add truncation comment

Fixes #1783

Cargo.toml

@@ -107,7 +107,7 @@ 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", "columnar", "tokenizer-api"]
 
 # Following the "fail" crate best practises, we isolate
 # tests that define specific behavior in fail check points

README.md

@@ -41,7 +41,7 @@ Your mileage WILL vary depending on the nature of queries and their load.
 - SIMD integer compression when the platform/CPU includes the SSE2 instruction set
 - Single valued and multivalued u64, i64, and f64 fast fields (equivalent of doc values in Lucene)
 - `&[u8]` fast fields
-- Text, i64, u64, f64, dates, ip, bool, and hierarchical facet fields
+- Text, i64, u64, f64, dates, and hierarchical facet fields
 - Compressed document store (LZ4, Zstd, None, Brotli, Snap)
 - Range queries
 - Faceted search
@@ -80,21 +80,56 @@ There are many ways to support this project.
 # Contributing code
 
 We use the GitHub Pull Request workflow: reference a GitHub ticket and/or include a comprehensive commit message when opening a PR.
-Feel free to update CHANGELOG.md with your contribution.
 
 ## Tokenizer
 
 When implementing a tokenizer for tantivy depend on the `tantivy-tokenizer-api` crate.
 
+## Minimum supported Rust version
+
+Tantivy currently requires at least Rust 1.62 or later to compile.
+
 ## Clone and build locally
 
 Tantivy compiles on stable Rust.
 To check out and run tests, you can simply run:
 
 ```bash
-git clone https://github.com/quickwit-oss/tantivy.git
-cd tantivy
-cargo test
+    git clone https://github.com/quickwit-oss/tantivy.git
+    cd tantivy
+    cargo build
+```
+
+## Run tests
+
+Some tests will not run with just `cargo test` because of `fail-rs`.
+To run the tests exhaustively, run `./run-tests.sh`.
+
+## Debug
+
+You might find it useful to step through the programme with a debugger.
+
+### A failing test
+
+Make sure you haven't run `cargo clean` after the most recent `cargo test` or `cargo build` to guarantee that the `target/` directory exists. Use this bash script to find the name of the most recent debug build of Tantivy and run it under `rust-gdb`:
+
+```bash
+find target/debug/ -maxdepth 1 -executable -type f -name "tantivy*" -printf '%TY-%Tm-%Td %TT %p\n' | sort -r | cut -d " " -f 3 | xargs -I RECENT_DBG_TANTIVY rust-gdb RECENT_DBG_TANTIVY
+```
+
+Now that you are in `rust-gdb`, you can set breakpoints on lines and methods that match your source code and run the debug executable with flags that you normally pass to `cargo test` like this:
+
+```bash
+$gdb run --test-threads 1 --test $NAME_OF_TEST
+```
+
+### An example
+
+By default, `rustc` compiles everything in the `examples/` directory in debug mode. This makes it easy for you to make examples to reproduce bugs:
+
+```bash
+rust-gdb target/debug/examples/$EXAMPLE_NAME
+$ gdb run
 ```
 
 # Companies Using Tantivy

benches/index-bench.rs

@@ -34,7 +34,7 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
             let index = Index::create_in_ram(schema.clone());
             let index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
             for _ in 0..NUM_REPEATS {
-                for doc_json in HDFS_LOGS.trim().split('\n') {
+                for doc_json in HDFS_LOGS.trim().split("\n") {
                     let doc = schema.parse_document(doc_json).unwrap();
                     index_writer.add_document(doc).unwrap();
                 }
@@ -46,7 +46,7 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
             let index = Index::create_in_ram(schema.clone());
             let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
             for _ in 0..NUM_REPEATS {
-                for doc_json in HDFS_LOGS.trim().split('\n') {
+                for doc_json in HDFS_LOGS.trim().split("\n") {
                     let doc = schema.parse_document(doc_json).unwrap();
                     index_writer.add_document(doc).unwrap();
                 }
@@ -59,7 +59,7 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
             let index = Index::create_in_ram(schema_with_store.clone());
             let index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
             for _ in 0..NUM_REPEATS {
-                for doc_json in HDFS_LOGS.trim().split('\n') {
+                for doc_json in HDFS_LOGS.trim().split("\n") {
                     let doc = schema.parse_document(doc_json).unwrap();
                     index_writer.add_document(doc).unwrap();
                 }
@@ -71,7 +71,7 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
             let index = Index::create_in_ram(schema_with_store.clone());
             let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
             for _ in 0..NUM_REPEATS {
-                for doc_json in HDFS_LOGS.trim().split('\n') {
+                for doc_json in HDFS_LOGS.trim().split("\n") {
                     let doc = schema.parse_document(doc_json).unwrap();
                     index_writer.add_document(doc).unwrap();
                 }
@@ -85,7 +85,7 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
             let json_field = dynamic_schema.get_field("json").unwrap();
             let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
             for _ in 0..NUM_REPEATS {
-                for doc_json in HDFS_LOGS.trim().split('\n') {
+                for doc_json in HDFS_LOGS.trim().split("\n") {
                     let json_val: serde_json::Map<String, serde_json::Value> =
                         serde_json::from_str(doc_json).unwrap();
                     let doc = tantivy::doc!(json_field=>json_val);
@@ -101,7 +101,7 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
             let json_field = dynamic_schema.get_field("json").unwrap();
             let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
             for _ in 0..NUM_REPEATS {
-                for doc_json in HDFS_LOGS.trim().split('\n') {
+                for doc_json in HDFS_LOGS.trim().split("\n") {
                     let json_val: serde_json::Map<String, serde_json::Value> =
                         serde_json::from_str(doc_json).unwrap();
                     let doc = tantivy::doc!(json_field=>json_val);

bitpacker/Cargo.toml

@@ -15,7 +15,3 @@ homepage = "https://github.com/quickwit-oss/tantivy"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-
-[dev-dependencies]
-rand = "0.8"
-proptest = "1"

bitpacker/benches/bench.rs

@@ -4,39 +4,9 @@ extern crate test;
 
 #[cfg(test)]
 mod tests {
-    use rand::seq::IteratorRandom;
-    use rand::thread_rng;
-    use tantivy_bitpacker::{BitPacker, BitUnpacker, BlockedBitpacker};
+    use tantivy_bitpacker::BlockedBitpacker;
     use test::Bencher;
 
-    #[inline(never)]
-    fn create_bitpacked_data(bit_width: u8, num_els: u32) -> Vec<u8> {
-        let mut bitpacker = BitPacker::new();
-        let mut buffer = Vec::new();
-        for _ in 0..num_els {
-            // the values do not matter.
-            bitpacker.write(0u64, bit_width, &mut buffer).unwrap();
-            bitpacker.flush(&mut buffer).unwrap();
-        }
-        buffer
-    }
-
-    #[bench]
-    fn bench_bitpacking_read(b: &mut Bencher) {
-        let bit_width = 3;
-        let num_els = 1_000_000u32;
-        let bit_unpacker = BitUnpacker::new(bit_width);
-        let data = create_bitpacked_data(bit_width, num_els);
-        let idxs: Vec<u32> = (0..num_els).choose_multiple(&mut thread_rng(), 100_000);
-        b.iter(|| {
-            let mut out = 0u64;
-            for &idx in &idxs {
-                out = out.wrapping_add(bit_unpacker.get(idx, &data[..]));
-            }
-            out
-        });
-    }
-
     #[bench]
     fn bench_blockedbitp_read(b: &mut Bencher) {
         let mut blocked_bitpacker = BlockedBitpacker::new();
@@ -44,9 +14,9 @@ mod tests {
             blocked_bitpacker.add(val * val);
         }
         b.iter(|| {
-            let mut out = 0u64;
+            let mut out = 0;
             for val in 0..=21500 {
-                out = out.wrapping_add(blocked_bitpacker.get(val));
+                out = blocked_bitpacker.get(val);
             }
             out
         });

bitpacker/src/bitpacker.rs

@@ -56,31 +56,27 @@ impl BitPacker {
 
     pub fn close<TWrite: io::Write>(&mut self, output: &mut TWrite) -> io::Result<()> {
         self.flush(output)?;
+        // Padding the write file to simplify reads.
+        output.write_all(&[0u8; 7])?;
         Ok(())
     }
 }
 
-#[derive(Clone, Debug, Default, Copy)]
+#[derive(Clone, Debug, Default)]
 pub struct BitUnpacker {
-    num_bits: u32,
+    num_bits: u64,
     mask: u64,
 }
 
 impl BitUnpacker {
-    /// Creates a bit unpacker, that assumes the same bitwidth for all values.
-    ///
-    /// The bitunpacker works by doing an unaligned read of 8 bytes.
-    /// For this reason, values of `num_bits` between
-    /// [57..63] are forbidden.
     pub fn new(num_bits: u8) -> BitUnpacker {
-        assert!(num_bits <= 7 * 8 || num_bits == 64);
         let mask: u64 = if num_bits == 64 {
             !0u64
         } else {
             (1u64 << num_bits) - 1u64
         };
         BitUnpacker {
-            num_bits: u32::from(num_bits),
+            num_bits: u64::from(num_bits),
             mask,
         }
     }
@@ -91,40 +87,28 @@ impl BitUnpacker {
 
