec takes a large number space (u128) and reduces it to a compact number space, by
+    /// removing the holes.
+    CompactSpace = 1,
+}
+
+impl BinarySerializable for U128FastFieldCodecType {
+    fn serialize<W: Write + ?Sized>(&self, wrt: &mut W) -> io::Result<()> {
+        self.to_code().serialize(wrt)
+    }
+
+    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
+        let code = u8::deserialize(reader)?;
+        let codec_type: Self = Self::from_code(code)
+            .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidData, "Unknown code `{code}.`"))?;
+        Ok(codec_type)
+    }
+}
+
+impl U128FastFieldCodecType {
+    pub(crate) fn to_code(self) -> u8 {
+        self as u8
+    }
+
+    pub(crate) fn from_code(code: u8) -> Option<Self> {
+        match code {
+            1 => Some(Self::CompactSpace),
+            _ => None,
+        }
+    }
+}
+
+/// Returns the correct codec reader wrapped in the `Arc` for the data.
+pub fn open_u128_mapped<T: MonotonicallyMappableToU128 + Debug>(
+    mut bytes: OwnedBytes,
+) -> io::Result<Arc<dyn ColumnValues<T>>> {
+    let header = U128Header::deserialize(&mut bytes)?;
+    assert_eq!(header.codec_type, U128FastFieldCodecType::CompactSpace);
+    let reader = CompactSpaceDecompressor::open(bytes)?;
+    let inverted: StrictlyMonotonicMappingInverter<StrictlyMonotonicMappingToInternal<T>> =
+        StrictlyMonotonicMappingToInternal::<T>::new().into();
+    Ok(Arc::new(monotonic_map_column(reader, inverted)))
+}
 #[cfg(test)]
 pub mod tests {
     use super::*;

columnar/src/column_values/u64_based/bitpacked.rs

@@ -4,7 +4,7 @@ use common::{BinarySerializable, OwnedBytes};
 use fastdivide::DividerU64;
 use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
 
-use crate::column_values::u64_based::{ColumnCodec, ColumnCodecEstimator, Stats};
+use crate::column_values::u64_based::{ColumnCodec, ColumnCodecEstimator, ColumnStats};
 use crate::{ColumnValues, RowId};
 
 /// Depending on the field type, a different
@@ -13,7 +13,7 @@ use crate::{ColumnValues, RowId};
 pub struct BitpackedReader {
     data: OwnedBytes,
     bit_unpacker: BitUnpacker,
-    stats: Stats,
+    stats: ColumnStats,
 }
 
 impl ColumnValues for BitpackedReader {
@@ -36,7 +36,7 @@ impl ColumnValues for BitpackedReader {
     }
 }
 
-fn num_bits(stats: &Stats) -> u8 {
+fn num_bits(stats: &ColumnStats) -> u8 {
     compute_num_bits(stats.amplitude() / stats.gcd)
 }
 
@@ -46,14 +46,14 @@ pub struct BitpackedCodecEstimator;
 impl ColumnCodecEstimator for BitpackedCodecEstimator {
     fn collect(&mut self, _value: u64) {}
 
-    fn estimate(&self, stats: &Stats) -> Option<u64> {
+    fn estimate(&self, stats: &ColumnStats) -> Option<u64> {
         let num_bits_per_value = num_bits(stats);
         Some(stats.num_bytes() + (stats.num_rows as u64 * (num_bits_per_value as u64) + 7) / 8)
     }
 
     fn serialize(
         &self,
-        stats: &Stats,
+        stats: &ColumnStats,
         vals: &mut dyn Iterator<Item = u64>,
         wrt: &mut dyn Write,
     ) -> io::Result<()> {
@@ -72,12 +72,12 @@ impl ColumnCodecEstimator for BitpackedCodecEstimator {
 pub struct BitpackedCodec;
 
 impl ColumnCodec for BitpackedCodec {
-    type Reader = BitpackedReader;
+    type ColumnValues = BitpackedReader;
     type Estimator = BitpackedCodecEstimator;
 
     /// Opens a fast field given a file.
-    fn load(mut data: OwnedBytes) -> io::Result<Self::Reader> {
-        let stats = Stats::deserialize(&mut data)?;
+    fn load(mut data: OwnedBytes) -> io::Result<Self::ColumnValues> {
+        let stats = ColumnStats::deserialize(&mut data)?;
         let num_bits = num_bits(&stats);
         let bit_unpacker = BitUnpacker::new(num_bits);
         Ok(BitpackedReader {

columnar/src/column_values/u64_based/blockwise_linear.rs

@@ -7,7 +7,7 @@ use fastdivide::DividerU64;
 use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
 
 use crate::column_values::u64_based::line::Line;
-use crate::column_values::u64_based::{ColumnCodec, ColumnCodecEstimator, Stats};
+use crate::column_values::u64_based::{ColumnCodec, ColumnCodecEstimator, ColumnStats};
 use crate::column_values::{ColumnValues, VecColumn};
 use crate::MonotonicallyMappableToU64;
 
@@ -84,7 +84,7 @@ impl ColumnCodecEstimator for BlockwiseLinearEstimator {
             self.block.clear();
         }
     }
-    fn estimate(&self, stats: &Stats) -> Option<u64> {
+    fn estimate(&self, stats: &ColumnStats) -> Option<u64> {
         let mut estimate = 4 + stats.num_bytes() + self.meta_num_bytes + self.values_num_bytes;
         if stats.gcd.get() > 1 {
             let estimate_gain_from_gcd =
@@ -100,7 +100,7 @@ impl ColumnCodecEstimator for BlockwiseLinearEstimator {
 
     fn serialize(
         &self,
-        stats: &Stats,
+        stats: &ColumnStats,
         mut vals: &mut dyn Iterator<Item = u64>,
         wrt: &mut dyn Write,
     ) -> io::Result<()> {
@@ -165,12 +165,12 @@ impl ColumnCodecEstimator for BlockwiseLinearEstimator {
 pub struct BlockwiseLinearCodec;
 
 impl ColumnCodec<u64> for BlockwiseLinearCodec {
-    type Reader = BlockwiseLinearReader;
+    type ColumnValues = BlockwiseLinearReader;
 
     type Estimator = BlockwiseLinearEstimator;
 
-    fn load(mut bytes: OwnedBytes) -> io::Result<Self::Reader> {
-        let stats = Stats::deserialize(&mut bytes)?;
+    fn load(mut bytes: OwnedBytes) -> io::Result<Self::ColumnValues> {
+        let stats = ColumnStats::deserialize(&mut bytes)?;
         let footer_len: u32 = (&bytes[bytes.len() - 4..]).deserialize()?;
         let footer_offset = bytes.len() - 4 - footer_len as usize;
         let (data, mut footer) = bytes.split(footer_offset);
@@ -195,14 +195,14 @@ impl ColumnCodec<u64> for BlockwiseLinearCodec {
 pub struct BlockwiseLinearReader {
     blocks: Arc<[Block]>,
     data: OwnedBytes,
-    stats: Stats,
+    stats: ColumnStats,
 }
 
 impl ColumnValues for BlockwiseLinearReader {
     #[inline(always)]
     fn get_val(&self, idx: u32) -> u64 {
-        let block_id = (idx / BLOCK_SIZE as u32) as usize;
-        let idx_within_block = idx % (BLOCK_SIZE as u32);
+        let block_id = (idx / BLOCK_SIZE) as usize;
+        let idx_within_block = idx % BLOCK_SIZE;
         let block = &self.blocks[block_id];
         let interpoled_val: u64 = block.line.eval(idx_within_block);
         let block_bytes = &self.data[block.data_start_offset..];

columnar/src/column_values/u64_based/linear.rs

@@ -5,7 +5,7 @@ use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
 
 use super::line::Line;
 use super::ColumnValues;
-use crate::column_values::u64_based::{ColumnCodec, ColumnCodecEstimator, Stats};
+use crate::column_values::u64_based::{ColumnCodec, ColumnCodecEstimator, ColumnStats};
 use crate::column_values::VecColumn;
 use crate::RowId;
 
@@ -18,7 +18,7 @@ const LINE_ESTIMATION_BLOCK_LEN: usize = 512;
 pub struct LinearReader {
     data: OwnedBytes,
     linear_params: LinearParams,
-    stats: Stats,
+    stats: ColumnStats,
 }
 
 impl ColumnValues for LinearReader {
@@ -106,7 +106,7 @@ impl ColumnCodecEstimator for LinearCodecEstimator {
         }
     }
 
-    fn estimate(&self, stats: &Stats) -> Option<u64> {
+    fn estimate(&self, stats: &ColumnStats) -> Option<u64> {
         let line = self.line?;
         let amplitude = self.max_deviation - self.min_deviation;
         let num_bits = compute_num_bits(amplitude);
@@ -123,7 +123,7 @@ impl ColumnCodecEstimator for LinearCodecEstimator {
 
     fn serialize(
         &self,
-        stats: &Stats,
+        stats: &ColumnStats,
         vals: &mut dyn Iterator<Item = u64>,
         wrt: &mut dyn io::Write,
     ) -> io::Result<()> {
@@ -184,12 +184,12 @@ impl LinearCodecEstimator {
 }
 
 impl ColumnCodec for LinearCodec {
-    type Reader = LinearReader;
+    type ColumnValues = LinearReader;
 
     type Estimator = LinearCodecEstimator;
 
-    fn load(mut data: OwnedBytes) -> io::Result<Self::Reader> {
-        let stats = Stats::deserialize(&mut data)?;
+    fn load(mut data: OwnedBytes) -> io::Result<Self::ColumnValues> {
+        let stats = ColumnStats::deserialize(&mut data)?;
         let linear_params = LinearParams::deserialize(&mut data)?;
         Ok(LinearReader {
             stats,

columnar/src/column_values/u64_based/mod.rs

@@ -13,35 +13,61 @@ use common::{BinarySerializable, OwnedBytes};
 use crate::column_values::monotonic_mapping::{
     StrictlyMonotonicMappingInverter, StrictlyMonotonicMappingToInternal,
 };
-use crate::column_values::u64_based::bitpacked::BitpackedCodec;
-use crate::column_values::u64_based::blockwise_linear::BlockwiseLinearCodec;
-use crate::column_values::u64_based::linear::LinearCodec;
-use crate::column_values::u64_based::stats_collector::StatsCollector;
-use crate::column_values::{monotonic_map_column, Stats};
+pub use crate::column_values::u64_based::bitpacked::BitpackedCodec;
+pub use crate::column_values::u64_based::blockwise_linear::BlockwiseLinearCodec;
+pub use crate::column_values::u64_based::linear::LinearCodec;
+pub use crate::column_values::u64_based::stats_collector::StatsCollector;
+use crate::column_values::{monotonic_map_column, ColumnStats};
 use crate::iterable::Iterable;
 use crate::{ColumnValues, MonotonicallyMappableToU64};
 
+/// A `ColumnCodecEstimator` is in charge of gathering all
+/// data required to serialize a column.
+///
+/// This happens during a first pass on data of the column elements.
+/// During that pass, all column estimators receive a call to their
+/// `.collect(el)`.
+///
+/// After this first pass, finalize is called.
+/// `.estimate(..)` then should return an accurate estimation of the
+/// size of the serialized column (were we to pick this codec.).
+/// `.serialize(..)` then serializes the column using this codec.
 pub trait ColumnCodecEstimator<T = u64>: 'static {
+    /// Records a new value for estimation.
+    /// This method will be called for each element of the column during
+    /// `estimation`.
     fn collect(&mut self, value: u64);
-    fn estimate(&self, stats: &Stats) -> Option<u64>;
+    /// Finalizes the first pass phase.
     fn finalize(&mut self) {}
+    /// Returns an accurate estimation of the number of bytes that will
+    /// be used to represent this column.
+    fn estimate(&self, stats: &ColumnStats) -> Option<u64>;
+    /// Serializes the column using the given codec.
+    /// This constitutes a second pass over the columns values.
     fn serialize(
         &self,
-        stats: &Stats,
+        stats: &ColumnStats,
         vals: &mut dyn Iterator<Item = T>,
         wrt: &mut dyn io::Write,
     ) -> io::Result<()>;
 }
 
+/// A column codec describes a colunm serialization format.
 pub trait ColumnCodec<T: PartialOrd = u64> {
-    type Reader: ColumnValues<T> + 'static;
+    /// Specialized `ColumnValues` type.
+    type ColumnValues: ColumnValues<T> + 'static;
+    /// `Estimator` for the given codec.
     type Estimator: ColumnCodecEstimator + Default;
 
-    fn load(bytes: OwnedBytes) -> io::Result<Self::Reader>;
+    /// Loads a column that has been serialized using this codec.
+    fn load(bytes: OwnedBytes) -> io::Result<Self::ColumnValues>;
 
+    /// Returns an estimator.
     fn estimator() -> Self::Estimator {
         Self::Estimator::default()
     }
+
+    /// Returns a boxed estimator.
     fn boxed_estimator() -> Box<dyn ColumnCodecEstimator> {
         Box::new(Self::estimator())
     }
@@ -62,6 +88,7 @@ pub enum CodecType {
     BlockwiseLinear = 2u8,
 }
 
+/// List of all available u64-base codecs.
 pub const ALL_U64_CODEC_TYPES: [CodecType; 3] = [
     CodecType::Bitpacked,
     CodecType::Linear,
@@ -106,6 +133,7 @@ fn load_specific_codec<C: ColumnCodec, T: MonotonicallyMappableToU64>(
 }
 
 impl CodecType {
+    /// Returns a boxed codec estimator associated to a given `CodecType`.
     pub fn estimator(&self) -> Box<dyn ColumnCodecEstimator> {
         match self {
             CodecType::Bitpacked => BitpackedCodec::boxed_estimator(),
@@ -115,7 +143,8 @@ impl CodecType {
     }
 }
 
-pub fn serialize_u64_based_column_values<'a, T: MonotonicallyMappableToU64>(
+/// Serializes a given column of u64-mapped values.
+pub fn serialize_u64_based_column_values<T: MonotonicallyMappableToU64>(
     vals: &dyn Iterable<T>,
     codec_types: &[CodecType],
     wrt: &mut dyn Write,
@@ -156,11 +185,14 @@ pub fn serialize_u64_based_column_values<'a, T: MonotonicallyMappableToU64>(
     Ok(())
 }
 
+/// Load u64-based column values.
+///
+/// This method first identifies the codec off the first byte.
 pub fn load_u64_based_column_values<T: MonotonicallyMappableToU64>(
     mut bytes: OwnedBytes,
 ) -> io::Result<Arc<dyn ColumnValues<T>>> {
     let codec_type: CodecType = bytes
-        .get(0)
+        .first()
         .copied()
         .and_then(CodecType::try_from_code)
         .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidData, "Failed to read codec type"))?;

columnar/src/column_values/u64_based/stats_collector.rs

@@ -2,7 +2,7 @@ use std::num::NonZeroU64;
 
 use fastdivide::DividerU64;
 
-use crate::column_values::Stats;
+use crate::column_values::ColumnStats;
 use crate::RowId;
 
 /// Compute the gcd of two non null numbers.
@@ -33,14 +33,14 @@ pub struct StatsCollector {
 }
 
 impl StatsCollector {
-    pub fn stats(&self) -> Stats {
+    pub fn stats(&self) -> ColumnStats {
         let (min_value, max_value) = self.min_max_opt.unwrap_or((0u64, 0u64));
         let increment_gcd = if let Some((increment_gcd, _)) = self.increment_gcd_opt {
             increment_gcd
         } else {
             NonZeroU64::new(1u64).unwrap()
         };
-        Stats {
+        ColumnStats {
             min_value,
             max_value,
             num_rows: self.num_rows,
@@ -97,9 +97,9 @@ mod tests {
     use std::num::NonZeroU64;
 
     use crate::column_values::u64_based::stats_collector::{compute_gcd, StatsCollector};
-    use crate::column_values::u64_based::Stats;
+    use crate::column_values::u64_based::ColumnStats;
 
-    fn compute_stats(vals: impl Iterator<Item = u64>) -> Stats {
+    fn compute_stats(vals: impl Iterator<Item = u64>) -> ColumnStats {
         let mut stats_collector = StatsCollector::default();
         for val in vals {
             stats_collector.collect(val);
@@ -144,7 +144,7 @@ mod tests {
     fn test_stats() {
         assert_eq!(
             compute_stats([].into_iter()),
-            Stats {
+            ColumnStats {
                 gcd: NonZeroU64::new(1).unwrap(),
                 min_value: 0,
                 max_value: 0,
@@ -153,7 +153,7 @@ mod tests {
         );
         assert_eq!(
             compute_stats([0, 1].into_iter()),
-            Stats {
+            ColumnStats {
                 gcd: NonZeroU64::new(1).unwrap(),
                 min_value: 0,
                 max_value: 1,
@@ -162,7 +162,7 @@ mod tests {
         );
         assert_eq!(
             compute_stats([0, 1].into_iter()),
-            Stats {
+            ColumnStats {
                 gcd: NonZeroU64::new(1).unwrap(),
                 min_value: 0,
                 max_value: 1,
@@ -171,7 +171,7 @@ mod tests {
         );
         assert_eq!(
             compute_stats([10, 20, 30].into_iter()),
-            Stats {
+            ColumnStats {
                 gcd: NonZeroU64::new(10).unwrap(),
                 min_value: 10,
                 max_value: 30,
@@ -180,7 +180,7 @@ mod tests {
         );
         assert_eq!(
             compute_stats([10, 50, 10, 30].into_iter()),
-            Stats {
+            ColumnStats {
                 gcd: NonZeroU64::new(20).unwrap(),
                 min_value: 10,
                 max_value: 50,
@@ -189,7 +189,7 @@ mod tests {
         );
         assert_eq!(
             compute_stats([10, 0, 30].into_iter()),
-            Stats {
+            ColumnStats {
                 gcd: NonZeroU64::new(10).unwrap(),
                 min_value: 0,
                 max_value: 30,

columnar/src/column_values/u64_based/tests.rs

@@ -1,6 +1,6 @@
 use proptest::prelude::*;
 use proptest::strategy::Strategy;
-use proptest::{prop_oneof, proptest};
+use proptest::{num, prop_oneof, proptest};
 
 #[test]
 fn test_serialize_and_load_simple() {
@@ -19,6 +19,62 @@ fn test_serialize_and_load_simple() {
     assert_eq!(col.get_val(1), 2);
     assert_eq!(col.get_val(2), 5);
 }
+
+#[test]
+fn test_empty_column_i64() {
+    let vals: [i64; 0] = [];
+    let mut num_acceptable_codecs = 0;
+    for codec in ALL_U64_CODEC_TYPES {
+        let mut buffer = Vec::new();
+        if serialize_u64_based_column_values(&&vals[..], &[codec], &mut buffer).is_err() {
+            continue;
+        }
+        num_acceptable_codecs += 1;
+        let col = load_u64_based_column_values::<i64>(OwnedBytes::new(buffer)).unwrap();
+        assert_eq!(col.num_vals(), 0);
+        assert_eq!(col.min_value(), i64::MIN);
+        assert_eq!(col.max_value(), i64::MIN);
+    }
+    assert!(num_acceptable_codecs > 0);
+}
+
+#[test]
+fn test_empty_column_u64() {
+    let vals: [u64; 0] = [];
+    let mut num_acceptable_codecs = 0;
+    for codec in ALL_U64_CODEC_TYPES {
+        let mut buffer = Vec::new();
+        if serialize_u64_based_column_values(&&vals[..], &[codec], &mut buffer).is_err() {
+            continue;
+        }
+        num_acceptable_codecs += 1;
+        let col = load_u64_based_column_values::<u64>(OwnedBytes::new(buffer)).unwrap();
+        assert_eq!(col.num_vals(), 0);
+        assert_eq!(col.min_value(), u64::MIN);
+        assert_eq!(col.max_value(), u64::MIN);
+    }
+    assert!(num_acceptable_codecs > 0);
+}
+
+#[test]
+fn test_empty_column_f64() {
+    let vals: [f64; 0] = [];
+    let mut num_acceptable_codecs = 0;
+    for codec in ALL_U64_CODEC_TYPES {
+        let mut buffer = Vec::new();
+        if serialize_u64_based_column_values(&&vals[..], &[codec], &mut buffer).is_err() {
+            continue;
+        }
+        num_acceptable_codecs += 1;
+        let col = load_u64_based_column_values::<f64>(OwnedBytes::new(buffer)).unwrap();
+        assert_eq!(col.num_vals(), 0);
+        // FIXME. f64::MIN would be better!
+        assert!(col.min_value().is_nan());
+        assert!(col.max_value().is_nan());
+    }
+    assert!(num_acceptable_codecs > 0);
+}
+
 pub(crate) fn create_and_validate<TColumnCodec: ColumnCodec>(
     vals: &[u64],
     name: &str,
@@ -60,7 +116,7 @@ pub(crate) fn create_and_validate<TColumnCodec: ColumnCodec>(
             .map(|(pos, _)| pos as u32)
             .collect();
         let mut positions = Vec::new();
-        reader.get_docids_for_value_range(
+        reader.get_row_ids_for_value_range(
             vals[test_rand_idx]..=vals[test_rand_idx],
             0..vals.len() as u32,
             &mut positions,

columnar/src/column_values/vec_column.rs

@@ -0,0 +1,52 @@
+use std::fmt::Debug;
+
+use tantivy_bitpacker::minmax;
+
+use crate::ColumnValues;
+
+/// VecColumn provides `Column` over a slice.
+pub struct VecColumn<'a, T = u64> {
+    pub(crate) values: &'a [T],
+    pub(crate) min_value: T,
+    pub(crate) max_value: T,
+}
+
+impl<'a, T: Copy + PartialOrd + Send + Sync + Debug> ColumnValues<T> for VecColumn<'a, T> {
+    fn get_val(&self, position: u32) -> T {
+        self.values[position as usize]
+    }
+
+    fn iter(&self) -> Box<dyn Iterator<Item = T> + '_> {
+        Box::new(self.values.iter().copied())
+    }
+
+    fn min_value(&self) -> T {
+        self.min_value
+    }
+
+    fn max_value(&self) -> T {
+        self.max_value
+    }
+
+    fn num_vals(&self) -> u32 {
+        self.values.len() as u32
+    }
+
+    fn get_range(&self, start: u64, output: &mut [T]) {
+        output.copy_from_slice(&self.values[start as usize..][..output.len()])
+    }
+}
+
+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();
+        let (min_value, max_value) = minmax(values.iter().copied()).unwrap_or_default();
+        Self {
+            values,
+            min_value,
+            max_value,
+        }
+    }
+}

columnar/src/columnar/column_type.rs

@@ -1,12 +1,14 @@
 use std::fmt::Debug;
 use std::net::Ipv6Addr;
 
+use serde::{Deserialize, Serialize};
+
 use crate::value::NumericalType;
 use crate::InvalidData;
 
 /// The column type represents the column type.
 /// Any changes need to be propagated to `COLUMN_TYPES`.
-#[derive(Hash, Eq, PartialEq, Debug, Clone, Copy, Ord, PartialOrd)]
+#[derive(Hash, Eq, PartialEq, Debug, Clone, Copy, Ord, PartialOrd, Serialize, Deserialize)]
 #[repr(u8)]
 pub enum ColumnType {
     I64 = 0u8,
@@ -111,7 +113,7 @@ impl HasAssociatedColumnType for bool {
     }
 }
 
-impl HasAssociatedColumnType for crate::DateTime {
+impl HasAssociatedColumnType for common::DateTime {
     fn column_type() -> ColumnType {
         ColumnType::DateTime
     }
@@ -143,7 +145,7 @@ mod tests {
             }
         }
         for code in COLUMN_TYPES.len() as u8..=u8::MAX {
-            assert!(ColumnType::try_from_code(code as u8).is_err());
+            assert!(ColumnType::try_from_code(code).is_err());
         }
     }
 

columnar/src/columnar/format_version.rs

@@ -4,7 +4,7 @@ pub const VERSION_FOOTER_NUM_BYTES: usize = MAGIC_BYTES.len() + std::mem::size_o
 
 /// We end the file by these 4 bytes just to somewhat identify that
 /// this is indeed a columnar file.
-const MAGIC_BYTES: [u8; 4] = [2, 113, 119, 066];
+const MAGIC_BYTES: [u8; 4] = [2, 113, 119, 66];
 
 pub fn footer() -> [u8; VERSION_FOOTER_NUM_BYTES] {
     let mut footer_bytes = [0u8; VERSION_FOOTER_NUM_BYTES];
@@ -27,8 +27,8 @@ pub enum Version {
 }
 
 impl Version {
-    fn to_bytes(&self) -> [u8; 4] {
-        (*self as u32).to_le_bytes()
+    fn to_bytes(self) -> [u8; 4] {
+        (self as u32).to_le_bytes()
     }
 
     fn try_from_bytes(bytes: [u8; 4]) -> Result<Version, InvalidData> {

columnar/src/columnar/merge/merge_dict_column.rs

@@ -58,7 +58,7 @@ impl<'a> RemappedTermOrdinalsValues<'a> {
             .enumerate()
             .flat_map(|(segment_ord, byte_column)| {
                 let segment_ord = self.term_ord_mapping.get_segment(segment_ord as u32);
-                byte_column.into_iter().flat_map(move |bytes_column| {
+                byte_column.iter().flat_map(move |bytes_column| {
                     bytes_column
                         .ords()
                         .values
@@ -96,7 +96,7 @@ fn compute_term_bitset(column: &BytesColumn, row_bitset: &ReadOnlyBitSet) -> Bit
     let num_terms = column.dictionary().num_terms();
     let mut term_bitset = BitSet::with_max_value(num_terms as u32);
     for row_id in row_bitset.iter() {
-        for term_ord in column.term_ord_column.values(row_id) {
+        for term_ord in column.term_ord_column.values_for_doc(row_id) {
             term_bitset.insert(term_ord as u32);
         }
     }
@@ -191,7 +191,7 @@ struct TermOrdinalMapping {
 impl TermOrdinalMapping {
     fn add_segment(&mut self, max_term_ord: usize) {
         self.per_segment_new_term_ordinals
-            .push(vec![TermOrdinal::default(); max_term_ord as usize]);
+            .push(vec![TermOrdinal::default(); max_term_ord]);
     }
 
     fn register_from_to(&mut self, segment_ord: usize, from_ord: TermOrdinal, to_ord: TermOrdinal) {

columnar/src/columnar/merge/mod.rs

@@ -2,8 +2,6 @@ mod merge_dict_column;
 mod merge_mapping;
 mod term_merger;
 
-// mod sorted_doc_id_column;
-
 use std::collections::{BTreeMap, HashMap, HashSet};
 use std::io;
 use std::net::Ipv6Addr;
@@ -54,14 +52,34 @@ impl From<ColumnType> for ColumnTypeCategory {
     }
 }
 
+/// Merge several columnar table together.
+///
+/// If several columns with the same name are conflicting with the numerical types in the
+/// input columnars, the first type compatible out of i64, u64, f64 in that order will be used.
+///
+/// `require_columns` makes it possible to ensure that some columns will be present in the
+/// resulting columnar. When a required column is a numerical column type, one of two things can
+/// happen:
+/// - If the required column type is compatible with all of the input columnar, the resulsting
+///   merged
+/// columnar will simply coerce the input column and use the required column type.
+/// - If the required column type is incompatible with one of the input columnar, the merged
+/// will fail with an InvalidData error.
+///
+/// `merge_row_order` makes it possible to remove or reorder row in the resulting
+/// `Columnar` table.
+///
+/// Reminder: a string and a numerical column may bare the same column name. This is not
+/// considered a conflict.
 pub fn merge_columnar(
     columnar_readers: &[&ColumnarReader],
+    required_columns: &[(String, ColumnType)],
     merge_row_order: MergeRowOrder,
     output: &mut impl io::Write,
 ) -> io::Result<()> {
     let mut serializer = ColumnarSerializer::new(output);
 
-    let columns_to_merge = group_columns_for_merge(columnar_readers)?;
+    let columns_to_merge = group_columns_for_merge(columnar_readers, required_columns)?;
     for ((column_name, column_type), columns) in columns_to_merge {
         let mut column_serializer =
             serializer.serialize_column(column_name.as_bytes(), column_type);
@@ -174,97 +192,183 @@ fn merge_column(
     Ok(())
 }
 
+struct GroupedColumns {
+    required_column_type: Option<ColumnType>,
+    columns: Vec<Option<DynamicColumn>>,
+    column_category: ColumnTypeCategory,
+}
+
+impl GroupedColumns {
+    fn for_category(column_category: ColumnTypeCategory, num_columnars: usize) -> Self {
+        GroupedColumns {
+            required_column_type: None,
+            columns: vec![None; num_columnars],
+            column_category,
+        }
+    }
+
+    /// Set the dynamic column for a given columnar.
+    fn set_column(&mut self, columnar_id: usize, column: DynamicColumn) {
+        self.columns[columnar_id] = Some(column);
+    }
+
+    /// Force the existence of a column, as well as its type.
+    fn require_type(&mut self, required_type: ColumnType) -> io::Result<()> {
+        if let Some(existing_required_type) = self.required_column_type {
+            if existing_required_type == required_type {
+                // This was just a duplicate in the `required_columns`.
+                // Nothing to do.
+                return Ok(());
+            } else {
+                return Err(io::Error::new(
+                    io::ErrorKind::InvalidInput,
+                    "Required column conflicts with another required column of the same type \
+                     category.",
+                ));
+            }
+        }
+        self.required_column_type = Some(required_type);
+        Ok(())
+    }
+
+    /// Returns the column type after merge.
+    ///
+    /// This method does not check if the column types can actually be coerced to
+    /// this type.
+    fn column_type_after_merge(&self) -> ColumnType {
+        if let Some(required_type) = self.required_column_type {
+            return required_type;
+        }
+        let column_type: HashSet<ColumnType> = self
+            .columns
+            .iter()
+            .flatten()
+            .map(|column| column.column_type())
+            .collect();
+        if column_type.len() == 1 {
+            return column_type.into_iter().next().unwrap();
+        }
+        // At the moment, only the numerical categorical column type has more than one possible
+        // column type.
+        assert_eq!(self.column_category, ColumnTypeCategory::Numerical);
+        merged_numerical_columns_type(self.columns.iter().flatten()).into()
+    }
+}
+
+/// Returns the type of the merged numerical column.
+///
+/// This function picks the first numerical type out of i64, u64, f64 (order matters
+/// here), that is compatible with all the `columns`.
+///
+/// # Panics
+/// Panics if one of the column is not numerical.
+fn merged_numerical_columns_type<'a>(
+    columns: impl Iterator<Item = &'a DynamicColumn>,
+) -> NumericalType {
+    let mut compatible_numerical_types = CompatibleNumericalTypes::default();
+    for column in columns {
+        let (min_value, max_value) =
+            min_max_if_numerical(column).expect("All columns re required to be numerical");
+        compatible_numerical_types.accept_value(min_value);
+        compatible_numerical_types.accept_value(max_value);
+    }
+    compatible_numerical_types.to_numerical_type()
+}
+
+#[allow(clippy::type_complexity)]
 fn group_columns_for_merge(
     columnar_readers: &[&ColumnarReader],
+    required_columns: &[(String, ColumnType)],
 ) -> io::Result<BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>>> {
     // Each column name may have multiple types of column associated.
     // For merging we are interested in the same column type category since they can be merged.
-    let mut columns_grouped: HashMap<(String, ColumnTypeCategory), Vec<Option<DynamicColumn>>> =
-        HashMap::new();
+    let mut columns_grouped: HashMap<(String, ColumnTypeCategory), GroupedColumns> = HashMap::new();
 
