Replugging facet collector

Cargo.toml

@@ -59,6 +59,7 @@ sstable = { version="0.1", path="./sstable", package ="tantivy-sstable", optiona
 stacker = { version="0.1", path="./stacker", package ="tantivy-stacker" }
 tantivy-query-grammar = { version= "0.19.0", path="./query-grammar" }
 tantivy-bitpacker = 		{ version= "0.3", path="./bitpacker" }
+columnar = 		{ version= "0.1", path="./columnar", package="tantivy-columnar" }
 common = 								{ version= "0.5", path = "./common/", package = "tantivy-common" }
 fastfield_codecs = 			{ version= "0.3", path="./fastfield_codecs", default-features = false }
 tokenizer-api = { version="0.1", path="./tokenizer-api", package="tantivy-tokenizer-api" }
@@ -107,7 +108,7 @@ unstable = [] # useful for benches.
 quickwit = ["sstable"]
 
 [workspace]
-members = ["query-grammar", "bitpacker", "common", "fastfield_codecs", "ownedbytes", "stacker", "sstable", "tokenizer-api"]
+members = ["query-grammar", "bitpacker", "common", "fastfield_codecs", "ownedbytes", "stacker", "sstable", "tokenizer-api", "columnar"]
 
 # Following the "fail" crate best practises, we isolate
 # tests that define specific behavior in fail check points

TODO.txt

@@ -0,0 +1,18 @@
+Make schema_builder API fluent.
+fix doc serialization and prevent compression problems
+
+u64 , etc. shoudl return Resutl<Option> now that we support optional missing a column is really not an error
+remove fastfield codecs
+ditch the first_or_default trick. if it is still useful, improve its implementation.
+rename FastFieldReaders::open to load
+
+
+remove fast field reader
+
+find a way to unify the two DateTime.
+readd type check in the filter wrapper
+
+add unit test on columnar list columns.
+
+make sure sort works
+

columnar/Cargo.toml

@@ -24,9 +24,5 @@ proptest = "1"
 more-asserts = "0.3.0"
 rand = "0.8.3"
 
-# temporary
-[workspace]
-members = []
-
 [features]
 unstable = []

columnar/Makefile

@@ -0,0 +1,6 @@
+test:
+	echo "Run test only... No examples."
+	cargo test --tests --lib
+
+fmt:
+	cargo +nightly fmt --all

columnar/benches/bench.rs

@@ -0,0 +1,311 @@
+#![feature(test)]
+
+extern crate test;
+
+#[cfg(test)]
+mod tests {
+    use std::ops::RangeInclusive;
+    use std::sync::Arc;
+
+    use common::OwnedBytes;
+    use rand::prelude::*;
+    use tantivy_columnar::*;
+    use test::Bencher;
+
+    use super::*;
+
+    // Warning: this generates the same permutation at each call
+    fn generate_permutation() -> Vec<u64> {
+        let mut permutation: Vec<u64> = (0u64..100_000u64).collect();
+        permutation.shuffle(&mut StdRng::from_seed([1u8; 32]));
+        permutation
+    }
+
+    fn generate_random() -> Vec<u64> {
+        let mut permutation: Vec<u64> = (0u64..100_000u64)
+            .map(|el| el + random::<u16>() as u64)
+            .collect();
+        permutation.shuffle(&mut StdRng::from_seed([1u8; 32]));
+        permutation
+    }
+
+    // Warning: this generates the same permutation at each call
+    fn generate_permutation_gcd() -> Vec<u64> {
+        let mut permutation: Vec<u64> = (1u64..100_000u64).map(|el| el * 1000).collect();
+        permutation.shuffle(&mut StdRng::from_seed([1u8; 32]));
+        permutation
+    }
+
+    pub fn serialize_and_load<T: MonotonicallyMappableToU64 + Ord + Default>(
+        column: &[T],
+    ) -> Arc<dyn Column<T>> {
+        let mut buffer = Vec::new();
+        serialize(VecColumn::from(&column), &mut buffer, &ALL_CODEC_TYPES).unwrap();
+        open(OwnedBytes::new(buffer)).unwrap()
+    }
+
+    #[bench]
+    fn bench_intfastfield_jumpy_veclookup(b: &mut Bencher) {
+        let permutation = generate_permutation();
+        let n = permutation.len();
+        b.iter(|| {
+            let mut a = 0u64;
+            for _ in 0..n {
+                a = permutation[a as usize];
+            }
+            a
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_jumpy_fflookup(b: &mut Bencher) {
+        let permutation = generate_permutation();
+        let n = permutation.len();
+        let column: Arc<dyn Column<u64>> = serialize_and_load(&permutation);
+        b.iter(|| {
+            let mut a = 0u64;
+            for _ in 0..n {
+                a = column.get_val(a as u32);
+            }
+            a
+        });
+    }
+
+    const FIFTY_PERCENT_RANGE: RangeInclusive<u64> = 1..=50;
+    const SINGLE_ITEM: u64 = 90;
+    const SINGLE_ITEM_RANGE: RangeInclusive<u64> = 90..=90;
+    const ONE_PERCENT_ITEM_RANGE: RangeInclusive<u64> = 49..=49;
+    fn get_data_50percent_item() -> Vec<u128> {
+        let mut rng = StdRng::from_seed([1u8; 32]);
+
+        let mut data = vec![];
+        for _ in 0..300_000 {
+            let val = rng.gen_range(1..=100);
+            data.push(val);
+        }
+        data.push(SINGLE_ITEM);
+
+        data.shuffle(&mut rng);
+        let data = data.iter().map(|el| *el as u128).collect::<Vec<_>>();
+        data
+    }
+    fn get_u128_column_random() -> Arc<dyn Column<u128>> {
+        let permutation = generate_random();
+        let permutation = permutation.iter().map(|el| *el as u128).collect::<Vec<_>>();
+        get_u128_column_from_data(&permutation)
+    }
+
+    fn get_u128_column_from_data(data: &[u128]) -> Arc<dyn Column<u128>> {
+        let mut out = vec![];
+        let iter_gen = || data.iter().cloned();
+        serialize_u128(iter_gen, data.len() as u32, &mut out).unwrap();
+        let out = OwnedBytes::new(out);
+        open_u128::<u128>(out).unwrap()
+    }
+
+    // U64 RANGE START
+    #[bench]
+    fn bench_intfastfield_getrange_u64_50percent_hit(b: &mut Bencher) {
+        let data = get_data_50percent_item();
+        let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
+        let column: Arc<dyn Column<u64>> = serialize_and_load(&data);
+
+        b.iter(|| {
+            let mut positions = Vec::new();
+            column.get_docids_for_value_range(
+                FIFTY_PERCENT_RANGE,
+                0..data.len() as u32,
+                &mut positions,
+            );
+            positions
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_getrange_u64_1percent_hit(b: &mut Bencher) {
+        let data = get_data_50percent_item();
+        let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
+        let column: Arc<dyn Column<u64>> = serialize_and_load(&data);
+
+        b.iter(|| {
+            let mut positions = Vec::new();
+            column.get_docids_for_value_range(
+                ONE_PERCENT_ITEM_RANGE,
+                0..data.len() as u32,
+                &mut positions,
+            );
+            positions
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_getrange_u64_single_hit(b: &mut Bencher) {
+        let data = get_data_50percent_item();
+        let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
+        let column: Arc<dyn Column<u64>> = serialize_and_load(&data);
+
+        b.iter(|| {
+            let mut positions = Vec::new();
+            column.get_docids_for_value_range(
+                SINGLE_ITEM_RANGE,
+                0..data.len() as u32,
+                &mut positions,
+            );
+            positions
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_getrange_u64_hit_all(b: &mut Bencher) {
+        let data = get_data_50percent_item();
+        let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
+        let column: Arc<dyn Column<u64>> = serialize_and_load(&data);
+
+        b.iter(|| {
+            let mut positions = Vec::new();
+            column.get_docids_for_value_range(0..=u64::MAX, 0..data.len() as u32, &mut positions);
+            positions
+        });
+    }
+    // U64 RANGE END
+
+    // U128 RANGE START
+    #[bench]
+    fn bench_intfastfield_getrange_u128_50percent_hit(b: &mut Bencher) {
+        let data = get_data_50percent_item();
+        let column = get_u128_column_from_data(&data);
+
+        b.iter(|| {
+            let mut positions = Vec::new();
+            column.get_docids_for_value_range(
+                *FIFTY_PERCENT_RANGE.start() as u128..=*FIFTY_PERCENT_RANGE.end() as u128,
+                0..data.len() as u32,
+                &mut positions,
+            );
+            positions
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_getrange_u128_single_hit(b: &mut Bencher) {
+        let data = get_data_50percent_item();
+        let column = get_u128_column_from_data(&data);
+
+        b.iter(|| {
+            let mut positions = Vec::new();
+            column.get_docids_for_value_range(
+                *SINGLE_ITEM_RANGE.start() as u128..=*SINGLE_ITEM_RANGE.end() as u128,
+                0..data.len() as u32,
+                &mut positions,
+            );
+            positions
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_getrange_u128_hit_all(b: &mut Bencher) {
+        let data = get_data_50percent_item();
+        let column = get_u128_column_from_data(&data);
+
+        b.iter(|| {
+            let mut positions = Vec::new();
+            column.get_docids_for_value_range(0..=u128::MAX, 0..data.len() as u32, &mut positions);
+            positions
+        });
+    }
+    // U128 RANGE END
+
+    #[bench]
+    fn bench_intfastfield_scan_all_fflookup_u128(b: &mut Bencher) {
+        let column = get_u128_column_random();
+
+        b.iter(|| {
+            let mut a = 0u128;
+            for i in 0u64..column.num_vals() as u64 {
+                a += column.get_val(i as u32);
+            }
+            a
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_jumpy_stride5_u128(b: &mut Bencher) {
+        let column = get_u128_column_random();
+
+        b.iter(|| {
+            let n = column.num_vals();
+            let mut a = 0u128;
+            for i in (0..n / 5).map(|val| val * 5) {
+                a += column.get_val(i);
+            }
+            a
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_stride7_vec(b: &mut Bencher) {
+        let permutation = generate_permutation();
+        let n = permutation.len();
+        b.iter(|| {
+            let mut a = 0u64;
+            for i in (0..n / 7).map(|val| val * 7) {
+                a += permutation[i as usize];
+            }
+            a
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_stride7_fflookup(b: &mut Bencher) {
+        let permutation = generate_permutation();
+        let n = permutation.len();
+        let column: Arc<dyn Column<u64>> = serialize_and_load(&permutation);
+        b.iter(|| {
+            let mut a = 0;
+            for i in (0..n / 7).map(|val| val * 7) {
+                a += column.get_val(i as u32);
+            }
+            a
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_scan_all_fflookup(b: &mut Bencher) {
+        let permutation = generate_permutation();
+        let n = permutation.len();
+        let column: Arc<dyn Column<u64>> = serialize_and_load(&permutation);
+        b.iter(|| {
+            let mut a = 0u64;
+            for i in 0u32..n as u32 {
+                a += column.get_val(i);
+            }
+            a
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_scan_all_fflookup_gcd(b: &mut Bencher) {
+        let permutation = generate_permutation_gcd();
+        let n = permutation.len();
+        let column: Arc<dyn Column<u64>> = serialize_and_load(&permutation);
+        b.iter(|| {
+            let mut a = 0u64;
+            for i in 0..n {
+                a += column.get_val(i as u32);
+            }
+            a
+        });
+    }
+
+    #[bench]
+    fn bench_intfastfield_scan_all_vec(b: &mut Bencher) {
+        let permutation = generate_permutation();
+        b.iter(|| {
+            let mut a = 0u64;
+            for i in 0..permutation.len() {
+                a += permutation[i as usize] as u64;
+            }
+            a
+        });
+    }
+}

columnar/src/TODO.md

@@ -9,6 +9,9 @@
     - indexing
     - aggregations
     - merge
+* replug facets
+* replug range queries
++ mutlivaued range queries restrat frm the beginning all of the time.
 
 # Perf and Size
 * re-add ZSTD compression for dictionaries
@@ -26,6 +29,7 @@ Add alignment?
 Consider another codec to bridge the gap between few and 5k elements
 
 # Cleanup and rationalization
+remove the 6 bit limitation of columntype. use 4 + 4 bits instead.
 in benchmark, unify percent vs ratio, f32 vs f64.
 investigate if should have better errors? io::Error is overused at the moment.
 rename rank/select in unit tests
@@ -36,6 +40,12 @@ use the rank & select naming in unit tests branch.
 multi-linear -> blockwise
 linear codec -> simply a multiplication for the index column
 rename columnar to something more explicit, like column_dictionary or columnar_table
+remove old column from the fast field API.
+remove the Column traits alias.
+rename fastfield -> column
+document changes
+rationalization FastFieldValue, HasColumnType
+
 
 # Other
 fix enhance column-cli
@@ -43,4 +53,3 @@ fix enhance column-cli
 # Santa claus
 
 autodetect datetime ipaddr, plug customizable tokenizer.
-

columnar/src/column/dictionary_encoded.rs

@@ -35,10 +35,22 @@ impl BytesColumn {
         self.term_ord_column.num_rows()
     }
 
+    pub fn term_ords(&self, row_id: RowId) -> impl Iterator<Item = u64> + '_ {
+        self.term_ord_column.values(row_id)
+    }
+
     /// Returns the column of ordinals
     pub fn ords(&self) -> &Column<u64> {
         &self.term_ord_column
     }
+
+    pub fn num_terms(&self) -> usize {
+        self.dictionary.num_terms()
+    }
+
+    pub fn dictionary(&self) -> &Dictionary<VoidSSTable> {
+        self.dictionary.as_ref()
+    }
 }
 
 #[derive(Clone)]

columnar/src/column/mod.rs

@@ -1,6 +1,7 @@
 mod dictionary_encoded;
 mod serialize;
 
+use std::fmt::Debug;
 use std::ops::Deref;
 use std::sync::Arc;
 
@@ -17,11 +18,11 @@ use crate::{Cardinality, RowId};
 
 #[derive(Clone)]
 pub struct Column<T> {
-    pub idx: ColumnIndex<'static>,
+    pub idx: ColumnIndex,
     pub values: Arc<dyn ColumnValues<T>>,
 }
 
-impl<T: PartialOrd> Column<T> {
+impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
     pub fn num_rows(&self) -> RowId {
         match &self.idx {
             ColumnIndex::Full => self.values.num_vals() as u32,
@@ -29,7 +30,7 @@ impl<T: PartialOrd> Column<T> {
             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_vals() - 1
+                col_index.num_rows()
             }
         }
     }
@@ -37,12 +38,11 @@ impl<T: PartialOrd> Column<T> {
     pub fn min_value(&self) -> T {
         self.values.min_value()
     }
+
     pub fn max_value(&self) -> T {
         self.values.max_value()
     }
-}
 
-impl<T: PartialOrd + Copy + Send + Sync + 'static> Column<T> {
     pub fn first(&self, row_id: RowId) -> Option<T> {
         self.values(row_id).next()
     }
@@ -61,7 +61,7 @@ impl<T: PartialOrd + Copy + Send + Sync + 'static> Column<T> {
 }
 
 impl<T> Deref for Column<T> {
-    type Target = ColumnIndex<'static>;
+    type Target = ColumnIndex;
 
     fn deref(&self) -> &Self::Target {
         &self.idx
@@ -86,7 +86,9 @@ struct FirstValueWithDefault<T: Copy> {
     default_value: T,
 }
 
-impl<T: PartialOrd + Send + Sync + Copy + 'static> ColumnValues<T> for FirstValueWithDefault<T> {
+impl<T: PartialOrd + Debug + Send + Sync + Copy + 'static> ColumnValues<T>
+    for FirstValueWithDefault<T>
+{
     fn get_val(&self, idx: u32) -> T {
         self.column.first(idx).unwrap_or(self.default_value)
     }

columnar/src/column/serialize.rs

@@ -1,3 +1,4 @@
+use std::fmt::Debug;
 use std::io;
 use std::io::Write;
 use std::sync::Arc;
@@ -33,7 +34,7 @@ pub fn serialize_column_mappable_to_u128<
     Ok(())
 }
 
-pub fn serialize_column_mappable_to_u64<T: MonotonicallyMappableToU64>(
+pub fn serialize_column_mappable_to_u64<T: MonotonicallyMappableToU64 + Debug>(
     column_index: SerializableColumnIndex<'_>,
     column_values: &impl ColumnValues<T>,
     output: &mut impl Write,

columnar/src/column_index/mod.rs

@@ -3,28 +3,23 @@ mod optional_index;
 mod serialize;
 
 use std::ops::Range;
-use std::sync::Arc;
 
 pub use optional_index::{OptionalIndex, SerializableOptionalIndex, Set};
 pub use serialize::{open_column_index, serialize_column_index, SerializableColumnIndex};
 
-use crate::column_values::ColumnValues;
+use crate::column_index::multivalued_index::MultiValueIndex;
 use crate::{Cardinality, RowId};
 
 #[derive(Clone)]
-pub enum ColumnIndex<'a> {
+pub enum ColumnIndex {
     Full,
     Optional(OptionalIndex),
-    // TODO Remove the static by fixing the codec if possible.
-    /// The column values enclosed contains for all row_id,
-    /// the value start_index.
-    ///
     /// In addition, at index num_rows, an extra value is added
     /// containing the overal number of values.
-    Multivalued(Arc<dyn ColumnValues<RowId> + 'a>),
+    Multivalued(MultiValueIndex),
 }
 
-impl<'a> ColumnIndex<'a> {
+impl ColumnIndex {
     pub fn get_cardinality(&self) -> Cardinality {
         match self {
             ColumnIndex::Full => Cardinality::Full,
@@ -43,11 +38,22 @@ impl<'a> ColumnIndex<'a> {
                     0..0
                 }
             }
+            ColumnIndex::Multivalued(multivalued_index) => multivalued_index.range(row_id),
+        }
+    }
+
+    pub fn select_batch_in_place(&self, rank_ids: &mut Vec<RowId>) {
+        match self {
+            ColumnIndex::Full => {
+                // No need to do anything:
+                // value_idx and row_idx are the same.
+            }
+            ColumnIndex::Optional(optional_index) => {
+                optional_index.select_batch(&mut rank_ids[..]);
+            }
             ColumnIndex::Multivalued(multivalued_index) => {
-                let multivalued_index_ref = &**multivalued_index;
-                let start: u32 = multivalued_index_ref.get_val(row_id);
-                let end: u32 = multivalued_index_ref.get_val(row_id + 1);
-                start..end
+                // TODO important: avoid using 0u32, and restart from the beginning all of the time.
+                multivalued_index.select_batch_in_place(0u32, rank_ids)
             }
         }
     }

columnar/src/column_index/multivalued_index.rs

@@ -1,5 +1,6 @@
 use std::io;
 use std::io::Write;
+use std::ops::Range;
 use std::sync::Arc;
 
 use common::OwnedBytes;
@@ -7,9 +8,6 @@ use common::OwnedBytes;
 use crate::column_values::{ColumnValues, FastFieldCodecType};
 use crate::RowId;
 
-#[derive(Clone)]
-pub struct MultivaluedIndex(Arc<dyn ColumnValues<RowId>>);
-
 pub fn serialize_multivalued_index(
     multivalued_index: &dyn ColumnValues<RowId>,
     output: &mut impl Write,
@@ -22,8 +20,113 @@ pub fn serialize_multivalued_index(
     Ok(())
 }
 
-pub fn open_multivalued_index(bytes: OwnedBytes) -> io::Result<Arc<dyn ColumnValues<RowId>>> {
+pub fn open_multivalued_index(bytes: OwnedBytes) -> io::Result<MultiValueIndex> {
     let start_index_column: Arc<dyn ColumnValues<RowId>> =
         crate::column_values::open_u64_mapped(bytes)?;
-    Ok(start_index_column)
+    Ok(MultiValueIndex { start_index_column })
+}
+
+#[derive(Clone)]
+/// Index to resolve value range for given doc_id.
+/// Starts at 0.
+pub struct MultiValueIndex {
+    start_index_column: Arc<dyn crate::ColumnValues<RowId>>,
+}
+
+impl From<Arc<dyn ColumnValues<RowId>>> for MultiValueIndex {
+    fn from(start_index_column: Arc<dyn ColumnValues<RowId>>) -> Self {
+        MultiValueIndex { start_index_column }
+    }
+}
+
+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);
+        start..end
+    }
+
+    /// Returns the number of documents in the index.
+    #[inline]
+    pub fn num_rows(&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.
+    ///
+    /// Since there is no index for value pos -> docid, but docid -> value pos range, we scan the
+    /// index.
+    ///
+    /// Correctness: positions needs to be sorted. idx_reader needs to contain monotonically
+    /// increasing positions.
+    ///
+    /// 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>) {
+        if ranks.is_empty() {
+            return;
+        }
+        let mut cur_doc = row_start;
+        let mut last_doc = None;
+
+        assert!(self.start_index_column.get_val(row_start) as u32 <= 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;
+                if end > pos {
+                    ranks[write_doc_pos] = cur_doc;
+                    write_doc_pos += if last_doc == Some(cur_doc) { 0 } else { 1 };
+                    last_doc = Some(cur_doc);
+                    break;
+                }
+                cur_doc += 1;
+            }
+        }
+        ranks.truncate(write_doc_pos);
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::ops::Range;
+    use std::sync::Arc;
+
+    use super::MultiValueIndex;
+    use crate::column_values::IterColumn;
+    use crate::{ColumnValues, RowId};
+
+    fn index_to_pos_helper(
+        index: &MultiValueIndex,
+        doc_id_range: Range<u32>,
+        positions: &[u32],
+    ) -> Vec<u32> {
+        let mut positions = positions.to_vec();
+        index.select_batch_in_place(doc_id_range.start, &mut positions);
+        positions
+    }
+
+    #[test]
+    fn test_positions_to_docid() {
+        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);
+        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]);
+        assert_eq!(index_to_pos_helper(&index, 0..5, &[9]), vec![0]);
+        assert_eq!(index_to_pos_helper(&index, 1..5, &[10]), vec![1]);
+        assert_eq!(index_to_pos_helper(&index, 1..5, &[11]), vec![1]);
+        assert_eq!(index_to_pos_helper(&index, 2..5, &[12]), vec![2]);
+        assert_eq!(index_to_pos_helper(&index, 2..5, &[12, 14]), vec![2]);
+        assert_eq!(index_to_pos_helper(&index, 2..5, &[12, 14, 15]), vec![2, 3]);
+    }
 }

columnar/src/column_index/optional_index/mod.rs

@@ -6,7 +6,7 @@ mod set;
 mod set_block;
 
 use common::{BinarySerializable, OwnedBytes, VInt};
-pub use set::{Set, SetCodec, SelectCursor};
+pub use set::{SelectCursor, Set, SetCodec};
 use set_block::{
     DenseBlock, DenseBlockCodec, SparseBlock, SparseBlockCodec, DENSE_BLOCK_NUM_BYTES,
 };
@@ -127,7 +127,6 @@ impl<'a> BlockSelectCursor<'a> {
             BlockSelectCursor::Sparse(sparse_select_cursor) => sparse_select_cursor.select(rank),
         }
     }
-
 }
 pub struct OptionalIndexSelectCursor<'a> {
     current_block_cursor: BlockSelectCursor<'a>,
@@ -146,7 +145,12 @@ impl<'a> OptionalIndexSelectCursor<'a> {
             return;
         }
         self.current_block_id = self.optional_index.find_block(rank, self.current_block_id);
-        self.current_block_end_rank = self.optional_index.block_metas.get(self.current_block_id as usize + 1).map(|block_meta| block_meta.non_null_rows_before_block).unwrap_or(u32::MAX);
+        self.current_block_end_rank = self
+            .optional_index
+            .block_metas
+            .get(self.current_block_id as usize + 1)
+            .map(|block_meta| block_meta.non_null_rows_before_block)
+            .unwrap_or(u32::MAX);
         self.block_doc_idx_start = (self.current_block_id as u32) * ELEMENTS_PER_BLOCK;
         let block_meta = self.optional_index.block_metas[self.current_block_id as usize];
         self.num_null_rows_before_block = block_meta.non_null_rows_before_block;
@@ -213,7 +217,9 @@ impl Set<RowId> for OptionalIndex {
 
     fn select_cursor<'b>(&'b self) -> OptionalIndexSelectCursor<'b> {
         OptionalIndexSelectCursor {
-            current_block_cursor: BlockSelectCursor::Sparse(SparseBlockCodec::open(b"").select_cursor()),
+            current_block_cursor: BlockSelectCursor::Sparse(
+                SparseBlockCodec::open(b"").select_cursor(),
+            ),
             current_block_id: 0u16,
             current_block_end_rank: 0u32, //< this is sufficient to force the first load
             optional_index: self,
@@ -224,7 +230,6 @@ impl Set<RowId> for OptionalIndex {
 }
 
 impl OptionalIndex {
-
     pub fn select_batch(&self, ranks: &mut [RowId]) {
         let mut select_cursor = self.select_cursor();
         for rank in ranks.iter_mut() {

columnar/src/column_index/optional_index/set.rs

@@ -13,7 +13,6 @@ pub trait SetCodec {
     fn open<'a>(data: &'a [u8]) -> Self::Reader<'a>;
 }
 
-
 /// Stateful object that makes it possible to compute several select in a row,
 /// provided the rank passed as argument are increasing.
 pub trait SelectCursor<T> {
@@ -23,8 +22,8 @@ pub trait SelectCursor<T> {
 }
 
 pub trait Set<T> {
-    type SelectCursor<'b>: SelectCursor<T> where Self: 'b;
-
+    type SelectCursor<'b>: SelectCursor<T>
+    where Self: 'b;
 
     /// Returns true if the elements is contained in the Set
     fn contains(&self, el: T) -> bool;
@@ -41,5 +40,5 @@ 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<'b>(&'b self) -> Self::SelectCursor<'b>;
 }

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

@@ -3,7 +3,7 @@ use std::io::{self, Write};
 
 use common::BinarySerializable;
 
-use crate::column_index::optional_index::{Set, SetCodec, SelectCursor, ELEMENTS_PER_BLOCK};
+use crate::column_index::optional_index::{SelectCursor, Set, SetCodec, ELEMENTS_PER_BLOCK};
 
 #[inline(always)]
 fn get_bit_at(input: u64, n: u16) -> bool {
@@ -113,7 +113,10 @@ pub struct DenseBlockSelectCursor<'a> {
 impl<'a> SelectCursor<u16> for DenseBlockSelectCursor<'a> {
     #[inline]
     fn select(&mut self, rank: u16) -> u16 {
-        self.block_id = self.dense_block.find_miniblock_containing_rank(rank, self.block_id).unwrap();
+        self.block_id = self
+            .dense_block
+            .find_miniblock_containing_rank(rank, self.block_id)
+            .unwrap();
         let index_block = self.dense_block.mini_block(self.block_id);
         let in_block_rank = rank - index_block.rank;
         self.block_id * ELEMENTS_PER_MINI_BLOCK + select_u64(index_block.bitvec, in_block_rank)
@@ -154,7 +157,7 @@ impl<'a> Set<u16> for DenseBlock<'a> {
     }
 
     #[inline(always)]
-    fn select_cursor<'b>(&'b self,) -> Self::SelectCursor<'b> {
+    fn select_cursor<'b>(&'b self) -> Self::SelectCursor<'b> {
         DenseBlockSelectCursor {
             block_id: 0,
             dense_block: *self,

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

@@ -1,4 +1,4 @@
-use crate::column_index::optional_index::{Set, SetCodec, SelectCursor};
+use crate::column_index::optional_index::{SelectCursor, Set, SetCodec};
 
 pub struct SparseBlockCodec;
 
@@ -32,7 +32,6 @@ impl<'a> SelectCursor<u16> for SparseBlock<'a> {
 }
 
 impl<'a> Set<u16> for SparseBlock<'a> {
-
     type SelectCursor<'b> = Self where Self: 'b;
 
     #[inline(always)]
@@ -52,10 +51,9 @@ impl<'a> Set<u16> for SparseBlock<'a> {
     }
 
     #[inline(always)]
-    fn select_cursor<'b>(&'b self,) -> Self::SelectCursor<'b> {
+    fn select_cursor<'b>(&'b self) -> Self::SelectCursor<'b> {
         *self
     }
-
 }
 
 #[inline(always)]

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

@@ -2,7 +2,7 @@ use std::collections::HashMap;
 
 use crate::column_index::optional_index::set_block::dense::DENSE_BLOCK_NUM_BYTES;
 use crate::column_index::optional_index::set_block::{DenseBlockCodec, SparseBlockCodec};
-use crate::column_index::optional_index::{Set, SetCodec, SelectCursor};
+use crate::column_index::optional_index::{SelectCursor, Set, SetCodec};
 
 fn test_set_helper<C: SetCodec<Item = u16>>(vals: &[u16]) -> usize {
     let mut buffer = Vec::new();

columnar/src/column_index/serialize.rs

@@ -47,7 +47,7 @@ pub fn serialize_column_index(
     Ok(column_index_num_bytes)
 }
 
-pub fn open_column_index(mut bytes: OwnedBytes) -> io::Result<ColumnIndex<'static>> {
+pub fn open_column_index(mut bytes: OwnedBytes) -> io::Result<ColumnIndex> {
     if bytes.is_empty() {
         return Err(io::Error::new(
             io::ErrorKind::UnexpectedEof,
@@ -64,8 +64,8 @@ pub fn open_column_index(mut bytes: OwnedBytes) -> io::Result<ColumnIndex<'stati
             Ok(ColumnIndex::Optional(optional_index))
         }
         Cardinality::Multivalued => {
-            let multivalued_index = super::multivalued_index::open_multivalued_index(bytes)?;
-            Ok(ColumnIndex::Multivalued(multivalued_index))
+            let multivalue_index = super::multivalued_index::open_multivalued_index(bytes)?;
+            Ok(ColumnIndex::Multivalued(multivalue_index))
         }
     }
 }

columnar/src/column_values/column.rs

@@ -1,3 +1,4 @@
+use std::fmt::Debug;
 use std::marker::PhantomData;
 use std::ops::{Range, RangeInclusive};
 
@@ -8,7 +9,7 @@ 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 {
+pub trait ColumnValues<T: PartialOrd + Debug = u64>: Send + Sync {
     /// Return the value associated with the given idx.
     ///
     /// This accessor should return as fast as possible.
@@ -44,7 +45,6 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
         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) {
@@ -78,7 +78,7 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
     }
 }
 
-impl<T: Copy + PartialOrd> ColumnValues<T> for std::sync::Arc<dyn ColumnValues<T>> {
+impl<T: Copy + PartialOrd + Debug> ColumnValues<T> for std::sync::Arc<dyn ColumnValues<T>> {
     fn get_val(&self, idx: u32) -> T {
         self.as_ref().get_val(idx)
     }
@@ -104,7 +104,7 @@ impl<T: Copy + PartialOrd> ColumnValues<T> for std::sync::Arc<dyn ColumnValues<T
     }
 }
 
-impl<'a, C: ColumnValues<T> + ?Sized, T: Copy + PartialOrd> ColumnValues<T> for &'a C {
+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)
     }
@@ -137,7 +137,7 @@ pub struct VecColumn<'a, T = u64> {
     pub(crate) max_value: T,
 }
 
-impl<'a, T: Copy + PartialOrd + Send + Sync> ColumnValues<T> for VecColumn<'a, 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]
     }
@@ -205,8 +205,8 @@ pub fn monotonic_map_column<C, T, Input, Output>(
 where
     C: ColumnValues<Input>,
     T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
-    Input: PartialOrd + Send + Sync + Clone,
-    Output: PartialOrd + Send + Sync + Clone,
+    Input: PartialOrd + Debug + Send + Sync + Clone,
+    Output: PartialOrd + Debug + Send + Sync + Clone,
 {
     MonotonicMappingColumn {
         from_column,
@@ -219,8 +219,8 @@ impl<C, T, Input, Output> ColumnValues<Output> for MonotonicMappingColumn<C, T,
 where
     C: ColumnValues<Input>,
     T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
-    Input: PartialOrd + Send + Sync + Clone,
-    Output: PartialOrd + Send + Sync + Clone,
+    Input: PartialOrd + Send + Debug + Sync + Clone,
+    Output: PartialOrd + Send + Debug + Sync + Clone,
 {
     #[inline]
     fn get_val(&self, idx: u32) -> Output {
@@ -282,7 +282,7 @@ where T: Iterator + Clone + ExactSizeIterator
 impl<T> ColumnValues<T::Item> for IterColumn<T>
 where
     T: Iterator + Clone + ExactSizeIterator + Send + Sync,
-    T::Item: PartialOrd,
+    T::Item: PartialOrd + Debug,
 {
     fn get_val(&self, idx: u32) -> T::Item {
         self.0.clone().nth(idx as usize).unwrap()

columnar/src/column_values/column_with_cardinality.rs

@@ -1,19 +0,0 @@
-// Copyright (C) 2022 Quickwit, Inc.
-//
-// Quickwit is offered under the AGPL v3.0 and as commercial software.
-// For commercial licensing, contact us at hello@quickwit.io.
-//
-// AGPL:
-// This program is free software: you can redistribute it and/or modify
-// it under the terms of the GNU Affero General Public License as
-// published by the Free Software Foundation, either version 3 of the
-// License, or (at your option) any later version.
-//
-// This program is distributed in the hope that it will be useful,
-// but WITHOUT ANY WARRANTY; without even the implied warranty of
-// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
-// GNU Affero General Public License for more details.
-//
-// You should have received a copy of the GNU Affero General Public License
-// along with this program. If not, see <http://www.gnu.org/licenses/>.
-//

columnar/src/column_values/mod.rs

@@ -10,16 +10,19 @@
 #[cfg(test)]
 mod tests;
 
+use std::fmt::Debug;
 use std::io;
 use std::io::Write;
 use std::sync::Arc;
 
 use common::{BinarySerializable, OwnedBytes};
 use compact_space::CompactSpaceDecompressor;
+pub use monotonic_mapping::{MonotonicallyMappableToU64, StrictlyMonotonicFn};
 use monotonic_mapping::{
     StrictlyMonotonicMappingInverter, StrictlyMonotonicMappingToInternal,
     StrictlyMonotonicMappingToInternalBaseval, StrictlyMonotonicMappingToInternalGCDBaseval,
 };
+pub use monotonic_mapping_u128::MonotonicallyMappableToU128;
 use serialize::{Header, U128Header};
 
 mod bitpacked;
@@ -31,13 +34,10 @@ pub(crate) mod monotonic_mapping;
 pub(crate) mod monotonic_mapping_u128;
 
 mod column;
-mod column_with_cardinality;
 mod gcd;
 pub mod serialize;
 
 pub use self::column::{monotonic_map_column, ColumnValues, IterColumn, VecColumn};
-pub use self::monotonic_mapping::{MonotonicallyMappableToU64, StrictlyMonotonicFn};
-pub use self::monotonic_mapping_u128::MonotonicallyMappableToU128;
 #[cfg(test)]
 pub use self::serialize::tests::serialize_and_load;
 pub use self::serialize::{serialize_column_values, NormalizedHeader};
@@ -124,7 +124,7 @@ impl U128FastFieldCodecType {
 }
 
 /// Returns the correct codec reader wrapped in the `Arc` for the data.
-pub fn open_u128_mapped<T: MonotonicallyMappableToU128>(
+pub fn open_u128_mapped<T: MonotonicallyMappableToU128 + Debug>(
     mut bytes: OwnedBytes,
 ) -> io::Result<Arc<dyn ColumnValues<T>>> {
     let header = U128Header::deserialize(&mut bytes)?;
@@ -137,7 +137,7 @@ pub fn open_u128_mapped<T: MonotonicallyMappableToU128>(
 }
 
 /// Returns the correct codec reader wrapped in the `Arc` for the data.
-pub fn open_u64_mapped<T: MonotonicallyMappableToU64>(
+pub fn open_u64_mapped<T: MonotonicallyMappableToU64 + Debug>(
     mut bytes: OwnedBytes,
 ) -> io::Result<Arc<dyn ColumnValues<T>>> {
     let header = Header::deserialize(&mut bytes)?;
@@ -150,7 +150,7 @@ pub fn open_u64_mapped<T: MonotonicallyMappableToU64>(
     }
 }
 
-fn open_specific_codec<C: FastFieldCodec, Item: MonotonicallyMappableToU64>(
+fn open_specific_codec<C: FastFieldCodec, Item: MonotonicallyMappableToU64 + Debug>(
     bytes: OwnedBytes,
     header: &Header,
 ) -> io::Result<Arc<dyn ColumnValues<Item>>> {

columnar/src/column_values/monotonic_mapping.rs

@@ -1,3 +1,4 @@
+use std::fmt::Debug;
 use std::marker::PhantomData;
 
 use fastdivide::DividerU64;
@@ -7,7 +8,7 @@ use crate::RowId;
 
 /// Monotonic maps a value to u64 value space.
 /// Monotonic mapping enables `PartialOrd` on u64 space without conversion to original space.
-pub trait MonotonicallyMappableToU64: 'static + PartialOrd + Copy + Send + Sync {
+pub trait MonotonicallyMappableToU64: 'static + PartialOrd + Debug + Copy + Send + Sync {
     /// Converts a value to u64.
     ///
     /// Internally all fast field values are encoded as u64.

columnar/src/column_values/monotonic_mapping_u128.rs

@@ -1,8 +1,9 @@
+use std::fmt::Debug;
 use std::net::Ipv6Addr;
 
 /// Montonic maps a value to u128 value space
 /// Monotonic mapping enables `PartialOrd` on u128 space without conversion to original space.
-pub trait MonotonicallyMappableToU128: 'static + PartialOrd + Copy + Send + Sync {
+pub trait MonotonicallyMappableToU128: 'static + PartialOrd + Copy + Debug + Send + Sync {
     /// Converts a value to u128.
     ///
     /// Internally all fast field values are encoded as u64.

columnar/src/column_values/serialize.rs

@@ -1,22 +1,4 @@
-// Copyright (C) 2022 Quickwit, Inc.
-//
-// Quickwit is offered under the AGPL v3.0 and as commercial software.
-// For commercial licensing, contact us at hello@quickwit.io.
-//
-// AGPL:
-// This program is free software: you can redistribute it and/or modify
-// it under the terms of the GNU Affero General Public License as
-// published by the Free Software Foundation, either version 3 of the
-// License, or (at your option) any later version.
-//
-// This program is distributed in the hope that it will be useful,
-// but WITHOUT ANY WARRANTY; without even the implied warranty of
-// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
-// GNU Affero General Public License for more details.
-//
-// You should have received a copy of the GNU Affero General Public License
-// along with this program. If not, see <http://www.gnu.org/licenses/>.
-
+use std::fmt::Debug;
 use std::io;
 use std::num::NonZeroU64;
 
@@ -178,7 +160,7 @@ pub fn serialize_column_values_u128<F: Fn() -> I, I: Iterator<Item = u128>>(
 }
 
 /// Serializes the column with the codec with the best estimate on the data.
-pub fn serialize_column_values<T: MonotonicallyMappableToU64>(
+pub fn serialize_column_values<T: MonotonicallyMappableToU64 + Debug>(
     typed_column: impl ColumnValues<T>,
     codecs: &[FastFieldCodecType],
     output: &mut impl io::Write,

columnar/src/columnar/column_type.rs

@@ -1,3 +1,4 @@
+use std::fmt::Debug;
 use std::net::Ipv6Addr;
 
 use crate::value::NumericalType;
@@ -91,7 +92,7 @@ impl ColumnType {
 }
 
 // TODO remove if possible
-pub trait HasAssociatedColumnType: 'static + Send + Sync + Copy + PartialOrd {
+pub trait HasAssociatedColumnType: 'static + Debug + Send + Sync + Copy + PartialOrd {
     fn column_type() -> ColumnType;
     fn default_value() -> Self;
 }

columnar/src/columnar/reader/mod.rs

@@ -73,7 +73,6 @@ impl ColumnarReader {
     ///
     /// There can be more than one column associated to a given column name, provided they have
     /// different types.
-    // TODO fix ugly API
     pub fn read_columns(&self, column_name: &str) -> io::Result<Vec<DynamicColumnHandle>> {
         // Each column is a associated to a given `column_key`,
         // that starts by `column_name\0column_header`.
@@ -120,3 +119,46 @@ impl ColumnarReader {
         self.column_dictionary.num_terms()
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use crate::{ColumnType, ColumnarReader, ColumnarWriter};
+
+    #[test]
+    fn test_list_columns() {
+        let mut columnar_writer = ColumnarWriter::default();
+        columnar_writer.record_column_type("col1", ColumnType::Str, false);
+        columnar_writer.record_column_type("col2", ColumnType::U64, false);
+        let mut buffer = Vec::new();
+        columnar_writer.serialize(1, &mut buffer).unwrap();
+        let columnar = ColumnarReader::open(buffer).unwrap();
+        let columns = columnar.list_columns().unwrap();
+        assert_eq!(columns.len(), 2);
+        assert_eq!(&columns[0].0, "col1");
+        assert_eq!(columns[0].1.column_type(), ColumnType::Str);
+        assert_eq!(&columns[1].0, "col2");
+        assert_eq!(columns[1].1.column_type(), ColumnType::U64);
+    }
+
+    #[test]
+    fn test_list_columns_strict_typing_prevents_coercion() {
+        let mut columnar_writer = ColumnarWriter::default();
+        columnar_writer.record_column_type("count", ColumnType::U64, false);
+        columnar_writer.record_numerical(1, "count", 1u64);
+        let mut buffer = Vec::new();
+        columnar_writer.serialize(2, &mut buffer).unwrap();
+        let columnar = ColumnarReader::open(buffer).unwrap();
+        let columns = columnar.list_columns().unwrap();
+        assert_eq!(columns.len(), 1);
+        assert_eq!(&columns[0].0, "count");
+        assert_eq!(columns[0].1.column_type(), ColumnType::U64);
+    }
+
+    #[test]
+    #[should_panic(expect = "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_writer.record_numerical(1, "count", 1i64);
+    }
+}

columnar/src/columnar/writer/column_writers.rs

@@ -168,7 +168,12 @@ impl CompatibleNumericalTypes {
                 }
             },
             CompatibleNumericalTypes::StaticType(typ) => {
-                assert_eq!(numerical_value.numerical_type(), *typ);
+                assert_eq!(
+                    numerical_value.numerical_type(),
+                    *typ,
+                    "Input type forbidden. This column has been forced to type {typ:?}, received \
+                     {numerical_value:?}"
+                );
             }
         }
     }
@@ -213,6 +218,14 @@ impl NumericalColumnWriter {
 pub(crate) struct StrOrBytesColumnWriter {
     pub(crate) dictionary_id: u32,
     pub(crate) column_writer: ColumnWriter,
+    // If true, when facing a multivalued cardinality,
+    // values associated to a given document will be sorted.
+    //
+    // This is useful for facets.
+    //
+    // If false, the order of appearance in the document will be
+    // observed.
+    pub(crate) sort_values_within_row: bool,
 }
 
 impl StrOrBytesColumnWriter {
@@ -220,6 +233,7 @@ impl StrOrBytesColumnWriter {
         StrOrBytesColumnWriter {
             dictionary_id,
             column_writer: Default::default(),
+            sort_values_within_row: false,
         }
     }
 

columnar/src/columnar/writer/mod.rs

@@ -29,10 +29,7 @@ use crate::{Cardinality, RowId};
 #[derive(Default)]
 struct SpareBuffers {
     value_index_builders: PreallocatedIndexBuilders,
-    i64_values: Vec<i64>,
     u64_values: Vec<u64>,
-    f64_values: Vec<f64>,
-    bool_values: Vec<bool>,
     ip_addr_values: Vec<Ipv6Addr>,
 }
 
@@ -106,7 +103,23 @@ impl ColumnarWriter {
             + self.datetime_field_hash_map.mem_usage()
     }
 
-    pub fn record_column_type(&mut self, column_name: &str, column_type: ColumnType) {
+    /// Records a column type. This is useful to bypass the coercion process,
+    /// makes sure the empty is present in the resulting columnar, or set
+    /// the `sort_values_within_row`.
+    ///
+    /// `sort_values_within_row` is only allowed for `Bytes` or `Str` columns.
+    pub fn record_column_type(
+        &mut self,
+        column_name: &str,
+        column_type: ColumnType,
+        sort_values_within_row: bool,
+    ) {
+        if sort_values_within_row {
+            assert!(
+                column_type == ColumnType::Bytes || column_type == ColumnType::Str,
+                "sort_values_within_row is only allowed for Bytes and Str columns",
+            );
+        }
         match column_type {
             ColumnType::Str | ColumnType::Bytes => {
                 let (hash_map, dictionaries) = (
@@ -121,13 +134,15 @@ impl ColumnarWriter {
                     hash_map,
                     column_name,
                     |column_opt: Option<StrOrBytesColumnWriter>| {
-                        if let Some(column_writer) = column_opt {
+                        let mut column_writer = if let Some(column_writer) = column_opt {
                             column_writer
                         } else {
                             let dictionary_id = dictionaries.len() as u32;
                             dictionaries.push(DictionaryBuilder::default());
                             StrOrBytesColumnWriter::with_dictionary_id(dictionary_id)
-                        }
+                        };
+                        column_writer.sort_values_within_row = sort_values_within_row;
+                        column_writer
                     },
                 );
             }
@@ -165,18 +180,6 @@ impl ColumnarWriter {
         }
     }
 
-    pub fn force_numerical_type(&mut self, column_name: &str, numerical_type: NumericalType) {
-        mutate_or_create_column(
-            &mut self.numerical_field_hash_map,
-            column_name,
-            |column_opt: Option<NumericalColumnWriter>| {
-                let mut column: NumericalColumnWriter = column_opt.unwrap_or_default();
-                column.force_numerical_type(numerical_type);
-                column
-            },
-        );
-    }
-
     pub fn record_numerical<T: Into<NumericalValue> + Copy>(
         &mut self,
         doc: RowId,
@@ -353,6 +356,7 @@ impl ColumnarWriter {
                     serialize_bytes_or_str_column(
                         cardinality,
                         num_docs,
+                        str_column_writer.sort_values_within_row,
                         dictionary_builder,
                         str_column_writer.operation_iterator(arena, &mut symbol_byte_buffer),
                         buffers,
@@ -399,6 +403,7 @@ impl ColumnarWriter {
 fn serialize_bytes_or_str_column(
     cardinality: Cardinality,
     num_docs: RowId,
+    sort_values_within_row: bool,
     dictionary_builder: &DictionaryBuilder,
     operation_it: impl Iterator<Item = ColumnOperation<UnorderedId>>,
     buffers: &mut SpareBuffers,
@@ -427,6 +432,7 @@ fn serialize_bytes_or_str_column(
         operation_iterator,
         cardinality,
         num_docs,
+        sort_values_within_row,
         value_index_builders,
         u64_values,
         &mut wrt,
@@ -446,8 +452,6 @@ fn serialize_numerical_column(
     let SpareBuffers {
         value_index_builders,
         u64_values,
-        i64_values,
-        f64_values,
         ..
     } = buffers;
     match numerical_type {
@@ -456,8 +460,9 @@ fn serialize_numerical_column(
                 coerce_numerical_symbol::<i64>(op_iterator),
                 cardinality,
                 num_docs,
+                false,
                 value_index_builders,
-                i64_values,
+                u64_values,
                 wrt,
             )?;
         }
@@ -466,6 +471,7 @@ fn serialize_numerical_column(
                 coerce_numerical_symbol::<u64>(op_iterator),
                 cardinality,
                 num_docs,
+                false,
                 value_index_builders,
                 u64_values,
                 wrt,
@@ -476,8 +482,9 @@ fn serialize_numerical_column(
                 coerce_numerical_symbol::<f64>(op_iterator),
                 cardinality,
                 num_docs,
+                false,
                 value_index_builders,
-                f64_values,
+                u64_values,
                 wrt,
             )?;
         }
@@ -494,15 +501,19 @@ fn serialize_bool_column(
 ) -> io::Result<()> {
     let SpareBuffers {
         value_index_builders,
-        bool_values,
+        u64_values,
         ..
     } = buffers;
     send_to_serialize_column_mappable_to_u64(
-        column_operations_it,
+        column_operations_it.map(|bool_column_operation| match bool_column_operation {
+            ColumnOperation::NewDoc(doc) => ColumnOperation::NewDoc(doc),
+            ColumnOperation::Value(bool_val) => ColumnOperation::Value(bool_val.to_u64()),
+        }),
         cardinality,
         num_docs,
+        false,
         value_index_builders,
-        bool_values,
+        u64_values,
         wrt,
     )?;
     Ok(())
@@ -532,7 +543,7 @@ fn serialize_ip_addr_column(
 }
 
 fn send_to_serialize_column_mappable_to_u128<
-    T: Copy + std::fmt::Debug + Send + Sync + MonotonicallyMappableToU128 + PartialOrd,
+    T: Copy + Ord + std::fmt::Debug + Send + Sync + MonotonicallyMappableToU128 + PartialOrd,
 >(
     op_iterator: impl Iterator<Item = ColumnOperation<T>>,
     cardinality: Cardinality,
@@ -577,18 +588,29 @@ where
     Ok(())
 }
 