     #[inline]
     pub fn get(&self, idx: u32, data: &[u8]) -> u64 {
-        let addr_in_bits = idx * self.num_bits;
-        let addr = (addr_in_bits >> 3) as usize;
-        if addr + 8 > data.len() {
-            if self.num_bits == 0 {
-                return 0;
-            }
-            let bit_shift = addr_in_bits & 7;
-            return self.get_slow_path(addr, bit_shift, data);
+        if self.num_bits == 0 {
+            return 0u64;
         }
+        let addr_in_bits = idx * self.num_bits as u32;
+        let addr = (addr_in_bits >> 3) as usize;
         let bit_shift = addr_in_bits & 7;
+        debug_assert!(
+            addr + 8 <= data.len(),
+            "The fast field field should have been padded with 7 bytes."
+        );
         let bytes: [u8; 8] = (&data[addr..addr + 8]).try_into().unwrap();
         let val_unshifted_unmasked: u64 = u64::from_le_bytes(bytes);
         let val_shifted = val_unshifted_unmasked >> bit_shift;
         val_shifted & self.mask
     }
-
-    #[inline(never)]
-    fn get_slow_path(&self, addr: usize, bit_shift: u32, data: &[u8]) -> u64 {
-        let mut bytes: [u8; 8] = [0u8; 8];
-        let available_bytes = data.len() - addr;
-        // This function is meant to only be called if we did not have 8 bytes to load.
-        debug_assert!(available_bytes < 8);
-        bytes[..available_bytes].copy_from_slice(&data[addr..]);
-        let val_unshifted_unmasked: u64 = u64::from_le_bytes(bytes);
-        let val_shifted = val_unshifted_unmasked >> bit_shift;
-        val_shifted & self.mask
-    }
 }
 
 #[cfg(test)]
 mod test {
     use super::{BitPacker, BitUnpacker};
 
-    fn create_bitpacker(len: usize, num_bits: u8) -> (BitUnpacker, Vec<u64>, Vec<u8>) {
+    fn create_fastfield_bitpacker(len: usize, num_bits: u8) -> (BitUnpacker, Vec<u64>, Vec<u8>) {
         let mut data = Vec::new();
         let mut bitpacker = BitPacker::new();
         let max_val: u64 = (1u64 << num_bits as u64) - 1u64;
@@ -135,13 +119,13 @@ mod test {
             bitpacker.write(val, num_bits, &mut data).unwrap();
         }
         bitpacker.close(&mut data).unwrap();
-        assert_eq!(data.len(), ((num_bits as usize) * len + 7) / 8);
+        assert_eq!(data.len(), ((num_bits as usize) * len + 7) / 8 + 7);
         let bitunpacker = BitUnpacker::new(num_bits);
         (bitunpacker, vals, data)
     }
 
     fn test_bitpacker_util(len: usize, num_bits: u8) {
-        let (bitunpacker, vals, data) = create_bitpacker(len, num_bits);
+        let (bitunpacker, vals, data) = create_fastfield_bitpacker(len, num_bits);
         for (i, val) in vals.iter().enumerate() {
             assert_eq!(bitunpacker.get(i as u32, &data), *val);
         }
@@ -155,49 +139,4 @@ mod test {
         test_bitpacker_util(6, 14);
         test_bitpacker_util(1000, 14);
     }
-
-    use proptest::prelude::*;
-
-    fn num_bits_strategy() -> impl Strategy<Value = u8> {
-        prop_oneof!(Just(0), Just(1), 2u8..56u8, Just(56), Just(64),)
-    }
-
-    fn vals_strategy() -> impl Strategy<Value = (u8, Vec<u64>)> {
-        (num_bits_strategy(), 0usize..100usize).prop_flat_map(|(num_bits, len)| {
-            let max_val = if num_bits == 64 {
-                u64::MAX
-            } else {
-                (1u64 << num_bits as u32) - 1
-            };
-            let vals = proptest::collection::vec(0..=max_val, len);
-            vals.prop_map(move |vals| (num_bits, vals))
-        })
-    }
-
-    fn test_bitpacker_aux(num_bits: u8, vals: &[u64]) {
-        let mut buffer: Vec<u8> = Vec::new();
-        let mut bitpacker = BitPacker::new();
-        for &val in vals {
-            bitpacker.write(val, num_bits, &mut buffer).unwrap();
-        }
-        bitpacker.flush(&mut buffer).unwrap();
-        assert_eq!(buffer.len(), (vals.len() * num_bits as usize + 7) / 8);
-        let bitunpacker = BitUnpacker::new(num_bits);
-        let max_val = if num_bits == 64 {
-            u64::MAX
-        } else {
-            (1u64 << num_bits) - 1
-        };
-        for (i, val) in vals.iter().copied().enumerate() {
-            assert!(val <= max_val);
-            assert_eq!(bitunpacker.get(i as u32, &buffer), val);
-        }
-    }
-
-    proptest::proptest! {
-        #[test]
-        fn test_bitpacker_proptest((num_bits, vals) in vals_strategy()) {
-            test_bitpacker_aux(num_bits, &vals);
-        }
-    }
 }

columnar/Cargo.toml

@@ -11,22 +11,8 @@ thiserror = "1"
 fnv = "1"
 sstable = { path = "../sstable", package = "tantivy-sstable" }
 common = { path = "../common", package = "tantivy-common" }
+fastfield_codecs = { path = "../fastfield_codecs"}
 itertools = "0.10"
-log = "0.4"
-tantivy-bitpacker = { version= "0.3", path = "../bitpacker/" }
-prettytable-rs = {version="0.10.0", optional= true}
-rand = {version="0.8.3", optional= true}
-fastdivide = "0.4"
-measure_time = { version="0.8.2", optional=true}
 
 [dev-dependencies]
 proptest = "1"
-more-asserts = "0.3.0"
-rand = "0.8.3"
-
-# temporary
-[workspace]
-members = []
-
-[features]
-unstable = []

columnar/README.md

@@ -6,12 +6,13 @@ This crate describes columnar format used in tantivy.
 
 This format is special in the following way.
 - it needs to be compact
-- accessing a specific column does not require to load the entire columnar. It can be done in 2 to 3 random access.
+- it does not required to be loaded in memory.
+- it is designed to fit well with quickwit's strange constraint:
+we need to be able to load columns rapidly.
 - columns of several types can be associated with the same column name.
 - it needs to support columns with different types `(str, u64, i64, f64)`
 and different cardinality `(required, optional, multivalued)`.
 - columns, once loaded, offer cheap random access.
-- it is designed to allow range queries.
 
 # Coercion rules
 
@@ -64,46 +65,3 @@ be done by listing all keys prefixed by
 
 The associated range of bytes refer to a range of bytes
 
-This crate exposes a columnar format for tantivy.
-This format is described in README.md
-
-
-The crate introduces the following concepts.
-
-`Columnar` is an equivalent of a dataframe.
-It maps `column_key` to `Column`.
-
-A `Column<T>` asssociates a `RowId` (u32) to any
-number of values.
-
-This is made possible by wrapping a `ColumnIndex` and a `ColumnValue` object.
-The `ColumnValue<T>` represents a mapping that associates each `RowId` to
-exactly one single value.
-
-The `ColumnIndex` then maps each RowId to a set of `RowId` in the
-`ColumnValue`.
-
-For optimization, and compression purposes, the `ColumnIndex` has three
-possible representation, each for different cardinalities.
-
-- Full
-
-All RowId have exactly one value. The ColumnIndex is the trivial mapping.
-
-- Optional
-
-All RowIds can have at most one value. The ColumnIndex is the trivial mapping `ColumnRowId -> Option<ColumnValueRowId>`.
-
-- Multivalued
-
-All RowIds can have any number of values.
-The column index is mapping values to a range.
-
-
-All these objects are implemented an unit tested independently
-in their own module:
-
-- columnar
-- column_index
-- column_values
-- column

columnar/src/TODO.md

@@ -1,46 +0,0 @@
-# zero to one
-* merges
-* full still needs a num_values
-* replug u128
-* add dictionary encoded stuff
-* fix multivalued
-* find a way to make columnar work with strict types
-* plug to tantivy
-    - indexing
-    - aggregations
-    - merge
-
-# Perf and Size
-* re-add ZSTD compression for dictionaries
-no systematic monotonic mapping
-consider removing multilinear
-f32?
-adhoc solution for bool?
-
-add metrics helper for aggregate. sum(row_id)
-review inline absence/presence
-improv perf of select using PDEP
-compare with roaring bitmap/elias fano etc etc.
-SIMD range? (see blog post)
-Add alignment?
-Consider another codec to bridge the gap between few and 5k elements
-
-# Cleanup and rationalization
-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
-Review the public API via cargo doc
-go through TODOs
-remove all  doc_id occurences -> row_id
-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
-
-# Other
-fix enhance column-cli
-
-# Santa claus
-
-autodetect datetime ipaddr, plug customizable tokenizer.
-