-    let num_columnars = columnar_readers.len();
+    for &(ref column_name, column_type) in required_columns {
+        columns_grouped
+            .entry((column_name.clone(), column_type.into()))
+            .or_insert_with(|| {
+                GroupedColumns::for_category(column_type.into(), columnar_readers.len())
+            })
+            .require_type(column_type)?;
+    }
 
     for (columnar_id, columnar_reader) in columnar_readers.iter().enumerate() {
         let column_name_and_handle = columnar_reader.list_columns()?;
         for (column_name, handle) in column_name_and_handle {
-            let column_type_category: ColumnTypeCategory = handle.column_type().into();
-            let columns = columns_grouped
-                .entry((column_name, column_type_category))
-                .or_insert_with(|| vec![None; num_columnars]);
+            let column_category: ColumnTypeCategory = handle.column_type().into();
             let column = handle.open()?;
-            columns[columnar_id] = Some(column);
+            columns_grouped
+                .entry((column_name, column_category))
+                .or_insert_with(|| {
+                    GroupedColumns::for_category(column_category, columnar_readers.len())
+                })
+                .set_column(columnar_id, column);
         }
     }
 
     let mut merge_columns: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
-        BTreeMap::default();
+        Default::default();
 
-    for ((column_name, col_category), mut columns) in columns_grouped {
-        if col_category == ColumnTypeCategory::Numerical {
-            coerce_numerical_columns_to_same_type(&mut columns);
-        }
-        let column_type = columns
-            .iter()
-            .flatten()
-            .map(|col| col.column_type())
-            .next()
-            .unwrap();
-        merge_columns.insert((column_name, column_type), columns);
+    for ((column_name, _), mut grouped_columns) in columns_grouped {
+        let column_type = grouped_columns.column_type_after_merge();
+        coerce_columns(column_type, &mut grouped_columns.columns)?;
+        merge_columns.insert((column_name, column_type), grouped_columns.columns);
     }
 
     Ok(merge_columns)
 }
 
-/// Coerce a set of numerical columns to the same type.
-///
-/// If all columns are already from the same type, keep this type
-/// (even if they could all be coerced to i64).
-fn coerce_numerical_columns_to_same_type(columns: &mut [Option<DynamicColumn>]) {
-    let mut column_types: HashSet<NumericalType> = HashSet::default();
-    let mut compatible_numerical_types = CompatibleNumericalTypes::default();
-    for column in columns.iter().flatten() {
-        let min_value: NumericalValue;
-        let max_value: NumericalValue;
-        match column {
-            DynamicColumn::I64(column) => {
-                min_value = column.min_value().into();
-                max_value = column.max_value().into();
-            }
-            DynamicColumn::U64(column) => {
-                min_value = column.min_value().into();
-                max_value = column.min_value().into();
-            }
-            DynamicColumn::F64(column) => {
-                min_value = column.min_value().into();
-                max_value = column.min_value().into();
-            }
-            DynamicColumn::Bool(_)
-            | DynamicColumn::IpAddr(_)
-            | DynamicColumn::DateTime(_)
-            | DynamicColumn::Bytes(_)
-            | DynamicColumn::Str(_) => {
-                panic!("We expected only numerical columns.");
-            }
-        }
-        column_types.insert(column.column_type().numerical_type().unwrap());
-        compatible_numerical_types.accept_value(min_value);
-        compatible_numerical_types.accept_value(max_value);
-    }
-    if column_types.len() <= 1 {
-        // No need to do anything. The columns are already all from the same type.
-        // This is necessary to let use force a given type.
-
-        // TODO This works in a world where we do not allow a change of schema,
-        // but in the future, we will have to pass some kind of schema to enforce
-        // the logic.
-        return;
-    }
-    let coerce_type = compatible_numerical_types.to_numerical_type();
+fn coerce_columns(
+    column_type: ColumnType,
+    columns: &mut [Option<DynamicColumn>],
+) -> io::Result<()> {
     for column_opt in columns.iter_mut() {
         if let Some(column) = column_opt.take() {
-            *column_opt = column.coerce_numerical(coerce_type);
+            *column_opt = Some(coerce_column(column_type, column)?);
         }
     }
+    Ok(())
+}
+
+fn coerce_column(column_type: ColumnType, column: DynamicColumn) -> io::Result<DynamicColumn> {
+    if let Some(numerical_type) = column_type.numerical_type() {
+        column
+            .coerce_numerical(numerical_type)
+            .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidInput, ""))
+    } else {
+        if column.column_type() != column_type {
+            return Err(io::Error::new(
+                io::ErrorKind::InvalidInput,
+                format!(
+                    "Cannot coerce column of type `{:?}` to `{column_type:?}`",
+                    column.column_type()
+                ),
+            ));
+        }
+        Ok(column)
+    }
+}
+
+/// Returns the (min, max) of a column provided it is numerical (i64, u64. f64).
+///
+/// The min and the max are simply the numerical value as defined by `ColumnValue::min_value()`,
+/// and `ColumnValue::max_value()`.
+///
+/// It is important to note that these values are only guaranteed to be lower/upper bound
+/// (as opposed to min/max value).
+/// If a column is empty, the min and max values are currently set to 0.
+fn min_max_if_numerical(column: &DynamicColumn) -> Option<(NumericalValue, NumericalValue)> {
+    match column {
+        DynamicColumn::I64(column) => Some((column.min_value().into(), column.max_value().into())),
+        DynamicColumn::U64(column) => Some((column.min_value().into(), column.min_value().into())),
+        DynamicColumn::F64(column) => Some((column.min_value().into(), column.min_value().into())),
+        DynamicColumn::Bool(_)
+        | DynamicColumn::IpAddr(_)
+        | DynamicColumn::DateTime(_)
+        | DynamicColumn::Bytes(_)
+        | DynamicColumn::Str(_) => None,
+    }
 }
 
 #[cfg(test)]

columnar/src/columnar/merge/sorted_doc_id_column.rs

@@ -1,107 +0,0 @@
-use std::sync::Arc;
-
-use fastfield_codecs::Column;
-use itertools::Itertools;
-
-use crate::indexer::doc_id_mapping::SegmentDocIdMapping;
-use crate::SegmentReader;
-
-pub(crate) struct RemappedDocIdColumn<'a> {
-    doc_id_mapping: &'a SegmentDocIdMapping,
-    fast_field_readers: Vec<Arc<dyn Column<u64>>>,
-    min_value: u64,
-    max_value: u64,
-    num_vals: u32,
-}
-
-fn compute_min_max_val(
-    u64_reader: &dyn Column<u64>,
-    segment_reader: &SegmentReader,
-) -> Option<(u64, u64)> {
-    if segment_reader.max_doc() == 0 {
-        return None;
-    }
-
-    if segment_reader.alive_bitset().is_none() {
-        // no deleted documents,
-        // we can use the previous min_val, max_val.
-        return Some((u64_reader.min_value(), u64_reader.max_value()));
-    }
-    // some deleted documents,
-    // we need to recompute the max / min
-    segment_reader
-        .doc_ids_alive()
-        .map(|doc_id| u64_reader.get_val(doc_id))
-        .minmax()
-        .into_option()
-}
-
-impl<'a> RemappedDocIdColumn<'a> {
-    pub(crate) fn new(
-        readers: &'a [SegmentReader],
-        doc_id_mapping: &'a SegmentDocIdMapping,
-        field: &str,
-    ) -> Self {
-        let (min_value, max_value) = readers
-            .iter()
-            .filter_map(|reader| {
-                let u64_reader: Arc<dyn Column<u64>> =
-                    reader.fast_fields().typed_fast_field_reader(field).expect(
-                        "Failed to find a reader for single fast field. This is a tantivy bug and \
-                         it should never happen.",
-                    );
-                compute_min_max_val(&*u64_reader, reader)
-            })
-            .reduce(|a, b| (a.0.min(b.0), a.1.max(b.1)))
-            .expect("Unexpected error, empty readers in IndexMerger");
-
-        let fast_field_readers = readers
-            .iter()
-            .map(|reader| {
-                let u64_reader: Arc<dyn Column<u64>> =
-                    reader.fast_fields().typed_fast_field_reader(field).expect(
-                        "Failed to find a reader for single fast field. This is a tantivy bug and \
-                         it should never happen.",
-                    );
-                u64_reader
-            })
-            .collect::<Vec<_>>();
-
-        RemappedDocIdColumn {
-            doc_id_mapping,
-            fast_field_readers,
-            min_value,
-            max_value,
-            num_vals: doc_id_mapping.len() as u32,
-        }
-    }
-}
-
-impl<'a> Column for RemappedDocIdColumn<'a> {
-    fn get_val(&self, _doc: u32) -> u64 {
-        unimplemented!()
-    }
-
-    fn iter(&self) -> Box<dyn Iterator<Item = u64> + '_> {
-        Box::new(
-            self.doc_id_mapping
-                .iter_old_doc_addrs()
-                .map(|old_doc_addr| {
-                    let fast_field_reader =
-                        &self.fast_field_readers[old_doc_addr.segment_ord as usize];
-                    fast_field_reader.get_val(old_doc_addr.doc_id)
-                }),
-        )
-    }
-    fn min_value(&self) -> u64 {
-        self.min_value
-    }
-
-    fn max_value(&self) -> u64 {
-        self.max_value
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.num_vals
-    }
-}

columnar/src/columnar/merge/tests.rs

@@ -24,7 +24,7 @@ fn test_column_coercion_to_u64() {
     // u64 type
     let columnar2 = make_columnar("numbers", &[u64::MAX]);
     let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
-        group_columns_for_merge(&[&columnar1, &columnar2]).unwrap();
+        group_columns_for_merge(&[&columnar1, &columnar2], &[]).unwrap();
     assert_eq!(column_map.len(), 1);
     assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
 }
@@ -34,7 +34,7 @@ fn test_column_no_coercion_if_all_the_same() {
     let columnar1 = make_columnar("numbers", &[1u64]);
     let columnar2 = make_columnar("numbers", &[2u64]);
     let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
-        group_columns_for_merge(&[&columnar1, &columnar2]).unwrap();
+        group_columns_for_merge(&[&columnar1, &columnar2], &[]).unwrap();
     assert_eq!(column_map.len(), 1);
     assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
 }
@@ -44,17 +44,74 @@ fn test_column_coercion_to_i64() {
     let columnar1 = make_columnar("numbers", &[-1i64]);
     let columnar2 = make_columnar("numbers", &[2u64]);
     let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
-        group_columns_for_merge(&[&columnar1, &columnar2]).unwrap();
+        group_columns_for_merge(&[&columnar1, &columnar2], &[]).unwrap();
     assert_eq!(column_map.len(), 1);
     assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::I64)));
 }
 
+#[test]
+fn test_impossible_coercion_returns_an_error() {
+    let columnar1 = make_columnar("numbers", &[u64::MAX]);
+    let group_error =
+        group_columns_for_merge(&[&columnar1], &[("numbers".to_string(), ColumnType::I64)])
+            .map(|_| ())
+            .unwrap_err();
+    assert_eq!(group_error.kind(), io::ErrorKind::InvalidInput);
+}
+
+#[test]
+fn test_group_columns_with_required_column() {
+    let columnar1 = make_columnar("numbers", &[1i64]);
+    let columnar2 = make_columnar("numbers", &[2u64]);
+    let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
+        group_columns_for_merge(
+            &[&columnar1, &columnar2],
+            &[("numbers".to_string(), ColumnType::U64)],
+        )
+        .unwrap();
+    assert_eq!(column_map.len(), 1);
+    assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
+}
+
+#[test]
+fn test_group_columns_required_column_with_no_existing_columns() {
+    let columnar1 = make_columnar("numbers", &[2u64]);
+    let columnar2 = make_columnar("numbers", &[2u64]);
+    let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
+        group_columns_for_merge(
+            &[&columnar1, &columnar2],
+            &[("required_col".to_string(), ColumnType::Str)],
+        )
+        .unwrap();
+    assert_eq!(column_map.len(), 2);
+    let columns = column_map
+        .get(&("required_col".to_string(), ColumnType::Str))
+        .unwrap();
+    assert_eq!(columns.len(), 2);
+    assert!(columns[0].is_none());
+    assert!(columns[1].is_none());
+}
+
+#[test]
+fn test_group_columns_required_column_is_above_all_columns_have_the_same_type_rule() {
+    let columnar1 = make_columnar("numbers", &[2i64]);
+    let columnar2 = make_columnar("numbers", &[2i64]);
+    let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
+        group_columns_for_merge(
+            &[&columnar1, &columnar2],
+            &[("numbers".to_string(), ColumnType::U64)],
+        )
+        .unwrap();
+    assert_eq!(column_map.len(), 1);
+    assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
+}
+
 #[test]
 fn test_missing_column() {
     let columnar1 = make_columnar("numbers", &[-1i64]);
     let columnar2 = make_columnar("numbers2", &[2u64]);
     let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
-        group_columns_for_merge(&[&columnar1, &columnar2]).unwrap();
+        group_columns_for_merge(&[&columnar1, &columnar2], &[]).unwrap();
     assert_eq!(column_map.len(), 2);
     assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::I64)));
     {
@@ -101,7 +158,7 @@ fn make_byte_columnar_multiple_columns(columns: &[(&str, &[&[&[u8]]])]) -> Colum
     for (column_name, column_values) in columns {
         for (row_id, vals) in column_values.iter().enumerate() {
             for val in vals.iter() {
-                dataframe_writer.record_bytes(row_id as u32, column_name, *val);
+                dataframe_writer.record_bytes(row_id as u32, column_name, val);
             }
         }
     }
@@ -122,7 +179,7 @@ fn make_text_columnar_multiple_columns(columns: &[(&str, &[&[&str]])]) -> Column
     for (column_name, column_values) in columns {
         for (row_id, vals) in column_values.iter().enumerate() {
             for val in vals.iter() {
-                dataframe_writer.record_str(row_id as u32, column_name, *val);
+                dataframe_writer.record_str(row_id as u32, column_name, val);
             }
         }
     }
@@ -151,6 +208,7 @@ fn test_merge_columnar_numbers() {
     let stack_merge_order = StackMergeOrder::stack(columnars);
     crate::columnar::merge_columnar(
         columnars,
+        &[],
         MergeRowOrder::Stack(stack_merge_order),
         &mut buffer,
     )
@@ -176,6 +234,7 @@ fn test_merge_columnar_texts() {
     let stack_merge_order = StackMergeOrder::stack(columnars);
     crate::columnar::merge_columnar(
         columnars,
+        &[],
         MergeRowOrder::Stack(stack_merge_order),
         &mut buffer,
     )
@@ -220,6 +279,7 @@ fn test_merge_columnar_byte() {
     let stack_merge_order = StackMergeOrder::stack(columnars);
     crate::columnar::merge_columnar(
         columnars,
+        &[],
         MergeRowOrder::Stack(stack_merge_order),
         &mut buffer,
     )

columnar/src/columnar/merge_index.rs

@@ -1 +0,0 @@
-

columnar/src/columnar/mod.rs

@@ -1,7 +1,6 @@
 mod column_type;
 mod format_version;
 mod merge;
-mod merge_index;
 mod reader;
 mod writer;
 

columnar/src/columnar/reader/mod.rs

@@ -21,6 +21,32 @@ pub struct ColumnarReader {
     num_rows: RowId,
 }
 
+/// Functions by both the async/sync code listing columns.
+/// It takes a stream from the column sstable and return the list of
+/// `DynamicColumn` available in it.
+fn read_all_columns_in_stream(
+    mut stream: sstable::Streamer<'_, RangeSSTable>,
+    column_data: &FileSlice,
+) -> io::Result<Vec<DynamicColumnHandle>> {
+    let mut results = Vec::new();
+    while stream.advance() {
+        let key_bytes: &[u8] = stream.key();
+        let Some(column_code) = key_bytes.last().copied() else {
+            return Err(io_invalid_data("Empty column name.".to_string()));
+        };
+        let column_type = ColumnType::try_from_code(column_code)
+            .map_err(|_| io_invalid_data(format!("Unknown column code `{column_code}`")))?;
+        let range = stream.value();
+        let file_slice = column_data.slice(range.start as usize..range.end as usize);
+        let dynamic_column_handle = DynamicColumnHandle {
+            file_slice,
+            column_type,
+        };
+        results.push(dynamic_column_handle);
+    }
+    Ok(results)
+}
+
 impl ColumnarReader {
     /// Opens a new Columnar file.
     pub fn open<F>(file_slice: F) -> io::Result<ColumnarReader>
@@ -76,11 +102,7 @@ impl ColumnarReader {
         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.
-    pub fn read_columns(&self, column_name: &str) -> io::Result<Vec<DynamicColumnHandle>> {
+    fn stream_for_column_range(&self, column_name: &str) -> sstable::StreamerBuilder<RangeSSTable> {
         // Each column is a associated to a given `column_key`,
         // that starts by `column_name\0column_header`.
         //
@@ -89,36 +111,35 @@ impl ColumnarReader {
         //
         // This is in turn equivalent to searching for the range
         // `[column_name,\0`..column_name\1)`.
-
-        // TODO can we get some more generic `prefix(..)` logic in the dictioanry.
+        // TODO can we get some more generic `prefix(..)` logic in the dictionary.
         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
+        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 = ColumnType::try_from_code(column_code)
-                .map_err(|_| io_invalid_data(format!("Unknown column code `{column_code}`")))?;
-            let range = stream.value().clone();
-            let file_slice = self
-                .column_data
-                .slice(range.start as usize..range.end as usize);
-            let dynamic_column_handle = DynamicColumnHandle {
-                file_slice,
-                column_type,
-            };
-            results.push(dynamic_column_handle);
-        }
-        Ok(results)
+    }
+
+    pub async fn read_columns_async(
+        &self,
+        column_name: &str,
+    ) -> io::Result<Vec<DynamicColumnHandle>> {
+        let stream = self
+            .stream_for_column_range(column_name)
+            .into_stream_async()
+            .await?;
+        read_all_columns_in_stream(stream, &self.column_data)
+    }
+
+    /// 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.
+    pub fn read_columns(&self, column_name: &str) -> io::Result<Vec<DynamicColumnHandle>> {
+        let stream = self.stream_for_column_range(column_name).into_stream()?;
+        read_all_columns_in_stream(stream, &self.column_data)
     }
 
     /// Return the number of columns in the columnar.
@@ -162,7 +183,7 @@ mod tests {
     }
 
     #[test]
-    #[should_panic(expect = "Input type forbidden")]
+    #[should_panic(expected = "Input type forbidden")]
     fn test_list_columns_strict_typing_panics_on_wrong_types() {
         let mut columnar_writer = ColumnarWriter::default();
         columnar_writer.record_column_type("count", ColumnType::U64, false);

columnar/src/columnar/writer/column_operation.rs

@@ -310,7 +310,7 @@ mod tests {
         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!(bytes.len() + buf.as_ref().len(), buffer.len());
         assert_eq!(column_op, serdeser_symbol);
     }
 
@@ -341,7 +341,7 @@ mod tests {
     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);
+        assert_eq!({ minibuf.as_ref().len() }, expected_len);
         let mut buf = minibuf.as_ref().to_vec();
         buf.extend_from_slice(&[2, 2, 2, 2, 2, 2]);
         let mut cursor = &buf[..];

columnar/src/columnar/writer/mod.rs

@@ -47,6 +47,7 @@ struct SpareBuffers {
 /// let mut wrt: Vec<u8> =  Vec::new();
 /// columnar_writer.serialize(2u32, None, &mut wrt).unwrap();
 /// ```
+#[derive(Default)]
 pub struct ColumnarWriter {
     numerical_field_hash_map: ArenaHashMap,
     datetime_field_hash_map: ArenaHashMap,
@@ -60,22 +61,6 @@ pub struct ColumnarWriter {
     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),
-            ip_addr_field_hash_map: ArenaHashMap::new(10_000),
-            bytes_field_hash_map: ArenaHashMap::new(10_000),
-            str_field_hash_map: ArenaHashMap::new(10_000),
-            datetime_field_hash_map: ArenaHashMap::new(10_000),
-            dictionaries: Vec::new(),
-            arena: MemoryArena::default(),
-            buffers: SpareBuffers::default(),
-        }
-    }
-}
-
 #[inline]
 fn mutate_or_create_column<V, TMutator>(
     arena_hash_map: &mut ArenaHashMap,
@@ -266,11 +251,15 @@ impl ColumnarWriter {
         });
     }
 
-    pub fn record_datetime(&mut self, doc: RowId, column_name: &str, datetime: crate::DateTime) {
+    pub fn record_datetime(&mut self, doc: RowId, column_name: &str, datetime: common::DateTime) {
         let (hash_map, arena) = (&mut self.datetime_field_hash_map, &mut self.arena);
         mutate_or_create_column(hash_map, column_name, |column_opt: Option<ColumnWriter>| {
             let mut column: ColumnWriter = column_opt.unwrap_or_default();
-            column.record(doc, NumericalValue::I64(datetime.timestamp_micros), arena);
+            column.record(
+                doc,
+                NumericalValue::I64(datetime.into_timestamp_micros()),
+                arena,
+            );
             column
         });
     }
@@ -667,7 +656,7 @@ where
     Ok(())
 }
 
-fn sort_values_within_row_in_place(multivalued_index: &[RowId], values: &mut Vec<u64>) {
+fn sort_values_within_row_in_place(multivalued_index: &[RowId], values: &mut [u64]) {
     let mut start_index: usize = 0;
     for end_index in multivalued_index.iter().copied() {
         let end_index = end_index as usize;
@@ -772,7 +761,7 @@ mod tests {
         assert_eq!(column_writer.get_cardinality(3), Cardinality::Full);
         let mut buffer = Vec::new();
         let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
-            .operation_iterator(&mut arena, None, &mut buffer)
+            .operation_iterator(&arena, None, &mut buffer)
             .collect();
         assert_eq!(symbols.len(), 6);
         assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));
@@ -801,7 +790,7 @@ mod tests {
         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, None, &mut buffer)
+            .operation_iterator(&arena, None, &mut buffer)
             .collect();
         assert_eq!(symbols.len(), 4);
         assert!(matches!(symbols[0], ColumnOperation::NewDoc(1u32)));
@@ -824,7 +813,7 @@ mod tests {
         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, None, &mut buffer)
+            .operation_iterator(&arena, None, &mut buffer)
             .collect();
         assert_eq!(symbols.len(), 2);
         assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));
@@ -843,7 +832,7 @@ mod tests {
         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, None, &mut buffer)
+            .operation_iterator(&arena, None, &mut buffer)
             .collect();
         assert_eq!(symbols.len(), 3);
         assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));

columnar/src/columnar/writer/value_index.rs

@@ -29,7 +29,7 @@ pub struct OptionalIndexBuilder {
 }
 
 impl OptionalIndexBuilder {
-    pub fn finish<'a>(&'a mut self, num_rows: RowId) -> impl Iterable<RowId> + 'a {
+    pub fn finish(&mut self, num_rows: RowId) -> impl Iterable<RowId> + '_ {
         debug_assert!(self
             .docs
             .last()
@@ -150,11 +150,7 @@ mod tests {
         multivalued_value_index_builder.record_row(2u32);
         multivalued_value_index_builder.record_value();
         assert_eq!(
-            multivalued_value_index_builder
-                .finish(4u32)
-                .iter()
-                .copied()
-                .collect::<Vec<u32>>(),
+            multivalued_value_index_builder.finish(4u32).to_vec(),
             vec![0, 0, 2, 3, 3]
         );
         multivalued_value_index_builder.reset();
@@ -162,11 +158,7 @@ mod tests {
         multivalued_value_index_builder.record_value();
         multivalued_value_index_builder.record_value();
         assert_eq!(
-            multivalued_value_index_builder
-                .finish(4u32)
-                .iter()
-                .copied()
-                .collect::<Vec<u32>>(),
+            multivalued_value_index_builder.finish(4u32).to_vec(),
             vec![0, 0, 0, 2, 2]
         );
     }

columnar/src/dynamic_column.rs

@@ -3,12 +3,12 @@ use std::net::Ipv6Addr;
 use std::sync::Arc;
 
 use common::file_slice::FileSlice;
-use common::{HasLen, OwnedBytes};
+use common::{DateTime, HasLen, OwnedBytes};
 
 use crate::column::{BytesColumn, Column, StrColumn};
 use crate::column_values::{monotonic_map_column, StrictlyMonotonicFn};
 use crate::columnar::ColumnType;
-use crate::{Cardinality, DateTime, NumericalType};
+use crate::{Cardinality, NumericalType};
 
 #[derive(Clone)]
 pub enum DynamicColumn {
@@ -166,9 +166,9 @@ impl StrictlyMonotonicFn<i64, u64> for MapI64ToU64 {
 
 macro_rules! static_dynamic_conversions {
     ($typ:ty, $enum_name:ident) => {
-        impl Into<Option<$typ>> for DynamicColumn {
-            fn into(self) -> Option<$typ> {
-                if let DynamicColumn::$enum_name(col) = self {
+        impl From<DynamicColumn> for Option<$typ> {
+            fn from(dynamic_column: DynamicColumn) -> Option<$typ> {
+                if let DynamicColumn::$enum_name(col) = dynamic_column {
                     Some(col)
                 } else {
                     None
@@ -188,7 +188,7 @@ static_dynamic_conversions!(Column<bool>, Bool);
 static_dynamic_conversions!(Column<u64>, U64);
 static_dynamic_conversions!(Column<i64>, I64);
 static_dynamic_conversions!(Column<f64>, F64);
-static_dynamic_conversions!(Column<crate::DateTime>, DateTime);
+static_dynamic_conversions!(Column<DateTime>, DateTime);
 static_dynamic_conversions!(StrColumn, Str);
 static_dynamic_conversions!(BytesColumn, Bytes);
 static_dynamic_conversions!(Column<Ipv6Addr>, IpAddr);
@@ -206,10 +206,9 @@ impl DynamicColumnHandle {
         self.open_internal(column_bytes)
     }
 
-    // TODO rename load_async
-    pub async fn open_async(&self) -> io::Result<DynamicColumn> {
-        let column_bytes: OwnedBytes = self.file_slice.read_bytes_async().await?;
-        self.open_internal(column_bytes)
+    #[doc(hidden)]
+    pub fn file_slice(&self) -> &FileSlice {
+        &self.file_slice
     }
 
     /// Returns the `u64` fast field reader reader associated with `fields` of types
@@ -243,7 +242,7 @@ impl DynamicColumnHandle {
             ColumnType::Bool => crate::column::open_column_u64::<bool>(column_bytes)?.into(),
             ColumnType::IpAddr => crate::column::open_column_u128::<Ipv6Addr>(column_bytes)?.into(),
             ColumnType::DateTime => {
-                crate::column::open_column_u64::<crate::DateTime>(column_bytes)?.into()
+                crate::column::open_column_u64::<DateTime>(column_bytes)?.into()
             }
         };
         Ok(dynamic_column)

columnar/src/lib.rs

@@ -32,6 +32,7 @@ pub use value::{NumericalType, NumericalValue};
 pub use self::dynamic_column::{DynamicColumn, DynamicColumnHandle};
 
 pub type RowId = u32;
+pub type DocId = u32;
 
 #[derive(Clone, Copy)]
 pub struct RowAddr {
@@ -42,16 +43,7 @@ pub struct RowAddr {
 pub use sstable::Dictionary;
 pub type Streamer<'a> = sstable::Streamer<'a, VoidSSTable>;
 
-#[derive(Clone, Copy, PartialOrd, PartialEq, Default, Debug)]
-pub struct DateTime {
-    pub timestamp_micros: i64,
-}
-
-impl DateTime {
-    pub fn into_timestamp_micros(self) -> i64 {
-        self.timestamp_micros
-    }
-}
+pub use common::DateTime;
 
 #[derive(Copy, Clone, Debug)]
 pub struct InvalidData;

columnar/src/tests.rs

@@ -75,7 +75,7 @@ fn test_dataframe_writer_u64_multivalued() {
         divisor_col.get_cardinality(),
         crate::Cardinality::Multivalued
     );
-    assert_eq!(divisor_col.num_rows(), 7);
+    assert_eq!(divisor_col.num_docs(), 7);
 }
 
 #[test]

columnar/src/value.rs

@@ -1,3 +1,5 @@
+use common::DateTime;
+
 use crate::InvalidData;
 
 #[derive(Copy, Clone, PartialEq, Debug)]
@@ -104,10 +106,10 @@ impl Coerce for f64 {
     }
 }
 
-impl Coerce for crate::DateTime {
+impl Coerce for DateTime {
     fn coerce(value: NumericalValue) -> Self {
         let timestamp_micros = i64::coerce(value);
-        crate::DateTime { timestamp_micros }
+        DateTime::from_timestamp_micros(timestamp_micros)
     }
 }
 

common/Cargo.toml

@@ -16,6 +16,8 @@ repository = "https://github.com/quickwit-oss/tantivy"
 byteorder = "1.4.3"
 ownedbytes = { version= "0.5", path="../ownedbytes" }
 async-trait = "0.1"
+time = { version = "0.3.10", features = ["serde-well-known"] }
+serde = { version = "1.0.136", features = ["derive"] }
 