-fn send_to_serialize_column_mappable_to_u64<
-    T: Copy + Default + std::fmt::Debug + Send + Sync + MonotonicallyMappableToU64 + PartialOrd,
->(
-    op_iterator: impl Iterator<Item = ColumnOperation<T>>,
+fn sort_values_within_row_in_place(
+    multivalued_index: &impl ColumnValues<RowId>,
+    values: &mut Vec<u64>,
+) {
+    let mut start_index: usize = 0;
+    for end_index in multivalued_index.iter() {
+        let end_index = end_index as usize;
+        values[start_index..end_index].sort_unstable();
+        start_index = end_index;
+    }
+}
+
+fn send_to_serialize_column_mappable_to_u64(
+    op_iterator: impl Iterator<Item = ColumnOperation<u64>>,
     cardinality: Cardinality,
     num_docs: RowId,
+    sort_values_within_row: bool,
     value_index_builders: &mut PreallocatedIndexBuilders,
-    values: &mut Vec<T>,
+    values: &mut Vec<u64>,
     mut wrt: impl io::Write,
 ) -> io::Result<()>
 where
-    for<'a> VecColumn<'a, T>: ColumnValues<T>,
+    for<'a> VecColumn<'a, u64>: ColumnValues<u64>,
 {
     values.clear();
     let serializable_column_index = match cardinality {
@@ -610,6 +632,9 @@ where
             let multivalued_index_builder = value_index_builders.borrow_multivalued_index_builder();
             consume_operation_iterator(op_iterator, multivalued_index_builder, values);
             let multivalued_index = multivalued_index_builder.finish(num_docs);
+            if sort_values_within_row {
+                sort_values_within_row_in_place(&multivalued_index, values);
+            }
             SerializableColumnIndex::Multivalued(Box::new(multivalued_index))
         }
     };
@@ -623,17 +648,17 @@ where
 
 fn coerce_numerical_symbol<T>(
     operation_iterator: impl Iterator<Item = ColumnOperation<NumericalValue>>,
-) -> impl Iterator<Item = ColumnOperation<T>>
-where T: Coerce {
+) -> impl Iterator<Item = ColumnOperation<u64>>
+where T: Coerce + MonotonicallyMappableToU64 {
     operation_iterator.map(|symbol| match symbol {
         ColumnOperation::NewDoc(doc) => ColumnOperation::NewDoc(doc),
         ColumnOperation::Value(numerical_value) => {
-            ColumnOperation::Value(Coerce::coerce(numerical_value))
+            ColumnOperation::Value(T::coerce(numerical_value).to_u64())
         }
     })
 }
 
-fn consume_operation_iterator<T: std::fmt::Debug, TIndexBuilder: IndexBuilder>(
+fn consume_operation_iterator<T: Ord, TIndexBuilder: IndexBuilder>(
     operation_iterator: impl Iterator<Item = ColumnOperation<T>>,
     index_builder: &mut TIndexBuilder,
     values: &mut Vec<T>,

columnar/src/lib.rs

@@ -19,16 +19,20 @@ pub(crate) mod utils;
 mod value;
 
 pub use column::{BytesColumn, Column, StrColumn};
-pub use column_values::ColumnValues;
+pub use column_index::ColumnIndex;
+pub use column_values::{ColumnValues, MonotonicallyMappableToU128, MonotonicallyMappableToU64};
 pub use columnar::{
     merge_columnar, ColumnType, ColumnarReader, ColumnarWriter, HasAssociatedColumnType,
     MergeDocOrder,
 };
+use sstable::VoidSSTable;
 pub use value::{NumericalType, NumericalValue};
 
 pub use self::dynamic_column::{DynamicColumn, DynamicColumnHandle};
 
 pub type RowId = u32;
+pub use sstable::Dictionary;
+pub type Streamer<'a> = sstable::Streamer<'a, VoidSSTable>;
 
 #[derive(Clone, Copy, PartialOrd, PartialEq, Default, Debug)]
 pub struct DateTime {

examples-disabled/aggregation.rs

@@ -13,7 +13,7 @@ use tantivy::aggregation::agg_result::AggregationResults;
 use tantivy::aggregation::metric::AverageAggregation;
 use tantivy::aggregation::AggregationCollector;
 use tantivy::query::TermQuery;
-use tantivy::schema::{self, Cardinality, IndexRecordOption, Schema, TextFieldIndexing};
+use tantivy::schema::{self, IndexRecordOption, Schema, TextFieldIndexing};
 use tantivy::{doc, Index, Term};
 
 fn main() -> tantivy::Result<()> {
@@ -25,7 +25,7 @@ fn main() -> tantivy::Result<()> {
         .set_stored();
     let text_field = schema_builder.add_text_field("text", text_fieldtype);
     let score_fieldtype =
-        crate::schema::NumericOptions::default().set_fast(Cardinality::SingleValue);
+        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);
 

examples-disabled/date_time_field.rs

@@ -4,7 +4,7 @@
 
 use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
-use tantivy::schema::{Cardinality, DateOptions, Schema, Value, INDEXED, STORED, STRING};
+use tantivy::schema::{DateOptions, Schema, Value, INDEXED, STORED, STRING};
 use tantivy::Index;
 
 fn main() -> tantivy::Result<()> {
@@ -12,7 +12,7 @@ fn main() -> tantivy::Result<()> {
     let mut schema_builder = Schema::builder();
     let opts = DateOptions::from(INDEXED)
         .set_stored()
-        .set_fast(Cardinality::SingleValue)
+        .set_fast()
         .set_precision(tantivy::DatePrecision::Seconds);
     let occurred_at = schema_builder.add_date_field("occurred_at", opts);
     let event_type = schema_builder.add_text_field("event", STRING | STORED);

fastfield_codecs/Cargo.toml

@@ -14,6 +14,7 @@ repository = "https://github.com/quickwit-oss/tantivy"
 [dependencies]
 common = { version = "0.5", path = "../common/", package = "tantivy-common" }
 tantivy-bitpacker = { version= "0.3", path = "../bitpacker/" }
+columnar = { version= "0.1", path="../columnar", package="tantivy-columnar" }
 prettytable-rs = {version="0.10.0", optional= true}
 rand = {version="0.8.3", optional= true}
 fastdivide = "0.4"

fastfield_codecs/src/bitpacked.rs

@@ -1,116 +0,0 @@
-use std::io::{self, Write};
-
-use common::OwnedBytes;
-use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
-
-use crate::serialize::NormalizedHeader;
-use crate::{Column, FastFieldCodec, FastFieldCodecType};
-
-/// Depending on the field type, a different
-/// fast field is required.
-#[derive(Clone)]
-pub struct BitpackedReader {
-    data: OwnedBytes,
-    bit_unpacker: BitUnpacker,
-    normalized_header: NormalizedHeader,
-}
-
-impl Column for BitpackedReader {
-    #[inline]
-    fn get_val(&self, doc: u32) -> u64 {
-        self.bit_unpacker.get(doc, &self.data)
-    }
-    #[inline]
-    fn min_value(&self) -> u64 {
-        // The BitpackedReader assumes a normalized vector.
-        0
-    }
-    #[inline]
-    fn max_value(&self) -> u64 {
-        self.normalized_header.max_value
-    }
-    #[inline]
-    fn num_vals(&self) -> u32 {
-        self.normalized_header.num_vals
-    }
-}
-
-pub struct BitpackedCodec;
-
-impl FastFieldCodec for BitpackedCodec {
-    /// The CODEC_TYPE is an enum value used for serialization.
-    const CODEC_TYPE: FastFieldCodecType = FastFieldCodecType::Bitpacked;
-
-    type Reader = BitpackedReader;
-
-    /// Opens a fast field given a file.
-    fn open_from_bytes(
-        data: OwnedBytes,
-        normalized_header: NormalizedHeader,
-    ) -> io::Result<Self::Reader> {
-        let num_bits = compute_num_bits(normalized_header.max_value);
-        let bit_unpacker = BitUnpacker::new(num_bits);
-        Ok(BitpackedReader {
-            data,
-            bit_unpacker,
-            normalized_header,
-        })
-    }
-
-    /// Serializes data with the BitpackedFastFieldSerializer.
-    ///
-    /// The bitpacker assumes that the column has been normalized.
-    /// i.e. It has already been shifted by its minimum value, so that its
-    /// current minimum value is 0.
-    ///
-    /// Ideally, we made a shift upstream on the column so that `col.min_value() == 0`.
-    fn serialize(column: &dyn Column, write: &mut impl Write) -> io::Result<()> {
-        assert_eq!(column.min_value(), 0u64);
-        let num_bits = compute_num_bits(column.max_value());
-        let mut bit_packer = BitPacker::new();
-        for val in column.iter() {
-            bit_packer.write(val, num_bits, write)?;
-        }
-        bit_packer.close(write)?;
-        Ok(())
-    }
-
-    fn estimate(column: &dyn Column) -> Option<f32> {
-        let num_bits = compute_num_bits(column.max_value());
-        let num_bits_uncompressed = 64;
-        Some(num_bits as f32 / num_bits_uncompressed as f32)
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::tests::get_codec_test_datasets;
-
-    fn create_and_validate(data: &[u64], name: &str) {
-        crate::tests::create_and_validate::<BitpackedCodec>(data, name);
-    }
-
-    #[test]
-    fn test_with_codec_data_sets() {
-        let data_sets = get_codec_test_datasets();
-        for (mut data, name) in data_sets {
-            create_and_validate(&data, name);
-            data.reverse();
-            create_and_validate(&data, name);
-        }
-    }
-
-    #[test]
-    fn bitpacked_fast_field_rand() {
-        for _ in 0..500 {
-            let mut data = (0..1 + rand::random::<u8>() as usize)
-                .map(|_| rand::random::<i64>() as u64 / 2)
-                .collect::<Vec<_>>();
-            create_and_validate(&data, "rand");
-
-            data.reverse();
-            create_and_validate(&data, "rand");
-        }
-    }
-}

fastfield_codecs/src/blockwise_linear.rs

@@ -1,188 +0,0 @@
-use std::sync::Arc;
-use std::{io, iter};
-
-use common::{BinarySerializable, CountingWriter, DeserializeFrom, OwnedBytes};
-use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
-
-use crate::line::Line;
-use crate::serialize::NormalizedHeader;
-use crate::{Column, FastFieldCodec, FastFieldCodecType, VecColumn};
-
-const CHUNK_SIZE: usize = 512;
-
-#[derive(Debug, Default)]
-struct Block {
-    line: Line,
-    bit_unpacker: BitUnpacker,
-    data_start_offset: usize,
-}
-
-impl BinarySerializable for Block {
-    fn serialize<W: io::Write>(&self, writer: &mut W) -> io::Result<()> {
-        self.line.serialize(writer)?;
-        self.bit_unpacker.bit_width().serialize(writer)?;
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let line = Line::deserialize(reader)?;
-        let bit_width = u8::deserialize(reader)?;
-        Ok(Block {
-            line,
-            bit_unpacker: BitUnpacker::new(bit_width),
-            data_start_offset: 0,
-        })
-    }
-}
-
-fn compute_num_blocks(num_vals: u32) -> usize {
-    (num_vals as usize + CHUNK_SIZE - 1) / CHUNK_SIZE
-}
-
-pub struct BlockwiseLinearCodec;
-
-impl FastFieldCodec for BlockwiseLinearCodec {
-    const CODEC_TYPE: crate::FastFieldCodecType = FastFieldCodecType::BlockwiseLinear;
-    type Reader = BlockwiseLinearReader;
-
-    fn open_from_bytes(
-        bytes: common::OwnedBytes,
-        normalized_header: NormalizedHeader,
-    ) -> io::Result<Self::Reader> {
-        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);
-        let num_blocks = compute_num_blocks(normalized_header.num_vals);
-        let mut blocks: Vec<Block> = iter::repeat_with(|| Block::deserialize(&mut footer))
-            .take(num_blocks)
-            .collect::<io::Result<_>>()?;
-
-        let mut start_offset = 0;
-        for block in &mut blocks {
-            block.data_start_offset = start_offset;
-            start_offset += (block.bit_unpacker.bit_width() as usize) * CHUNK_SIZE / 8;
-        }
-        Ok(BlockwiseLinearReader {
-            blocks: Arc::new(blocks),
-            data,
-            normalized_header,
-        })
-    }
-
-    // Estimate first_chunk and extrapolate
-    fn estimate(column: &dyn crate::Column) -> Option<f32> {
-        if column.num_vals() < 10 * CHUNK_SIZE as u32 {
-            return None;
-        }
-        let mut first_chunk: Vec<u64> = column.iter().take(CHUNK_SIZE).collect();
-        let line = Line::train(&VecColumn::from(&first_chunk));
-        for (i, buffer_val) in first_chunk.iter_mut().enumerate() {
-            let interpolated_val = line.eval(i as u32);
-            *buffer_val = buffer_val.wrapping_sub(interpolated_val);
-        }
-        let estimated_bit_width = first_chunk
-            .iter()
-            .map(|el| ((el + 1) as f32 * 3.0) as u64)
-            .map(compute_num_bits)
-            .max()
-            .unwrap();
-
-        let metadata_per_block = {
-            let mut out = vec![];
-            Block::default().serialize(&mut out).unwrap();
-            out.len()
-        };
-        let num_bits = estimated_bit_width as u64 * column.num_vals() as u64
-            // function metadata per block
-            + metadata_per_block as u64 * (column.num_vals() as u64 / CHUNK_SIZE as u64);
-        let num_bits_uncompressed = 64 * column.num_vals();
-        Some(num_bits as f32 / num_bits_uncompressed as f32)
-    }
-
-    fn serialize(column: &dyn Column, wrt: &mut impl io::Write) -> io::Result<()> {
-        // The BitpackedReader assumes a normalized vector.
-        assert_eq!(column.min_value(), 0);
-        let mut buffer = Vec::with_capacity(CHUNK_SIZE);
-        let num_vals = column.num_vals();
-
-        let num_blocks = compute_num_blocks(num_vals);
-        let mut blocks = Vec::with_capacity(num_blocks);
-
-        let mut vals = column.iter();
-
-        let mut bit_packer = BitPacker::new();
-
-        for _ in 0..num_blocks {
-            buffer.clear();
-            buffer.extend((&mut vals).take(CHUNK_SIZE));
-            let line = Line::train(&VecColumn::from(&buffer));
-
-            assert!(!buffer.is_empty());
-
-            for (i, buffer_val) in buffer.iter_mut().enumerate() {
-                let interpolated_val = line.eval(i as u32);
-                *buffer_val = buffer_val.wrapping_sub(interpolated_val);
-            }
-            let bit_width = buffer.iter().copied().map(compute_num_bits).max().unwrap();
-
-            for &buffer_val in &buffer {
-                bit_packer.write(buffer_val, bit_width, wrt)?;
-            }
-
-            blocks.push(Block {
-                line,
-                bit_unpacker: BitUnpacker::new(bit_width),
-                data_start_offset: 0,
-            });
-        }
-
-        bit_packer.close(wrt)?;
-
-        assert_eq!(blocks.len(), compute_num_blocks(num_vals));
-
-        let mut counting_wrt = CountingWriter::wrap(wrt);
-        for block in &blocks {
-            block.serialize(&mut counting_wrt)?;
-        }
-        let footer_len = counting_wrt.written_bytes();
-        (footer_len as u32).serialize(&mut counting_wrt)?;
-
-        Ok(())
-    }
-}
-
-#[derive(Clone)]
-pub struct BlockwiseLinearReader {
-    blocks: Arc<Vec<Block>>,
-    normalized_header: NormalizedHeader,
-    data: OwnedBytes,
-}
-
-impl Column for BlockwiseLinearReader {
-    #[inline(always)]
-    fn get_val(&self, idx: u32) -> u64 {
-        let block_id = (idx / CHUNK_SIZE as u32) as usize;
-        let idx_within_block = idx % (CHUNK_SIZE as u32);
-        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..];
-        let bitpacked_diff = block.bit_unpacker.get(idx_within_block, block_bytes);
-        interpoled_val.wrapping_add(bitpacked_diff)
-    }
-
-    #[inline(always)]
-    fn min_value(&self) -> u64 {
-        // The BlockwiseLinearReader assumes a normalized vector.
-        0u64
-    }
-
-    #[inline(always)]
-    fn max_value(&self) -> u64 {
-        self.normalized_header.max_value
-    }
-
-    #[inline(always)]
-    fn num_vals(&self) -> u32 {
-        self.normalized_header.num_vals
-    }
-}

fastfield_codecs/src/column.rs

@@ -1,352 +0,0 @@
-use std::fmt::{self, Debug};
-use std::marker::PhantomData;
-use std::ops::{Range, RangeInclusive};
-
-use tantivy_bitpacker::minmax;
-
-use crate::monotonic_mapping::StrictlyMonotonicFn;
-
-/// `Column` provides columnar access on a field.
-pub trait Column<T: PartialOrd + Debug = u64>: Send + Sync {
-    /// Return the value associated with the given idx.
-    ///
-    /// This accessor should return as fast as possible.
-    ///
-    /// # 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]
-    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]
-    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)))
-    }
-}
-
-/// VecColumn provides `Column` over a slice.
-pub struct VecColumn<'a, T = u64> {
-    values: &'a [T],
-    min_value: T,
-    max_value: T,
-}
-
-impl<'a, C: Column<T>, T: Copy + PartialOrd + fmt::Debug> Column<T> for &'a C {
-    fn get_val(&self, idx: u32) -> T {
-        (*self).get_val(idx)
-    }
-
-    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)
-    }
-}
-
-impl<'a, T: Copy + PartialOrd + Send + Sync + Debug> Column<T> for VecColumn<'a, T> {
-    fn get_val(&self, position: u32) -> T {
-        self.values[position as usize]
-    }
-
-    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 Column<Output>
-where
-    C: Column<Input>,
-    T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
-    Input: PartialOrd + Send + Sync + Copy + Debug,
-    Output: PartialOrd + Send + Sync + Copy + Debug,
-{
-    MonotonicMappingColumn {
-        from_column,
-        monotonic_mapping,
-        _phantom: PhantomData,
-    }
-}
-
-impl<C, T, Input, Output> Column<Output> for MonotonicMappingColumn<C, T, Input>
-where
-    C: Column<Input>,
-    T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
-    Input: PartialOrd + Send + Sync + Copy + Debug,
-    Output: PartialOrd + Send + Sync + Copy + Debug,
-{
-    #[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>,
-    ) {
-        if range.start() > &self.max_value() || range.end() < &self.min_value() {
-            return;
-        }
-        let range = self.monotonic_mapping.inverse_coerce(range);
-        if range.start() > range.end() {
-            return;
-        }
-        self.from_column
-            .get_docids_for_value_range(range, doc_id_range, positions)
-    }
-
-    // We voluntarily do not implement get_range as it yields a regression,
-    // 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> Column<T::Item> for IterColumn<T>
-where
-    T: Iterator + Clone + ExactSizeIterator + Send + Sync,
-    T::Item: PartialOrd + fmt::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::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);
-    }
-}

fastfield_codecs/src/compact_space/blank_range.rs

@@ -1,43 +0,0 @@
-use std::ops::RangeInclusive;
-
-/// The range of a blank in value space.
-///
-/// A blank is an unoccupied space in the data.
-/// Use try_into() to construct.
-/// A range has to have at least length of 3. Invalid ranges will be rejected.
-///
-/// Ordered by range length.
-#[derive(Debug, Eq, PartialEq, Clone)]
-pub(crate) struct BlankRange {
-    blank_range: RangeInclusive<u128>,
-}
-impl TryFrom<RangeInclusive<u128>> for BlankRange {
-    type Error = &'static str;
-    fn try_from(range: RangeInclusive<u128>) -> Result<Self, Self::Error> {
-        let blank_size = range.end().saturating_sub(*range.start());
-        if blank_size < 2 {
-            Err("invalid range")
-        } else {
-            Ok(BlankRange { blank_range: range })
-        }
-    }
-}
-impl BlankRange {
-    pub(crate) fn blank_size(&self) -> u128 {
-        self.blank_range.end() - self.blank_range.start() + 1
-    }
-    pub(crate) fn blank_range(&self) -> RangeInclusive<u128> {
-        self.blank_range.clone()
-    }
-}
-
-impl Ord for BlankRange {
-    fn cmp(&self, other: &Self) -> std::cmp::Ordering {
-        self.blank_size().cmp(&other.blank_size())
-    }
-}
-impl PartialOrd for BlankRange {
-    fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
-        Some(self.blank_size().cmp(&other.blank_size()))
-    }
-}

fastfield_codecs/src/compact_space/build_compact_space.rs

@@ -1,231 +0,0 @@
-use std::collections::{BTreeSet, BinaryHeap};
-use std::iter;
-use std::ops::RangeInclusive;
-
-use itertools::Itertools;
-
-use super::blank_range::BlankRange;
-use super::{CompactSpace, RangeMapping};
-
-/// Put the blanks for the sorted values into a binary heap
-fn get_blanks(values_sorted: &BTreeSet<u128>) -> BinaryHeap<BlankRange> {
-    let mut blanks: BinaryHeap<BlankRange> = BinaryHeap::new();
-    for (first, second) in values_sorted.iter().tuple_windows() {
-        // Correctness Overflow: the values are deduped and sorted (BTreeSet property), that means
-        // there's always space between two values.
-        let blank_range = first + 1..=second - 1;
-        let blank_range: Result<BlankRange, _> = blank_range.try_into();
-        if let Ok(blank_range) = blank_range {
-            blanks.push(blank_range);
-        }
-    }
-
-    blanks
-}
-
-struct BlankCollector {
-    blanks: Vec<BlankRange>,
-    staged_blanks_sum: u128,
-}
-impl BlankCollector {
-    fn new() -> Self {
-        Self {
-            blanks: vec![],
-            staged_blanks_sum: 0,
-        }
-    }
-    fn stage_blank(&mut self, blank: BlankRange) {
-        self.staged_blanks_sum += blank.blank_size();
-        self.blanks.push(blank);
-    }
-    fn drain(&mut self) -> impl Iterator<Item = BlankRange> + '_ {
-        self.staged_blanks_sum = 0;
-        self.blanks.drain(..)
-    }
-    fn staged_blanks_sum(&self) -> u128 {
-        self.staged_blanks_sum
-    }
-    fn num_staged_blanks(&self) -> usize {
-        self.blanks.len()
-    }
-}
-fn num_bits(val: u128) -> u8 {
-    (128u32 - val.leading_zeros()) as u8
-}
-
-/// Will collect blanks and add them to compact space if more bits are saved than cost from
-/// metadata.
-pub fn get_compact_space(
-    values_deduped_sorted: &BTreeSet<u128>,
-    total_num_values: u32,
-    cost_per_blank: usize,
-) -> CompactSpace {
-    let mut compact_space_builder = CompactSpaceBuilder::new();
-    if values_deduped_sorted.is_empty() {
-        return compact_space_builder.finish();
-    }
-
-    let mut blanks: BinaryHeap<BlankRange> = get_blanks(values_deduped_sorted);
-    // Replace after stabilization of https://github.com/rust-lang/rust/issues/62924
-
-    // We start by space that's limited to min_value..=max_value
-    let min_value = *values_deduped_sorted.iter().next().unwrap_or(&0);
-    let max_value = *values_deduped_sorted.iter().last().unwrap_or(&0);
-
-    // +1 for null, in case min and max covers the whole space, we are off by one.
-    let mut amplitude_compact_space = (max_value - min_value).saturating_add(1);
-    if min_value != 0 {
-        compact_space_builder.add_blanks(iter::once(0..=min_value - 1));
-    }
-    if max_value != u128::MAX {
-        compact_space_builder.add_blanks(iter::once(max_value + 1..=u128::MAX));
-    }
-
-    let mut amplitude_bits: u8 = num_bits(amplitude_compact_space);
-
-    let mut blank_collector = BlankCollector::new();
-    // We will stage blanks until they reduce the compact space by at least 1 bit and then flush
-    // them if the metadata cost is lower than the total number of saved bits.
-    // Binary heap to process the gaps by their size
-    while let Some(blank_range) = blanks.pop() {
-        blank_collector.stage_blank(blank_range);
-
-        let staged_spaces_sum: u128 = blank_collector.staged_blanks_sum();
-        let amplitude_new_compact_space = amplitude_compact_space - staged_spaces_sum;
-        let amplitude_new_bits = num_bits(amplitude_new_compact_space);
-        if amplitude_bits == amplitude_new_bits {
-            continue;
-        }
-        let saved_bits = (amplitude_bits - amplitude_new_bits) as usize * total_num_values as usize;
-        // TODO: Maybe calculate exact cost of blanks and run this more expensive computation only,
-        // when amplitude_new_bits changes
-        let cost = blank_collector.num_staged_blanks() * cost_per_blank;
-        if cost >= saved_bits {
-            // Continue here, since although we walk over the blanks by size,
-            // we can potentially save a lot at the last bits, which are smaller blanks
-            //
-            // E.g. if the first range reduces the compact space by 1000 from 2000 to 1000, which
-            // saves 11-10=1 bit and the next range reduces the compact space by 950 to
-            // 50, which saves 10-6=4 bit
-            continue;
-        }
-
-        amplitude_compact_space = amplitude_new_compact_space;
-        amplitude_bits = amplitude_new_bits;
-        compact_space_builder.add_blanks(blank_collector.drain().map(|blank| blank.blank_range()));
-    }
-
-    // special case, when we don't collected any blanks because:
-    // * the data is empty (early exit)
-    // * the algorithm did decide it's not worth the cost, which can be the case for single values
-    //
-    // We drain one collected blank unconditionally, so the empty case is reserved for empty
-    // data, and therefore empty compact_space means the data is empty and no data is covered
-    // (conversely to all data) and we can assign null to it.
-    if compact_space_builder.is_empty() {
-        compact_space_builder.add_blanks(
-            blank_collector
-                .drain()
-                .map(|blank| blank.blank_range())
-                .take(1),
-        );
-    }
-
-    let compact_space = compact_space_builder.finish();
-    if max_value - min_value != u128::MAX {
-        debug_assert_eq!(
-            compact_space.amplitude_compact_space(),
-            amplitude_compact_space
-        );
-    }
-    compact_space
-}
-
-#[derive(Debug, Clone, Eq, PartialEq)]
-struct CompactSpaceBuilder {
-    blanks: Vec<RangeInclusive<u128>>,
-}
-
-impl CompactSpaceBuilder {
-    /// Creates a new compact space builder which will initially cover the whole space.
-    fn new() -> Self {
-        Self { blanks: Vec::new() }
-    }
-
-    /// Assumes that repeated add_blank calls don't overlap and are not adjacent,
-    /// e.g. [3..=5, 5..=10] is not allowed
-    ///
-    /// Both of those assumptions are true when blanks are produced from sorted values.
-    fn add_blanks(&mut self, blank: impl Iterator<Item = RangeInclusive<u128>>) {
-        self.blanks.extend(blank);
-    }
-
-    fn is_empty(&self) -> bool {
-        self.blanks.is_empty()
-    }
-
-    /// Convert blanks to covered space and assign null value
-    fn finish(mut self) -> CompactSpace {
-        // sort by start. ranges are not allowed to overlap
-        self.blanks.sort_unstable_by_key(|blank| *blank.start());
-
-        let mut covered_space = Vec::with_capacity(self.blanks.len());
-
-        // begining of the blanks
-        if let Some(first_blank_start) = self.blanks.first().map(RangeInclusive::start) {
-            if *first_blank_start != 0 {
-                covered_space.push(0..=first_blank_start - 1);
-            }
-        }
-
-        // Between the blanks
-        let between_blanks = self.blanks.iter().tuple_windows().map(|(left, right)| {
-            assert!(
-                left.end() < right.start(),
-                "overlapping or adjacent ranges detected"
-            );
-            *left.end() + 1..=*right.start() - 1
-        });
-        covered_space.extend(between_blanks);
-
-        // end of the blanks
-        if let Some(last_blank_end) = self.blanks.last().map(RangeInclusive::end) {
-            if *last_blank_end != u128::MAX {
-                covered_space.push(last_blank_end + 1..=u128::MAX);
-            }
-        }
-
-        if covered_space.is_empty() {
-            covered_space.push(0..=0); // empty data case
-        };
-
-        let mut compact_start: u64 = 1; // 0 is reserved for `null`
-        let mut ranges_mapping: Vec<RangeMapping> = Vec::with_capacity(covered_space.len());
-        for cov in covered_space {
-            let range_mapping = super::RangeMapping {
-                value_range: cov,
-                compact_start,
-            };
-            let covered_range_len = range_mapping.range_length();
-            ranges_mapping.push(range_mapping);
-            compact_start += covered_range_len;
-        }
-        // println!("num ranges {}", ranges_mapping.len());
-        CompactSpace { ranges_mapping }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_binary_heap_pop_order() {
-        let mut blanks: BinaryHeap<BlankRange> = BinaryHeap::new();
-        blanks.push((0..=10).try_into().unwrap());
-        blanks.push((100..=200).try_into().unwrap());
-        blanks.push((100..=110).try_into().unwrap());
-        assert_eq!(blanks.pop().unwrap().blank_size(), 101);
-        assert_eq!(blanks.pop().unwrap().blank_size(), 11);
-    }
-}

fastfield_codecs/src/compact_space/mod.rs

@@ -1,815 +0,0 @@
-/// This codec takes a large number space (u128) and reduces it to a compact number space.
-///
-/// It will find spaces in the number range. For example:
-///
-/// 100, 101, 102, 103, 104, 50000, 50001
-/// could be mapped to
-/// 100..104 -> 0..4
-/// 50000..50001 -> 5..6
-///
-/// Compact space 0..=6 requires much less bits than 100..=50001
-///
-/// The codec is created to compress ip addresses, but may be employed in other use cases.
-use std::{
-    cmp::Ordering,
-    collections::BTreeSet,
-    io::{self, Write},
-    ops::{Range, RangeInclusive},
-};
-
-use common::{BinarySerializable, CountingWriter, OwnedBytes, VInt, VIntU128};
-use tantivy_bitpacker::{self, BitPacker, BitUnpacker};
-
-use crate::compact_space::build_compact_space::get_compact_space;
-use crate::Column;
-
-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.
-///
-/// The number is taken by looking at a real dataset. It is optimized for larger datasets.
-const COST_PER_BLANK_IN_BITS: usize = 36;
-
-#[derive(Debug, Clone, Eq, PartialEq)]
-pub struct CompactSpace {
-    ranges_mapping: Vec<RangeMapping>,
-}
-
-/// Maps the range from the original space to compact_start + range.len()
-#[derive(Debug, Clone, Eq, PartialEq)]
-struct RangeMapping {
-    value_range: RangeInclusive<u128>,
-    compact_start: u64,
-}
-impl RangeMapping {
-    fn range_length(&self) -> u64 {
-        (self.value_range.end() - self.value_range.start()) as u64 + 1
-    }
-
-    // The last value of the compact space in this range
-    fn compact_end(&self) -> u64 {
-        self.compact_start + self.range_length() - 1
-    }
-}
-
-impl BinarySerializable for CompactSpace {
-    fn serialize<W: io::Write>(&self, writer: &mut W) -> io::Result<()> {
-        VInt(self.ranges_mapping.len() as u64).serialize(writer)?;
-
-        let mut prev_value = 0;
-        for value_range in self
-            .ranges_mapping
-            .iter()
-            .map(|range_mapping| &range_mapping.value_range)
-        {
-            let blank_delta_start = value_range.start() - prev_value;
-            VIntU128(blank_delta_start).serialize(writer)?;
-            prev_value = *value_range.start();
-
-            let blank_delta_end = value_range.end() - prev_value;
-            VIntU128(blank_delta_end).serialize(writer)?;
-            prev_value = *value_range.end();
-        }
-
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let num_ranges = VInt::deserialize(reader)?.0;
-        let mut ranges_mapping: Vec<RangeMapping> = vec![];
-        let mut value = 0u128;
-        let mut compact_start = 1u64; // 0 is reserved for `null`
-        for _ in 0..num_ranges {
-            let blank_delta_start = VIntU128::deserialize(reader)?.0;
-            value += blank_delta_start;
-            let blank_start = value;
-
-            let blank_delta_end = VIntU128::deserialize(reader)?.0;
-            value += blank_delta_end;
-            let blank_end = value;
-
-            let range_mapping = RangeMapping {
-                value_range: blank_start..=blank_end,
-                compact_start,
-            };
-            let range_length = range_mapping.range_length();
-            ranges_mapping.push(range_mapping);
-            compact_start += range_length;
-        }
-
-        Ok(Self { ranges_mapping })
-    }
-}
-
-impl CompactSpace {
-    /// Amplitude is the value range of the compact space including the sentinel value used to
-    /// identify null values. The compact space is 0..=amplitude .
-    ///
-    /// It's only used to verify we don't exceed u64 number space, which would indicate a bug.
-    fn amplitude_compact_space(&self) -> u128 {
-        self.ranges_mapping
-            .last()
-            .map(|last_range| last_range.compact_end() as u128)
-            .unwrap_or(1) // compact space starts at 1, 0 == null
-    }
-
-    fn get_range_mapping(&self, pos: usize) -> &RangeMapping {
-        &self.ranges_mapping[pos]
-    }
-
-    /// Returns either Ok(the value in the compact space) or if it is outside the compact space the
-    /// Err(position where it would be inserted)
-    fn u128_to_compact(&self, value: u128) -> Result<u64, usize> {
-        self.ranges_mapping
-            .binary_search_by(|probe| {
-                let value_range = &probe.value_range;
-                if value < *value_range.start() {
-                    Ordering::Greater
-                } else if value > *value_range.end() {
-                    Ordering::Less
-                } else {
-                    Ordering::Equal
-                }
-            })
-            .map(|pos| {
-                let range_mapping = &self.ranges_mapping[pos];
-                let pos_in_range = (value - range_mapping.value_range.start()) as u64;
-                range_mapping.compact_start + pos_in_range
-            })
-    }
-
-    /// Unpacks a value from compact space u64 to u128 space
-    fn compact_to_u128(&self, compact: u64) -> u128 {
-        let pos = self
-            .ranges_mapping
-            .binary_search_by_key(&compact, |range_mapping| range_mapping.compact_start)
-            // Correctness: Overflow. The first range starts at compact space 0, the error from
-            // binary search can never be 0
-            .map_or_else(|e| e - 1, |v| v);
-
-        let range_mapping = &self.ranges_mapping[pos];
-        let diff = compact - range_mapping.compact_start;
-        range_mapping.value_range.start() + diff as u128
-    }
-}
-
-pub struct CompactSpaceCompressor {
-    params: IPCodecParams,
-}
-#[derive(Debug, Clone)]
-pub struct IPCodecParams {
-    compact_space: CompactSpace,
-    bit_unpacker: BitUnpacker,
-    min_value: u128,
-    max_value: u128,
-    num_vals: u32,
-    num_bits: u8,
-}
-
-impl CompactSpaceCompressor {
-    /// Taking the vals as Vec may cost a lot of memory. It is used to sort the vals.
-    pub fn train_from(iter: impl Iterator<Item = u128>, num_vals: u32) -> Self {
-        let mut values_sorted = BTreeSet::new();
-        values_sorted.extend(iter);
-        let total_num_values = num_vals;
-
-        let compact_space =
-            get_compact_space(&values_sorted, total_num_values, COST_PER_BLANK_IN_BITS);
-        let amplitude_compact_space = compact_space.amplitude_compact_space();
-
-        assert!(
-            amplitude_compact_space <= u64::MAX as u128,
-            "case unsupported."
-        );
-
-        let num_bits = tantivy_bitpacker::compute_num_bits(amplitude_compact_space as u64);
-        let min_value = *values_sorted.iter().next().unwrap_or(&0);
-        let max_value = *values_sorted.iter().last().unwrap_or(&0);
-        assert_eq!(
-            compact_space
-                .u128_to_compact(max_value)
-                .expect("could not convert max value to compact space"),
-            amplitude_compact_space as u64
-        );
-        CompactSpaceCompressor {
-            params: IPCodecParams {
-                compact_space,
-                bit_unpacker: BitUnpacker::new(num_bits),
-                min_value,
-                max_value,
-                num_vals: total_num_values,
-                num_bits,
-            },
-        }
-    }
-
-    fn write_footer(self, writer: &mut impl Write) -> io::Result<()> {
-        let writer = &mut CountingWriter::wrap(writer);
-        self.params.serialize(writer)?;
-
-        let footer_len = writer.written_bytes() as u32;
-        footer_len.serialize(writer)?;
-
-        Ok(())
-    }
-
-    pub fn compress_into(
-        self,
-        vals: impl Iterator<Item = u128>,
-        write: &mut impl Write,
-    ) -> io::Result<()> {
-        let mut bitpacker = BitPacker::default();
-        for val in vals {
-            let compact = self
-                .params
-                .compact_space
-                .u128_to_compact(val)
-                .map_err(|_| {
-                    io::Error::new(
-                        io::ErrorKind::InvalidData,
-                        "Could not convert value to compact_space. This is a bug.",
-                    )
-                })?;
-            bitpacker.write(compact, self.params.num_bits, write)?;
-        }
-        bitpacker.close(write)?;
-        self.write_footer(write)?;
-        Ok(())
-    }
-}
-
-#[derive(Debug, Clone)]
-pub struct CompactSpaceDecompressor {
-    data: OwnedBytes,
-    params: IPCodecParams,
-}
-
-impl BinarySerializable for IPCodecParams {
-    fn serialize<W: io::Write>(&self, writer: &mut W) -> io::Result<()> {
-        // header flags for future optional dictionary encoding
-        let footer_flags = 0u64;
-        footer_flags.serialize(writer)?;
-
-        VIntU128(self.min_value).serialize(writer)?;
-        VIntU128(self.max_value).serialize(writer)?;
-        VIntU128(self.num_vals as u128).serialize(writer)?;
-        self.num_bits.serialize(writer)?;
-
-        self.compact_space.serialize(writer)?;
-
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let _header_flags = u64::deserialize(reader)?;
-        let min_value = VIntU128::deserialize(reader)?.0;
-        let max_value = VIntU128::deserialize(reader)?.0;
-        let num_vals = VIntU128::deserialize(reader)?.0 as u32;
-        let num_bits = u8::deserialize(reader)?;
-        let compact_space = CompactSpace::deserialize(reader)?;
-
-        Ok(Self {
-            compact_space,
-            bit_unpacker: BitUnpacker::new(num_bits),
-            min_value,
-            max_value,
-            num_vals,
-            num_bits,
-        })
-    }
-}
-
-impl Column<u128> for CompactSpaceDecompressor {
-    #[inline]
-    fn get_val(&self, doc: u32) -> u128 {
-        self.get(doc)
-    }
-
-    fn min_value(&self) -> u128 {
-        self.min_value()
-    }
-
-    fn max_value(&self) -> u128 {
-        self.max_value()
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.params.num_vals
-    }
-
-    #[inline]
-    fn iter(&self) -> Box<dyn Iterator<Item = u128> + '_> {
-        Box::new(self.iter())
-    }
-
-    #[inline]
-    fn get_docids_for_value_range(
-        &self,
-        value_range: RangeInclusive<u128>,
-        positions_range: Range<u32>,
-        positions: &mut Vec<u32>,
-    ) {
-        self.get_positions_for_value_range(value_range, positions_range, positions)
-    }
-}
-
-impl CompactSpaceDecompressor {
-    pub fn open(data: OwnedBytes) -> io::Result<CompactSpaceDecompressor> {
-        let (data_slice, footer_len_bytes) = data.split_at(data.len() - 4);
-        let footer_len = u32::deserialize(&mut &footer_len_bytes[..])?;
-
-        let data_footer = &data_slice[data_slice.len() - footer_len as usize..];
-        let params = IPCodecParams::deserialize(&mut &data_footer[..])?;
-        let decompressor = CompactSpaceDecompressor { data, params };
-
-        Ok(decompressor)
-    }
-
-    /// Converting to compact space for the decompressor is more complex, since we may get values
-    /// which are outside the compact space. e.g. if we map
-    /// 1000 => 5
-    /// 2000 => 6
-    ///
-    /// and we want a mapping for 1005, there is no equivalent compact space. We instead return an
-    /// error with the index of the next range.
-    fn u128_to_compact(&self, value: u128) -> Result<u64, usize> {
-        self.params.compact_space.u128_to_compact(value)
-    }
-
-    fn compact_to_u128(&self, compact: u64) -> u128 {
-        self.params.compact_space.compact_to_u128(compact)
-    }
-
-    /// Comparing on compact space: Random dataset 0,24 (50% random hit) - 1.05 GElements/s
-    /// Comparing on compact space: Real dataset 1.08 GElements/s
-    ///
-    /// Comparing on original space: Real dataset .06 GElements/s (not completely optimized)
-    #[inline]
-    pub fn get_positions_for_value_range(
-        &self,
-        value_range: RangeInclusive<u128>,
-        position_range: Range<u32>,
-        positions: &mut Vec<u32>,
-    ) {
-        if value_range.start() > value_range.end() {
-            return;
-        }
-        let position_range = position_range.start..position_range.end.min(self.num_vals());
-        let from_value = *value_range.start();
-        let to_value = *value_range.end();
-        assert!(to_value >= from_value);
-        let compact_from = self.u128_to_compact(from_value);
-        let compact_to = self.u128_to_compact(to_value);
-
-        // Quick return, if both ranges fall into the same non-mapped space, the range can't cover
-        // any values, so we can early exit
-        match (compact_to, compact_from) {
-            (Err(pos1), Err(pos2)) if pos1 == pos2 => return,
-            _ => {}
-        }
-
-        let compact_from = compact_from.unwrap_or_else(|pos| {
-            // Correctness: Out of bounds, if this value is Err(last_index + 1), we early exit,
-            // since the to_value also mapps into the same non-mapped space
-            let range_mapping = self.params.compact_space.get_range_mapping(pos);
-            range_mapping.compact_start
-        });
-        // If there is no compact space, we go to the closest upperbound compact space
-        let compact_to = compact_to.unwrap_or_else(|pos| {
-            // Correctness: Overflow, if this value is Err(0), we early exit,
-            // since the from_value also mapps into the same non-mapped space
-
-            // Get end of previous range
-            let pos = pos - 1;
-            let range_mapping = self.params.compact_space.get_range_mapping(pos);
-            range_mapping.compact_end()
-        });
-
-        let range = compact_from..=compact_to;
-
-        let scan_num_docs = position_range.end - position_range.start;
-
-        let step_size = 4;
-        let cutoff = position_range.start + scan_num_docs - scan_num_docs % step_size;
-
-        let mut push_if_in_range = |idx, val| {
-            if range.contains(&val) {
-                positions.push(idx);
-            }
-        };
-        let get_val = |idx| self.params.bit_unpacker.get(idx, &self.data);
-        // unrolled loop
-        for idx in (position_range.start..cutoff).step_by(step_size as usize) {
-            let idx1 = idx;
-            let idx2 = idx + 1;
-            let idx3 = idx + 2;
-            let idx4 = idx + 3;
-            let val1 = get_val(idx1);
-            let val2 = get_val(idx2);
-            let val3 = get_val(idx3);
-            let val4 = get_val(idx4);
-            push_if_in_range(idx1, val1);
-            push_if_in_range(idx2, val2);
-            push_if_in_range(idx3, val3);
-            push_if_in_range(idx4, val4);
-        }
-
-        // handle rest
-        for idx in cutoff..position_range.end {
-            push_if_in_range(idx, get_val(idx));
-        }
-    }
-
-    #[inline]
-    fn iter_compact(&self) -> impl Iterator<Item = u64> + '_ {
-        (0..self.params.num_vals).map(move |idx| self.params.bit_unpacker.get(idx, &self.data))
-    }
-
-    #[inline]
-    fn iter(&self) -> impl Iterator<Item = u128> + '_ {
-        // TODO: Performance. It would be better to iterate on the ranges and check existence via
-        // the bit_unpacker.
-        self.iter_compact()
-            .map(|compact| self.compact_to_u128(compact))
-    }
-
-    #[inline]
-    pub fn get(&self, idx: u32) -> u128 {
-        let compact = self.params.bit_unpacker.get(idx, &self.data);
-        self.compact_to_u128(compact)
-    }
-
-    pub fn min_value(&self) -> u128 {
-        self.params.min_value
-    }
-
-    pub fn max_value(&self) -> u128 {
-        self.params.max_value
-    }
-}
-
-#[cfg(test)]
-mod tests {
-
-    use std::fmt;
-
-    use super::*;
-    use crate::format_version::read_format_version;
-    use crate::null_index_footer::read_null_index_footer;
-    use crate::serialize::U128Header;
-    use crate::{open_u128, serialize_u128};
-
-    #[test]
-    fn compact_space_test() {
-        let ips = &[
-            2u128, 4u128, 1000, 1001, 1002, 1003, 1004, 1005, 1008, 1010, 1012, 1260,
-        ]
-        .into_iter()
-        .collect();
-        let compact_space = get_compact_space(ips, ips.len() as u32, 11);
-        let amplitude = compact_space.amplitude_compact_space();
-        assert_eq!(amplitude, 17);
-        assert_eq!(1, compact_space.u128_to_compact(2).unwrap());
-        assert_eq!(2, compact_space.u128_to_compact(3).unwrap());
-        assert_eq!(compact_space.u128_to_compact(100).unwrap_err(), 1);
-
-        for (num1, num2) in (0..3).tuple_windows() {
-            assert_eq!(
-                compact_space.get_range_mapping(num1).compact_end() + 1,
-                compact_space.get_range_mapping(num2).compact_start
-            );
-        }
-
-        let mut output: Vec<u8> = Vec::new();
-        compact_space.serialize(&mut output).unwrap();
-
-        assert_eq!(
-            compact_space,
-            CompactSpace::deserialize(&mut &output[..]).unwrap()
-        );
-
-        for ip in ips {
-            let compact = compact_space.u128_to_compact(*ip).unwrap();
-            assert_eq!(compact_space.compact_to_u128(compact), *ip);
-        }
-    }
-
-    #[test]
-    fn compact_space_amplitude_test() {
-        let ips = &[100000u128, 1000000].into_iter().collect();
-        let compact_space = get_compact_space(ips, ips.len() as u32, 1);
-        let amplitude = compact_space.amplitude_compact_space();
-        assert_eq!(amplitude, 2);
-    }
-
-    fn test_all(mut data: OwnedBytes, expected: &[u128]) {
-        let _header = U128Header::deserialize(&mut data);
-        let decompressor = CompactSpaceDecompressor::open(data).unwrap();
-        for (idx, expected_val) in expected.iter().cloned().enumerate() {
-            let val = decompressor.get(idx as u32);
-            assert_eq!(val, expected_val);
-
-            let test_range = |range: RangeInclusive<u128>| {
-                let expected_positions = expected
-                    .iter()
-                    .positions(|val| range.contains(val))
-                    .map(|pos| pos as u32)
-                    .collect::<Vec<_>>();
-                let mut positions = Vec::new();
-                decompressor.get_positions_for_value_range(
-                    range,
-                    0..decompressor.num_vals(),
-                    &mut positions,
-                );
-                assert_eq!(positions, expected_positions);
-            };
-
-            test_range(expected_val.saturating_sub(1)..=expected_val);
-            test_range(expected_val..=expected_val);
-            test_range(expected_val..=expected_val.saturating_add(1));
-            test_range(expected_val.saturating_sub(1)..=expected_val.saturating_add(1));
-        }
-    }
-
-    fn test_aux_vals(u128_vals: &[u128]) -> OwnedBytes {
-        let mut out = Vec::new();
-        serialize_u128(
-            || u128_vals.iter().cloned(),
-            u128_vals.len() as u32,
-            &mut out,
-        )
-        .unwrap();
-
-        let data = OwnedBytes::new(out);
-        let (data, _format_version) = read_format_version(data).unwrap();
-        let (data, _null_index_footer) = read_null_index_footer(data).unwrap();
-        test_all(data.clone(), u128_vals);
-
-        data
-    }
-
-    #[test]
-    fn test_range_1() {
-        let vals = &[
-            1u128,
-            100u128,
-            3u128,
-            99999u128,
-            100000u128,
-            100001u128,
-            4_000_211_221u128,
-            4_000_211_222u128,
-            333u128,
-        ];
-        let mut data = test_aux_vals(vals);
-
-        let _header = U128Header::deserialize(&mut data);
-        let decomp = CompactSpaceDecompressor::open(data).unwrap();
-        let complete_range = 0..vals.len() as u32;
-        for (pos, val) in vals.iter().enumerate() {
-            let val = *val;
-            let pos = pos as u32;
-            let mut positions = Vec::new();
-            decomp.get_positions_for_value_range(val..=val, pos..pos + 1, &mut positions);
-            assert_eq!(positions, vec![pos]);
-        }
-
-        // handle docid range out of bounds
-        let positions = get_positions_for_value_range_helper(&decomp, 0..=1, 1..u32::MAX);
-        assert_eq!(positions, vec![]);
-
-        let positions =
-            get_positions_for_value_range_helper(&decomp, 0..=1, complete_range.clone());
-        assert_eq!(positions, vec![0]);
-        let positions =
-            get_positions_for_value_range_helper(&decomp, 0..=2, complete_range.clone());
-        assert_eq!(positions, vec![0]);
-        let positions =
-            get_positions_for_value_range_helper(&decomp, 0..=3, complete_range.clone());
-        assert_eq!(positions, vec![0, 2]);
-        assert_eq!(
-            get_positions_for_value_range_helper(
-                &decomp,
-                99999u128..=99999u128,
-                complete_range.clone()
-            ),
-            vec![3]
-        );
-        assert_eq!(
-            get_positions_for_value_range_helper(
-                &decomp,
-                99999u128..=100000u128,
-                complete_range.clone()
-            ),
-            vec![3, 4]
-        );
-        assert_eq!(
-            get_positions_for_value_range_helper(
-                &decomp,
-                99998u128..=100000u128,
-                complete_range.clone()
-            ),
-            vec![3, 4]
-        );
-        assert_eq!(
-            get_positions_for_value_range_helper(
-                &decomp,
-                99998u128..=99999u128,
-                complete_range.clone()
-            ),
-            vec![3]
-        );
-        assert_eq!(
-            get_positions_for_value_range_helper(
-                &decomp,
-                99998u128..=99998u128,
-                complete_range.clone()
-            ),
-            vec![]
-        );
-        assert_eq!(
-            get_positions_for_value_range_helper(
-                &decomp,
-                333u128..=333u128,
-                complete_range.clone()
-            ),
-            vec![8]
-        );
-        assert_eq!(
-            get_positions_for_value_range_helper(
-                &decomp,
-                332u128..=333u128,
-                complete_range.clone()
-            ),
-            vec![8]
-        );
-        assert_eq!(
-            get_positions_for_value_range_helper(
-                &decomp,
-                332u128..=334u128,
-                complete_range.clone()
-            ),
-            vec![8]
-        );
-        assert_eq!(
-            get_positions_for_value_range_helper(
-                &decomp,
-                333u128..=334u128,
-                complete_range.clone()
-            ),
-            vec![8]
-        );
-
-        assert_eq!(
-            get_positions_for_value_range_helper(
-                &decomp,
-                4_000_211_221u128..=5_000_000_000u128,
-                complete_range
-            ),
-            vec![6, 7]
-        );
-    }
-
-    #[test]
-    fn test_empty() {
-        let vals = &[];
-        let data = test_aux_vals(vals);
-        let _decomp = CompactSpaceDecompressor::open(data).unwrap();
-    }
-
-    #[test]
-    fn test_range_2() {
-        let vals = &[
-            100u128,
-            99999u128,
-            100000u128,
-            100001u128,
-            4_000_211_221u128,
-            4_000_211_222u128,
-            333u128,
-        ];
-        let mut data = test_aux_vals(vals);
-        let _header = U128Header::deserialize(&mut data);
-        let decomp = CompactSpaceDecompressor::open(data).unwrap();
-        let complete_range = 0..vals.len() as u32;
-        assert_eq!(
-            get_positions_for_value_range_helper(&decomp, 0..=5, complete_range.clone()),
-            vec![]
-        );
-        assert_eq!(
-            get_positions_for_value_range_helper(&decomp, 0..=100, complete_range.clone()),
-            vec![0]
-        );
-        assert_eq!(
-            get_positions_for_value_range_helper(&decomp, 0..=105, complete_range),
-            vec![0]
-        );
-    }
-
-    fn get_positions_for_value_range_helper<C: Column<T> + ?Sized, T: PartialOrd + fmt::Debug>(
-        column: &C,
-        value_range: RangeInclusive<T>,
-        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);
-        positions
-    }
-
-    #[test]
-    fn test_range_3() {
-        let vals = &[
-            200u128,
-            201,
-            202,
-            203,
-            204,
-            204,
-            206,
-            207,
-            208,
-            209,
-            210,
-            1_000_000,
-            5_000_000_000,
-        ];
-        let mut out = Vec::new();
-        serialize_u128(|| vals.iter().cloned(), vals.len() as u32, &mut out).unwrap();
-        let decomp = open_u128::<u128>(OwnedBytes::new(out)).unwrap();
-        let complete_range = 0..vals.len() as u32;
-
-        assert_eq!(
-            get_positions_for_value_range_helper(&*decomp, 199..=200, complete_range.clone()),
-            vec![0]
-        );
-
-        assert_eq!(
-            get_positions_for_value_range_helper(&*decomp, 199..=201, complete_range.clone()),
-            vec![0, 1]
-        );
-
-        assert_eq!(
-            get_positions_for_value_range_helper(&*decomp, 200..=200, complete_range.clone()),
-            vec![0]
-        );
-
-        assert_eq!(
-            get_positions_for_value_range_helper(&*decomp, 1_000_000..=1_000_000, complete_range),
-            vec![11]
-        );
-    }
-
-    #[test]
-    fn test_bug1() {
-        let vals = &[9223372036854775806];
-        let _data = test_aux_vals(vals);
-    }
-
-    #[test]
-    fn test_bug2() {
-        let vals = &[340282366920938463463374607431768211455u128];
-        let _data = test_aux_vals(vals);
-    }
-
-    #[test]
-    fn test_bug3() {
-        let vals = &[340282366920938463463374607431768211454];
-        let _data = test_aux_vals(vals);
-    }
-
-    #[test]
-    fn test_bug4() {
-        let vals = &[340282366920938463463374607431768211455, 0];
-        let _data = test_aux_vals(vals);
-    }
-
-    #[test]
-    fn test_first_large_gaps() {
-        let vals = &[1_000_000_000u128; 100];
-        let _data = test_aux_vals(vals);
-    }
-    use itertools::Itertools;
-    use proptest::prelude::*;
-
-    fn num_strategy() -> impl Strategy<Value = u128> {
-        prop_oneof![
-            1 => prop::num::u128::ANY.prop_map(|num| u128::MAX - (num % 10) ),
-            1 => prop::num::u128::ANY.prop_map(|num| i64::MAX as u128 + 5 - (num % 10) ),
-            1 => prop::num::u128::ANY.prop_map(|num| i128::MAX as u128 + 5 - (num % 10) ),
-            1 => prop::num::u128::ANY.prop_map(|num| num % 10 ),
-            20 => prop::num::u128::ANY,
-        ]
-    }
-
-    proptest! {
-        #![proptest_config(ProptestConfig::with_cases(10))]
-
-            #[test]
-            fn compress_decompress_random(vals in proptest::collection::vec(num_strategy()
-    , 1..1000)) {
-                let _data = test_aux_vals(&vals);
-            }
-        }
-}