columnar/src/column/dictionary_encoded.rs

@@ -1,78 +0,0 @@
-use std::io;
-use std::ops::Deref;
-use std::sync::Arc;
-
-use sstable::{Dictionary, VoidSSTable};
-
-use crate::column::Column;
-use crate::RowId;
-
-/// Dictionary encoded column.
-///
-/// The column simply gives access to a regular u64-column that, in
-/// which the values are term-ordinals.
-///
-/// These ordinals are ids uniquely identify the bytes that are stored in
-/// the column. These ordinals are small, and sorted in the same order
-/// as the term_ord_column.
-#[derive(Clone)]
-pub struct BytesColumn {
-    pub(crate) dictionary: Arc<Dictionary<VoidSSTable>>,
-    pub(crate) term_ord_column: Column<u64>,
-}
-
-impl BytesColumn {
-    /// Fills the given `output` buffer with the term associated to the ordinal `ord`.
-    ///
-    /// Returns `false` if the term does not exist (e.g. `term_ord` is greater or equal to the
-    /// overll number of terms).
-    pub fn ord_to_bytes(&self, ord: u64, output: &mut Vec<u8>) -> io::Result<bool> {
-        self.dictionary.ord_to_term(ord, output)
-    }
-
-    /// Returns the number of rows in the column.
-    pub fn num_rows(&self) -> RowId {
-        self.term_ord_column.num_rows()
-    }
-
-    /// Returns the column of ordinals
-    pub fn ords(&self) -> &Column<u64> {
-        &self.term_ord_column
-    }
-}
-
-#[derive(Clone)]
-pub struct StrColumn(BytesColumn);
-
-impl From<BytesColumn> for StrColumn {
-    fn from(bytes_col: BytesColumn) -> Self {
-        StrColumn(bytes_col)
-    }
-}
-
-impl StrColumn {
-    /// Fills the buffer
-    pub fn ord_to_str(&self, term_ord: u64, output: &mut String) -> io::Result<bool> {
-        unsafe {
-            let buf = output.as_mut_vec();
-            self.0.dictionary.ord_to_term(term_ord, buf)?;
-            // TODO consider remove checks if it hurts performance.
-            if std::str::from_utf8(buf.as_slice()).is_err() {
-                buf.clear();
-                return Err(io::Error::new(
-                    io::ErrorKind::InvalidData,
-                    "Not valid utf-8",
-                ));
-            }
-        }
-        Ok(true)
-    }
-}
-
-impl Deref for StrColumn {
-    type Target = BytesColumn;
-
-    fn deref(&self) -> &Self::Target {
-        &self.0
-    }
-}

columnar/src/column/mod.rs

@@ -1,109 +0,0 @@
-mod dictionary_encoded;
-mod serialize;
-
-use std::ops::Deref;
-use std::sync::Arc;
-
-use common::BinarySerializable;
-pub use dictionary_encoded::{BytesColumn, StrColumn};
-pub use serialize::{
-    open_column_bytes, open_column_u128, open_column_u64, serialize_column_mappable_to_u128,
-    serialize_column_mappable_to_u64,
-};
-
-use crate::column_index::ColumnIndex;
-use crate::column_values::ColumnValues;
-use crate::{Cardinality, RowId};
-
-#[derive(Clone)]
-pub struct Column<T> {
-    pub idx: ColumnIndex<'static>,
-    pub values: Arc<dyn ColumnValues<T>>,
-}
-
-impl<T: PartialOrd> Column<T> {
-    pub fn num_rows(&self) -> RowId {
-        match &self.idx {
-            ColumnIndex::Full => self.values.num_vals() as u32,
-            ColumnIndex::Optional(optional_index) => optional_index.num_rows(),
-            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
-            }
-        }
-    }
-
-    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()
-    }
-
-    pub fn values(&self, row_id: RowId) -> impl Iterator<Item = T> + '_ {
-        self.value_row_ids(row_id)
-            .map(|value_row_id: RowId| self.values.get_val(value_row_id))
-    }
-
-    pub fn first_or_default_col(self, default_value: T) -> Arc<dyn ColumnValues<T>> {
-        Arc::new(FirstValueWithDefault {
-            column: self,
-            default_value,
-        })
-    }
-}
-
-impl<T> Deref for Column<T> {
-    type Target = ColumnIndex<'static>;
-
-    fn deref(&self) -> &Self::Target {
-        &self.idx
-    }
-}
-
-impl BinarySerializable for Cardinality {
-    fn serialize<W: std::io::Write>(&self, writer: &mut W) -> std::io::Result<()> {
-        self.to_code().serialize(writer)
-    }
-
-    fn deserialize<R: std::io::Read>(reader: &mut R) -> std::io::Result<Self> {
-        let cardinality_code = u8::deserialize(reader)?;
-        let cardinality = Cardinality::try_from_code(cardinality_code)?;
-        Ok(cardinality)
-    }
-}
-
-// TODO simplify or optimize
-struct FirstValueWithDefault<T: Copy> {
-    column: Column<T>,
-    default_value: T,
-}
-
-impl<T: PartialOrd + 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)
-    }
-
-    fn min_value(&self) -> T {
-        self.column.values.min_value()
-    }
-
-    fn max_value(&self) -> T {
-        self.column.values.max_value()
-    }
-
-    fn num_vals(&self) -> u32 {
-        match &self.column.idx {
-            ColumnIndex::Full => self.column.values.num_vals(),
-            ColumnIndex::Optional(optional_idx) => optional_idx.num_rows(),
-            ColumnIndex::Multivalued(_) => todo!(),
-        }
-    }
-}

columnar/src/column/serialize.rs

@@ -1,101 +0,0 @@
-use std::io;
-use std::io::Write;
-use std::sync::Arc;
-
-use common::OwnedBytes;
-use sstable::Dictionary;
-
-use crate::column::{BytesColumn, Column};
-use crate::column_index::{serialize_column_index, SerializableColumnIndex};
-use crate::column_values::serialize::serialize_column_values_u128;
-use crate::column_values::{
-    serialize_column_values, ColumnValues, FastFieldCodecType, MonotonicallyMappableToU128,
-    MonotonicallyMappableToU64,
-};
-
-pub fn serialize_column_mappable_to_u128<
-    F: Fn() -> I,
-    I: Iterator<Item = T>,
-    T: MonotonicallyMappableToU128,
->(
-    column_index: SerializableColumnIndex<'_>,
-    column_values: F,
-    num_vals: u32,
-    output: &mut impl Write,
-) -> io::Result<()> {
-    let column_index_num_bytes = serialize_column_index(column_index, output)?;
-    serialize_column_values_u128(
-        || column_values().map(|val| val.to_u128()),
-        num_vals,
-        output,
-    )?;
-    output.write_all(&column_index_num_bytes.to_le_bytes())?;
-    Ok(())
-}
-
-pub fn serialize_column_mappable_to_u64<T: MonotonicallyMappableToU64>(
-    column_index: SerializableColumnIndex<'_>,
-    column_values: &impl ColumnValues<T>,
-    output: &mut impl Write,
-) -> io::Result<()> {
-    let column_index_num_bytes = serialize_column_index(column_index, output)?;
-    serialize_column_values(
-        column_values,
-        &[
-            FastFieldCodecType::Bitpacked,
-            FastFieldCodecType::BlockwiseLinear,
-        ],
-        output,
-    )?;
-    output.write_all(&column_index_num_bytes.to_le_bytes())?;
-    Ok(())
-}
-
-pub fn open_column_u64<T: MonotonicallyMappableToU64>(bytes: OwnedBytes) -> io::Result<Column<T>> {
-    let (body, column_index_num_bytes_payload) = bytes.rsplit(4);
-    let column_index_num_bytes = u32::from_le_bytes(
-        column_index_num_bytes_payload
-            .as_slice()
-            .try_into()
-            .unwrap(),
-    );
-    let (column_index_data, column_values_data) = body.split(column_index_num_bytes as usize);
-    let column_index = crate::column_index::open_column_index(column_index_data)?;
-    let column_values = crate::column_values::open_u64_mapped(column_values_data)?;
-    Ok(Column {
-        idx: column_index,
-        values: column_values,
-    })
-}
-
-pub fn open_column_u128<T: MonotonicallyMappableToU128>(
-    bytes: OwnedBytes,
-) -> io::Result<Column<T>> {
-    let (body, column_index_num_bytes_payload) = bytes.rsplit(4);
-    let column_index_num_bytes = u32::from_le_bytes(
-        column_index_num_bytes_payload
-            .as_slice()
-            .try_into()
-            .unwrap(),
-    );
-    let (column_index_data, column_values_data) = body.split(column_index_num_bytes as usize);
-    let column_index = crate::column_index::open_column_index(column_index_data)?;
-    let column_values = crate::column_values::open_u128_mapped(column_values_data)?;
-    Ok(Column {
-        idx: column_index,
-        values: column_values,
-    })
-}
-
-pub fn open_column_bytes<T: From<BytesColumn>>(data: OwnedBytes) -> io::Result<T> {
-    let (body, dictionary_len_bytes) = data.rsplit(4);
-    let dictionary_len = u32::from_le_bytes(dictionary_len_bytes.as_slice().try_into().unwrap());
-    let (dictionary_bytes, column_bytes) = body.split(dictionary_len as usize);
-    let dictionary = Arc::new(Dictionary::from_bytes(dictionary_bytes)?);
-    let term_ord_column = crate::column::open_column_u64::<u64>(column_bytes)?;
-    let bytes_column = BytesColumn {
-        dictionary,
-        term_ord_column,
-    };
-    Ok(bytes_column.into())
-}