 [dev-dependencies]
 proptest = "1.0.0"

common/src/datetime.rs

@@ -0,0 +1,136 @@
+use std::fmt;
+
+use serde::{Deserialize, Serialize};
+use time::format_description::well_known::Rfc3339;
+use time::{OffsetDateTime, PrimitiveDateTime, UtcOffset};
+
+/// DateTime Precision
+#[derive(
+    Clone, Copy, Debug, Hash, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize, Default,
+)]
+#[serde(rename_all = "lowercase")]
+pub enum DatePrecision {
+    /// Seconds precision
+    #[default]
+    Seconds,
+    /// Milli-seconds precision.
+    Milliseconds,
+    /// Micro-seconds precision.
+    Microseconds,
+}
+
+/// A date/time value with microsecond precision.
+///
+/// This timestamp does not carry any explicit time zone information.
+/// Users are responsible for applying the provided conversion
+/// functions consistently. Internally the time zone is assumed
+/// to be UTC, which is also used implicitly for JSON serialization.
+///
+/// All constructors and conversions are provided as explicit
+/// functions and not by implementing any `From`/`Into` traits
+/// to prevent unintended usage.
+#[derive(Clone, Default, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
+pub struct DateTime {
+    // Timestamp in microseconds.
+    pub(crate) timestamp_micros: i64,
+}
+
+impl DateTime {
+    /// Create new from UNIX timestamp in seconds
+    pub const fn from_timestamp_secs(seconds: i64) -> Self {
+        Self {
+            timestamp_micros: seconds * 1_000_000,
+        }
+    }
+
+    /// Create new from UNIX timestamp in milliseconds
+    pub const fn from_timestamp_millis(milliseconds: i64) -> Self {
+        Self {
+            timestamp_micros: milliseconds * 1_000,
+        }
+    }
+
+    /// Create new from UNIX timestamp in microseconds.
+    pub const fn from_timestamp_micros(microseconds: i64) -> Self {
+        Self {
+            timestamp_micros: microseconds,
+        }
+    }
+
+    /// Create new from `OffsetDateTime`
+    ///
+    /// 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() * 1_000_000 + dt.microsecond() as i64;
+        Self { timestamp_micros }
+    }
+
+    /// Create new from `PrimitiveDateTime`
+    ///
+    /// Implicitly assumes that the given date/time is in UTC!
+    /// Otherwise the original value must only be reobtained with
+    /// [`Self::into_primitive()`].
+    pub fn from_primitive(dt: PrimitiveDateTime) -> Self {
+        Self::from_utc(dt.assume_utc())
+    }
+
+    /// Convert to UNIX timestamp in seconds.
+    pub const fn into_timestamp_secs(self) -> i64 {
+        self.timestamp_micros / 1_000_000
+    }
+
+    /// Convert to UNIX timestamp in milliseconds.
+    pub const fn into_timestamp_millis(self) -> i64 {
+        self.timestamp_micros / 1_000
+    }
+
+    /// Convert to UNIX timestamp in microseconds.
+    pub const fn into_timestamp_micros(self) -> i64 {
+        self.timestamp_micros
+    }
+
+    /// Convert to UTC `OffsetDateTime`
+    pub fn into_utc(self) -> OffsetDateTime {
+        let timestamp_nanos = self.timestamp_micros as i128 * 1000;
+        let utc_datetime = OffsetDateTime::from_unix_timestamp_nanos(timestamp_nanos)
+            .expect("valid UNIX timestamp");
+        debug_assert_eq!(UtcOffset::UTC, utc_datetime.offset());
+        utc_datetime
+    }
+
+    /// Convert to `OffsetDateTime` with the given time zone
+    pub fn into_offset(self, offset: UtcOffset) -> OffsetDateTime {
+        self.into_utc().to_offset(offset)
+    }
+
+    /// Convert to `PrimitiveDateTime` without any time zone
+    ///
+    /// The value should have been constructed with [`Self::from_primitive()`].
+    /// Otherwise the time zone is implicitly assumed to be UTC.
+    pub fn into_primitive(self) -> PrimitiveDateTime {
+        let utc_datetime = self.into_utc();
+        // Discard the UTC time zone offset
+        debug_assert_eq!(UtcOffset::UTC, utc_datetime.offset());
+        PrimitiveDateTime::new(utc_datetime.date(), utc_datetime.time())
+    }
+
+    /// Truncates the microseconds value to the corresponding precision.
+    pub fn truncate(self, precision: DatePrecision) -> Self {
+        let truncated_timestamp_micros = match precision {
+            DatePrecision::Seconds => (self.timestamp_micros / 1_000_000) * 1_000_000,
+            DatePrecision::Milliseconds => (self.timestamp_micros / 1_000) * 1_000,
+            DatePrecision::Microseconds => self.timestamp_micros,
+        };
+        Self {
+            timestamp_micros: truncated_timestamp_micros,
+        }
+    }
+}
+
+impl fmt::Debug for DateTime {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        let utc_rfc3339 = self.into_utc().format(&Rfc3339).map_err(|_| fmt::Error)?;
+        f.write_str(&utc_rfc3339)
+    }
+}

common/src/lib.rs

@@ -5,12 +5,14 @@ use std::ops::Deref;
 pub use byteorder::LittleEndian as Endianness;
 
 mod bitset;
+mod datetime;
 pub mod file_slice;
 mod group_by;
 mod serialize;
 mod vint;
 mod writer;
 pub use bitset::*;
+pub use datetime::{DatePrecision, DateTime};
 pub use group_by::GroupByIteratorExtended;
 pub use ownedbytes::{OwnedBytes, StableDeref};
 pub use serialize::{BinarySerializable, DeserializeFrom, FixedSize};
@@ -107,6 +109,21 @@ pub fn u64_to_f64(val: u64) -> f64 {
     })
 }
 
+/// Replaces a given byte in the `bytes` slice of bytes.
+///
+/// This function assumes that the needle is rarely contained in the bytes string
+/// and offers a fast path if the needle is not present.
+pub fn replace_in_place(needle: u8, replacement: u8, bytes: &mut [u8]) {
+    if !bytes.contains(&needle) {
+        return;
+    }
+    for b in bytes {
+        if *b == needle {
+            *b = replacement;
+        }
+    }
+}
+
 #[cfg(test)]
 pub mod test {
 
@@ -171,4 +188,20 @@ pub mod test {
         assert!(f64_to_u64(-2.0) < f64_to_u64(1.0));
         assert!(f64_to_u64(-2.0) < f64_to_u64(-1.5));
     }
+
+    #[test]
+    fn test_replace_in_place() {
+        let test_aux = |before_replacement: &[u8], expected: &[u8]| {
+            let mut bytes: Vec<u8> = before_replacement.to_vec();
+            super::replace_in_place(b'b', b'c', &mut bytes);
+            assert_eq!(&bytes[..], expected);
+        };
+        test_aux(b"", b"");
+        test_aux(b"b", b"c");
+        test_aux(b"baaa", b"caaa");
+        test_aux(b"aaab", b"aaac");
+        test_aux(b"aaabaa", b"aaacaa");
+        test_aux(b"aaaaaa", b"aaaaaa");
+        test_aux(b"bbbb", b"cccc");
+    }
 }

examples-disabled/aggregation.rs

@@ -1,130 +0,0 @@
-// # Aggregation example
-//
-// This example shows how you can use built-in aggregations.
-// We will use range buckets and compute the average in each bucket.
-//
-
-use serde_json::Value;
-use tantivy::aggregation::agg_req::{
-    Aggregation, Aggregations, BucketAggregation, BucketAggregationType, MetricAggregation,
-    RangeAggregation,
-};
-use tantivy::aggregation::agg_result::AggregationResults;
-use tantivy::aggregation::metric::AverageAggregation;
-use tantivy::aggregation::AggregationCollector;
-use tantivy::query::TermQuery;
-use tantivy::schema::{self, IndexRecordOption, Schema, TextFieldIndexing};
-use tantivy::{doc, Index, Term};
-
-fn main() -> tantivy::Result<()> {
-    let mut schema_builder = Schema::builder();
-    let text_fieldtype = schema::TextOptions::default()
-        .set_indexing_options(
-            TextFieldIndexing::default().set_index_option(IndexRecordOption::WithFreqs),
-        )
-        .set_stored();
-    let text_field = schema_builder.add_text_field("text", text_fieldtype);
-    let score_fieldtype =
-        crate::schema::NumericOptions::default().set_fast();
-    let highscore_field = schema_builder.add_f64_field("highscore", score_fieldtype.clone());
-    let price_field = schema_builder.add_f64_field("price", score_fieldtype);
-
-    let schema = schema_builder.build();
-
-    // # Indexing documents
-    //
-    // Lets index a bunch of documents for this example.
-    let index = Index::create_in_ram(schema);
-
-    let mut index_writer = index.writer(50_000_000)?;
-    // writing the segment
-    index_writer.add_document(doc!(
-        text_field => "cool",
-        highscore_field => 1f64,
-        price_field => 0f64,
-    ))?;
-    index_writer.add_document(doc!(
-        text_field => "cool",
-        highscore_field => 3f64,
-        price_field => 1f64,
-    ))?;
-    index_writer.add_document(doc!(
-        text_field => "cool",
-        highscore_field => 5f64,
-        price_field => 1f64,
-    ))?;
-    index_writer.add_document(doc!(
-        text_field => "nohit",
-        highscore_field => 6f64,
-        price_field => 2f64,
-    ))?;
-    index_writer.add_document(doc!(
-        text_field => "cool",
-        highscore_field => 7f64,
-        price_field => 2f64,
-    ))?;
-    index_writer.commit()?;
-    index_writer.add_document(doc!(
-        text_field => "cool",
-        highscore_field => 11f64,
-        price_field => 10f64,
-    ))?;
-    index_writer.add_document(doc!(
-        text_field => "cool",
-        highscore_field => 14f64,
-        price_field => 15f64,
-    ))?;
-
-    index_writer.add_document(doc!(
-        text_field => "cool",
-        highscore_field => 15f64,
-        price_field => 20f64,
-    ))?;
-
-    index_writer.commit()?;
-
-    let reader = index.reader()?;
-    let text_field = reader.searcher().schema().get_field("text").unwrap();
-
-    let term_query = TermQuery::new(
-        Term::from_field_text(text_field, "cool"),
-        IndexRecordOption::Basic,
-    );
-
-    let sub_agg_req_1: Aggregations = vec![(
-        "average_price".to_string(),
-        Aggregation::Metric(MetricAggregation::Average(
-            AverageAggregation::from_field_name("price".to_string()),
-        )),
-    )]
-    .into_iter()
-    .collect();
-
-    let agg_req_1: Aggregations = vec![(
-        "score_ranges".to_string(),
-        Aggregation::Bucket(BucketAggregation {
-            bucket_agg: BucketAggregationType::Range(RangeAggregation {
-                field: "highscore".to_string(),
-                ranges: vec![
-                    (-1f64..9f64).into(),
-                    (9f64..14f64).into(),
-                    (14f64..20f64).into(),
-                ],
-                ..Default::default()
-            }),
-            sub_aggregation: sub_agg_req_1,
-        }),
-    )]
-    .into_iter()
-    .collect();
-
-    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();
-
-    let res: Value = serde_json::to_value(agg_res)?;
-    println!("{}", serde_json::to_string_pretty(&res)?);
-
-    Ok(())
-}

examples-disabled/ip_field.rs

@@ -1,73 +0,0 @@
-// # IP Address example
-//
-// This example shows how the ip field can be used
-// with IpV6 and IpV4.
-
-use tantivy::collector::{Count, TopDocs};
-use tantivy::query::QueryParser;
-use tantivy::schema::{Schema, FAST, INDEXED, STORED, STRING};
-use tantivy::Index;
-
-fn main() -> tantivy::Result<()> {
-    // # Defining the schema
-    let mut schema_builder = Schema::builder();
-    let event_type = schema_builder.add_text_field("event_type", STRING | STORED);
-    let ip = schema_builder.add_ip_addr_field("ip", STORED | INDEXED | FAST);
-    let schema = schema_builder.build();
-
-    // # Indexing documents
-    let index = Index::create_in_ram(schema.clone());
-
-    let mut index_writer = index.writer(50_000_000)?;
-    let doc = schema.parse_document(
-        r#"{
-        "ip": "192.168.0.33",
-        "event_type": "login"
-    }"#,
-    )?;
-    index_writer.add_document(doc)?;
-    let doc = schema.parse_document(
-        r#"{
-        "ip": "192.168.0.80",
-        "event_type": "checkout"
-    }"#,
-    )?;
-    index_writer.add_document(doc)?;
-    let doc = schema.parse_document(
-        r#"{
-        "ip": "2001:0db8:85a3:0000:0000:8a2e:0370:7334",
-        "event_type": "checkout"
-    }"#,
-    )?;
-
-    index_writer.add_document(doc)?;
-    index_writer.commit()?;
-
-    let reader = index.reader()?;
-    let searcher = reader.searcher();
-
-    let query_parser = QueryParser::for_index(&index, vec![event_type, ip]);
-    {
-        let query = query_parser.parse_query("ip:[192.168.0.0 TO 192.168.0.100]")?;
-        let count_docs = searcher.search(&*query, &TopDocs::with_limit(5))?;
-        assert_eq!(count_docs.len(), 2);
-    }
-    {
-        let query = query_parser.parse_query("ip:[192.168.1.0 TO 192.168.1.100]")?;
-        let count_docs = searcher.search(&*query, &TopDocs::with_limit(2))?;
-        assert_eq!(count_docs.len(), 0);
-    }
-    {
-        let query = query_parser.parse_query("ip:192.168.0.80")?;
-        let count_docs = searcher.search(&*query, &Count)?;
-        assert_eq!(count_docs, 1);
-    }
-    {
-        // IpV6 needs to be escaped because it contains `:`
-        let query = query_parser.parse_query("ip:\"2001:0db8:85a3:0000:0000:8a2e:0370:7334\"")?;
-        let count_docs = searcher.search(&*query, &Count)?;
-        assert_eq!(count_docs, 1);
-    }
-
-    Ok(())
-}

examples/aggregation.rs

@@ -0,0 +1,319 @@
+// # Aggregation example
+//
+// This example shows how you can use built-in aggregations.
+// We will use nested aggregations with buckets and metrics:
+// - Range buckets and compute the average in each bucket.
+// - Term aggregation and compute the min price in each bucket
+// ---
+
+use serde_json::{Deserializer, Value};
+use tantivy::aggregation::agg_req::{
+    Aggregation, Aggregations, BucketAggregation, BucketAggregationType, MetricAggregation,
+    RangeAggregation,
+};
+use tantivy::aggregation::agg_result::AggregationResults;
+use tantivy::aggregation::bucket::RangeAggregationRange;
+use tantivy::aggregation::metric::AverageAggregation;
+use tantivy::aggregation::AggregationCollector;
+use tantivy::query::AllQuery;
+use tantivy::schema::{self, IndexRecordOption, Schema, TextFieldIndexing, FAST};
+use tantivy::Index;
+
+fn main() -> tantivy::Result<()> {
+    // # Create Schema
+    //
+    // Lets create a schema for a footwear shop, with 4 fields: name, category, stock and price.
+    // category, stock and price will be fast fields as that's the requirement
+    // for aggregation queries.
+    //
+
+    let mut schema_builder = Schema::builder();
+    // In preparation of the `TermsAggregation`, the category field is configured with:
+    // - `set_fast`
+    // - `raw` tokenizer
+    //
+    // The tokenizer is set to "raw", because the fast field uses the same dictionary as the
+    // inverted index. (This behaviour will change in tantivy 0.20, where the fast field will
+    // always be raw tokenized independent from the regular tokenizing)
+    //
+    let text_fieldtype = schema::TextOptions::default()
+        .set_indexing_options(
+            TextFieldIndexing::default()
+                .set_index_option(IndexRecordOption::WithFreqs)
+                .set_tokenizer("raw"),
+        )
+        .set_fast()
+        .set_stored();
+    schema_builder.add_text_field("category", text_fieldtype);
+    schema_builder.add_f64_field("stock", FAST);
+    schema_builder.add_f64_field("price", FAST);
+
+    let schema = schema_builder.build();
+
+    // # Indexing documents
+    //
+    // Lets index a bunch of documents for this example.
+    let index = Index::create_in_ram(schema.clone());
+
+    let data = r#"{
+        "name": "Almond Toe Court Shoes, Patent Black",
+        "category": "Womens Footwear",
+        "price": 99.00,
+        "stock": 5
+    }
+    {
+        "name": "Suede Shoes, Blue",
+        "category": "Womens Footwear",
+        "price": 42.00,
+        "stock": 4
+    }
+    {
+        "name": "Leather Driver Saddle Loafers, Tan",
+        "category": "Mens Footwear",
+        "price": 34.00,
+        "stock": 12
+    }
+    {
+        "name": "Flip Flops, Red",
+        "category": "Mens Footwear",
+        "price": 19.00,
+        "stock": 6
+    }
+    {
+        "name": "Flip Flops, Blue",
+        "category": "Mens Footwear",
+        "price": 19.00,
+        "stock": 0
+    }
+    {
+        "name": "Gold Button Cardigan, Black",
+        "category": "Womens Casualwear",
+        "price": 167.00,
+        "stock": 6
+    }
+    {
+        "name": "Cotton Shorts, Medium Red",
+        "category": "Womens Casualwear",
+        "price": 30.00,
+        "stock": 5
+    }
+    {
+        "name": "Fine Stripe Short SleeveShirt, Grey",
+        "category": "Mens Casualwear",
+        "price": 49.99,
+        "stock": 9
+    }
+    {
+        "name": "Fine Stripe Short SleeveShirt, Green",
+        "category": "Mens Casualwear",
+        "price": 49.99,
+        "offer": 39.99,
+        "stock": 9
+    }
+    {
+        "name": "Sharkskin Waistcoat, Charcoal",
+        "category": "Mens Formalwear",
+        "price": 75.00,
+        "stock": 2
+    }
+    {
+        "name": "Lightweight Patch PocketBlazer, Deer",
+        "category": "Mens Formalwear",
+        "price": 175.50,
+        "stock": 1
+    }
+    {
+        "name": "Bird Print Dress, Black",
+        "category": "Womens Formalwear",
+        "price": 270.00,
+        "stock": 10
+    }
+    {
+        "name": "Mid Twist Cut-Out Dress, Pink",
+        "category": "Womens Formalwear",
+        "price": 540.00,
+        "stock": 5
+    }"#;
+
+    let stream = Deserializer::from_str(data).into_iter::<Value>();
+
+    let mut index_writer = index.writer(50_000_000)?;
+    let mut num_indexed = 0;
+    for value in stream {
+        let doc = schema.parse_document(&serde_json::to_string(&value.unwrap())?)?;
+        index_writer.add_document(doc)?;
+        num_indexed += 1;
+        if num_indexed > 4 {
+            // Writing the first segment
+            index_writer.commit()?;
+        }
+    }
+
+    // Writing the second segment
+    index_writer.commit()?;
+
+    // We have two segments now. The `AggregationCollector` will run the aggregation on each
+    // segment and then merge the results into an `IntermediateAggregationResult`.
+
+    let reader = index.reader()?;
+    let searcher = reader.searcher();
+    // ---
+    // # Aggregation Query
+    //
+    //
+    // We can construct the query by building the request structure or by deserializing from JSON.
+    // The JSON API is more stable and therefore recommended.
+    //
+    // ## Request 1
+
+    let agg_req_str = r#"
+    {
+      "group_by_stock": {
+        "aggs": {
+          "average_price": { "avg": { "field": "price" } }
+        },
+        "range": {
+          "field": "stock",
+          "ranges": [
+            { "key": "few", "to": 1.0 },
+            { "key": "some", "from": 1.0, "to": 10.0 },
+            { "key": "many", "from": 10.0 }
+          ]
+        }
+      }
+    } "#;
+
+    // In this Aggregation we want to get the average price for different groups, depending on how
+    // many items are in stock. We define custom ranges `few`, `some`, `many` via the
+    // range aggregation.
+    // For every bucket we want the average price, so we create a nested metric aggregation on the
+    // range bucket aggregation. Only buckets support nested aggregations.
+    // ### Request JSON API
+    //
+
+    let agg_req: Aggregations = serde_json::from_str(agg_req_str)?;
+    let collector = AggregationCollector::from_aggs(agg_req, None);
+
+    let agg_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
+    let res2: Value = serde_json::to_value(agg_res)?;
+
+    // ### Request Rust API
+    //
+    // This is exactly the same request as above, but via the rust structures.
+    //
+
+    let agg_req: Aggregations = vec![(
+        "group_by_stock".to_string(),
+        Aggregation::Bucket(BucketAggregation {
+            bucket_agg: BucketAggregationType::Range(RangeAggregation {
+                field: "stock".to_string(),
+                ranges: vec![
+                    RangeAggregationRange {
+                        key: Some("few".into()),
+                        from: None,
+                        to: Some(1f64),
+                    },
+                    RangeAggregationRange {
+                        key: Some("some".into()),
+                        from: Some(1f64),
+                        to: Some(10f64),
+                    },
+                    RangeAggregationRange {
+                        key: Some("many".into()),
+                        from: Some(10f64),
+                        to: None,
+                    },
+                ],
+                ..Default::default()
+            }),
+            sub_aggregation: vec![(
+                "average_price".to_string(),
+                Aggregation::Metric(MetricAggregation::Average(
+                    AverageAggregation::from_field_name("price".to_string()),
+                )),
+            )]
+            .into_iter()
+            .collect(),
+        }),
+    )]
+    .into_iter()
+    .collect();
+
+    let collector = AggregationCollector::from_aggs(agg_req, None);
+    // We use the `AllQuery` which will pass all documents to the AggregationCollector.
+    let agg_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
+
+    let res1: Value = serde_json::to_value(agg_res)?;
+
+    // ### Aggregation Result
+    //
+    // The resulting structure deserializes in the same JSON format as elastic search.
+    //
+    let expected_res = r#"
+    {
+        "group_by_stock":{
+            "buckets":[
+                {"average_price":{"value":19.0},"doc_count":1,"key":"few","to":1.0},
+                {"average_price":{"value":124.748},"doc_count":10,"from":1.0,"key":"some","to":10.0},
+                {"average_price":{"value":152.0},"doc_count":2,"from":10.0,"key":"many"}
+            ]
+        }
+    }
+    "#;
+    let expected_json: Value = serde_json::from_str(expected_res)?;
+    assert_eq!(expected_json, res1);
+    assert_eq!(expected_json, res2);
+
+    // ### Request 2
+    //
+    // Now we are interested in the minimum price per category, so we create a bucket per
+    // category via `TermsAggregation`. We are interested in the highest minimum prices, and set the
+    // order of the buckets `"order": { "min_price": "desc" }` to be sorted by the the metric of
+    // the sub aggregation. (awesome)
+    //
+    let agg_req_str = r#"
+    {
+      "min_price_per_category": {
+        "aggs": {
+          "min_price": { "min": { "field": "price" } }
+        },
+        "terms": {
+          "field": "category",
+          "min_doc_count": 1,
+          "order": { "min_price": "desc" }
+        }
+      }
+    } "#;
+
+    let agg_req: Aggregations = serde_json::from_str(agg_req_str)?;
+
+    let collector = AggregationCollector::from_aggs(agg_req, None);
+
+    let agg_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
+    let res: Value = serde_json::to_value(agg_res)?;
+
+    // Minimum price per category, sorted by minimum price descending
+    //
+    // As you can see, the starting prices for `Formalwear` are higher than `Casualwear`.
+    //
+    let expected_res = r#"
+    {
+      "min_price_per_category": {
+        "buckets": [
+          { "doc_count": 2, "key": "Womens Formalwear", "min_price": { "value": 270.0 } },
+          { "doc_count": 2, "key": "Mens Formalwear", "min_price": { "value": 75.0 } },
+          { "doc_count": 2, "key": "Mens Casualwear", "min_price": { "value": 49.99 } },
+          { "doc_count": 2, "key": "Womens Footwear", "min_price": { "value": 42.0 } },
+          { "doc_count": 2, "key": "Womens Casualwear", "min_price": { "value": 30.0 } },
+          { "doc_count": 3, "key": "Mens Footwear", "min_price": { "value": 19.0 } }
+        ],
+        "sum_other_doc_count": 0
+      }
+    }
+    "#;
+    let expected_json: Value = serde_json::from_str(expected_res)?;
+
+    assert_eq!(expected_json, res);
+
+    Ok(())
+}

examples/custom_collector.rs

@@ -7,9 +7,7 @@
 // Of course, you can have a look at the tantivy's built-in collectors
 // such as the `CountCollector` for more examples.
 
-use std::sync::Arc;
-
-use fastfield_codecs::Column;
+use columnar::Column;
 // ---
 // Importing tantivy...
 use tantivy::collector::{Collector, SegmentCollector};
@@ -97,7 +95,7 @@ impl Collector for StatsCollector {
 }
 
 struct StatsSegmentCollector {
-    fast_field_reader: Arc<dyn Column<u64>>,
+    fast_field_reader: Column,
     stats: Stats,
 }
 
@@ -105,10 +103,14 @@ impl SegmentCollector for StatsSegmentCollector {
     type Fruit = Option<Stats>;
 
     fn collect(&mut self, doc: u32, _score: Score) {
-        let value = self.fast_field_reader.get_val(doc) as f64;
-        self.stats.count += 1;
-        self.stats.sum += value;
-        self.stats.squared_sum += value * value;
+        // Since we know the values are single value, we could call `first_or_default_col` on the
+        // column and fetch single values.
+        for value in self.fast_field_reader.values_for_doc(doc) {
+            let value = value as f64;
+            self.stats.count += 1;
+            self.stats.sum += value;
+            self.stats.squared_sum += value * value;
+        }
     }
 
     fn harvest(self) -> <Self as SegmentCollector>::Fruit {
@@ -169,7 +171,7 @@ fn main() -> tantivy::Result<()> {
     let searcher = reader.searcher();
     let query_parser = QueryParser::for_index(&index, vec![product_name, product_description]);
 
-    // here we want to get a hit on the 'ken' in Frankenstein
+    // here we want to search for `broom` and use `StatsCollector` on the hits.
     let query = query_parser.parse_query("broom")?;
     if let Some(stats) =
         searcher.search(&query, &StatsCollector::with_field("price".to_string()))?

examples/custom_tokenizer.rs

@@ -1,7 +1,7 @@
 // # Defining a tokenizer pipeline
 //
-// In this example, we'll see how to define a tokenizer pipeline
-// by aligning a bunch of `TokenFilter`.
+// In this example, we'll see how to define a tokenizer
+// by creating a custom `NgramTokenizer`.
 use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
 use tantivy::schema::*;

examples/date_time_field.rs

@@ -14,6 +14,7 @@ fn main() -> tantivy::Result<()> {
         .set_stored()
         .set_fast()
         .set_precision(tantivy::DatePrecision::Seconds);
+    // Add `occurred_at` date field type
     let occurred_at = schema_builder.add_date_field("occurred_at", opts);
     let event_type = schema_builder.add_text_field("event", STRING | STORED);
     let schema = schema_builder.build();
@@ -22,6 +23,7 @@ fn main() -> tantivy::Result<()> {
     let index = Index::create_in_ram(schema.clone());
 
     let mut index_writer = index.writer(50_000_000)?;
+    // The dates are passed as string in the RFC3339 format
     let doc = schema.parse_document(
         r#"{
         "occurred_at": "2022-06-22T12:53:50.53Z",
@@ -41,14 +43,16 @@ fn main() -> tantivy::Result<()> {
     let reader = index.reader()?;
     let searcher = reader.searcher();
 
-    // # Default fields: event_type
+    // # Search
     let query_parser = QueryParser::for_index(&index, vec![event_type]);
     {
-        let query = query_parser.parse_query("event:comment")?;
+        // Simple exact search on the date
+        let query = query_parser.parse_query("occurred_at:\"2022-06-22T12:53:50.53Z\"")?;
         let count_docs = searcher.search(&*query, &TopDocs::with_limit(5))?;
         assert_eq!(count_docs.len(), 1);
     }
     {
+        // Range query on the date field
         let query = query_parser
             .parse_query(r#"occurred_at:[2022-06-22T12:58:00Z TO 2022-06-23T00:00:00Z}"#)?;
         let count_docs = searcher.search(&*query, &TopDocs::with_limit(4))?;

examples/faceted_search.rs

@@ -71,7 +71,7 @@ fn main() -> tantivy::Result<()> {
     let reader = index.reader()?;
     let searcher = reader.searcher();
     {
-        let mut facet_collector = FacetCollector::for_field(classification);
+        let mut facet_collector = FacetCollector::for_field("classification");
         facet_collector.add_facet("/Felidae");
         let facet_counts = searcher.search(&AllQuery, &facet_collector)?;
         // This lists all of the facet counts, right below "/Felidae".
@@ -97,7 +97,7 @@ fn main() -> tantivy::Result<()> {
         let facet = Facet::from("/Felidae/Pantherinae");
         let facet_term = Term::from_facet(classification, &facet);
         let facet_term_query = TermQuery::new(facet_term, IndexRecordOption::Basic);
-        let mut facet_collector = FacetCollector::for_field(classification);
+        let mut facet_collector = FacetCollector::for_field("classification");
         facet_collector.add_facet("/Felidae/Pantherinae");
         let facet_counts = searcher.search(&facet_term_query, &facet_collector)?;
         let facets: Vec<(&Facet, u64)> = facet_counts.get("/Felidae/Pantherinae").collect();

examples/faceted_search_with_tweaked_score.rs

@@ -1,3 +1,12 @@
+// # Faceted Search With Tweak Score
+//
+// This example covers the faceted search functionalities of
+// tantivy.
+//
+// We will :
+// - define a text field "name" in our schema
+// - define a facet field "classification" in our schema
+
 use std::collections::HashSet;
 
 use tantivy::collector::TopDocs;
@@ -55,8 +64,9 @@ fn main() -> tantivy::Result<()> {
                 .collect(),
         );
         let top_docs_by_custom_score =
+            // Call TopDocs with a custom tweak score
             TopDocs::with_limit(2).tweak_score(move |segment_reader: &SegmentReader| {
-                let ingredient_reader = segment_reader.facet_reader(ingredient).unwrap();
+                let ingredient_reader = segment_reader.facet_reader("ingredient").unwrap();
                 let facet_dict = ingredient_reader.facet_dict();
 
                 let query_ords: HashSet<u64> = facets
@@ -64,12 +74,10 @@ fn main() -> tantivy::Result<()> {
                     .filter_map(|key| facet_dict.term_ord(key.encoded_str()).unwrap())
                     .collect();
 