fastfield_codecs/src/format_version.rs

@@ -1,38 +0,0 @@
-use std::io;
-
-use common::{BinarySerializable, OwnedBytes};
-
-const MAGIC_NUMBER: u16 = 4335u16;
-const FASTFIELD_FORMAT_VERSION: u8 = 1;
-
-pub(crate) fn append_format_version(output: &mut impl io::Write) -> io::Result<()> {
-    FASTFIELD_FORMAT_VERSION.serialize(output)?;
-    MAGIC_NUMBER.serialize(output)?;
-
-    Ok(())
-}
-
-pub(crate) fn read_format_version(data: OwnedBytes) -> io::Result<(OwnedBytes, u8)> {
-    let (data, magic_number_bytes) = data.rsplit(2);
-
-    let magic_number = u16::deserialize(&mut magic_number_bytes.as_slice())?;
-    if magic_number != MAGIC_NUMBER {
-        return Err(io::Error::new(
-            io::ErrorKind::InvalidData,
-            format!("magic number mismatch {} != {}", magic_number, MAGIC_NUMBER),
-        ));
-    }
-    let (data, format_version_bytes) = data.rsplit(1);
-    let format_version = u8::deserialize(&mut format_version_bytes.as_slice())?;
-    if format_version > FASTFIELD_FORMAT_VERSION {
-        return Err(io::Error::new(
-            io::ErrorKind::InvalidData,
-            format!(
-                "Unsupported fastfield format version: {}. Max supported version: {}",
-                format_version, FASTFIELD_FORMAT_VERSION
-            ),
-        ));
-    }
-
-    Ok((data, format_version))
-}

fastfield_codecs/src/gcd.rs

@@ -1,170 +0,0 @@
-use std::num::NonZeroU64;
-
-use fastdivide::DividerU64;
-
-/// Compute the gcd of two non null numbers.
-///
-/// It is recommended, but not required, to feed values such that `large >= small`.
-fn compute_gcd(mut large: NonZeroU64, mut small: NonZeroU64) -> NonZeroU64 {
-    loop {
-        let rem: u64 = large.get() % small;
-        if let Some(new_small) = NonZeroU64::new(rem) {
-            (large, small) = (small, new_small);
-        } else {
-            return small;
-        }
-    }
-}
-
-// Find GCD for iterator of numbers
-pub fn find_gcd(numbers: impl Iterator<Item = u64>) -> Option<NonZeroU64> {
-    let mut numbers = numbers.flat_map(NonZeroU64::new);
-    let mut gcd: NonZeroU64 = numbers.next()?;
-    if gcd.get() == 1 {
-        return Some(gcd);
-    }
-
-    let mut gcd_divider = DividerU64::divide_by(gcd.get());
-    for val in numbers {
-        let remainder = val.get() - (gcd_divider.divide(val.get())) * gcd.get();
-        if remainder == 0 {
-            continue;
-        }
-        gcd = compute_gcd(val, gcd);
-        if gcd.get() == 1 {
-            return Some(gcd);
-        }
-
-        gcd_divider = DividerU64::divide_by(gcd.get());
-    }
-    Some(gcd)
-}
-
-#[cfg(test)]
-mod tests {
-    use std::io;
-    use std::num::NonZeroU64;
-
-    use common::OwnedBytes;
-
-    use crate::gcd::{compute_gcd, find_gcd};
-    use crate::{FastFieldCodecType, VecColumn};
-
-    fn test_fastfield_gcd_i64_with_codec(
-        codec_type: FastFieldCodecType,
-        num_vals: usize,
-    ) -> io::Result<()> {
-        let mut vals: Vec<i64> = (-4..=(num_vals as i64) - 5).map(|val| val * 1000).collect();
-        let mut buffer: Vec<u8> = Vec::new();
-        crate::serialize(VecColumn::from(&vals), &mut buffer, &[codec_type])?;
-        let buffer = OwnedBytes::new(buffer);
-        let column = crate::open::<i64>(buffer.clone())?;
-        assert_eq!(column.get_val(0), -4000i64);
-        assert_eq!(column.get_val(1), -3000i64);
-        assert_eq!(column.get_val(2), -2000i64);
-        assert_eq!(column.max_value(), (num_vals as i64 - 5) * 1000);
-        assert_eq!(column.min_value(), -4000i64);
-
-        // Can't apply gcd
-        let mut buffer_without_gcd = Vec::new();
-        vals.pop();
-        vals.push(1001i64);
-        crate::serialize(
-            VecColumn::from(&vals),
-            &mut buffer_without_gcd,
-            &[codec_type],
-        )?;
-        let buffer_without_gcd = OwnedBytes::new(buffer_without_gcd);
-        assert!(buffer_without_gcd.len() > buffer.len());
-
-        Ok(())
-    }
-
-    #[test]
-    fn test_fastfield_gcd_i64() -> io::Result<()> {
-        for &codec_type in &[
-            FastFieldCodecType::Bitpacked,
-            FastFieldCodecType::BlockwiseLinear,
-            FastFieldCodecType::Linear,
-        ] {
-            test_fastfield_gcd_i64_with_codec(codec_type, 5500)?;
-        }
-        Ok(())
-    }
-
-    fn test_fastfield_gcd_u64_with_codec(
-        codec_type: FastFieldCodecType,
-        num_vals: usize,
-    ) -> io::Result<()> {
-        let mut vals: Vec<u64> = (1..=num_vals).map(|i| i as u64 * 1000u64).collect();
-        let mut buffer: Vec<u8> = Vec::new();
-        crate::serialize(VecColumn::from(&vals), &mut buffer, &[codec_type])?;
-        let buffer = OwnedBytes::new(buffer);
-        let column = crate::open::<u64>(buffer.clone())?;
-        assert_eq!(column.get_val(0), 1000u64);
-        assert_eq!(column.get_val(1), 2000u64);
-        assert_eq!(column.get_val(2), 3000u64);
-        assert_eq!(column.max_value(), num_vals as u64 * 1000);
-        assert_eq!(column.min_value(), 1000u64);
-
-        // Can't apply gcd
-        let mut buffer_without_gcd = Vec::new();
-        vals.pop();
-        vals.push(1001u64);
-        crate::serialize(
-            VecColumn::from(&vals),
-            &mut buffer_without_gcd,
-            &[codec_type],
-        )?;
-        let buffer_without_gcd = OwnedBytes::new(buffer_without_gcd);
-        assert!(buffer_without_gcd.len() > buffer.len());
-        Ok(())
-    }
-
-    #[test]
-    fn test_fastfield_gcd_u64() -> io::Result<()> {
-        for &codec_type in &[
-            FastFieldCodecType::Bitpacked,
-            FastFieldCodecType::BlockwiseLinear,
-            FastFieldCodecType::Linear,
-        ] {
-            test_fastfield_gcd_u64_with_codec(codec_type, 5500)?;
-        }
-        Ok(())
-    }
-
-    #[test]
-    pub fn test_fastfield2() {
-        let test_fastfield = crate::serialize_and_load(&[100u64, 200u64, 300u64]);
-        assert_eq!(test_fastfield.get_val(0), 100);
-        assert_eq!(test_fastfield.get_val(1), 200);
-        assert_eq!(test_fastfield.get_val(2), 300);
-    }
-
-    #[test]
-    fn test_compute_gcd() {
-        let test_compute_gcd_aux = |large, small, expected| {
-            let large = NonZeroU64::new(large).unwrap();
-            let small = NonZeroU64::new(small).unwrap();
-            let expected = NonZeroU64::new(expected).unwrap();
-            assert_eq!(compute_gcd(small, large), expected);
-            assert_eq!(compute_gcd(large, small), expected);
-        };
-        test_compute_gcd_aux(1, 4, 1);
-        test_compute_gcd_aux(2, 4, 2);
-        test_compute_gcd_aux(10, 25, 5);
-        test_compute_gcd_aux(25, 25, 25);
-    }
-
-    #[test]
-    fn find_gcd_test() {
-        assert_eq!(find_gcd([0].into_iter()), None);
-        assert_eq!(find_gcd([0, 10].into_iter()), NonZeroU64::new(10));
-        assert_eq!(find_gcd([10, 0].into_iter()), NonZeroU64::new(10));
-        assert_eq!(find_gcd([].into_iter()), None);
-        assert_eq!(find_gcd([15, 30, 5, 10].into_iter()), NonZeroU64::new(5));
-        assert_eq!(find_gcd([15, 16, 10].into_iter()), NonZeroU64::new(1));
-        assert_eq!(find_gcd([0, 5, 5, 5].into_iter()), NonZeroU64::new(5));
-        assert_eq!(find_gcd([0, 0].into_iter()), None);
-    }
-}

fastfield_codecs/src/lib.rs

@@ -7,562 +7,4 @@
 //! - Encode data in different codecs.
 //! - Monotonically map values to u64/u128
 
-#[cfg(test)]
-#[macro_use]
-extern crate more_asserts;
-
-#[cfg(all(test, feature = "unstable"))]
-extern crate test;
-
-use std::io::Write;
-use std::sync::Arc;
-use std::{fmt, io};
-
-use common::{BinarySerializable, OwnedBytes};
-use compact_space::CompactSpaceDecompressor;
-use format_version::read_format_version;
-use monotonic_mapping::{
-    StrictlyMonotonicMappingInverter, StrictlyMonotonicMappingToInternal,
-    StrictlyMonotonicMappingToInternalBaseval, StrictlyMonotonicMappingToInternalGCDBaseval,
-};
-use null_index_footer::read_null_index_footer;
-use serialize::{Header, U128Header};
-
-mod bitpacked;
-mod blockwise_linear;
-mod compact_space;
-mod format_version;
-mod line;
-mod linear;
-mod monotonic_mapping;
-mod monotonic_mapping_u128;
-#[allow(dead_code)]
-mod null_index;
-mod null_index_footer;
-
-mod column;
-mod gcd;
-pub mod serialize;
-
-use self::bitpacked::BitpackedCodec;
-use self::blockwise_linear::BlockwiseLinearCodec;
-pub use self::column::{monotonic_map_column, Column, IterColumn, VecColumn};
-use self::linear::LinearCodec;
-pub use self::monotonic_mapping::{MonotonicallyMappableToU64, StrictlyMonotonicFn};
-pub use self::monotonic_mapping_u128::MonotonicallyMappableToU128;
-pub use self::serialize::{
-    estimate, serialize, serialize_and_load, serialize_u128, NormalizedHeader,
-};
-
-#[derive(PartialEq, Eq, PartialOrd, Ord, Debug, Clone, Copy)]
-#[repr(u8)]
-/// Available codecs to use to encode the u64 (via [`MonotonicallyMappableToU64`]) converted data.
-pub enum FastFieldCodecType {
-    /// Bitpack all values in the value range. The number of bits is defined by the amplitude
-    /// `column.max_value() - column.min_value()`
-    Bitpacked = 1,
-    /// Linear interpolation puts a line between the first and last value and then bitpacks the
-    /// values by the offset from the line. The number of bits is defined by the max deviation from
-    /// the line.
-    Linear = 2,
-    /// Same as [`FastFieldCodecType::Linear`], but encodes in blocks of 512 elements.
-    BlockwiseLinear = 3,
-}
-
-impl BinarySerializable for FastFieldCodecType {
-    fn serialize<W: Write>(&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 FastFieldCodecType {
-    pub(crate) fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    pub(crate) fn from_code(code: u8) -> Option<Self> {
-        match code {
-            1 => Some(Self::Bitpacked),
-            2 => Some(Self::Linear),
-            3 => Some(Self::BlockwiseLinear),
-            _ => None,
-        }
-    }
-}
-
-#[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>(&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<Item: MonotonicallyMappableToU128 + fmt::Debug>(
-    bytes: OwnedBytes,
-) -> io::Result<Arc<dyn Column<Item>>> {
-    let (bytes, _format_version) = read_format_version(bytes)?;
-    let (mut bytes, _null_index_footer) = read_null_index_footer(bytes)?;
-    let header = U128Header::deserialize(&mut bytes)?;
-    assert_eq!(header.codec_type, U128FastFieldCodecType::CompactSpace);
-    let reader = CompactSpaceDecompressor::open(bytes)?;
-    let inverted: StrictlyMonotonicMappingInverter<StrictlyMonotonicMappingToInternal<Item>> =
-        StrictlyMonotonicMappingToInternal::<Item>::new().into();
-    Ok(Arc::new(monotonic_map_column(reader, inverted)))
-}
-
-/// Returns the correct codec reader wrapped in the `Arc` for the data.
-pub fn open<T: MonotonicallyMappableToU64 + fmt::Debug>(
-    bytes: OwnedBytes,
-) -> io::Result<Arc<dyn Column<T>>> {
-    let (bytes, _format_version) = read_format_version(bytes)?;
-    let (mut bytes, _null_index_footer) = read_null_index_footer(bytes)?;
-    let header = Header::deserialize(&mut bytes)?;
-    match header.codec_type {
-        FastFieldCodecType::Bitpacked => open_specific_codec::<BitpackedCodec, _>(bytes, &header),
-        FastFieldCodecType::Linear => open_specific_codec::<LinearCodec, _>(bytes, &header),
-        FastFieldCodecType::BlockwiseLinear => {
-            open_specific_codec::<BlockwiseLinearCodec, _>(bytes, &header)
-        }
-    }
-}
-
-fn open_specific_codec<C: FastFieldCodec, Item: MonotonicallyMappableToU64 + fmt::Debug>(
-    bytes: OwnedBytes,
-    header: &Header,
-) -> io::Result<Arc<dyn Column<Item>>> {
-    let normalized_header = header.normalized();
-    let reader = C::open_from_bytes(bytes, normalized_header)?;
-    let min_value = header.min_value;
-    if let Some(gcd) = header.gcd {
-        let mapping = StrictlyMonotonicMappingInverter::from(
-            StrictlyMonotonicMappingToInternalGCDBaseval::new(gcd.get(), min_value),
-        );
-        Ok(Arc::new(monotonic_map_column(reader, mapping)))
-    } else {
-        let mapping = StrictlyMonotonicMappingInverter::from(
-            StrictlyMonotonicMappingToInternalBaseval::new(min_value),
-        );
-        Ok(Arc::new(monotonic_map_column(reader, mapping)))
-    }
-}
-
-/// The FastFieldSerializerEstimate trait is required on all variants
-/// of fast field compressions, to decide which one to choose.
-trait FastFieldCodec: 'static {
-    /// A codex needs to provide a unique name and id, which is
-    /// used for debugging and de/serialization.
-    const CODEC_TYPE: FastFieldCodecType;
-
-    type Reader: Column<u64> + 'static;
-
-    /// Reads the metadata and returns the CodecReader
-    fn open_from_bytes(bytes: OwnedBytes, header: NormalizedHeader) -> io::Result<Self::Reader>;
-
-    /// Serializes the data using the serializer into write.
-    ///
-    /// The column iterator should be preferred over using column `get_val` method for
-    /// performance reasons.
-    fn serialize(column: &dyn Column, write: &mut impl Write) -> io::Result<()>;
-
-    /// Returns an estimate of the compression ratio.
-    /// If the codec is not applicable, returns `None`.
-    ///
-    /// The baseline is uncompressed 64bit data.
-    ///
-    /// It could make sense to also return a value representing
-    /// computational complexity.
-    fn estimate(column: &dyn Column) -> Option<f32>;
-}
-
-/// The list of all available codecs for u64 convertible data.
-pub const ALL_CODEC_TYPES: [FastFieldCodecType; 3] = [
-    FastFieldCodecType::Bitpacked,
-    FastFieldCodecType::BlockwiseLinear,
-    FastFieldCodecType::Linear,
-];
-
-#[cfg(test)]
-mod tests {
-
-    use proptest::prelude::*;
-    use proptest::strategy::Strategy;
-    use proptest::{prop_oneof, proptest};
-
-    use crate::bitpacked::BitpackedCodec;
-    use crate::blockwise_linear::BlockwiseLinearCodec;
-    use crate::linear::LinearCodec;
-    use crate::serialize::Header;
-
-    pub(crate) fn create_and_validate<Codec: FastFieldCodec>(
-        data: &[u64],
-        name: &str,
-    ) -> Option<(f32, f32)> {
-        let col = &VecColumn::from(data);
-        let header = Header::compute_header(col, &[Codec::CODEC_TYPE])?;
-        let normalized_col = header.normalize_column(col);
-        let estimation = Codec::estimate(&normalized_col)?;
-
-        let mut out = Vec::new();
-        let col = VecColumn::from(data);
-        serialize(col, &mut out, &[Codec::CODEC_TYPE]).unwrap();
-
-        let actual_compression = out.len() as f32 / (data.len() as f32 * 8.0);
-
-        let reader = crate::open::<u64>(OwnedBytes::new(out)).unwrap();
-        assert_eq!(reader.num_vals(), data.len() as u32);
-        for (doc, orig_val) in data.iter().copied().enumerate() {
-            let val = reader.get_val(doc as u32);
-            assert_eq!(
-                val, orig_val,
-                "val `{val}` does not match orig_val {orig_val:?}, in data set {name}, data \
-                 `{data:?}`",
-            );
-        }
-
-        if !data.is_empty() {
-            let test_rand_idx = rand::thread_rng().gen_range(0..=data.len() - 1);
-            let expected_positions: Vec<u32> = data
-                .iter()
-                .enumerate()
-                .filter(|(_, el)| **el == data[test_rand_idx])
-                .map(|(pos, _)| pos as u32)
-                .collect();
-            let mut positions = Vec::new();
-            reader.get_docids_for_value_range(
-                data[test_rand_idx]..=data[test_rand_idx],
-                0..data.len() as u32,
-                &mut positions,
-            );
-            assert_eq!(expected_positions, positions);
-        }
-        Some((estimation, actual_compression))
-    }
-
-    proptest! {
-        #![proptest_config(ProptestConfig::with_cases(100))]
-
-        #[test]
-        fn test_proptest_small_bitpacked(data in proptest::collection::vec(num_strategy(), 1..10)) {
-            create_and_validate::<BitpackedCodec>(&data, "proptest bitpacked");
-        }
-
-        #[test]
-        fn test_proptest_small_linear(data in proptest::collection::vec(num_strategy(), 1..10)) {
-            create_and_validate::<LinearCodec>(&data, "proptest linearinterpol");
-        }
-
-        #[test]
-        fn test_proptest_small_blockwise_linear(data in proptest::collection::vec(num_strategy(), 1..10)) {
-            create_and_validate::<BlockwiseLinearCodec>(&data, "proptest multilinearinterpol");
-        }
-    }
-
-    proptest! {
-        #![proptest_config(ProptestConfig::with_cases(10))]
-
-        #[test]
-        fn test_proptest_large_bitpacked(data in proptest::collection::vec(num_strategy(), 1..6000)) {
-            create_and_validate::<BitpackedCodec>(&data, "proptest bitpacked");
-        }
-
-        #[test]
-        fn test_proptest_large_linear(data in proptest::collection::vec(num_strategy(), 1..6000)) {
-            create_and_validate::<LinearCodec>(&data, "proptest linearinterpol");
-        }
-
-        #[test]
-        fn test_proptest_large_blockwise_linear(data in proptest::collection::vec(num_strategy(), 1..6000)) {
-            create_and_validate::<BlockwiseLinearCodec>(&data, "proptest multilinearinterpol");
-        }
-    }
-
-    fn num_strategy() -> impl Strategy<Value = u64> {
-        prop_oneof![
-            1 => prop::num::u64::ANY.prop_map(|num| u64::MAX - (num % 10) ),
-            1 => prop::num::u64::ANY.prop_map(|num| num % 10 ),
-            20 => prop::num::u64::ANY,
-        ]
-    }
-
-    pub fn get_codec_test_datasets() -> Vec<(Vec<u64>, &'static str)> {
-        let mut data_and_names = vec![];
-
-        let data = vec![10];
-        data_and_names.push((data, "minimal test"));
-
-        let data = (10..=10_000_u64).collect::<Vec<_>>();
-        data_and_names.push((data, "simple monotonically increasing"));
-
-        data_and_names.push((
-            vec![5, 6, 7, 8, 9, 10, 99, 100],
-            "offset in linear interpol",
-        ));
-
-        data_and_names.push((vec![3, 18446744073709551613, 5], "docid range regression"));
-
-        data_and_names.push((vec![5, 50, 3, 13, 1, 1000, 35], "rand small"));
-        data_and_names.push((vec![10], "single value"));
-
-        data_and_names.push((
-            vec![1572656989877777, 1170935903116329, 720575940379279, 0],
-            "overflow error",
-        ));
-
-        data_and_names
-    }
-
-    fn test_codec<C: FastFieldCodec>() {
-        let codec_name = format!("{:?}", C::CODEC_TYPE);
-        for (data, dataset_name) in get_codec_test_datasets() {
-            let estimate_actual_opt: Option<(f32, f32)> =
-                crate::tests::create_and_validate::<C>(&data, dataset_name);
-            let result = if let Some((estimate, actual)) = estimate_actual_opt {
-                format!("Estimate `{estimate}` Actual `{actual}`")
-            } else {
-                "Disabled".to_string()
-            };
-            println!("Codec {codec_name}, DataSet {dataset_name}, {result}");
-        }
-    }
-    #[test]
-    fn test_codec_bitpacking() {
-        test_codec::<BitpackedCodec>();
-    }
-    #[test]
-    fn test_codec_interpolation() {
-        test_codec::<LinearCodec>();
-    }
-    #[test]
-    fn test_codec_multi_interpolation() {
-        test_codec::<BlockwiseLinearCodec>();
-    }
-
-    use super::*;
-
-    #[test]
-    fn estimation_good_interpolation_case() {
-        let data = (10..=20000_u64).collect::<Vec<_>>();
-        let data: VecColumn = data.as_slice().into();
-
-        let linear_interpol_estimation = LinearCodec::estimate(&data).unwrap();
-        assert_le!(linear_interpol_estimation, 0.01);
-
-        let multi_linear_interpol_estimation = BlockwiseLinearCodec::estimate(&data).unwrap();
-        assert_le!(multi_linear_interpol_estimation, 0.2);
-        assert_lt!(linear_interpol_estimation, multi_linear_interpol_estimation);
-
-        let bitpacked_estimation = BitpackedCodec::estimate(&data).unwrap();
-        assert_lt!(linear_interpol_estimation, bitpacked_estimation);
-    }
-    #[test]
-    fn estimation_test_bad_interpolation_case() {
-        let data: &[u64] = &[200, 10, 10, 10, 10, 1000, 20];
-
-        let data: VecColumn = data.into();
-        let linear_interpol_estimation = LinearCodec::estimate(&data).unwrap();
-        assert_le!(linear_interpol_estimation, 0.34);
-
-        let bitpacked_estimation = BitpackedCodec::estimate(&data).unwrap();
-        assert_lt!(bitpacked_estimation, linear_interpol_estimation);
-    }
-
-    #[test]
-    fn estimation_prefer_bitpacked() {
-        let data = VecColumn::from(&[10, 10, 10, 10]);
-        let linear_interpol_estimation = LinearCodec::estimate(&data).unwrap();
-        let bitpacked_estimation = BitpackedCodec::estimate(&data).unwrap();
-        assert_lt!(bitpacked_estimation, linear_interpol_estimation);
-    }
-
-    #[test]
-    fn estimation_test_bad_interpolation_case_monotonically_increasing() {
-        let mut data: Vec<u64> = (201..=20000_u64).collect();
-        data.push(1_000_000);
-        let data: VecColumn = data.as_slice().into();
-
-        // in this case the linear interpolation can't in fact not be worse than bitpacking,
-        // but the estimator adds some threshold, which leads to estimated worse behavior
-        let linear_interpol_estimation = LinearCodec::estimate(&data).unwrap();
-        assert_le!(linear_interpol_estimation, 0.35);
-
-        let bitpacked_estimation = BitpackedCodec::estimate(&data).unwrap();
-        assert_le!(bitpacked_estimation, 0.32);
-        assert_le!(bitpacked_estimation, linear_interpol_estimation);
-    }
-
-    #[test]
-    fn test_fast_field_codec_type_to_code() {
-        let mut count_codec = 0;
-        for code in 0..=255 {
-            if let Some(codec_type) = FastFieldCodecType::from_code(code) {
-                assert_eq!(codec_type.to_code(), code);
-                count_codec += 1;
-            }
-        }
-        assert_eq!(count_codec, 3);
-    }
-}
-
-#[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};
-
-    use super::*;
-    use crate::Column;
-
-    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 = crate::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 Column>) {
-        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);
-    }
-}
+pub use columnar::ColumnValues as Column;

fastfield_codecs/src/line.rs

@@ -1,222 +0,0 @@
-use std::io;
-use std::num::NonZeroU32;
-
-use common::{BinarySerializable, VInt};
-
-use crate::Column;
-
-const MID_POINT: u64 = (1u64 << 32) - 1u64;
-
-/// `Line` describes a line function `y: ax + b` using integer
-/// arithmetics.
-///
-/// The slope is in fact a decimal split into a 32 bit integer value,
-/// and a 32-bit decimal value.
-///
-/// The multiplication then becomes.
-/// `y = m * x >> 32 + b`
-#[derive(Debug, Clone, Copy, Default)]
-pub struct Line {
-    slope: u64,
-    intercept: u64,
-}
-
-/// Compute the line slope.
-///
-/// This function has the nice property of being
-/// invariant by translation.
-/// `
-///   compute_slope(y0, y1)
-/// = compute_slope(y0 + X % 2^64, y1 + X % 2^64)
-/// `
-fn compute_slope(y0: u64, y1: u64, num_vals: NonZeroU32) -> u64 {
-    let dy = y1.wrapping_sub(y0);
-    let sign = dy <= (1 << 63);
-    let abs_dy = if sign {
-        y1.wrapping_sub(y0)
-    } else {
-        y0.wrapping_sub(y1)
-    };
-    if abs_dy >= 1 << 32 {
-        // This is outside of realm we handle.
-        // Let's just bail.
-        return 0u64;
-    }
-
-    let abs_slope = (abs_dy << 32) / num_vals.get() as u64;
-    if sign {
-        abs_slope
-    } else {
-        // The complement does indeed create the
-        // opposite decreasing slope...
-        //
-        // Intuitively (without the bitshifts and % u64::MAX)
-        // ```
-        //    (x + shift)*(u64::MAX - abs_slope)
-        // -  (x * (u64::MAX - abs_slope))
-        // = - shift * abs_slope
-        // ```
-        u64::MAX - abs_slope
-    }
-}
-
-impl Line {
-    #[inline(always)]
-    pub fn eval(&self, x: u32) -> u64 {
-        let linear_part = ((x as u64).wrapping_mul(self.slope) >> 32) as i32 as u64;
-        self.intercept.wrapping_add(linear_part)
-    }
-
-    // Same as train, but the intercept is only estimated from provided sample positions
-    pub fn estimate(sample_positions_and_values: &[(u64, u64)]) -> Self {
-        let first_val = sample_positions_and_values[0].1;
-        let last_val = sample_positions_and_values[sample_positions_and_values.len() - 1].1;
-        let num_vals = sample_positions_and_values[sample_positions_and_values.len() - 1].0 + 1;
-        Self::train_from(
-            first_val,
-            last_val,
-            num_vals as u32,
-            sample_positions_and_values.iter().cloned(),
-        )
-    }
-
-    // Intercept is only computed from provided positions
-    fn train_from(
-        first_val: u64,
-        last_val: u64,
-        num_vals: u32,
-        positions_and_values: impl Iterator<Item = (u64, u64)>,
-    ) -> Self {
-        // TODO replace with let else
-        let idx_last_val = if let Some(idx_last_val) = NonZeroU32::new(num_vals - 1) {
-            idx_last_val
-        } else {
-            return Line::default();
-        };
-
-        let y0 = first_val;
-        let y1 = last_val;
-
-        // We first independently pick our slope.
-        let slope = compute_slope(y0, y1, idx_last_val);
-
-        // We picked our slope. Note that it does not have to be perfect.
-        // Now we need to compute the best intercept.
-        //
-        // Intuitively, the best intercept is such that line passes through one of the
-        // `(i, ys[])`.
-        //
-        // The best intercept therefore has the form
-        // `y[i] - line.eval(i)` (using wrapping arithmetics).
-        // In other words, the best intercept is one of the `y - Line::eval(ys[i])`
-        // and our task is just to pick the one that minimizes our error.
-        //
-        // Without sorting our values, this is a difficult problem.
-        // We however rely on the following trick...
-        //
-        // We only focus on the case where the interpolation is half decent.
-        // If the line interpolation is doing its job on a dataset suited for it,
-        // we can hope that the maximum error won't be larger than `u64::MAX / 2`.
-        //
-        // In other words, even without the intercept the values `y - Line::eval(ys[i])` will all be
-        // within an interval that takes less than half of the modulo space of `u64`.
-        //
-        // Our task is therefore to identify this interval.
-        // Here we simply translate all of our values by `y0 - 2^63` and pick the min.
-        let mut line = Line {
-            slope,
-            intercept: 0,
-        };
-        let heuristic_shift = y0.wrapping_sub(MID_POINT);
-        line.intercept = positions_and_values
-            .map(|(pos, y)| y.wrapping_sub(line.eval(pos as u32)))
-            .min_by_key(|&val| val.wrapping_sub(heuristic_shift))
-            .unwrap_or(0u64); //< Never happens.
-        line
-    }
-
-    /// Returns a line that attemps to approximate a function
-    /// f: i in 0..[ys.num_vals()) -> ys[i].
-    ///
-    /// - The approximation is always lower than the actual value.
-    /// Or more rigorously, formally `f(i).wrapping_sub(ys[i])` is small
-    /// for any i in [0..ys.len()).
-    /// - It computes without panicking for any value of it.
-    ///
-    /// This function is only invariable by translation if all of the
-    /// `ys` are packaged into half of the space. (See heuristic below)
-    pub fn train(ys: &dyn Column) -> Self {
-        let first_val = ys.iter().next().unwrap();
-        let last_val = ys.iter().nth(ys.num_vals() as usize - 1).unwrap();
-        Self::train_from(
-            first_val,
-            last_val,
-            ys.num_vals(),
-            ys.iter().enumerate().map(|(pos, val)| (pos as u64, val)),
-        )
-    }
-}
-
-impl BinarySerializable for Line {
-    fn serialize<W: io::Write>(&self, writer: &mut W) -> io::Result<()> {
-        VInt(self.slope).serialize(writer)?;
-        VInt(self.intercept).serialize(writer)?;
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let slope = VInt::deserialize(reader)?.0;
-        let intercept = VInt::deserialize(reader)?.0;
-        Ok(Line { slope, intercept })
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::VecColumn;
-
-    /// Test training a line and ensuring that the maximum difference between
-    /// the data points and the line is `expected`.
-    ///
-    /// This function operates translation over the data for better coverage.
-    #[track_caller]
-    fn test_line_interpol_with_translation(ys: &[u64], expected: Option<u64>) {
-        let mut translations = vec![0, 100, u64::MAX / 2, u64::MAX, u64::MAX - 1];
-        translations.extend_from_slice(ys);
-        for translation in translations {
-            let translated_ys: Vec<u64> = ys
-                .iter()
-                .copied()
-                .map(|y| y.wrapping_add(translation))
-                .collect();
-            let largest_err = test_eval_max_err(&translated_ys);
-            assert_eq!(largest_err, expected);
-        }
-    }
-
-    fn test_eval_max_err(ys: &[u64]) -> Option<u64> {
-        let line = Line::train(&VecColumn::from(&ys));
-        ys.iter()
-            .enumerate()
-            .map(|(x, y)| y.wrapping_sub(line.eval(x as u32)))
-            .max()
-    }
-
-    #[test]
-    fn test_train() {
-        test_line_interpol_with_translation(&[11, 11, 11, 12, 12, 13], Some(1));
-        test_line_interpol_with_translation(&[13, 12, 12, 11, 11, 11], Some(1));
-        test_line_interpol_with_translation(&[13, 13, 12, 11, 11, 11], Some(1));
-        test_line_interpol_with_translation(&[13, 13, 12, 11, 11, 11], Some(1));
-        test_line_interpol_with_translation(&[u64::MAX - 1, 0, 0, 1], Some(1));
-        test_line_interpol_with_translation(&[u64::MAX - 1, u64::MAX, 0, 1], Some(0));
-        test_line_interpol_with_translation(&[0, 1, 2, 3, 5], Some(0));
-        test_line_interpol_with_translation(&[1, 2, 3, 4], Some(0));
-
-        let data: Vec<u64> = (0..255).collect();
-        test_line_interpol_with_translation(&data, Some(0));
-        let data: Vec<u64> = (0..255).map(|el| el * 2).collect();
-        test_line_interpol_with_translation(&data, Some(0));
-    }
-}

fastfield_codecs/src/linear.rs

@@ -1,230 +0,0 @@
-use std::io::{self, Write};
-
-use common::{BinarySerializable, OwnedBytes};
-use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
-
-use crate::line::Line;
-use crate::serialize::NormalizedHeader;
-use crate::{Column, FastFieldCodec, FastFieldCodecType};
-
-/// Depending on the field type, a different
-/// fast field is required.
-#[derive(Clone)]
-pub struct LinearReader {
-    data: OwnedBytes,
-    linear_params: LinearParams,
-    header: NormalizedHeader,
-}
-
-impl Column for LinearReader {
-    #[inline]
-    fn get_val(&self, doc: u32) -> u64 {
-        let interpoled_val: u64 = self.linear_params.line.eval(doc);
-        let bitpacked_diff = self.linear_params.bit_unpacker.get(doc, &self.data);
-        interpoled_val.wrapping_add(bitpacked_diff)
-    }
-
-    #[inline(always)]
-    fn min_value(&self) -> u64 {
-        // The LinearReader assumes a normalized vector.
-        0u64
-    }
-
-    #[inline(always)]
-    fn max_value(&self) -> u64 {
-        self.header.max_value
-    }
-
-    #[inline]
-    fn num_vals(&self) -> u32 {
-        self.header.num_vals
-    }
-}
-
-/// Fastfield serializer, which tries to guess values by linear interpolation
-/// and stores the difference bitpacked.
-pub struct LinearCodec;
-
-#[derive(Debug, Clone)]
-struct LinearParams {
-    line: Line,
-    bit_unpacker: BitUnpacker,
-}
-
-impl BinarySerializable for LinearParams {
-    fn serialize<W: io::Write>(&self, writer: &mut W) -> io::Result<()> {
-        self.line.serialize(writer)?;
-        self.bit_unpacker.bit_width().serialize(writer)?;
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let line = Line::deserialize(reader)?;
-        let bit_width = u8::deserialize(reader)?;
-        Ok(Self {
-            line,
-            bit_unpacker: BitUnpacker::new(bit_width),
-        })
-    }
-}
-
-impl FastFieldCodec for LinearCodec {
-    const CODEC_TYPE: FastFieldCodecType = FastFieldCodecType::Linear;
-
-    type Reader = LinearReader;
-
-    /// Opens a fast field given a file.
-    fn open_from_bytes(mut data: OwnedBytes, header: NormalizedHeader) -> io::Result<Self::Reader> {
-        let linear_params = LinearParams::deserialize(&mut data)?;
-        Ok(LinearReader {
-            data,
-            linear_params,
-            header,
-        })
-    }
-
-    /// Creates a new fast field serializer.
-    fn serialize(column: &dyn Column, write: &mut impl Write) -> io::Result<()> {
-        assert_eq!(column.min_value(), 0);
-        let line = Line::train(column);
-
-        let max_offset_from_line = column
-            .iter()
-            .enumerate()
-            .map(|(pos, actual_value)| {
-                let calculated_value = line.eval(pos as u32);
-                actual_value.wrapping_sub(calculated_value)
-            })
-            .max()
-            .unwrap();
-
-        let num_bits = compute_num_bits(max_offset_from_line);
-        let linear_params = LinearParams {
-            line,
-            bit_unpacker: BitUnpacker::new(num_bits),
-        };
-        linear_params.serialize(write)?;
-
-        let mut bit_packer = BitPacker::new();
-        for (pos, actual_value) in column.iter().enumerate() {
-            let calculated_value = line.eval(pos as u32);
-            let offset = actual_value.wrapping_sub(calculated_value);
-            bit_packer.write(offset, num_bits, write)?;
-        }
-        bit_packer.close(write)?;
-
-        Ok(())
-    }
-
-    /// estimation for linear interpolation is hard because, you don't know
-    /// where the local maxima for the deviation of the calculated value are and
-    /// the offset to shift all values to >=0 is also unknown.
-    #[allow(clippy::question_mark)]
-    fn estimate(column: &dyn Column) -> Option<f32> {
-        if column.num_vals() < 3 {
-            return None; // disable compressor for this case
-        }
-
-        let limit_num_vals = column.num_vals().min(100_000);
-
-        let num_samples = 100;
-        let step_size = (limit_num_vals / num_samples).max(1); // 20 samples
-        let mut sample_positions_and_values: Vec<_> = Vec::new();
-        for (pos, val) in column.iter().enumerate().step_by(step_size as usize) {
-            sample_positions_and_values.push((pos as u64, val));
-        }
-
-        let line = Line::estimate(&sample_positions_and_values);
-
-        let estimated_bit_width = sample_positions_and_values
-            .into_iter()
-            .map(|(pos, actual_value)| {
-                let interpolated_val = line.eval(pos as u32);
-                actual_value.wrapping_sub(interpolated_val)
-            })
-            .map(|diff| ((diff as f32 * 1.5) * 2.0) as u64)
-            .map(compute_num_bits)
-            .max()
-            .unwrap_or(0);
-
-        // Extrapolate to whole column
-        let num_bits = (estimated_bit_width as u64 * column.num_vals() as u64) + 64;
-        let num_bits_uncompressed = 64 * column.num_vals();
-        Some(num_bits as f32 / num_bits_uncompressed as f32)
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use rand::RngCore;
-
-    use super::*;
-    use crate::tests::get_codec_test_datasets;
-
-    fn create_and_validate(data: &[u64], name: &str) -> Option<(f32, f32)> {
-        crate::tests::create_and_validate::<LinearCodec>(data, name)
-    }
-
-    #[test]
-    fn test_compression() {
-        let data = (10..=6_000_u64).collect::<Vec<_>>();
-        let (estimate, actual_compression) =
-            create_and_validate(&data, "simple monotonically large").unwrap();
-
-        assert_le!(actual_compression, 0.001);
-        assert_le!(estimate, 0.02);
-    }
-
-    #[test]
-    fn test_with_codec_datasets() {
-        let data_sets = get_codec_test_datasets();
-        for (mut data, name) in data_sets {
-            create_and_validate(&data, name);
-            data.reverse();
-            create_and_validate(&data, name);
-        }
-    }
-    #[test]
-    fn linear_interpol_fast_field_test_large_amplitude() {
-        let data = vec![
-            i64::MAX as u64 / 2,
-            i64::MAX as u64 / 3,
-            i64::MAX as u64 / 2,
-        ];
-
-        create_and_validate(&data, "large amplitude");
-    }
-
-    #[test]
-    fn overflow_error_test() {
-        let data = vec![1572656989877777, 1170935903116329, 720575940379279, 0];
-        create_and_validate(&data, "overflow test");
-    }
-
-    #[test]
-    fn linear_interpol_fast_concave_data() {
-        let data = vec![0, 1, 2, 5, 8, 10, 20, 50];
-        create_and_validate(&data, "concave data");
-    }
-    #[test]
-    fn linear_interpol_fast_convex_data() {
-        let data = vec![0, 40, 60, 70, 75, 77];
-        create_and_validate(&data, "convex data");
-    }
-    #[test]
-    fn linear_interpol_fast_field_test_simple() {
-        let data = (10..=20_u64).collect::<Vec<_>>();
-        create_and_validate(&data, "simple monotonically");
-    }
-
-    #[test]
-    fn linear_interpol_fast_field_rand() {
-        let mut rng = rand::thread_rng();
-        for _ in 0..50 {
-            let mut data = (0..10_000).map(|_| rng.next_u64()).collect::<Vec<_>>();
-            create_and_validate(&data, "random");
-            data.reverse();
-            create_and_validate(&data, "random");
-        }
-    }
-}