columnar/src/column_index/mod.rs

@@ -1,54 +0,0 @@
-mod multivalued_index;
-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::{Cardinality, RowId};
-
-#[derive(Clone)]
-pub enum ColumnIndex<'a> {
-    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>),
-}
-
-impl<'a> ColumnIndex<'a> {
-    pub fn get_cardinality(&self) -> Cardinality {
-        match self {
-            ColumnIndex::Full => Cardinality::Full,
-            ColumnIndex::Optional(_) => Cardinality::Optional,
-            ColumnIndex::Multivalued(_) => Cardinality::Multivalued,
-        }
-    }
-
-    pub fn value_row_ids(&self, row_id: RowId) -> Range<RowId> {
-        match self {
-            ColumnIndex::Full => row_id..row_id + 1,
-            ColumnIndex::Optional(optional_index) => {
-                if let Some(val) = optional_index.rank_if_exists(row_id) {
-                    val..val + 1
-                } else {
-                    0..0
-                }
-            }
-            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
-            }
-        }
-    }
-}

columnar/src/column_index/multivalued_index.rs

@@ -1,29 +0,0 @@
-use std::io;
-use std::io::Write;
-use std::sync::Arc;
-
-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,
-) -> io::Result<()> {
-    crate::column_values::serialize_column_values(
-        &*multivalued_index,
-        &[FastFieldCodecType::Bitpacked, FastFieldCodecType::Linear],
-        output,
-    )?;
-    Ok(())
-}
-
-pub fn open_multivalued_index(bytes: OwnedBytes) -> io::Result<Arc<dyn ColumnValues<RowId>>> {
-    let start_index_column: Arc<dyn ColumnValues<RowId>> =
-        crate::column_values::open_u64_mapped(bytes)?;
-    Ok(start_index_column)
-}