-                let mut facet_ords_buffer: Vec<u64> = Vec::with_capacity(20);
-
                 move |doc: DocId, original_score: Score| {
-                    ingredient_reader.facet_ords(doc, &mut facet_ords_buffer);
-                    let missing_ingredients = facet_ords_buffer
-                        .iter()
+                    // Update the original score with a tweaked score
+                    let missing_ingredients = ingredient_reader
+                        .facet_ords(doc)
                         .filter(|ord| !query_ords.contains(ord))
                         .count();
                     let tweak = 1.0 / 4_f32.powi(missing_ingredients as i32);

examples/fuzzy_search.rs

@@ -0,0 +1,167 @@
+// # Basic Example
+//
+// This example covers the basic functionalities of
+// tantivy.
+//
+// We will :
+// - define our schema
+// - create an index in a directory
+// - index a few documents into our index
+// - search for the best document matching a basic query
+// - retrieve the best document's original content.
+// ---
+// Importing tantivy...
+use tantivy::collector::{Count, TopDocs};
+use tantivy::query::FuzzyTermQuery;
+use tantivy::schema::*;
+use tantivy::{doc, Index, ReloadPolicy};
+use tempfile::TempDir;
+
+fn main() -> tantivy::Result<()> {
+    // Let's create a temporary directory for the
+    // sake of this example
+    let index_path = TempDir::new()?;
+
+    // # Defining the schema
+    //
+    // The Tantivy index requires a very strict schema.
+    // The schema declares which fields are in the index,
+    // and for each field, its type and "the way it should
+    // be indexed".
+
+    // First we need to define a schema ...
+    let mut schema_builder = Schema::builder();
+
+    // Our first field is title.
+    // We want full-text search for it, and we also want
+    // to be able to retrieve the document after the search.
+    //
+    // `TEXT | STORED` is some syntactic sugar to describe
+    // that.
+    //
+    // `TEXT` means the field should be tokenized and indexed,
+    // along with its term frequency and term positions.
+    //
+    // `STORED` means that the field will also be saved
+    // in a compressed, row-oriented key-value store.
+    // This store is useful for reconstructing the
+    // documents that were selected during the search phase.
+    let title = schema_builder.add_text_field("title", TEXT | STORED);
+
+    let schema = schema_builder.build();
+
+    // # Indexing documents
+    //
+    // Let's create a brand new index.
+    //
+    // This will actually just save a meta.json
+    // with our schema in the directory.
+    let index = Index::create_in_dir(&index_path, schema.clone())?;
+
+    // To insert a document we will need an index writer.
+    // There must be only one writer at a time.
+    // This single `IndexWriter` is already
+    // multithreaded.
+    //
+    // Here we give tantivy a budget of `50MB`.
+    // Using a bigger memory_arena for the indexer may increase
+    // throughput, but 50 MB is already plenty.
+    let mut index_writer = index.writer(50_000_000)?;
+
+    // Let's index our documents!
+    // We first need a handle on the title and the body field.
+
+    // ### Adding documents
+    //
+    index_writer.add_document(doc!(
+        title => "The Name of the Wind",
+    ))?;
+    index_writer.add_document(doc!(
+        title => "The Diary of Muadib",
+    ))?;
+    index_writer.add_document(doc!(
+        title => "A Dairy Cow",
+    ))?;
+    index_writer.add_document(doc!(
+        title => "The Diary of a Young Girl",
+    ))?;
+    index_writer.commit()?;
+
+    // ### Committing
+    //
+    // At this point our documents are not searchable.
+    //
+    //
+    // We need to call `.commit()` explicitly to force the
+    // `index_writer` to finish processing the documents in the queue,
+    // flush the current index to the disk, and advertise
+    // the existence of new documents.
+    //
+    // This call is blocking.
+    index_writer.commit()?;
+
+    // If `.commit()` returns correctly, then all of the
+    // documents that have been added are guaranteed to be
+    // persistently indexed.
+    //
+    // In the scenario of a crash or a power failure,
+    // tantivy behaves as if it has rolled back to its last
+    // commit.
+
+    // # Searching
+    //
+    // ### Searcher
+    //
+    // A reader is required first in order to search an index.
+    // It acts as a `Searcher` pool that reloads itself,
+    // depending on a `ReloadPolicy`.
+    //
+    // For a search server you will typically create one reader for the entire lifetime of your
+    // program, and acquire a new searcher for every single request.
+    //
+    // In the code below, we rely on the 'ON_COMMIT' policy: the reader
+    // will reload the index automatically after each commit.
+    let reader = index
+        .reader_builder()
+        .reload_policy(ReloadPolicy::OnCommit)
+        .try_into()?;
+
+    // We now need to acquire a searcher.
+    //
+    // A searcher points to a snapshotted, immutable version of the index.
+    //
+    // Some search experience might require more than
+    // one query. Using the same searcher ensures that all of these queries will run on the
+    // same version of the index.
+    //
+    // Acquiring a `searcher` is very cheap.
+    //
+    // You should acquire a searcher every time you start processing a request and
+    // and release it right after your query is finished.
+    let searcher = reader.searcher();
+
+    // ### FuzzyTermQuery
+    {
+        let term = Term::from_field_text(title, "Diary");
+        let query = FuzzyTermQuery::new(term, 2, true);
+
+        let (top_docs, count) = searcher
+            .search(&query, &(TopDocs::with_limit(5), Count))
+            .unwrap();
+        assert_eq!(count, 3);
+        assert_eq!(top_docs.len(), 3);
+        for (score, doc_address) in top_docs {
+            let retrieved_doc = searcher.doc(doc_address)?;
+            // Note that the score is not lower for the fuzzy hit.
+            // There's an issue open for that: https://github.com/quickwit-oss/tantivy/issues/563
+            println!("score {score:?} doc {}", schema.to_json(&retrieved_doc));
+            // score 1.0 doc {"title":["The Diary of Muadib"]}
+            //
+            // score 1.0 doc {"title":["The Diary of a Young Girl"]}
+            //
+            // score 1.0 doc {"title":["A Dairy Cow"]}
+        }
+    }
+
+    Ok(())
+}

examples/ip_field.rs

@@ -0,0 +1,107 @@
+// # IP Address example
+//
+// This example shows how the ip field can be used
+// with IpV6 and IpV4.
+
+use tantivy::collector::{Count, TopDocs};
+use tantivy::query::QueryParser;
+use tantivy::schema::{Schema, FAST, INDEXED, STORED, STRING};
+use tantivy::Index;
+
+fn main() -> tantivy::Result<()> {
+    // # Defining the schema
+    // We set the IP field as `INDEXED`, so it can be searched
+    // `FAST` will create a fast field. The fast field will be used to execute search queries.
+    // `FAST` is not a requirement for range queries, it can also be executed on the inverted index
+    // which is created by `INDEXED`.
+    let mut schema_builder = Schema::builder();
+    let event_type = schema_builder.add_text_field("event_type", STRING | STORED);
+    let ip = schema_builder.add_ip_addr_field("ip", STORED | INDEXED | FAST);
+    let schema = schema_builder.build();
+
+    // # Indexing documents
+    let index = Index::create_in_ram(schema.clone());
+
+    let mut index_writer = index.writer(50_000_000)?;
+
+    // ### IPv4
+    // Adding documents that contain an IPv4 address. Notice that the IP addresses are passed as
+    // `String`. Since the field is of type ip, we parse the IP address from the string and store it
+    // internally as IPv6.
+    let doc = schema.parse_document(
+        r#"{
+            "ip": "192.168.0.33",
+            "event_type": "login"
+        }"#,
+    )?;
+    index_writer.add_document(doc)?;
+    let doc = schema.parse_document(
+        r#"{
+            "ip": "192.168.0.80",
+            "event_type": "checkout"
+        }"#,
+    )?;
+    index_writer.add_document(doc)?;
+    // ### IPv6
+    // Adding a document that contains an IPv6 address.
+    let doc = schema.parse_document(
+        r#"{
+            "ip": "2001:0db8:85a3:0000:0000:8a2e:0370:7334",
+            "event_type": "checkout"
+        }"#,
+    )?;
+
+    index_writer.add_document(doc)?;
+    // Commit will create a segment containing our documents.
+    index_writer.commit()?;
+
+    let reader = index.reader()?;
+    let searcher = reader.searcher();
+
+    // # Search
+    // Range queries on IPv4. Since we created a fast field, the fast field will be used to execute
+    // the search.
+    // ### Range Queries
+    let query_parser = QueryParser::for_index(&index, vec![event_type, ip]);
+    {
+        // Inclusive range queries
+        let query = query_parser.parse_query("ip:[192.168.0.80 TO 192.168.0.100]")?;
+        let count_docs = searcher.search(&*query, &TopDocs::with_limit(5))?;
+        assert_eq!(count_docs.len(), 1);
+    }
+    {
+        // Exclusive range queries
+        let query = query_parser.parse_query("ip:{192.168.0.80 TO 192.168.1.100]")?;
+        let count_docs = searcher.search(&*query, &TopDocs::with_limit(2))?;
+        assert_eq!(count_docs.len(), 0);
+    }
+    {
+        // Find docs with IP addresses smaller equal 192.168.1.100
+        let query = query_parser.parse_query("ip:[* TO 192.168.1.100]")?;
+        let count_docs = searcher.search(&*query, &TopDocs::with_limit(2))?;
+        assert_eq!(count_docs.len(), 2);
+    }
+    {
+        // Find docs with IP addresses smaller than 192.168.1.100
+        let query = query_parser.parse_query("ip:[* TO 192.168.1.100}")?;
+        let count_docs = searcher.search(&*query, &TopDocs::with_limit(2))?;
+        assert_eq!(count_docs.len(), 2);
+    }
+
+    // ### Exact Queries
+    // Exact search on IPv4.
+    {
+        let query = query_parser.parse_query("ip:192.168.0.80")?;
+        let count_docs = searcher.search(&*query, &Count)?;
+        assert_eq!(count_docs, 1);
+    }
+    // Exact search on IPv6.
+    // IpV6 addresses need to be quoted because they contain `:`
+    {
+        let query = query_parser.parse_query("ip:\"2001:0db8:85a3:0000:0000:8a2e:0370:7334\"")?;
+        let count_docs = searcher.search(&*query, &Count)?;
+        assert_eq!(count_docs, 1);
+    }
+
+    Ok(())
+}

examples/warmer.rs

@@ -17,7 +17,6 @@ use tantivy::{
 
 type ProductId = u64;
 
-/// Price
 type Price = u32;
 
 pub trait PriceFetcher: Send + Sync + 'static {
@@ -48,7 +47,10 @@ impl Warmer for DynamicPriceColumn {
     fn warm(&self, searcher: &Searcher) -> tantivy::Result<()> {
         for segment in searcher.segment_readers() {
             let key = (segment.segment_id(), segment.delete_opstamp());
-            let product_id_reader = segment.fast_fields().u64(&self.field)?;
+            let product_id_reader = segment
+                .fast_fields()
+                .u64(&self.field)?
+                .first_or_default_col(0);
             let product_ids: Vec<ProductId> = segment
                 .doc_ids_alive()
                 .map(|doc| product_id_reader.get_val(doc))
@@ -87,10 +89,10 @@ impl Warmer for DynamicPriceColumn {
     }
 }
 
-/// For the sake of this example, the table is just an editable HashMap behind a RwLock.
-/// This map represents a map (ProductId -> Price)
-///
-/// In practise, it could be fetching things from an external service, like a SQL table.
+// For the sake of this example, the table is just an editable HashMap behind a RwLock.
+// This map represents a map (ProductId -> Price)
+//
+// In practise, it could be fetching things from an external service, like a SQL table.
 #[derive(Default, Clone)]
 pub struct ExternalPriceTable {
     prices: Arc<RwLock<HashMap<ProductId, Price>>>,

src/aggregation/agg_req.rs

@@ -50,7 +50,7 @@ use std::collections::{HashMap, HashSet};
 use serde::{Deserialize, Serialize};
 
 pub use super::bucket::RangeAggregation;
-use super::bucket::{HistogramAggregation, TermsAggregation};
+use super::bucket::{DateHistogramAggregationReq, HistogramAggregation, TermsAggregation};
 use super::metric::{
     AverageAggregation, CountAggregation, MaxAggregation, MinAggregation, StatsAggregation,
     SumAggregation,
@@ -110,10 +110,13 @@ impl BucketAggregationInternal {
             _ => None,
         }
     }
-    pub(crate) fn as_histogram(&self) -> Option<&HistogramAggregation> {
+    pub(crate) fn as_histogram(&self) -> crate::Result<Option<HistogramAggregation>> {
         match &self.bucket_agg {
-            BucketAggregationType::Histogram(histogram) => Some(histogram),
-            _ => None,
+            BucketAggregationType::Histogram(histogram) => Ok(Some(histogram.clone())),
+            BucketAggregationType::DateHistogram(histogram) => {
+                Ok(Some(histogram.to_histogram_req()?))
+            }
+            _ => Ok(None),
         }
     }
     pub(crate) fn as_term(&self) -> Option<&TermsAggregation> {
@@ -124,15 +127,6 @@ impl BucketAggregationInternal {
     }
 }
 
-/// Extract all fields, where the term directory is used in the tree.
-pub fn get_term_dict_field_names(aggs: &Aggregations) -> HashSet<String> {
-    let mut term_dict_field_names = Default::default();
-    for el in aggs.values() {
-        el.get_term_dict_field_names(&mut term_dict_field_names)
-    }
-    term_dict_field_names
-}
-
 /// Extract all fast field names used in the tree.
 pub fn get_fast_field_names(aggs: &Aggregations) -> HashSet<String> {
     let mut fast_field_names = Default::default();
@@ -155,16 +149,12 @@ pub enum Aggregation {
 }
 
 impl Aggregation {
-    fn get_term_dict_field_names(&self, term_field_names: &mut HashSet<String>) {
-        if let Aggregation::Bucket(bucket) = self {
-            bucket.get_term_dict_field_names(term_field_names)
-        }
-    }
-
     fn get_fast_field_names(&self, fast_field_names: &mut HashSet<String>) {
         match self {
             Aggregation::Bucket(bucket) => bucket.get_fast_field_names(fast_field_names),
-            Aggregation::Metric(metric) => metric.get_fast_field_names(fast_field_names),
+            Aggregation::Metric(metric) => {
+                fast_field_names.insert(metric.get_fast_field_name().to_string());
+            }
         }
     }
 }
@@ -193,14 +183,9 @@ pub struct BucketAggregation {
 }
 
 impl BucketAggregation {
-    fn get_term_dict_field_names(&self, term_dict_field_names: &mut HashSet<String>) {
-        if let BucketAggregationType::Terms(terms) = &self.bucket_agg {
-            term_dict_field_names.insert(terms.field.to_string());
-        }
-        term_dict_field_names.extend(get_term_dict_field_names(&self.sub_aggregation));
-    }
     fn get_fast_field_names(&self, fast_field_names: &mut HashSet<String>) {
-        self.bucket_agg.get_fast_field_names(fast_field_names);
+        let fast_field_name = self.bucket_agg.get_fast_field_name();
+        fast_field_names.insert(fast_field_name.to_string());
         fast_field_names.extend(get_fast_field_names(&self.sub_aggregation));
     }
 }
@@ -214,20 +199,22 @@ pub enum BucketAggregationType {
     /// Put data into buckets of user-defined ranges.
     #[serde(rename = "histogram")]
     Histogram(HistogramAggregation),
+    /// Put data into buckets of user-defined ranges.
+    #[serde(rename = "date_histogram")]
+    DateHistogram(DateHistogramAggregationReq),
     /// Put data into buckets of terms.
     #[serde(rename = "terms")]
     Terms(TermsAggregation),
 }
 
 impl BucketAggregationType {
-    fn get_fast_field_names(&self, fast_field_names: &mut HashSet<String>) {
+    fn get_fast_field_name(&self) -> &str {
         match self {
-            BucketAggregationType::Terms(terms) => fast_field_names.insert(terms.field.to_string()),
-            BucketAggregationType::Range(range) => fast_field_names.insert(range.field.to_string()),
-            BucketAggregationType::Histogram(histogram) => {
-                fast_field_names.insert(histogram.field.to_string())
-            }
-        };
+            BucketAggregationType::Terms(terms) => terms.field.as_str(),
+            BucketAggregationType::Range(range) => range.field.as_str(),
+            BucketAggregationType::Histogram(histogram) => histogram.field.as_str(),
+            BucketAggregationType::DateHistogram(histogram) => histogram.field.as_str(),
+        }
     }
 }
 
@@ -262,16 +249,15 @@ pub enum MetricAggregation {
 }
 
 impl MetricAggregation {
-    fn get_fast_field_names(&self, fast_field_names: &mut HashSet<String>) {
-        let fast_field_name = match self {
+    fn get_fast_field_name(&self) -> &str {
+        match self {
             MetricAggregation::Average(avg) => avg.field_name(),
             MetricAggregation::Count(count) => count.field_name(),
             MetricAggregation::Max(max) => max.field_name(),
             MetricAggregation::Min(min) => min.field_name(),
             MetricAggregation::Stats(stats) => stats.field_name(),
             MetricAggregation::Sum(sum) => sum.field_name(),
-        };
-        fast_field_names.insert(fast_field_name.to_string());
+        }
     }
 }
 

src/aggregation/agg_req_with_accessor.rs

@@ -3,17 +3,18 @@
 use std::rc::Rc;
 use std::sync::atomic::AtomicU32;
 
-use columnar::{Column, StrColumn};
+use columnar::{Column, ColumnType, StrColumn};
 
 use super::agg_req::{Aggregation, Aggregations, BucketAggregationType, MetricAggregation};
-use super::bucket::{HistogramAggregation, RangeAggregation, TermsAggregation};
+use super::bucket::{
+    DateHistogramAggregationReq, HistogramAggregation, RangeAggregation, TermsAggregation,
+};
 use super::metric::{
     AverageAggregation, CountAggregation, MaxAggregation, MinAggregation, StatsAggregation,
     SumAggregation,
 };
 use super::segment_agg_result::BucketCount;
 use super::VecWithNames;
-use crate::schema::Type;
 use crate::{SegmentReader, TantivyError};
 
 #[derive(Clone, Default)]
@@ -41,7 +42,7 @@ pub struct BucketAggregationWithAccessor {
     /// based on search terms. So eventually this needs to be Option or moved.
     pub(crate) accessor: Column<u64>,
     pub(crate) str_dict_column: Option<StrColumn>,
-    pub(crate) field_type: Type,
+    pub(crate) field_type: ColumnType,
     pub(crate) bucket_agg: BucketAggregationType,
     pub(crate) sub_aggregation: AggregationsWithAccessor,
     pub(crate) bucket_count: BucketCount,
@@ -63,10 +64,14 @@ impl BucketAggregationWithAccessor {
             BucketAggregationType::Histogram(HistogramAggregation {
                 field: field_name, ..
             }) => get_ff_reader_and_validate(reader, field_name)?,
+            BucketAggregationType::DateHistogram(DateHistogramAggregationReq {
+                field: field_name,
+                ..
+            }) => get_ff_reader_and_validate(reader, field_name)?,
             BucketAggregationType::Terms(TermsAggregation {
                 field: field_name, ..
             }) => {
-                str_dict_column = reader.fast_fields().str(&field_name)?;
+                str_dict_column = reader.fast_fields().str(field_name)?;
                 get_ff_reader_and_validate(reader, field_name)?
             }
         };
@@ -94,7 +99,7 @@ impl BucketAggregationWithAccessor {
 #[derive(Clone)]
 pub struct MetricAggregationWithAccessor {
     pub metric: MetricAggregation,
-    pub field_type: Type,
+    pub field_type: ColumnType,
     pub accessor: Column<u64>,
 }
 
@@ -158,22 +163,12 @@ pub(crate) fn get_aggs_with_accessor_and_validate(
 fn get_ff_reader_and_validate(
     reader: &SegmentReader,
     field_name: &str,
-) -> crate::Result<(columnar::Column<u64>, Type)> {
-    let field = reader.schema().get_field(field_name)?;
-    // TODO we should get type metadata from columnar
-    let field_type = reader
-        .schema()
-        .get_field_entry(field)
-        .field_type()
-        .value_type();
-    // TODO Do validation
-
+) -> crate::Result<(columnar::Column<u64>, ColumnType)> {
     let ff_fields = reader.fast_fields();
-    let ff_field = ff_fields.u64_lenient(field_name)?.ok_or_else(|| {
-        TantivyError::InvalidArgument(format!(
-            "No numerical fast field found for field: {}",
-            field_name
-        ))
-    })?;
-    Ok((ff_field, field_type))
+    let ff_field_with_type = ff_fields
+        .u64_lenient_with_type(field_name)?
+        .ok_or_else(|| {
+            TantivyError::InvalidArgument(format!("No fast field found for field: {}", field_name))
+        })?;
+    Ok(ff_field_with_type)
 }

src/aggregation/agg_result.rs

@@ -12,7 +12,6 @@ use super::bucket::GetDocCount;
 use super::intermediate_agg_result::{IntermediateBucketResult, IntermediateMetricResult};
 use super::metric::{SingleMetricResult, Stats};
 use super::Key;
-use crate::schema::Schema;
 use crate::TantivyError;
 
 #[derive(Clone, Default, Debug, PartialEq, Serialize, Deserialize)]
@@ -154,12 +153,9 @@ pub enum BucketResult {
 }
 
 impl BucketResult {
-    pub(crate) fn empty_from_req(
-        req: &BucketAggregationInternal,
-        schema: &Schema,
-    ) -> crate::Result<Self> {
+    pub(crate) fn empty_from_req(req: &BucketAggregationInternal) -> crate::Result<Self> {
         let empty_bucket = IntermediateBucketResult::empty_from_req(&req.bucket_agg);
-        empty_bucket.into_final_bucket_result(req, schema)
+        empty_bucket.into_final_bucket_result(req)
     }
 }
 

src/aggregation/bucket/histogram/date_histogram.rs

@@ -1,5 +1,8 @@
 use serde::{Deserialize, Serialize};
 
+use super::{HistogramAggregation, HistogramBounds};
+use crate::aggregation::AggregationError;
+
 /// DateHistogramAggregation is similar to `HistogramAggregation`, but it can only be used with date
 /// type.
 ///
@@ -29,8 +32,16 @@ use serde::{Deserialize, Serialize};
 /// See [`BucketEntry`](crate::aggregation::agg_result::BucketEntry)
 #[derive(Clone, Debug, Default, PartialEq, Serialize, Deserialize)]
 pub struct DateHistogramAggregationReq {
+    #[doc(hidden)]
+    /// Only for validation
+    interval: Option<String>,
+    #[doc(hidden)]
+    /// Only for validation
+    date_interval: Option<String>,
     /// The field to aggregate on.
     pub field: String,
+    /// The format to format dates.
+    pub format: Option<String>,
     /// The interval to chunk your data range. Each bucket spans a value range of
     /// [0..fixed_interval). Accepted values
     ///
@@ -55,72 +66,410 @@ pub struct DateHistogramAggregationReq {
     /// Intervals implicitly defines an absolute grid of buckets `[interval * k, interval * (k +
     /// 1))`.
     pub offset: Option<String>,
+    /// The minimum number of documents in a bucket to be returned. Defaults to 0.
+    pub min_doc_count: Option<u64>,
+    /// Limits the data range to `[min, max]` closed interval.
+    ///
+    /// This can be used to filter values if they are not in the data range.
+    ///
+    /// hard_bounds only limits the buckets, to force a range set both extended_bounds and
+    /// hard_bounds to the same range.
+    ///
+    /// Needs to be provided as timestamp in microseconds precision.
+    ///
+    /// ## Example
+    /// ```json
+    /// {
+    ///     "sales_over_time": {
+    ///        "date_histogram": {
+    ///            "field": "dates",
+    ///            "interval": "1d",
+    ///            "hard_bounds": {
+    ///                "min": 0,
+    ///                "max": 1420502400000000
+    ///            }
+    ///        }
+    ///    }
+    /// }
+    /// ```
+    pub hard_bounds: Option<HistogramBounds>,
+    /// Can be set to extend your bounds. The range of the buckets is by default defined by the
+    /// data range of the values of the documents. As the name suggests, this can only be used to
+    /// extend the value range. If the bounds for min or max are not extending the range, the value
+    /// has no effect on the returned buckets.
+    ///
+    /// Cannot be set in conjunction with min_doc_count > 0, since the empty buckets from extended
+    /// bounds would not be returned.
+    pub extended_bounds: Option<HistogramBounds>,
+
     /// Whether to return the buckets as a hash map
     #[serde(default)]
     pub keyed: bool,
 }
 
 impl DateHistogramAggregationReq {
+    pub(crate) fn to_histogram_req(&self) -> crate::Result<HistogramAggregation> {
+        self.validate()?;
+        Ok(HistogramAggregation {
+            field: self.field.to_string(),
+            interval: parse_into_microseconds(&self.fixed_interval)? as f64,
+            offset: self
+                .offset
+                .as_ref()
+                .map(|offset| parse_offset_into_microseconds(offset))
+                .transpose()?
+                .map(|el| el as f64),
+            min_doc_count: self.min_doc_count,
+            hard_bounds: None,
+            extended_bounds: None,
+            keyed: self.keyed,
+        })
+    }
+
     fn validate(&self) -> crate::Result<()> {
+        if self.interval.is_some() {
+            return Err(crate::TantivyError::InvalidArgument(format!(
+                "`interval` parameter {:?} in date histogram is unsupported, only \
+                 `fixed_interval` is supported",
+                self.interval
+            )));
+        }
+        if self.format.is_some() {
+            return Err(crate::TantivyError::InvalidArgument(
+                "format parameter on date_histogram is unsupported".to_string(),
+            ));
+        }
+
+        if self.date_interval.is_some() {
+            return Err(crate::TantivyError::InvalidArgument(
+                "date_interval in date histogram is unsupported, only `fixed_interval` is \
+                 supported"
+                    .to_string(),
+            ));
+        }
+
+        parse_into_microseconds(&self.fixed_interval)?;
+
         Ok(())
     }
 }
 
-#[derive(Debug, PartialEq, Eq)]
+#[derive(Debug, Clone, PartialEq, Eq, Error)]
 /// Errors when parsing the fixed interval for `DateHistogramAggregationReq`.
 pub enum DateHistogramParseError {
     /// Unit not recognized in passed String
+    #[error("Unit not recognized in passed String {0:?}")]
     UnitNotRecognized(String),
     /// Number not found in passed String
+    #[error("Number not found in passed String {0:?}")]
     NumberMissing(String),
     /// Unit not found in passed String
+    #[error("Unit not found in passed String {0:?}")]
     UnitMissing(String),
+    /// Offset invalid
+    #[error("passed offset is invalid {0:?}")]
+    InvalidOffset(String),
 }
 
-fn parse_into_milliseconds(input: &str) -> Result<u64, DateHistogramParseError> {
+fn parse_offset_into_microseconds(input: &str) -> Result<i64, AggregationError> {
+    let is_sign = |byte| &[byte] == b"-" || &[byte] == b"+";
+    if input.is_empty() {
+        return Err(DateHistogramParseError::InvalidOffset(input.to_string()).into());
+    }
+
+    let has_sign = is_sign(input.as_bytes()[0]);
+    if has_sign {
+        let (sign, input) = input.split_at(1);
+        let val = parse_into_microseconds(input)?;
+        if sign == "-" {
+            Ok(-val)
+        } else {
+            Ok(val)
+        }
+    } else {
+        parse_into_microseconds(input)
+    }
+}
+
+fn parse_into_microseconds(input: &str) -> Result<i64, AggregationError> {
     let split_boundary = input
-        .char_indices()
-        .take_while(|(pos, el)| el.is_numeric())
+        .as_bytes()
+        .iter()
+        .take_while(|byte| byte.is_ascii_digit())
         .count();
     let (number, unit) = input.split_at(split_boundary);
     if number.is_empty() {
-        return Err(DateHistogramParseError::NumberMissing(input.to_string()));
+        return Err(DateHistogramParseError::NumberMissing(input.to_string()).into());
     }
     if unit.is_empty() {
-        return Err(DateHistogramParseError::UnitMissing(input.to_string()));
+        return Err(DateHistogramParseError::UnitMissing(input.to_string()).into());
     }
-    let number: u64 = number.parse().unwrap();
+    let number: i64 = number
+        .parse()
+        // Technically this should never happen, but there was a bug
+        // here and being defensive does not hurt.
+        .map_err(|_err| DateHistogramParseError::NumberMissing(input.to_string()))?;
+
     let multiplier_from_unit = match unit {
         "ms" => 1,
         "s" => 1000,
         "m" => 60 * 1000,
         "h" => 60 * 60 * 1000,
         "d" => 24 * 60 * 60 * 1000,
-        _ => return Err(DateHistogramParseError::UnitNotRecognized(unit.to_string())),
+        _ => return Err(DateHistogramParseError::UnitNotRecognized(unit.to_string()).into()),
     };
 
-    Ok(number * multiplier_from_unit)
+    Ok(number * multiplier_from_unit * 1000)
 }
 
 #[cfg(test)]
 mod tests {
+    use pretty_assertions::assert_eq;
+
     use super::*;
+    use crate::aggregation::agg_req::Aggregations;
+    use crate::aggregation::tests::exec_request;
+    use crate::indexer::NoMergePolicy;
+    use crate::schema::{Schema, FAST};
+    use crate::Index;
 