fastfield_codecs/src/main.rs

@@ -1,222 +0,0 @@
-#[macro_use]
-extern crate prettytable;
-use std::collections::HashSet;
-use std::env;
-use std::io::BufRead;
-use std::net::{IpAddr, Ipv6Addr};
-use std::str::FromStr;
-
-use common::OwnedBytes;
-use fastfield_codecs::{open_u128, serialize_u128, Column, FastFieldCodecType, VecColumn};
-use itertools::Itertools;
-use measure_time::print_time;
-use prettytable::{Cell, Row, Table};
-
-fn print_set_stats(ip_addrs: &[u128]) {
-    println!("NumIps\t{}", ip_addrs.len());
-    let ip_addr_set: HashSet<u128> = ip_addrs.iter().cloned().collect();
-    println!("NumUniqueIps\t{}", ip_addr_set.len());
-    let ratio_unique = ip_addr_set.len() as f64 / ip_addrs.len() as f64;
-    println!("RatioUniqueOverTotal\t{ratio_unique:.4}");
-
-    // histogram
-    let mut ip_addrs = ip_addrs.to_vec();
-    ip_addrs.sort();
-    let mut cnts: Vec<usize> = ip_addrs
-        .into_iter()
-        .dedup_with_count()
-        .map(|(cnt, _)| cnt)
-        .collect();
-    cnts.sort();
-
-    let top_256_cnt: usize = cnts.iter().rev().take(256).sum();
-    let top_128_cnt: usize = cnts.iter().rev().take(128).sum();
-    let top_64_cnt: usize = cnts.iter().rev().take(64).sum();
-    let top_8_cnt: usize = cnts.iter().rev().take(8).sum();
-    let total: usize = cnts.iter().sum();
-
-    println!("{}", total);
-    println!("{}", top_256_cnt);
-    println!("{}", top_128_cnt);
-    println!("Percentage Top8 {:02}", top_8_cnt as f32 / total as f32);
-    println!("Percentage Top64 {:02}", top_64_cnt as f32 / total as f32);
-    println!("Percentage Top128 {:02}", top_128_cnt as f32 / total as f32);
-    println!("Percentage Top256 {:02}", top_256_cnt as f32 / total as f32);
-
-    let mut cnts: Vec<(usize, usize)> = cnts.into_iter().dedup_with_count().collect();
-    cnts.sort_by(|a, b| {
-        if a.1 == b.1 {
-            a.0.cmp(&b.0)
-        } else {
-            b.1.cmp(&a.1)
-        }
-    });
-}
-
-fn ip_dataset() -> Vec<u128> {
-    let mut ip_addr_v4 = 0;
-
-    let stdin = std::io::stdin();
-    let ip_addrs: Vec<u128> = stdin
-        .lock()
-        .lines()
-        .flat_map(|line| {
-            let line = line.unwrap();
-            let line = line.trim();
-            let ip_addr = IpAddr::from_str(line.trim()).ok()?;
-            if ip_addr.is_ipv4() {
-                ip_addr_v4 += 1;
-            }
-            let ip_addr_v6: Ipv6Addr = match ip_addr {
-                IpAddr::V4(v4) => v4.to_ipv6_mapped(),
-                IpAddr::V6(v6) => v6,
-            };
-            Some(ip_addr_v6)
-        })
-        .map(|ip_v6| u128::from_be_bytes(ip_v6.octets()))
-        .collect();
-
-    println!("IpAddrsAny\t{}", ip_addrs.len());
-    println!("IpAddrsV4\t{}", ip_addr_v4);
-
-    ip_addrs
-}
-
-fn bench_ip() {
-    let dataset = ip_dataset();
-    print_set_stats(&dataset);
-
-    // Chunks
-    {
-        let mut data = vec![];
-        for dataset in dataset.chunks(500_000) {
-            serialize_u128(|| dataset.iter().cloned(), dataset.len() as u32, &mut data).unwrap();
-        }
-        let compression = data.len() as f64 / (dataset.len() * 16) as f64;
-        println!("Compression 50_000 chunks {:.4}", compression);
-        println!(
-            "Num Bits per elem {:.2}",
-            (data.len() * 8) as f32 / dataset.len() as f32
-        );
-    }
-
-    let mut data = vec![];
-    {
-        print_time!("creation");
-        serialize_u128(|| dataset.iter().cloned(), dataset.len() as u32, &mut data).unwrap();
-    }
-
-    let compression = data.len() as f64 / (dataset.len() * 16) as f64;
-    println!("Compression {:.2}", compression);
-    println!(
-        "Num Bits per elem {:.2}",
-        (data.len() * 8) as f32 / dataset.len() as f32
-    );
-
-    let decompressor = open_u128::<u128>(OwnedBytes::new(data)).unwrap();
-    // Sample some ranges
-    let mut doc_values = Vec::new();
-    for value in dataset.iter().take(1110).skip(1100).cloned() {
-        doc_values.clear();
-        print_time!("get range");
-        decompressor.get_docids_for_value_range(
-            value..=value,
-            0..decompressor.num_vals(),
-            &mut doc_values,
-        );
-        println!("{:?}", doc_values.len());
-    }
-}
-
-fn main() {
-    if env::args().nth(1).unwrap() == "bench_ip" {
-        bench_ip();
-        return;
-    }
-
-    let mut table = Table::new();
-
-    // Add a row per time
-    table.add_row(row!["", "Compression Ratio", "Compression Estimation"]);
-
-    for (data, data_set_name) in get_codec_test_data_sets() {
-        let results: Vec<(f32, f32, FastFieldCodecType)> = [
-            serialize_with_codec(&data, FastFieldCodecType::Bitpacked),
-            serialize_with_codec(&data, FastFieldCodecType::Linear),
-            serialize_with_codec(&data, FastFieldCodecType::BlockwiseLinear),
-        ]
-        .into_iter()
-        .flatten()
-        .collect();
-        let best_compression_ratio_codec = results
-            .iter()
-            .min_by(|&res1, &res2| res1.partial_cmp(res2).unwrap())
-            .cloned()
-            .unwrap();
-
-        table.add_row(Row::new(vec![Cell::new(data_set_name).style_spec("Bbb")]));
-        for (est, comp, codec_type) in results {
-            let est_cell = est.to_string();
-            let ratio_cell = comp.to_string();
-            let style = if comp == best_compression_ratio_codec.1 {
-                "Fb"
-            } else {
-                ""
-            };
-            table.add_row(Row::new(vec![
-                Cell::new(&format!("{codec_type:?}")).style_spec("bFg"),
-                Cell::new(&ratio_cell).style_spec(style),
-                Cell::new(&est_cell).style_spec(""),
-            ]));
-        }
-    }
-
-    table.printstd();
-}
-
-pub fn get_codec_test_data_sets() -> Vec<(Vec<u64>, &'static str)> {
-    let mut data_and_names = vec![];
-
-    let data = (1000..=200_000_u64).collect::<Vec<_>>();
-    data_and_names.push((data, "Autoincrement"));
-
-    let mut current_cumulative = 0;
-    let data = (1..=200_000_u64)
-        .map(|num| {
-            let num = (num as f32 + num as f32).log10() as u64;
-            current_cumulative += num;
-            current_cumulative
-        })
-        .collect::<Vec<_>>();
-    // let data = (1..=200000_u64).map(|num| num + num).collect::<Vec<_>>();
-    data_and_names.push((data, "Monotonically increasing concave"));
-
-    let mut current_cumulative = 0;
-    let data = (1..=200_000_u64)
-        .map(|num| {
-            let num = (200_000.0 - num as f32).log10() as u64;
-            current_cumulative += num;
-            current_cumulative
-        })
-        .collect::<Vec<_>>();
-    data_and_names.push((data, "Monotonically increasing convex"));
-
-    let data = (1000..=200_000_u64)
-        .map(|num| num + rand::random::<u8>() as u64)
-        .collect::<Vec<_>>();
-    data_and_names.push((data, "Almost monotonically increasing"));
-
-    data_and_names
-}
-
-pub fn serialize_with_codec(
-    data: &[u64],
-    codec_type: FastFieldCodecType,
-) -> Option<(f32, f32, FastFieldCodecType)> {
-    let col = VecColumn::from(data);
-    let estimation = fastfield_codecs::estimate(&col, codec_type)?;
-    let mut out = Vec::new();
-    fastfield_codecs::serialize(&col, &mut out, &[codec_type]).ok()?;
-    let actual_compression = out.len() as f32 / (col.num_vals() * 8) as f32;
-    Some((estimation, actual_compression, codec_type))
-}

fastfield_codecs/src/monotonic_mapping.rs

@@ -1,320 +0,0 @@
-use std::fmt;
-use std::marker::PhantomData;
-use std::ops::RangeInclusive;
-
-use fastdivide::DividerU64;
-
-use crate::MonotonicallyMappableToU128;
-
-/// Monotonic maps a value to u64 value space.
-/// Monotonic mapping enables `PartialOrd` on u64 space without conversion to original space.
-pub trait MonotonicallyMappableToU64:
-    'static + PartialOrd + Copy + Send + Sync + fmt::Debug
-{
-    /// Converts a value to u64.
-    ///
-    /// Internally all fast field values are encoded as u64.
-    fn to_u64(self) -> u64;
-
-    /// Converts a value from u64
-    ///
-    /// Internally all fast field values are encoded as u64.
-    /// **Note: To be used for converting encoded Term, Posting values.**
-    fn from_u64(val: u64) -> Self;
-}
-
-/// Values need to be strictly monotonic mapped to a `Internal` value (u64 or u128) that can be
-/// used in fast field codecs.
-///
-/// The monotonic mapping is required so that `PartialOrd` can be used on `Internal` without
-/// converting to `External`.
-///
-/// All strictly monotonic functions are invertible because they are guaranteed to have a one-to-one
-/// mapping from their range to their domain. The `inverse` method is required when opening a codec,
-/// so a value can be converted back to its original domain (e.g. ip address or f64) from its
-/// internal representation.
-pub trait StrictlyMonotonicFn<External: Copy, Internal: Copy> {
-    /// Strictly monotonically maps the value from External to Internal.
-    fn mapping(&self, inp: External) -> Internal;
-    /// Inverse of `mapping`. Maps the value from Internal to External.
-    fn inverse(&self, out: Internal) -> External;
-
-    /// Maps a user provded value from External to Internal.
-    /// It may be necessary to coerce the value if it is outside the value space.
-    /// In that case it tries to find the next greater value in the value space.
-    ///
-    /// Returns a bool to mark if a value was outside the value space and had to be coerced _up_.
-    /// With that information we can detect if two values in a range both map outside the same value
-    /// space.
-    ///
-    /// coerce_up means the next valid upper value in the value space will be chosen if the value
-    /// has to be coerced.
-    fn mapping_coerce(&self, inp: RangeInclusive<External>) -> RangeInclusive<Internal> {
-        self.mapping(*inp.start())..=self.mapping(*inp.end())
-    }
-    /// Inverse of `mapping_coerce`.
-    fn inverse_coerce(&self, out: RangeInclusive<Internal>) -> RangeInclusive<External> {
-        self.inverse(*out.start())..=self.inverse(*out.end())
-    }
-}
-
-/// Inverts a strictly monotonic mapping from `StrictlyMonotonicFn<A, B>` to
-/// `StrictlyMonotonicFn<B, A>`.
-///
-/// # Warning
-///
-/// This type comes with a footgun. A type being strictly monotonic does not impose that the inverse
-/// mapping is strictly monotonic over the entire space External. e.g. a -> a * 2. Use at your own
-/// risks.
-pub(crate) struct StrictlyMonotonicMappingInverter<T> {
-    orig_mapping: T,
-}
-impl<T> From<T> for StrictlyMonotonicMappingInverter<T> {
-    fn from(orig_mapping: T) -> Self {
-        Self { orig_mapping }
-    }
-}
-
-impl<From, To, T> StrictlyMonotonicFn<To, From> for StrictlyMonotonicMappingInverter<T>
-where
-    T: StrictlyMonotonicFn<From, To>,
-    From: Copy,
-    To: Copy,
-{
-    #[inline(always)]
-    fn mapping(&self, val: To) -> From {
-        self.orig_mapping.inverse(val)
-    }
-
-    #[inline(always)]
-    fn inverse(&self, val: From) -> To {
-        self.orig_mapping.mapping(val)
-    }
-
-    #[inline]
-    fn mapping_coerce(&self, inp: RangeInclusive<To>) -> RangeInclusive<From> {
-        self.orig_mapping.inverse_coerce(inp)
-    }
-    #[inline]
-    fn inverse_coerce(&self, out: RangeInclusive<From>) -> RangeInclusive<To> {
-        self.orig_mapping.mapping_coerce(out)
-    }
-}
-
-/// Applies the strictly monotonic mapping from `T` without any additional changes.
-pub(crate) struct StrictlyMonotonicMappingToInternal<T> {
-    _phantom: PhantomData<T>,
-}
-
-impl<T> StrictlyMonotonicMappingToInternal<T> {
-    pub(crate) fn new() -> StrictlyMonotonicMappingToInternal<T> {
-        Self {
-            _phantom: PhantomData,
-        }
-    }
-}
-
-impl<External: MonotonicallyMappableToU128, T: MonotonicallyMappableToU128>
-    StrictlyMonotonicFn<External, u128> for StrictlyMonotonicMappingToInternal<T>
-where T: MonotonicallyMappableToU128
-{
-    #[inline(always)]
-    fn mapping(&self, inp: External) -> u128 {
-        External::to_u128(inp)
-    }
-
-    #[inline(always)]
-    fn inverse(&self, out: u128) -> External {
-        External::from_u128(out)
-    }
-}
-
-impl<External: MonotonicallyMappableToU64, T: MonotonicallyMappableToU64>
-    StrictlyMonotonicFn<External, u64> for StrictlyMonotonicMappingToInternal<T>
-where T: MonotonicallyMappableToU64
-{
-    #[inline(always)]
-    fn mapping(&self, inp: External) -> u64 {
-        External::to_u64(inp)
-    }
-
-    #[inline(always)]
-    fn inverse(&self, out: u64) -> External {
-        External::from_u64(out)
-    }
-}
-
-/// 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)
-    }
-
-    #[inline]
-    #[allow(clippy::reversed_empty_ranges)]
-    fn mapping_coerce(&self, inp: RangeInclusive<External>) -> RangeInclusive<u64> {
-        let end = External::to_u64(*inp.end());
-        if end < self.min_value || inp.end() < inp.start() {
-            return 1..=0;
-        }
-        let map_coerce = |mut inp, coerce_up| {
-            let inp_lower_bound = self.inverse(0);
-            if inp < inp_lower_bound {
-                inp = inp_lower_bound;
-            }
-            let val = External::to_u64(inp);
-            let need_coercion = coerce_up && (val - self.min_value) % self.gcd != 0;
-            let mut mapped_val = self.mapping(inp);
-            if need_coercion {
-                mapped_val += 1;
-            }
-            mapped_val
-        };
-        let start = map_coerce(*inp.start(), true);
-        let end = map_coerce(*inp.end(), false);
-        start..=end
-    }
-}
-
-/// Strictly monotonic mapping with a base value.
-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]
-    #[allow(clippy::reversed_empty_ranges)]
-    fn mapping_coerce(&self, inp: RangeInclusive<External>) -> RangeInclusive<u64> {
-        if External::to_u64(*inp.end()) < self.min_value {
-            return 1..=0;
-        }
-        let start = self.mapping(External::to_u64(*inp.start()).max(self.min_value));
-        let end = self.mapping(External::to_u64(*inp.end()));
-        start..=end
-    }
-
-    #[inline(always)]
-    fn mapping(&self, val: External) -> u64 {
-        External::to_u64(val) - self.min_value
-    }
-
-    #[inline(always)]
-    fn inverse(&self, val: u64) -> External {
-        External::from_u64(self.min_value + val)
-    }
-}
-
-impl MonotonicallyMappableToU64 for u64 {
-    #[inline(always)]
-    fn to_u64(self) -> u64 {
-        self
-    }
-
-    #[inline(always)]
-    fn from_u64(val: u64) -> Self {
-        val
-    }
-}
-
-impl MonotonicallyMappableToU64 for i64 {
-    #[inline(always)]
-    fn to_u64(self) -> u64 {
-        common::i64_to_u64(self)
-    }
-
-    #[inline(always)]
-    fn from_u64(val: u64) -> Self {
-        common::u64_to_i64(val)
-    }
-}
-
-impl MonotonicallyMappableToU64 for bool {
-    #[inline(always)]
-    fn to_u64(self) -> u64 {
-        u64::from(self)
-    }
-
-    #[inline(always)]
-    fn from_u64(val: u64) -> Self {
-        val > 0
-    }
-}
-
-// TODO remove me.
-// Tantivy should refuse NaN values and work with NotNaN internally.
-impl MonotonicallyMappableToU64 for f64 {
-    #[inline(always)]
-    fn to_u64(self) -> u64 {
-        common::f64_to_u64(self)
-    }
-
-    #[inline(always)]
-    fn from_u64(val: u64) -> Self {
-        common::u64_to_f64(val)
-    }
-}
-
-#[cfg(test)]
-mod tests {
-
-    use super::*;
-
-    #[test]
-    fn strictly_monotonic_test() {
-        // identity mapping
-        test_round_trip(&StrictlyMonotonicMappingToInternal::<u64>::new(), 100u64);
-        // round trip to i64
-        test_round_trip(&StrictlyMonotonicMappingToInternal::<i64>::new(), 100u64);
-        // 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: Copy>(
-        mapping: &T,
-        test_val: K,
-    ) {
-        assert_eq!(mapping.inverse(mapping.mapping(test_val)), test_val);
-    }
-}

fastfield_codecs/src/monotonic_mapping_u128.rs

@@ -1,43 +0,0 @@
-use std::fmt;
-use std::net::Ipv6Addr;
-
-/// Montonic maps a value to u128 value space
-/// Monotonic mapping enables `PartialOrd` on u128 space without conversion to original space.
-pub trait MonotonicallyMappableToU128:
-    'static + PartialOrd + Copy + Send + Sync + fmt::Debug
-{
-    /// Converts a value to u128.
-    ///
-    /// Internally all fast field values are encoded as u64.
-    fn to_u128(self) -> u128;
-
-    /// Converts a value from u128
-    ///
-    /// Internally all fast field values are encoded as u64.
-    /// **Note: To be used for converting encoded Term, Posting values.**
-    fn from_u128(val: u128) -> Self;
-}
-
-impl MonotonicallyMappableToU128 for u128 {
-    fn to_u128(self) -> u128 {
-        self
-    }
-
-    fn from_u128(val: u128) -> Self {
-        val
-    }
-}
-
-impl MonotonicallyMappableToU128 for Ipv6Addr {
-    fn to_u128(self) -> u128 {
-        ip_to_u128(self)
-    }
-
-    fn from_u128(val: u128) -> Self {
-        Ipv6Addr::from(val.to_be_bytes())
-    }
-}
-
-fn ip_to_u128(ip_addr: Ipv6Addr) -> u128 {
-    u128::from_be_bytes(ip_addr.octets())
-}

fastfield_codecs/src/null_index/dense.rs

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

fastfield_codecs/src/null_index/mod.rs

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

fastfield_codecs/src/null_index/sparse.rs

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

fastfield_codecs/src/null_index_footer.rs

@@ -1,145 +0,0 @@
-use std::io::{self, Write};
-use std::ops::Range;
-
-use common::{BinarySerializable, CountingWriter, OwnedBytes, VInt};
-
-#[derive(Debug, Clone, Copy, Eq, PartialEq)]
-pub(crate) enum FastFieldCardinality {
-    Single = 1,
-    Multi = 2,
-}
-
-impl BinarySerializable for FastFieldCardinality {
-    fn serialize<W: Write>(&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 FastFieldCardinality {
-    pub(crate) fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    pub(crate) fn from_code(code: u8) -> Option<Self> {
-        match code {
-            1 => Some(Self::Single),
-            2 => Some(Self::Multi),
-            _ => None,
-        }
-    }
-}
-
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
-pub(crate) enum NullIndexCodec {
-    Full = 1,
-}
-
-impl BinarySerializable for NullIndexCodec {
-    fn serialize<W: Write>(&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 NullIndexCodec {
-    pub(crate) fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    pub(crate) fn from_code(code: u8) -> Option<Self> {
-        match code {
-            1 => Some(Self::Full),
-            _ => None,
-        }
-    }
-}
-
-#[derive(Debug, Clone, Eq, PartialEq)]
-pub(crate) struct NullIndexFooter {
-    pub(crate) cardinality: FastFieldCardinality,
-    pub(crate) null_index_codec: NullIndexCodec,
-    // Unused for NullIndexCodec::Full
-    pub(crate) null_index_byte_range: Range<u64>,
-}
-
-impl BinarySerializable for NullIndexFooter {
-    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
-        self.cardinality.serialize(writer)?;
-        self.null_index_codec.serialize(writer)?;
-        VInt(self.null_index_byte_range.start).serialize(writer)?;
-        VInt(self.null_index_byte_range.end - self.null_index_byte_range.start)
-            .serialize(writer)?;
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let cardinality = FastFieldCardinality::deserialize(reader)?;
-        let null_index_codec = NullIndexCodec::deserialize(reader)?;
-        let null_index_byte_range_start = VInt::deserialize(reader)?.0;
-        let null_index_byte_range_end = VInt::deserialize(reader)?.0 + null_index_byte_range_start;
-        Ok(Self {
-            cardinality,
-            null_index_codec,
-            null_index_byte_range: null_index_byte_range_start..null_index_byte_range_end,
-        })
-    }
-}
-
-pub(crate) fn append_null_index_footer(
-    output: &mut impl io::Write,
-    null_index_footer: NullIndexFooter,
-) -> io::Result<()> {
-    let mut counting_write = CountingWriter::wrap(output);
-    null_index_footer.serialize(&mut counting_write)?;
-    let footer_payload_len = counting_write.written_bytes();
-    BinarySerializable::serialize(&(footer_payload_len as u16), &mut counting_write)?;
-
-    Ok(())
-}
-
-pub(crate) fn read_null_index_footer(
-    data: OwnedBytes,
-) -> io::Result<(OwnedBytes, NullIndexFooter)> {
-    let (data, null_footer_length_bytes) = data.rsplit(2);
-
-    let footer_length = u16::deserialize(&mut null_footer_length_bytes.as_slice())?;
-    let (data, null_index_footer_bytes) = data.rsplit(footer_length as usize);
-    let null_index_footer = NullIndexFooter::deserialize(&mut null_index_footer_bytes.as_ref())?;
-
-    Ok((data, null_index_footer))
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn null_index_footer_deser_test() {
-        let null_index_footer = NullIndexFooter {
-            cardinality: FastFieldCardinality::Single,
-            null_index_codec: NullIndexCodec::Full,
-            null_index_byte_range: 100..120,
-        };
-
-        let mut out = vec![];
-        null_index_footer.serialize(&mut out).unwrap();
-
-        assert_eq!(
-            null_index_footer,
-            NullIndexFooter::deserialize(&mut &out[..]).unwrap()
-        );
-    }
-}

fastfield_codecs/src/serialize.rs

@@ -1,427 +0,0 @@
-// Copyright (C) 2022 Quickwit, Inc.
-//
-// Quickwit is offered under the AGPL v3.0 and as commercial software.
-// For commercial licensing, contact us at hello@quickwit.io.
-//
-// AGPL:
-// This program is free software: you can redistribute it and/or modify
-// it under the terms of the GNU Affero General Public License as
-// published by the Free Software Foundation, either version 3 of the
-// License, or (at your option) any later version.
-//
-// This program is distributed in the hope that it will be useful,
-// but WITHOUT ANY WARRANTY; without even the implied warranty of
-// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
-// GNU Affero General Public License for more details.
-//
-// You should have received a copy of the GNU Affero General Public License
-// along with this program. If not, see <http://www.gnu.org/licenses/>.
-
-use std::num::NonZeroU64;
-use std::sync::Arc;
-use std::{fmt, io};
-
-use common::{BinarySerializable, OwnedBytes, VInt};
-use log::warn;
-
-use crate::bitpacked::BitpackedCodec;
-use crate::blockwise_linear::BlockwiseLinearCodec;
-use crate::compact_space::CompactSpaceCompressor;
-use crate::format_version::append_format_version;
-use crate::linear::LinearCodec;
-use crate::monotonic_mapping::{
-    StrictlyMonotonicFn, StrictlyMonotonicMappingToInternal,
-    StrictlyMonotonicMappingToInternalGCDBaseval,
-};
-use crate::null_index_footer::{
-    append_null_index_footer, FastFieldCardinality, NullIndexCodec, NullIndexFooter,
-};
-use crate::{
-    monotonic_map_column, Column, FastFieldCodec, FastFieldCodecType, MonotonicallyMappableToU64,
-    U128FastFieldCodecType, VecColumn, ALL_CODEC_TYPES,
-};
-
-/// 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,
-}
-
-#[derive(Debug, Copy, Clone)]
-pub(crate) struct Header {
-    pub num_vals: u32,
-    pub min_value: u64,
-    pub max_value: u64,
-    pub gcd: Option<NonZeroU64>,
-    pub codec_type: FastFieldCodecType,
-}
-
-impl Header {
-    pub fn normalized(self) -> NormalizedHeader {
-        let gcd = self.gcd.map(|gcd| gcd.get()).unwrap_or(1);
-        let gcd_min_val_mapping =
-            StrictlyMonotonicMappingToInternalGCDBaseval::new(gcd, self.min_value);
-
-        let max_value = gcd_min_val_mapping.mapping(self.max_value);
-        NormalizedHeader {
-            num_vals: self.num_vals,
-            max_value,
-        }
-    }
-
-    pub fn normalize_column<C: Column>(&self, from_column: C) -> impl Column {
-        normalize_column(from_column, self.min_value, self.gcd)
-    }
-
-    pub fn compute_header(
-        column: impl Column<u64>,
-        codecs: &[FastFieldCodecType],
-    ) -> Option<Header> {
-        let num_vals = column.num_vals();
-        let min_value = column.min_value();
-        let max_value = column.max_value();
-        let gcd = crate::gcd::find_gcd(column.iter().map(|val| val - min_value))
-            .filter(|gcd| gcd.get() > 1u64);
-        let normalized_column = normalize_column(column, min_value, gcd);
-        let codec_type = detect_codec(normalized_column, codecs)?;
-        Some(Header {
-            num_vals,
-            min_value,
-            max_value,
-            gcd,
-            codec_type,
-        })
-    }
-}
-
-#[derive(Debug, Copy, Clone, PartialEq, Eq)]
-pub(crate) struct U128Header {
-    pub num_vals: u32,
-    pub codec_type: U128FastFieldCodecType,
-}
-
-impl BinarySerializable for U128Header {
-    fn serialize<W: io::Write>(&self, writer: &mut W) -> io::Result<()> {
-        VInt(self.num_vals as u64).serialize(writer)?;
-        self.codec_type.serialize(writer)?;
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let num_vals = VInt::deserialize(reader)?.0 as u32;
-        let codec_type = U128FastFieldCodecType::deserialize(reader)?;
-        Ok(U128Header {
-            num_vals,
-            codec_type,
-        })
-    }
-}
-
-pub fn normalize_column<C: Column>(
-    from_column: C,
-    min_value: u64,
-    gcd: Option<NonZeroU64>,
-) -> impl Column {
-    let gcd = gcd.map(|gcd| gcd.get()).unwrap_or(1);
-    let mapping = StrictlyMonotonicMappingToInternalGCDBaseval::new(gcd, min_value);
-    monotonic_map_column(from_column, mapping)
-}
-
-impl BinarySerializable for Header {
-    fn serialize<W: io::Write>(&self, writer: &mut W) -> io::Result<()> {
-        VInt(self.num_vals as u64).serialize(writer)?;
-        VInt(self.min_value).serialize(writer)?;
-        VInt(self.max_value - self.min_value).serialize(writer)?;
-        if let Some(gcd) = self.gcd {
-            VInt(gcd.get()).serialize(writer)?;
-        } else {
-            VInt(0u64).serialize(writer)?;
-        }
-        self.codec_type.serialize(writer)?;
-        Ok(())
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let num_vals = VInt::deserialize(reader)?.0 as u32;
-        let min_value = VInt::deserialize(reader)?.0;
-        let amplitude = VInt::deserialize(reader)?.0;
-        let max_value = min_value + amplitude;
-        let gcd_u64 = VInt::deserialize(reader)?.0;
-        let codec_type = FastFieldCodecType::deserialize(reader)?;
-        Ok(Header {
-            num_vals,
-            min_value,
-            max_value,
-            gcd: NonZeroU64::new(gcd_u64),
-            codec_type,
-        })
-    }
-}
-
-/// Return estimated compression for given codec in the value range [0.0..1.0], where 1.0 means no
-/// compression.
-pub fn estimate<T: MonotonicallyMappableToU64 + fmt::Debug>(
-    typed_column: impl Column<T>,
-    codec_type: FastFieldCodecType,
-) -> Option<f32> {
-    let column = monotonic_map_column(typed_column, StrictlyMonotonicMappingToInternal::<T>::new());
-    let min_value = column.min_value();
-    let gcd = crate::gcd::find_gcd(column.iter().map(|val| val - min_value))
-        .filter(|gcd| gcd.get() > 1u64);
-    let mapping = StrictlyMonotonicMappingToInternalGCDBaseval::new(
-        gcd.map(|gcd| gcd.get()).unwrap_or(1u64),
-        min_value,
-    );
-    let normalized_column = monotonic_map_column(&column, mapping);
-    match codec_type {
-        FastFieldCodecType::Bitpacked => BitpackedCodec::estimate(&normalized_column),
-        FastFieldCodecType::Linear => LinearCodec::estimate(&normalized_column),
-        FastFieldCodecType::BlockwiseLinear => BlockwiseLinearCodec::estimate(&normalized_column),
-    }
-}
-
-/// Serializes u128 values with the compact space codec.
-pub fn serialize_u128<F: Fn() -> I, I: Iterator<Item = u128>>(
-    iter_gen: F,
-    num_vals: u32,
-    output: &mut impl io::Write,
-) -> io::Result<()> {
-    serialize_u128_new(ValueIndexInfo::default(), iter_gen, num_vals, output)
-}
-
-#[allow(dead_code)]
-pub enum ValueIndexInfo<'a> {
-    MultiValue(Box<dyn MultiValueIndexInfo + 'a>),
-    SingleValue(Box<dyn SingleValueIndexInfo + 'a>),
-}
-
-// TODO Remove me
-impl Default for ValueIndexInfo<'static> {
-    fn default() -> Self {
-        struct Dummy {}
-        impl SingleValueIndexInfo for Dummy {
-            fn num_vals(&self) -> u32 {
-                todo!()
-            }
-            fn num_non_nulls(&self) -> u32 {
-                todo!()
-            }
-            fn iter(&self) -> Box<dyn Iterator<Item = u32>> {
-                todo!()
-            }
-        }
-
-        Self::SingleValue(Box::new(Dummy {}))
-    }
-}
-
-impl<'a> ValueIndexInfo<'a> {
-    fn get_cardinality(&self) -> FastFieldCardinality {
-        match self {
-            ValueIndexInfo::MultiValue(_) => FastFieldCardinality::Multi,
-            ValueIndexInfo::SingleValue(_) => FastFieldCardinality::Single,
-        }
-    }
-}
-
-pub trait MultiValueIndexInfo {
-    /// The number of docs in the column.
-    fn num_docs(&self) -> u32;
-    /// The number of values in the column.
-    fn num_vals(&self) -> u32;
-    /// Return the start index of the values for each doc
-    fn iter(&self) -> Box<dyn Iterator<Item = u32> + '_>;
-}
-
-pub trait SingleValueIndexInfo {
-    /// The number of values including nulls in the column.
-    fn num_vals(&self) -> u32;
-    /// The number of non-null values in the column.
-    fn num_non_nulls(&self) -> u32;
-    /// Return a iterator of the positions of docs with a value
-    fn iter(&self) -> Box<dyn Iterator<Item = u32> + '_>;
-}
-
-/// Serializes u128 values with the compact space codec.
-pub fn serialize_u128_new<F: Fn() -> I, I: Iterator<Item = u128>>(
-    value_index: ValueIndexInfo,
-    iter_gen: F,
-    num_vals: u32,
-    output: &mut impl io::Write,
-) -> io::Result<()> {
-    let header = U128Header {
-        num_vals,
-        codec_type: U128FastFieldCodecType::CompactSpace,
-    };
-    header.serialize(output)?;
-    let compressor = CompactSpaceCompressor::train_from(iter_gen(), num_vals);
-    compressor.compress_into(iter_gen(), output).unwrap();
-
-    let null_index_footer = NullIndexFooter {
-        cardinality: value_index.get_cardinality(),
-        null_index_codec: NullIndexCodec::Full,
-        null_index_byte_range: 0..0,
-    };
-    append_null_index_footer(output, null_index_footer)?;
-    append_format_version(output)?;
-
-    Ok(())
-}
-
-/// Serializes the column with the codec with the best estimate on the data.
-pub fn serialize<T: MonotonicallyMappableToU64 + fmt::Debug>(
-    typed_column: impl Column<T>,
-    output: &mut impl io::Write,
-    codecs: &[FastFieldCodecType],
-) -> io::Result<()> {
-    serialize_new(ValueIndexInfo::default(), typed_column, output, codecs)
-}
-
-/// Serializes the column with the codec with the best estimate on the data.
-pub fn serialize_new<T: MonotonicallyMappableToU64 + fmt::Debug>(
-    value_index: ValueIndexInfo,
-    typed_column: impl Column<T>,
-    output: &mut impl io::Write,
-    codecs: &[FastFieldCodecType],
-) -> io::Result<()> {
-    let column = monotonic_map_column(typed_column, StrictlyMonotonicMappingToInternal::<T>::new());
-    let header = Header::compute_header(&column, codecs).ok_or_else(|| {
-        io::Error::new(
-            io::ErrorKind::InvalidInput,
-            format!(
-                "Data cannot be serialized with this list of codec. {:?}",
-                codecs
-            ),
-        )
-    })?;
-    header.serialize(output)?;
-    let normalized_column = header.normalize_column(column);
-    assert_eq!(normalized_column.min_value(), 0u64);
-    serialize_given_codec(normalized_column, header.codec_type, output)?;
-
-    let null_index_footer = NullIndexFooter {
-        cardinality: value_index.get_cardinality(),
-        null_index_codec: NullIndexCodec::Full,
-        null_index_byte_range: 0..0,
-    };
-    append_null_index_footer(output, null_index_footer)?;
-    append_format_version(output)?;
-
-    Ok(())
-}
-
-fn detect_codec(
-    column: impl Column<u64>,
-    codecs: &[FastFieldCodecType],
-) -> Option<FastFieldCodecType> {
-    let mut estimations = Vec::new();
-    for &codec in codecs {
-        let estimation_opt = match codec {
-            FastFieldCodecType::Bitpacked => BitpackedCodec::estimate(&column),
-            FastFieldCodecType::Linear => LinearCodec::estimate(&column),
-            FastFieldCodecType::BlockwiseLinear => BlockwiseLinearCodec::estimate(&column),
-        };
-        if let Some(estimation) = estimation_opt {
-            estimations.push((estimation, codec));
-        }
-    }
-    if let Some(broken_estimation) = estimations.iter().find(|estimation| estimation.0.is_nan()) {
-        warn!(
-            "broken estimation for fast field codec {:?}",
-            broken_estimation.1
-        );
-    }
-    // removing nan values for codecs with broken calculations, and max values which disables
-    // codecs
-    estimations.retain(|estimation| !estimation.0.is_nan() && estimation.0 != f32::MAX);
-    estimations.sort_by(|(score_left, _), (score_right, _)| score_left.total_cmp(score_right));
-    Some(estimations.first()?.1)
-}
-
-fn serialize_given_codec(
-    column: impl Column<u64>,
-    codec_type: FastFieldCodecType,
-    output: &mut impl io::Write,
-) -> io::Result<()> {
-    match codec_type {
-        FastFieldCodecType::Bitpacked => {
-            BitpackedCodec::serialize(&column, output)?;
-        }
-        FastFieldCodecType::Linear => {
-            LinearCodec::serialize(&column, output)?;
-        }
-        FastFieldCodecType::BlockwiseLinear => {
-            BlockwiseLinearCodec::serialize(&column, output)?;
-        }
-    }
-    output.flush()?;
-    Ok(())
-}
-
-/// Helper function to serialize a column (autodetect from all codecs) and then open it
-pub fn serialize_and_load<T: MonotonicallyMappableToU64 + Ord + Default + fmt::Debug>(
-    column: &[T],
-) -> Arc<dyn Column<T>> {
-    let mut buffer = Vec::new();
-    super::serialize(VecColumn::from(&column), &mut buffer, &ALL_CODEC_TYPES).unwrap();
-    super::open(OwnedBytes::new(buffer)).unwrap()
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_serialize_deserialize_u128_header() {
-        let original = U128Header {
-            num_vals: 11,
-            codec_type: U128FastFieldCodecType::CompactSpace,
-        };
-        let mut out = Vec::new();
-        original.serialize(&mut out).unwrap();
-        let restored = U128Header::deserialize(&mut &out[..]).unwrap();
-        assert_eq!(restored, original);
-    }
-
-    #[test]
-    fn test_serialize_deserialize() {
-        let original = [1u64, 5u64, 10u64];
-        let restored: Vec<u64> = serialize_and_load(&original[..]).iter().collect();
-        assert_eq!(&restored, &original[..]);
-    }
-
-    #[test]
-    fn test_fastfield_bool_size_bitwidth_1() {
-        let mut buffer = Vec::new();
-        let col = VecColumn::from(&[false, true][..]);
-        serialize(col, &mut buffer, &ALL_CODEC_TYPES).unwrap();
-        // 5 bytes of header, 1 byte of value
-        assert_eq!(buffer.len(), 3 + 5 + 1 + 4 + 2);
-    }
-
-    #[test]
-    fn test_fastfield_bool_bit_size_bitwidth_0() {
-        let mut buffer = Vec::new();
-        let col = VecColumn::from(&[true][..]);
-        serialize(col, &mut buffer, &ALL_CODEC_TYPES).unwrap();
-        // 5 bytes of header, 0 bytes of value
-        assert_eq!(buffer.len(), 3 + 5 + 4 + 2);
-    }
-
-    #[test]
-    fn test_fastfield_gcd() {
-        let mut buffer = Vec::new();
-        let vals: Vec<u64> = (0..80).map(|val| (val % 7) * 1_000u64).collect();
-        let col = VecColumn::from(&vals[..]);
-        serialize(col, &mut buffer, &[FastFieldCodecType::Bitpacked]).unwrap();
-        // Values are stored over 3 bits.
-        assert_eq!(buffer.len(), 3 + 7 + (3 * 80 / 8) + 4 + 2);
-    }
-}

src/aggregation/agg_req_with_accessor.rs

@@ -15,7 +15,7 @@ use super::metric::{
 use super::segment_agg_result::BucketCount;
 use super::VecWithNames;
 use crate::fastfield::{type_and_cardinality, MultiValuedFastFieldReader};
-use crate::schema::{Cardinality, Type};
+use crate::schema::Type;
 use crate::{InvertedIndexReader, SegmentReader, TantivyError};
 
 #[derive(Clone, Default)]

src/aggregation/metric/mod.rs

@@ -43,13 +43,13 @@ mod tests {
     use crate::aggregation::agg_result::AggregationResults;
     use crate::aggregation::AggregationCollector;
     use crate::query::AllQuery;
-    use crate::schema::{Cardinality, NumericOptions, Schema};
+    use crate::schema::{NumericOptions, Schema};
     use crate::Index;
 
     #[test]
     fn test_metric_aggregations() {
         let mut schema_builder = Schema::builder();
-        let field_options = NumericOptions::default().set_fast(Cardinality::SingleValue);
+        let field_options = NumericOptions::default().set_fast();
         let field = schema_builder.add_f64_field("price", field_options);
         let index = Index::create_in_ram(schema_builder.build());
         let mut index_writer = index.writer_for_tests().unwrap();

src/aggregation/mod.rs

@@ -433,13 +433,13 @@ mod tests {
         let text_field_id = schema_builder.add_text_field("text_id", text_fieldtype);
         let string_field_id = schema_builder.add_text_field("string_id", STRING | FAST);
         let score_fieldtype =
-            crate::schema::NumericOptions::default().set_fast(Cardinality::SingleValue);
+            crate::schema::NumericOptions::default().set_fast();
         let score_field = schema_builder.add_u64_field("score", score_fieldtype.clone());
         let score_field_f64 = schema_builder.add_f64_field("score_f64", score_fieldtype.clone());
         let score_field_i64 = schema_builder.add_i64_field("score_i64", score_fieldtype);
         let fraction_field = schema_builder.add_f64_field(
             "fraction_f64",
-            crate::schema::NumericOptions::default().set_fast(Cardinality::SingleValue),
+            crate::schema::NumericOptions::default().set_fast(),
         );
         let index = Index::create_in_ram(schema_builder.build());
         {
@@ -657,12 +657,12 @@ mod tests {
         let date_field = schema_builder.add_date_field("date", FAST);
         schema_builder.add_text_field("dummy_text", STRING);
         let score_fieldtype =
-            crate::schema::NumericOptions::default().set_fast(Cardinality::SingleValue);
+            crate::schema::NumericOptions::default().set_fast();
         let score_field = schema_builder.add_u64_field("score", score_fieldtype.clone());
         let score_field_f64 = schema_builder.add_f64_field("score_f64", score_fieldtype.clone());
 
         let multivalue =
-            crate::schema::NumericOptions::default().set_fast(Cardinality::MultiValues);
+            crate::schema::NumericOptions::default().set_fast();
         let scores_field_i64 = schema_builder.add_i64_field("scores_i64", multivalue);
 
         let score_field_i64 = schema_builder.add_i64_field("score_i64", score_fieldtype);
@@ -1190,7 +1190,7 @@ mod tests {
             let text_field_few_terms =
                 schema_builder.add_text_field("text_few_terms", STRING | FAST);
             let score_fieldtype =
-                crate::schema::NumericOptions::default().set_fast(Cardinality::SingleValue);
+                crate::schema::NumericOptions::default().set_fast();
             let score_field = schema_builder.add_u64_field("score", score_fieldtype.clone());
             let score_field_f64 =
                 schema_builder.add_f64_field("score_f64", score_fieldtype.clone());

src/collector/facet_collector.rs

@@ -1,12 +1,11 @@
 use std::cmp::Ordering;
 use std::collections::{btree_map, BTreeMap, BTreeSet, BinaryHeap};
-use std::iter::Peekable;
 use std::ops::Bound;
-use std::{u64, usize};
+use std::{io, u64, usize};
 
 use crate::collector::{Collector, SegmentCollector};
 use crate::fastfield::FacetReader;
-use crate::schema::{Facet, Field};
+use crate::schema::Facet;
 use crate::{DocId, Score, SegmentOrdinal, SegmentReader};
 
 struct Hit<'a> {
@@ -119,7 +118,7 @@ fn facet_depth(facet_bytes: &[u8]) -> usize {
 ///     let searcher = reader.searcher();
 ///
 ///     {
-///         let mut facet_collector = FacetCollector::for_field(facet);
+///         let mut facet_collector = FacetCollector::for_field("facet");
 ///         facet_collector.add_facet("/lang");
 ///         facet_collector.add_facet("/category");
 ///         let facet_counts = searcher.search(&AllQuery, &facet_collector)?;
@@ -135,7 +134,7 @@ fn facet_depth(facet_bytes: &[u8]) -> usize {
 ///     }
 ///
 ///     {
-///         let mut facet_collector = FacetCollector::for_field(facet);
+///         let mut facet_collector = FacetCollector::for_field("facet");
 ///         facet_collector.add_facet("/category/fiction");
 ///         let facet_counts = searcher.search(&AllQuery, &facet_collector)?;
 ///
@@ -167,47 +166,18 @@ fn facet_depth(facet_bytes: &[u8]) -> usize {
 /// # assert!(example().is_ok());
 /// ```
 pub struct FacetCollector {
-    field: Field,
+    field_name: String,
     facets: BTreeSet<Facet>,
 }
 
 pub struct FacetSegmentCollector {
     reader: FacetReader,
-    facet_ords_buf: Vec<u64>,
-    // facet_ord -> collapse facet_id
-    collapse_mapping: Vec<usize>,
     // collapse facet_id -> count
     counts: Vec<u64>,
-    // collapse facet_id -> facet_ord
-    collapse_facet_ords: Vec<u64>,
-}
-
-enum SkipResult {
-    Found,
-    NotFound,
-}
-
-fn skip<'a, I: Iterator<Item = &'a Facet>>(
-    target: &[u8],
-    collapse_it: &mut Peekable<I>,
-) -> SkipResult {
-    loop {
-        match collapse_it.peek() {
-            Some(facet_bytes) => match facet_bytes.encoded_str().as_bytes().cmp(target) {
-                Ordering::Less => {}
-                Ordering::Greater => {
-                    return SkipResult::NotFound;
-                }
-                Ordering::Equal => {
-                    return SkipResult::Found;
-                }
-            },
-            None => {
-                return SkipResult::NotFound;
-            }
-        }
-        collapse_it.next();
-    }
+    // facet_ord -> compressed collapse facet_id
+    compressed_collapse_mapping: Vec<usize>,
+    // compressed collapse facet_id -> facet_ord
+    unique_facet_ords: Vec<(u64, usize)>,
 }
 
 impl FacetCollector {
@@ -216,9 +186,9 @@ impl FacetCollector {
     ///
     /// This function does not check whether the field
     /// is of the proper type.
-    pub fn for_field(field: Field) -> FacetCollector {
+    pub fn for_field(field_name: impl ToString) -> FacetCollector {
         FacetCollector {
-            field,
+            field_name: field_name.to_string(),
             facets: BTreeSet::default(),
         }
     }
@@ -249,6 +219,29 @@ impl FacetCollector {
     }
 }
 
+fn compress_mapping(mapping: &[(u64, usize)]) -> (Vec<usize>, Vec<(u64, usize)>) {
+    // facet_ord -> collapse facet_id
+    let mut compressed_collapse_mapping: Vec<usize> = Vec::with_capacity(mapping.len());
+    // collapse facet_id -> facet_ord
+    let mut unique_facet_ords: Vec<(u64, usize)> = Vec::new();
+    if mapping.is_empty() {
+        return (Vec::new(), Vec::new());
+    }
+    compressed_collapse_mapping.push(0);
+    unique_facet_ords.push(mapping[0]);
+    let mut last_facet_ord = mapping[0];
+    let mut last_facet_id = 0;
+    for &facet_ord in &mapping[1..] {
+        if facet_ord != last_facet_ord {
+            last_facet_id += 1;
+            last_facet_ord = facet_ord;
+            unique_facet_ords.push(facet_ord);
+        }
+        compressed_collapse_mapping.push(last_facet_id);
+    }
+    (compressed_collapse_mapping, unique_facet_ords)
+}
+
 impl Collector for FacetCollector {
     type Fruit = FacetCounts;
 
@@ -259,59 +252,17 @@ impl Collector for FacetCollector {
         _: SegmentOrdinal,
         reader: &SegmentReader,
     ) -> crate::Result<FacetSegmentCollector> {
-        let facet_reader = reader.facet_reader(self.field)?;
-
-        let mut collapse_mapping = Vec::new();
-        let mut counts = Vec::new();
-        let mut collapse_facet_ords = Vec::new();
-
-        let mut collapse_facet_it = self.facets.iter().peekable();
-        collapse_facet_ords.push(0);
-        {
-            let mut facet_streamer = facet_reader.facet_dict().range().into_stream()?;
-            if facet_streamer.advance() {
-                'outer: loop {
-                    // at the beginning of this loop, facet_streamer
-                    // is positioned on a term that has not been processed yet.
-                    let skip_result = skip(facet_streamer.key(), &mut collapse_facet_it);
-                    match skip_result {
-                        SkipResult::Found => {
-                            // we reach a facet we decided to collapse.
-                            let collapse_depth = facet_depth(facet_streamer.key());
-                            let mut collapsed_id = 0;
-                            collapse_mapping.push(0);
-                            while facet_streamer.advance() {
-                                let depth = facet_depth(facet_streamer.key());
-                                if depth <= collapse_depth {
-                                    continue 'outer;
-                                }
-                                if depth == collapse_depth + 1 {
-                                    collapsed_id = collapse_facet_ords.len();
-                                    collapse_facet_ords.push(facet_streamer.term_ord());
-                                }
-                                collapse_mapping.push(collapsed_id);
-                            }
-                            break;
-                        }
-                        SkipResult::NotFound => {
-                            collapse_mapping.push(0);
-                            if !facet_streamer.advance() {
-                                break;
-                            }
-                        }
-                    }
-                }
-            }
-        }
-
-        counts.resize(collapse_facet_ords.len(), 0);
-
+        let facet_reader = reader.facet_reader(&self.field_name)?;
+        let facet_dict = facet_reader.facet_dict();
+        let collapse_mapping: Vec<(u64, usize)> =
+            compute_collapse_mapping(facet_dict, &self.facets)?;
+        let (compressed_collapse_mapping, unique_facet_ords) = compress_mapping(&collapse_mapping);
+        let counts = vec![0u64; unique_facet_ords.len()];
         Ok(FacetSegmentCollector {
             reader: facet_reader,
-            facet_ords_buf: Vec::with_capacity(255),
-            collapse_mapping,
+            compressed_collapse_mapping,
             counts,
-            collapse_facet_ords,
+            unique_facet_ords,
         })
     }
 