columnar/src/column_index/optional_index/mod.rs

@@ -1,493 +0,0 @@
-use std::io::{self, Write};
-use std::ops::Range;
-use std::sync::Arc;
-
-mod set;
-mod set_block;
-
-use common::{BinarySerializable, OwnedBytes, VInt};
-pub use set::{Set, SetCodec, SelectCursor};
-use set_block::{
-    DenseBlock, DenseBlockCodec, SparseBlock, SparseBlockCodec, DENSE_BLOCK_NUM_BYTES,
-};
-
-use crate::{InvalidData, RowId};
-
-/// The threshold for for number of elements after which we switch to dense block encoding.
-///
-/// We simply pick the value that minimize the size of the blocks.
-const DENSE_BLOCK_THRESHOLD: u32 =
-    set_block::DENSE_BLOCK_NUM_BYTES / std::mem::size_of::<u16>() as u32; //< 5_120
-
-const ELEMENTS_PER_BLOCK: u32 = u16::MAX as u32 + 1;
-
-const BLOCK_SIZE: RowId = 1 << 16;
-
-#[derive(Copy, Clone, Debug)]
-struct BlockMeta {
-    non_null_rows_before_block: u32,
-    start_byte_offset: u32,
-    block_variant: BlockVariant,
-}
-
-#[derive(Clone, Copy, Debug)]
-enum BlockVariant {
-    Dense,
-    Sparse { num_vals: u16 },
-}
-
-impl BlockVariant {
-    pub fn empty() -> Self {
-        Self::Sparse { num_vals: 0 }
-    }
-    pub fn num_bytes_in_block(&self) -> u32 {
-        match *self {
-            BlockVariant::Dense => set_block::DENSE_BLOCK_NUM_BYTES,
-            BlockVariant::Sparse { num_vals } => num_vals as u32 * 2,
-        }
-    }
-}
-
-/// This codec is inspired by roaring bitmaps.
-/// In the dense blocks, however, in order to accelerate `select`
-/// we interleave an offset over two bytes. (more on this lower)
-///
-/// 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.
-#[derive(Clone)]
-pub struct OptionalIndex {
-    num_rows: RowId,
-    num_non_null_rows: RowId,
-    block_data: OwnedBytes,
-    block_metas: Arc<[BlockMeta]>,
-}
-
-impl OptionalIndex {
-    pub fn num_rows(&self) -> RowId {
-        self.num_rows
-    }
-
-    pub fn num_non_nulls(&self) -> RowId {
-        self.num_non_null_rows
-    }
-}
-
-/// 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(Copy, Debug, Clone)]
-struct RowAddr {
-    block_id: u16,
-    in_block_row_id: u16,
-}
-
-#[inline(always)]
-fn row_addr_from_row_id(row_id: RowId) -> RowAddr {
-    RowAddr {
-        block_id: (row_id / BLOCK_SIZE) as u16,
-        in_block_row_id: (row_id % BLOCK_SIZE) as u16,
-    }
-}
-
-enum BlockSelectCursor<'a> {
-    Dense(<DenseBlock<'a> as Set<u16>>::SelectCursor<'a>),
-    Sparse(<SparseBlock<'a> as Set<u16>>::SelectCursor<'a>),
-}
-
-impl<'a> BlockSelectCursor<'a> {
-    fn select(&mut self, rank: u16) -> u16 {
-        match self {
-            BlockSelectCursor::Dense(dense_select_cursor) => dense_select_cursor.select(rank),
-            BlockSelectCursor::Sparse(sparse_select_cursor) => sparse_select_cursor.select(rank),
-        }
-    }
-
-}
-pub struct OptionalIndexSelectCursor<'a> {
-    current_block_cursor: BlockSelectCursor<'a>,
-    current_block_id: u16,
-    // The current block is guaranteed to contain ranks < end_rank.
-    current_block_end_rank: RowId,
-    optional_index: &'a OptionalIndex,
-    block_doc_idx_start: RowId,
-    num_null_rows_before_block: RowId,
-}
-
-impl<'a> OptionalIndexSelectCursor<'a> {
-    fn search_and_load_block(&mut self, rank: RowId) {
-        if rank < self.current_block_end_rank {
-            // we are already in the right block
-            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.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;
-        let block: Block<'_> = self.optional_index.block(block_meta);
-        self.current_block_cursor = match block {
-            Block::Dense(dense_block) => BlockSelectCursor::Dense(dense_block.select_cursor()),
-            Block::Sparse(sparse_block) => BlockSelectCursor::Sparse(sparse_block.select_cursor()),
-        };
-    }
-}
-
-impl<'a> SelectCursor<RowId> for OptionalIndexSelectCursor<'a> {
-    fn select(&mut self, rank: RowId) -> RowId {
-        self.search_and_load_block(rank);
-        let index_in_block = (rank - self.num_null_rows_before_block) as u16;
-        self.current_block_cursor.select(index_in_block) as RowId + self.block_doc_idx_start
-    }
-}
-
-impl Set<RowId> for OptionalIndex {
-    type SelectCursor<'b> = OptionalIndexSelectCursor<'b> where Self: 'b;
-    // Check if value at position is not null.
-    #[inline]
-    fn contains(&self, row_id: RowId) -> bool {
-        let RowAddr {
-            block_id,
-            in_block_row_id,
-        } = row_addr_from_row_id(row_id);
-        let block_meta = self.block_metas[block_id as usize];
-        match self.block(block_meta) {
-            Block::Dense(dense_block) => dense_block.contains(in_block_row_id),
-            Block::Sparse(sparse_block) => sparse_block.contains(in_block_row_id),
-        }
-    }
-
-    #[inline]
-    fn rank_if_exists(&self, row_id: RowId) -> Option<RowId> {
-        let RowAddr {
-            block_id,
-            in_block_row_id,
-        } = row_addr_from_row_id(row_id);
-        let block_meta = self.block_metas[block_id as usize];
-        let block = self.block(block_meta);
-        let block_offset_row_id = match block {
-            Block::Dense(dense_block) => dense_block.rank_if_exists(in_block_row_id),
-            Block::Sparse(sparse_block) => sparse_block.rank_if_exists(in_block_row_id),
-        }? as u32;
-        Some(block_meta.non_null_rows_before_block + block_offset_row_id)
-    }
-
-    #[inline]
-    fn select(&self, rank: RowId) -> RowId {
-        let block_pos = self.find_block(rank, 0);
-        let block_doc_idx_start = (block_pos as u32) * ELEMENTS_PER_BLOCK;
-        let block_meta = self.block_metas[block_pos as usize];
-        let block: Block<'_> = self.block(block_meta);
-        let index_in_block = (rank - block_meta.non_null_rows_before_block) as u16;
-        let in_block_rank = match block {
-            Block::Dense(dense_block) => dense_block.select(index_in_block),
-            Block::Sparse(sparse_block) => sparse_block.select(index_in_block),
-        };
-        block_doc_idx_start + in_block_rank as u32
-    }
-
-    fn select_cursor<'b>(&'b self) -> OptionalIndexSelectCursor<'b> {
-        OptionalIndexSelectCursor {
-            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,
-            block_doc_idx_start: 0u32,
-            num_null_rows_before_block: 0u32,
-        }
-    }
-}
-
-impl OptionalIndex {
-
-    pub fn select_batch(&self, ranks: &mut [RowId]) {
-        let mut select_cursor = self.select_cursor();
-        for rank in ranks.iter_mut() {
-            *rank = select_cursor.select(*rank);
-        }
-    }
-
-    #[inline]
-    fn block<'a>(&'a self, block_meta: BlockMeta) -> Block<'a> {
-        let BlockMeta {
-            start_byte_offset,
-            block_variant,
-            ..
-        } = block_meta;
-        let start_byte_offset = start_byte_offset as usize;
-        let bytes = self.block_data.as_slice();
-        match block_variant {
-            BlockVariant::Dense => Block::Dense(DenseBlockCodec::open(
-                &bytes[start_byte_offset..start_byte_offset + DENSE_BLOCK_NUM_BYTES as usize],
-            )),
-            BlockVariant::Sparse { num_vals } => {
-                let end_byte_offset = start_byte_offset + num_vals as usize * 2;
-                let sparse_bytes = &bytes[start_byte_offset..end_byte_offset];
-                Block::Sparse(SparseBlockCodec::open(sparse_bytes))
-            }
-        }
-    }
-
-    #[inline]
-    fn find_block(&self, dense_idx: u32, start_block_pos: u16) -> u16 {
-        for block_pos in start_block_pos..self.block_metas.len() as u16 {
-            let offset = self.block_metas[block_pos as usize].non_null_rows_before_block;
-            if offset > dense_idx {
-                return block_pos - 1u16;
-            }
-        }
-        self.block_metas.len() as u16 - 1u16
-    }
-
-    // TODO Add a good API for the codec_idx to original_idx translation.
-    // The Iterator API is a probably a bad idea
-}
-
-#[derive(Copy, Clone)]
-enum Block<'a> {
-    Dense(DenseBlock<'a>),
-    Sparse(SparseBlock<'a>),
-}
-
-#[derive(Debug, Copy, Clone)]
-enum OptionalIndexCodec {
-    Dense = 0,
-    Sparse = 1,
-}
-
-impl OptionalIndexCodec {
-    fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    fn try_from_code(code: u8) -> Result<Self, InvalidData> {
-        match code {
-            0 => Ok(Self::Dense),
-            1 => Ok(Self::Sparse),
-            _ => Err(InvalidData),
-        }
-    }
-}
-
-impl BinarySerializable for OptionalIndexCodec {
-    fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
-        writer.write_all(&[self.to_code()])
-    }
-
-    fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
-        let optional_codec_code = u8::deserialize(reader)?;
-        let optional_codec = Self::try_from_code(optional_codec_code)?;
-        Ok(optional_codec)
-    }
-}
-
-fn serialize_optional_index_block(block_els: &[u16], out: &mut impl io::Write) -> io::Result<()> {
-    let is_sparse = is_sparse(block_els.len() as u32);
-    if is_sparse {
-        SparseBlockCodec::serialize(block_els.iter().copied(), out)?;
-    } else {
-        DenseBlockCodec::serialize(block_els.iter().copied(), out)?;
-    }
-    Ok(())
-}
-
-pub fn serialize_optional_index<'a, W: io::Write>(
-    serializable_optional_index: &dyn SerializableOptionalIndex<'a>,
-    output: &mut W,
-) -> io::Result<()> {
-    VInt(serializable_optional_index.num_rows() as u64).serialize(output)?;
-
-    let mut rows_it = serializable_optional_index.non_null_rows();
-    let mut block_metadata: Vec<SerializedBlockMeta> = 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.
-    let Some(idx) = rows_it.next() else {
-        output.write_all(&0u16.to_le_bytes())?;
-        return Ok(());
-    };
-
-    let row_addr = row_addr_from_row_id(idx);
-
-    let mut current_block_id = row_addr.block_id;
-    current_block.push(row_addr.in_block_row_id);
-
-    for idx in rows_it {
-        let value_addr = row_addr_from_row_id(idx);
-        if current_block_id != value_addr.block_id {
-            serialize_optional_index_block(&current_block[..], output)?;
-            block_metadata.push(SerializedBlockMeta {
-                block_id: current_block_id,
-                num_non_null_rows: current_block.len() as u32,
-            });
-            current_block.clear();
-            current_block_id = value_addr.block_id;
-        }
-        current_block.push(value_addr.in_block_row_id);
-    }
-
-    // handle last block
-    serialize_optional_index_block(&current_block[..], output)?;
-
-    block_metadata.push(SerializedBlockMeta {
-        block_id: current_block_id,
-        num_non_null_rows: current_block.len() as u32,
-    });
-
-    for block in &block_metadata {
-        output.write_all(&block.to_bytes())?;
-    }
-
-    output.write_all((block_metadata.len() as u16).to_le_bytes().as_ref())?;
-
-    Ok(())
-}
-
-const SERIALIZED_BLOCK_META_NUM_BYTES: usize = 4;
-
-#[derive(Clone, Copy, Debug)]
-struct SerializedBlockMeta {
-    block_id: u16,
-    num_non_null_rows: u32, //< takes values in 1..=u16::MAX
-}
-
-// TODO unit tests
-impl SerializedBlockMeta {
-    #[inline]
-    fn from_bytes(bytes: [u8; SERIALIZED_BLOCK_META_NUM_BYTES]) -> SerializedBlockMeta {
-        let block_id = u16::from_le_bytes(bytes[0..2].try_into().unwrap());
-        let num_non_null_rows: u32 =
-            u16::from_le_bytes(bytes[2..4].try_into().unwrap()) as u32 + 1u32;
-        SerializedBlockMeta {
-            block_id,
-            num_non_null_rows,
-        }
-    }
-
-    #[inline]
-    fn to_bytes(&self) -> [u8; SERIALIZED_BLOCK_META_NUM_BYTES] {
-        assert!(self.num_non_null_rows > 0);
-        let mut bytes = [0u8; SERIALIZED_BLOCK_META_NUM_BYTES];
-        bytes[0..2].copy_from_slice(&self.block_id.to_le_bytes());
-        // 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
-        bytes[2..4].copy_from_slice(&((self.num_non_null_rows - 1u32) as u16).to_le_bytes());
-        bytes
-    }
-}
-
-#[inline]
-fn is_sparse(num_rows_in_block: u32) -> bool {
-    num_rows_in_block < DENSE_BLOCK_THRESHOLD as u32
-}
-
-fn deserialize_optional_index_block_metadatas(
-    data: &[u8],
-    num_rows: u32,
-) -> (Box<[BlockMeta]>, u32) {
-    let num_blocks = data.len() / SERIALIZED_BLOCK_META_NUM_BYTES;
-    let mut block_metas = Vec::with_capacity(num_blocks as usize + 1);
-    let mut start_byte_offset = 0;
-    let mut non_null_rows_before_block = 0;
-    for block_meta_bytes in data.chunks_exact(SERIALIZED_BLOCK_META_NUM_BYTES) {
-        let block_meta_bytes: [u8; SERIALIZED_BLOCK_META_NUM_BYTES] =
-            block_meta_bytes.try_into().unwrap();
-        let SerializedBlockMeta {
-            block_id,
-            num_non_null_rows,
-        } = SerializedBlockMeta::from_bytes(block_meta_bytes);
-        block_metas.resize(
-            block_id as usize,
-            BlockMeta {
-                non_null_rows_before_block,
-                start_byte_offset,
-                block_variant: BlockVariant::empty(),
-            },
-        );
-        let block_variant = if is_sparse(num_non_null_rows) {
-            BlockVariant::Sparse {
-                num_vals: num_non_null_rows as u16,
-            }
-        } else {
-            BlockVariant::Dense
-        };
-        block_metas.push(BlockMeta {
-            non_null_rows_before_block,
-            start_byte_offset,
-            block_variant,
-        });
-        start_byte_offset += block_variant.num_bytes_in_block();
-        non_null_rows_before_block += num_non_null_rows as u32;
-    }
-    block_metas.resize(
-        ((num_rows + BLOCK_SIZE - 1) / BLOCK_SIZE) as usize,
-        BlockMeta {
-            non_null_rows_before_block,
-            start_byte_offset,
-            block_variant: BlockVariant::empty(),
-        },
-    );
-    (block_metas.into_boxed_slice(), non_null_rows_before_block)
-}
-
-pub fn open_optional_index(bytes: OwnedBytes) -> io::Result<OptionalIndex> {
-    let (mut bytes, num_non_empty_blocks_bytes) = bytes.rsplit(2);
-    let num_non_empty_block_bytes =
-        u16::from_le_bytes(num_non_empty_blocks_bytes.as_slice().try_into().unwrap());
-    let num_rows = VInt::deserialize_u64(&mut bytes)? as u32;
-    let block_metas_num_bytes =
-        num_non_empty_block_bytes as usize * SERIALIZED_BLOCK_META_NUM_BYTES;
-    let (block_data, block_metas) = bytes.rsplit(block_metas_num_bytes);
-    let (block_metas, num_non_null_rows) =
-        deserialize_optional_index_block_metadatas(block_metas.as_slice(), num_rows).into();
-    let optional_index = OptionalIndex {
-        num_rows,
-        num_non_null_rows,
-        block_data,
-        block_metas: block_metas.into(),
-    };
-    Ok(optional_index)
-}
-
-pub trait SerializableOptionalIndex<'a> {
-    fn num_rows(&self) -> RowId;
-    fn non_null_rows(&self) -> Box<dyn Iterator<Item = RowId> + 'a>;
-}
-
-impl SerializableOptionalIndex<'static> for Range<u32> {
-    fn num_rows(&self) -> RowId {
-        self.end
-    }
-    fn non_null_rows(&self) -> Box<dyn Iterator<Item = RowId> + 'static> {
-        Box::new(self.clone())
-    }
-}
-
-#[cfg(test)]
-mod tests;