     #[test]
-    fn parser_test() {
-        assert_eq!(parse_into_milliseconds("1m").unwrap(), 60_000);
-        assert_eq!(parse_into_milliseconds("2m").unwrap(), 120_000);
+    fn test_parse_into_microseconds() {
+        assert_eq!(parse_into_microseconds("1m").unwrap(), 60_000_000);
+        assert_eq!(parse_into_microseconds("2m").unwrap(), 120_000_000);
         assert_eq!(
-            parse_into_milliseconds("2y").unwrap_err(),
-            DateHistogramParseError::UnitNotRecognized("y".to_string())
+            parse_into_microseconds("2y").unwrap_err(),
+            DateHistogramParseError::UnitNotRecognized("y".to_string()).into()
         );
         assert_eq!(
-            parse_into_milliseconds("2000").unwrap_err(),
-            DateHistogramParseError::UnitMissing("2000".to_string())
+            parse_into_microseconds("2000").unwrap_err(),
+            DateHistogramParseError::UnitMissing("2000".to_string()).into()
         );
         assert_eq!(
-            parse_into_milliseconds("ms").unwrap_err(),
-            DateHistogramParseError::NumberMissing("ms".to_string())
+            parse_into_microseconds("ms").unwrap_err(),
+            DateHistogramParseError::NumberMissing("ms".to_string()).into()
         );
     }
+
+    #[test]
+    fn test_parse_offset_into_microseconds() {
+        assert_eq!(parse_offset_into_microseconds("1m").unwrap(), 60_000_000);
+        assert_eq!(parse_offset_into_microseconds("+1m").unwrap(), 60_000_000);
+        assert_eq!(parse_offset_into_microseconds("-1m").unwrap(), -60_000_000);
+        assert_eq!(parse_offset_into_microseconds("2m").unwrap(), 120_000_000);
+        assert_eq!(parse_offset_into_microseconds("+2m").unwrap(), 120_000_000);
+        assert_eq!(parse_offset_into_microseconds("-2m").unwrap(), -120_000_000);
+        assert_eq!(parse_offset_into_microseconds("-2ms").unwrap(), -2_000);
+        assert_eq!(
+            parse_offset_into_microseconds("2y").unwrap_err(),
+            DateHistogramParseError::UnitNotRecognized("y".to_string()).into()
+        );
+        assert_eq!(
+            parse_offset_into_microseconds("2000").unwrap_err(),
+            DateHistogramParseError::UnitMissing("2000".to_string()).into()
+        );
+        assert_eq!(
+            parse_offset_into_microseconds("ms").unwrap_err(),
+            DateHistogramParseError::NumberMissing("ms".to_string()).into()
+        );
+    }
+
+    #[test]
+    fn test_parse_into_milliseconds_do_not_accept_non_ascii() {
+        assert!(parse_into_microseconds("１m").is_err());
+    }
+
+    pub fn get_test_index_from_docs(
+        merge_segments: bool,
+        segment_and_docs: &[Vec<&str>],
+    ) -> crate::Result<Index> {
+        let mut schema_builder = Schema::builder();
+        schema_builder.add_date_field("date", FAST);
+        schema_builder.add_text_field("text", FAST);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema.clone());
+        {
+            let mut index_writer = index.writer_with_num_threads(1, 30_000_000)?;
+            index_writer.set_merge_policy(Box::new(NoMergePolicy));
+            for values in segment_and_docs {
+                for doc_str in values {
+                    let doc = schema.parse_document(doc_str)?;
+                    index_writer.add_document(doc)?;
+                }
+                // writing the segment
+                index_writer.commit()?;
+            }
+        }
+        if merge_segments {
+            let segment_ids = index
+                .searchable_segment_ids()
+                .expect("Searchable segments failed.");
+            if segment_ids.len() > 1 {
+                let mut index_writer = index.writer_for_tests()?;
+                index_writer.merge(&segment_ids).wait()?;
+                index_writer.wait_merging_threads()?;
+            }
+        }
+
+        Ok(index)
+    }
+
+    #[test]
+    fn histogram_test_date_force_merge_segments() -> crate::Result<()> {
+        histogram_test_date_merge_segments(true)
+    }
+
+    #[test]
+    fn histogram_test_date() -> crate::Result<()> {
+        histogram_test_date_merge_segments(false)
+    }
+    fn histogram_test_date_merge_segments(merge_segments: bool) -> crate::Result<()> {
+        let docs = vec![
+            vec![r#"{ "date": "2015-01-01T12:10:30Z", "text": "aaa" }"#],
+            vec![r#"{ "date": "2015-01-01T11:11:30Z", "text": "bbb" }"#],
+            vec![r#"{ "date": "2015-01-02T00:00:00Z", "text": "bbb" }"#],
+            vec![r#"{ "date": "2015-01-06T00:00:00Z", "text": "ccc" }"#],
+        ];
+
+        let index = get_test_index_from_docs(merge_segments, &docs)?;
+        // 30day + offset
+        let elasticsearch_compatible_json = json!(
+            {
+              "sales_over_time": {
+                "date_histogram": {
+                  "field": "date",
+                  "fixed_interval": "30d",
+                  "offset": "-4d"
+                }
+              }
+            }
+        );
+
+        let agg_req: Aggregations =
+            serde_json::from_str(&serde_json::to_string(&elasticsearch_compatible_json).unwrap())
+                .unwrap();
+        let res = exec_request(agg_req, &index)?;
+        let expected_res = json!({
+          "sales_over_time" : {
+          "buckets" : [
+            {
+              "key_as_string" : "2015-01-01T00:00:00Z",
+              "key" : 1420070400000000.0,
+              "doc_count" : 4
+            }
+          ]
+        }
+        });
+        assert_eq!(res, expected_res);
+
+        // 30day + offset + sub_agg
+        let elasticsearch_compatible_json = json!(
+            {
+              "sales_over_time": {
+                "date_histogram": {
+                  "field": "date",
+                  "fixed_interval": "30d",
+                  "offset": "-4d"
+                },
+                "aggs": {
+                    "texts": {
+                        "terms": {"field": "text"}
+                    }
+                }
+              }
+            }
+        );
+
+        let agg_req: Aggregations =
+            serde_json::from_str(&serde_json::to_string(&elasticsearch_compatible_json).unwrap())
+                .unwrap();
+        let res = exec_request(agg_req, &index)?;
+        println!("{}", serde_json::to_string_pretty(&res).unwrap());
+        let expected_res = json!({
+          "sales_over_time" : {
+          "buckets" : [
+            {
+              "key_as_string" : "2015-01-01T00:00:00Z",
+              "key" : 1420070400000000.0,
+              "doc_count" : 4,
+              "texts": {
+                  "buckets": [
+                    {
+                      "doc_count": 2,
+                      "key": "bbb"
+                    },
+                    {
+                      "doc_count": 1,
+                      "key": "ccc"
+                    },
+                    {
+                      "doc_count": 1,
+                      "key": "aaa"
+                    }
+                  ],
+                  "doc_count_error_upper_bound": 0,
+                  "sum_other_doc_count": 0
+                }
+            }
+          ]
+        }
+        });
+        assert_eq!(res, expected_res);
+
+        // 1day
+        let elasticsearch_compatible_json = json!(
+            {
+              "sales_over_time": {
+                "date_histogram": {
+                  "field": "date",
+                  "fixed_interval": "1d"
+                }
+              }
+            }
+        );
+
+        let agg_req: Aggregations =
+            serde_json::from_str(&serde_json::to_string(&elasticsearch_compatible_json).unwrap())
+                .unwrap();
+        let res = exec_request(agg_req, &index)?;
+        let expected_res = json!( {
+          "sales_over_time": {
+            "buckets": [
+              {
+                "doc_count": 2,
+                "key": 1420070400000000.0,
+                "key_as_string": "2015-01-01T00:00:00Z"
+              },
+              {
+                "doc_count": 1,
+                "key": 1420156800000000.0,
+                "key_as_string": "2015-01-02T00:00:00Z"
+              },
+              {
+                "doc_count": 0,
+                "key": 1420243200000000.0,
+                "key_as_string": "2015-01-03T00:00:00Z"
+              },
+              {
+                "doc_count": 0,
+                "key": 1420329600000000.0,
+                "key_as_string": "2015-01-04T00:00:00Z"
+              },
+              {
+                "doc_count": 0,
+                "key": 1420416000000000.0,
+                "key_as_string": "2015-01-05T00:00:00Z"
+              },
+              {
+                "doc_count": 1,
+                "key": 1420502400000000.0,
+                "key_as_string": "2015-01-06T00:00:00Z"
+              }
+            ]
+          }
+        });
+        assert_eq!(res, expected_res);
+
+        Ok(())
+    }
 }

src/aggregation/bucket/histogram/histogram.rs

@@ -1,9 +1,11 @@
 use std::cmp::Ordering;
 use std::fmt::Display;
 
-use columnar::Column;
+use columnar::ColumnType;
 use itertools::Itertools;
+use rustc_hash::FxHashMap;
 use serde::{Deserialize, Serialize};
+use tantivy_bitpacker::minmax;
 
 use crate::aggregation::agg_req::AggregationsInternal;
 use crate::aggregation::agg_req_with_accessor::{
@@ -14,10 +16,9 @@ use crate::aggregation::intermediate_agg_result::{
     IntermediateAggregationResults, IntermediateBucketResult, IntermediateHistogramBucketEntry,
 };
 use crate::aggregation::segment_agg_result::{
-    GenericSegmentAggregationResultsCollector, SegmentAggregationCollector,
+    build_segment_agg_collector, SegmentAggregationCollector,
 };
-use crate::aggregation::{f64_from_fastfield_u64, format_date};
-use crate::schema::{Schema, Type};
+use crate::aggregation::{f64_from_fastfield_u64, format_date, VecWithNames};
 use crate::{DocId, TantivyError};
 
 /// Histogram is a bucket aggregation, where buckets are created dynamically for given `interval`.
@@ -176,7 +177,7 @@ impl HistogramBounds {
     }
 }
 
-#[derive(Clone, Debug, PartialEq)]
+#[derive(Default, Clone, Debug, PartialEq)]
 pub(crate) struct SegmentHistogramBucketEntry {
     pub key: f64,
     pub doc_count: u64,
@@ -185,7 +186,7 @@ pub(crate) struct SegmentHistogramBucketEntry {
 impl SegmentHistogramBucketEntry {
     pub(crate) fn into_intermediate_bucket_entry(
         self,
-        sub_aggregation: GenericSegmentAggregationResultsCollector,
+        sub_aggregation: Box<dyn SegmentAggregationCollector>,
         agg_with_accessor: &AggregationsWithAccessor,
     ) -> crate::Result<IntermediateHistogramBucketEntry> {
         Ok(IntermediateHistogramBucketEntry {
@@ -202,14 +203,85 @@ impl SegmentHistogramBucketEntry {
 #[derive(Clone, Debug)]
 pub struct SegmentHistogramCollector {
     /// The buckets containing the aggregation data.
-    buckets: Vec<SegmentHistogramBucketEntry>,
-    sub_aggregations: Option<Vec<GenericSegmentAggregationResultsCollector>>,
-    field_type: Type,
+    buckets: FxHashMap<i64, SegmentHistogramBucketEntry>,
+    sub_aggregations: FxHashMap<i64, Box<dyn SegmentAggregationCollector>>,
+    sub_aggregation_blueprint: Option<Box<dyn SegmentAggregationCollector>>,
+    column_type: ColumnType,
     interval: f64,
     offset: f64,
-    min_doc_count: u64,
-    first_bucket_num: i64,
     bounds: HistogramBounds,
+    accessor_idx: usize,
+}
+
+impl SegmentAggregationCollector for SegmentHistogramCollector {
+    fn into_intermediate_aggregations_result(
+        self: Box<Self>,
+        agg_with_accessor: &AggregationsWithAccessor,
+    ) -> crate::Result<IntermediateAggregationResults> {
+        let name = agg_with_accessor.buckets.keys[self.accessor_idx].to_string();
+        let agg_with_accessor = &agg_with_accessor.buckets.values[self.accessor_idx];
+
+        let bucket = self.into_intermediate_bucket_result(agg_with_accessor)?;
+        let buckets = Some(VecWithNames::from_entries(vec![(name, bucket)]));
+
+        Ok(IntermediateAggregationResults {
+            metrics: None,
+            buckets,
+        })
+    }
+
+    fn collect(
+        &mut self,
+        doc: crate::DocId,
+        agg_with_accessor: &AggregationsWithAccessor,
+    ) -> crate::Result<()> {
+        self.collect_block(&[doc], agg_with_accessor)
+    }
+
+    fn collect_block(
+        &mut self,
+        docs: &[crate::DocId],
+        agg_with_accessor: &AggregationsWithAccessor,
+    ) -> crate::Result<()> {
+        let accessor = &agg_with_accessor.buckets.values[self.accessor_idx].accessor;
+        let sub_aggregation_accessor =
+            &agg_with_accessor.buckets.values[self.accessor_idx].sub_aggregation;
+
+        let bounds = self.bounds;
+        let interval = self.interval;
+        let offset = self.offset;
+        let get_bucket_pos = |val| (get_bucket_pos_f64(val, interval, offset) as i64);
+
+        for doc in docs {
+            for val in accessor.values_for_doc(*doc) {
+                let val = self.f64_from_fastfield_u64(val);
+
+                let bucket_pos = get_bucket_pos(val);
+
+                if bounds.contains(val) {
+                    self.increment_bucket(
+                        bucket_pos,
+                        *doc,
+                        sub_aggregation_accessor,
+                        interval,
+                        offset,
+                    )?;
+                }
+            }
+        }
+        Ok(())
+    }
+
+    fn flush(&mut self, agg_with_accessor: &AggregationsWithAccessor) -> crate::Result<()> {
+        let sub_aggregation_accessor =
+            &agg_with_accessor.buckets.values[self.accessor_idx].sub_aggregation;
+
+        for sub_aggregation in self.sub_aggregations.values_mut() {
+            sub_aggregation.flush(sub_aggregation_accessor)?;
+        }
+
+        Ok(())
+    }
 }
 
 impl SegmentHistogramCollector {
@@ -217,210 +289,96 @@ impl SegmentHistogramCollector {
         self,
         agg_with_accessor: &BucketAggregationWithAccessor,
     ) -> crate::Result<IntermediateBucketResult> {
-        // Compute the number of buckets to validate against max num buckets
-        // Note: We use min_doc_count here, but it's only an lowerbound here, since were are on the
-        // intermediate level and after merging the number of documents of a bucket could exceed
-        // `min_doc_count`.
-        {
-            let cut_off_buckets_front = self
-                .buckets
-                .iter()
-                .take_while(|bucket| bucket.doc_count <= self.min_doc_count)
-                .count();
-            let cut_off_buckets_back = self.buckets[cut_off_buckets_front..]
-                .iter()
-                .rev()
-                .take_while(|bucket| bucket.doc_count <= self.min_doc_count)
-                .count();
-            let estimate_num_buckets =
-                self.buckets.len() - cut_off_buckets_front - cut_off_buckets_back;
+        let mut buckets = Vec::with_capacity(self.buckets.len());
 
-            agg_with_accessor
-                .bucket_count
-                .add_count(estimate_num_buckets as u32);
-            agg_with_accessor.bucket_count.validate_bucket_count()?;
-        }
+        if self.sub_aggregation_blueprint.is_some() {
+            for (bucket_pos, bucket) in self.buckets.into_iter() {
+                let bucket_res = bucket.into_intermediate_bucket_entry(
+                    self.sub_aggregations.get(&bucket_pos).unwrap().clone(),
+                    &agg_with_accessor.sub_aggregation,
+                );
 
-        let mut buckets = Vec::with_capacity(
-            self.buckets
-                .iter()
-                .filter(|bucket| bucket.doc_count != 0)
-                .count(),
-        );
-
-        // Below we remove empty buckets for two reasons
-        // 1. To reduce the size of the intermediate result, which may be passed on the wire.
-        // 2. To mimic elasticsearch, there are no empty buckets at the start and end.
-        //
-        // Empty buckets may be added later again in the final result, depending on the request.
-        if let Some(sub_aggregations) = self.sub_aggregations {
-            for bucket_res in self
-                .buckets
-                .into_iter()
-                .zip(sub_aggregations.into_iter())
-                .filter(|(bucket, _sub_aggregation)| bucket.doc_count != 0)
-                .map(|(bucket, sub_aggregation)| {
-                    bucket.into_intermediate_bucket_entry(
-                        sub_aggregation,
-                        &agg_with_accessor.sub_aggregation,
-                    )
-                })
-            {
                 buckets.push(bucket_res?);
             }
         } else {
-            buckets.extend(
-                self.buckets
-                    .into_iter()
-                    .filter(|bucket| bucket.doc_count != 0)
-                    .map(|bucket| bucket.into()),
-            );
+            buckets.extend(self.buckets.into_values().map(|bucket| bucket.into()));
         };
+        buckets.sort_unstable_by(|b1, b2| b1.key.partial_cmp(&b2.key).unwrap_or(Ordering::Equal));
 
-        Ok(IntermediateBucketResult::Histogram { buckets })
+        Ok(IntermediateBucketResult::Histogram {
+            buckets,
+            column_type: Some(self.column_type),
+        })
     }
 
     pub(crate) fn from_req_and_validate(
         req: &HistogramAggregation,
         sub_aggregation: &AggregationsWithAccessor,
-        field_type: Type,
-        accessor: &Column<u64>,
+        field_type: ColumnType,
+        accessor_idx: usize,
     ) -> crate::Result<Self> {
         req.validate()?;
-        let min = f64_from_fastfield_u64(accessor.min_value(), &field_type);
-        let max = f64_from_fastfield_u64(accessor.max_value(), &field_type);
 
-        let (min, max) = get_req_min_max(req, Some((min, max)));
-
-        // We compute and generate the buckets range (min, max) based on the request and the min
-        // max in the fast field, but this is likely not ideal when this is a subbucket, where many
-        // unnecessary buckets may be generated.
-        let buckets = generate_buckets(req, min, max);
-
-        let sub_aggregations = if sub_aggregation.is_empty() {
+        let sub_aggregation_blueprint = if sub_aggregation.is_empty() {
             None
         } else {
-            let sub_aggregation =
-                GenericSegmentAggregationResultsCollector::from_req_and_validate(sub_aggregation)?;
-            Some(buckets.iter().map(|_| sub_aggregation.clone()).collect())
+            let sub_aggregation = build_segment_agg_collector(sub_aggregation)?;
+            Some(sub_aggregation)
         };
 
-        let buckets = buckets
-            .iter()
-            .map(|bucket| SegmentHistogramBucketEntry {
-                key: *bucket,
-                doc_count: 0,
-            })
-            .collect();
-
-        let first_bucket_num =
-            get_bucket_num_f64(min, req.interval, req.offset.unwrap_or(0.0)) as i64;
-
         let bounds = req.hard_bounds.unwrap_or(HistogramBounds {
             min: f64::MIN,
             max: f64::MAX,
         });
 
         Ok(Self {
-            buckets,
-            field_type,
+            buckets: Default::default(),
+            column_type: field_type,
             interval: req.interval,
             offset: req.offset.unwrap_or(0.0),
-            first_bucket_num,
             bounds,
-            sub_aggregations,
-            min_doc_count: req.min_doc_count(),
+            sub_aggregations: Default::default(),
+            sub_aggregation_blueprint,
+            accessor_idx,
         })
     }
 
-    #[inline]
-    pub(crate) fn collect_block(
-        &mut self,
-        docs: &[DocId],
-        bucket_with_accessor: &BucketAggregationWithAccessor,
-        force_flush: bool,
-    ) -> crate::Result<()> {
-        let bounds = self.bounds;
-        let interval = self.interval;
-        let offset = self.offset;
-        let first_bucket_num = self.first_bucket_num;
-        let get_bucket_num =
-            |val| (get_bucket_num_f64(val, interval, offset) as i64 - first_bucket_num) as usize;
-
-        let accessor = &bucket_with_accessor.accessor;
-        for doc in docs {
-            for val in accessor.values(*doc) {
-                let val = self.f64_from_fastfield_u64(val);
-
-                let bucket_pos = get_bucket_num(val);
-                self.increment_bucket_if_in_bounds(
-                    val,
-                    &bounds,
-                    bucket_pos,
-                    *doc,
-                    &bucket_with_accessor.sub_aggregation,
-                )?;
-            }
-        }
-        if force_flush {
-            if let Some(sub_aggregations) = self.sub_aggregations.as_mut() {
-                for sub_aggregation in sub_aggregations {
-                    sub_aggregation
-                        .flush_staged_docs(&bucket_with_accessor.sub_aggregation, force_flush)?;
-                }
-            }
-        }
-        Ok(())
-    }
-
-    #[inline]
-    fn increment_bucket_if_in_bounds(
-        &mut self,
-        val: f64,
-        bounds: &HistogramBounds,
-        bucket_pos: usize,
-        doc: DocId,
-        bucket_with_accessor: &AggregationsWithAccessor,
-    ) -> crate::Result<()> {
-        if bounds.contains(val) {
-            debug_assert_eq!(
-                self.buckets[bucket_pos].key,
-                get_bucket_val(val, self.interval, self.offset)
-            );
-
-            self.increment_bucket(bucket_pos, doc, bucket_with_accessor)?;
-        }
-        Ok(())
-    }
-
     #[inline]
     fn increment_bucket(
         &mut self,
-        bucket_pos: usize,
+        bucket_pos: i64,
         doc: DocId,
         bucket_with_accessor: &AggregationsWithAccessor,
+        interval: f64,
+        offset: f64,
     ) -> crate::Result<()> {
-        let bucket = &mut self.buckets[bucket_pos];
+        let bucket = self.buckets.entry(bucket_pos).or_insert_with(|| {
+            let key = get_bucket_key_from_pos(bucket_pos as f64, interval, offset);
+            SegmentHistogramBucketEntry { key, doc_count: 0 }
+        });
         bucket.doc_count += 1;
-        if let Some(sub_aggregation) = self.sub_aggregations.as_mut() {
-            sub_aggregation[bucket_pos].collect(doc, bucket_with_accessor)?;
+        if let Some(sub_aggregation_blueprint) = self.sub_aggregation_blueprint.as_mut() {
+            self.sub_aggregations
+                .entry(bucket_pos)
+                .or_insert_with(|| sub_aggregation_blueprint.clone())
+                .collect(doc, bucket_with_accessor)?;
         }
         Ok(())
     }
 
+    #[inline]
     fn f64_from_fastfield_u64(&self, val: u64) -> f64 {
-        f64_from_fastfield_u64(val, &self.field_type)
+        f64_from_fastfield_u64(val, &self.column_type)
     }
 }
 
 #[inline]
-fn get_bucket_num_f64(val: f64, interval: f64, offset: f64) -> f64 {
+fn get_bucket_pos_f64(val: f64, interval: f64, offset: f64) -> f64 {
     ((val - offset) / interval).floor()
 }
 
 #[inline]
-fn get_bucket_val(val: f64, interval: f64, offset: f64) -> f64 {
-    let bucket_pos = get_bucket_num_f64(val, interval, offset);
+fn get_bucket_key_from_pos(bucket_pos: f64, interval: f64, offset: f64) -> f64 {
     bucket_pos * interval + offset
 }
 
@@ -429,19 +387,14 @@ fn intermediate_buckets_to_final_buckets_fill_gaps(
     buckets: Vec<IntermediateHistogramBucketEntry>,
     histogram_req: &HistogramAggregation,
     sub_aggregation: &AggregationsInternal,
-    schema: &Schema,
 ) -> crate::Result<Vec<BucketEntry>> {
     // Generate the full list of buckets without gaps.
     //
     // The bounds are the min max from the current buckets, optionally extended by
     // extended_bounds from the request
-    let min_max = if buckets.is_empty() {
-        None
-    } else {
-        let min = buckets[0].key;
-        let max = buckets[buckets.len() - 1].key;
-        Some((min, max))
-    };
+    let min_max = minmax(buckets.iter().map(|bucket| bucket.key));
+
+    // TODO add memory check
     let fill_gaps_buckets = generate_buckets_with_opt_minmax(histogram_req, min_max);
 
     let empty_sub_aggregation = IntermediateAggregationResults::empty_from_req(sub_aggregation);
@@ -470,43 +423,33 @@ fn intermediate_buckets_to_final_buckets_fill_gaps(
                 sub_aggregation: empty_sub_aggregation.clone(),
             },
         })
-        .map(|intermediate_bucket| {
-            intermediate_bucket.into_final_bucket_entry(sub_aggregation, schema)
-        })
+        .map(|intermediate_bucket| intermediate_bucket.into_final_bucket_entry(sub_aggregation))
         .collect::<crate::Result<Vec<_>>>()
 }
 
 // Convert to BucketEntry
 pub(crate) fn intermediate_histogram_buckets_to_final_buckets(
     buckets: Vec<IntermediateHistogramBucketEntry>,
+    column_type: Option<ColumnType>,
     histogram_req: &HistogramAggregation,
     sub_aggregation: &AggregationsInternal,
-    schema: &Schema,
 ) -> crate::Result<Vec<BucketEntry>> {
     let mut buckets = if histogram_req.min_doc_count() == 0 {
         // With min_doc_count != 0, we may need to add buckets, so that there are no
         // gaps, since intermediate result does not contain empty buckets (filtered to
         // reduce serialization size).
 
-        intermediate_buckets_to_final_buckets_fill_gaps(
-            buckets,
-            histogram_req,
-            sub_aggregation,
-            schema,
-        )?
+        intermediate_buckets_to_final_buckets_fill_gaps(buckets, histogram_req, sub_aggregation)?
     } else {
         buckets
             .into_iter()
             .filter(|histogram_bucket| histogram_bucket.doc_count >= histogram_req.min_doc_count())
-            .map(|histogram_bucket| {
-                histogram_bucket.into_final_bucket_entry(sub_aggregation, schema)
-            })
+            .map(|histogram_bucket| histogram_bucket.into_final_bucket_entry(sub_aggregation))
             .collect::<crate::Result<Vec<_>>>()?
     };
 
     // If we have a date type on the histogram buckets, we add the `key_as_string` field as rfc339
-    let field = schema.get_field(&histogram_req.field)?;
-    if schema.get_field_entry(field).field_type().is_date() {
+    if column_type == Some(ColumnType::DateTime) {
         for bucket in buckets.iter_mut() {
             if let crate::aggregation::Key::F64(val) = bucket.key {
                 let key_as_string = format_date(val as i64)?;
@@ -537,12 +480,6 @@ fn get_req_min_max(req: &HistogramAggregation, min_max: Option<(f64, f64)>) -> (
     (min, max)
 }
 
-/// Generates buckets with req.interval
-/// range is computed for provided min_max and request extended_bounds/hard_bounds
-pub(crate) fn generate_buckets(req: &HistogramAggregation, min: f64, max: f64) -> Vec<f64> {
-    generate_buckets_with_opt_minmax(req, Some((min, max)))
-}
-
 /// Generates buckets with req.interval
 /// Range is computed for provided min_max and request extended_bounds/hard_bounds
 /// returns empty vec when there is no range to span
@@ -553,8 +490,8 @@ pub(crate) fn generate_buckets_with_opt_minmax(
     let (min, max) = get_req_min_max(req, min_max);
 
     let offset = req.offset.unwrap_or(0.0);
-    let first_bucket_num = get_bucket_num_f64(min, req.interval, offset) as i64;
-    let last_bucket_num = get_bucket_num_f64(max, req.interval, offset) as i64;
+    let first_bucket_num = get_bucket_pos_f64(min, req.interval, offset) as i64;
+    let last_bucket_num = get_bucket_pos_f64(max, req.interval, offset) as i64;
     let mut buckets = Vec::with_capacity((first_bucket_num..=last_bucket_num).count());
     for bucket_pos in first_bucket_num..=last_bucket_num {
         let bucket_key = bucket_pos as f64 * req.interval + offset;
@@ -564,118 +501,6 @@ pub(crate) fn generate_buckets_with_opt_minmax(
     buckets
 }
 
-#[test]
-fn generate_buckets_test() {
-    let histogram_req = HistogramAggregation {
-        field: "dummy".to_string(),
-        interval: 2.0,
-        ..Default::default()
-    };
-
-    let buckets = generate_buckets(&histogram_req, 0.0, 10.0);
-    assert_eq!(buckets, vec![0.0, 2.0, 4.0, 6.0, 8.0, 10.0]);
-
-    let buckets = generate_buckets(&histogram_req, 2.5, 5.5);
-    assert_eq!(buckets, vec![2.0, 4.0]);
-
-    // Single bucket
-    let buckets = generate_buckets(&histogram_req, 0.5, 0.75);
-    assert_eq!(buckets, vec![0.0]);
-
-    // With offset
-    let histogram_req = HistogramAggregation {
-        field: "dummy".to_string(),
-        interval: 2.0,
-        offset: Some(0.5),
-        ..Default::default()
-    };
-
-    let buckets = generate_buckets(&histogram_req, 0.0, 10.0);
-    assert_eq!(buckets, vec![-1.5, 0.5, 2.5, 4.5, 6.5, 8.5]);
-
-    let buckets = generate_buckets(&histogram_req, 2.5, 5.5);
-    assert_eq!(buckets, vec![2.5, 4.5]);
-
-    // Single bucket
-    let buckets = generate_buckets(&histogram_req, 0.5, 0.75);
-    assert_eq!(buckets, vec![0.5]);
-
-    // no bucket
-    let buckets = generate_buckets(&histogram_req, f64::MAX, f64::MIN);
-    assert_eq!(buckets, vec![] as Vec<f64>);
-
-    // With extended_bounds
-    let histogram_req = HistogramAggregation {
-        field: "dummy".to_string(),
-        interval: 2.0,
-        extended_bounds: Some(HistogramBounds {
-            min: 0.0,
-            max: 10.0,
-        }),
-        ..Default::default()
-    };
-
-    let buckets = generate_buckets(&histogram_req, 0.0, 10.0);
-    assert_eq!(buckets, vec![0.0, 2.0, 4.0, 6.0, 8.0, 10.0]);
-
-    let buckets = generate_buckets(&histogram_req, 2.5, 5.5);
-    assert_eq!(buckets, vec![0.0, 2.0, 4.0, 6.0, 8.0, 10.0]);
-
-    // Single bucket, but extended_bounds
-    let buckets = generate_buckets(&histogram_req, 0.5, 0.75);
-    assert_eq!(buckets, vec![0.0, 2.0, 4.0, 6.0, 8.0, 10.0]);
-
-    // no bucket, but extended_bounds
-    let buckets = generate_buckets(&histogram_req, f64::MAX, f64::MIN);
-    assert_eq!(buckets, vec![0.0, 2.0, 4.0, 6.0, 8.0, 10.0]);
-
-    // With invalid extended_bounds
-    let histogram_req = HistogramAggregation {
-        field: "dummy".to_string(),
-        interval: 2.0,
-        extended_bounds: Some(HistogramBounds { min: 3.0, max: 5.0 }),
-        ..Default::default()
-    };
-
-    let buckets = generate_buckets(&histogram_req, 0.0, 10.0);
-    assert_eq!(buckets, vec![0.0, 2.0, 4.0, 6.0, 8.0, 10.0]);
-
-    // With hard_bounds reducing
-    let histogram_req = HistogramAggregation {
-        field: "dummy".to_string(),
-        interval: 2.0,
-        hard_bounds: Some(HistogramBounds { min: 3.0, max: 5.0 }),
-        ..Default::default()
-    };
-
-    let buckets = generate_buckets(&histogram_req, 0.0, 10.0);
-    assert_eq!(buckets, vec![2.0, 4.0]);
-
-    // With hard_bounds, extending has no effect
-    let histogram_req = HistogramAggregation {
-        field: "dummy".to_string(),
-        interval: 2.0,
-        hard_bounds: Some(HistogramBounds {
-            min: 0.0,
-            max: 10.0,
-        }),
-        ..Default::default()
-    };
-
-    let buckets = generate_buckets(&histogram_req, 2.5, 5.5);
-    assert_eq!(buckets, vec![2.0, 4.0]);
-
-    // Blubber
-    let histogram_req = HistogramAggregation {
-        field: "dummy".to_string(),
-        interval: 2.0,
-        ..Default::default()
-    };
-
-    let buckets = generate_buckets(&histogram_req, 4.0, 10.0);
-    assert_eq!(buckets, vec![4.0, 6.0, 8.0, 10.0]);
-}
-
 #[cfg(test)]
 mod tests {
 