@@ -330,14 +281,78 @@ impl Collector for FacetCollector {
     }
 }
 
+fn is_child_facet(parent_facet: &[u8], possible_child_facet: &[u8]) -> bool {
+    if !possible_child_facet.starts_with(parent_facet) {
+        return false;
+    }
+    possible_child_facet.get(parent_facet.len()).copied() == Some(0u8)
+}
+
+fn compute_collapse_mapping_one(
+    facet_terms: &mut columnar::Streamer,
+    facet_bytes: &[u8],
+    collapsed: &mut [(u64, usize)],
+) -> io::Result<bool> {
+    let mut facet_child: Vec<u8> = Vec::new();
+    let mut term_ord = 0;
+    let offset = facet_bytes.len() + 1;
+    let depth = facet_depth(facet_bytes);
+    loop {
+        match facet_terms.key().cmp(facet_bytes) {
+            Ordering::Less | Ordering::Equal => {}
+            Ordering::Greater => {
+                if !is_child_facet(facet_bytes, facet_terms.key()) {
+                    return Ok(true);
+                }
+                let suffix = &facet_terms.key()[offset..];
+                if facet_child.is_empty() || !is_child_facet(&facet_child, suffix) {
+                    facet_child.clear();
+                    term_ord = facet_terms.term_ord();
+                    let end = suffix
+                        .iter()
+                        .position(|b| *b == 0u8)
+                        .unwrap_or(suffix.len());
+                    facet_child.extend(&suffix[..end]);
+                }
+                collapsed[facet_terms.term_ord() as usize] = (term_ord, depth);
+            }
+        }
+        if !facet_terms.advance() {
+            return Ok(false);
+        }
+    }
+}
+
+fn compute_collapse_mapping(
+    facet_dict: &columnar::Dictionary,
+    facets: &BTreeSet<Facet>,
+) -> io::Result<Vec<(u64, usize)>> {
+    let mut collapsed = vec![(u64::MAX, 0); facet_dict.num_terms()];
+    if facets.is_empty() {
+        return Ok(collapsed);
+    }
+    let mut facet_terms: columnar::Streamer = facet_dict.range().into_stream()?;
+    if !facet_terms.advance() {
+        return Ok(collapsed);
+    }
+    let mut facet_bytes = Vec::new();
+    for facet in facets {
+        facet_bytes.clear();
+        facet_bytes.extend(facet.encoded_str().as_bytes());
+        if !compute_collapse_mapping_one(&mut facet_terms, &facet_bytes, &mut collapsed[..])? {
+            break;
+        }
+    }
+    Ok(collapsed)
+}
+
 impl SegmentCollector for FacetSegmentCollector {
     type Fruit = FacetCounts;
 
     fn collect(&mut self, doc: DocId, _: Score) {
-        self.reader.facet_ords(doc, &mut self.facet_ords_buf);
         let mut previous_collapsed_ord: usize = usize::MAX;
-        for &facet_ord in &self.facet_ords_buf {
-            let collapsed_ord = self.collapse_mapping[facet_ord as usize];
+        for facet_ord in self.reader.facet_ords(doc) {
+            let collapsed_ord = self.compressed_collapse_mapping[facet_ord as usize];
             self.counts[collapsed_ord] += u64::from(collapsed_ord != previous_collapsed_ord);
             previous_collapsed_ord = collapsed_ord;
         }
@@ -355,9 +370,17 @@ impl SegmentCollector for FacetSegmentCollector {
                 continue;
             }
             let mut facet = vec![];
-            let facet_ord = self.collapse_facet_ords[collapsed_facet_ord];
+            let (facet_ord, facet_depth) = self.unique_facet_ords[collapsed_facet_ord];
             // TODO handle errors.
             if facet_dict.ord_to_term(facet_ord, &mut facet).is_ok() {
+                if let Some((end_collapsed_facet, _)) = facet
+                    .iter()
+                    .enumerate()
+                    .filter(|(_pos, &b)| b == 0u8)
+                    .nth(facet_depth)
+                {
+                    facet.truncate(end_collapsed_facet);
+                }
                 if let Ok(facet) = Facet::from_encoded(facet) {
                     facet_counts.insert(facet, count);
                 }
@@ -441,27 +464,114 @@ impl FacetCounts {
 
 #[cfg(test)]
 mod tests {
+    use std::collections::BTreeSet;
     use std::iter;
 
+    use columnar::Dictionary;
     use rand::distributions::Uniform;
     use rand::prelude::SliceRandom;
     use rand::{thread_rng, Rng};
 
     use super::{FacetCollector, FacetCounts};
+    use crate::collector::facet_collector::compress_mapping;
     use crate::collector::Count;
     use crate::core::Index;
     use crate::query::{AllQuery, QueryParser, TermQuery};
-    use crate::schema::{Document, Facet, FacetOptions, Field, IndexRecordOption, Schema};
+    use crate::schema::{Document, Facet, FacetOptions, IndexRecordOption, Schema};
     use crate::Term;
 
+    fn test_collapse_mapping_aux(
+        facet_terms: &[&str],
+        facet_params: &[&str],
+        expected_collapsed_mapping: &[(u64, usize)],
+    ) {
+        let mut facets: Vec<Facet> = facet_terms.iter().map(Facet::from).collect();
+        facets.sort();
+        let facet_terms: Vec<&str> = facets.iter().map(|facet| facet.encoded_str()).collect();
+        let dictionary = Dictionary::build_for_tests(&facet_terms);
+        let facet_params: BTreeSet<Facet> = facet_params.iter().map(Facet::from).collect();
+        let collapse_mapping = super::compute_collapse_mapping(&dictionary, &facet_params).unwrap();
+        assert_eq!(&collapse_mapping[..], expected_collapsed_mapping);
+    }
+
     #[test]
-    fn test_facet_collector_drilldown() -> crate::Result<()> {
+    fn test_collapse_simple() {
+        test_collapse_mapping_aux(&["/facet/a", "/facet/b"], &["/facet"], &[(0, 1), (1, 1)]);
+        test_collapse_mapping_aux(
+            &["/facet/a", "/facet/a2", "/facet/b"],
+            &["/facet"],
+            &[(0, 1), (1, 1), (2, 1)],
+        );
+        test_collapse_mapping_aux(&["/facet/a", "/facet/a/2"], &["/facet"], &[(0, 1), (0, 1)]);
+        test_collapse_mapping_aux(
+            &["/facet/a", "/facet/a/2", "/facet/b"],
+            &["/facet"],
+            &[(0, 1), (0, 1), (2, 1)],
+        );
+    }
+
+    fn test_compress_mapping_aux(
+        collapsed_mapping: &[(u64, usize)],
+        expected_compressed_collapsed_mapping: &[usize],
+        expected_unique_facet_ords: &[(u64, usize)],
+    ) {
+        let (compressed_collapsed_mapping, unique_facet_ords) =
+            compress_mapping(&collapsed_mapping);
+        assert_eq!(
+            compressed_collapsed_mapping,
+            expected_compressed_collapsed_mapping
+        );
+        assert_eq!(unique_facet_ords, expected_unique_facet_ords);
+    }
+
+    #[test]
+    fn test_compress_mapping() {
+        test_compress_mapping_aux(&[], &[], &[]);
+        test_compress_mapping_aux(&[(1, 2)], &[0], &[(1, 2)]);
+        test_compress_mapping_aux(&[(1, 2), (1, 2)], &[0, 0], &[(1, 2)]);
+        test_compress_mapping_aux(
+            &[(1, 2), (5, 2), (5, 2), (6, 3), (8, 3)],
+            &[0, 1, 1, 2, 3],
+            &[(1, 2), (5, 2), (6, 3), (8, 3)],
+        );
+    }
+
+    #[test]
+    fn test_facet_collector_simple() {
+        let mut schema_builder = Schema::builder();
+        let facet_field = schema_builder.add_facet_field("facet", FacetOptions::default());
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_for_tests().unwrap();
+        index_writer
+            .add_document(doc!(facet_field=>Facet::from("/facet/a")))
+            .unwrap();
+        index_writer
+            .add_document(doc!(facet_field=>Facet::from("/facet/b")))
+            .unwrap();
+        index_writer
+            .add_document(doc!(facet_field=>Facet::from("/facet/b")))
+            .unwrap();
+        index_writer
+            .add_document(doc!(facet_field=>Facet::from("/facet/c")))
+            .unwrap();
+        index_writer.commit().unwrap();
+        let searcher = index.reader().unwrap().searcher();
+        let mut facet_collector = FacetCollector::for_field("facet");
+        facet_collector.add_facet("/facet");
+        let counts: FacetCounts = searcher.search(&AllQuery, &facet_collector).unwrap();
+        let facets: Vec<(&Facet, u64)> = counts.top_k("/facet", 1);
+        assert_eq!(facets, vec![(&Facet::from("/facet/b"), 2)]);
+    }
+
+    #[test]
+    fn test_facet_collector_drilldown() {
         let mut schema_builder = Schema::builder();
         let facet_field = schema_builder.add_facet_field("facet", FacetOptions::default());
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
 
-        let mut index_writer = index.writer_for_tests()?;
+        let mut index_writer = index.writer_for_tests().unwrap();
         let num_facets: usize = 3 * 4 * 5;
         let facets: Vec<Facet> = (0..num_facets)
             .map(|mut n| {
@@ -476,14 +586,14 @@ mod tests {
         for i in 0..num_facets * 10 {
             let mut doc = Document::new();
             doc.add_facet(facet_field, facets[i % num_facets].clone());
-            index_writer.add_document(doc)?;
+            index_writer.add_document(doc).unwrap();
         }
-        index_writer.commit()?;
-        let reader = index.reader()?;
+        index_writer.commit().unwrap();
+        let reader = index.reader().unwrap();
         let searcher = reader.searcher();
-        let mut facet_collector = FacetCollector::for_field(facet_field);
+        let mut facet_collector = FacetCollector::for_field("facet");
         facet_collector.add_facet(Facet::from("/top1"));
-        let counts = searcher.search(&AllQuery, &facet_collector)?;
+        let counts = searcher.search(&AllQuery, &facet_collector).unwrap();
 
         {
             let facets: Vec<(String, u64)> = counts
@@ -503,7 +613,6 @@ mod tests {
                 .collect::<Vec<_>>()
             );
         }
-        Ok(())
     }
 
     #[test]
@@ -511,7 +620,7 @@ mod tests {
         expected = "Tried to add a facet which is a descendant of an already added facet."
     )]
     fn test_misused_facet_collector() {
-        let mut facet_collector = FacetCollector::for_field(Field::from_field_id(0));
+        let mut facet_collector = FacetCollector::for_field("facet");
         facet_collector.add_facet(Facet::from("/country"));
         facet_collector.add_facet(Facet::from("/country/europe"));
     }
@@ -533,7 +642,7 @@ mod tests {
         let reader = index.reader()?;
         let searcher = reader.searcher();
         assert_eq!(searcher.num_docs(), 1);
-        let mut facet_collector = FacetCollector::for_field(facet_field);
+        let mut facet_collector = FacetCollector::for_field("facets");
         facet_collector.add_facet("/subjects");
         let counts = searcher.search(&AllQuery, &facet_collector)?;
         let facets: Vec<(&Facet, u64)> = counts.get("/subjects").collect();
@@ -593,7 +702,7 @@ mod tests {
 
     #[test]
     fn test_non_used_facet_collector() {
-        let mut facet_collector = FacetCollector::for_field(Field::from_field_id(0));
+        let mut facet_collector = FacetCollector::for_field("facet");
         facet_collector.add_facet(Facet::from("/country"));
         facet_collector.add_facet(Facet::from("/countryeurope"));
     }
@@ -630,7 +739,7 @@ mod tests {
         index_writer.commit().unwrap();
         let searcher = index.reader().unwrap().searcher();
 
-        let mut facet_collector = FacetCollector::for_field(facet_field);
+        let mut facet_collector = FacetCollector::for_field("facet");
         facet_collector.add_facet("/facet");
         let counts: FacetCounts = searcher.search(&AllQuery, &facet_collector).unwrap();
 
@@ -670,7 +779,7 @@ mod tests {
         index_writer.commit()?;
 
         let searcher = index.reader()?.searcher();
-        let mut facet_collector = FacetCollector::for_field(facet_field);
+        let mut facet_collector = FacetCollector::for_field("facet");
         facet_collector.add_facet("/facet");
         let counts: FacetCounts = searcher.search(&AllQuery, &facet_collector)?;
 

src/collector/filter_collector_wrapper.rs

@@ -12,10 +12,10 @@
 use std::marker::PhantomData;
 use std::sync::Arc;
 
+use columnar::{DynamicColumn, HasAssociatedColumnType};
 use fastfield_codecs::Column;
 
 use crate::collector::{Collector, SegmentCollector};
-use crate::fastfield::FastValue;
 use crate::schema::Field;
 use crate::{Score, SegmentReader, TantivyError};
 
@@ -61,7 +61,7 @@ use crate::{Score, SegmentReader, TantivyError};
 /// # Ok(())
 /// # }
 /// ```
-pub struct FilterCollector<TCollector, TPredicate, TPredicateValue: FastValue>
+pub struct FilterCollector<TCollector, TPredicate, TPredicateValue: Default>
 where TPredicate: 'static + Clone
 {
     field: Field,
@@ -70,7 +70,7 @@ where TPredicate: 'static + Clone
     t_predicate_value: PhantomData<TPredicateValue>,
 }
 
-impl<TCollector, TPredicate, TPredicateValue: FastValue>
+impl<TCollector, TPredicate, TPredicateValue: Default>
     FilterCollector<TCollector, TPredicate, TPredicateValue>
 where
     TCollector: Collector + Send + Sync,
@@ -91,12 +91,13 @@ where
     }
 }
 
-impl<TCollector, TPredicate, TPredicateValue: FastValue> Collector
+impl<TCollector, TPredicate, TPredicateValue: Default> Collector
     for FilterCollector<TCollector, TPredicate, TPredicateValue>
 where
     TCollector: Collector + Send + Sync,
     TPredicate: 'static + Fn(TPredicateValue) -> bool + Send + Sync + Clone,
-    TPredicateValue: FastValue,
+    TPredicateValue: HasAssociatedColumnType,
+    DynamicColumn: Into<Option<columnar::Column<TPredicateValue>>>,
 {
     // That's the type of our result.
     // Our standard deviation will be a float.
@@ -117,20 +118,10 @@ where
                 field_entry.name()
             )));
         }
-        let requested_type = TPredicateValue::to_type();
-        let field_schema_type = field_entry.field_type().value_type();
-        if requested_type != field_schema_type {
-            return Err(TantivyError::SchemaError(format!(
-                "Field {:?} is of type {:?}!={:?}",
-                field_entry.name(),
-                requested_type,
-                field_schema_type
-            )));
-        }
 
         let fast_field_reader = segment_reader
             .fast_fields()
-            .typed_fast_field_reader(schema.get_field_name(self.field))?;
+            .typed_column_first_or_default(schema.get_field_name(self.field))?;
 
         let segment_collector = self
             .collector
@@ -159,7 +150,7 @@ where
 pub struct FilterSegmentCollector<TSegmentCollector, TPredicate, TPredicateValue>
 where
     TPredicate: 'static,
-    TPredicateValue: FastValue,
+    DynamicColumn: Into<Option<columnar::Column<TPredicateValue>>>,
 {
     fast_field_reader: Arc<dyn Column<TPredicateValue>>,
     segment_collector: TSegmentCollector,
@@ -171,8 +162,9 @@ impl<TSegmentCollector, TPredicate, TPredicateValue> SegmentCollector
     for FilterSegmentCollector<TSegmentCollector, TPredicate, TPredicateValue>
 where
     TSegmentCollector: SegmentCollector,
+    TPredicateValue: HasAssociatedColumnType,
     TPredicate: 'static + Fn(TPredicateValue) -> bool + Send + Sync,
-    TPredicateValue: FastValue,
+    DynamicColumn: Into<Option<columnar::Column<TPredicateValue>>>,
 {
     type Fruit = TSegmentCollector::Fruit;
 

src/collector/histogram_collector.rs

@@ -4,7 +4,7 @@ use fastdivide::DividerU64;
 use fastfield_codecs::Column;
 
 use crate::collector::{Collector, SegmentCollector};
-use crate::fastfield::FastValue;
+use crate::fastfield::{FastFieldNotAvailableError, FastValue};
 use crate::schema::Type;
 use crate::{DocId, Score};
 
@@ -87,14 +87,14 @@ impl HistogramComputer {
 }
 pub struct SegmentHistogramCollector {
     histogram_computer: HistogramComputer,
-    ff_reader: Arc<dyn Column<u64>>,
+    column_u64: Arc<dyn Column<u64>>,
 }
 
 impl SegmentCollector for SegmentHistogramCollector {
     type Fruit = Vec<u64>;
 
     fn collect(&mut self, doc: DocId, _score: Score) {
-        let value = self.ff_reader.get_val(doc);
+        let value = self.column_u64.get_val(doc);
         self.histogram_computer.add_value(value);
     }
 
@@ -112,14 +112,18 @@ impl Collector for HistogramCollector {
         _segment_local_id: crate::SegmentOrdinal,
         segment: &crate::SegmentReader,
     ) -> crate::Result<Self::Child> {
-        let ff_reader = segment.fast_fields().u64_lenient(&self.field)?;
+        let column_opt = segment.fast_fields().u64_lenient(&self.field)?;
+        let column = column_opt.ok_or_else(|| FastFieldNotAvailableError {
+            field_name: self.field.clone(),
+        })?;
+        let column_u64 = column.first_or_default_col(0u64);
         Ok(SegmentHistogramCollector {
             histogram_computer: HistogramComputer {
                 counts: vec![0; self.num_buckets],
                 min_value: self.min_value,
                 divider: self.divider,
             },
-            ff_reader,
+            column_u64,
         })
     }
 

src/collector/mod.rs

@@ -104,7 +104,6 @@ pub use self::custom_score_top_collector::{CustomScorer, CustomSegmentScorer};
 
 mod tweak_score_top_collector;
 pub use self::tweak_score_top_collector::{ScoreSegmentTweaker, ScoreTweaker};
-
 mod facet_collector;
 pub use self::facet_collector::{FacetCollector, FacetCounts};
 use crate::query::Weight;

src/collector/tests.rs

@@ -5,7 +5,6 @@ use fastfield_codecs::Column;
 use super::*;
 use crate::collector::{Count, FilterCollector, TopDocs};
 use crate::core::SegmentReader;
-use crate::fastfield::BytesFastFieldReader;
 use crate::query::{AllQuery, QueryParser};
 use crate::schema::{Field, Schema, FAST, TEXT};
 use crate::time::format_description::well_known::Rfc3339;
@@ -58,9 +57,10 @@ pub fn test_filter_collector() -> crate::Result<()> {
 
     assert_eq!(filtered_top_docs.len(), 0);
 
-    fn date_filter(value: DateTime) -> bool {
-        (value.into_utc() - OffsetDateTime::parse("2019-04-09T00:00:00+00:00", &Rfc3339).unwrap())
-            .whole_weeks()
+    fn date_filter(value: columnar::DateTime) -> bool {
+        (crate::DateTime::from(value).into_utc()
+            - OffsetDateTime::parse("2019-04-09T00:00:00+00:00", &Rfc3339).unwrap())
+        .whole_weeks()
             > 0
     }
 
@@ -164,8 +164,10 @@ pub struct FastFieldSegmentCollector {
 }
 
 impl FastFieldTestCollector {
-    pub fn for_field(field: String) -> FastFieldTestCollector {
-        FastFieldTestCollector { field }
+    pub fn for_field(field: impl ToString) -> FastFieldTestCollector {
+        FastFieldTestCollector {
+            field: field.to_string(),
+        }
     }
 }
 
@@ -210,64 +212,62 @@ impl SegmentCollector for FastFieldSegmentCollector {
     }
 }
 
-/// Collects in order all of the fast field bytes for all of the
-/// docs in the `DocSet`
-///
-/// This collector is mainly useful for tests.
-pub struct BytesFastFieldTestCollector {
-    field: Field,
-}
+// /// Collects in order all of the fast field bytes for all of the
+// /// docs in the `DocSet`
+// ///
+// /// This collector is mainly useful for tests.
+// pub struct BytesFastFieldTestCollector {
+//     field: Field,
+// }
 
-pub struct BytesFastFieldSegmentCollector {
-    vals: Vec<u8>,
-    reader: BytesFastFieldReader,
-}
+// pub struct BytesFastFieldSegmentCollector {
+//     vals: Vec<u8>,
+//     reader: BytesFastFieldReader,
+// }
 
-impl BytesFastFieldTestCollector {
-    pub fn for_field(field: Field) -> BytesFastFieldTestCollector {
-        BytesFastFieldTestCollector { field }
-    }
-}
+// impl BytesFastFieldTestCollector {
+//     pub fn for_field(field: Field) -> BytesFastFieldTestCollector {
+//         BytesFastFieldTestCollector { field }
+//     }
+// }
 
-impl Collector for BytesFastFieldTestCollector {
-    type Fruit = Vec<u8>;
-    type Child = BytesFastFieldSegmentCollector;
+// impl Collector for BytesFastFieldTestCollector {
+//     type Fruit = Vec<u8>;
+//     type Child = BytesFastFieldSegmentCollector;
 
-    fn for_segment(
-        &self,
-        _segment_local_id: u32,
-        segment_reader: &SegmentReader,
-    ) -> crate::Result<BytesFastFieldSegmentCollector> {
-        let reader = segment_reader
-            .fast_fields()
-            .bytes(segment_reader.schema().get_field_name(self.field))?;
-        Ok(BytesFastFieldSegmentCollector {
-            vals: Vec::new(),
-            reader,
-        })
-    }
+//     fn for_segment(
+//         &self,
+//         _segment_local_id: u32,
+//         segment_reader: &SegmentReader,
+//     ) -> crate::Result<BytesFastFieldSegmentCollector> {
+//         let reader = segment_reader.fast_fields().bytes(self.field)?;
+//         Ok(BytesFastFieldSegmentCollector {
+//             vals: Vec::new(),
+//             reader,
+//         })
+//     }
 
-    fn requires_scoring(&self) -> bool {
-        false
-    }
+//     fn requires_scoring(&self) -> bool {
+//         false
+//     }
 
-    fn merge_fruits(&self, children: Vec<Vec<u8>>) -> crate::Result<Vec<u8>> {
-        Ok(children.into_iter().flat_map(|c| c.into_iter()).collect())
-    }
-}
+//     fn merge_fruits(&self, children: Vec<Vec<u8>>) -> crate::Result<Vec<u8>> {
+//         Ok(children.into_iter().flat_map(|c| c.into_iter()).collect())
+//     }
+// }
 
-impl SegmentCollector for BytesFastFieldSegmentCollector {
-    type Fruit = Vec<u8>;
+// impl SegmentCollector for BytesFastFieldSegmentCollector {
+//     type Fruit = Vec<u8>;
 
-    fn collect(&mut self, doc: u32, _score: Score) {
-        let data = self.reader.get_bytes(doc);
-        self.vals.extend(data);
-    }
+//     fn collect(&mut self, doc: u32, _score: Score) {
+//         let data = self.reader.get_bytes(doc);
+//         self.vals.extend(data);
+//     }
 
-    fn harvest(self) -> <Self as SegmentCollector>::Fruit {
-        self.vals
-    }
-}
+//     fn harvest(self) -> <Self as SegmentCollector>::Fruit {
+//         self.vals
+//     }
+// }
 
 fn make_test_searcher() -> crate::Result<Searcher> {
     let schema = Schema::builder().build();

src/collector/top_score_collector.rs

@@ -12,7 +12,7 @@ use crate::collector::tweak_score_top_collector::TweakedScoreTopCollector;
 use crate::collector::{
     CustomScorer, CustomSegmentScorer, ScoreSegmentTweaker, ScoreTweaker, SegmentCollector,
 };
-use crate::fastfield::FastValue;
+use crate::fastfield::{FastFieldNotAvailableError, FastValue};
 use crate::query::Weight;
 use crate::schema::Field;
 use crate::{DocAddress, DocId, Score, SegmentOrdinal, SegmentReader, TantivyError};
@@ -22,7 +22,7 @@ struct FastFieldConvertCollector<
     TFastValue: FastValue,
 > {
     pub collector: TCollector,
-    pub field: Field,
+    pub field: String,
     pub fast_value: std::marker::PhantomData<TFastValue>,
 }
 
@@ -41,7 +41,8 @@ where
         segment: &SegmentReader,
     ) -> crate::Result<Self::Child> {
         let schema = segment.schema();
-        let field_entry = schema.get_field_entry(self.field);
+        let field = schema.get_field(&self.field)?;
+        let field_entry = schema.get_field_entry(field);
         if !field_entry.is_fast() {
             return Err(TantivyError::SchemaError(format!(
                 "Field {:?} is not a fast field.",
@@ -132,17 +133,17 @@ impl fmt::Debug for TopDocs {
 }
 
 struct ScorerByFastFieldReader {
-    ff_reader: Arc<dyn Column<u64>>,
+    sort_column: Arc<dyn Column<u64>>,
 }
 
 impl CustomSegmentScorer<u64> for ScorerByFastFieldReader {
     fn score(&mut self, doc: DocId) -> u64 {
-        self.ff_reader.get_val(doc)
+        self.sort_column.get_val(doc)
     }
 }
 
 struct ScorerByField {
-    field: Field,
+    field: String,
 }
 
 impl CustomScorer<u64> for ScorerByField {
@@ -154,10 +155,13 @@ impl CustomScorer<u64> for ScorerByField {
         // mapping is monotonic, so it is sufficient to compute our top-K docs.
         //
         // The conversion will then happen only on the top-K docs.
-        let ff_reader = segment_reader
-            .fast_fields()
-            .typed_fast_field_reader(segment_reader.schema().get_field_name(self.field))?;
-        Ok(ScorerByFastFieldReader { ff_reader })
+        let sort_column_opt = segment_reader.fast_fields().u64_lenient(&self.field)?;
+        let sort_column = sort_column_opt
+            .ok_or_else(|| FastFieldNotAvailableError {
+                field_name: self.field.clone(),
+            })?
+            .first_or_default_col(0u64);
+        Ok(ScorerByFastFieldReader { sort_column })
     }
 }
 
@@ -290,9 +294,14 @@ impl TopDocs {
     /// the [.order_by_fast_field(...)](TopDocs::order_by_fast_field) method.
     pub fn order_by_u64_field(
         self,
-        field: Field,
+        field: impl ToString,
     ) -> impl Collector<Fruit = Vec<(u64, DocAddress)>> {
-        CustomScoreTopCollector::new(ScorerByField { field }, self.0.into_tscore())
+        CustomScoreTopCollector::new(
+            ScorerByField {
+                field: field.to_string(),
+            },
+            self.0.into_tscore(),
+        )
     }
 
     /// Set top-K to rank documents by a given fast field.
@@ -367,15 +376,15 @@ impl TopDocs {
     /// ```
     pub fn order_by_fast_field<TFastValue>(
         self,
-        fast_field: Field,
+        fast_field: impl ToString,
     ) -> impl Collector<Fruit = Vec<(TFastValue, DocAddress)>>
     where
         TFastValue: FastValue,
     {
-        let u64_collector = self.order_by_u64_field(fast_field);
+        let u64_collector = self.order_by_u64_field(fast_field.to_string());
         FastFieldConvertCollector {
             collector: u64_collector,
-            field: fast_field,
+            field: fast_field.to_string(),
             fast_value: PhantomData,
         }
     }
@@ -877,7 +886,7 @@ mod tests {
         });
         let searcher = index.reader()?.searcher();
 
-        let top_collector = TopDocs::with_limit(4).order_by_u64_field(size);
+        let top_collector = TopDocs::with_limit(4).order_by_u64_field(SIZE);
         let top_docs: Vec<(u64, DocAddress)> = searcher.search(&query, &top_collector)?;
         assert_eq!(
             &top_docs[..],
@@ -916,7 +925,7 @@ mod tests {
         ))?;
         index_writer.commit()?;
         let searcher = index.reader()?.searcher();
-        let top_collector = TopDocs::with_limit(3).order_by_fast_field(birthday);
+        let top_collector = TopDocs::with_limit(3).order_by_fast_field("birthday");
         let top_docs: Vec<(DateTime, DocAddress)> = searcher.search(&AllQuery, &top_collector)?;
         assert_eq!(
             &top_docs[..],
@@ -946,7 +955,7 @@ mod tests {
         ))?;
         index_writer.commit()?;
         let searcher = index.reader()?.searcher();
-        let top_collector = TopDocs::with_limit(3).order_by_fast_field(altitude);
+        let top_collector = TopDocs::with_limit(3).order_by_fast_field("altitude");
         let top_docs: Vec<(i64, DocAddress)> = searcher.search(&AllQuery, &top_collector)?;
         assert_eq!(
             &top_docs[..],
@@ -976,7 +985,7 @@ mod tests {
         ))?;
         index_writer.commit()?;
         let searcher = index.reader()?.searcher();
-        let top_collector = TopDocs::with_limit(3).order_by_fast_field(altitude);
+        let top_collector = TopDocs::with_limit(3).order_by_fast_field("altitude");
         let top_docs: Vec<(f64, DocAddress)> = searcher.search(&AllQuery, &top_collector)?;
         assert_eq!(
             &top_docs[..],
@@ -1004,7 +1013,7 @@ mod tests {
                 .unwrap();
         });
         let searcher = index.reader().unwrap().searcher();
-        let top_collector = TopDocs::with_limit(4).order_by_u64_field(Field::from_field_id(2));
+        let top_collector = TopDocs::with_limit(4).order_by_u64_field("missing_field");
         let segment_reader = searcher.segment_reader(0u32);
         top_collector
             .for_segment(0, segment_reader)
@@ -1022,7 +1031,7 @@ mod tests {
         index_writer.commit()?;
         let searcher = index.reader()?.searcher();
         let segment = searcher.segment_reader(0);
-        let top_collector = TopDocs::with_limit(4).order_by_u64_field(size);
+        let top_collector = TopDocs::with_limit(4).order_by_u64_field(SIZE);
         let err = top_collector.for_segment(0, segment).err().unwrap();
         assert!(matches!(err, crate::TantivyError::SchemaError(_)));
         Ok(())
@@ -1039,7 +1048,7 @@ mod tests {
         index_writer.commit()?;
         let searcher = index.reader()?.searcher();
         let segment = searcher.segment_reader(0);
-        let top_collector = TopDocs::with_limit(4).order_by_fast_field::<i64>(size);
+        let top_collector = TopDocs::with_limit(4).order_by_fast_field::<i64>(SIZE);
         let err = top_collector.for_segment(0, segment).err().unwrap();
         assert!(
             matches!(err, crate::TantivyError::SchemaError(msg) if msg == "Field \"size\" is not a fast field.")

src/core/index.rs

@@ -19,7 +19,7 @@ use crate::error::{DataCorruption, TantivyError};
 use crate::indexer::index_writer::{MAX_NUM_THREAD, MEMORY_ARENA_NUM_BYTES_MIN};
 use crate::indexer::segment_updater::save_metas;
 use crate::reader::{IndexReader, IndexReaderBuilder};
-use crate::schema::{Cardinality, Field, FieldType, Schema};
+use crate::schema::{Field, FieldType, Schema};
 use crate::tokenizer::{TextAnalyzer, TokenizerManager};
 use crate::IndexWriter;
 
@@ -93,7 +93,7 @@ fn save_new_metas(
 /// let body_field = schema_builder.add_text_field("body", TEXT);
 /// let number_field = schema_builder.add_u64_field(
 ///     "number",
-///     NumericOptions::default().set_fast(Cardinality::SingleValue),
+///     NumericOptions::default().set_fast(),
 /// );
 ///
 /// let schema = schema_builder.build();
@@ -245,12 +245,6 @@ impl IndexBuilder {
                         sort_by_field.field
                     )));
                 }
-                if entry.field_type().fastfield_cardinality() != Some(Cardinality::SingleValue) {
-                    return Err(TantivyError::InvalidArgument(format!(
-                        "Only single value fast field Cardinality supported for sorting index {}",
-                        sort_by_field.field
-                    )));
-                }
             }
             Ok(())
         } else {

src/core/segment_reader.rs

@@ -9,7 +9,7 @@ use crate::directory::{CompositeFile, FileSlice};
 use crate::error::DataCorruption;
 use crate::fastfield::{intersect_alive_bitsets, AliveBitSet, FacetReader, FastFieldReaders};
 use crate::fieldnorm::{FieldNormReader, FieldNormReaders};
-use crate::schema::{Field, FieldType, IndexRecordOption, Schema};
+use crate::schema::{Field, FieldType, IndexRecordOption, Schema, Type};
 use crate::space_usage::SegmentSpaceUsage;
 use crate::store::StoreReader;
 use crate::termdict::TermDictionary;
@@ -90,25 +90,19 @@ impl SegmentReader {
     }
 
     /// Accessor to the `FacetReader` associated with a given `Field`.
-    pub fn facet_reader(&self, field: Field) -> crate::Result<FacetReader> {
-        let field_entry = self.schema.get_field_entry(field);
-
-        match field_entry.field_type() {
-            FieldType::Facet(_) => {
-                let term_ords_reader =
-                    self.fast_fields().u64s(self.schema.get_field_name(field))?;
-                let termdict = self
-                    .termdict_composite
-                    .open_read(field)
-                    .map(TermDictionary::open)
-                    .unwrap_or_else(|| Ok(TermDictionary::empty()))?;
-                Ok(FacetReader::new(term_ords_reader, termdict))
-            }
-            _ => Err(crate::TantivyError::InvalidArgument(format!(
-                "Field {:?} is not a facet field.",
-                field_entry.name()
-            ))),
+    pub fn facet_reader(&self, field_name: &str) -> crate::Result<FacetReader> {
+        let schema = self.schema();
+        let field = schema.get_field(field_name)?;
+        let field_entry = schema.get_field_entry(field);
+        if field_entry.field_type().value_type() != Type::Facet {
+            return Err(crate::TantivyError::SchemaError(format!(
+                "`{field_name}` is not a facet field.`"
+            )));
         }
+        let Some(facet_column) = self.fast_fields().str_column_opt(field_name)? else {
+            panic!("Facet Field `{field_name}` is missing. This should not happen");
+        };
+        Ok(FacetReader::new(facet_column))
     }
 
     /// Accessor to the segment's `Field norms`'s reader.
@@ -170,9 +164,7 @@ impl SegmentReader {
         let schema = segment.schema();
 
         let fast_fields_data = segment.open_read(SegmentComponent::FastFields)?;
-        let fast_fields_composite = CompositeFile::open(&fast_fields_data)?;
-        let fast_fields_readers =
-            Arc::new(FastFieldReaders::new(schema.clone(), fast_fields_composite));
+        let fast_fields_readers = Arc::new(FastFieldReaders::open(fast_fields_data)?);
         let fieldnorm_data = segment.open_read(SegmentComponent::FieldNorms)?;
         let fieldnorm_readers = FieldNormReaders::open(fieldnorm_data)?;
 
@@ -326,7 +318,7 @@ impl SegmentReader {
             self.termdict_composite.space_usage(),
             self.postings_composite.space_usage(),
             self.positions_composite.space_usage(),
-            self.fast_fields_readers.space_usage(),
+            self.fast_fields_readers.space_usage(self.schema())?,
             self.fieldnorm_readers.space_usage(),
             self.get_store_reader(0)?.space_usage(),
             self.alive_bitset_opt

src/directory/composite_file.rs

@@ -169,12 +169,11 @@ impl CompositeFile {
     }
 
     pub fn space_usage(&self) -> PerFieldSpaceUsage {
-        let mut fields = HashMap::new();
+        let mut fields = Vec::new();
         for (&field_addr, byte_range) in &self.offsets_index {
-            fields
-                .entry(field_addr.field)
-                .or_insert_with(|| FieldUsage::empty(field_addr.field))
-                .add_field_idx(field_addr.idx, byte_range.len());
+            let mut field_usage = FieldUsage::empty(field_addr.field);
+            field_usage.add_field_idx(field_addr.idx, byte_range.len());
+            fields.push(field_usage);
         }
         PerFieldSpaceUsage::new(fields)
     }

src/fastfield/bytes/mod.rs

@@ -1,116 +0,0 @@
-mod reader;
-mod writer;
-
-pub use self::reader::BytesFastFieldReader;
-pub use self::writer::BytesFastFieldWriter;
-
-#[cfg(test)]
-mod tests {
-    use crate::query::{EnableScoring, TermQuery};
-    use crate::schema::{BytesOptions, IndexRecordOption, Schema, Value, FAST, INDEXED, STORED};
-    use crate::{DocAddress, DocSet, Index, Searcher, Term};
-
-    #[test]
-    fn test_bytes() -> crate::Result<()> {
-        let mut schema_builder = Schema::builder();
-        let bytes_field = schema_builder.add_bytes_field("bytesfield", FAST);
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
-        index_writer.add_document(doc!(bytes_field=>vec![0u8, 1, 2, 3]))?;
-        index_writer.add_document(doc!(bytes_field=>vec![]))?;
-        index_writer.add_document(doc!(bytes_field=>vec![255u8]))?;
-        index_writer.add_document(doc!(bytes_field=>vec![1u8, 3, 5, 7, 9]))?;
-        index_writer.add_document(doc!(bytes_field=>vec![0u8; 1000]))?;
-        index_writer.commit()?;
-        let searcher = index.reader()?.searcher();
-        let segment_reader = searcher.segment_reader(0);
-        let bytes_reader = segment_reader.fast_fields().bytes("bytesfield").unwrap();
-        assert_eq!(bytes_reader.get_bytes(0), &[0u8, 1, 2, 3]);
-        assert!(bytes_reader.get_bytes(1).is_empty());
-        assert_eq!(bytes_reader.get_bytes(2), &[255u8]);
-        assert_eq!(bytes_reader.get_bytes(3), &[1u8, 3, 5, 7, 9]);
-        let long = vec![0u8; 1000];
-        assert_eq!(bytes_reader.get_bytes(4), long.as_slice());
-        Ok(())
-    }
-
-    fn create_index_for_test<T: Into<BytesOptions>>(byte_options: T) -> crate::Result<Searcher> {
-        let mut schema_builder = Schema::builder();
-        let field = schema_builder.add_bytes_field("string_bytes", byte_options.into());
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
-        index_writer.add_document(doc!(
-                field => b"tantivy".as_ref(),
-                field => b"lucene".as_ref()
-        ))?;
-        index_writer.commit()?;
-        Ok(index.reader()?.searcher())
-    }
-
-    #[test]
-    fn test_stored_bytes() -> crate::Result<()> {
-        let searcher = create_index_for_test(STORED)?;
-        assert_eq!(searcher.num_docs(), 1);
-        let retrieved_doc = searcher.doc(DocAddress::new(0u32, 0u32))?;
-        let field = searcher.schema().get_field("string_bytes").unwrap();
-        let values: Vec<&Value> = retrieved_doc.get_all(field).collect();
-        assert_eq!(values.len(), 2);
-        let values_bytes: Vec<&[u8]> = values
-            .into_iter()
-            .flat_map(|value| value.as_bytes())
-            .collect();
-        assert_eq!(values_bytes, &[&b"tantivy"[..], &b"lucene"[..]]);
-        Ok(())
-    }
-
-    #[test]
-    fn test_non_stored_bytes() -> crate::Result<()> {
-        let searcher = create_index_for_test(INDEXED)?;
-        assert_eq!(searcher.num_docs(), 1);
-        let retrieved_doc = searcher.doc(DocAddress::new(0u32, 0u32))?;
-        let field = searcher.schema().get_field("string_bytes").unwrap();
-        assert!(retrieved_doc.get_first(field).is_none());
-        Ok(())
-    }
-
-    #[test]
-    fn test_index_bytes() -> crate::Result<()> {
-        let searcher = create_index_for_test(INDEXED)?;
-        assert_eq!(searcher.num_docs(), 1);
-        let field = searcher.schema().get_field("string_bytes").unwrap();
-        let term = Term::from_field_bytes(field, b"lucene".as_ref());
-        let term_query = TermQuery::new(term, IndexRecordOption::Basic);
-        let term_weight = term_query.specialized_weight(EnableScoring::Enabled(&searcher))?;
-        let term_scorer = term_weight.specialized_scorer(searcher.segment_reader(0), 1.0)?;
-        assert_eq!(term_scorer.doc(), 0u32);
-        Ok(())
-    }
-
-    #[test]
-    fn test_non_index_bytes() -> crate::Result<()> {
-        let searcher = create_index_for_test(STORED)?;
-        assert_eq!(searcher.num_docs(), 1);
-        let field = searcher.schema().get_field("string_bytes").unwrap();
-        let term = Term::from_field_bytes(field, b"lucene".as_ref());
-        let term_query = TermQuery::new(term, IndexRecordOption::Basic);
-        let term_weight_err =
-            term_query.specialized_weight(EnableScoring::disabled_from_schema(searcher.schema()));
-        assert!(matches!(
-            term_weight_err,
-            Err(crate::TantivyError::SchemaError(_))
-        ));
-        Ok(())
-    }
-
-    #[test]
-    fn test_fast_bytes_multivalue_value() -> crate::Result<()> {
-        let searcher = create_index_for_test(FAST)?;
-        assert_eq!(searcher.num_docs(), 1);
-        let fast_fields = searcher.segment_reader(0u32).fast_fields();
-        let fast_field_reader = fast_fields.bytes("string_bytes").unwrap();
-        assert_eq!(fast_field_reader.get_bytes(0u32), b"tantivy");
-        Ok(())
-    }
-}

src/fastfield/bytes/reader.rs

@@ -1,58 +0,0 @@
-use std::sync::Arc;
-
-use fastfield_codecs::Column;
-
-use crate::directory::{FileSlice, OwnedBytes};
-use crate::fastfield::MultiValueIndex;
-use crate::DocId;
-
-/// Reader for byte array fast fields
-///
-/// The reader is implemented as a `u64` fast field and a separate collection of bytes.
-///
-/// The `vals_reader` will access the concatenated list of all values for all documents.
-///
-/// The `idx_reader` associates, for each document, the index of its first value.
-///
-/// Reading the value for a document is done by reading the start index for it,
-/// and the start index for the next document, and keeping the bytes in between.
-#[derive(Clone)]
-pub struct BytesFastFieldReader {
-    idx_reader: MultiValueIndex,
-    values: OwnedBytes,
-}
-
-impl BytesFastFieldReader {
-    pub(crate) fn open(
-        idx_reader: Arc<dyn Column<u64>>,
-        values_file: FileSlice,
-    ) -> crate::Result<BytesFastFieldReader> {
-        let values = values_file.read_bytes()?;
-        Ok(BytesFastFieldReader {
-            idx_reader: MultiValueIndex::new(idx_reader),
-            values,
-        })
-    }
-
-    /// returns the multivalue index
-    pub fn get_index_reader(&self) -> &MultiValueIndex {
-        &self.idx_reader
-    }
-
-    /// Returns the bytes associated with the given `doc`
-    pub fn get_bytes(&self, doc: DocId) -> &[u8] {
-        let range = self.idx_reader.range(doc);
-        &self.values.as_slice()[range.start as usize..range.end as usize]
-    }
-
-    /// Returns the length of the bytes associated with the given `doc`
-    pub fn num_bytes(&self, doc: DocId) -> u64 {
-        let range = self.idx_reader.range(doc);
-        (range.end - range.start) as u64
-    }
-
-    /// Returns the overall number of bytes in this bytes fast field.
-    pub fn total_num_bytes(&self) -> u32 {
-        self.values.len() as u32
-    }
-}

src/fastfield/bytes/writer.rs

@@ -1,145 +0,0 @@
-use std::io::{self, Write};
-
-use fastfield_codecs::VecColumn;
-
-use crate::fastfield::serializer::CompositeFastFieldSerializer;
-use crate::fastfield::MultivalueStartIndex;
-use crate::indexer::doc_id_mapping::DocIdMapping;
-use crate::schema::{Document, Field, Value};
-use crate::DocId;
-
-/// Writer for byte array (as in, any number of bytes per document) fast fields
-///
-/// This `BytesFastFieldWriter` is only useful for advanced users.
-/// The normal way to get your associated bytes in your index
-/// is to
-/// - declare your field with fast set to
-/// [`Cardinality::SingleValue`](crate::schema::Cardinality::SingleValue)
-/// in your schema
-/// - add your document simply by calling `.add_document(...)` with associating bytes to the field.
-///
-/// The `BytesFastFieldWriter` can be acquired from the
-/// fast field writer by calling
-/// [`.get_bytes_writer_mut(...)`](crate::fastfield::FastFieldsWriter::get_bytes_writer_mut).
-///
-/// Once acquired, writing is done by calling
-/// [`.add_document_val(&[u8])`](BytesFastFieldWriter::add_document_val)
-/// once per document, even if there are no bytes associated with it.
-pub struct BytesFastFieldWriter {
-    field: Field,
-    vals: Vec<u8>,
-    doc_index: Vec<u64>,
-}
-
-impl BytesFastFieldWriter {
-    /// Creates a new `BytesFastFieldWriter`
-    pub fn new(field: Field) -> Self {
-        BytesFastFieldWriter {
-            field,
-            vals: Vec::new(),
-            doc_index: Vec::new(),
-        }
-    }
-
-    /// The memory used (inclusive childs)
-    pub fn mem_usage(&self) -> usize {
-        self.vals.capacity() + self.doc_index.capacity() * std::mem::size_of::<u64>()
-    }
-    /// Access the field associated with the `BytesFastFieldWriter`
-    pub fn field(&self) -> Field {
-        self.field
-    }
-
-    /// Finalize the current document.
-    pub(crate) fn next_doc(&mut self) {
-        self.doc_index.push(self.vals.len() as u64);
-    }
-
-    /// Shift to the next document and add all of the
-    /// matching field values present in the document.
-    pub fn add_document(&mut self, doc: &Document) -> crate::Result<()> {
-        self.next_doc();
-        for field_value in doc.get_all(self.field) {
-            if let Value::Bytes(ref bytes) = field_value {
-                self.vals.extend_from_slice(bytes);
-                return Ok(());
-            }
-        }
-        Ok(())
-    }
-
-    /// Register the bytes associated with a document.
-    ///
-    /// The method returns the `DocId` of the document that was
-    /// just written.
-    pub fn add_document_val(&mut self, val: &[u8]) -> DocId {
-        let doc = self.doc_index.len() as DocId;
-        self.next_doc();
-        self.vals.extend_from_slice(val);
-        doc
-    }
-
-    /// Returns an iterator over values per doc_id in ascending doc_id order.
-    ///
-    /// Normally the order is simply iterating self.doc_id_index.
-    /// With doc_id_map it accounts for the new mapping, returning values in the order of the
-    /// new doc_ids.
-    fn get_ordered_values<'a: 'b, 'b>(
-        &'a self,
-        doc_id_map: Option<&'b DocIdMapping>,
-    ) -> impl Iterator<Item = &'b [u8]> {
-        let doc_id_iter: Box<dyn Iterator<Item = u32>> = if let Some(doc_id_map) = doc_id_map {
-            Box::new(doc_id_map.iter_old_doc_ids())
-        } else {
-            let max_doc = self.doc_index.len() as u32;
-            Box::new(0..max_doc)
-        };
-        doc_id_iter.map(move |doc_id| self.get_values_for_doc_id(doc_id))
-    }
-
-    /// returns all values for a doc_ids
-    fn get_values_for_doc_id(&self, doc_id: u32) -> &[u8] {
-        let start_pos = self.doc_index[doc_id as usize] as usize;
-        let end_pos = self
-            .doc_index
-            .get(doc_id as usize + 1)
-            .cloned()
-            .unwrap_or(self.vals.len() as u64) as usize; // special case, last doc_id has no offset information
-        &self.vals[start_pos..end_pos]
-    }
-
-    /// Serializes the fast field values by pushing them to the `FastFieldSerializer`.
-    pub fn serialize(
-        mut self,
-        serializer: &mut CompositeFastFieldSerializer,
-        doc_id_map: Option<&DocIdMapping>,
-    ) -> io::Result<()> {
-        // writing the offset index
-        {
-            self.doc_index.push(self.vals.len() as u64);
-            let col = VecColumn::from(&self.doc_index[..]);
-            if let Some(doc_id_map) = doc_id_map {
-                let multi_value_start_index = MultivalueStartIndex::new(&col, doc_id_map);
-                serializer.create_auto_detect_u64_fast_field_with_idx(
-                    self.field,
-                    multi_value_start_index,
-                    0,
-                )?;
-            } else {
-                serializer.create_auto_detect_u64_fast_field_with_idx(self.field, col, 0)?;
-            }
-        }
-        // writing the values themselves
-        let mut value_serializer = serializer.new_bytes_fast_field(self.field);
-        // the else could be removed, but this is faster (difference not benchmarked)
-        if let Some(doc_id_map) = doc_id_map {
-            for vals in self.get_ordered_values(Some(doc_id_map)) {
-                // sort values in case of remapped doc_ids?
-                value_serializer.write_all(vals)?;
-            }
-        } else {
-            value_serializer.write_all(&self.vals)?;
-        }
-        Ok(())
-    }
-}

src/fastfield/error.rs

@@ -8,7 +8,7 @@ use crate::schema::FieldEntry;
 #[derive(Debug, Error)]
 #[error("Fast field not available: '{field_name:?}'")]
 pub struct FastFieldNotAvailableError {
-    field_name: String,
+    pub(crate) field_name: String,
 }
 
 impl FastFieldNotAvailableError {

src/fastfield/facet_reader.rs

@@ -1,9 +1,7 @@
-use std::str;
+use columnar::StrColumn;
 
-use super::MultiValuedFastFieldReader;
-use crate::error::DataCorruption;
 use crate::schema::Facet;
-use crate::termdict::{TermDictionary, TermOrdinal};
+use crate::termdict::TermOrdinal;
 use crate::DocId;
 
 /// The facet reader makes it possible to access the list of
@@ -20,9 +18,7 @@ use crate::DocId;
 /// list of facets. This ordinal is segment local and
 /// only makes sense for a given segment.
 pub struct FacetReader {
-    term_ords: MultiValuedFastFieldReader<u64>,
-    term_dict: TermDictionary,
-    buffer: Vec<u8>,
+    facet_column: StrColumn,
 }
 
 impl FacetReader {
@@ -33,15 +29,8 @@ impl FacetReader {
     /// access the list of facet ords for a given document.
     /// - a `TermDictionary` that helps associating a facet to
     /// an ordinal and vice versa.
-    pub fn new(
-        term_ords: MultiValuedFastFieldReader<u64>,
-        term_dict: TermDictionary,
-    ) -> FacetReader {
-        FacetReader {
-            term_ords,
-            term_dict,
-            buffer: vec![],
-        }
+    pub fn new(facet_column: StrColumn) -> FacetReader {
+        FacetReader { facet_column }
     }
 