columnar/src/column_index/optional_index/set.rs

@@ -1,45 +0,0 @@
-use std::io;
-
-/// A codec makes it possible to serialize a set of
-/// elements, and open the resulting Set representation.
-pub trait SetCodec {
-    type Item: Copy + TryFrom<usize> + Eq + std::hash::Hash + std::fmt::Debug;
-    type Reader<'a>: Set<Self::Item>;
-
-    /// Serializes a set of unique sorted u16 elements.
-    ///
-    /// May panic if the elements are not sorted.
-    fn serialize(els: impl Iterator<Item = Self::Item>, wrt: impl io::Write) -> io::Result<()>;
-    fn open<'a>(data: &'a [u8]) -> Self::Reader<'a>;
-}
-
-
-/// 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> {
-    // May panic if rank is greater than the number of elements in the Set,
-    // or if rank is < than value provided in the previous call.
-    fn select(&mut self, rank: T) -> T;
-}
-
-pub trait Set<T> {
-    type SelectCursor<'b>: SelectCursor<T> where Self: 'b;
-
-
-    /// Returns true if the elements is contained in the Set
-    fn contains(&self, el: T) -> bool;
-
-    /// If the set contains `el` returns its position in the sortd set of elements.
-    /// If the set does not contain the element, it returns `None`.
-    fn rank_if_exists(&self, el: T) -> Option<T>;
-
-    /// Return the rank-th value stored in this bitmap.
-    ///
-    /// # Panics
-    ///
-    /// May panic if rank is greater than the number of elements in the Set.
-    fn select(&self, rank: T) -> T;
-
-    /// Creates a brand new select cursor.
-    fn select_cursor<'b>(&'b self,) -> Self::SelectCursor<'b>;
-}

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

@@ -1,266 +0,0 @@
-use std::convert::TryInto;
-use std::io::{self, Write};
-
-use common::BinarySerializable;
-
-use crate::column_index::optional_index::{Set, SetCodec, SelectCursor, ELEMENTS_PER_BLOCK};
-
-#[inline(always)]
-fn get_bit_at(input: u64, n: u16) -> bool {
-    input & (1 << n) != 0
-}
-
-#[inline]
-fn set_bit_at(input: &mut u64, n: u16) {
-    *input |= 1 << n;
-}
-
-/// 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.
-
-const ELEMENTS_PER_MINI_BLOCK: u16 = 64;
-const MINI_BLOCK_BITVEC_NUM_BYTES: usize = 8;
-const MINI_BLOCK_OFFSET_NUM_BYTES: usize = 2;
-pub const MINI_BLOCK_NUM_BYTES: usize = MINI_BLOCK_BITVEC_NUM_BYTES + MINI_BLOCK_OFFSET_NUM_BYTES;
-
-/// Number of bytes in a dense block.
-pub const DENSE_BLOCK_NUM_BYTES: u32 =
-    (ELEMENTS_PER_BLOCK as u32 / ELEMENTS_PER_MINI_BLOCK as u32) * MINI_BLOCK_NUM_BYTES as u32;
-
-pub struct DenseBlockCodec;
-
-impl SetCodec for DenseBlockCodec {
-    type Item = u16;
-    type Reader<'a> = DenseBlock<'a>;
-
-    fn serialize(els: impl Iterator<Item = u16>, wrt: impl io::Write) -> io::Result<()> {
-        serialize_dense_codec(els, wrt)
-    }
-
-    #[inline]
-    fn open<'a>(data: &'a [u8]) -> Self::Reader<'a> {
-        assert_eq!(data.len(), DENSE_BLOCK_NUM_BYTES as usize);
-        DenseBlock(data)
-    }
-}
-
-/// Interpreting the bitvec as a set of integer within 0..=63
-/// and given an element, returns the number of elements in the
-/// set lesser than the element.
-///
-/// # Panics
-///
-/// May panic or return a wrong result if el <= 64.
-#[inline(always)]
-fn rank_u64(bitvec: u64, el: u16) -> u16 {
-    debug_assert!(el < 64);
-    let mask = (1u64 << el) - 1;
-    let masked_bitvec = bitvec & mask;
-    masked_bitvec.count_ones() as u16
-}
-
-#[inline(always)]
-fn select_u64(mut bitvec: u64, rank: u16) -> u16 {
-    for _ in 0..rank {
-        bitvec &= bitvec - 1;
-    }
-    bitvec.trailing_zeros() as u16
-}
-
-// TODO test the following solution on Intel... on Ryzen Zen <3 it is a catastrophy.
-// #[target_feature(enable = "bmi2")]
-// unsafe fn select_bitvec_unsafe(bitvec: u64, rank: u16) -> u16 {
-//     let pdep = _pdep_u64(1u64 << rank, bitvec);
-//     pdep.trailing_zeros() as u16
-// }
-
-#[derive(Clone, Copy, Debug)]
-struct DenseMiniBlock {
-    bitvec: u64,
-    rank: u16,
-}
-
-impl DenseMiniBlock {
-    fn from_bytes(data: [u8; MINI_BLOCK_NUM_BYTES]) -> Self {
-        let bitvec = u64::from_le_bytes(data[..MINI_BLOCK_BITVEC_NUM_BYTES].try_into().unwrap());
-        let rank = u16::from_le_bytes(data[MINI_BLOCK_BITVEC_NUM_BYTES..].try_into().unwrap());
-        Self { bitvec, rank }
-    }
-
-    fn to_bytes(&self) -> [u8; MINI_BLOCK_NUM_BYTES] {
-        let mut bytes = [0u8; MINI_BLOCK_NUM_BYTES];
-        bytes[..MINI_BLOCK_BITVEC_NUM_BYTES].copy_from_slice(&self.bitvec.to_le_bytes());
-        bytes[MINI_BLOCK_BITVEC_NUM_BYTES..].copy_from_slice(&self.rank.to_le_bytes());
-        bytes
-    }
-}
-
-#[derive(Copy, Clone)]
-pub struct DenseBlock<'a>(&'a [u8]);
-
-pub struct DenseBlockSelectCursor<'a> {
-    block_id: u16,
-    dense_block: DenseBlock<'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();
-        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)
-    }
-}
-
-impl<'a> Set<u16> for DenseBlock<'a> {
-    type SelectCursor<'b> = DenseBlockSelectCursor<'a> where Self: 'b;
-
-    #[inline(always)]
-    fn contains(&self, el: u16) -> bool {
-        let mini_block_id = el / ELEMENTS_PER_MINI_BLOCK;
-        let bitvec = self.mini_block(mini_block_id).bitvec;
-        let pos_in_bitvec = el % ELEMENTS_PER_MINI_BLOCK;
-        get_bit_at(bitvec, pos_in_bitvec)
-    }
-
-    #[inline(always)]
-    fn rank_if_exists(&self, el: u16) -> Option<u16> {
-        let block_pos = el / ELEMENTS_PER_MINI_BLOCK;
-        let index_block = self.mini_block(block_pos);
-        let pos_in_block_bit_vec = el % ELEMENTS_PER_MINI_BLOCK;
-        let ones_in_block = rank_u64(index_block.bitvec, pos_in_block_bit_vec);
-        let rank = index_block.rank + ones_in_block;
-        if get_bit_at(index_block.bitvec, pos_in_block_bit_vec) {
-            Some(rank)
-        } else {
-            None
-        }
-    }
-
-    #[inline(always)]
-    fn select(&self, rank: u16) -> u16 {
-        let block_id = self.find_miniblock_containing_rank(rank, 0).unwrap();
-        let index_block = self.mini_block(block_id);
-        let in_block_rank = rank - index_block.rank;
-        block_id * ELEMENTS_PER_MINI_BLOCK + select_u64(index_block.bitvec, in_block_rank)
-    }
-
-    #[inline(always)]
-    fn select_cursor<'b>(&'b self,) -> Self::SelectCursor<'b> {
-        DenseBlockSelectCursor {
-            block_id: 0,
-            dense_block: *self,
-        }
-    }
-}
-
-impl<'a> DenseBlock<'a> {
-    #[inline]
-    fn mini_block(&self, mini_block_id: u16) -> DenseMiniBlock {
-        let data_start_pos = mini_block_id as usize * MINI_BLOCK_NUM_BYTES;
-        DenseMiniBlock::from_bytes(
-            self.0[data_start_pos..data_start_pos + MINI_BLOCK_NUM_BYTES]
-                .try_into()
-                .unwrap(),
-        )
-    }
-
-    #[inline]
-    fn iter_miniblocks(
-        &self,
-        from_block_id: u16,
-    ) -> impl Iterator<Item = (u16, DenseMiniBlock)> + '_ {
-        self.0
-            .chunks_exact(MINI_BLOCK_NUM_BYTES)
-            .enumerate()
-            .skip(from_block_id as usize)
-            .map(|(block_id, bytes)| {
-                let mini_block = DenseMiniBlock::from_bytes(bytes.try_into().unwrap());
-                (block_id as u16, mini_block)
-            })
-    }
-
-    /// 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.
-    #[inline]
-    fn find_miniblock_containing_rank(&self, rank: u16, from_block_id: u16) -> Option<u16> {
-        self.iter_miniblocks(from_block_id)
-            .take_while(|(_, block)| block.rank <= rank)
-            .map(|(block_id, _)| block_id)
-            .last()
-    }
-}
-
-/// Iterator over all values, true if set, otherwise false
-pub fn serialize_dense_codec(
-    els: impl Iterator<Item = u16>,
-    mut output: impl Write,
-) -> io::Result<()> {
-    let mut non_null_rows_before: u16 = 0u16;
-    let mut block = 0u64;
-    let mut current_block_id = 0u16;
-    for el in els {
-        let block_id = el / ELEMENTS_PER_MINI_BLOCK;
-        let in_offset = el % ELEMENTS_PER_MINI_BLOCK;
-        while block_id > current_block_id {
-            let dense_mini_block = DenseMiniBlock {
-                bitvec: block,
-                rank: non_null_rows_before as u16,
-            };
-            output.write_all(&dense_mini_block.to_bytes())?;
-            non_null_rows_before += block.count_ones() as u16;
-            block = 0u64;
-            current_block_id += 1u16;
-        }
-        set_bit_at(&mut block, in_offset);
-    }
-    while current_block_id <= u16::MAX / ELEMENTS_PER_MINI_BLOCK {
-        block.serialize(&mut output)?;
-        non_null_rows_before.serialize(&mut output)?;
-        // This will overflow to 0 exactly if all bits are set.
-        // This is however not problem as we won't use this last value.
-        non_null_rows_before = non_null_rows_before.wrapping_add(block.count_ones() as u16);
-        block = 0u64;
-        current_block_id += 1u16;
-    }
-    Ok(())
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_select_bitvec() {
-        assert_eq!(select_u64(1u64, 0), 0);
-        assert_eq!(select_u64(2u64, 0), 1);
-        assert_eq!(select_u64(4u64, 0), 2);
-        assert_eq!(select_u64(8u64, 0), 3);
-        assert_eq!(select_u64(1 | 8u64, 0), 0);
-        assert_eq!(select_u64(1 | 8u64, 1), 3);
-    }
-
-    #[test]
-    fn test_count_ones() {
-        for i in 0..=63 {
-            assert_eq!(rank_u64(u64::MAX, i), i);
-        }
-    }
-
-    #[test]
-    fn test_dense() {
-        assert_eq!(DENSE_BLOCK_NUM_BYTES, 10_240);
-    }
-}