@@ -1496,36 +1321,4 @@ mod tests {
 
         Ok(())
     }
-
-    #[test]
-    fn histogram_test_max_buckets_segments() -> crate::Result<()> {
-        let values = vec![0.0, 70000.0];
-
-        let index = get_test_index_from_values(true, &values)?;
-
-        let agg_req: Aggregations = vec![(
-            "my_interval".to_string(),
-            Aggregation::Bucket(BucketAggregation {
-                bucket_agg: BucketAggregationType::Histogram(HistogramAggregation {
-                    field: "score_f64".to_string(),
-                    interval: 1.0,
-                    ..Default::default()
-                }),
-                sub_aggregation: Default::default(),
-            }),
-        )]
-        .into_iter()
-        .collect();
-
-        let res = exec_request(agg_req, &index);
-
-        assert_eq!(
-            res.unwrap_err().to_string(),
-            "An invalid argument was passed: 'Aborting aggregation because too many buckets were \
-             created'"
-                .to_string()
-        );
-
-        Ok(())
-    }
 }

src/aggregation/bucket/mod.rs

@@ -21,28 +21,25 @@ use serde::{de, Deserialize, Deserializer, Serialize, Serializer};
 pub use term_agg::*;
 
 /// Order for buckets in a bucket aggregation.
-#[derive(Clone, Copy, Debug, PartialEq, Serialize, Deserialize)]
+#[derive(Clone, Copy, Debug, PartialEq, Serialize, Deserialize, Default)]
 pub enum Order {
     /// Asc order
     #[serde(rename = "asc")]
     Asc,
     /// Desc order
     #[serde(rename = "desc")]
+    #[default]
     Desc,
 }
 
-impl Default for Order {
-    fn default() -> Self {
-        Order::Desc
-    }
-}
-
 #[derive(Clone, Debug, PartialEq)]
 /// Order property by which to apply the order
+#[derive(Default)]
 pub enum OrderTarget {
     /// The key of the bucket
     Key,
     /// The doc count of the bucket
+    #[default]
     Count,
     /// Order by value of the sub aggregation metric with identified by given `String`.
     ///
@@ -50,11 +47,6 @@ pub enum OrderTarget {
     SubAggregation(String),
 }
 
-impl Default for OrderTarget {
-    fn default() -> Self {
-        OrderTarget::Count
-    }
-}
 impl From<&str> for OrderTarget {
     fn from(val: &str) -> Self {
         match val {

src/aggregation/bucket/range.rs

@@ -1,24 +1,22 @@
 use std::fmt::Debug;
 use std::ops::Range;
 
-use columnar::MonotonicallyMappableToU64;
+use columnar::{ColumnType, MonotonicallyMappableToU64};
 use rustc_hash::FxHashMap;
 use serde::{Deserialize, Serialize};
 
-use crate::aggregation::agg_req_with_accessor::{
-    AggregationsWithAccessor, BucketAggregationWithAccessor,
-};
+use crate::aggregation::agg_req_with_accessor::AggregationsWithAccessor;
 use crate::aggregation::intermediate_agg_result::{
-    IntermediateBucketResult, IntermediateRangeBucketEntry, IntermediateRangeBucketResult,
+    IntermediateAggregationResults, IntermediateBucketResult, IntermediateRangeBucketEntry,
+    IntermediateRangeBucketResult,
 };
 use crate::aggregation::segment_agg_result::{
-    BucketCount, GenericSegmentAggregationResultsCollector, SegmentAggregationCollector,
+    build_segment_agg_collector, BucketCount, SegmentAggregationCollector,
 };
 use crate::aggregation::{
-    f64_from_fastfield_u64, f64_to_fastfield_u64, format_date, Key, SerializedKey,
+    f64_from_fastfield_u64, f64_to_fastfield_u64, format_date, Key, SerializedKey, VecWithNames,
 };
-use crate::schema::Type;
-use crate::{DocId, TantivyError};
+use crate::TantivyError;
 
 /// Provide user-defined buckets to aggregate on.
 /// Two special buckets will automatically be created to cover the whole range of values.
@@ -128,14 +126,15 @@ pub(crate) struct SegmentRangeAndBucketEntry {
 pub struct SegmentRangeCollector {
     /// The buckets containing the aggregation data.
     buckets: Vec<SegmentRangeAndBucketEntry>,
-    field_type: Type,
+    column_type: ColumnType,
+    pub(crate) accessor_idx: usize,
 }
 
 #[derive(Clone)]
 pub(crate) struct SegmentRangeBucketEntry {
     pub key: Key,
     pub doc_count: u64,
-    pub sub_aggregation: Option<GenericSegmentAggregationResultsCollector>,
+    pub sub_aggregation: Option<Box<dyn SegmentAggregationCollector>>,
     /// The from range of the bucket. Equals `f64::MIN` when `None`.
     pub from: Option<f64>,
     /// The to range of the bucket. Equals `f64::MAX` when `None`. Open interval, `to` is not
@@ -174,12 +173,14 @@ impl SegmentRangeBucketEntry {
     }
 }
 
-impl SegmentRangeCollector {
-    pub fn into_intermediate_bucket_result(
-        self,
-        agg_with_accessor: &BucketAggregationWithAccessor,
-    ) -> crate::Result<IntermediateBucketResult> {
-        let field_type = self.field_type;
+impl SegmentAggregationCollector for SegmentRangeCollector {
+    fn into_intermediate_aggregations_result(
+        self: Box<Self>,
+        agg_with_accessor: &AggregationsWithAccessor,
+    ) -> crate::Result<IntermediateAggregationResults> {
+        let field_type = self.column_type;
+        let name = agg_with_accessor.buckets.keys[self.accessor_idx].to_string();
+        let sub_agg = &agg_with_accessor.buckets.values[self.accessor_idx].sub_aggregation;
 
         let buckets: FxHashMap<SerializedKey, IntermediateRangeBucketEntry> = self
             .buckets
@@ -189,21 +190,77 @@ impl SegmentRangeCollector {
                     range_to_string(&range_bucket.range, &field_type)?,
                     range_bucket
                         .bucket
-                        .into_intermediate_bucket_entry(&agg_with_accessor.sub_aggregation)?,
+                        .into_intermediate_bucket_entry(sub_agg)?,
                 ))
             })
             .collect::<crate::Result<_>>()?;
 
-        Ok(IntermediateBucketResult::Range(
-            IntermediateRangeBucketResult { buckets },
-        ))
+        let bucket = IntermediateBucketResult::Range(IntermediateRangeBucketResult {
+            buckets,
+            column_type: Some(self.column_type),
+        });
+
+        let buckets = Some(VecWithNames::from_entries(vec![(name, bucket)]));
+
+        Ok(IntermediateAggregationResults {
+            metrics: None,
+            buckets,
+        })
     }
 
+    fn collect(
+        &mut self,
+        doc: crate::DocId,
+        agg_with_accessor: &AggregationsWithAccessor,
+    ) -> crate::Result<()> {
+        self.collect_block(&[doc], agg_with_accessor)
+    }
+
+    fn collect_block(
+        &mut self,
+        docs: &[crate::DocId],
+        agg_with_accessor: &AggregationsWithAccessor,
+    ) -> crate::Result<()> {
+        let accessor = &agg_with_accessor.buckets.values[self.accessor_idx].accessor;
+        let sub_aggregation_accessor =
+            &agg_with_accessor.buckets.values[self.accessor_idx].sub_aggregation;
+        for doc in docs {
+            for val in accessor.values_for_doc(*doc) {
+                let bucket_pos = self.get_bucket_pos(val);
+
+                let bucket = &mut self.buckets[bucket_pos];
+
+                bucket.bucket.doc_count += 1;
+                if let Some(sub_aggregation) = &mut bucket.bucket.sub_aggregation {
+                    sub_aggregation.collect(*doc, sub_aggregation_accessor)?;
+                }
+            }
+        }
+
+        Ok(())
+    }
+
+    fn flush(&mut self, agg_with_accessor: &AggregationsWithAccessor) -> crate::Result<()> {
+        let sub_aggregation_accessor =
+            &agg_with_accessor.buckets.values[self.accessor_idx].sub_aggregation;
+
+        for bucket in self.buckets.iter_mut() {
+            if let Some(sub_agg) = bucket.bucket.sub_aggregation.as_mut() {
+                sub_agg.flush(sub_aggregation_accessor)?;
+            }
+        }
+
+        Ok(())
+    }
+}
+
+impl SegmentRangeCollector {
     pub(crate) fn from_req_and_validate(
         req: &RangeAggregation,
         sub_aggregation: &AggregationsWithAccessor,
         bucket_count: &BucketCount,
-        field_type: Type,
+        field_type: ColumnType,
+        accessor_idx: usize,
     ) -> crate::Result<Self> {
         // The range input on the request is f64.
         // We need to convert to u64 ranges, because we read the values as u64.
@@ -229,11 +286,7 @@ impl SegmentRangeCollector {
                 let sub_aggregation = if sub_aggregation.is_empty() {
                     None
                 } else {
-                    Some(
-                        GenericSegmentAggregationResultsCollector::from_req_and_validate(
-                            sub_aggregation,
-                        )?,
-                    )
+                    Some(build_segment_agg_collector(sub_aggregation)?)
                 };
 
                 Ok(SegmentRangeAndBucketEntry {
@@ -254,52 +307,11 @@ impl SegmentRangeCollector {
 
         Ok(SegmentRangeCollector {
             buckets,
-            field_type,
+            column_type: field_type,
+            accessor_idx,
         })
     }
 
-    #[inline]
-    pub(crate) fn collect_block(
-        &mut self,
-        docs: &[DocId],
-        bucket_with_accessor: &BucketAggregationWithAccessor,
-        force_flush: bool,
-    ) -> crate::Result<()> {
-        let accessor = &bucket_with_accessor.accessor;
-        for doc in docs {
-            for val in accessor.values(*doc) {
-                let bucket_pos = self.get_bucket_pos(val);
-                self.increment_bucket(bucket_pos, *doc, &bucket_with_accessor.sub_aggregation)?;
-            }
-        }
-
-        if force_flush {
-            for bucket in &mut self.buckets {
-                if let Some(sub_aggregation) = &mut bucket.bucket.sub_aggregation {
-                    sub_aggregation
-                        .flush_staged_docs(&bucket_with_accessor.sub_aggregation, force_flush)?;
-                }
-            }
-        }
-        Ok(())
-    }
-
-    #[inline]
-    fn increment_bucket(
-        &mut self,
-        bucket_pos: usize,
-        doc: DocId,
-        bucket_with_accessor: &AggregationsWithAccessor,
-    ) -> crate::Result<()> {
-        let bucket = &mut self.buckets[bucket_pos];
-
-        bucket.bucket.doc_count += 1;
-        if let Some(sub_aggregation) = &mut bucket.bucket.sub_aggregation {
-            sub_aggregation.collect(doc, bucket_with_accessor)?;
-        }
-        Ok(())
-    }
-
     #[inline]
     fn get_bucket_pos(&self, val: u64) -> usize {
         let pos = self
@@ -325,7 +337,7 @@ impl SegmentRangeCollector {
 /// more computational expensive when many documents are hit.
 fn to_u64_range(
     range: &RangeAggregationRange,
-    field_type: &Type,
+    field_type: &ColumnType,
 ) -> crate::Result<InternalRangeAggregationRange> {
     let start = if let Some(from) = range.from {
         f64_to_fastfield_u64(from, field_type)
@@ -351,7 +363,7 @@ fn to_u64_range(
 /// beginning and end and filling gaps.
 fn extend_validate_ranges(
     buckets: &[RangeAggregationRange],
-    field_type: &Type,
+    field_type: &ColumnType,
 ) -> crate::Result<Vec<InternalRangeAggregationRange>> {
     let mut converted_buckets = buckets
         .iter()
@@ -393,13 +405,16 @@ fn extend_validate_ranges(
     Ok(converted_buckets)
 }
 
-pub(crate) fn range_to_string(range: &Range<u64>, field_type: &Type) -> crate::Result<String> {
+pub(crate) fn range_to_string(
+    range: &Range<u64>,
+    field_type: &ColumnType,
+) -> crate::Result<String> {
     // is_start is there for malformed requests, e.g. ig the user passes the range u64::MIN..0.0,
     // it should be rendered as "*-0" and not "*-*"
     let to_str = |val: u64, is_start: bool| {
         if (is_start && val == u64::MIN) || (!is_start && val == u64::MAX) {
             Ok("*".to_string())
-        } else if *field_type == Type::Date {
+        } else if *field_type == ColumnType::DateTime {
             let val = i64::from_u64(val);
             format_date(val)
         } else {
@@ -414,7 +429,7 @@ pub(crate) fn range_to_string(range: &Range<u64>, field_type: &Type) -> crate::R
     ))
 }
 
-pub(crate) fn range_to_key(range: &Range<u64>, field_type: &Type) -> crate::Result<Key> {
+pub(crate) fn range_to_key(range: &Range<u64>, field_type: &ColumnType) -> crate::Result<Key> {
     Ok(Key::Str(range_to_string(range, field_type)?))
 }
 
@@ -426,8 +441,9 @@ mod tests {
 
     use super::*;
     use crate::aggregation::agg_req::{
-        Aggregation, Aggregations, BucketAggregation, BucketAggregationType,
+        Aggregation, Aggregations, BucketAggregation, BucketAggregationType, MetricAggregation,
     };
+    use crate::aggregation::metric::AverageAggregation;
     use crate::aggregation::tests::{
         exec_request, exec_request_with_query, get_test_index_2_segments,
         get_test_index_with_num_docs,
@@ -435,7 +451,7 @@ mod tests {
 
     pub fn get_collector_from_ranges(
         ranges: Vec<RangeAggregationRange>,
-        field_type: Type,
+        field_type: ColumnType,
     ) -> SegmentRangeCollector {
         let req = RangeAggregation {
             field: "dummy".to_string(),
@@ -448,6 +464,7 @@ mod tests {
             &Default::default(),
             &Default::default(),
             field_type,
+            0,
         )
         .expect("unexpected error")
     }
@@ -484,6 +501,47 @@ mod tests {
         Ok(())
     }
 
+    #[test]
+    fn range_fraction_test_with_sub_agg() -> crate::Result<()> {
+        let index = get_test_index_with_num_docs(false, 100)?;
+
+        let sub_agg_req: Aggregations = vec![(
+            "score_f64".to_string(),
+            Aggregation::Metric(MetricAggregation::Average(
+                AverageAggregation::from_field_name("score_f64".to_string()),
+            )),
+        )]
+        .into_iter()
+        .collect();
+
+        let agg_req: Aggregations = vec![(
+            "range".to_string(),
+            Aggregation::Bucket(BucketAggregation {
+                bucket_agg: BucketAggregationType::Range(RangeAggregation {
+                    field: "fraction_f64".to_string(),
+                    ranges: vec![(0f64..0.1f64).into(), (0.1f64..0.2f64).into()],
+                    ..Default::default()
+                }),
+                sub_aggregation: sub_agg_req,
+            }),
+        )]
+        .into_iter()
+        .collect();
+
+        let res = exec_request_with_query(agg_req, &index, None)?;
+
+        assert_eq!(res["range"]["buckets"][0]["key"], "*-0");
+        assert_eq!(res["range"]["buckets"][0]["doc_count"], 0);
+        assert_eq!(res["range"]["buckets"][1]["key"], "0-0.1");
+        assert_eq!(res["range"]["buckets"][1]["doc_count"], 10);
+        assert_eq!(res["range"]["buckets"][2]["key"], "0.1-0.2");
+        assert_eq!(res["range"]["buckets"][2]["doc_count"], 10);
+        assert_eq!(res["range"]["buckets"][3]["key"], "0.2-*");
+        assert_eq!(res["range"]["buckets"][3]["doc_count"], 80);
+
+        Ok(())
+    }
+
     #[test]
     fn range_keyed_buckets_test() -> crate::Result<()> {
         let index = get_test_index_with_num_docs(false, 100)?;
@@ -683,7 +741,7 @@ mod tests {
     #[test]
     fn bucket_test_extend_range_hole() {
         let buckets = vec![(10f64..20f64).into(), (30f64..40f64).into()];
-        let collector = get_collector_from_ranges(buckets, Type::F64);
+        let collector = get_collector_from_ranges(buckets, ColumnType::F64);
 
         let buckets = collector.buckets;
         assert_eq!(buckets[0].range.start, u64::MIN);
@@ -706,7 +764,7 @@ mod tests {
             (10f64..20f64).into(),
             (20f64..f64::MAX).into(),
         ];
-        let collector = get_collector_from_ranges(buckets, Type::F64);
+        let collector = get_collector_from_ranges(buckets, ColumnType::F64);
 
         let buckets = collector.buckets;
         assert_eq!(buckets[0].range.start, u64::MIN);
@@ -721,7 +779,7 @@ mod tests {
     #[test]
     fn bucket_range_test_negative_vals() {
         let buckets = vec![(-10f64..-1f64).into()];
-        let collector = get_collector_from_ranges(buckets, Type::F64);
+        let collector = get_collector_from_ranges(buckets, ColumnType::F64);
 
         let buckets = collector.buckets;
         assert_eq!(&buckets[0].bucket.key.to_string(), "*--10");
@@ -730,7 +788,7 @@ mod tests {
     #[test]
     fn bucket_range_test_positive_vals() {
         let buckets = vec![(0f64..10f64).into()];
-        let collector = get_collector_from_ranges(buckets, Type::F64);
+        let collector = get_collector_from_ranges(buckets, ColumnType::F64);
 
         let buckets = collector.buckets;
         assert_eq!(&buckets[0].bucket.key.to_string(), "*-0");
@@ -740,7 +798,7 @@ mod tests {
     #[test]
     fn range_binary_search_test_u64() {
         let check_ranges = |ranges: Vec<RangeAggregationRange>| {
-            let collector = get_collector_from_ranges(ranges, Type::U64);
+            let collector = get_collector_from_ranges(ranges, ColumnType::U64);
             let search = |val: u64| collector.get_bucket_pos(val);
 
             assert_eq!(search(u64::MIN), 0);
@@ -786,7 +844,7 @@ mod tests {
     fn range_binary_search_test_f64() {
         let ranges = vec![(10.0..100.0).into()];
 
-        let collector = get_collector_from_ranges(ranges, Type::F64);
+        let collector = get_collector_from_ranges(ranges, ColumnType::F64);
         let search = |val: u64| collector.get_bucket_pos(val);
 
         assert_eq!(search(u64::MIN), 0);
@@ -821,7 +879,7 @@ mod bench {
             buckets.push((bucket_start..bucket_start + bucket_size as f64).into())
         }
 
-        get_collector_from_ranges(buckets, Type::U64)
+        get_collector_from_ranges(buckets, ColumnType::U64)
     }
 
     fn get_rand_docs(total_docs: u64, num_docs_returned: u64) -> Vec<u64> {

src/aggregation/bucket/term_agg.rs

@@ -1,7 +1,6 @@
 use std::fmt::Debug;
 
-use columnar::Column;
-use itertools::Itertools;
+use columnar::{Cardinality, ColumnType};
 use rustc_hash::FxHashMap;
 use serde::{Deserialize, Serialize};
 
@@ -10,15 +9,15 @@ use crate::aggregation::agg_req_with_accessor::{
     AggregationsWithAccessor, BucketAggregationWithAccessor,
 };
 use crate::aggregation::intermediate_agg_result::{
-    IntermediateBucketResult, IntermediateTermBucketEntry, IntermediateTermBucketResult,
+    IntermediateAggregationResults, IntermediateBucketResult, IntermediateTermBucketEntry,
+    IntermediateTermBucketResult,
 };
 use crate::aggregation::segment_agg_result::{
-    build_segment_agg_collector, GenericSegmentAggregationResultsCollector,
-    SegmentAggregationCollector,
+    build_segment_agg_collector, SegmentAggregationCollector,
 };
+use crate::aggregation::{f64_from_fastfield_u64, Key, VecWithNames};
 use crate::error::DataCorruption;
-use crate::schema::Type;
-use crate::{DocId, TantivyError};
+use crate::TantivyError;
 
 /// Creates a bucket for every unique term and counts the number of occurences.
 /// Note that doc_count in the response buckets equals term count here.
@@ -26,6 +25,10 @@ use crate::{DocId, TantivyError};
 /// If the text is untokenized and single value, that means one term per document and therefore it
 /// is in fact doc count.
 ///
+/// ## Prerequisite
+/// Term aggregations work only on [fast fields](`crate::fastfield`) of type `u64`, `f64`, `i64` and
+/// text.
+///
 /// ### Terminology
 /// Shard parameters are supposed to be equivalent to elasticsearch shard parameter.
 /// Since they are
@@ -78,9 +81,9 @@ use crate::{DocId, TantivyError};
 ///     ...
 ///     "aggregations": {
 ///         "genres": {
-///             "doc_count_error_upper_bound": 0,   
-///             "sum_other_doc_count": 0,           
-///             "buckets": [                        
+///             "doc_count_error_upper_bound": 0,
+///             "sum_other_doc_count": 0,
+///             "buckets": [
 ///                 { "key": "drumnbass", "doc_count": 6 },
 ///                 { "key": "raggae", "doc_count": 4 },
 ///                 { "key": "jazz", "doc_count": 2 }
@@ -200,9 +203,9 @@ impl TermsAggregationInternal {
 }
 
 #[derive(Clone, Debug, Default)]
-/// Container to store term_ids and their buckets.
+/// Container to store term_ids/or u64 values and their buckets.
 struct TermBuckets {
-    pub(crate) entries: FxHashMap<u32, TermBucketEntry>,
+    pub(crate) entries: FxHashMap<u64, TermBucketEntry>,
 }
 
 #[derive(Clone, Default)]
@@ -245,19 +248,10 @@ impl TermBucketEntry {
 }
 
 impl TermBuckets {
-    pub(crate) fn from_req_and_validate(
-        sub_aggregation: &AggregationsWithAccessor,
-        _max_term_id: usize,
-    ) -> crate::Result<Self> {
-        Ok(TermBuckets {
-            entries: Default::default(),
-        })
-    }
-
     fn force_flush(&mut self, agg_with_accessor: &AggregationsWithAccessor) -> crate::Result<()> {
         for entry in &mut self.entries.values_mut() {
             if let Some(sub_aggregations) = entry.sub_aggregations.as_mut() {
-                sub_aggregations.flush_staged_docs(agg_with_accessor, false)?;
+                sub_aggregations.flush(agg_with_accessor)?;
             }
         }
         Ok(())
@@ -272,6 +266,8 @@ pub struct SegmentTermCollector {
     term_buckets: TermBuckets,
     req: TermsAggregationInternal,
     blueprint: Option<Box<dyn SegmentAggregationCollector>>,
+    field_type: ColumnType,
+    accessor_idx: usize,
 }
 
 pub(crate) fn get_agg_name_and_property(name: &str) -> (&str, &str) {
@@ -279,10 +275,86 @@ pub(crate) fn get_agg_name_and_property(name: &str) -> (&str, &str) {
     (agg_name, agg_property)
 }
 
+impl SegmentAggregationCollector for SegmentTermCollector {
+    fn into_intermediate_aggregations_result(
+        self: Box<Self>,
+        agg_with_accessor: &AggregationsWithAccessor,
+    ) -> crate::Result<IntermediateAggregationResults> {
+        let name = agg_with_accessor.buckets.keys[self.accessor_idx].to_string();
+        let agg_with_accessor = &agg_with_accessor.buckets.values[self.accessor_idx];
+
+        let bucket = self.into_intermediate_bucket_result(agg_with_accessor)?;
+        let buckets = Some(VecWithNames::from_entries(vec![(name, bucket)]));
+
+        Ok(IntermediateAggregationResults {
+            metrics: None,
+            buckets,
+        })
+    }
+
+    fn collect(
+        &mut self,
+        doc: crate::DocId,
+        agg_with_accessor: &AggregationsWithAccessor,
+    ) -> crate::Result<()> {
+        self.collect_block(&[doc], agg_with_accessor)
+    }
+
+    fn collect_block(
+        &mut self,
+        docs: &[crate::DocId],
+        agg_with_accessor: &AggregationsWithAccessor,
+    ) -> crate::Result<()> {
+        let accessor = &agg_with_accessor.buckets.values[self.accessor_idx].accessor;
+        let sub_aggregation_accessor =
+            &agg_with_accessor.buckets.values[self.accessor_idx].sub_aggregation;
+
+        if accessor.get_cardinality() == Cardinality::Full {
+            for doc in docs {
+                let term_id = accessor.values.get_val(*doc);
+                let entry = self
+                    .term_buckets
+                    .entries
+                    .entry(term_id)
+                    .or_insert_with(|| TermBucketEntry::from_blueprint(&self.blueprint));
+                entry.doc_count += 1;
+                if let Some(sub_aggregations) = entry.sub_aggregations.as_mut() {
+                    sub_aggregations.collect(*doc, sub_aggregation_accessor)?;
+                }
+            }
+        } else {
+            for doc in docs {
+                for term_id in accessor.values_for_doc(*doc) {
+                    let entry = self
+                        .term_buckets
+                        .entries
+                        .entry(term_id)
+                        .or_insert_with(|| TermBucketEntry::from_blueprint(&self.blueprint));
+                    entry.doc_count += 1;
+                    if let Some(sub_aggregations) = entry.sub_aggregations.as_mut() {
+                        sub_aggregations.collect(*doc, sub_aggregation_accessor)?;
+                    }
+                }
+            }
+        }
+        Ok(())
+    }
+
+    fn flush(&mut self, agg_with_accessor: &AggregationsWithAccessor) -> crate::Result<()> {
+        let sub_aggregation_accessor =
+            &agg_with_accessor.buckets.values[self.accessor_idx].sub_aggregation;
+
+        self.term_buckets.force_flush(sub_aggregation_accessor)?;
+        Ok(())
+    }
+}
+
 impl SegmentTermCollector {
     pub(crate) fn from_req_and_validate(
         req: &TermsAggregation,
         sub_aggregations: &AggregationsWithAccessor,
+        field_type: ColumnType,
+        accessor_idx: usize,
     ) -> crate::Result<Self> {
         let term_buckets = TermBuckets::default();
 
@@ -312,6 +384,8 @@ impl SegmentTermCollector {
             req: TermsAggregationInternal::from_req(req),
             term_buckets,
             blueprint,
+            field_type,
+            accessor_idx,
         })
     }
 
@@ -319,10 +393,9 @@ impl SegmentTermCollector {
         self,
         agg_with_accessor: &BucketAggregationWithAccessor,
     ) -> crate::Result<IntermediateBucketResult> {
-        let mut entries: Vec<(u32, TermBucketEntry)> =
+        let mut entries: Vec<(u64, TermBucketEntry)> =
             self.term_buckets.entries.into_iter().collect();
 
-        let order_by_key = self.req.order.target == OrderTarget::Key;
         let order_by_sub_aggregation =
             matches!(self.req.order.target, OrderTarget::SubAggregation(_));
 
@@ -351,61 +424,58 @@ impl SegmentTermCollector {
             }
         }
 
-        let (term_doc_count_before_cutoff, mut sum_other_doc_count) = if order_by_sub_aggregation {
+        let (term_doc_count_before_cutoff, sum_other_doc_count) = if order_by_sub_aggregation {
             (0, 0)
         } else {
             cut_off_buckets(&mut entries, self.req.segment_size as usize)
         };
 
-        let inverted_index = agg_with_accessor
-            .str_dict_column
-            .as_ref()
-            .expect("internal error: inverted index not loaded for term aggregation");
-        let term_dict = inverted_index;
+        let mut dict: FxHashMap<Key, IntermediateTermBucketEntry> = Default::default();
+        dict.reserve(entries.len());
+        if self.field_type == ColumnType::Str {
+            let term_dict = agg_with_accessor
+                .str_dict_column
+                .as_ref()
+                .expect("internal error: term dictionary not found for term aggregation");
 
-        let mut dict: FxHashMap<String, IntermediateTermBucketEntry> = Default::default();
-        let mut buffer = String::new();
-        for (term_id, entry) in entries {
-            if !term_dict.ord_to_str(term_id as u64, &mut buffer)? {
-                return Err(TantivyError::InternalError(format!(
-                    "Couldn't find term_id {} in dict",
-                    term_id
-                )));
-            }
-            dict.insert(
-                buffer.to_string(),
-                entry.into_intermediate_bucket_entry(&agg_with_accessor.sub_aggregation)?,
-            );
-        }
-        if self.req.min_doc_count == 0 {
-            // TODO: Handle rev streaming for descending sorting by keys
-            let mut stream = term_dict.dictionary().stream()?;
-            while let Some((key, _ord)) = stream.next() {
-                if dict.len() >= self.req.segment_size as usize {
-                    break;
+            let mut buffer = String::new();
+            for (term_id, entry) in entries {
+                if !term_dict.ord_to_str(term_id, &mut buffer)? {
+                    return Err(TantivyError::InternalError(format!(
+                        "Couldn't find term_id {} in dict",
+                        term_id
+                    )));
                 }
+                dict.insert(
+                    Key::Str(buffer.to_string()),
+                    entry.into_intermediate_bucket_entry(&agg_with_accessor.sub_aggregation)?,
+                );
+            }
+            if self.req.min_doc_count == 0 {
+                // TODO: Handle rev streaming for descending sorting by keys
+                let mut stream = term_dict.dictionary().stream()?;
+                while let Some((key, _ord)) = stream.next() {
+                    if dict.len() >= self.req.segment_size as usize {
+                        break;
+                    }
 