     /// Returns the size of the sets of facets in the segment.
@@ -50,31 +39,23 @@ impl FacetReader {
     ///
     /// `Facet` ordinals range from `0` to `num_facets() - 1`.
     pub fn num_facets(&self) -> usize {
-        self.term_dict.num_terms()
-    }
-
-    /// Accessor for the facet term dictionary.
-    pub fn facet_dict(&self) -> &TermDictionary {
-        &self.term_dict
+        self.facet_column.num_terms()
     }
 
     /// Given a term ordinal returns the term associated with it.
-    pub fn facet_from_ord(
-        &mut self,
-        facet_ord: TermOrdinal,
-        output: &mut Facet,
-    ) -> crate::Result<()> {
-        let found_term = self.term_dict.ord_to_term(facet_ord, &mut self.buffer)?;
-        assert!(found_term, "Term ordinal {} no found.", facet_ord);
-        let facet_str = str::from_utf8(&self.buffer[..])
-            .map_err(|utf8_err| DataCorruption::comment_only(utf8_err.to_string()))?;
-        output.set_facet_str(facet_str);
+    pub fn facet_from_ord(&self, facet_ord: TermOrdinal, output: &mut Facet) -> crate::Result<()> {
+        let found_term = self.facet_column.ord_to_str(facet_ord, &mut output.0)?;
+        assert!(found_term, "Term ordinal {facet_ord} no found.");
         Ok(())
     }
 
     /// Return the list of facet ordinals associated with a document.
-    pub fn facet_ords(&self, doc: DocId, output: &mut Vec<u64>) {
-        self.term_ords.get_vals(doc, output);
+    pub fn facet_ords(&self, doc: DocId) -> impl Iterator<Item = u64> + '_ {
+        self.facet_column.ords().values(doc)
+    }
+
+    pub fn facet_dict(&self) -> &columnar::Dictionary {
+        self.facet_column.dictionary()
     }
 }
 
@@ -84,26 +65,66 @@ mod tests {
     use crate::{DocAddress, Document, Index};
 
     #[test]
-    fn test_facet_only_indexed() -> crate::Result<()> {
+    fn test_facet_only_indexed() {
         let mut schema_builder = SchemaBuilder::default();
         let facet_field = schema_builder.add_facet_field("facet", FacetOptions::default());
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
-        index_writer.add_document(doc!(facet_field=>Facet::from_text("/a/b").unwrap()))?;
-        index_writer.commit()?;
-        let searcher = index.reader()?.searcher();
-        let facet_reader = searcher
-            .segment_reader(0u32)
-            .facet_reader(facet_field)
+        let mut index_writer = index.writer_for_tests().unwrap();
+        index_writer
+            .add_document(doc!(facet_field=>Facet::from_text("/a/b").unwrap()))
             .unwrap();
+        index_writer.commit().unwrap();
+        let searcher = index.reader().unwrap().searcher();
+        let facet_reader = searcher.segment_reader(0u32).facet_reader("facet").unwrap();
         let mut facet_ords = Vec::new();
-        facet_reader.facet_ords(0u32, &mut facet_ords);
-        assert_eq!(&facet_ords, &[2u64]);
-        let doc = searcher.doc(DocAddress::new(0u32, 0u32))?;
+        facet_ords.extend(facet_reader.facet_ords(0u32));
+        assert_eq!(&facet_ords, &[0u64]);
+        assert_eq!(facet_reader.num_facets(), 1);
+        let mut facet = Facet::default();
+        facet_reader.facet_from_ord(0, &mut facet).unwrap();
+        assert_eq!(facet.to_path_string(), "/a/b");
+        let doc = searcher.doc(DocAddress::new(0u32, 0u32)).unwrap();
         let value = doc.get_first(facet_field).and_then(Value::as_facet);
         assert_eq!(value, None);
-        Ok(())
+    }
+
+    #[test]
+    fn test_facet_several_facets_sorted() {
+        let mut schema_builder = SchemaBuilder::default();
+        let facet_field = schema_builder.add_facet_field("facet", FacetOptions::default());
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_for_tests().unwrap();
+        index_writer
+            .add_document(doc!(facet_field=>Facet::from_text("/parent/child1").unwrap()))
+            .unwrap();
+        index_writer
+            .add_document(doc!(
+                facet_field=>Facet::from_text("/parent/child2").unwrap(),
+                facet_field=>Facet::from_text("/parent/child1/blop").unwrap(),
+            ))
+            .unwrap();
+        index_writer.commit().unwrap();
+        let searcher = index.reader().unwrap().searcher();
+        let facet_reader = searcher.segment_reader(0u32).facet_reader("facet").unwrap();
+        let mut facet_ords = Vec::new();
+
+        facet_ords.extend(facet_reader.facet_ords(0u32));
+        assert_eq!(&facet_ords, &[0u64]);
+
+        facet_ords.clear();
+        facet_ords.extend(facet_reader.facet_ords(1u32));
+        assert_eq!(&facet_ords, &[1u64, 2u64]);
+
+        assert_eq!(facet_reader.num_facets(), 3);
+        let mut facet = Facet::default();
+        facet_reader.facet_from_ord(0, &mut facet).unwrap();
+        assert_eq!(facet.to_path_string(), "/parent/child1");
+        facet_reader.facet_from_ord(1, &mut facet).unwrap();
+        assert_eq!(facet.to_path_string(), "/parent/child1/blop");
+        facet_reader.facet_from_ord(2, &mut facet).unwrap();
+        assert_eq!(facet.to_path_string(), "/parent/child2");
     }
 
     #[test]
@@ -116,13 +137,10 @@ mod tests {
         index_writer.add_document(doc!(facet_field=>Facet::from_text("/a/b").unwrap()))?;
         index_writer.commit()?;
         let searcher = index.reader()?.searcher();
-        let facet_reader = searcher
-            .segment_reader(0u32)
-            .facet_reader(facet_field)
-            .unwrap();
+        let facet_reader = searcher.segment_reader(0u32).facet_reader("facet").unwrap();
         let mut facet_ords = Vec::new();
-        facet_reader.facet_ords(0u32, &mut facet_ords);
-        assert_eq!(&facet_ords, &[2u64]);
+        facet_ords.extend(facet_reader.facet_ords(0u32));
+        assert_eq!(&facet_ords, &[0u64]);
         let doc = searcher.doc(DocAddress::new(0u32, 0u32))?;
         let value: Option<&Facet> = doc.get_first(facet_field).and_then(Value::as_facet);
         assert_eq!(value, Facet::from_text("/a/b").ok().as_ref());
@@ -140,14 +158,12 @@ mod tests {
         index_writer.add_document(Document::default())?;
         index_writer.commit()?;
         let searcher = index.reader()?.searcher();
-        let facet_reader = searcher
-            .segment_reader(0u32)
-            .facet_reader(facet_field)
-            .unwrap();
+        let facet_reader = searcher.segment_reader(0u32).facet_reader("facet").unwrap();
         let mut facet_ords = Vec::new();
-        facet_reader.facet_ords(0u32, &mut facet_ords);
-        assert_eq!(&facet_ords, &[2u64]);
-        facet_reader.facet_ords(1u32, &mut facet_ords);
+        facet_ords.extend(facet_reader.facet_ords(0u32));
+        assert_eq!(&facet_ords, &[0u64]);
+        facet_ords.clear();
+        facet_ords.extend(facet_reader.facet_ords(1u32));
         assert!(facet_ords.is_empty());
         Ok(())
     }
@@ -155,7 +171,7 @@ mod tests {
     #[test]
     fn test_facet_not_populated_for_any_docs() -> crate::Result<()> {
         let mut schema_builder = SchemaBuilder::default();
-        let facet_field = schema_builder.add_facet_field("facet", FacetOptions::default());
+        schema_builder.add_facet_field("facet", FacetOptions::default());
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
         let mut index_writer = index.writer_for_tests()?;
@@ -163,15 +179,9 @@ mod tests {
         index_writer.add_document(Document::default())?;
         index_writer.commit()?;
         let searcher = index.reader()?.searcher();
-        let facet_reader = searcher
-            .segment_reader(0u32)
-            .facet_reader(facet_field)
-            .unwrap();
-        let mut facet_ords = Vec::new();
-        facet_reader.facet_ords(0u32, &mut facet_ords);
-        assert!(facet_ords.is_empty());
-        facet_reader.facet_ords(1u32, &mut facet_ords);
-        assert!(facet_ords.is_empty());
+        let facet_reader = searcher.segment_reader(0u32).facet_reader("facet").unwrap();
+        assert!(facet_reader.facet_ords(0u32).next().is_none());
+        assert!(facet_reader.facet_ords(1u32).next().is_none());
         Ok(())
     }
 }

src/fastfield/multivalued/index.rs

@@ -1,149 +0,0 @@
-use std::ops::Range;
-use std::sync::Arc;
-
-use fastfield_codecs::Column;
-
-use crate::DocId;
-
-#[derive(Clone)]
-/// Index to resolve value range for given doc_id.
-/// Starts at 0.
-pub struct MultiValueIndex {
-    idx: Arc<dyn Column<u64>>,
-}
-
-impl MultiValueIndex {
-    pub(crate) fn new(idx: Arc<dyn Column<u64>>) -> Self {
-        Self { idx }
-    }
-
-    /// Returns `[start, end)`, such that the values associated with
-    /// the given document are `start..end`.
-    #[inline]
-    pub(crate) fn range(&self, doc: DocId) -> Range<u32> {
-        let start = self.idx.get_val(doc) as u32;
-        let end = self.idx.get_val(doc + 1) as u32;
-        start..end
-    }
-
-    /// Given a range of documents, returns the Range of value offsets fo
-    /// these documents.
-    ///
-    /// For instance, `given start_doc..end_doc`,
-    /// if we assume Document #start_doc end #end_doc both
-    /// have values, this function returns `start..end`
-    /// such that `value_column.get(start_doc)` is the first value of
-    /// `start_doc` (well, if there is one), and `value_column.get(end_doc - 1)`
-    /// is the last value of `end_doc`.
-    ///
-    /// The passed end range is allowed to be out of bounds, in which case
-    /// it will be clipped to make it valid.
-    #[inline]
-    pub(crate) fn docid_range_to_position_range(&self, range: Range<DocId>) -> Range<u32> {
-        let end_docid = range.end.min(self.num_docs() - 1) + 1;
-        let start_docid = range.start.min(end_docid);
-
-        let start = self.idx.get_val(start_docid) as u32;
-        let end = self.idx.get_val(end_docid) as u32;
-        assert!(start <= end);
-
-        start..end
-    }
-
-    /// returns the num of values associated with a doc_id
-    pub(crate) fn num_vals_for_doc(&self, doc: DocId) -> u32 {
-        let range = self.range(doc);
-        range.end - range.start
-    }
-
-    /// Returns the overall number of values in this field.
-    #[inline]
-    pub fn total_num_vals(&self) -> u32 {
-        self.idx.max_value() as u32
-    }
-
-    /// Returns the number of documents in the index.
-    #[inline]
-    pub fn num_docs(&self) -> u32 {
-        self.idx.num_vals() - 1
-    }
-
-    /// Converts a list of positions of values in a 1:n index to the corresponding list of 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.
-    ///
-    /// Correctness: positions needs to be sorted. idx_reader needs to contain monotonically
-    /// increasing positions.
-    ///
-    ///
-    /// TODO: Instead of a linear scan we can employ a exponential search into binary search to
-    /// match a docid to its value position.
-    #[allow(clippy::bool_to_int_with_if)]
-    pub(crate) fn positions_to_docids(&self, doc_id_range: Range<u32>, positions: &mut Vec<u32>) {
-        if positions.is_empty() {
-            return;
-        }
-        let mut cur_doc = doc_id_range.start;
-        let mut last_doc = None;
-
-        assert!(self.idx.get_val(doc_id_range.start) as u32 <= positions[0]);
-
-        let mut write_doc_pos = 0;
-        for i in 0..positions.len() {
-            let pos = positions[i];
-            loop {
-                let end = self.idx.get_val(cur_doc + 1) as u32;
-                if end > pos {
-                    positions[write_doc_pos] = cur_doc;
-                    write_doc_pos += if last_doc == Some(cur_doc) { 0 } else { 1 };
-                    last_doc = Some(cur_doc);
-                    break;
-                }
-                cur_doc += 1;
-            }
-        }
-        positions.truncate(write_doc_pos);
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use std::ops::Range;
-    use std::sync::Arc;
-
-    use fastfield_codecs::IterColumn;
-
-    use crate::fastfield::MultiValueIndex;
-
-    fn index_to_pos_helper(
-        index: &MultiValueIndex,
-        doc_id_range: Range<u32>,
-        positions: &[u32],
-    ) -> Vec<u32> {
-        let mut positions = positions.to_vec();
-        index.positions_to_docids(doc_id_range, &mut positions);
-        positions
-    }
-
-    #[test]
-    fn test_positions_to_docid() {
-        let offsets = vec![0, 10, 12, 15, 22, 23]; // docid values are [0..10, 10..12, 12..15, etc.]
-        let column = IterColumn::from(offsets.into_iter());
-        let index = MultiValueIndex::new(Arc::new(column));
-        assert_eq!(index.num_docs(), 5);
-        {
-            let positions = vec![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]);
-            assert_eq!(index_to_pos_helper(&index, 0..5, &[9]), vec![0]);
-            assert_eq!(index_to_pos_helper(&index, 1..5, &[10]), vec![1]);
-            assert_eq!(index_to_pos_helper(&index, 1..5, &[11]), vec![1]);
-            assert_eq!(index_to_pos_helper(&index, 2..5, &[12]), vec![2]);
-            assert_eq!(index_to_pos_helper(&index, 2..5, &[12, 14]), vec![2]);
-            assert_eq!(index_to_pos_helper(&index, 2..5, &[12, 14, 15]), vec![2, 3]);
-        }
-    }
-}

src/fastfield/multivalued/mod.rs

@@ -1,619 +0,0 @@
-mod index;
-mod reader;
-mod writer;
-
-use fastfield_codecs::FastFieldCodecType;
-pub use index::MultiValueIndex;
-
-pub use self::reader::MultiValuedFastFieldReader;
-pub(crate) use self::writer::MultivalueStartIndex;
-pub use self::writer::{MultiValueU128FastFieldWriter, MultiValuedFastFieldWriter};
-
-/// The valid codecs for multivalue values excludes the linear interpolation codec.
-///
-/// This limitation is only valid for the values, not the offset index of the multivalue index.
-pub(crate) fn get_fastfield_codecs_for_multivalue() -> [FastFieldCodecType; 2] {
-    [
-        FastFieldCodecType::Bitpacked,
-        FastFieldCodecType::BlockwiseLinear,
-    ]
-}
-
-#[cfg(test)]
-mod tests {
-    use proptest::strategy::Strategy;
-    use proptest::{prop_oneof, proptest};
-    use test_log::test;
-
-    use crate::collector::TopDocs;
-    use crate::indexer::NoMergePolicy;
-    use crate::query::QueryParser;
-    use crate::schema::{Cardinality, DateOptions, Facet, FacetOptions, NumericOptions, Schema};
-    use crate::time::format_description::well_known::Rfc3339;
-    use crate::time::{Duration, OffsetDateTime};
-    use crate::{DateTime, Document, Index, Term};
-
-    #[test]
-    fn test_multivalued_u64() -> crate::Result<()> {
-        let mut schema_builder = Schema::builder();
-        let field = schema_builder.add_u64_field(
-            "multifield",
-            NumericOptions::default().set_fast(Cardinality::MultiValues),
-        );
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
-        index_writer.add_document(doc!(field=>1u64, field=>3u64))?;
-        index_writer.add_document(doc!())?;
-        index_writer.add_document(doc!(field=>4u64))?;
-        index_writer.add_document(doc!(field=>5u64, field=>20u64,field=>1u64))?;
-        index_writer.commit()?;
-
-        let searcher = index.reader()?.searcher();
-        let segment_reader = searcher.segment_reader(0);
-        let mut vals = Vec::new();
-        let multi_value_reader = segment_reader.fast_fields().u64s("multifield")?;
-        {
-            multi_value_reader.get_vals(2, &mut vals);
-            assert_eq!(&vals, &[4u64]);
-        }
-        {
-            multi_value_reader.get_vals(0, &mut vals);
-            assert_eq!(&vals, &[1u64, 3u64]);
-        }
-        {
-            multi_value_reader.get_vals(1, &mut vals);
-            assert!(vals.is_empty());
-        }
-        Ok(())
-    }
-
-    #[test]
-    fn test_multivalued_date() -> crate::Result<()> {
-        let mut schema_builder = Schema::builder();
-        let date_field = schema_builder.add_date_field(
-            "multi_date_field",
-            DateOptions::default()
-                .set_fast(Cardinality::MultiValues)
-                .set_indexed()
-                .set_fieldnorm()
-                .set_stored(),
-        );
-        let time_i =
-            schema_builder.add_i64_field("time_stamp_i", NumericOptions::default().set_stored());
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
-        let first_time_stamp = OffsetDateTime::now_utc();
-        index_writer.add_document(doc!(
-                date_field => DateTime::from_utc(first_time_stamp),
-                date_field => DateTime::from_utc(first_time_stamp),
-                time_i=>1i64))?;
-        index_writer.add_document(doc!(time_i => 0i64))?;
-        // add one second
-        index_writer.add_document(doc!(
-            date_field => DateTime::from_utc(first_time_stamp + Duration::seconds(1)),
-            time_i => 2i64))?;
-        // add another second
-        let two_secs_ahead = first_time_stamp + Duration::seconds(2);
-        index_writer.add_document(doc!(
-            date_field => DateTime::from_utc(two_secs_ahead),
-            date_field => DateTime::from_utc(two_secs_ahead),
-            date_field => DateTime::from_utc(two_secs_ahead),
-            time_i => 3i64))?;
-        // add three seconds
-        index_writer.add_document(doc!(
-                date_field => DateTime::from_utc(first_time_stamp + Duration::seconds(3)),
-                time_i => 4i64))?;
-        index_writer.commit()?;
-
-        let reader = index.reader()?;
-        let searcher = reader.searcher();
-        let reader = searcher.segment_reader(0);
-        assert_eq!(reader.num_docs(), 5);
-
-        {
-            let parser = QueryParser::for_index(&index, vec![]);
-            let query = parser.parse_query(&format!(
-                "multi_date_field:\"{}\"",
-                first_time_stamp.format(&Rfc3339)?,
-            ))?;
-            let results = searcher.search(&query, &TopDocs::with_limit(5))?;
-            assert_eq!(results.len(), 1);
-            for (_score, doc_address) in results {
-                let retrieved_doc = searcher.doc(doc_address)?;
-                assert_eq!(
-                    retrieved_doc
-                        .get_first(date_field)
-                        .expect("cannot find value")
-                        .as_date()
-                        .unwrap(),
-                    DateTime::from_utc(first_time_stamp),
-                );
-                assert_eq!(
-                    retrieved_doc
-                        .get_first(time_i)
-                        .expect("cannot find value")
-                        .as_i64(),
-                    Some(1i64)
-                );
-            }
-        }
-
-        {
-            let parser = QueryParser::for_index(&index, vec![date_field]);
-            let query = parser.parse_query(&format!("\"{}\"", two_secs_ahead.format(&Rfc3339)?))?;
-            let results = searcher.search(&query, &TopDocs::with_limit(5))?;
-
-            assert_eq!(results.len(), 1);
-
-            for (_score, doc_address) in results {
-                let retrieved_doc = searcher.doc(doc_address).expect("cannot fetch doc");
-                assert_eq!(
-                    retrieved_doc
-                        .get_first(date_field)
-                        .expect("cannot find value")
-                        .as_date()
-                        .unwrap(),
-                    DateTime::from_utc(two_secs_ahead)
-                );
-                assert_eq!(
-                    retrieved_doc
-                        .get_first(time_i)
-                        .expect("cannot find value")
-                        .as_i64(),
-                    Some(3i64)
-                );
-            }
-        }
-
-        {
-            let parser = QueryParser::for_index(&index, vec![date_field]);
-            let range_q = format!(
-                "multi_date_field:[{} TO {}}}",
-                (first_time_stamp + Duration::seconds(1)).format(&Rfc3339)?,
-                (first_time_stamp + Duration::seconds(3)).format(&Rfc3339)?
-            );
-            let query = parser.parse_query(&range_q)?;
-            let results = searcher.search(&query, &TopDocs::with_limit(5))?;
-
-            assert_eq!(results.len(), 2);
-            for (i, doc_pair) in results.iter().enumerate() {
-                let retrieved_doc = searcher.doc(doc_pair.1).expect("cannot fetch doc");
-                let offset_sec = match i {
-                    0 => 1,
-                    1 => 2,
-                    _ => panic!("should not have more than 2 docs"),
-                };
-                let time_i_val = match i {
-                    0 => 2,
-                    1 => 3,
-                    _ => panic!("should not have more than 2 docs"),
-                };
-                assert_eq!(
-                    retrieved_doc
-                        .get_first(date_field)
-                        .expect("cannot find value")
-                        .as_date()
-                        .expect("value not of Date type"),
-                    DateTime::from_utc(first_time_stamp + Duration::seconds(offset_sec)),
-                );
-                assert_eq!(
-                    retrieved_doc
-                        .get_first(time_i)
-                        .expect("cannot find value")
-                        .as_i64(),
-                    Some(time_i_val)
-                );
-            }
-        }
-        Ok(())
-    }
-
-    #[test]
-    fn test_multivalued_i64() -> crate::Result<()> {
-        let mut schema_builder = Schema::builder();
-        let field = schema_builder.add_i64_field(
-            "multifield",
-            NumericOptions::default().set_fast(Cardinality::MultiValues),
-        );
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
-        index_writer.add_document(doc!(field=> 1i64, field => 3i64))?;
-        index_writer.add_document(doc!())?;
-        index_writer.add_document(doc!(field=> -4i64))?;
-        index_writer.add_document(doc!(field=> -5i64, field => -20i64, field=>1i64))?;
-        index_writer.commit()?;
-
-        let searcher = index.reader()?.searcher();
-        let segment_reader = searcher.segment_reader(0);
-        let mut vals = Vec::new();
-        let multi_value_reader = segment_reader.fast_fields().i64s("multifield").unwrap();
-        multi_value_reader.get_vals(2, &mut vals);
-        assert_eq!(&vals, &[-4i64]);
-        multi_value_reader.get_vals(0, &mut vals);
-        assert_eq!(&vals, &[1i64, 3i64]);
-        multi_value_reader.get_vals(1, &mut vals);
-        assert!(vals.is_empty());
-        multi_value_reader.get_vals(3, &mut vals);
-        assert_eq!(&vals, &[-5i64, -20i64, 1i64]);
-        Ok(())
-    }
-
-    #[test]
-    fn test_multivalued_bool() -> crate::Result<()> {
-        let mut schema_builder = Schema::builder();
-        let bool_field = schema_builder.add_bool_field(
-            "multifield",
-            NumericOptions::default().set_fast(Cardinality::MultiValues),
-        );
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
-        index_writer.add_document(doc!(bool_field=> true, bool_field => false))?;
-        index_writer.add_document(doc!())?;
-        index_writer.add_document(doc!(bool_field=> false))?;
-        index_writer
-            .add_document(doc!(bool_field=> true, bool_field => true, bool_field => false))?;
-        index_writer.commit()?;
-
-        let searcher = index.reader()?.searcher();
-        let segment_reader = searcher.segment_reader(0);
-        let mut vals = Vec::new();
-        let multi_value_reader = segment_reader.fast_fields().bools("multifield").unwrap();
-        multi_value_reader.get_vals(2, &mut vals);
-        assert_eq!(&vals, &[false]);
-        multi_value_reader.get_vals(0, &mut vals);
-        assert_eq!(&vals, &[true, false]);
-        multi_value_reader.get_vals(1, &mut vals);
-        assert!(vals.is_empty());
-        multi_value_reader.get_vals(3, &mut vals);
-        assert_eq!(&vals, &[true, true, false]);
-        Ok(())
-    }
-
-    fn test_multivalued_no_panic(ops: &[IndexingOp]) -> crate::Result<()> {
-        let mut schema_builder = Schema::builder();
-        let field = schema_builder.add_u64_field(
-            "multifield",
-            NumericOptions::default()
-                .set_fast(Cardinality::MultiValues)
-                .set_indexed(),
-        );
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
-        index_writer.set_merge_policy(Box::new(NoMergePolicy));
-
-        for &op in ops {
-            match op {
-                IndexingOp::AddDoc { id } => {
-                    match id % 3 {
-                        0 => {
-                            index_writer.add_document(doc!())?;
-                        }
-                        1 => {
-                            let mut doc = Document::new();
-                            for _ in 0..5001 {
-                                doc.add_u64(field, id as u64);
-                            }
-                            index_writer.add_document(doc)?;
-                        }
-                        _ => {
-                            let mut doc = Document::new();
-                            doc.add_u64(field, id as u64);
-                            index_writer.add_document(doc)?;
-                        }
-                    };
-                }
-                IndexingOp::DeleteDoc { id } => {
-                    index_writer.delete_term(Term::from_field_u64(field, id as u64));
-                }
-                IndexingOp::Commit => {
-                    index_writer.commit().unwrap();
-                }
-                IndexingOp::Merge => {
-                    let segment_ids = index.searchable_segment_ids()?;
-                    if segment_ids.len() >= 2 {
-                        index_writer.merge(&segment_ids).wait()?;
-                        index_writer.segment_updater().wait_merging_thread()?;
-                    }
-                }
-            }
-        }
-
-        index_writer.commit()?;
-
-        // Merging the segments
-        {
-            let segment_ids = index
-                .searchable_segment_ids()
-                .expect("Searchable segments failed.");
-            if !segment_ids.is_empty() {
-                index_writer.merge(&segment_ids).wait()?;
-                assert!(index_writer.wait_merging_threads().is_ok());
-            }
-        }
-        Ok(())
-    }
-
-    #[derive(Debug, Clone, Copy)]
-    enum IndexingOp {
-        AddDoc { id: u32 },
-        DeleteDoc { id: u32 },
-        Commit,
-        Merge,
-    }
-
-    fn operation_strategy() -> impl Strategy<Value = IndexingOp> {
-        prop_oneof![
-            (0u32..10u32).prop_map(|id| IndexingOp::DeleteDoc { id }),
-            (0u32..10u32).prop_map(|id| IndexingOp::AddDoc { id }),
-            (0u32..2u32).prop_map(|_| IndexingOp::Commit),
-            (0u32..1u32).prop_map(|_| IndexingOp::Merge),
-        ]
-    }
-
-    proptest! {
-        #![proptest_config(proptest::prelude::ProptestConfig::with_cases(5))]
-        #[test]
-        fn test_multivalued_proptest(ops in proptest::collection::vec(operation_strategy(), 1..10)) {
-            assert!(test_multivalued_no_panic(&ops[..]).is_ok());
-        }
-    }
-
-    #[test]
-    fn test_multivalued_proptest_gcd() {
-        use IndexingOp::*;
-        let ops = [AddDoc { id: 9 }, AddDoc { id: 9 }, Merge];
-
-        assert!(test_multivalued_no_panic(&ops[..]).is_ok());
-    }
-
-    #[test]
-    fn test_multivalued_proptest_off_by_one_bug_1151() {
-        use IndexingOp::*;
-        let ops = [
-            AddDoc { id: 3 },
-            AddDoc { id: 1 },
-            AddDoc { id: 3 },
-            Commit,
-            Merge,
-        ];
-
-        assert!(test_multivalued_no_panic(&ops[..]).is_ok());
-    }
-
-    #[test]
-    #[ignore]
-    fn test_many_facets() -> crate::Result<()> {
-        let mut schema_builder = Schema::builder();
-        let field = schema_builder.add_facet_field("facetfield", FacetOptions::default());
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
-        for i in 0..100_000 {
-            index_writer
-                .add_document(doc!(field=> Facet::from(format!("/lang/{}", i).as_str())))?;
-        }
-        index_writer.commit()?;
-        Ok(())
-    }
-}
-
-#[cfg(all(test, feature = "unstable"))]
-mod bench {
-    use std::collections::HashMap;
-    use std::path::Path;
-
-    use test::{self, Bencher};
-
-    use super::*;
-    use crate::directory::{CompositeFile, Directory, RamDirectory, WritePtr};
-    use crate::fastfield::{CompositeFastFieldSerializer, FastFieldsWriter};
-    use crate::indexer::doc_id_mapping::DocIdMapping;
-    use crate::schema::{Cardinality, NumericOptions, Schema};
-    use crate::Document;
-
-    fn bench_multi_value_ff_merge_opt(
-        num_docs: usize,
-        segments_every_n_docs: usize,
-        merge_policy: impl crate::indexer::MergePolicy + 'static,
-    ) {
-        let mut builder = crate::schema::SchemaBuilder::new();
-
-        let fast_multi =
-            crate::schema::NumericOptions::default().set_fast(Cardinality::MultiValues);
-        let multi_field = builder.add_f64_field("f64s", fast_multi);
-
-        let index = crate::Index::create_in_ram(builder.build());
-
-        let mut writer = index.writer_for_tests().unwrap();
-        writer.set_merge_policy(Box::new(merge_policy));
-
-        for i in 0..num_docs {
-            let mut doc = crate::Document::new();
-            doc.add_f64(multi_field, 0.24);
-            doc.add_f64(multi_field, 0.27);
-            doc.add_f64(multi_field, 0.37);
-            if i % 3 == 0 {
-                doc.add_f64(multi_field, 0.44);
-            }
-
-            writer.add_document(doc).unwrap();
-            if i % segments_every_n_docs == 0 {
-                writer.commit().unwrap();
-            }
-        }
-
-        {
-            writer.wait_merging_threads().unwrap();
-            let mut writer = index.writer_for_tests().unwrap();
-            let segment_ids = index.searchable_segment_ids().unwrap();
-            writer.merge(&segment_ids).wait().unwrap();
-        }
-
-        // If a merging thread fails, we should end up with more
-        // than one segment here
-        assert_eq!(1, index.searchable_segments().unwrap().len());
-    }
-
-    #[bench]
-    fn bench_multi_value_ff_merge_many_segments(b: &mut Bencher) {
-        let num_docs = 100_000;
-        b.iter(|| {
-            bench_multi_value_ff_merge_opt(num_docs, 1_000, crate::indexer::NoMergePolicy);
-        });
-    }
-
-    #[bench]
-    fn bench_multi_value_ff_merge_many_segments_log_merge(b: &mut Bencher) {
-        let num_docs = 100_000;
-        b.iter(|| {
-            let merge_policy = crate::indexer::LogMergePolicy::default();
-            bench_multi_value_ff_merge_opt(num_docs, 1_000, merge_policy);
-        });
-    }
-
-    #[bench]
-    fn bench_multi_value_ff_merge_few_segments(b: &mut Bencher) {
-        let num_docs = 100_000;
-        b.iter(|| {
-            bench_multi_value_ff_merge_opt(num_docs, 33_000, crate::indexer::NoMergePolicy);
-        });
-    }
-
-    fn multi_values(num_docs: usize, vals_per_doc: usize) -> Vec<Vec<u64>> {
-        let mut vals = vec![];
-        for _i in 0..num_docs {
-            let mut block = vec![];
-            for j in 0..vals_per_doc {
-                block.push(j as u64);
-            }
-            vals.push(block);
-        }
-
-        vals
-    }
-
-    #[bench]
-    fn bench_multi_value_fflookup(b: &mut Bencher) {
-        let num_docs = 100_000;
-
-        let path = Path::new("test");
-        let directory: RamDirectory = RamDirectory::create();
-        let field = {
-            let options = NumericOptions::default().set_fast(Cardinality::MultiValues);
-            let mut schema_builder = Schema::builder();
-            let field = schema_builder.add_u64_field("field", options);
-            let schema = schema_builder.build();
-
-            let write: WritePtr = directory.open_write(Path::new("test")).unwrap();
-            let mut serializer = CompositeFastFieldSerializer::from_write(write).unwrap();
-            let mut fast_field_writers = FastFieldsWriter::from_schema(&schema);
-            for block in &multi_values(num_docs, 3) {
-                let mut doc = Document::new();
-                for val in block {
-                    doc.add_u64(field, *val);
-                }
-                fast_field_writers.add_document(&doc).unwrap();
-            }
-            fast_field_writers
-                .serialize(&mut serializer, &HashMap::new(), None)
-                .unwrap();
-            serializer.close().unwrap();
-            field
-        };
-        let file = directory.open_read(path).unwrap();
-        {
-            let fast_fields_composite = CompositeFile::open(&file).unwrap();
-            let data_idx = fast_fields_composite
-                .open_read_with_idx(field, 0)
-                .unwrap()
-                .read_bytes()
-                .unwrap();
-            let idx_reader = fastfield_codecs::open(data_idx).unwrap();
-
-            let data_vals = fast_fields_composite
-                .open_read_with_idx(field, 1)
-                .unwrap()
-                .read_bytes()
-                .unwrap();
-            let vals_reader = fastfield_codecs::open(data_vals).unwrap();
-            let fast_field_reader = MultiValuedFastFieldReader::open(idx_reader, vals_reader);
-            b.iter(|| {
-                let mut sum = 0u64;
-                let mut data = Vec::with_capacity(10);
-                for i in 0u32..num_docs as u32 {
-                    fast_field_reader.get_vals(i, &mut data);
-                    sum += data.iter().sum::<u64>();
-                }
-                sum
-            });
-        }
-    }
-
-    #[bench]
-    fn bench_multi_value_ff_creation(b: &mut Bencher) {
-        // 3 million ff entries
-        let num_docs = 1_000_000;
-        let multi_values = multi_values(num_docs, 3);
-
-        b.iter(|| {
-            let directory: RamDirectory = RamDirectory::create();
-            let options = NumericOptions::default().set_fast(Cardinality::MultiValues);
-            let mut schema_builder = Schema::builder();
-            let field = schema_builder.add_u64_field("field", options);
-            let schema = schema_builder.build();
-
-            let write: WritePtr = directory.open_write(Path::new("test")).unwrap();
-            let mut serializer = CompositeFastFieldSerializer::from_write(write).unwrap();
-            let mut fast_field_writers = FastFieldsWriter::from_schema(&schema);
-            for block in &multi_values {
-                let mut doc = Document::new();
-                for val in block {
-                    doc.add_u64(field, *val);
-                }
-                fast_field_writers.add_document(&doc).unwrap();
-            }
-            fast_field_writers
-                .serialize(&mut serializer, &HashMap::new(), None)
-                .unwrap();
-            serializer.close().unwrap();
-        });
-    }
-
-    #[bench]
-    fn bench_multi_value_ff_creation_with_sorting(b: &mut Bencher) {
-        // 3 million ff entries
-        let num_docs = 1_000_000;
-        let multi_values = multi_values(num_docs, 3);
-
-        let doc_id_mapping =
-            DocIdMapping::from_new_id_to_old_id((0..1_000_000).collect::<Vec<_>>());
-
-        b.iter(|| {
-            let directory: RamDirectory = RamDirectory::create();
-            let options = NumericOptions::default().set_fast(Cardinality::MultiValues);
-            let mut schema_builder = Schema::builder();
-            let field = schema_builder.add_u64_field("field", options);
-            let schema = schema_builder.build();
-
-            let write: WritePtr = directory.open_write(Path::new("test")).unwrap();
-            let mut serializer = CompositeFastFieldSerializer::from_write(write).unwrap();
-            let mut fast_field_writers = FastFieldsWriter::from_schema(&schema);
-            for block in &multi_values {
-                let mut doc = Document::new();
-                for val in block {
-                    doc.add_u64(field, *val);
-                }
-                fast_field_writers.add_document(&doc).unwrap();
-            }
-            fast_field_writers
-                .serialize(&mut serializer, &HashMap::new(), Some(&doc_id_mapping))
-                .unwrap();
-            serializer.close().unwrap();
-        });
-    }
-}

src/fastfield/multivalued/reader.rs

@@ -1,333 +0,0 @@
-use core::fmt;
-use std::ops::{Range, RangeInclusive};
-use std::sync::Arc;
-
-use fastfield_codecs::Column;
-
-use super::MultiValueIndex;
-use crate::fastfield::MakeZero;
-use crate::DocId;
-
-/// Reader for a multivalued fast field.
-///
-/// The reader is implemented as two fast fields, one u64 fast field for the index and one for the
-/// values.
-///
-/// The `vals_reader` will access the concatenated list of all values.
-/// The `idx_reader` associates, for each document, the index of its first value.
-#[derive(Clone)]
-pub struct MultiValuedFastFieldReader<T> {
-    idx_reader: MultiValueIndex,
-    vals_reader: Arc<dyn Column<T>>,
-}
-
-impl<T: PartialOrd + MakeZero + Copy + fmt::Debug> MultiValuedFastFieldReader<T> {
-    pub(crate) fn open(
-        idx_reader: Arc<dyn Column<u64>>,
-        vals_reader: Arc<dyn Column<T>>,
-    ) -> MultiValuedFastFieldReader<T> {
-        Self {
-            idx_reader: MultiValueIndex::new(idx_reader),
-            vals_reader,
-        }
-    }
-
-    /// Returns the array of values associated to the given `doc`.
-    #[inline]
-    pub fn get_first_val(&self, doc: DocId) -> Option<T> {
-        let range = self.idx_reader.range(doc);
-        if range.is_empty() {
-            return None;
-        }
-        Some(self.vals_reader.get_val(range.start))
-    }
-
-    /// Returns the array of values associated to the given `doc`.
-    #[inline]
-    fn get_vals_for_range(&self, range: Range<u32>, vals: &mut Vec<T>) {
-        let len = (range.end - range.start) as usize;
-        vals.resize(len, T::make_zero());
-        self.vals_reader
-            .get_range(range.start as u64, &mut vals[..]);
-    }
-
-    /// Returns the index reader
-    pub fn get_index_reader(&self) -> &MultiValueIndex {
-        &self.idx_reader
-    }
-
-    /// Returns the array of values associated to the given `doc`.
-    #[inline]
-    pub fn get_vals(&self, doc: DocId, vals: &mut Vec<T>) {
-        let range = self.idx_reader.range(doc);
-        self.get_vals_for_range(range, vals);
-    }
-
-    /// Iterates over all elements in the fast field
-    pub fn iter(&self) -> impl Iterator<Item = T> + '_ {
-        self.vals_reader.iter()
-    }
-
-    /// Returns the minimum value for this fast field.
-    ///
-    /// The min value does not take in account of possible
-    /// deleted document, and should be considered as a lower bound
-    /// of the actual mimimum value.
-    pub fn min_value(&self) -> T {
-        self.vals_reader.min_value()
-    }
-
-    /// Returns the maximum value for this fast field.
-    ///
-    /// The max value does not take in account of possible
-    /// deleted document, and should be considered as an upper bound
-    /// of the actual maximum value.
-    pub fn max_value(&self) -> T {
-        self.vals_reader.max_value()
-    }
-
-    /// Returns the number of values associated with the document `DocId`.
-    #[inline]
-    pub fn num_vals(&self, doc: DocId) -> u32 {
-        self.idx_reader.num_vals_for_doc(doc)
-    }
-
-    /// Returns the overall number of values in this field. It does not include deletes.
-    #[inline]
-    pub fn total_num_vals(&self) -> u32 {
-        assert_eq!(
-            self.vals_reader.num_vals(),
-            self.get_index_reader().total_num_vals()
-        );
-        self.idx_reader.total_num_vals()
-    }
-
-    /// Returns the docids matching given doc_id_range and value_range.
-    #[inline]
-    pub fn get_docids_for_value_range(
-        &self,
-        value_range: RangeInclusive<T>,
-        doc_id_range: Range<u32>,
-        positions: &mut Vec<u32>,
-    ) {
-        let position_range = self
-            .get_index_reader()
-            .docid_range_to_position_range(doc_id_range.clone());
-        self.vals_reader
-            .get_docids_for_value_range(value_range, position_range, positions);
-
-        self.idx_reader.positions_to_docids(doc_id_range, positions);
-    }
-}
-
-#[cfg(test)]
-mod tests {
-
-    use time::{Duration, OffsetDateTime};
-
-    use crate::collector::Count;
-    use crate::core::Index;
-    use crate::query::RangeQuery;
-    use crate::schema::{Cardinality, Facet, FacetOptions, NumericOptions, Schema};
-    use crate::{DateOptions, DatePrecision, DateTime};
-
-    #[test]
-    fn test_multivalued_date_docids_for_value_range_1() -> crate::Result<()> {
-        let mut schema_builder = Schema::builder();
-        let date_field = schema_builder.add_date_field(
-            "multi_date_field",
-            DateOptions::default()
-                .set_fast(Cardinality::MultiValues)
-                .set_indexed()
-                .set_fieldnorm()
-                .set_precision(DatePrecision::Microseconds)
-                .set_stored(),
-        );
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
-        let first_time_stamp = OffsetDateTime::now_utc();
-        index_writer.add_document(doc!(
-            date_field => DateTime::from_utc(first_time_stamp),
-            date_field => DateTime::from_utc(first_time_stamp),
-        ))?;
-        // add another second
-        let two_secs_ahead = first_time_stamp + Duration::seconds(2);
-        index_writer.commit()?;
-
-        let reader = index.reader()?;
-        let searcher = reader.searcher();
-        let reader = searcher.segment_reader(0);
-
-        let date_ff_reader = reader.fast_fields().dates("multi_date_field").unwrap();
-        let mut docids = vec![];
-        date_ff_reader.get_docids_for_value_range(
-            DateTime::from_utc(first_time_stamp)..=DateTime::from_utc(two_secs_ahead),
-            0..5,
-            &mut docids,
-        );
-        assert_eq!(docids, vec![0]);
-
-        let count_multiples =
-            |range_query: RangeQuery| searcher.search(&range_query, &Count).unwrap();
-
-        assert_eq!(
-            count_multiples(RangeQuery::new_date(
-                "multi_date_field".to_string(),
-                DateTime::from_utc(first_time_stamp)..DateTime::from_utc(two_secs_ahead)
-            )),
-            1
-        );
-
-        Ok(())
-    }
-
-    #[test]
-    fn test_multivalued_date_docids_for_value_range_2() -> crate::Result<()> {
-        let mut schema_builder = Schema::builder();
-        let date_field = schema_builder.add_date_field(
-            "multi_date_field",
-            DateOptions::default()
-                .set_fast(Cardinality::MultiValues)
-                // TODO: Test different precision after fixing https://github.com/quickwit-oss/tantivy/issues/1783
-                .set_precision(DatePrecision::Microseconds)
-                .set_indexed()
-                .set_fieldnorm()
-                .set_stored(),
-        );
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
-        let first_time_stamp = OffsetDateTime::now_utc();
-        index_writer.add_document(doc!(
-            date_field => DateTime::from_utc(first_time_stamp),
-            date_field => DateTime::from_utc(first_time_stamp),
-        ))?;
-        index_writer.add_document(doc!())?;
-        // add one second
-        index_writer.add_document(doc!(
-            date_field => DateTime::from_utc(first_time_stamp + Duration::seconds(1)),
-        ))?;
-        // add another second
-        let two_secs_ahead = first_time_stamp + Duration::seconds(2);
-        index_writer.add_document(doc!(
-            date_field => DateTime::from_utc(two_secs_ahead),
-            date_field => DateTime::from_utc(two_secs_ahead),
-            date_field => DateTime::from_utc(two_secs_ahead),
-        ))?;
-        // add three seconds
-        index_writer.add_document(doc!(
-            date_field => DateTime::from_utc(first_time_stamp + Duration::seconds(3)),
-        ))?;
-        index_writer.commit()?;
-
-        let reader = index.reader()?;
-        let searcher = reader.searcher();
-        let reader = searcher.segment_reader(0);
-        assert_eq!(reader.num_docs(), 5);
-
-        let date_ff_reader = reader.fast_fields().dates("multi_date_field").unwrap();
-        let mut docids = vec![];
-        date_ff_reader.get_docids_for_value_range(
-            DateTime::from_utc(first_time_stamp)..=DateTime::from_utc(two_secs_ahead),
-            0..5,
-            &mut docids,
-        );
-        assert_eq!(docids, vec![0, 2, 3]);
-
-        let count_multiples =
-            |range_query: RangeQuery| searcher.search(&range_query, &Count).unwrap();
-
-        assert_eq!(
-            count_multiples(RangeQuery::new_date(
-                "multi_date_field".to_string(),
-                DateTime::from_utc(first_time_stamp)..DateTime::from_utc(two_secs_ahead)
-            )),
-            2
-        );
-
-        Ok(())
-    }
-
-    #[test]
-    fn test_multifastfield_reader() -> crate::Result<()> {
-        let mut schema_builder = Schema::builder();
-        let facet_field = schema_builder.add_facet_field("facets", FacetOptions::default());
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests()?;
-        index_writer.add_document(doc!(
-            facet_field => Facet::from("/category/cat2"),
-            facet_field => Facet::from("/category/cat1"),
-        ))?;
-        index_writer.add_document(doc!(facet_field => Facet::from("/category/cat2")))?;
-        index_writer.add_document(doc!(facet_field => Facet::from("/category/cat3")))?;
-        index_writer.commit()?;
-        let searcher = index.reader()?.searcher();
-        let segment_reader = searcher.segment_reader(0);
-        let mut facet_reader = segment_reader.facet_reader(facet_field)?;
-
-        let mut facet = Facet::root();
-        {
-            facet_reader.facet_from_ord(1, &mut facet).unwrap();
-            assert_eq!(facet, Facet::from("/category"));
-        }
-        {
-            facet_reader.facet_from_ord(2, &mut facet).unwrap();
-            assert_eq!(facet, Facet::from("/category/cat1"));
-        }
-        {
-            facet_reader.facet_from_ord(3, &mut facet).unwrap();
-            assert_eq!(format!("{}", facet), "/category/cat2");
-            assert_eq!(facet, Facet::from("/category/cat2"));
-        }
-        {
-            facet_reader.facet_from_ord(4, &mut facet).unwrap();
-            assert_eq!(facet, Facet::from("/category/cat3"));
-        }
-
-        let mut vals = Vec::new();
-        {
-            facet_reader.facet_ords(0, &mut vals);
-            assert_eq!(&vals[..], &[2, 3]);
-        }
-        {
-            facet_reader.facet_ords(1, &mut vals);
-            assert_eq!(&vals[..], &[3]);
-        }
-        {
-            facet_reader.facet_ords(2, &mut vals);
-            assert_eq!(&vals[..], &[4]);
-        }
-        Ok(())
-    }
-
-    #[test]
-    fn test_multifastfield_reader_min_max() -> crate::Result<()> {
-        let mut schema_builder = Schema::builder();
-        let field_options = NumericOptions::default()
-            .set_indexed()
-            .set_fast(Cardinality::MultiValues);
-        let item_field = schema_builder.add_i64_field("items", field_options);
-        let schema = schema_builder.build();
-        let index = Index::create_in_ram(schema);
-        let mut index_writer = index
-            .writer_for_tests()
-            .expect("Failed to create index writer.");
-        index_writer.add_document(doc!(
-            item_field => 2i64,
-            item_field => 3i64,
-            item_field => -2i64,
-        ))?;
-        index_writer.add_document(doc!(item_field => 6i64, item_field => 3i64))?;
-        index_writer.add_document(doc!(item_field => 4i64))?;
-        index_writer.commit()?;
-        let searcher = index.reader()?.searcher();
-        let segment_reader = searcher.segment_reader(0);
-        let field_reader = segment_reader.fast_fields().i64s("items")?;
-
-        assert_eq!(field_reader.min_value(), -2);
-        assert_eq!(field_reader.max_value(), 6);
-        Ok(())
-    }
-}