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

@@ -1,8 +0,0 @@
-mod dense;
-mod sparse;
-
-pub use dense::{DenseBlock, DenseBlockCodec, DENSE_BLOCK_NUM_BYTES};
-pub use sparse::{SparseBlock, SparseBlockCodec};
-
-#[cfg(test)]
-mod tests;

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

@@ -1,108 +0,0 @@
-use crate::column_index::optional_index::{Set, SetCodec, SelectCursor};
-
-pub struct SparseBlockCodec;
-
-impl SetCodec for SparseBlockCodec {
-    type Item = u16;
-    type Reader<'a> = SparseBlock<'a>;
-
-    fn serialize(
-        els: impl Iterator<Item = u16>,
-        mut wrt: impl std::io::Write,
-    ) -> std::io::Result<()> {
-        for el in els {
-            wrt.write_all(&el.to_le_bytes())?;
-        }
-        Ok(())
-    }
-
-    fn open<'a>(data: &'a [u8]) -> Self::Reader<'a> {
-        SparseBlock(data)
-    }
-}
-
-#[derive(Copy, Clone)]
-pub struct SparseBlock<'a>(&'a [u8]);
-
-impl<'a> SelectCursor<u16> for SparseBlock<'a> {
-    #[inline]
-    fn select(&mut self, rank: u16) -> u16 {
-        <SparseBlock<'a> as Set<u16>>::select(self, rank)
-    }
-}
-
-impl<'a> Set<u16> for SparseBlock<'a> {
-
-    type SelectCursor<'b> = Self where Self: 'b;
-
-    #[inline(always)]
-    fn contains(&self, el: u16) -> bool {
-        self.binary_search(el).is_ok()
-    }
-
-    #[inline(always)]
-    fn rank_if_exists(&self, el: u16) -> Option<u16> {
-        self.binary_search(el).ok()
-    }
-
-    #[inline(always)]
-    fn select(&self, rank: u16) -> u16 {
-        let offset = rank as usize * 2;
-        u16::from_le_bytes(self.0[offset..offset + 2].try_into().unwrap())
-    }
-
-    #[inline(always)]
-    fn select_cursor<'b>(&'b self,) -> Self::SelectCursor<'b> {
-        *self
-    }
-
-}
-
-#[inline(always)]
-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)
-}
-
-impl<'a> SparseBlock<'a> {
-    #[inline(always)]
-    fn value_at_idx(&self, data: &[u8], idx: u16) -> u16 {
-        let start_offset: usize = idx as usize * 2;
-        get_u16(data, start_offset)
-    }
-
-    #[inline]
-    fn num_vals(&self) -> u16 {
-        (self.0.len() / 2) as u16
-    }
-
-    #[inline]
-    #[allow(clippy::comparison_chain)]
-    // Looks for the element in the block. Returns the positions if found.
-    fn binary_search(&self, target: u16) -> Result<u16, u16> {
-        let data = &self.0;
-        let mut size = self.num_vals();
-        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 Ok(mid);
-            }
-
-            size = right - left;
-        }
-        Err(left)
-    }
-}