-                let key = std::str::from_utf8(key)
-                    .map_err(|utf8_err| DataCorruption::comment_only(utf8_err.to_string()))?;
-                if !dict.contains_key(key) {
-                    dict.insert(key.to_owned(), Default::default());
+                    let key = Key::Str(
+                        std::str::from_utf8(key)
+                            .map_err(|utf8_err| DataCorruption::comment_only(utf8_err.to_string()))?
+                            .to_string(),
+                    );
+                    dict.entry(key).or_default();
                 }
             }
-        }
-
-        if order_by_key {
-            let mut dict_entries = dict.into_iter().collect_vec();
-            if self.req.order.order == Order::Desc {
-                dict_entries.sort_unstable_by(|(key1, _), (key2, _)| key1.cmp(key2));
-            } else {
-                dict_entries.sort_unstable_by(|(key1, _), (key2, _)| key2.cmp(key1));
+        } else {
+            for (val, entry) in entries {
+                let val = f64_from_fastfield_u64(val, &self.field_type);
+                dict.insert(
+                    Key::F64(val),
+                    entry.into_intermediate_bucket_entry(&agg_with_accessor.sub_aggregation)?,
+                );
             }
-            let (_, sum_other_docs) =
-                cut_off_buckets(&mut dict_entries, self.req.segment_size as usize);
-
-            sum_other_doc_count += sum_other_docs;
-            dict = dict_entries.into_iter().collect();
-        }
+        };
 
         Ok(IntermediateBucketResult::Terms(
             IntermediateTermBucketResult {
@@ -415,36 +485,6 @@ impl SegmentTermCollector {
             },
         ))
     }
-
-    #[inline]
-    pub(crate) fn collect_block(
-        &mut self,
-        docs: &[DocId],
-        bucket_with_accessor: &BucketAggregationWithAccessor,
-        force_flush: bool,
-    ) -> crate::Result<()> {
-        let accessor = &bucket_with_accessor.accessor;
-
-        for doc in docs {
-            for term_id in accessor.values(*doc) {
-                let entry = self
-                    .term_buckets
-                    .entries
-                    .entry(term_id as u32)
-                    .or_insert_with(|| TermBucketEntry::from_blueprint(&self.blueprint));
-                entry.doc_count += 1;
-                if let Some(sub_aggregations) = entry.sub_aggregations.as_mut() {
-                    sub_aggregations.collect(*doc, &bucket_with_accessor.sub_aggregation)?;
-                }
-            }
-        }
-
-        if force_flush {
-            self.term_buckets
-                .force_flush(&bucket_with_accessor.sub_aggregation)?;
-        }
-        Ok(())
-    }
 }
 
 pub(crate) trait GetDocCount {
@@ -455,6 +495,11 @@ impl GetDocCount for (u32, TermBucketEntry) {
         self.1.doc_count
     }
 }
+impl GetDocCount for (u64, TermBucketEntry) {
+    fn doc_count(&self) -> u64 {
+        self.1.doc_count
+    }
+}
 impl GetDocCount for (String, IntermediateTermBucketEntry) {
     fn doc_count(&self) -> u64 {
         self.1.doc_count
@@ -483,8 +528,7 @@ pub(crate) fn cut_off_buckets<T: GetDocCount + Debug>(
 mod tests {
     use super::*;
     use crate::aggregation::agg_req::{
-        get_term_dict_field_names, Aggregation, Aggregations, BucketAggregation,
-        BucketAggregationType, MetricAggregation,
+        Aggregation, Aggregations, BucketAggregation, BucketAggregationType, MetricAggregation,
     };
     use crate::aggregation::metric::{AverageAggregation, StatsAggregation};
     use crate::aggregation::tests::{
@@ -585,12 +629,6 @@ mod tests {
             serde_json::Value::Null
         );
         assert_eq!(res["my_texts"]["sum_other_doc_count"], 0); // TODO sum_other_doc_count with min_doc_count
-
-        assert_eq!(
-            get_term_dict_field_names(&agg_req),
-            vec!["string_id".to_string(),].into_iter().collect()
-        );
-
         Ok(())
     }
 
@@ -605,7 +643,8 @@ mod tests {
     fn terms_aggregation_test_order_count_merge_segment(merge_segments: bool) -> crate::Result<()> {
         let segment_and_terms = vec![
             vec![(5.0, "terma".to_string())],
-            vec![(4.0, "termb".to_string())],
+            vec![(2.0, "termb".to_string())],
+            vec![(2.0, "terma".to_string())],
             vec![(1.0, "termc".to_string())],
             vec![(1.0, "termc".to_string())],
             vec![(1.0, "termc".to_string())],
@@ -646,7 +685,7 @@ mod tests {
                     }),
                     ..Default::default()
                 }),
-                sub_aggregation: sub_agg,
+                sub_aggregation: sub_agg.clone(),
             }),
         )]
         .into_iter()
@@ -655,15 +694,114 @@ mod tests {
         let res = exec_request(agg_req, &index)?;
         assert_eq!(res["my_texts"]["buckets"][0]["key"], "termb");
         assert_eq!(res["my_texts"]["buckets"][0]["doc_count"], 2);
+        assert_eq!(res["my_texts"]["buckets"][0]["avg_score"]["value"], 5.0);
 
         assert_eq!(res["my_texts"]["buckets"][1]["key"], "termc");
         assert_eq!(res["my_texts"]["buckets"][1]["doc_count"], 3);
+        assert_eq!(res["my_texts"]["buckets"][1]["avg_score"]["value"], 1.0);
 
         assert_eq!(res["my_texts"]["buckets"][2]["key"], "terma");
-        assert_eq!(res["my_texts"]["buckets"][2]["doc_count"], 5);
+        assert_eq!(res["my_texts"]["buckets"][2]["doc_count"], 6);
+        assert_eq!(res["my_texts"]["buckets"][2]["avg_score"]["value"], 4.5);
 
         assert_eq!(res["my_texts"]["sum_other_doc_count"], 0);
 
+        // Agg on non string
+        //
+        let agg_req: Aggregations = vec![
+            (
+                "my_scores1".to_string(),
+                Aggregation::Bucket(BucketAggregation {
+                    bucket_agg: BucketAggregationType::Terms(TermsAggregation {
+                        field: "score".to_string(),
+                        order: Some(CustomOrder {
+                            order: Order::Asc,
+                            target: OrderTarget::Count,
+                        }),
+                        ..Default::default()
+                    }),
+                    sub_aggregation: sub_agg.clone(),
+                }),
+            ),
+            (
+                "my_scores2".to_string(),
+                Aggregation::Bucket(BucketAggregation {
+                    bucket_agg: BucketAggregationType::Terms(TermsAggregation {
+                        field: "score_f64".to_string(),
+                        order: Some(CustomOrder {
+                            order: Order::Asc,
+                            target: OrderTarget::Count,
+                        }),
+                        ..Default::default()
+                    }),
+                    sub_aggregation: sub_agg.clone(),
+                }),
+            ),
+            (
+                "my_scores3".to_string(),
+                Aggregation::Bucket(BucketAggregation {
+                    bucket_agg: BucketAggregationType::Terms(TermsAggregation {
+                        field: "score_i64".to_string(),
+                        order: Some(CustomOrder {
+                            order: Order::Asc,
+                            target: OrderTarget::Count,
+                        }),
+                        ..Default::default()
+                    }),
+                    sub_aggregation: sub_agg,
+                }),
+            ),
+        ]
+        .into_iter()
+        .collect();
+
+        let res = exec_request(agg_req, &index)?;
+        assert_eq!(res["my_scores1"]["buckets"][0]["key"], 8.0);
+        assert_eq!(res["my_scores1"]["buckets"][0]["doc_count"], 1);
+        assert_eq!(res["my_scores1"]["buckets"][0]["avg_score"]["value"], 8.0);
+
+        assert_eq!(res["my_scores1"]["buckets"][1]["key"], 2.0);
+        assert_eq!(res["my_scores1"]["buckets"][1]["doc_count"], 2);
+        assert_eq!(res["my_scores1"]["buckets"][1]["avg_score"]["value"], 2.0);
+
+        assert_eq!(res["my_scores1"]["buckets"][2]["key"], 1.0);
+        assert_eq!(res["my_scores1"]["buckets"][2]["doc_count"], 3);
+        assert_eq!(res["my_scores1"]["buckets"][2]["avg_score"]["value"], 1.0);
+
+        assert_eq!(res["my_scores1"]["buckets"][3]["key"], 5.0);
+        assert_eq!(res["my_scores1"]["buckets"][3]["doc_count"], 5);
+        assert_eq!(res["my_scores1"]["buckets"][3]["avg_score"]["value"], 5.0);
+
+        assert_eq!(res["my_scores1"]["sum_other_doc_count"], 0);
+
+        assert_eq!(res["my_scores2"]["buckets"][0]["key"], 8.0);
+        assert_eq!(res["my_scores2"]["buckets"][0]["doc_count"], 1);
+        assert_eq!(res["my_scores2"]["buckets"][0]["avg_score"]["value"], 8.0);
+
+        assert_eq!(res["my_scores2"]["buckets"][1]["key"], 2.0);
+        assert_eq!(res["my_scores2"]["buckets"][1]["doc_count"], 2);
+        assert_eq!(res["my_scores2"]["buckets"][1]["avg_score"]["value"], 2.0);
+
+        assert_eq!(res["my_scores2"]["buckets"][2]["key"], 1.0);
+        assert_eq!(res["my_scores2"]["buckets"][2]["doc_count"], 3);
+        assert_eq!(res["my_scores2"]["buckets"][2]["avg_score"]["value"], 1.0);
+
+        assert_eq!(res["my_scores2"]["sum_other_doc_count"], 0);
+
+        assert_eq!(res["my_scores3"]["buckets"][0]["key"], 8.0);
+        assert_eq!(res["my_scores3"]["buckets"][0]["doc_count"], 1);
+        assert_eq!(res["my_scores3"]["buckets"][0]["avg_score"]["value"], 8.0);
+
+        assert_eq!(res["my_scores3"]["buckets"][1]["key"], 2.0);
+        assert_eq!(res["my_scores3"]["buckets"][1]["doc_count"], 2);
+        assert_eq!(res["my_scores3"]["buckets"][1]["avg_score"]["value"], 2.0);
+
+        assert_eq!(res["my_scores3"]["buckets"][2]["key"], 1.0);
+        assert_eq!(res["my_scores3"]["buckets"][2]["doc_count"], 3);
+        assert_eq!(res["my_scores3"]["buckets"][2]["avg_score"]["value"], 1.0);
+
+        assert_eq!(res["my_scores3"]["sum_other_doc_count"], 0);
+
         Ok(())
     }
 
@@ -857,14 +995,14 @@ mod tests {
         ];
         let index = get_test_index_from_values_and_terms(merge_segments, &segment_and_terms)?;
 
-        // key desc
+        // key asc
         let agg_req: Aggregations = vec![(
             "my_texts".to_string(),
             Aggregation::Bucket(BucketAggregation {
                 bucket_agg: BucketAggregationType::Terms(TermsAggregation {
                     field: "string_id".to_string(),
                     order: Some(CustomOrder {
-                        order: Order::Desc,
+                        order: Order::Asc,
                         target: OrderTarget::Key,
                     }),
                     ..Default::default()
@@ -891,7 +1029,7 @@ mod tests {
                 bucket_agg: BucketAggregationType::Terms(TermsAggregation {
                     field: "string_id".to_string(),
                     order: Some(CustomOrder {
-                        order: Order::Desc,
+                        order: Order::Asc,
                         target: OrderTarget::Key,
                     }),
                     size: Some(2),
@@ -915,14 +1053,14 @@ mod tests {
 
         assert_eq!(res["my_texts"]["sum_other_doc_count"], 3);
 
-        // key desc and segment_size cut_off
+        // key asc and segment_size cut_off
         let agg_req: Aggregations = vec![(
             "my_texts".to_string(),
             Aggregation::Bucket(BucketAggregation {
                 bucket_agg: BucketAggregationType::Terms(TermsAggregation {
                     field: "string_id".to_string(),
                     order: Some(CustomOrder {
-                        order: Order::Desc,
+                        order: Order::Asc,
                         target: OrderTarget::Key,
                     }),
                     size: Some(2),
@@ -945,14 +1083,14 @@ mod tests {
             serde_json::Value::Null
         );
 
-        // key asc
+        // key desc
         let agg_req: Aggregations = vec![(
             "my_texts".to_string(),
             Aggregation::Bucket(BucketAggregation {
                 bucket_agg: BucketAggregationType::Terms(TermsAggregation {
                     field: "string_id".to_string(),
                     order: Some(CustomOrder {
-                        order: Order::Asc,
+                        order: Order::Desc,
                         target: OrderTarget::Key,
                     }),
                     ..Default::default()
@@ -972,14 +1110,14 @@ mod tests {
         assert_eq!(res["my_texts"]["buckets"][2]["doc_count"], 5);
         assert_eq!(res["my_texts"]["sum_other_doc_count"], 0);
 
-        // key asc, size cut_off
+        // key desc, size cut_off
         let agg_req: Aggregations = vec![(
             "my_texts".to_string(),
             Aggregation::Bucket(BucketAggregation {
                 bucket_agg: BucketAggregationType::Terms(TermsAggregation {
                     field: "string_id".to_string(),
                     order: Some(CustomOrder {
-                        order: Order::Asc,
+                        order: Order::Desc,
                         target: OrderTarget::Key,
                     }),
                     size: Some(2),
@@ -1002,14 +1140,14 @@ mod tests {
         );
         assert_eq!(res["my_texts"]["sum_other_doc_count"], 5);
 
-        // key asc, segment_size cut_off
+        // key desc, segment_size cut_off
         let agg_req: Aggregations = vec![(
             "my_texts".to_string(),
             Aggregation::Bucket(BucketAggregation {
                 bucket_agg: BucketAggregationType::Terms(TermsAggregation {
                     field: "string_id".to_string(),
                     order: Some(CustomOrder {
-                        order: Order::Asc,
+                        order: Order::Desc,
                         target: OrderTarget::Key,
                     }),
                     size: Some(2),

src/aggregation/buf_collector.rs

@@ -0,0 +1,79 @@
+use super::agg_req_with_accessor::AggregationsWithAccessor;
+use super::intermediate_agg_result::IntermediateAggregationResults;
+use super::segment_agg_result::SegmentAggregationCollector;
+use crate::DocId;
+
+pub(crate) const DOC_BLOCK_SIZE: usize = 64;
+pub(crate) type DocBlock = [DocId; DOC_BLOCK_SIZE];
+
+/// BufAggregationCollector buffers documents before calling collect_block().
+#[derive(Clone)]
+pub(crate) struct BufAggregationCollector {
+    pub(crate) collector: Box<dyn SegmentAggregationCollector>,
+    staged_docs: DocBlock,
+    num_staged_docs: usize,
+}
+
+impl std::fmt::Debug for BufAggregationCollector {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("SegmentAggregationResultsCollector")
+            .field("staged_docs", &&self.staged_docs[..self.num_staged_docs])
+            .field("num_staged_docs", &self.num_staged_docs)
+            .finish()
+    }
+}
+
+impl BufAggregationCollector {
+    pub fn new(collector: Box<dyn SegmentAggregationCollector>) -> Self {
+        Self {
+            collector,
+            num_staged_docs: 0,
+            staged_docs: [0; DOC_BLOCK_SIZE],
+        }
+    }
+}
+
+impl SegmentAggregationCollector for BufAggregationCollector {
+    fn into_intermediate_aggregations_result(
+        self: Box<Self>,
+        agg_with_accessor: &AggregationsWithAccessor,
+    ) -> crate::Result<IntermediateAggregationResults> {
+        Box::new(self.collector).into_intermediate_aggregations_result(agg_with_accessor)
+    }
+
+    fn collect(
+        &mut self,
+        doc: crate::DocId,
+        agg_with_accessor: &AggregationsWithAccessor,
+    ) -> crate::Result<()> {
+        self.staged_docs[self.num_staged_docs] = doc;
+        self.num_staged_docs += 1;
+        if self.num_staged_docs == self.staged_docs.len() {
+            self.collector
+                .collect_block(&self.staged_docs[..self.num_staged_docs], agg_with_accessor)?;
+            self.num_staged_docs = 0;
+        }
+        Ok(())
+    }
+
+    fn collect_block(
+        &mut self,
+        docs: &[crate::DocId],
+        agg_with_accessor: &AggregationsWithAccessor,
+    ) -> crate::Result<()> {
+        for doc in docs {
+            self.collect(*doc, agg_with_accessor)?;
+        }
+        Ok(())
+    }
+
+    fn flush(&mut self, agg_with_accessor: &AggregationsWithAccessor) -> crate::Result<()> {
+        self.collector
+            .collect_block(&self.staged_docs[..self.num_staged_docs], agg_with_accessor)?;
+        self.num_staged_docs = 0;
+
+        self.collector.flush(agg_with_accessor)?;
+
+        Ok(())
+    }
+}

src/aggregation/collector.rs

@@ -3,14 +3,11 @@ use std::rc::Rc;
 use super::agg_req::Aggregations;
 use super::agg_req_with_accessor::AggregationsWithAccessor;
 use super::agg_result::AggregationResults;
+use super::buf_collector::BufAggregationCollector;
 use super::intermediate_agg_result::IntermediateAggregationResults;
-use super::segment_agg_result::{
-    build_segment_agg_collector, GenericSegmentAggregationResultsCollector,
-    SegmentAggregationCollector,
-};
+use super::segment_agg_result::{build_segment_agg_collector, SegmentAggregationCollector};
 use crate::aggregation::agg_req_with_accessor::get_aggs_with_accessor_and_validate;
 use crate::collector::{Collector, SegmentCollector};
-use crate::schema::Schema;
 use crate::{SegmentReader, TantivyError};
 
 /// The default max bucket count, before the aggregation fails.
@@ -20,7 +17,6 @@ pub const MAX_BUCKET_COUNT: u32 = 65000;
 ///
 /// The collector collects all aggregations by the underlying aggregation request.
 pub struct AggregationCollector {
-    schema: Schema,
     agg: Aggregations,
     max_bucket_count: u32,
 }
@@ -30,9 +26,8 @@ impl AggregationCollector {
     ///
     /// Aggregation fails when the total bucket count is higher than max_bucket_count.
     /// max_bucket_count will default to `MAX_BUCKET_COUNT` (65000) when unset
-    pub fn from_aggs(agg: Aggregations, max_bucket_count: Option<u32>, schema: Schema) -> Self {
+    pub fn from_aggs(agg: Aggregations, max_bucket_count: Option<u32>) -> Self {
         Self {
-            schema,
             agg,
             max_bucket_count: max_bucket_count.unwrap_or(MAX_BUCKET_COUNT),
         }
@@ -119,7 +114,7 @@ impl Collector for AggregationCollector {
         segment_fruits: Vec<<Self::Child as SegmentCollector>::Fruit>,
     ) -> crate::Result<Self::Fruit> {
         let res = merge_fruits(segment_fruits)?;
-        res.into_final_bucket_result(self.agg.clone(), &self.schema)
+        res.into_final_bucket_result(self.agg.clone())
     }
 }
 
@@ -140,7 +135,7 @@ fn merge_fruits(
 /// `AggregationSegmentCollector` does the aggregation collection on a segment.
 pub struct AggregationSegmentCollector {
     aggs_with_accessor: AggregationsWithAccessor,
-    result: Box<dyn SegmentAggregationCollector>,
+    result: BufAggregationCollector,
     error: Option<TantivyError>,
 }
 
@@ -154,7 +149,8 @@ impl AggregationSegmentCollector {
     ) -> crate::Result<Self> {
         let aggs_with_accessor =
             get_aggs_with_accessor_and_validate(agg, reader, Rc::default(), max_bucket_count)?;
-        let result = build_segment_agg_collector(&aggs_with_accessor)?;
+        let result =
+            BufAggregationCollector::new(build_segment_agg_collector(&aggs_with_accessor)?);
         Ok(AggregationSegmentCollector {
             aggs_with_accessor,
             result,
@@ -180,9 +176,7 @@ impl SegmentCollector for AggregationSegmentCollector {
         if let Some(err) = self.error {
             return Err(err);
         }
-        self.result
-            .flush_staged_docs(&self.aggs_with_accessor, true)?;
-        self.result
-            .into_intermediate_aggregations_result(&self.aggs_with_accessor)
+        self.result.flush(&self.aggs_with_accessor)?;
+        Box::new(self.result).into_intermediate_aggregations_result(&self.aggs_with_accessor)
     }
 }

src/aggregation/error.rs

@@ -0,0 +1,9 @@
+use super::bucket::DateHistogramParseError;
+
+/// Error that may occur when opening a directory
+#[derive(Debug, Clone, PartialEq, Eq, Error)]
+pub enum AggregationError {
+    /// Failed to open the directory.
+    #[error("Date histogram parse error: {0:?}")]
+    DateHistogramParseError(#[from] DateHistogramParseError),
+}

src/aggregation/intermediate_agg_result.rs

@@ -4,6 +4,7 @@
 
 use std::cmp::Ordering;
 
+use columnar::ColumnType;
 use itertools::Itertools;
 use rustc_hash::FxHashMap;
 use serde::{Deserialize, Serialize};
@@ -21,11 +22,9 @@ use super::metric::{
     IntermediateAverage, IntermediateCount, IntermediateMax, IntermediateMin, IntermediateStats,
     IntermediateSum,
 };
-use super::segment_agg_result::SegmentMetricResultCollector;
 use super::{format_date, Key, SerializedKey, VecWithNames};
 use crate::aggregation::agg_result::{AggregationResults, BucketEntries, BucketEntry};
 use crate::aggregation::bucket::TermsAggregationInternal;
-use crate::schema::Schema;
 
 /// Contains the intermediate aggregation result, which is optimized to be merged with other
 /// intermediate results.
@@ -39,12 +38,8 @@ pub struct IntermediateAggregationResults {
 
 impl IntermediateAggregationResults {
     /// Convert intermediate result and its aggregation request to the final result.
-    pub fn into_final_bucket_result(
-        self,
-        req: Aggregations,
-        schema: &Schema,
-    ) -> crate::Result<AggregationResults> {
-        self.into_final_bucket_result_internal(&(req.into()), schema)
+    pub fn into_final_bucket_result(self, req: Aggregations) -> crate::Result<AggregationResults> {
+        self.into_final_bucket_result_internal(&(req.into()))
     }
 
     /// Convert intermediate result and its aggregation request to the final result.
@@ -54,7 +49,6 @@ impl IntermediateAggregationResults {
     pub(crate) fn into_final_bucket_result_internal(
         self,
         req: &AggregationsInternal,
-        schema: &Schema,
     ) -> crate::Result<AggregationResults> {
         // Important assumption:
         // When the tree contains buckets/metric, we expect it to have all buckets/metrics from the
@@ -62,11 +56,11 @@ impl IntermediateAggregationResults {
         let mut results: FxHashMap<String, AggregationResult> = FxHashMap::default();
 
         if let Some(buckets) = self.buckets {
-            convert_and_add_final_buckets_to_result(&mut results, buckets, &req.buckets, schema)?
+            convert_and_add_final_buckets_to_result(&mut results, buckets, &req.buckets)?
         } else {
             // When there are no buckets, we create empty buckets, so that the serialized json
             // format is constant
-            add_empty_final_buckets_to_result(&mut results, &req.buckets, schema)?
+            add_empty_final_buckets_to_result(&mut results, &req.buckets)?
         };
 
         if let Some(metrics) = self.metrics {
@@ -167,12 +161,10 @@ fn add_empty_final_metrics_to_result(
 fn add_empty_final_buckets_to_result(
     results: &mut FxHashMap<String, AggregationResult>,
     req_buckets: &VecWithNames<BucketAggregationInternal>,
-    schema: &Schema,
 ) -> crate::Result<()> {
     let requested_buckets = req_buckets.iter();
     for (key, req) in requested_buckets {
-        let empty_bucket =
-            AggregationResult::BucketResult(BucketResult::empty_from_req(req, schema)?);
+        let empty_bucket = AggregationResult::BucketResult(BucketResult::empty_from_req(req)?);
         results.insert(key.to_string(), empty_bucket);
     }
     Ok(())
@@ -182,13 +174,12 @@ fn convert_and_add_final_buckets_to_result(
     results: &mut FxHashMap<String, AggregationResult>,
     buckets: VecWithNames<IntermediateBucketResult>,
     req_buckets: &VecWithNames<BucketAggregationInternal>,
-    schema: &Schema,
 ) -> crate::Result<()> {
     assert_eq!(buckets.len(), req_buckets.len());
 
     let buckets_with_request = buckets.into_iter().zip(req_buckets.values());
     for ((key, bucket), req) in buckets_with_request {
-        let result = AggregationResult::BucketResult(bucket.into_final_bucket_result(req, schema)?);
+        let result = AggregationResult::BucketResult(bucket.into_final_bucket_result(req)?);
         results.insert(key, result);
     }
     Ok(())
@@ -220,32 +211,6 @@ pub enum IntermediateMetricResult {
     Sum(IntermediateSum),
 }
 
-impl From<SegmentMetricResultCollector> for IntermediateMetricResult {
-    fn from(tree: SegmentMetricResultCollector) -> Self {
-        use super::metric::SegmentStatsType;
-        match tree {
-            SegmentMetricResultCollector::Stats(collector) => match collector.collecting_for {
-                SegmentStatsType::Average => IntermediateMetricResult::Average(
-                    IntermediateAverage::from_collector(collector),
-                ),
-                SegmentStatsType::Count => {
-                    IntermediateMetricResult::Count(IntermediateCount::from_collector(collector))
-                }
-                SegmentStatsType::Max => {
-                    IntermediateMetricResult::Max(IntermediateMax::from_collector(collector))
-                }
-                SegmentStatsType::Min => {
-                    IntermediateMetricResult::Min(IntermediateMin::from_collector(collector))
-                }
-                SegmentStatsType::Stats => IntermediateMetricResult::Stats(collector.stats),
-                SegmentStatsType::Sum => {
-                    IntermediateMetricResult::Sum(IntermediateSum::from_collector(collector))
-                }
-            },
-        }
-    }
-}
-
 impl IntermediateMetricResult {
     pub(crate) fn empty_from_req(req: &MetricAggregation) -> Self {
         match req {
@@ -309,6 +274,8 @@ pub enum IntermediateBucketResult {
     /// This is the histogram entry for a bucket, which contains a key, count, and optionally
     /// sub_aggregations.
     Histogram {
+        /// The column_type of the underlying `Column`
+        column_type: Option<ColumnType>,
         /// The buckets
         buckets: Vec<IntermediateHistogramBucketEntry>,
     },
@@ -320,7 +287,6 @@ impl IntermediateBucketResult {
     pub(crate) fn into_final_bucket_result(
         self,
         req: &BucketAggregationInternal,
-        schema: &Schema,
     ) -> crate::Result<BucketResult> {
         match self {
             IntermediateBucketResult::Range(range_res) => {
@@ -330,9 +296,9 @@ impl IntermediateBucketResult {
                     .map(|bucket| {
                         bucket.into_final_bucket_entry(
                             &req.sub_aggregation,
-                            schema,
                             req.as_range()
                                 .expect("unexpected aggregation, expected histogram aggregation"),
+                            range_res.column_type,
                         )
                     })
                     .collect::<crate::Result<Vec<_>>>()?;
@@ -359,16 +325,21 @@ impl IntermediateBucketResult {
                 };
                 Ok(BucketResult::Range { buckets })
             }
-            IntermediateBucketResult::Histogram { buckets } => {
+            IntermediateBucketResult::Histogram {
+                column_type,
+                buckets,
+            } => {
+                let histogram_req = &req
+                    .as_histogram()?
+                    .expect("unexpected aggregation, expected histogram aggregation");
                 let buckets = intermediate_histogram_buckets_to_final_buckets(
                     buckets,
-                    req.as_histogram()
-                        .expect("unexpected aggregation, expected histogram aggregation"),
+                    column_type,
+                    histogram_req,
                     &req.sub_aggregation,
-                    schema,
                 )?;
 
-                let buckets = if req.as_histogram().unwrap().keyed {
+                let buckets = if histogram_req.keyed {
                     let mut bucket_map =
                         FxHashMap::with_capacity_and_hasher(buckets.len(), Default::default());
                     for bucket in buckets {
@@ -384,7 +355,6 @@ impl IntermediateBucketResult {
                 req.as_term()
                     .expect("unexpected aggregation, expected term aggregation"),
                 &req.sub_aggregation,
-                schema,
             ),
         }
     }
@@ -393,8 +363,11 @@ impl IntermediateBucketResult {
         match req {
             BucketAggregationType::Terms(_) => IntermediateBucketResult::Terms(Default::default()),
             BucketAggregationType::Range(_) => IntermediateBucketResult::Range(Default::default()),
-            BucketAggregationType::Histogram(_) => {
-                IntermediateBucketResult::Histogram { buckets: vec![] }
+            BucketAggregationType::Histogram(_) | BucketAggregationType::DateHistogram(_) => {
+                IntermediateBucketResult::Histogram {
+                    buckets: vec![],
+                    column_type: None,
+                }
             }
         }
     }
@@ -404,7 +377,7 @@ impl IntermediateBucketResult {
                 IntermediateBucketResult::Terms(term_res_left),
                 IntermediateBucketResult::Terms(term_res_right),
             ) => {
-                merge_maps(&mut term_res_left.entries, term_res_right.entries);
+                merge_key_maps(&mut term_res_left.entries, term_res_right.entries);
                 term_res_left.sum_other_doc_count += term_res_right.sum_other_doc_count;
                 term_res_left.doc_count_error_upper_bound +=
                     term_res_right.doc_count_error_upper_bound;
@@ -414,7 +387,7 @@ impl IntermediateBucketResult {
                 IntermediateBucketResult::Range(range_res_left),
                 IntermediateBucketResult::Range(range_res_right),
             ) => {
-                merge_maps(&mut range_res_left.buckets, range_res_right.buckets);
+                merge_serialized_key_maps(&mut range_res_left.buckets, range_res_right.buckets);
             }
             (
                 IntermediateBucketResult::Histogram {
@@ -460,12 +433,13 @@ impl IntermediateBucketResult {
 /// Range aggregation including error counts
 pub struct IntermediateRangeBucketResult {
     pub(crate) buckets: FxHashMap<SerializedKey, IntermediateRangeBucketEntry>,
+    pub(crate) column_type: Option<ColumnType>,
 }
 