src/fastfield/multivalued/writer.rs

@@ -1,442 +0,0 @@
-use std::io;
-
-use fastfield_codecs::{
-    Column, MonotonicallyMappableToU128, MonotonicallyMappableToU64, VecColumn,
-};
-use rustc_hash::FxHashMap;
-
-use super::get_fastfield_codecs_for_multivalue;
-use crate::fastfield::writer::unexpected_value;
-use crate::fastfield::{value_to_u64, CompositeFastFieldSerializer, FastFieldType};
-use crate::indexer::doc_id_mapping::DocIdMapping;
-use crate::postings::UnorderedTermId;
-use crate::schema::{Document, Field, Value};
-use crate::termdict::TermOrdinal;
-use crate::{DatePrecision, DocId};
-
-/// Writer for multi-valued (as in, more than one value per document)
-/// int fast field.
-///
-/// This `Writer` is only useful for advanced users.
-/// The normal way to get your multivalued int in your index
-/// is to
-/// - declare your field with fast set to
-///   [`Cardinality::MultiValues`](crate::schema::Cardinality::MultiValues) in your schema
-/// - add your document simply by calling `.add_document(...)`.
-///
-/// The `MultiValuedFastFieldWriter` can be acquired from the fastfield writer, by calling
-/// [`FastFieldWriter::get_multivalue_writer_mut()`](crate::fastfield::FastFieldsWriter::get_multivalue_writer_mut).
-///
-/// Once acquired, writing is done by calling
-/// [`.add_document(&Document)`](MultiValuedFastFieldWriter::add_document) once per value.
-///
-/// The serializer makes it possible to remap all of the values
-/// that were pushed to the writer using a mapping.
-/// This makes it possible to push unordered term ids,
-/// during indexing and remap them to their respective
-/// term ids when the segment is getting serialized.
-pub struct MultiValuedFastFieldWriter {
-    field: Field,
-    precision_opt: Option<DatePrecision>,
-    vals: Vec<UnorderedTermId>,
-    doc_index: Vec<u64>,
-    fast_field_type: FastFieldType,
-}
-
-impl MultiValuedFastFieldWriter {
-    /// Creates a new `MultiValuedFastFieldWriter`
-    pub(crate) fn new(
-        field: Field,
-        fast_field_type: FastFieldType,
-        precision_opt: Option<DatePrecision>,
-    ) -> Self {
-        MultiValuedFastFieldWriter {
-            field,
-            precision_opt,
-            vals: Vec::new(),
-            doc_index: Vec::new(),
-            fast_field_type,
-        }
-    }
-
-    /// The memory used (inclusive childs)
-    pub fn mem_usage(&self) -> usize {
-        self.vals.capacity() * std::mem::size_of::<UnorderedTermId>()
-            + self.doc_index.capacity() * std::mem::size_of::<u64>()
-    }
-
-    /// Access the field associated with the `MultiValuedFastFieldWriter`
-    pub fn field(&self) -> Field {
-        self.field
-    }
-
-    /// Finalize the current document.
-    pub(crate) fn next_doc(&mut self) {
-        self.doc_index.push(self.vals.len() as u64);
-    }
-
-    /// Pushes a new value to the current document.
-    pub(crate) fn add_val(&mut self, val: UnorderedTermId) {
-        self.vals.push(val);
-    }
-
-    /// Shift to the next document and adds
-    /// all of the matching field values present in the document.
-    pub fn add_document(&mut self, doc: &Document) -> crate::Result<()> {
-        self.next_doc();
-        // facets/texts are indexed in the `SegmentWriter` as we encode their unordered id.
-        if self.fast_field_type.is_storing_term_ids() {
-            return Ok(());
-        }
-        for field_value in doc.field_values() {
-            if field_value.field == self.field {
-                let value = field_value.value();
-                let value_u64 = match (self.precision_opt, value) {
-                    (Some(precision), Value::Date(date_val)) => {
-                        date_val.truncate(precision).to_u64()
-                    }
-                    _ => value_to_u64(value)?,
-                };
-                self.add_val(value_u64);
-            }
-        }
-        Ok(())
-    }
-
-    /// Returns an iterator over values per doc_id in ascending doc_id order.
-    ///
-    /// Normally the order is simply iterating self.doc_id_index.
-    /// With doc_id_map it accounts for the new mapping, returning values in the order of the
-    /// new doc_ids.
-    fn get_ordered_values<'a: 'b, 'b>(
-        &'a self,
-        doc_id_map: Option<&'b DocIdMapping>,
-    ) -> impl Iterator<Item = &'b [u64]> {
-        let doc_id_iter: Box<dyn Iterator<Item = u32>> = if let Some(doc_id_map) = doc_id_map {
-            Box::new(doc_id_map.iter_old_doc_ids())
-        } else {
-            let max_doc = self.doc_index.len() as DocId;
-            Box::new(0..max_doc)
-        };
-        doc_id_iter.map(move |doc_id| self.get_values_for_doc_id(doc_id))
-    }
-
-    /// returns all values for a doc_ids
-    fn get_values_for_doc_id(&self, doc_id: u32) -> &[u64] {
-        let start_pos = self.doc_index[doc_id as usize] as usize;
-        let end_pos = self
-            .doc_index
-            .get(doc_id as usize + 1)
-            .cloned()
-            .unwrap_or(self.vals.len() as u64) as usize; // special case, last doc_id has no offset information
-        &self.vals[start_pos..end_pos]
-    }
-    /// Serializes fast field values by pushing them to the `FastFieldSerializer`.
-    ///
-    /// If a mapping is given, the values are remapped *and sorted* before serialization.
-    /// This is used when serializing `facets`. Specifically their terms are
-    /// first stored in the writer as their position in the `IndexWriter`'s `HashMap`.
-    /// This value is called an `UnorderedTermId`.
-    ///
-    /// During the serialization of the segment, terms gets sorted and
-    /// `tantivy` builds a mapping to convert this `UnorderedTermId` into
-    /// term ordinals.
-    pub fn serialize(
-        mut self,
-        serializer: &mut CompositeFastFieldSerializer,
-        term_mapping_opt: Option<&FxHashMap<UnorderedTermId, TermOrdinal>>,
-        doc_id_map: Option<&DocIdMapping>,
-    ) -> io::Result<()> {
-        {
-            self.doc_index.push(self.vals.len() as u64);
-            let col = VecColumn::from(&self.doc_index[..]);
-            if let Some(doc_id_map) = doc_id_map {
-                let multi_value_start_index = MultivalueStartIndex::new(&col, doc_id_map);
-                serializer.create_auto_detect_u64_fast_field_with_idx(
-                    self.field,
-                    multi_value_start_index,
-                    0,
-                )?;
-            } else {
-                serializer.create_auto_detect_u64_fast_field_with_idx(self.field, col, 0)?;
-            }
-        }
-        {
-            // Writing the values themselves.
-            // TODO FIXME: Use less memory.
-            let mut values: Vec<u64> = Vec::new();
-            if let Some(term_mapping) = term_mapping_opt {
-                if self.fast_field_type.is_facet() {
-                    let mut doc_vals: Vec<u64> = Vec::with_capacity(100);
-                    for vals in self.get_ordered_values(doc_id_map) {
-                        // In the case of facets, we want a vec of facet ord that is sorted.
-                        doc_vals.clear();
-                        let remapped_vals = vals
-                            .iter()
-                            .map(|val| *term_mapping.get(val).expect("Missing term ordinal"));
-                        doc_vals.extend(remapped_vals);
-                        doc_vals.sort_unstable();
-                        for &val in &doc_vals {
-                            values.push(val);
-                        }
-                    }
-                } else {
-                    for vals in self.get_ordered_values(doc_id_map) {
-                        let remapped_vals = vals
-                            .iter()
-                            .map(|val| *term_mapping.get(val).expect("Missing term ordinal"));
-                        for val in remapped_vals {
-                            values.push(val);
-                        }
-                    }
-                }
-            } else {
-                for vals in self.get_ordered_values(doc_id_map) {
-                    // sort values in case of remapped doc_ids?
-                    for &val in vals {
-                        values.push(val);
-                    }
-                }
-            }
-            let col = VecColumn::from(&values[..]);
-            serializer.create_auto_detect_u64_fast_field_with_idx_and_codecs(
-                self.field,
-                col,
-                1,
-                &get_fastfield_codecs_for_multivalue(),
-            )?;
-        }
-        Ok(())
-    }
-}
-
-pub(crate) struct MultivalueStartIndex<'a, C: Column> {
-    column: &'a C,
-    doc_id_map: &'a DocIdMapping,
-    min: u64,
-    max: u64,
-}
-
-impl<'a, C: Column> MultivalueStartIndex<'a, C> {
-    pub fn new(column: &'a C, doc_id_map: &'a DocIdMapping) -> Self {
-        assert_eq!(column.num_vals(), doc_id_map.num_old_doc_ids() as u32 + 1);
-        let (min, max) =
-            tantivy_bitpacker::minmax(iter_remapped_multivalue_index(doc_id_map, column))
-                .unwrap_or((0u64, 0u64));
-        MultivalueStartIndex {
-            column,
-            doc_id_map,
-            min,
-            max,
-        }
-    }
-}
-impl<'a, C: Column> Column for MultivalueStartIndex<'a, C> {
-    fn get_val(&self, _idx: u32) -> u64 {
-        unimplemented!()
-    }
-
-    fn min_value(&self) -> u64 {
-        self.min
-    }
-
-    fn max_value(&self) -> u64 {
-        self.max
-    }
-
-    fn num_vals(&self) -> u32 {
-        (self.doc_id_map.num_new_doc_ids() + 1) as u32
-    }
-
-    fn iter(&self) -> Box<dyn Iterator<Item = u64> + '_> {
-        Box::new(iter_remapped_multivalue_index(
-            self.doc_id_map,
-            &self.column,
-        ))
-    }
-}
-
-fn iter_remapped_multivalue_index<'a, C: Column>(
-    doc_id_map: &'a DocIdMapping,
-    column: &'a C,
-) -> impl Iterator<Item = u64> + 'a {
-    let mut offset = 0;
-    std::iter::once(0).chain(doc_id_map.iter_old_doc_ids().map(move |old_doc| {
-        let num_vals_for_doc = column.get_val(old_doc + 1) - column.get_val(old_doc);
-        offset += num_vals_for_doc;
-        offset
-    }))
-}
-
-/// Writer for multi-valued (as in, more than one value per document)
-/// int fast field.
-///
-/// This `Writer` is only useful for advanced users.
-/// The normal way to get your multivalued int in your index
-/// is to
-/// - declare your field with fast set to `Cardinality::MultiValues`
-/// in your schema
-/// - add your document simply by calling `.add_document(...)`.
-///
-/// The `MultiValuedFastFieldWriter` can be acquired from the
-
-pub struct MultiValueU128FastFieldWriter {
-    field: Field,
-    vals: Vec<u128>,
-    doc_index: Vec<u64>,
-}
-
-impl MultiValueU128FastFieldWriter {
-    /// Creates a new `U128MultiValueFastFieldWriter`
-    pub(crate) fn new(field: Field) -> Self {
-        MultiValueU128FastFieldWriter {
-            field,
-            vals: Vec::new(),
-            doc_index: Vec::new(),
-        }
-    }
-
-    /// The memory used (inclusive childs)
-    pub fn mem_usage(&self) -> usize {
-        self.vals.capacity() * std::mem::size_of::<UnorderedTermId>()
-            + self.doc_index.capacity() * std::mem::size_of::<u64>()
-    }
-
-    /// Finalize the current document.
-    pub(crate) fn next_doc(&mut self) {
-        self.doc_index.push(self.vals.len() as u64);
-    }
-
-    /// Pushes a new value to the current document.
-    pub(crate) fn add_val(&mut self, val: u128) {
-        self.vals.push(val);
-    }
-
-    /// Shift to the next document and adds
-    /// all of the matching field values present in the document.
-    pub fn add_document(&mut self, doc: &Document) -> crate::Result<()> {
-        self.next_doc();
-        for field_value in doc.field_values() {
-            if field_value.field == self.field {
-                let value = field_value.value();
-                let ip_addr = value
-                    .as_ip_addr()
-                    .ok_or_else(|| unexpected_value("ip", value))?;
-                let ip_addr_u128 = ip_addr.to_u128();
-                self.add_val(ip_addr_u128);
-            }
-        }
-        Ok(())
-    }
-
-    /// Returns an iterator over values per doc_id in ascending doc_id order.
-    ///
-    /// Normally the order is simply iterating self.doc_id_index.
-    /// With doc_id_map it accounts for the new mapping, returning values in the order of the
-    /// new doc_ids.
-    fn get_ordered_values<'a: 'b, 'b>(
-        &'a self,
-        doc_id_map: Option<&'b DocIdMapping>,
-    ) -> impl Iterator<Item = &'b [u128]> {
-        get_ordered_values(&self.vals, &self.doc_index, doc_id_map)
-    }
-
-    /// Serializes fast field values.
-    pub fn serialize(
-        mut self,
-        serializer: &mut CompositeFastFieldSerializer,
-        doc_id_map: Option<&DocIdMapping>,
-    ) -> io::Result<()> {
-        {
-            // writing the offset index
-            //
-            self.doc_index.push(self.vals.len() as u64);
-            let col = VecColumn::from(&self.doc_index[..]);
-            if let Some(doc_id_map) = doc_id_map {
-                let multi_value_start_index = MultivalueStartIndex::new(&col, doc_id_map);
-                serializer.create_auto_detect_u64_fast_field_with_idx(
-                    self.field,
-                    multi_value_start_index,
-                    0,
-                )?;
-            } else {
-                serializer.create_auto_detect_u64_fast_field_with_idx(self.field, col, 0)?;
-            }
-        }
-        {
-            let iter_gen = || self.get_ordered_values(doc_id_map).flatten().cloned();
-
-            serializer.create_u128_fast_field_with_idx(
-                self.field,
-                iter_gen,
-                self.vals.len() as u32,
-                1,
-            )?;
-        }
-        Ok(())
-    }
-}
-
-/// Returns an iterator over values per doc_id in ascending doc_id order.
-///
-/// Normally the order is simply iterating self.doc_id_index.
-/// With doc_id_map it accounts for the new mapping, returning values in the order of the
-/// new doc_ids.
-fn get_ordered_values<'a: 'b, 'b, T>(
-    vals: &'a [T],
-    doc_index: &'a [u64],
-    doc_id_map: Option<&'b DocIdMapping>,
-) -> impl Iterator<Item = &'b [T]> {
-    let doc_id_iter: Box<dyn Iterator<Item = u32>> = if let Some(doc_id_map) = doc_id_map {
-        Box::new(doc_id_map.iter_old_doc_ids())
-    } else {
-        let max_doc = doc_index.len() as DocId;
-        Box::new(0..max_doc)
-    };
-    doc_id_iter.map(move |doc_id| get_values_for_doc_id(doc_id, vals, doc_index))
-}
-
-/// returns all values for a doc_id
-fn get_values_for_doc_id<'a, T>(doc_id: u32, vals: &'a [T], doc_index: &'a [u64]) -> &'a [T] {
-    let start_pos = doc_index[doc_id as usize] as usize;
-    let end_pos = doc_index
-        .get(doc_id as usize + 1)
-        .cloned()
-        .unwrap_or(vals.len() as u64) as usize; // special case, last doc_id has no offset information
-    &vals[start_pos..end_pos]
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_multivalue_start_index() {
-        let doc_id_mapping = DocIdMapping::from_new_id_to_old_id(vec![4, 1, 2]);
-        assert_eq!(doc_id_mapping.num_old_doc_ids(), 5);
-        let col = VecColumn::from(&[0u64, 3, 5, 10, 12, 16][..]);
-        let multivalue_start_index = MultivalueStartIndex::new(
-            &col, // 3, 2, 5, 2, 4
-            &doc_id_mapping,
-        );
-        assert_eq!(multivalue_start_index.num_vals(), 4);
-        assert_eq!(
-            multivalue_start_index.iter().collect::<Vec<u64>>(),
-            vec![0, 4, 6, 11]
-        ); // 4, 2, 5
-    }
-
-    #[test]
-    fn test_multivalue_get_vals() {
-        let doc_id_mapping =
-            DocIdMapping::from_new_id_to_old_id(vec![0, 1, 2, 3, 4, 5, 6, 7, 8, 9]);
-        assert_eq!(doc_id_mapping.num_old_doc_ids(), 10);
-        let col = VecColumn::from(&[0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55][..]);
-        let multivalue_start_index = MultivalueStartIndex::new(&col, &doc_id_mapping);
-        assert_eq!(
-            multivalue_start_index.iter().collect::<Vec<u64>>(),
-            vec![0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55]
-        );
-        assert_eq!(multivalue_start_index.num_vals(), 11);
-    }
-}

src/fastfield/readers.rs

@@ -1,14 +1,17 @@
+use std::collections::HashMap;
+use std::io;
 use std::net::Ipv6Addr;
 use std::sync::Arc;
 
-use fastfield_codecs::{open, open_u128, Column};
+use columnar::{
+    BytesColumn, ColumnType, ColumnValues, ColumnarReader, DynamicColumn, DynamicColumnHandle,
+    HasAssociatedColumnType, StrColumn,
+};
+use fastfield_codecs::Column;
 
-use super::multivalued::MultiValuedFastFieldReader;
-use crate::directory::{CompositeFile, FileSlice};
-use crate::fastfield::{BytesFastFieldReader, FastFieldNotAvailableError, FastValue};
-use crate::schema::{Cardinality, Field, FieldType, Schema};
-use crate::space_usage::PerFieldSpaceUsage;
-use crate::{DateTime, TantivyError};
+use crate::directory::FileSlice;
+use crate::schema::{Field, Schema};
+use crate::space_usage::{FieldUsage, PerFieldSpaceUsage};
 
 /// Provides access to all of the BitpackedFastFieldReader.
 ///
@@ -16,315 +19,159 @@ use crate::{DateTime, TantivyError};
 /// and just wraps several `HashMap`.
 #[derive(Clone)]
 pub struct FastFieldReaders {
-    schema: Schema,
-    fast_fields_composite: CompositeFile,
-}
-#[derive(Eq, PartialEq, Debug)]
-pub(crate) enum FastType {
-    I64,
-    U64,
-    U128,
-    F64,
-    Bool,
-    Date,
-}
-
-pub(crate) fn type_and_cardinality(field_type: &FieldType) -> Option<(FastType, Cardinality)> {
-    match field_type {
-        FieldType::U64(options) => options
-            .get_fastfield_cardinality()
-            .map(|cardinality| (FastType::U64, cardinality)),
-        FieldType::I64(options) => options
-            .get_fastfield_cardinality()
-            .map(|cardinality| (FastType::I64, cardinality)),
-        FieldType::F64(options) => options
-            .get_fastfield_cardinality()
-            .map(|cardinality| (FastType::F64, cardinality)),
-        FieldType::Bool(options) => options
-            .get_fastfield_cardinality()
-            .map(|cardinality| (FastType::Bool, cardinality)),
-        FieldType::Date(options) => options
-            .get_fastfield_cardinality()
-            .map(|cardinality| (FastType::Date, cardinality)),
-        FieldType::Facet(_) => Some((FastType::U64, Cardinality::MultiValues)),
-        FieldType::Str(options) if options.is_fast() => {
-            Some((FastType::U64, Cardinality::MultiValues))
-        }
-        FieldType::IpAddr(options) => options
-            .get_fastfield_cardinality()
-            .map(|cardinality| (FastType::U128, cardinality)),
-        _ => None,
-    }
+    columnar: Arc<ColumnarReader>,
 }
 
 impl FastFieldReaders {
-    pub(crate) fn new(schema: Schema, fast_fields_composite: CompositeFile) -> FastFieldReaders {
-        FastFieldReaders {
-            schema,
-            fast_fields_composite,
+    pub(crate) fn open(fast_field_file: FileSlice) -> io::Result<FastFieldReaders> {
+        let columnar = Arc::new(ColumnarReader::open(fast_field_file)?);
+        Ok(FastFieldReaders { columnar })
+    }
+
+    pub(crate) fn space_usage(&self, schema: &Schema) -> io::Result<PerFieldSpaceUsage> {
+        let mut per_field_usages: Vec<FieldUsage> = Default::default();
+        for (field, field_entry) in schema.fields() {
+            let column_handles = self.columnar.read_columns(field_entry.name())?;
+            let num_bytes: usize = column_handles
+                .iter()
+                .map(|column_handle| column_handle.num_bytes())
+                .sum();
+            let mut field_usage = FieldUsage::empty(field);
+            field_usage.add_field_idx(0, num_bytes);
+            per_field_usages.push(field_usage);
         }
+        // TODO fix space usage for JSON fields.
+        Ok(PerFieldSpaceUsage::new(per_field_usages))
     }
 
-    pub(crate) fn space_usage(&self) -> PerFieldSpaceUsage {
-        self.fast_fields_composite.space_usage()
-    }
-
-    #[doc(hidden)]
-    pub fn fast_field_data(&self, field: Field, idx: usize) -> crate::Result<FileSlice> {
-        self.fast_fields_composite
-            .open_read_with_idx(field, idx)
-            .ok_or_else(|| {
-                let field_name = self.schema.get_field_entry(field).name();
-                TantivyError::SchemaError(format!("Field({}) data was not found", field_name))
-            })
-    }
-
-    fn check_type(
-        &self,
-        field: Field,
-        expected_fast_type: FastType,
-        expected_cardinality: Cardinality,
-    ) -> crate::Result<()> {
-        let field_entry = self.schema.get_field_entry(field);
-        let (fast_type, cardinality) =
-            type_and_cardinality(field_entry.field_type()).ok_or_else(|| {
-                crate::TantivyError::SchemaError(format!(
-                    "Field {:?} is not a fast field.",
-                    field_entry.name()
-                ))
-            })?;
-        if fast_type != expected_fast_type {
-            return Err(crate::TantivyError::SchemaError(format!(
-                "Field {:?} is of type {:?}, expected {:?}.",
-                field_entry.name(),
-                fast_type,
-                expected_fast_type
-            )));
-        }
-        if cardinality != expected_cardinality {
-            return Err(crate::TantivyError::SchemaError(format!(
-                "Field {:?} is of cardinality {:?}, expected {:?}.",
-                field_entry.name(),
-                cardinality,
-                expected_cardinality
-            )));
-        }
-        Ok(())
-    }
-
-    pub(crate) fn typed_fast_field_reader_with_idx<TFastValue: FastValue>(
+    pub fn typed_column_opt<T>(
         &self,
         field_name: &str,
-        index: usize,
-    ) -> crate::Result<Arc<dyn Column<TFastValue>>> {
-        let field = self.schema.get_field(field_name)?;
-
-        let fast_field_slice = self.fast_field_data(field, index)?;
-        let bytes = fast_field_slice.read_bytes()?;
-        let column = fastfield_codecs::open(bytes)?;
-        Ok(column)
+    ) -> crate::Result<Option<columnar::Column<T>>>
+    where
+        T: PartialOrd + Copy + HasAssociatedColumnType + Send + Sync + 'static,
+        DynamicColumn: Into<Option<columnar::Column<T>>>,
+    {
+        let column_type = T::column_type();
+        let Some(dynamic_column_handle) = self.column_handle(field_name, column_type)?
+        else {
+            return Ok(None);
+        };
+        let dynamic_column = dynamic_column_handle.open()?;
+        Ok(dynamic_column.into())
     }
 
-    pub(crate) fn typed_fast_field_reader<TFastValue: FastValue>(
-        &self,
-        field_name: &str,
-    ) -> crate::Result<Arc<dyn Column<TFastValue>>> {
-        self.typed_fast_field_reader_with_idx(field_name, 0)
+    pub fn bytes_column_opt(&self, field_name: &str) -> crate::Result<Option<BytesColumn>> {
+        let Some(dynamic_column_handle) = self.column_handle(field_name, ColumnType::Bytes)?
+        else {
+            return Ok(None);
+        };
+        let dynamic_column = dynamic_column_handle.open()?;
+        Ok(dynamic_column.into())
+    }
+    pub fn str_column_opt(&self, field_name: &str) -> crate::Result<Option<StrColumn>> {
+        let Some(dynamic_column_handle) = self.column_handle(field_name, ColumnType::Str)?
+        else {
+            return Ok(None);
+        };
+        let dynamic_column = dynamic_column_handle.open()?;
+        Ok(dynamic_column.into())
     }
 
-    pub(crate) fn typed_fast_field_multi_reader<TFastValue: FastValue>(
-        &self,
-        field_name: &str,
-    ) -> crate::Result<MultiValuedFastFieldReader<TFastValue>> {
-        let idx_reader = self.typed_fast_field_reader(field_name)?;
-        let vals_reader = self.typed_fast_field_reader_with_idx(field_name, 1)?;
-        Ok(MultiValuedFastFieldReader::open(idx_reader, vals_reader))
+    pub fn column_num_bytes(&self, field: &str) -> crate::Result<usize> {
+        Ok(self
+            .columnar
+            .read_columns(field)?
+            .into_iter()
+            .map(|column_handle| column_handle.num_bytes())
+            .sum())
+    }
+
+    pub fn typed_column_first_or_default<T>(&self, field: &str) -> crate::Result<Arc<dyn Column<T>>>
+    where
+        T: PartialOrd + Copy + HasAssociatedColumnType + Send + Sync + 'static,
+        DynamicColumn: Into<Option<columnar::Column<T>>>,
+    {
+        let col_opt: Option<columnar::Column<T>> = self.typed_column_opt(field)?;
+        if let Some(col) = col_opt {
+            Ok(col.first_or_default_col(T::default_value()))
+        } else {
+            Err(crate::TantivyError::SchemaError(format!(
+                "Field `{field}` is missing or is not configured as a fast field."
+            )))
+        }
     }
 
     /// Returns the `u64` fast field reader reader associated with `field`.
     ///
     /// If `field` is not a u64 fast field, this method returns an Error.
-    pub fn u64(&self, field_name: &str) -> crate::Result<Arc<dyn Column<u64>>> {
-        self.check_type(
-            self.schema.get_field(field_name)?,
-            FastType::U64,
-            Cardinality::SingleValue,
-        )?;
-        self.typed_fast_field_reader(field_name)
+    pub fn u64(&self, field: &str) -> crate::Result<Arc<dyn ColumnValues<u64>>> {
+        self.typed_column_first_or_default(field)
+    }
+
+    /// Returns the `date` fast field reader reader associated with `field`.
+    ///
+    /// If `field` is not a date fast field, this method returns an Error.
+    pub fn date(&self, field: &str) -> crate::Result<Arc<dyn ColumnValues<columnar::DateTime>>> {
+        self.typed_column_first_or_default(field)
     }
 
     /// Returns the `ip` fast field reader reader associated to `field`.
     ///
     /// If `field` is not a u128 fast field, this method returns an Error.
-    pub fn ip_addr(&self, field_name: &str) -> crate::Result<Arc<dyn Column<Ipv6Addr>>> {
-        let field = self.schema.get_field(field_name)?;
-        self.check_type(field, FastType::U128, Cardinality::SingleValue)?;
-        let bytes = self.fast_field_data(field, 0)?.read_bytes()?;
-        Ok(open_u128::<Ipv6Addr>(bytes)?)
+    pub fn ip_addr(&self, field: &str) -> crate::Result<Arc<dyn Column<Ipv6Addr>>> {
+        self.typed_column_first_or_default(field)
     }
 
-    /// Returns the `ip` fast field reader reader associated to `field`.
-    ///
-    /// If `field` is not a u128 fast field, this method returns an Error.
-    pub fn ip_addrs(
+    pub fn str(&self, field: &str) -> crate::Result<Option<columnar::StrColumn>> {
+        self.str_column_opt(field)
+    }
+
+    pub fn bytes(&self, field: &str) -> crate::Result<Option<columnar::BytesColumn>> {
+        self.bytes_column_opt(field)
+    }
+
+    pub fn column_handle(
         &self,
         field_name: &str,
-    ) -> crate::Result<MultiValuedFastFieldReader<Ipv6Addr>> {
-        let field = self.schema.get_field(field_name)?;
-        self.check_type(field, FastType::U128, Cardinality::MultiValues)?;
-        let idx_reader: Arc<dyn Column<u64>> = self.typed_fast_field_reader(field_name)?;
-
-        let bytes = self.fast_field_data(field, 1)?.read_bytes()?;
-        let vals_reader = open_u128::<Ipv6Addr>(bytes)?;
-
-        Ok(MultiValuedFastFieldReader::open(idx_reader, vals_reader))
+        column_type: ColumnType,
+    ) -> crate::Result<Option<DynamicColumnHandle>> {
+        let dynamic_column_handle_opt = self
+            .columnar
+            .read_columns(field_name)?
+            .into_iter()
+            .filter(|column| column.column_type() == column_type)
+            .next();
+        Ok(dynamic_column_handle_opt)
     }
 
-    /// Returns the `u128` fast field reader reader associated to `field`.
-    ///
-    /// If `field` is not a u128 fast field, this method returns an Error.
-    pub(crate) fn u128(&self, field_name: &str) -> crate::Result<Arc<dyn Column<u128>>> {
-        let field = self.schema.get_field(field_name)?;
-        self.check_type(field, FastType::U128, Cardinality::SingleValue)?;
-        let bytes = self.fast_field_data(field, 0)?.read_bytes()?;
-        Ok(open_u128::<u128>(bytes)?)
-    }
-
-    /// Returns the `u128` multi-valued fast field reader reader associated to `field`.
-    ///
-    /// If `field` is not a u128 multi-valued fast field, this method returns an Error.
-    pub fn u128s(&self, field_name: &str) -> crate::Result<MultiValuedFastFieldReader<u128>> {
-        let field = self.schema.get_field(field_name)?;
-        self.check_type(field, FastType::U128, Cardinality::MultiValues)?;
-        let idx_reader: Arc<dyn Column<u64>> =
-            self.typed_fast_field_reader(self.schema.get_field_name(field))?;
-
-        let bytes = self.fast_field_data(field, 1)?.read_bytes()?;
-        let vals_reader = open_u128::<u128>(bytes)?;
-
-        Ok(MultiValuedFastFieldReader::open(idx_reader, vals_reader))
-    }
-
-    /// Returns the `u64` fast field reader reader associated with `field`, regardless of whether
-    /// the given field is effectively of type `u64` or not.
-    ///
-    /// If not, the fastfield reader will returns the u64-value associated with the original
-    /// FastValue.
-    pub fn u64_lenient(&self, field_name: &str) -> crate::Result<Arc<dyn Column<u64>>> {
-        self.typed_fast_field_reader(field_name)
+    pub fn u64_lenient(&self, field_name: &str) -> crate::Result<Option<columnar::Column<u64>>> {
+        for col in self.columnar.read_columns(field_name)? {
+            if let Some(col_u64) = col.open_u64_lenient()? {
+                return Ok(Some(col_u64));
+            }
+        }
+        Ok(None)
     }
 
     /// Returns the `i64` fast field reader reader associated with `field`.
     ///
     /// If `field` is not a i64 fast field, this method returns an Error.
     pub fn i64(&self, field_name: &str) -> crate::Result<Arc<dyn Column<i64>>> {
-        let field = self.schema.get_field(field_name)?;
-        self.check_type(field, FastType::I64, Cardinality::SingleValue)?;
-        self.typed_fast_field_reader(self.schema.get_field_name(field))
-    }
-
-    /// Returns the `date` fast field reader reader associated with `field`.
-    ///
-    /// If `field` is not a date fast field, this method returns an Error.
-    pub fn date(&self, field_name: &str) -> crate::Result<Arc<dyn Column<DateTime>>> {
-        let field = self.schema.get_field(field_name)?;
-        self.check_type(field, FastType::Date, Cardinality::SingleValue)?;
-        self.typed_fast_field_reader(field_name)
+        self.typed_column_first_or_default(field_name)
     }
 
     /// Returns the `f64` fast field reader reader associated with `field`.
     ///
     /// If `field` is not a f64 fast field, this method returns an Error.
     pub fn f64(&self, field_name: &str) -> crate::Result<Arc<dyn Column<f64>>> {
-        let field = self.schema.get_field(field_name)?;
-        self.check_type(field, FastType::F64, Cardinality::SingleValue)?;
-        self.typed_fast_field_reader(field_name)
+        self.typed_column_first_or_default(field_name)
     }
 
     /// Returns the `bool` fast field reader reader associated with `field`.
     ///
     /// If `field` is not a bool fast field, this method returns an Error.
     pub fn bool(&self, field_name: &str) -> crate::Result<Arc<dyn Column<bool>>> {
-        let field = self.schema.get_field(field_name)?;
-        self.check_type(field, FastType::Bool, Cardinality::SingleValue)?;
-        self.typed_fast_field_reader(field_name)
-    }
-
-    /// Returns a `u64s` multi-valued fast field reader reader associated with `field`.
-    ///
-    /// If `field` is not a u64 multi-valued fast field, this method returns an Error.
-    pub fn u64s(&self, field_name: &str) -> crate::Result<MultiValuedFastFieldReader<u64>> {
-        let field = self.schema.get_field(field_name)?;
-        self.check_type(field, FastType::U64, Cardinality::MultiValues)?;
-        self.typed_fast_field_multi_reader(field_name)
-    }
-
-    /// Returns a `u64s` multi-valued fast field reader reader associated with `field`, regardless
-    /// of whether the given field is effectively of type `u64` or not.
-    ///
-    /// If `field` is not a u64 multi-valued fast field, this method returns an Error.
-    pub fn u64s_lenient(&self, field_name: &str) -> crate::Result<MultiValuedFastFieldReader<u64>> {
-        self.typed_fast_field_multi_reader(field_name)
-    }
-
-    /// Returns a `i64s` multi-valued fast field reader reader associated with `field`.
-    ///
-    /// If `field` is not a i64 multi-valued fast field, this method returns an Error.
-    pub fn i64s(&self, field_name: &str) -> crate::Result<MultiValuedFastFieldReader<i64>> {
-        let field = self.schema.get_field(field_name)?;
-        self.check_type(field, FastType::I64, Cardinality::MultiValues)?;
-        self.typed_fast_field_multi_reader(self.schema.get_field_name(field))
-    }
-
-    /// Returns a `f64s` multi-valued fast field reader reader associated with `field`.
-    ///
-    /// If `field` is not a f64 multi-valued fast field, this method returns an Error.
-    pub fn f64s(&self, field_name: &str) -> crate::Result<MultiValuedFastFieldReader<f64>> {
-        let field = self.schema.get_field(field_name)?;
-        self.check_type(field, FastType::F64, Cardinality::MultiValues)?;
-        self.typed_fast_field_multi_reader(self.schema.get_field_name(field))
-    }
-
-    /// Returns a `bools` multi-valued fast field reader reader associated with `field`.
-    ///
-    /// If `field` is not a bool multi-valued fast field, this method returns an Error.
-    pub fn bools(&self, field_name: &str) -> crate::Result<MultiValuedFastFieldReader<bool>> {
-        let field = self.schema.get_field(field_name)?;
-        self.check_type(field, FastType::Bool, Cardinality::MultiValues)?;
-        self.typed_fast_field_multi_reader(self.schema.get_field_name(field))
-    }
-
-    /// Returns a `time::OffsetDateTime` multi-valued fast field reader reader associated with
-    /// `field`.
-    ///
-    /// If `field` is not a `time::OffsetDateTime` multi-valued fast field, this method returns an
-    /// Error.
-    pub fn dates(&self, field_name: &str) -> crate::Result<MultiValuedFastFieldReader<DateTime>> {
-        let field = self.schema.get_field(field_name)?;
-        self.check_type(field, FastType::Date, Cardinality::MultiValues)?;
-        self.typed_fast_field_multi_reader(self.schema.get_field_name(field))
-    }
-
-    /// Returns the `bytes` fast field reader associated with `field`.
-    ///
-    /// If `field` is not a bytes fast field, returns an Error.
-    pub fn bytes(&self, field_name: &str) -> crate::Result<BytesFastFieldReader> {
-        let field = self.schema.get_field(field_name)?;
-        let field_entry = self.schema.get_field_entry(field);
-        if let FieldType::Bytes(bytes_option) = field_entry.field_type() {
-            if !bytes_option.is_fast() {
-                return Err(crate::TantivyError::SchemaError(format!(
-                    "Field {:?} is not a fast field.",
-                    field_entry.name()
-                )));
-            }
-            let fast_field_idx_file = self.fast_field_data(field, 0)?;
-            let fast_field_idx_bytes = fast_field_idx_file.read_bytes()?;
-            let idx_reader = open(fast_field_idx_bytes)?;
-            let data = self.fast_field_data(field, 1)?;
-            BytesFastFieldReader::open(idx_reader, data)
-        } else {
-            Err(FastFieldNotAvailableError::new(field_entry).into())
-        }
+        self.typed_column_first_or_default(field_name)
     }
 }

src/fastfield/serializer/mod.rs

@@ -1,122 +0,0 @@
-use std::fmt;
-use std::io::{self, Write};
-
-pub use fastfield_codecs::Column;
-use fastfield_codecs::{FastFieldCodecType, MonotonicallyMappableToU64, ALL_CODEC_TYPES};
-
-use crate::directory::{CompositeWrite, WritePtr};
-use crate::schema::Field;
-
-/// `CompositeFastFieldSerializer` is in charge of serializing
-/// fastfields on disk.
-///
-/// Fast fields have different encodings like bit-packing.
-///
-/// `FastFieldWriter`s are in charge of pushing the data to
-/// the serializer.
-/// The serializer expects to receive the following calls.
-///
-/// * `create_auto_detect_u64_fast_field(...)`
-/// * `create_auto_detect_u64_fast_field(...)`
-/// * ...
-/// * `let bytes_fastfield = new_bytes_fast_field(...)`
-/// * `bytes_fastfield.write_all(...)`
-/// * `bytes_fastfield.write_all(...)`
-/// * `bytes_fastfield.flush()`
-/// * ...
-/// * `close()`
-pub struct CompositeFastFieldSerializer {
-    composite_write: CompositeWrite<WritePtr>,
-    codec_types: Vec<FastFieldCodecType>,
-}
-
-impl CompositeFastFieldSerializer {
-    /// New fast field serializer with all codec types
-    pub fn from_write(write: WritePtr) -> io::Result<CompositeFastFieldSerializer> {
-        Self::from_write_with_codec(write, &ALL_CODEC_TYPES)
-    }
-
-    /// New fast field serializer with allowed codec types
-    pub fn from_write_with_codec(
-        write: WritePtr,
-        codec_types: &[FastFieldCodecType],
-    ) -> io::Result<CompositeFastFieldSerializer> {
-        let composite_write = CompositeWrite::wrap(write);
-        Ok(CompositeFastFieldSerializer {
-            composite_write,
-            codec_types: codec_types.to_vec(),
-        })
-    }
-
-    /// Serialize data into a new u64 fast field. The best compression codec will be chosen
-    /// automatically.
-    pub fn create_auto_detect_u64_fast_field<T: MonotonicallyMappableToU64 + fmt::Debug>(
-        &mut self,
-        field: Field,
-        fastfield_accessor: impl Column<T>,
-    ) -> io::Result<()> {
-        self.create_auto_detect_u64_fast_field_with_idx(field, fastfield_accessor, 0)
-    }
-
-    /// Serialize data into a new u64 fast field. The best compression codec will be chosen
-    /// automatically.
-    pub fn create_auto_detect_u64_fast_field_with_idx<
-        T: MonotonicallyMappableToU64 + fmt::Debug,
-    >(
-        &mut self,
-        field: Field,
-        fastfield_accessor: impl Column<T>,
-        idx: usize,
-    ) -> io::Result<()> {
-        let field_write = self.composite_write.for_field_with_idx(field, idx);
-        fastfield_codecs::serialize(fastfield_accessor, field_write, &self.codec_types)?;
-        Ok(())
-    }
-
-    /// Serialize data into a new u64 fast field. The best compression codec of the the provided
-    /// will be chosen.
-    pub fn create_auto_detect_u64_fast_field_with_idx_and_codecs<
-        T: MonotonicallyMappableToU64 + fmt::Debug,
-    >(
-        &mut self,
-        field: Field,
-        fastfield_accessor: impl Column<T>,
-        idx: usize,
-        codec_types: &[FastFieldCodecType],
-    ) -> io::Result<()> {
-        let field_write = self.composite_write.for_field_with_idx(field, idx);
-        fastfield_codecs::serialize(fastfield_accessor, field_write, codec_types)?;
-        Ok(())
-    }
-
-    /// Serialize data into a new u128 fast field. The codec will be compact space compressor,
-    /// which is optimized for scanning the fast field for a given range.
-    pub fn create_u128_fast_field_with_idx<F: Fn() -> I, I: Iterator<Item = u128>>(
-        &mut self,
-        field: Field,
-        iter_gen: F,
-        num_vals: u32,
-        idx: usize,
-    ) -> io::Result<()> {
-        let field_write = self.composite_write.for_field_with_idx(field, idx);
-        fastfield_codecs::serialize_u128(iter_gen, num_vals, field_write)?;
-
-        Ok(())
-    }
-
-    /// Start serializing a new [u8] fast field. Use the returned writer to write data into the
-    /// bytes field. To associate the bytes with documents a seperate index must be created on
-    /// index 0. See bytes/writer.rs::serialize for an example.
-    ///
-    /// The bytes will be stored as is, no compression will be applied.
-    pub fn new_bytes_fast_field(&mut self, field: Field) -> impl Write + '_ {
-        self.composite_write.for_field_with_idx(field, 1)
-    }
-
-    /// Closes the serializer
-    ///
-    /// After this call the data must be persistently saved on disk.
-    pub fn close(self) -> io::Result<()> {
-        self.composite_write.close()
-    }
-}

src/fastfield/writer.rs

@@ -1,558 +1,153 @@
-use std::collections::HashMap;
 use std::io;
 
-use common;
-use fastfield_codecs::{Column, MonotonicallyMappableToU128, MonotonicallyMappableToU64};
-use rustc_hash::FxHashMap;
-use tantivy_bitpacker::BlockedBitpacker;
+use columnar::{ColumnType, ColumnarWriter, NumericalValue};
 
-use super::multivalued::{MultiValueU128FastFieldWriter, MultiValuedFastFieldWriter};
-use super::FastFieldType;
-use crate::fastfield::{BytesFastFieldWriter, CompositeFastFieldSerializer};
 use crate::indexer::doc_id_mapping::DocIdMapping;
-use crate::postings::UnorderedTermId;
-use crate::schema::{Cardinality, Document, Field, FieldEntry, FieldType, Schema, Value};
-use crate::termdict::TermOrdinal;
-use crate::DatePrecision;
+use crate::schema::{Document, FieldType, Schema, Type, Value};
+use crate::{DatePrecision, DocId};
 
 /// The `FastFieldsWriter` groups all of the fast field writers.
 pub struct FastFieldsWriter {
-    term_id_writers: Vec<MultiValuedFastFieldWriter>,
-    single_value_writers: Vec<IntFastFieldWriter>,
-    u128_value_writers: Vec<U128FastFieldWriter>,
-    u128_multi_value_writers: Vec<MultiValueU128FastFieldWriter>,
-    multi_values_writers: Vec<MultiValuedFastFieldWriter>,
-    bytes_value_writers: Vec<BytesFastFieldWriter>,
-}
-
-pub(crate) fn unexpected_value(expected: &str, actual: &Value) -> crate::TantivyError {
-    crate::TantivyError::SchemaError(format!(
-        "Expected a {:?} in fast field, but got {:?}",
-        expected, actual
-    ))
-}
-
-fn fast_field_default_value(field_entry: &FieldEntry) -> u64 {
-    match *field_entry.field_type() {
-        FieldType::I64(_) | FieldType::Date(_) => common::i64_to_u64(0i64),
-        FieldType::F64(_) => common::f64_to_u64(0.0f64),
-        _ => 0u64,
-    }
+    columnar_writer: ColumnarWriter,
+    fast_field_names: Vec<Option<String>>, //< TODO see if we can cash the field name hash too.
+    date_precisions: Vec<DatePrecision>,
+    num_docs: DocId,
 }
 
 impl FastFieldsWriter {
     /// Create all `FastFieldWriter` required by the schema.
     pub fn from_schema(schema: &Schema) -> FastFieldsWriter {
-        let mut u128_value_writers = Vec::new();
-        let mut u128_multi_value_writers = Vec::new();
-        let mut single_value_writers = Vec::new();
-        let mut term_id_writers = Vec::new();
-        let mut multi_values_writers = Vec::new();
-        let mut bytes_value_writers = Vec::new();
-
-        for (field, field_entry) in schema.fields() {
-            match field_entry.field_type() {
-                FieldType::I64(ref int_options)
-                | FieldType::U64(ref int_options)
-                | FieldType::F64(ref int_options)
-                | FieldType::Bool(ref int_options) => {
-                    match int_options.get_fastfield_cardinality() {
-                        Some(Cardinality::SingleValue) => {
-                            let mut fast_field_writer = IntFastFieldWriter::new(field, None);
-                            let default_value = fast_field_default_value(field_entry);
-                            fast_field_writer.set_val_if_missing(default_value);
-                            single_value_writers.push(fast_field_writer);
-                        }
-                        Some(Cardinality::MultiValues) => {
-                            let fast_field_writer = MultiValuedFastFieldWriter::new(
-                                field,
-                                FastFieldType::Numeric,
-                                None,
-                            );
-                            multi_values_writers.push(fast_field_writer);
-                        }
-                        None => {}
-                    }
-                }
-                FieldType::Date(ref options) => match options.get_fastfield_cardinality() {
-                    Some(Cardinality::SingleValue) => {
-                        let mut fast_field_writer =
-                            IntFastFieldWriter::new(field, Some(options.get_precision()));
-                        let default_value = fast_field_default_value(field_entry);
-                        fast_field_writer.set_val_if_missing(default_value);
-                        single_value_writers.push(fast_field_writer);
-                    }
-                    Some(Cardinality::MultiValues) => {
-                        let fast_field_writer = MultiValuedFastFieldWriter::new(
-                            field,
-                            FastFieldType::Numeric,
-                            Some(options.get_precision()),
-                        );
-                        multi_values_writers.push(fast_field_writer);
-                    }
-                    None => {}
-                },
-                FieldType::Facet(_) => {
-                    let fast_field_writer =
-                        MultiValuedFastFieldWriter::new(field, FastFieldType::Facet, None);
-                    term_id_writers.push(fast_field_writer);
-                }
-                FieldType::Str(_) if field_entry.is_fast() => {
-                    let fast_field_writer =
-                        MultiValuedFastFieldWriter::new(field, FastFieldType::String, None);
-                    term_id_writers.push(fast_field_writer);
-                }
-                FieldType::Bytes(bytes_option) => {
-                    if bytes_option.is_fast() {
-                        let fast_field_writer = BytesFastFieldWriter::new(field);
-                        bytes_value_writers.push(fast_field_writer);
-                    }
-                }
-                FieldType::IpAddr(opt) => {
-                    if opt.is_fast() {
-                        match opt.get_fastfield_cardinality() {
-                            Some(Cardinality::SingleValue) => {
-                                let fast_field_writer = U128FastFieldWriter::new(field);
-                                u128_value_writers.push(fast_field_writer);
-                            }
-                            Some(Cardinality::MultiValues) => {
-                                let fast_field_writer = MultiValueU128FastFieldWriter::new(field);
-                                u128_multi_value_writers.push(fast_field_writer);
-                            }
-                            None => {}
-                        }
-                    }
-                }
-                FieldType::Str(_) | FieldType::JsonObject(_) => {}
+        let mut columnar_writer = ColumnarWriter::default();
+        let mut fast_fields: Vec<Option<String>> = vec![None; schema.num_fields()];
+        let mut date_precisions: Vec<DatePrecision> =
+            std::iter::repeat_with(DatePrecision::default)
+                .take(schema.num_fields())
+                .collect();
+        // TODO see other types
+        for (field_id, field_entry) in schema.fields() {
+            if !field_entry.field_type().is_fast() {
+                continue;
             }
+            fast_fields[field_id.field_id() as usize] = Some(field_entry.name().to_string());
+            let value_type = field_entry.field_type().value_type();
+            let column_type = match value_type {
+                Type::Str => ColumnType::Str,
+                Type::U64 => ColumnType::U64,
+                Type::I64 => ColumnType::I64,
+                Type::F64 => ColumnType::F64,
+                Type::Bool => ColumnType::Bool,
+                Type::Date => ColumnType::DateTime,
+                Type::Facet => ColumnType::Str,
+                Type::Bytes => ColumnType::Bytes,
+                Type::Json => {
+                    continue;
+                }
+                Type::IpAddr => ColumnType::IpAddr,
+            };
+            if let FieldType::Date(date_options) = field_entry.field_type() {
+                date_precisions[field_id.field_id() as usize] = date_options.get_precision();
+            }
+            let sort_values_within_row = value_type == Type::Facet;
+            columnar_writer.record_column_type(
+                field_entry.name(),
+                column_type,
+                sort_values_within_row,
+            );
         }
         FastFieldsWriter {
-            u128_value_writers,
-            u128_multi_value_writers,
-            term_id_writers,
-            single_value_writers,
-            multi_values_writers,
-            bytes_value_writers,
+            columnar_writer,
+            fast_field_names: fast_fields,
+            num_docs: 0u32,
+            date_precisions,
         }
     }
 
     /// The memory used (inclusive childs)
     pub fn mem_usage(&self) -> usize {
-        self.term_id_writers
-            .iter()
-            .map(|w| w.mem_usage())
-            .sum::<usize>()
-            + self
-                .single_value_writers
-                .iter()
-                .map(|w| w.mem_usage())
-                .sum::<usize>()
-            + self
-                .multi_values_writers
-                .iter()
-                .map(|w| w.mem_usage())
-                .sum::<usize>()
-            + self
-                .bytes_value_writers
-                .iter()
-                .map(|w| w.mem_usage())
-                .sum::<usize>()
-            + self
-                .u128_value_writers
-                .iter()
-                .map(|w| w.mem_usage())
-                .sum::<usize>()
-            + self
-                .u128_multi_value_writers
-                .iter()
-                .map(|w| w.mem_usage())
-                .sum::<usize>()
+        self.columnar_writer.mem_usage()
     }
 
-    /// Get the `FastFieldWriter` associated with a field.
-    pub fn get_term_id_writer(&self, field: Field) -> Option<&MultiValuedFastFieldWriter> {
-        // TODO optimize
-        self.term_id_writers
-            .iter()
-            .find(|field_writer| field_writer.field() == field)
-    }
-
-    /// Get the `FastFieldWriter` associated with a field.
-    pub fn get_field_writer(&self, field: Field) -> Option<&IntFastFieldWriter> {
-        // TODO optimize
-        self.single_value_writers
-            .iter()
-            .find(|field_writer| field_writer.field() == field)
-    }
-
-    /// Get the `FastFieldWriter` associated with a field.
-    pub fn get_field_writer_mut(&mut self, field: Field) -> Option<&mut IntFastFieldWriter> {
-        // TODO optimize
-        self.single_value_writers
-            .iter_mut()
-            .find(|field_writer| field_writer.field() == field)
-    }
-
-    /// Get the `FastFieldWriter` associated with a field.
-    pub fn get_term_id_writer_mut(
-        &mut self,
-        field: Field,
-    ) -> Option<&mut MultiValuedFastFieldWriter> {
-        // TODO optimize
-        self.term_id_writers
-            .iter_mut()
-            .find(|field_writer| field_writer.field() == field)
-    }
-
-    /// Returns the fast field multi-value writer for the given field.
-    ///
-    /// Returns `None` if the field does not exist, or is not
-    /// configured as a multivalued fastfield in the schema.
-    pub fn get_multivalue_writer_mut(
-        &mut self,
-        field: Field,
-    ) -> Option<&mut MultiValuedFastFieldWriter> {
-        // TODO optimize
-        self.multi_values_writers
-            .iter_mut()
-            .find(|multivalue_writer| multivalue_writer.field() == field)
-    }
-
-    /// Returns the bytes fast field writer for the given field.
-    ///
-    /// Returns `None` if the field does not exist, or is not
-    /// configured as a bytes fastfield in the schema.
-    pub fn get_bytes_writer_mut(&mut self, field: Field) -> Option<&mut BytesFastFieldWriter> {
-        // TODO optimize
-        self.bytes_value_writers
-            .iter_mut()
-            .find(|field_writer| field_writer.field() == field)
-    }
     /// Indexes all of the fastfields of a new document.
     pub fn add_document(&mut self, doc: &Document) -> crate::Result<()> {
-        for field_writer in &mut self.term_id_writers {
-            field_writer.add_document(doc)?;
-        }
-        for field_writer in &mut self.single_value_writers {
-            field_writer.add_document(doc)?;
-        }
-        for field_writer in &mut self.multi_values_writers {
-            field_writer.add_document(doc)?;
-        }
-        for field_writer in &mut self.bytes_value_writers {
-            field_writer.add_document(doc)?;
-        }
-        for field_writer in &mut self.u128_value_writers {
-            field_writer.add_document(doc)?;
-        }
-        for field_writer in &mut self.u128_multi_value_writers {
-            field_writer.add_document(doc)?;
+        let doc_id = self.num_docs;
+        for field_value in doc.field_values() {
+            if let Some(field_name) =
+                self.fast_field_names[field_value.field().field_id() as usize].as_ref()
+            {
+                match &field_value.value {
+                    Value::U64(u64_val) => {
+                        self.columnar_writer.record_numerical(
+                            doc_id,
+                            field_name.as_str(),
+                            NumericalValue::from(*u64_val),
+                        );
+                    }
+                    Value::I64(i64_val) => {
+                        self.columnar_writer.record_numerical(
+                            doc_id,
+                            field_name.as_str(),
+                            NumericalValue::from(*i64_val),
+                        );
+                    }
+                    Value::F64(f64_val) => {
+                        self.columnar_writer.record_numerical(
+                            doc_id,
+                            field_name.as_str(),
+                            NumericalValue::from(*f64_val),
+                        );
+                    }
+                    Value::Str(text_val) => {
+                        self.columnar_writer
+                            .record_str(doc_id, field_name.as_str(), text_val);
+                    }
+                    Value::Bytes(bytes_val) => {
+                        self.columnar_writer
+                            .record_bytes(doc_id, field_name.as_str(), bytes_val);
+                    }
+                    Value::PreTokStr(_) => todo!(),
+                    Value::Bool(bool_val) => {
+                        self.columnar_writer
+                            .record_bool(doc_id, field_name.as_str(), *bool_val);
+                    }
+                    Value::Date(datetime) => {
+                        let date_precision =
+                            self.date_precisions[field_value.field().field_id() as usize];
+                        let truncated_datetime = datetime.truncate(date_precision);
+                        self.columnar_writer.record_datetime(
+                            doc_id,
+                            field_name.as_str(),
+                            truncated_datetime.into(),
+                        );
+                    }
+                    Value::Facet(facet) => {
+                        self.columnar_writer.record_str(
+                            doc_id,
+                            field_name.as_str(),
+                            facet.encoded_str(),
+                        );
+                    }
+                    Value::JsonObject(_) => todo!(),
+                    Value::IpAddr(ip_addr) => {
+                        self.columnar_writer
+                            .record_ip_addr(doc_id, field_name.as_str(), *ip_addr);
+                    }
+                }
+            }
         }
+        self.num_docs += 1;
         Ok(())
     }
 
     /// Serializes all of the `FastFieldWriter`s by pushing them in
     /// order to the fast field serializer.
     pub fn serialize(
-        self,
-        serializer: &mut CompositeFastFieldSerializer,
-        mapping: &HashMap<Field, FxHashMap<UnorderedTermId, TermOrdinal>>,
+        mut self,
+        wrt: &mut dyn io::Write,
         doc_id_map: Option<&DocIdMapping>,
     ) -> io::Result<()> {
-        for field_writer in self.term_id_writers {
-            let field = field_writer.field();
-            field_writer.serialize(serializer, mapping.get(&field), doc_id_map)?;
-        }
-        for field_writer in &self.single_value_writers {
-            field_writer.serialize(serializer, doc_id_map)?;
-        }
-
-        for field_writer in self.multi_values_writers {
-            let field = field_writer.field();
-            field_writer.serialize(serializer, mapping.get(&field), doc_id_map)?;
-        }
-        for field_writer in self.bytes_value_writers {
-            field_writer.serialize(serializer, doc_id_map)?;
-        }
-        for field_writer in self.u128_value_writers {
-            field_writer.serialize(serializer, doc_id_map)?;
-        }
-        for field_writer in self.u128_multi_value_writers {
-            field_writer.serialize(serializer, doc_id_map)?;
-        }
-
+        assert!(doc_id_map.is_none()); // TODO handle doc id map
+        let num_docs = self.num_docs;
+        self.columnar_writer.serialize(num_docs, wrt)?;
         Ok(())
     }
 }
-
-/// Fast field writer for u128 values.
-/// The fast field writer just keeps the values in memory.
-///
-/// Only when the segment writer can be closed and
-/// persisted on disk, the fast field writer is
-/// sent to a `FastFieldSerializer` via the `.serialize(...)`
-/// method.
-///
-/// We cannot serialize earlier as the values are
-/// compressed to a compact number space and the number of
-/// bits required for bitpacking can only been known once
-/// we have seen all of the values.
-pub struct U128FastFieldWriter {
-    field: Field,
-    vals: Vec<u128>,
-    val_count: u32,
-}
-
-impl U128FastFieldWriter {
-    /// Creates a new `IntFastFieldWriter`
-    pub fn new(field: Field) -> Self {
-        Self {
-            field,
-            vals: vec![],
-            val_count: 0,
-        }
-    }
-
-    /// The memory used (inclusive childs)
-    pub fn mem_usage(&self) -> usize {
-        self.vals.len() * 16
-    }
-
-    /// Records a new value.
-    ///
-    /// The n-th value being recorded is implicitely
-    /// associated to the document with the `DocId` n.
-    /// (Well, `n-1` actually because of 0-indexing)
-    pub fn add_val(&mut self, val: u128) {
-        self.vals.push(val);
-    }
-
-    /// Extract the fast field value from the document
-    /// (or use the default value) and records it.
-    ///
-    /// Extract the value associated to the fast field for
-    /// this document.
-    pub fn add_document(&mut self, doc: &Document) -> crate::Result<()> {
-        match doc.get_first(self.field) {
-            Some(v) => {
-                let ip_addr = v.as_ip_addr().ok_or_else(|| unexpected_value("ip", v))?;
-                let value = ip_addr.to_u128();
-                self.add_val(value);
-            }
-            None => {
-                self.add_val(0); // TODO fix null handling
-            }
-        };
-        self.val_count += 1;
-        Ok(())
-    }
-
-    /// Push the fast fields value to the `FastFieldWriter`.
-    pub fn serialize(
-        &self,
-        serializer: &mut CompositeFastFieldSerializer,
-        doc_id_map: Option<&DocIdMapping>,
-    ) -> io::Result<()> {
-        if let Some(doc_id_map) = doc_id_map {
-            let iter_gen = || {
-                doc_id_map
-                    .iter_old_doc_ids()
-                    .map(|idx| self.vals[idx as usize])
-            };
-
-            serializer.create_u128_fast_field_with_idx(self.field, iter_gen, self.val_count, 0)?;
-        } else {
-            let iter_gen = || self.vals.iter().cloned();
-            serializer.create_u128_fast_field_with_idx(self.field, iter_gen, self.val_count, 0)?;
-        }
-
-        Ok(())
-    }
-}
-
-/// Fast field writer for ints.
-/// The fast field writer just keeps the values in memory.
-///
-/// Only when the segment writer can be closed and
-/// persisted on disk, the fast field writer is
-/// sent to a `FastFieldSerializer` via the `.serialize(...)`
-/// method.
-///
-/// We cannot serialize earlier as the values are
-/// bitpacked and the number of bits required for bitpacking
-/// can only been known once we have seen all of the values.
-///
-/// Both u64, i64 and f64 use the same writer.
-/// i64 and f64 are just remapped to the `0..2^64 - 1`
-/// using `common::i64_to_u64` and `common::f64_to_u64`.
-pub struct IntFastFieldWriter {
-    field: Field,
-    precision_opt: Option<DatePrecision>,
-    vals: BlockedBitpacker,
-    val_count: usize,
-    val_if_missing: u64,
-    val_min: u64,
-    val_max: u64,
-}
-
-impl IntFastFieldWriter {
-    /// Creates a new `IntFastFieldWriter`
-    pub fn new(field: Field, precision_opt: Option<DatePrecision>) -> IntFastFieldWriter {
-        IntFastFieldWriter {
-            field,
-            precision_opt,
-            vals: BlockedBitpacker::new(),
-            val_count: 0,
-            val_if_missing: 0u64,
-            val_min: u64::MAX,
-            val_max: 0,
-        }
-    }
-
-    /// The memory used (inclusive childs)
-    pub fn mem_usage(&self) -> usize {
-        self.vals.mem_usage()
-    }
-
-    /// Returns the field that this writer is targeting.
-    pub fn field(&self) -> Field {
-        self.field
-    }
-
-    /// Sets the default value.
-    ///
-    /// This default value is recorded for documents if
-    /// a document does not have any value.
-    fn set_val_if_missing(&mut self, val_if_missing: u64) {
-        self.val_if_missing = val_if_missing;
-    }
-
-    /// Records a new value.
-    ///
-    /// The n-th value being recorded is implicitly
-    /// associated with the document with the `DocId` n.
-    /// (Well, `n-1` actually because of 0-indexing)
-    pub fn add_val(&mut self, val: u64) {
-        self.vals.add(val);
-
-        if val > self.val_max {
-            self.val_max = val;
-        }
-        if val < self.val_min {
-            self.val_min = val;
-        }
-
-        self.val_count += 1;
-    }
-
-    /// Extract the fast field value from the document
-    /// (or use the default value) and records it.
-    ///
-    ///
-    /// Extract the value associated with the fast field for
-    /// this document.
-    ///
-    /// i64 and f64 are remapped to u64 using the logic
-    /// in `common::i64_to_u64` and `common::f64_to_u64`.
-    ///
-    /// If the value is missing, then the default value is used
-    /// instead.
-    /// If the document has more than one value for the given field,
-    /// only the first one is taken in account.
-    ///
-    /// Values on text fast fields are skipped.
-    pub fn add_document(&mut self, doc: &Document) -> crate::Result<()> {
-        match doc.get_first(self.field) {
-            Some(v) => {
-                let value = match (self.precision_opt, v) {
-                    (Some(precision), Value::Date(date_val)) => {
-                        date_val.truncate(precision).to_u64()
-                    }
-                    _ => super::value_to_u64(v)?,
-                };
-                self.add_val(value);
-            }
-            None => {
-                self.add_val(self.val_if_missing);
-            }
-        };
-        Ok(())
-    }
-
-    /// get iterator over the data
-    pub(crate) fn iter(&self) -> impl Iterator<Item = u64> + '_ {
-        self.vals.iter()
-    }
-
-    /// Push the fast fields value to the `FastFieldWriter`.
-    pub fn serialize(
-        &self,
-        serializer: &mut CompositeFastFieldSerializer,
-        doc_id_map: Option<&DocIdMapping>,
-    ) -> io::Result<()> {
-        let (min, max) = if self.val_min > self.val_max {
-            (0, 0)
-        } else {
-            (self.val_min, self.val_max)
-        };
-
-        let fastfield_accessor = WriterFastFieldAccessProvider {
-            doc_id_map,
-            vals: &self.vals,
-            min_value: min,
-            max_value: max,
-            num_vals: self.val_count as u32,
-        };
-
-        serializer.create_auto_detect_u64_fast_field(self.field, fastfield_accessor)?;
-
-        Ok(())
-    }
-}
-
-#[derive(Clone)]
-struct WriterFastFieldAccessProvider<'map, 'bitp> {
-    doc_id_map: Option<&'map DocIdMapping>,
-    vals: &'bitp BlockedBitpacker,
-    min_value: u64,
-    max_value: u64,
-    num_vals: u32,
-}
-
-impl<'map, 'bitp> Column for WriterFastFieldAccessProvider<'map, 'bitp> {
-    /// Return the value associated with the given doc.
-    ///
-    /// Whenever possible use the Iterator passed to the fastfield creation instead, for performance
-    /// reasons.
-    ///
-    /// # Panics
-    ///
-    /// May panic if `doc` is greater than the index.
-    fn get_val(&self, _doc: u32) -> u64 {
-        unimplemented!()
-    }
-
-    fn iter(&self) -> Box<dyn Iterator<Item = u64> + '_> {
-        if let Some(doc_id_map) = self.doc_id_map {
-            Box::new(
-                doc_id_map
-                    .iter_old_doc_ids()
-                    .map(|doc_id| self.vals.get(doc_id as usize)),
-            )
-        } else {
-            Box::new(self.vals.iter())
-        }
-    }
-
-    fn min_value(&self) -> u64 {
-        self.min_value
-    }
-
-    fn max_value(&self) -> u64 {
-        self.max_value
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.num_vals
-    }
-}

src/indexer/doc_id_mapping.rs

@@ -113,34 +113,36 @@ pub(crate) fn get_doc_id_mapping_from_field(
     sort_by_field: IndexSortByField,
     segment_writer: &SegmentWriter,
 ) -> crate::Result<DocIdMapping> {
-    let schema = segment_writer.segment_serializer.segment().schema();
-    let field_id = expect_field_id_for_sort_field(&schema, &sort_by_field)?; // for now expect fastfield, but not strictly required
-    let fast_field = segment_writer
-        .fast_field_writers
-        .get_field_writer(field_id)
-        .ok_or_else(|| {
-            TantivyError::InvalidArgument(format!(
-                "sort index by field is required to be a fast field {:?}",
-                sort_by_field.field
-            ))
-        })?;
+    todo!()
+    // let schema = segment_writer.segment_serializer.segment().schema();
+    // let field_id = expect_field_id_for_sort_field(&schema, &sort_by_field)?; // for now expect
+    // fastfield, but not strictly required
+    // let fast_field = segment_writer
+    //     .fast_field_writers
+    //     .get_field_writer(field_id)
+    //     .ok_or_else(|| {
+    //         TantivyError::InvalidArgument(format!(
+    //             "sort index by field is required to be a fast field {:?}",
+    //             sort_by_field.field
+    //         ))
+    //     })?;
 
-    // create new doc_id to old doc_id index (used in fast_field_writers)
-    let mut doc_id_and_data = fast_field
-        .iter()
-        .enumerate()
-        .map(|el| (el.0 as DocId, el.1))
-        .collect::<Vec<_>>();
-    if sort_by_field.order == Order::Desc {
-        doc_id_and_data.sort_by_key(|k| Reverse(k.1));
-    } else {
-        doc_id_and_data.sort_by_key(|k| k.1);
-    }
-    let new_doc_id_to_old = doc_id_and_data
-        .into_iter()
-        .map(|el| el.0)
-        .collect::<Vec<_>>();
-    Ok(DocIdMapping::from_new_id_to_old_id(new_doc_id_to_old))
+    // // create new doc_id to old doc_id index (used in fast_field_writers)
+    // let mut doc_id_and_data = fast_field
+    //     .iter()
+    //     .enumerate()
+    //     .map(|el| (el.0 as DocId, el.1))
+    //     .collect::<Vec<_>>();
+    // if sort_by_field.order == Order::Desc {
+    //     doc_id_and_data.sort_by_key(|k| Reverse(k.1));
+    // } else {
+    //     doc_id_and_data.sort_by_key(|k| k.1);
+    // }
+    // let new_doc_id_to_old = doc_id_and_data
+    //     .into_iter()
+    //     .map(|el| el.0)
+    //     .collect::<Vec<_>>();
+    // Ok(DocIdMapping::from_new_id_to_old_id(new_doc_id_to_old))
 }
 
 #[cfg(test)]
@@ -159,15 +161,11 @@ mod tests_indexsorting {
 
         let my_text_field = schema_builder.add_text_field("text_field", text_field_options);
         let my_string_field = schema_builder.add_text_field("string_field", STRING | STORED);
-        let my_number = schema_builder.add_u64_field(
-            "my_number",
-            NumericOptions::default().set_fast(Cardinality::SingleValue),
-        );
+        let my_number =
+            schema_builder.add_u64_field("my_number", NumericOptions::default().set_fast());
 
-        let multi_numbers = schema_builder.add_u64_field(
-            "multi_numbers",
-            NumericOptions::default().set_fast(Cardinality::MultiValues),
-        );
+        let multi_numbers =
+            schema_builder.add_u64_field("multi_numbers", NumericOptions::default().set_fast());
 
         let schema = schema_builder.build();
         let mut index_builder = Index::builder().schema(schema);
@@ -441,47 +439,48 @@ mod tests_indexsorting {
         Ok(())
     }
 
-    #[test]
-    fn test_sort_index_fast_field() -> crate::Result<()> {
-        let index = create_test_index(
-            Some(IndexSettings {
-                sort_by_field: Some(IndexSortByField {
-                    field: "my_number".to_string(),
-                    order: Order::Asc,
-                }),
-                ..Default::default()
-            }),
-            get_text_options(),
-        )?;
-        assert_eq!(
-            index.settings().sort_by_field.as_ref().unwrap().field,
-            "my_number".to_string()
-        );
+    // #[test]
+    // fn test_sort_index_fast_field() -> crate::Result<()> {
+    //     let index = create_test_index(
+    //         Some(IndexSettings {
+    //             sort_by_field: Some(IndexSortByField {
+    //                 field: "my_number".to_string(),
+    //                 order: Order::Asc,
+    //             }),
+    //             ..Default::default()
+    //         }),
+    //         get_text_options(),
+    //     )?;
+    //     assert_eq!(
+    //         index.settings().sort_by_field.as_ref().unwrap().field,
+    //         "my_number".to_string()
+    //     );
 
-        let searcher = index.reader()?.searcher();
-        assert_eq!(searcher.segment_readers().len(), 1);
-        let segment_reader = searcher.segment_reader(0);
-        let fast_fields = segment_reader.fast_fields();
-        index.schema().get_field("my_number").unwrap();
+    //     let searcher = index.reader()?.searcher();
+    //     assert_eq!(searcher.segment_readers().len(), 1);
+    //     let segment_reader = searcher.segment_reader(0);
+    //     let fast_fields = segment_reader.fast_fields();
+    //     let my_number = index.schema().get_field("my_number").unwrap();
 
-        let fast_field = fast_fields.u64("my_number").unwrap();
-        assert_eq!(fast_field.get_val(0), 10u64);
-        assert_eq!(fast_field.get_val(1), 20u64);
-        assert_eq!(fast_field.get_val(2), 30u64);
+    //     let fast_field = fast_fields.u64(my_number).unwrap();
+    //     assert_eq!(fast_field.get_val(0), 10u64);
+    //     assert_eq!(fast_field.get_val(1), 20u64);
+    //     assert_eq!(fast_field.get_val(2), 30u64);
 
-        let multifield = fast_fields.u64s("multi_numbers").unwrap();
-        let mut vals = vec![];
-        multifield.get_vals(0u32, &mut vals);
-        assert_eq!(vals, &[] as &[u64]);
-        let mut vals = vec![];
-        multifield.get_vals(1u32, &mut vals);
-        assert_eq!(vals, &[5, 6]);
+    //     let multi_numbers = index.schema().get_field("multi_numbers").unwrap();
+    //     let multifield = fast_fields.u64s(multi_numbers).unwrap();
+    //     let mut vals = vec![];
+    //     multifield.get_vals(0u32, &mut vals);
+    //     assert_eq!(vals, &[] as &[u64]);
+    //     let mut vals = vec![];
+    //     multifield.get_vals(1u32, &mut vals);
+    //     assert_eq!(vals, &[5, 6]);
 
-        let mut vals = vec![];
-        multifield.get_vals(2u32, &mut vals);
-        assert_eq!(vals, &[3]);
-        Ok(())
-    }
+    //     let mut vals = vec![];
+    //     multifield.get_vals(2u32, &mut vals);
+    //     assert_eq!(vals, &[3]);
+    //     Ok(())
+    // }
 
     #[test]
     fn test_doc_mapping() {

src/indexer/json_term_writer.rs

@@ -1,4 +1,4 @@
-use fastfield_codecs::MonotonicallyMappableToU64;
+use columnar::MonotonicallyMappableToU64;
 use murmurhash32::murmurhash2;
 use rustc_hash::FxHashMap;
 
@@ -150,7 +150,6 @@ fn index_json_value(
                     json_term_writer.term_buffer,
                     ctx,
                     indexing_position,
-                    None,
                 );
             }
             TextOrDateTime::DateTime(dt) => {

src/indexer/merger_sorted_index_test.rs

@@ -2,19 +2,17 @@
 mod tests {
     use crate::collector::TopDocs;
     use crate::core::Index;
-    use crate::fastfield::{AliveBitSet, MultiValuedFastFieldReader};
+    use crate::fastfield::AliveBitSet;
     use crate::query::QueryParser;
     use crate::schema::{
-        self, BytesOptions, Cardinality, Facet, FacetOptions, IndexRecordOption, NumericOptions,
+        self, BytesOptions, Facet, FacetOptions, IndexRecordOption, NumericOptions,
         TextFieldIndexing, TextOptions,
     };
     use crate::{DocAddress, DocSet, IndexSettings, IndexSortByField, Order, Postings, Term};
 
     fn create_test_index_posting_list_issue(index_settings: Option<IndexSettings>) -> Index {
         let mut schema_builder = schema::Schema::builder();
-        let int_options = NumericOptions::default()
-            .set_fast(Cardinality::SingleValue)
-            .set_indexed();
+        let int_options = NumericOptions::default().set_fast().set_indexed();
         let int_field = schema_builder.add_u64_field("intval", int_options);
 
         let facet_field = schema_builder.add_facet_field("facet", FacetOptions::default());
@@ -62,7 +60,7 @@ mod tests {
     ) -> crate::Result<Index> {
         let mut schema_builder = schema::Schema::builder();
         let int_options = NumericOptions::default()
-            .set_fast(Cardinality::SingleValue)
+            .set_fast()
             .set_stored()
             .set_indexed();
         let int_field = schema_builder.add_u64_field("intval", int_options);
@@ -71,10 +69,8 @@ mod tests {
         let bytes_field = schema_builder.add_bytes_field("bytes", bytes_options);
         let facet_field = schema_builder.add_facet_field("facet", FacetOptions::default());
 
-        let multi_numbers = schema_builder.add_u64_field(
-            "multi_numbers",
-            NumericOptions::default().set_fast(Cardinality::MultiValues),
-        );
+        let multi_numbers =
+            schema_builder.add_u64_field("multi_numbers", NumericOptions::default().set_fast());
         let text_field_options = TextOptions::default()
             .set_indexing_options(
                 TextFieldIndexing::default()
@@ -349,128 +345,130 @@ mod tests {
         }
     }
 
-    #[test]
-    fn test_merge_sorted_index_asc() {
-        let index = create_test_index(
-            Some(IndexSettings {
-                sort_by_field: Some(IndexSortByField {
-                    field: "intval".to_string(),
-                    order: Order::Asc,
-                }),
-                ..Default::default()
-            }),
-            false,
-        )
-        .unwrap();
+    // #[test]
+    // fn test_merge_sorted_index_asc() {
+    //     let index = create_test_index(
+    //         Some(IndexSettings {
+    //             sort_by_field: Some(IndexSortByField {
+    //                 field: "intval".to_string(),
+    //                 order: Order::Asc,
+    //             }),
+    //             ..Default::default()
+    //         }),
+    //         false,
+    //     )
+    //     .unwrap();
 
-        let int_field = index.schema().get_field("intval").unwrap();
-        let reader = index.reader().unwrap();
-        let searcher = reader.searcher();
-        assert_eq!(searcher.segment_readers().len(), 1);
-        let segment_reader = searcher.segment_readers().last().unwrap();
+    //     let int_field = index.schema().get_field("intval").unwrap();
+    //     let multi_numbers = index.schema().get_field("multi_numbers").unwrap();
+    //     let bytes_field = index.schema().get_field("bytes").unwrap();
+    //     let reader = index.reader().unwrap();
+    //     let searcher = reader.searcher();
+    //     assert_eq!(searcher.segment_readers().len(), 1);
+    //     let segment_reader = searcher.segment_readers().last().unwrap();
 
-        let fast_fields = segment_reader.fast_fields();
-        let fast_field = fast_fields.u64("intval").unwrap();
-        assert_eq!(fast_field.get_val(0), 1u64);
-        assert_eq!(fast_field.get_val(1), 2u64);
-        assert_eq!(fast_field.get_val(2), 3u64);
-        assert_eq!(fast_field.get_val(3), 10u64);
-        assert_eq!(fast_field.get_val(4), 20u64);
-        assert_eq!(fast_field.get_val(5), 1_000u64);
+    //     let fast_fields = segment_reader.fast_fields();
+    //     let fast_field = fast_fields.u64(int_field).unwrap();
+    //     assert_eq!(fast_field.get_val(0), 1u64);
+    //     assert_eq!(fast_field.get_val(1), 2u64);
+    //     assert_eq!(fast_field.get_val(2), 3u64);
+    //     assert_eq!(fast_field.get_val(3), 10u64);
+    //     assert_eq!(fast_field.get_val(4), 20u64);
+    //     assert_eq!(fast_field.get_val(5), 1_000u64);
 
-        let get_vals = |fast_field: &MultiValuedFastFieldReader<u64>, doc_id: u32| -> Vec<u64> {
-            let mut vals = vec![];
-            fast_field.get_vals(doc_id, &mut vals);
-            vals
-        };
-        let fast_fields = segment_reader.fast_fields();
-        let fast_field = fast_fields.u64s("multi_numbers").unwrap();
-        assert_eq!(&get_vals(&fast_field, 0), &[] as &[u64]);
-        assert_eq!(&get_vals(&fast_field, 1), &[2, 3]);
-        assert_eq!(&get_vals(&fast_field, 2), &[3, 4]);
-        assert_eq!(&get_vals(&fast_field, 3), &[10, 11]);
-        assert_eq!(&get_vals(&fast_field, 4), &[20]);
-        assert_eq!(&get_vals(&fast_field, 5), &[1001, 1002]);
+    //     let get_vals = |fast_field: &MultiValuedFastFieldReader<u64>, doc_id: u32| -> Vec<u64> {
+    //         let mut vals = vec![];
+    //         fast_field.get_vals(doc_id, &mut vals);
+    //         vals
+    //     };
+    //     let fast_fields = segment_reader.fast_fields();
+    //     let fast_field = fast_fields.u64s(multi_numbers).unwrap();
+    //     assert_eq!(&get_vals(&fast_field, 0), &[] as &[u64]);
+    //     assert_eq!(&get_vals(&fast_field, 1), &[2, 3]);
+    //     assert_eq!(&get_vals(&fast_field, 2), &[3, 4]);
+    //     assert_eq!(&get_vals(&fast_field, 3), &[10, 11]);
+    //     assert_eq!(&get_vals(&fast_field, 4), &[20]);
+    //     assert_eq!(&get_vals(&fast_field, 5), &[1001, 1002]);
 
-        let fast_field = fast_fields.bytes("bytes").unwrap();
-        assert_eq!(fast_field.get_bytes(0), &[] as &[u8]);
-        assert_eq!(fast_field.get_bytes(2), &[1, 2, 3]);
-        assert_eq!(fast_field.get_bytes(5), &[5, 5]);
+    //     let fast_field = fast_fields.bytes(bytes_field).unwrap();
+    //     assert_eq!(fast_field.get_bytes(0), &[] as &[u8]);
+    //     assert_eq!(fast_field.get_bytes(2), &[1, 2, 3]);
+    //     assert_eq!(fast_field.get_bytes(5), &[5, 5]);
 
-        // test new field norm mapping
-        {
-            let my_text_field = index.schema().get_field("text_field").unwrap();
-            let fieldnorm_reader = segment_reader.get_fieldnorms_reader(my_text_field).unwrap();
-            assert_eq!(fieldnorm_reader.fieldnorm(0), 0);
-            assert_eq!(fieldnorm_reader.fieldnorm(1), 4);
-            assert_eq!(fieldnorm_reader.fieldnorm(2), 2); // some text
-            assert_eq!(fieldnorm_reader.fieldnorm(3), 1);
-            assert_eq!(fieldnorm_reader.fieldnorm(5), 3); // the biggest num
-        }
+    //     // test new field norm mapping
+    //     {
+    //         let my_text_field = index.schema().get_field("text_field").unwrap();
+    //         let fieldnorm_reader = segment_reader.get_fieldnorms_reader(my_text_field).unwrap();
+    //         assert_eq!(fieldnorm_reader.fieldnorm(0), 0);
+    //         assert_eq!(fieldnorm_reader.fieldnorm(1), 4);
+    //         assert_eq!(fieldnorm_reader.fieldnorm(2), 2); // some text
+    //         assert_eq!(fieldnorm_reader.fieldnorm(3), 1);
+    //         assert_eq!(fieldnorm_reader.fieldnorm(5), 3); // the biggest num
+    //     }
 
-        let searcher = index.reader().unwrap().searcher();
-        {
-            let my_text_field = index.schema().get_field("text_field").unwrap();
+    //     let searcher = index.reader().unwrap().searcher();
+    //     {
+    //         let my_text_field = index.schema().get_field("text_field").unwrap();
 
-            let do_search = |term: &str| {
-                let query = QueryParser::for_index(&index, vec![my_text_field])
-                    .parse_query(term)
-                    .unwrap();
-                let top_docs: Vec<(f32, DocAddress)> =
-                    searcher.search(&query, &TopDocs::with_limit(3)).unwrap();
+    //         let do_search = |term: &str| {
+    //             let query = QueryParser::for_index(&index, vec![my_text_field])
+    //                 .parse_query(term)
+    //                 .unwrap();
+    //             let top_docs: Vec<(f32, DocAddress)> =
+    //                 searcher.search(&query, &TopDocs::with_limit(3)).unwrap();
 
-                top_docs.iter().map(|el| el.1.doc_id).collect::<Vec<_>>()
-            };
+    //             top_docs.iter().map(|el| el.1.doc_id).collect::<Vec<_>>()
+    //         };
 
-            assert_eq!(do_search("some"), vec![2]);
-            assert_eq!(do_search("blubber"), vec![3]);
-            assert_eq!(do_search("biggest"), vec![5]);
-        }
+    //         assert_eq!(do_search("some"), vec![2]);
+    //         assert_eq!(do_search("blubber"), vec![3]);
+    //         assert_eq!(do_search("biggest"), vec![5]);
+    //     }
 
-        // postings file
-        {
-            let my_text_field = index.schema().get_field("text_field").unwrap();
-            let term_a = Term::from_field_text(my_text_field, "text");
-            let inverted_index = segment_reader.inverted_index(my_text_field).unwrap();
-            let mut postings = inverted_index
-                .read_postings(&term_a, IndexRecordOption::WithFreqsAndPositions)
-                .unwrap()
-                .unwrap();
+    //     // postings file
+    //     {
+    //         let my_text_field = index.schema().get_field("text_field").unwrap();
+    //         let term_a = Term::from_field_text(my_text_field, "text");
+    //         let inverted_index = segment_reader.inverted_index(my_text_field).unwrap();
+    //         let mut postings = inverted_index
+    //             .read_postings(&term_a, IndexRecordOption::WithFreqsAndPositions)
+    //             .unwrap()
+    //             .unwrap();
 
-            assert_eq!(postings.doc_freq(), 2);
-            let fallback_bitset = AliveBitSet::for_test_from_deleted_docs(&[0], 100);
-            assert_eq!(
-                postings.doc_freq_given_deletes(
-                    segment_reader.alive_bitset().unwrap_or(&fallback_bitset)
-                ),
-                2
-            );
+    //         assert_eq!(postings.doc_freq(), 2);
+    //         let fallback_bitset = AliveBitSet::for_test_from_deleted_docs(&[0], 100);
+    //         assert_eq!(
+    //             postings.doc_freq_given_deletes(
+    //                 segment_reader.alive_bitset().unwrap_or(&fallback_bitset)
+    //             ),
+    //             2
+    //         );
 
-            let mut output = vec![];
-            postings.positions(&mut output);
-            assert_eq!(output, vec![1, 3]);
-            postings.advance();
+    //         let mut output = vec![];
+    //         postings.positions(&mut output);
+    //         assert_eq!(output, vec![1, 3]);
+    //         postings.advance();
 
-            postings.positions(&mut output);
-            assert_eq!(output, vec![1]);
-        }
+    //         postings.positions(&mut output);
+    //         assert_eq!(output, vec![1]);
+    //     }
 
-        // access doc store
-        {
-            let doc = searcher.doc(DocAddress::new(0, 0)).unwrap();
-            assert_eq!(doc.get_first(int_field).unwrap().as_u64(), Some(1));
-            let doc = searcher.doc(DocAddress::new(0, 1)).unwrap();
-            assert_eq!(doc.get_first(int_field).unwrap().as_u64(), Some(2));
-            let doc = searcher.doc(DocAddress::new(0, 2)).unwrap();
-            assert_eq!(doc.get_first(int_field).unwrap().as_u64(), Some(3));
-            let doc = searcher.doc(DocAddress::new(0, 3)).unwrap();
-            assert_eq!(doc.get_first(int_field).unwrap().as_u64(), Some(10));
-            let doc = searcher.doc(DocAddress::new(0, 4)).unwrap();
-            assert_eq!(doc.get_first(int_field).unwrap().as_u64(), Some(20));
-            let doc = searcher.doc(DocAddress::new(0, 5)).unwrap();
-            assert_eq!(doc.get_first(int_field).unwrap().as_u64(), Some(1_000));
-        }
-    }
+    //     // access doc store
+    //     {
+    //         let doc = searcher.doc(DocAddress::new(0, 0)).unwrap();
+    //         assert_eq!(doc.get_first(int_field).unwrap().as_u64(), Some(1));
+    //         let doc = searcher.doc(DocAddress::new(0, 1)).unwrap();
+    //         assert_eq!(doc.get_first(int_field).unwrap().as_u64(), Some(2));
+    //         let doc = searcher.doc(DocAddress::new(0, 2)).unwrap();
+    //         assert_eq!(doc.get_first(int_field).unwrap().as_u64(), Some(3));
+    //         let doc = searcher.doc(DocAddress::new(0, 3)).unwrap();
+    //         assert_eq!(doc.get_first(int_field).unwrap().as_u64(), Some(10));
+    //         let doc = searcher.doc(DocAddress::new(0, 4)).unwrap();
+    //         assert_eq!(doc.get_first(int_field).unwrap().as_u64(), Some(20));
+    //         let doc = searcher.doc(DocAddress::new(0, 5)).unwrap();
+    //         assert_eq!(doc.get_first(int_field).unwrap().as_u64(), Some(1_000));
+    //     }
+    // }
 }
 
 #[cfg(all(test, feature = "unstable"))]
@@ -487,9 +485,7 @@ mod bench_sorted_index_merge {
     use crate::{IndexSettings, IndexSortByField, IndexWriter, Order};
     fn create_index(sort_by_field: Option<IndexSortByField>) -> Index {
         let mut schema_builder = Schema::builder();
-        let int_options = NumericOptions::default()
-            .set_fast(Cardinality::SingleValue)
-            .set_indexed();
+        let int_options = NumericOptions::default().set_fast().set_indexed();
         let int_field = schema_builder.add_u64_field("intval", int_options);
         let schema = schema_builder.build();
 

src/indexer/mod.rs

@@ -19,8 +19,8 @@ mod segment_register;
 pub mod segment_serializer;
 pub mod segment_updater;
 mod segment_writer;
-mod sorted_doc_id_column;
-mod sorted_doc_id_multivalue_column;
+// mod sorted_doc_id_column;
+// mod sorted_doc_id_multivalue_column;
 mod stamper;
 
 use crossbeam_channel as channel;
@@ -58,7 +58,7 @@ type AddBatchReceiver = channel::Receiver<AddBatch>;
 #[cfg(test)]
 mod tests_mmap {
     use crate::collector::Count;
-    use crate::query::QueryParser;
+    // use crate::query::QueryParser;
     use crate::schema::{JsonObjectOptions, Schema, TEXT};
     use crate::{Index, Term};
 
@@ -79,45 +79,45 @@ mod tests_mmap {
         Ok(())
     }
 
-    #[test]
-    fn test_json_field_expand_dots_disabled_dot_escaped_required() {
-        let mut schema_builder = Schema::builder();
-        let json_field = schema_builder.add_json_field("json", TEXT);
-        let index = Index::create_in_ram(schema_builder.build());
-        let mut index_writer = index.writer_for_tests().unwrap();
-        let json = serde_json::json!({"k8s.container.name": "prometheus", "val": "hello"});
-        index_writer.add_document(doc!(json_field=>json)).unwrap();
-        index_writer.commit().unwrap();
-        let reader = index.reader().unwrap();
-        let searcher = reader.searcher();
-        assert_eq!(searcher.num_docs(), 1);
-        let parse_query = QueryParser::for_index(&index, Vec::new());
-        let query = parse_query
-            .parse_query(r#"json.k8s\.container\.name:prometheus"#)
-            .unwrap();
-        let num_docs = searcher.search(&query, &Count).unwrap();
-        assert_eq!(num_docs, 1);
-    }
+    // #[test]
+    // fn test_json_field_expand_dots_disabled_dot_escaped_required() {
+    //     let mut schema_builder = Schema::builder();
+    //     let json_field = schema_builder.add_json_field("json", TEXT);
+    //     let index = Index::create_in_ram(schema_builder.build());
+    //     let mut index_writer = index.writer_for_tests().unwrap();
+    //     let json = serde_json::json!({"k8s.container.name": "prometheus", "val": "hello"});
+    //     index_writer.add_document(doc!(json_field=>json)).unwrap();
+    //     index_writer.commit().unwrap();
+    //     let reader = index.reader().unwrap();
+    //     let searcher = reader.searcher();
+    //     assert_eq!(searcher.num_docs(), 1);
+    //     let parse_query = QueryParser::for_index(&index, Vec::new());
+    //     let query = parse_query
+    //         .parse_query(r#"json.k8s\.container\.name:prometheus"#)
+    //         .unwrap();
+    //     let num_docs = searcher.search(&query, &Count).unwrap();
+    //     assert_eq!(num_docs, 1);
+    // }
 
-    #[test]
-    fn test_json_field_expand_dots_enabled_dot_escape_not_required() {
-        let mut schema_builder = Schema::builder();
-        let json_options: JsonObjectOptions =
-            JsonObjectOptions::from(TEXT).set_expand_dots_enabled();
-        let json_field = schema_builder.add_json_field("json", json_options);
-        let index = Index::create_in_ram(schema_builder.build());
-        let mut index_writer = index.writer_for_tests().unwrap();
-        let json = serde_json::json!({"k8s.container.name": "prometheus", "val": "hello"});
-        index_writer.add_document(doc!(json_field=>json)).unwrap();
-        index_writer.commit().unwrap();
-        let reader = index.reader().unwrap();
-        let searcher = reader.searcher();
-        assert_eq!(searcher.num_docs(), 1);
-        let parse_query = QueryParser::for_index(&index, Vec::new());
-        let query = parse_query
-            .parse_query(r#"json.k8s.container.name:prometheus"#)
-            .unwrap();
-        let num_docs = searcher.search(&query, &Count).unwrap();
-        assert_eq!(num_docs, 1);
-    }
+    // #[test]
+    // fn test_json_field_expand_dots_enabled_dot_escape_not_required() {
+    //     let mut schema_builder = Schema::builder();
+    //     let json_options: JsonObjectOptions =
+    //         JsonObjectOptions::from(TEXT).set_expand_dots_enabled();
+    //     let json_field = schema_builder.add_json_field("json", json_options);
+    //     let index = Index::create_in_ram(schema_builder.build());
+    //     let mut index_writer = index.writer_for_tests().unwrap();
+    //     let json = serde_json::json!({"k8s.container.name": "prometheus", "val": "hello"});
+    //     index_writer.add_document(doc!(json_field=>json)).unwrap();
+    //     index_writer.commit().unwrap();
+    //     let reader = index.reader().unwrap();
+    //     let searcher = reader.searcher();
+    //     assert_eq!(searcher.num_docs(), 1);
+    //     let parse_query = QueryParser::for_index(&index, Vec::new());
+    //     let query = parse_query
+    //         .parse_query(r#"json.k8s.container.name:prometheus"#)
+    //         .unwrap();
+    //     let num_docs = searcher.search(&query, &Count).unwrap();
+    //     assert_eq!(num_docs, 1);
+    // }
 }

src/indexer/segment_serializer.rs

@@ -1,5 +1,7 @@
+use common::TerminatingWrite;
+
 use crate::core::{Segment, SegmentComponent};
-use crate::fastfield::CompositeFastFieldSerializer;
+use crate::directory::WritePtr;
 use crate::fieldnorm::FieldNormsSerializer;
 use crate::postings::InvertedIndexSerializer;
 use crate::store::StoreWriter;
@@ -9,7 +11,7 @@ use crate::store::StoreWriter;
 pub struct SegmentSerializer {
     segment: Segment,
     pub(crate) store_writer: StoreWriter,
-    fast_field_serializer: CompositeFastFieldSerializer,
+    fast_field_write: WritePtr,
     fieldnorms_serializer: Option<FieldNormsSerializer>,
     postings_serializer: InvertedIndexSerializer,
 }
@@ -47,7 +49,6 @@ impl SegmentSerializer {
         };
 
         let fast_field_write = segment.open_write(SegmentComponent::FastFields)?;
-        let fast_field_serializer = CompositeFastFieldSerializer::from_write(fast_field_write)?;
 
         let fieldnorms_write = segment.open_write(SegmentComponent::FieldNorms)?;
         let fieldnorms_serializer = FieldNormsSerializer::from_write(fieldnorms_write)?;
@@ -56,7 +57,7 @@ impl SegmentSerializer {
         Ok(SegmentSerializer {
             segment,
             store_writer,
-            fast_field_serializer,
+            fast_field_write,
             fieldnorms_serializer: Some(fieldnorms_serializer),
             postings_serializer,
         })
@@ -81,8 +82,8 @@ impl SegmentSerializer {
     }
 
     /// Accessor to the `FastFieldSerializer`.
-    pub fn get_fast_field_serializer(&mut self) -> &mut CompositeFastFieldSerializer {
-        &mut self.fast_field_serializer
+    pub fn get_fast_field_write(&mut self) -> &mut WritePtr {
+        &mut self.fast_field_write
     }
 
     /// Extract the field norm serializer.
@@ -102,7 +103,7 @@ impl SegmentSerializer {
         if let Some(fieldnorms_serializer) = self.extract_fieldnorms_serializer() {
             fieldnorms_serializer.close()?;
         }
-        self.fast_field_serializer.close()?;
+        self.fast_field_write.terminate()?;
         self.postings_serializer.close()?;
         self.store_writer.close()?;
         Ok(())

src/indexer/segment_writer.rs

@@ -1,4 +1,4 @@
-use fastfield_codecs::MonotonicallyMappableToU64;
+use columnar::MonotonicallyMappableToU64;
 use itertools::Itertools;
 
 use super::doc_id_mapping::{get_doc_id_mapping_from_field, DocIdMapping};
@@ -139,7 +139,6 @@ impl SegmentWriter {
             self.ctx,
             self.fast_field_writers,
             &self.fieldnorms_writer,
-            &self.schema,
             self.segment_serializer,
             mapping.as_ref(),
         )?;
@@ -185,22 +184,15 @@ impl SegmentWriter {
                     for value in values {
                         let facet = value.as_facet().ok_or_else(make_schema_error)?;
                         let facet_str = facet.encoded_str();
-                        let mut unordered_term_id_opt = None;
-                        FacetTokenizer
-                            .token_stream(facet_str)
-                            .process(&mut |token| {
-                                term_buffer.set_text(&token.text);
-                                let unordered_term_id =
-                                    postings_writer.subscribe(doc_id, 0u32, term_buffer, ctx);
-                                // TODO pass indexing context directly in subscribe function
-                                unordered_term_id_opt = Some(unordered_term_id);
-                            });
-                        if let Some(unordered_term_id) = unordered_term_id_opt {
-                            self.fast_field_writers
-                                .get_term_id_writer_mut(field)
-                                .expect("writer for facet missing")
-                                .add_val(unordered_term_id);
-                        }
+                        let mut facet_tokenizer = FacetTokenizer.token_stream(facet_str);
+                        let mut indexing_position = IndexingPosition::default();
+                        postings_writer.index_text(
+                            doc_id,
+                            &mut *facet_tokenizer,
+                            term_buffer,
+                            ctx,
+                            &mut indexing_position,
+                        );
                     }
                 }
                 FieldType::Str(_) => {
@@ -227,7 +219,6 @@ impl SegmentWriter {
                             term_buffer,
                             ctx,
                             &mut indexing_position,
-                            self.fast_field_writers.get_term_id_writer_mut(field),
                         );
                     }
                     if field_entry.has_fieldnorms() {
@@ -383,7 +374,6 @@ fn remap_and_write(
     ctx: IndexingContext,
     fast_field_writers: FastFieldsWriter,
     fieldnorms_writer: &FieldNormsWriter,
-    schema: &Schema,
     mut serializer: SegmentSerializer,
     doc_id_map: Option<&DocIdMapping>,
 ) -> crate::Result<()> {
@@ -395,20 +385,15 @@ fn remap_and_write(
         .segment()
         .open_read(SegmentComponent::FieldNorms)?;
     let fieldnorm_readers = FieldNormReaders::open(fieldnorm_data)?;
-    let term_ord_map = serialize_postings(
+    serialize_postings(
         ctx,
         per_field_postings_writers,
         fieldnorm_readers,
         doc_id_map,
-        schema,
         serializer.get_postings_serializer(),
     )?;
     debug!("fastfield-serialize");
-    fast_field_writers.serialize(
-        serializer.get_fast_field_serializer(),
-        &term_ord_map,
-        doc_id_map,
-    )?;
+    fast_field_writers.serialize(serializer.get_fast_field_write(), doc_id_map)?;
 
     // finalize temp docstore and create version, which reflects the doc_id_map
     if let Some(doc_id_map) = doc_id_map {

src/lib.rs

@@ -147,6 +147,22 @@ pub struct DateTime {
     pub(crate) timestamp_micros: i64,
 }
 
+impl From<columnar::DateTime> for DateTime {
+    fn from(columnar_datetime: columnar::DateTime) -> Self {
+        DateTime {
+            timestamp_micros: columnar_datetime.timestamp_micros,
+        }
+    }
+}
+
+impl From<DateTime> for columnar::DateTime {
+    fn from(datetime: crate::DateTime) -> Self {
+        columnar::DateTime {
+            timestamp_micros: datetime.timestamp_micros,
+        }
+    }
+}
+
 impl DateTime {
     /// Create new from UNIX timestamp in seconds
     pub const fn from_timestamp_secs(seconds: i64) -> Self {
@@ -263,7 +279,7 @@ mod indexer;
 pub mod error;
 pub mod tokenizer;
 
-pub mod aggregation;
+// pub mod aggregation;
 pub mod collector;
 pub mod directory;
 pub mod fastfield;