 #[derive(Default, Clone, Debug, PartialEq, Serialize, Deserialize)]
 /// Term aggregation including error counts
 pub struct IntermediateTermBucketResult {
-    pub(crate) entries: FxHashMap<String, IntermediateTermBucketEntry>,
+    pub(crate) entries: FxHashMap<Key, IntermediateTermBucketEntry>,
     pub(crate) sum_other_doc_count: u64,
     pub(crate) doc_count_error_upper_bound: u64,
 }
@@ -475,7 +449,6 @@ impl IntermediateTermBucketResult {
         self,
         req: &TermsAggregation,
         sub_aggregation_req: &AggregationsInternal,
-        schema: &Schema,
     ) -> crate::Result<BucketResult> {
         let req = TermsAggregationInternal::from_req(req);
         let mut buckets: Vec<BucketEntry> = self
@@ -485,11 +458,11 @@ impl IntermediateTermBucketResult {
             .map(|(key, entry)| {
                 Ok(BucketEntry {
                     key_as_string: None,
-                    key: Key::Str(key),
+                    key,
                     doc_count: entry.doc_count,
                     sub_aggregation: entry
                         .sub_aggregation
-                        .into_final_bucket_result_internal(sub_aggregation_req, schema)?,
+                        .into_final_bucket_result_internal(sub_aggregation_req)?,
                 })
             })
             .collect::<crate::Result<_>>()?;
@@ -498,7 +471,7 @@ impl IntermediateTermBucketResult {
         match req.order.target {
             OrderTarget::Key => {
                 buckets.sort_by(|left, right| {
-                    if req.order.order == Order::Desc {
+                    if req.order.order == Order::Asc {
                         left.key.partial_cmp(&right.key)
                     } else {
                         right.key.partial_cmp(&left.key)
@@ -563,7 +536,7 @@ trait MergeFruits {
     fn merge_fruits(&mut self, other: Self);
 }
 
-fn merge_maps<V: MergeFruits + Clone>(
+fn merge_serialized_key_maps<V: MergeFruits + Clone>(
     entries_left: &mut FxHashMap<SerializedKey, V>,
     mut entries_right: FxHashMap<SerializedKey, V>,
 ) {
@@ -578,6 +551,21 @@ fn merge_maps<V: MergeFruits + Clone>(
     }
 }
 
+fn merge_key_maps<V: MergeFruits + Clone>(
+    entries_left: &mut FxHashMap<Key, V>,
+    mut entries_right: FxHashMap<Key, V>,
+) {
+    for (name, entry_left) in entries_left.iter_mut() {
+        if let Some(entry_right) = entries_right.remove(name) {
+            entry_left.merge_fruits(entry_right);
+        }
+    }
+
+    for (key, res) in entries_right.into_iter() {
+        entries_left.entry(key).or_insert(res);
+    }
+}
+
 /// This is the histogram entry for a bucket, which contains a key, count, and optionally
 /// sub_aggregations.
 #[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
@@ -594,7 +582,6 @@ impl IntermediateHistogramBucketEntry {
     pub(crate) fn into_final_bucket_entry(
         self,
         req: &AggregationsInternal,
-        schema: &Schema,
     ) -> crate::Result<BucketEntry> {
         Ok(BucketEntry {
             key_as_string: None,
@@ -602,7 +589,7 @@ impl IntermediateHistogramBucketEntry {
             doc_count: self.doc_count,
             sub_aggregation: self
                 .sub_aggregation
-                .into_final_bucket_result_internal(req, schema)?,
+                .into_final_bucket_result_internal(req)?,
         })
     }
 }
@@ -639,15 +626,15 @@ impl IntermediateRangeBucketEntry {
     pub(crate) fn into_final_bucket_entry(
         self,
         req: &AggregationsInternal,
-        schema: &Schema,
-        range_req: &RangeAggregation,
+        _range_req: &RangeAggregation,
+        column_type: Option<ColumnType>,
     ) -> crate::Result<RangeBucketEntry> {
         let mut range_bucket_entry = RangeBucketEntry {
             key: self.key,
             doc_count: self.doc_count,
             sub_aggregation: self
                 .sub_aggregation
-                .into_final_bucket_result_internal(req, schema)?,
+                .into_final_bucket_result_internal(req)?,
             to: self.to,
             from: self.from,
             to_as_string: None,
@@ -656,8 +643,7 @@ impl IntermediateRangeBucketEntry {
 
         // If we have a date type on the histogram buckets, we add the `key_as_string` field as
         // rfc339
-        let field = schema.get_field(&range_req.field)?;
-        if schema.get_field_entry(field).field_type().is_date() {
+        if column_type == Some(ColumnType::DateTime) {
             if let Some(val) = range_bucket_entry.to {
                 let key_as_string = format_date(val as i64)?;
                 range_bucket_entry.to_as_string = Some(key_as_string);
@@ -728,7 +714,10 @@ mod tests {
         }
         map.insert(
             "my_agg_level2".to_string(),
-            IntermediateBucketResult::Range(IntermediateRangeBucketResult { buckets }),
+            IntermediateBucketResult::Range(IntermediateRangeBucketResult {
+                buckets,
+                column_type: None,
+            }),
         );
         IntermediateAggregationResults {
             buckets: Some(VecWithNames::from_entries(map.into_iter().collect())),
@@ -758,7 +747,10 @@ mod tests {
         }
         map.insert(
             "my_agg_level1".to_string(),
-            IntermediateBucketResult::Range(IntermediateRangeBucketResult { buckets }),
+            IntermediateBucketResult::Range(IntermediateRangeBucketResult {
+                buckets,
+                column_type: None,
+            }),
         );
         IntermediateAggregationResults {
             buckets: Some(VecWithNames::from_entries(map.into_iter().collect())),

src/aggregation/metric/mod.rs

@@ -81,7 +81,7 @@ mod tests {
             "price_sum": { "sum": { "field": "price" } }
         }"#;
         let aggregations: Aggregations = serde_json::from_str(aggregations_json).unwrap();
-        let collector = AggregationCollector::from_aggs(aggregations, None, index.schema());
+        let collector = AggregationCollector::from_aggs(aggregations, None);
         let reader = index.reader().unwrap();
         let searcher = reader.searcher();
         let aggregations_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();

src/aggregation/metric/stats.rs

@@ -1,4 +1,4 @@
-use columnar::Column;
+use columnar::{Cardinality, Column, ColumnType};
 use serde::{Deserialize, Serialize};
 
 use super::*;
@@ -8,7 +8,6 @@ use crate::aggregation::intermediate_agg_result::{
 };
 use crate::aggregation::segment_agg_result::SegmentAggregationCollector;
 use crate::aggregation::{f64_from_fastfield_u64, VecWithNames};
-use crate::schema::Type;
 use crate::{DocId, TantivyError};
 
 /// A multi-value metric aggregation that computes a collection of statistics on numeric values that
@@ -153,26 +152,40 @@ pub(crate) enum SegmentStatsType {
 
 #[derive(Clone, Debug, PartialEq)]
 pub(crate) struct SegmentStatsCollector {
-    field_type: Type,
+    field_type: ColumnType,
     pub(crate) collecting_for: SegmentStatsType,
     pub(crate) stats: IntermediateStats,
+    pub(crate) accessor_idx: usize,
 }
 
 impl SegmentStatsCollector {
-    pub fn from_req(field_type: Type, collecting_for: SegmentStatsType) -> Self {
+    pub fn from_req(
+        field_type: ColumnType,
+        collecting_for: SegmentStatsType,
+        accessor_idx: usize,
+    ) -> Self {
         Self {
             field_type,
             collecting_for,
             stats: IntermediateStats::default(),
+            accessor_idx,
         }
     }
-    pub(crate) fn collect_block(&mut self, docs: &[DocId], field: &Column<u64>) {
-        // TODO special case for Required, Optional column type
-        for doc in docs {
-            for val in field.values(*doc) {
+    #[inline]
+    pub(crate) fn collect_block_with_field(&mut self, docs: &[DocId], field: &Column<u64>) {
+        if field.get_cardinality() == Cardinality::Full {
+            for doc in docs {
+                let val = field.values.get_val(*doc);
                 let val1 = f64_from_fastfield_u64(val, &self.field_type);
                 self.stats.collect(val1);
             }
+        } else {
+            for doc in docs {
+                for val in field.values_for_doc(*doc) {
+                    let val1 = f64_from_fastfield_u64(val, &self.field_type);
+                    self.stats.collect(val1);
+                }
+            }
         }
     }
 }
@@ -182,7 +195,7 @@ impl SegmentAggregationCollector for SegmentStatsCollector {
         self: Box<Self>,
         agg_with_accessor: &AggregationsWithAccessor,
     ) -> crate::Result<IntermediateAggregationResults> {
-        let name = agg_with_accessor.metrics.keys[0].to_string();
+        let name = agg_with_accessor.metrics.keys[self.accessor_idx].to_string();
 
         let intermediate_metric_result = match self.collecting_for {
             SegmentStatsType::Average => {
@@ -219,8 +232,9 @@ impl SegmentAggregationCollector for SegmentStatsCollector {
         doc: crate::DocId,
         agg_with_accessor: &AggregationsWithAccessor,
     ) -> crate::Result<()> {
-        let accessor = &agg_with_accessor.metrics.values[0].accessor;
-        for val in accessor.values(doc) {
+        let field = &agg_with_accessor.metrics.values[self.accessor_idx].accessor;
+
+        for val in field.values_for_doc(doc) {
             let val1 = f64_from_fastfield_u64(val, &self.field_type);
             self.stats.collect(val1);
         }
@@ -228,11 +242,14 @@ impl SegmentAggregationCollector for SegmentStatsCollector {
         Ok(())
     }
 
-    fn flush_staged_docs(
+    #[inline]
+    fn collect_block(
         &mut self,
-        _agg_with_accessor: &AggregationsWithAccessor,
-        _force_flush: bool,
+        docs: &[crate::DocId],
+        agg_with_accessor: &AggregationsWithAccessor,
     ) -> crate::Result<()> {
+        let field = &agg_with_accessor.metrics.values[self.accessor_idx].accessor;
+        self.collect_block_with_field(docs, field);
         Ok(())
     }
 }
@@ -272,7 +289,7 @@ mod tests {
         .into_iter()
         .collect();
 
-        let collector = AggregationCollector::from_aggs(agg_req_1, None, index.schema());
+        let collector = AggregationCollector::from_aggs(agg_req_1, None);
 
         let reader = index.reader()?;
         let searcher = reader.searcher();
@@ -293,6 +310,43 @@ mod tests {
         Ok(())
     }
 
+    #[test]
+    fn test_aggregation_stats_simple() -> crate::Result<()> {
+        // test index without segments
+        let values = vec![10.0];
+
+        let index = get_test_index_from_values(false, &values)?;
+
+        let agg_req_1: Aggregations = vec![(
+            "stats".to_string(),
+            Aggregation::Metric(MetricAggregation::Stats(StatsAggregation::from_field_name(
+                "score".to_string(),
+            ))),
+        )]
+        .into_iter()
+        .collect();
+
+        let collector = AggregationCollector::from_aggs(agg_req_1, None);
+
+        let reader = index.reader()?;
+        let searcher = reader.searcher();
+        let agg_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
+
+        let res: Value = serde_json::from_str(&serde_json::to_string(&agg_res)?)?;
+        assert_eq!(
+            res["stats"],
+            json!({
+                "avg": 10.0,
+                "count": 1,
+                "max": 10.0,
+                "min": 10.0,
+                "sum": 10.0
+            })
+        );
+
+        Ok(())
+    }
+
     #[test]
     fn test_aggregation_stats() -> crate::Result<()> {
         let index = get_test_index_2_segments(false)?;
@@ -349,7 +403,7 @@ mod tests {
         .into_iter()
         .collect();
 
-        let collector = AggregationCollector::from_aggs(agg_req_1, None, index.schema());
+        let collector = AggregationCollector::from_aggs(agg_req_1, None);
 
         let searcher = reader.searcher();
         let agg_res: AggregationResults = searcher.search(&term_query, &collector).unwrap();

src/aggregation/segment_agg_result.rs

@@ -13,17 +13,14 @@ use super::agg_req_with_accessor::{
 };
 use super::bucket::{SegmentHistogramCollector, SegmentRangeCollector, SegmentTermCollector};
 use super::collector::MAX_BUCKET_COUNT;
-use super::intermediate_agg_result::{IntermediateAggregationResults, IntermediateBucketResult};
+use super::intermediate_agg_result::IntermediateAggregationResults;
 use super::metric::{
     AverageAggregation, CountAggregation, MaxAggregation, MinAggregation, SegmentStatsCollector,
     SegmentStatsType, StatsAggregation, SumAggregation,
 };
 use super::VecWithNames;
 use crate::aggregation::agg_req::BucketAggregationType;
-use crate::{DocId, TantivyError};
-
-pub(crate) const DOC_BLOCK_SIZE: usize = 64;
-pub(crate) type DocBlock = [DocId; DOC_BLOCK_SIZE];
+use crate::TantivyError;
 
 pub(crate) trait SegmentAggregationCollector: CollectorClone + Debug {
     fn into_intermediate_aggregations_result(
@@ -37,11 +34,17 @@ pub(crate) trait SegmentAggregationCollector: CollectorClone + Debug {
         agg_with_accessor: &AggregationsWithAccessor,
     ) -> crate::Result<()>;
 
-    fn flush_staged_docs(
+    fn collect_block(
         &mut self,
+        docs: &[crate::DocId],
         agg_with_accessor: &AggregationsWithAccessor,
-        force_flush: bool,
     ) -> crate::Result<()>;
+
+    /// Finalize method. Some Aggregator collect blocks of docs before calling `collect_block`.
+    /// This method ensures those staged docs will be collected.
+    fn flush(&mut self, _agg_with_accessor: &AggregationsWithAccessor) -> crate::Result<()> {
+        Ok(())
+    }
 }
 
 pub(crate) trait CollectorClone {
@@ -68,54 +71,97 @@ pub(crate) fn build_segment_agg_collector(
     // Single metric special case
     if req.buckets.is_empty() && req.metrics.len() == 1 {
         let req = &req.metrics.values[0];
-        let stats_collector = match &req.metric {
-            MetricAggregation::Average(AverageAggregation { .. }) => {
-                SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Average)
-            }
-            MetricAggregation::Count(CountAggregation { .. }) => {
-                SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Count)
-            }
-            MetricAggregation::Max(MaxAggregation { .. }) => {
-                SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Max)
-            }
-            MetricAggregation::Min(MinAggregation { .. }) => {
-                SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Min)
-            }
-            MetricAggregation::Stats(StatsAggregation { .. }) => {
-                SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Stats)
-            }
-            MetricAggregation::Sum(SumAggregation { .. }) => {
-                SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Sum)
-            }
-        };
+        let accessor_idx = 0;
+        return build_metric_segment_agg_collector(req, accessor_idx);
+    }
 
-        return Ok(Box::new(stats_collector));
+    // Single bucket special case
+    if req.metrics.is_empty() && req.buckets.len() == 1 {
+        let req = &req.buckets.values[0];
+        let accessor_idx = 0;
+        return build_bucket_segment_agg_collector(req, accessor_idx);
     }
 
     let agg = GenericSegmentAggregationResultsCollector::from_req_and_validate(req)?;
     Ok(Box::new(agg))
 }
 
-#[derive(Clone)]
+pub(crate) fn build_metric_segment_agg_collector(
+    req: &MetricAggregationWithAccessor,
+    accessor_idx: usize,
+) -> crate::Result<Box<dyn SegmentAggregationCollector>> {
+    let stats_collector = match &req.metric {
+        MetricAggregation::Average(AverageAggregation { .. }) => {
+            SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Average, accessor_idx)
+        }
+        MetricAggregation::Count(CountAggregation { .. }) => {
+            SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Count, accessor_idx)
+        }
+        MetricAggregation::Max(MaxAggregation { .. }) => {
+            SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Max, accessor_idx)
+        }
+        MetricAggregation::Min(MinAggregation { .. }) => {
+            SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Min, accessor_idx)
+        }
+        MetricAggregation::Stats(StatsAggregation { .. }) => {
+            SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Stats, accessor_idx)
+        }
+        MetricAggregation::Sum(SumAggregation { .. }) => {
+            SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Sum, accessor_idx)
+        }
+    };
+
+    Ok(Box::new(stats_collector))
+}
+
+pub(crate) fn build_bucket_segment_agg_collector(
+    req: &BucketAggregationWithAccessor,
+    accessor_idx: usize,
+) -> crate::Result<Box<dyn SegmentAggregationCollector>> {
+    match &req.bucket_agg {
+        BucketAggregationType::Terms(terms_req) => {
+            Ok(Box::new(SegmentTermCollector::from_req_and_validate(
+                terms_req,
+                &req.sub_aggregation,
+                req.field_type,
+                accessor_idx,
+            )?))
+        }
+        BucketAggregationType::Range(range_req) => {
+            Ok(Box::new(SegmentRangeCollector::from_req_and_validate(
+                range_req,
+                &req.sub_aggregation,
+                &req.bucket_count,
+                req.field_type,
+                accessor_idx,
+            )?))
+        }
+        BucketAggregationType::Histogram(histogram) => {
+            Ok(Box::new(SegmentHistogramCollector::from_req_and_validate(
+                histogram,
+                &req.sub_aggregation,
+                req.field_type,
+                accessor_idx,
+            )?))
+        }
+        BucketAggregationType::DateHistogram(histogram) => {
+            Ok(Box::new(SegmentHistogramCollector::from_req_and_validate(
+                &histogram.to_histogram_req()?,
+                &req.sub_aggregation,
+                req.field_type,
+                accessor_idx,
+            )?))
+        }
+    }
+}
+
+#[derive(Clone, Default)]
 /// The GenericSegmentAggregationResultsCollector is the generic version of the collector, which
 /// can handle arbitrary complexity of  sub-aggregations. Ideally we never have to pick this one
 /// and can provide specialized versions instead, that remove some of its overhead.
 pub(crate) struct GenericSegmentAggregationResultsCollector {
-    pub(crate) metrics: Option<VecWithNames<SegmentMetricResultCollector>>,
-    pub(crate) buckets: Option<VecWithNames<SegmentBucketResultCollector>>,
-    staged_docs: DocBlock,
-    num_staged_docs: usize,
-}
-
-impl Default for GenericSegmentAggregationResultsCollector {
-    fn default() -> Self {
-        Self {
-            metrics: Default::default(),
-            buckets: Default::default(),
-            staged_docs: [0; DOC_BLOCK_SIZE],
-            num_staged_docs: Default::default(),
-        }
-    }
+    pub(crate) metrics: Option<Vec<Box<dyn SegmentAggregationCollector>>>,
+    pub(crate) buckets: Option<Vec<Box<dyn SegmentAggregationCollector>>>,
 }
 
 impl Debug for GenericSegmentAggregationResultsCollector {
@@ -123,8 +169,6 @@ impl Debug for GenericSegmentAggregationResultsCollector {
         f.debug_struct("SegmentAggregationResultsCollector")
             .field("metrics", &self.metrics)
             .field("buckets", &self.buckets)
-            .field("staged_docs", &&self.staged_docs[..self.num_staged_docs])
-            .field("num_staged_docs", &self.num_staged_docs)
             .finish()
     }
 }
@@ -135,16 +179,29 @@ impl SegmentAggregationCollector for GenericSegmentAggregationResultsCollector {
         agg_with_accessor: &AggregationsWithAccessor,
     ) -> crate::Result<IntermediateAggregationResults> {
         let buckets = if let Some(buckets) = self.buckets {
-            let entries = buckets
-                .into_iter()
-                .zip(agg_with_accessor.buckets.values())
-                .map(|((key, bucket), acc)| Ok((key, bucket.into_intermediate_bucket_result(acc)?)))
-                .collect::<crate::Result<Vec<(String, _)>>>()?;
-            Some(VecWithNames::from_entries(entries))
+            let mut intermeditate_buckets = VecWithNames::default();
+            for bucket in buckets {
+                // TODO too many allocations?
+                let res = bucket.into_intermediate_aggregations_result(agg_with_accessor)?;
+                // unwrap is fine since we only have buckets here
+                intermeditate_buckets.extend(res.buckets.unwrap());
+            }
+            Some(intermeditate_buckets)
+        } else {
+            None
+        };
+        let metrics = if let Some(metrics) = self.metrics {
+            let mut intermeditate_metrics = VecWithNames::default();
+            for metric in metrics {
+                // TODO too many allocations?
+                let res = metric.into_intermediate_aggregations_result(agg_with_accessor)?;
+                // unwrap is fine since we only have metrics here
+                intermeditate_metrics.extend(res.metrics.unwrap());
+            }
+            Some(intermeditate_metrics)
         } else {
             None
         };
-        let metrics = self.metrics.map(VecWithNames::from_other);
 
         Ok(IntermediateAggregationResults { metrics, buckets })
     }
@@ -154,229 +211,77 @@ impl SegmentAggregationCollector for GenericSegmentAggregationResultsCollector {
         doc: crate::DocId,
         agg_with_accessor: &AggregationsWithAccessor,
     ) -> crate::Result<()> {
-        self.staged_docs[self.num_staged_docs] = doc;
-        self.num_staged_docs += 1;
-        if self.num_staged_docs == self.staged_docs.len() {
-            self.flush_staged_docs(agg_with_accessor, false)?;
-        }
+        self.collect_block(&[doc], agg_with_accessor)?;
+
         Ok(())
     }
 
-    fn flush_staged_docs(
+    fn collect_block(
         &mut self,
+        docs: &[crate::DocId],
         agg_with_accessor: &AggregationsWithAccessor,
-        force_flush: bool,
     ) -> crate::Result<()> {
-        if self.num_staged_docs == 0 {
-            return Ok(());
+        if let Some(metrics) = self.metrics.as_mut() {
+            for collector in metrics {
+                collector.collect_block(docs, agg_with_accessor)?;
+            }
         }
+
+        if let Some(buckets) = self.buckets.as_mut() {
+            for collector in buckets {
+                collector.collect_block(docs, agg_with_accessor)?;
+            }
+        }
+
+        Ok(())
+    }
+
+    fn flush(&mut self, agg_with_accessor: &AggregationsWithAccessor) -> crate::Result<()> {
         if let Some(metrics) = &mut self.metrics {
-            for (collector, agg_with_accessor) in
-                metrics.values_mut().zip(agg_with_accessor.metrics.values())
-            {
-                collector
-                    .collect_block(&self.staged_docs[..self.num_staged_docs], agg_with_accessor);
+            for collector in metrics {
+                collector.flush(agg_with_accessor)?;
             }
         }
-
         if let Some(buckets) = &mut self.buckets {
-            for (collector, agg_with_accessor) in
-                buckets.values_mut().zip(agg_with_accessor.buckets.values())
-            {
-                collector.collect_block(
-                    &self.staged_docs[..self.num_staged_docs],
-                    agg_with_accessor,
-                    force_flush,
-                )?;
+            for collector in buckets {
+                collector.flush(agg_with_accessor)?;
             }
         }
-
-        self.num_staged_docs = 0;
         Ok(())
     }
 }
 
 impl GenericSegmentAggregationResultsCollector {
-    pub fn into_intermediate_aggregations_result(
-        self,
-        agg_with_accessor: &AggregationsWithAccessor,
-    ) -> crate::Result<IntermediateAggregationResults> {
-        let buckets = if let Some(buckets) = self.buckets {
-            let entries = buckets
-                .into_iter()
-                .zip(agg_with_accessor.buckets.values())
-                .map(|((key, bucket), acc)| Ok((key, bucket.into_intermediate_bucket_result(acc)?)))
-                .collect::<crate::Result<Vec<(String, _)>>>()?;
-            Some(VecWithNames::from_entries(entries))
-        } else {
-            None
-        };
-        let metrics = self.metrics.map(VecWithNames::from_other);
-
-        Ok(IntermediateAggregationResults { metrics, buckets })
-    }
-
     pub(crate) fn from_req_and_validate(req: &AggregationsWithAccessor) -> crate::Result<Self> {
         let buckets = req
             .buckets
             .iter()
-            .map(|(key, req)| {
-                Ok((
-                    key.to_string(),
-                    SegmentBucketResultCollector::from_req_and_validate(req)?,
-                ))
+            .enumerate()
+            .map(|(accessor_idx, (_key, req))| {
+                build_bucket_segment_agg_collector(req, accessor_idx)
             })
-            .collect::<crate::Result<Vec<(String, _)>>>()?;
+            .collect::<crate::Result<Vec<Box<dyn SegmentAggregationCollector>>>>()?;
         let metrics = req
             .metrics
             .iter()
-            .map(|(key, req)| {
-                Ok((
-                    key.to_string(),
-                    SegmentMetricResultCollector::from_req_and_validate(req)?,
-                ))
+            .enumerate()
+            .map(|(accessor_idx, (_key, req))| {
+                build_metric_segment_agg_collector(req, accessor_idx)
             })
-            .collect::<crate::Result<Vec<(String, _)>>>()?;
+            .collect::<crate::Result<Vec<Box<dyn SegmentAggregationCollector>>>>()?;
+
         let metrics = if metrics.is_empty() {
             None
         } else {
-            Some(VecWithNames::from_entries(metrics))
+            Some(metrics)
         };
+
         let buckets = if buckets.is_empty() {
             None
         } else {
-            Some(VecWithNames::from_entries(buckets))
+            Some(buckets)
         };
-        Ok(GenericSegmentAggregationResultsCollector {
-            metrics,
-            buckets,
-            staged_docs: [0; DOC_BLOCK_SIZE],
-            num_staged_docs: 0,
-        })
-    }
-}
-
-#[derive(Clone, Debug, PartialEq)]
-pub(crate) enum SegmentMetricResultCollector {
-    Stats(SegmentStatsCollector),
-}
-
-impl SegmentMetricResultCollector {
-    pub fn from_req_and_validate(req: &MetricAggregationWithAccessor) -> crate::Result<Self> {
-        match &req.metric {
-            MetricAggregation::Average(AverageAggregation { .. }) => {
-                Ok(SegmentMetricResultCollector::Stats(
-                    SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Average),
-                ))
-            }
-            MetricAggregation::Count(CountAggregation { .. }) => {
-                Ok(SegmentMetricResultCollector::Stats(
-                    SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Count),
-                ))
-            }
-            MetricAggregation::Max(MaxAggregation { .. }) => {
-                Ok(SegmentMetricResultCollector::Stats(
-                    SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Max),
-                ))
-            }
-            MetricAggregation::Min(MinAggregation { .. }) => {
-                Ok(SegmentMetricResultCollector::Stats(
-                    SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Min),
-                ))
-            }
-            MetricAggregation::Stats(StatsAggregation { .. }) => {
-                Ok(SegmentMetricResultCollector::Stats(
-                    SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Stats),
-                ))
-            }
-            MetricAggregation::Sum(SumAggregation { .. }) => {
-                Ok(SegmentMetricResultCollector::Stats(
-                    SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Sum),
-                ))
-            }
-        }
-    }
-    pub(crate) fn collect_block(&mut self, doc: &[DocId], metric: &MetricAggregationWithAccessor) {
-        match self {
-            SegmentMetricResultCollector::Stats(stats_collector) => {
-                stats_collector.collect_block(doc, &metric.accessor);
-            }
-        }
-    }
-}
-
-/// SegmentBucketAggregationResultCollectors will have specialized buckets for collection inside
-/// segments.
-/// The typical structure of Map<Key, Bucket> is not suitable during collection for performance
-/// reasons.
-#[derive(Clone, Debug)]
-pub(crate) enum SegmentBucketResultCollector {
-    Range(SegmentRangeCollector),
-    Histogram(Box<SegmentHistogramCollector>),
-    Terms(Box<SegmentTermCollector>),
-}
-
-impl SegmentBucketResultCollector {
-    pub fn into_intermediate_bucket_result(
-        self,
-        agg_with_accessor: &BucketAggregationWithAccessor,
-    ) -> crate::Result<IntermediateBucketResult> {
-        match self {
-            SegmentBucketResultCollector::Terms(terms) => {
-                terms.into_intermediate_bucket_result(agg_with_accessor)
-            }
-            SegmentBucketResultCollector::Range(range) => {
-                range.into_intermediate_bucket_result(agg_with_accessor)
-            }
-            SegmentBucketResultCollector::Histogram(histogram) => {
-                histogram.into_intermediate_bucket_result(agg_with_accessor)
-            }
-        }
-    }
-
-    pub fn from_req_and_validate(req: &BucketAggregationWithAccessor) -> crate::Result<Self> {
-        match &req.bucket_agg {
-            BucketAggregationType::Terms(terms_req) => Ok(Self::Terms(Box::new(
-                SegmentTermCollector::from_req_and_validate(terms_req, &req.sub_aggregation)?,
-            ))),
-            BucketAggregationType::Range(range_req) => {
-                Ok(Self::Range(SegmentRangeCollector::from_req_and_validate(
-                    range_req,
-                    &req.sub_aggregation,
-                    &req.bucket_count,
-                    req.field_type,
-                )?))
-            }
-            BucketAggregationType::Histogram(histogram) => Ok(Self::Histogram(Box::new(
-                SegmentHistogramCollector::from_req_and_validate(
-                    histogram,
-                    &req.sub_aggregation,
-                    req.field_type,
-                    &req.accessor,
-                )?,
-            ))),
-        }
-    }
-
-    #[inline]
-    pub(crate) fn collect_block(
-        &mut self,
-        doc: &[DocId],
-        bucket_with_accessor: &BucketAggregationWithAccessor,
-        force_flush: bool,
-    ) -> crate::Result<()> {
-        match self {
-            SegmentBucketResultCollector::Range(range) => {
-                range.collect_block(doc, bucket_with_accessor, force_flush)?;
-            }
-            SegmentBucketResultCollector::Histogram(histogram) => {
-                histogram.collect_block(doc, bucket_with_accessor, force_flush)?;
-            }
-            SegmentBucketResultCollector::Terms(terms) => {
-                terms.collect_block(doc, bucket_with_accessor, force_flush)?;
-            }
-        }
-        Ok(())
+        Ok(GenericSegmentAggregationResultsCollector { metrics, buckets })
     }
 }
 

src/collector/facet_collector.rs

@@ -515,8 +515,7 @@ mod tests {
         expected_compressed_collapsed_mapping: &[usize],
         expected_unique_facet_ords: &[(u64, usize)],
     ) {
-        let (compressed_collapsed_mapping, unique_facet_ords) =
-            compress_mapping(&collapsed_mapping);
+        let (compressed_collapsed_mapping, unique_facet_ords) = compress_mapping(collapsed_mapping);
         assert_eq!(
             compressed_collapsed_mapping,
             expected_compressed_collapsed_mapping