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

@@ -1,105 +0,0 @@
-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};
-
-fn test_set_helper<C: SetCodec<Item = u16>>(vals: &[u16]) -> usize {
-    let mut buffer = Vec::new();
-    C::serialize(vals.iter().copied(), &mut buffer).unwrap();
-    let tested_set = C::open(buffer.as_slice());
-    let hash_set: HashMap<C::Item, C::Item> = vals
-        .iter()
-        .copied()
-        .enumerate()
-        .map(|(ord, val)| (val, C::Item::try_from(ord).ok().unwrap()))
-        .collect();
-    for val in 0u16..=u16::MAX {
-        assert_eq!(tested_set.contains(val), hash_set.contains_key(&val));
-        assert_eq!(tested_set.rank_if_exists(val), hash_set.get(&val).copied());
-    }
-    for rank in 0..vals.len() {
-        assert_eq!(tested_set.select(rank as u16), vals[rank]);
-    }
-    buffer.len()
-}
-
-#[test]
-fn test_dense_block_set_u16_empty() {
-    let buffer_len = test_set_helper::<DenseBlockCodec>(&[]);
-    assert_eq!(buffer_len, DENSE_BLOCK_NUM_BYTES as usize);
-}
-
-#[test]
-fn test_dense_block_set_u16_max() {
-    let buffer_len = test_set_helper::<DenseBlockCodec>(&[u16::MAX]);
-    assert_eq!(buffer_len, DENSE_BLOCK_NUM_BYTES as usize);
-}
-
-#[test]
-fn test_sparse_block_set_u16_empty() {
-    let buffer_len = test_set_helper::<SparseBlockCodec>(&[]);
-    assert_eq!(buffer_len, 0);
-}
-
-#[test]
-fn test_sparse_block_set_u16_max() {
-    let buffer_len = test_set_helper::<SparseBlockCodec>(&[u16::MAX]);
-    assert_eq!(buffer_len, 2);
-}
-
-use proptest::prelude::*;
-
-proptest! {
-    #![proptest_config(ProptestConfig::with_cases(1))]
-    #[test]
-    fn test_prop_test_dense(els in proptest::collection::btree_set(0..=u16::MAX, 0..=u16::MAX as usize)) {
-        let vals: Vec<u16> = els.into_iter().collect();
-        let buffer_len = test_set_helper::<DenseBlockCodec>(&vals);
-        assert_eq!(buffer_len, DENSE_BLOCK_NUM_BYTES as usize);
-    }
-
-    #[test]
-    fn test_prop_test_sparse(els in proptest::collection::btree_set(0..=u16::MAX, 0..=u16::MAX as usize)) {
-        let vals: Vec<u16> = els.into_iter().collect();
-        let buffer_len = test_set_helper::<SparseBlockCodec>(&vals);
-        assert_eq!(buffer_len, vals.len() * 2);
-    }
-}
-
-#[test]
-fn test_simple_translate_codec_codec_idx_to_original_idx_dense() {
-    let mut buffer = Vec::new();
-    DenseBlockCodec::serialize([1, 3, 17, 32, 30_000, 30_001].iter().copied(), &mut buffer)
-        .unwrap();
-    let tested_set = DenseBlockCodec::open(buffer.as_slice());
-    assert!(tested_set.contains(1));
-    let mut select_cursor = tested_set.select_cursor();
-    assert_eq!(select_cursor.select(0), 1);
-    assert_eq!(select_cursor.select(1), 3);
-    assert_eq!(select_cursor.select(2), 17);
-}
-
-#[test]
-fn test_simple_translate_codec_idx_to_original_idx_sparse() {
-    let mut buffer = Vec::new();
-    SparseBlockCodec::serialize([1, 3, 17].iter().copied(), &mut buffer).unwrap();
-    let tested_set = SparseBlockCodec::open(buffer.as_slice());
-    assert!(tested_set.contains(1));
-    let mut select_cursor = tested_set.select_cursor();
-    assert_eq!(SelectCursor::select(&mut select_cursor, 0), 1);
-    assert_eq!(SelectCursor::select(&mut select_cursor, 1), 3);
-    assert_eq!(SelectCursor::select(&mut select_cursor, 2), 17);
-}
-
-#[test]
-fn test_simple_translate_codec_idx_to_original_idx_dense() {
-    let mut buffer = Vec::new();
-    DenseBlockCodec::serialize(0u16..150u16, &mut buffer).unwrap();
-    let tested_set = DenseBlockCodec::open(buffer.as_slice());
-    assert!(tested_set.contains(1));
-    let mut select_cursor = tested_set.select_cursor();
-    for i in 0..150 {
-        assert_eq!(i, select_cursor.select(i));
-    }
-}

columnar/src/column_index/optional_index/tests.rs

@@ -1,327 +0,0 @@
-use proptest::prelude::{any, prop, *};
-use proptest::strategy::Strategy;
-use proptest::{prop_oneof, proptest};
-
-use super::*;
-
-#[test]
-fn test_dense_block_threshold() {
-    assert_eq!(super::DENSE_BLOCK_THRESHOLD, 5_120);
-}
-
-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 test_with_random_sets_simple() {
-    let vals = 10..BLOCK_SIZE * 2;
-    let mut out: Vec<u8> = Vec::new();
-    serialize_optional_index(&vals.clone(), &mut out).unwrap();
-    let null_index = open_optional_index(OwnedBytes::new(out)).unwrap();
-    let ranks: Vec<u32> = (65_472u32..65_473u32).collect();
-    let els: Vec<u32> = ranks.iter().copied().map(|rank| rank + 10).collect();
-    let mut select_cursor = null_index.select_cursor();
-    for (rank, el) in ranks.iter().copied().zip(els.iter().copied()) {
-        assert_eq!(select_cursor.select(rank), el);
-    }
-}
-
-#[test]
-fn test_optional_index_trailing_empty_blocks() {
-    test_null_index(&[false]);
-}
-
-#[test]
-fn test_optional_index_one_block_false() {
-    let mut iter = vec![false; ELEMENTS_PER_BLOCK as usize];
-    iter.push(true);
-    test_null_index(&iter[..]);
-}
-
-#[test]
-fn test_optional_index_one_block_true() {
-    let mut iter = vec![true; ELEMENTS_PER_BLOCK as usize];
-    iter.push(true);
-    test_null_index(&iter[..]);
-}
-
-impl<'a> SerializableOptionalIndex<'a> for &'a [bool] {
-    fn num_rows(&self) -> RowId {
-        self.len() as u32
-    }
-
-    fn non_null_rows(&self) -> Box<dyn Iterator<Item = RowId> + 'a> {
-        Box::new(
-            self.iter()
-                .cloned()
-                .enumerate()
-                .filter(|(_pos, val)| *val)
-                .map(|(pos, _val)| pos as u32),
-        )
-    }
-}
-
-fn test_null_index(data: &[bool]) {
-    let mut out: Vec<u8> = Vec::new();
-    serialize_optional_index(&data, &mut out).unwrap();
-    let null_index = open_optional_index(OwnedBytes::new(out)).unwrap();
-    let orig_idx_with_value: Vec<u32> = data
-        .iter()
-        .enumerate()
-        .filter(|(_pos, val)| **val)
-        .map(|(pos, _val)| pos as u32)
-        .collect();
-    let mut select_iter = null_index.select_cursor();
-    for i in 0..orig_idx_with_value.len() {
-        assert_eq!(select_iter.select(i as u32), orig_idx_with_value[i]);
-    }
-
-    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.rank_if_exists(*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.contains(pos as u32), *value);
-    }
-}
-
-#[test]
-fn test_optional_index_test_translation() {
-    let mut out = vec![];
-    let iter = &[true, false, true, false];
-    serialize_optional_index(&&iter[..], &mut out).unwrap();
-    let null_index = open_optional_index(OwnedBytes::new(out)).unwrap();
-    let mut select_cursor = null_index.select_cursor();
-    assert_eq!(select_cursor.select(0), 0);
-    assert_eq!(select_cursor.select(1), 2);
-}
-
-#[test]
-fn test_optional_index_translate() {
-    let mut out = vec![];
-    let iter = &[true, false, true, false];
-    serialize_optional_index(&&iter[..], &mut out).unwrap();
-    let null_index = open_optional_index(OwnedBytes::new(out)).unwrap();
-    assert_eq!(null_index.rank_if_exists(0), Some(0));
-    assert_eq!(null_index.rank_if_exists(2), Some(1));
-}
-
-#[test]
-fn test_optional_index_small() {
-    let mut out = vec![];
-    let iter = &[true, false, true, false];
-    serialize_optional_index(&&iter[..], &mut out).unwrap();
-    let null_index = open_optional_index(OwnedBytes::new(out)).unwrap();
-    assert!(null_index.contains(0));
-    assert!(!null_index.contains(1));
-    assert!(null_index.contains(2));
-    assert!(!null_index.contains(3));
-}
-
-#[test]
-fn test_optional_index_large() {
-    let mut docs = vec![];
-    docs.extend((0..ELEMENTS_PER_BLOCK).map(|_idx| false));
-    docs.extend((0..=1).map(|_idx| true));
-
-    let mut out = vec![];
-    serialize_optional_index(&&docs[..], &mut out).unwrap();
-    let null_index = open_optional_index(OwnedBytes::new(out)).unwrap();
-    assert!(!null_index.contains(0));
-    assert!(!null_index.contains(100));
-    assert!(!null_index.contains(ELEMENTS_PER_BLOCK - 1));
-    assert!(null_index.contains(ELEMENTS_PER_BLOCK));
-    assert!(null_index.contains(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) -> OptionalIndex {
-        let mut out = Vec::new();
-        let mut rng: StdRng = StdRng::from_seed([1u8; 32]);
-        let vals: Vec<bool> = (0..TOTAL_NUM_VALUES)
-            .map(|_| rng.gen_bool(fill_ratio))
-            .collect();
-        serialize_optional_index(&&vals[..], &mut out).unwrap();
-        let codec = open_optional_index(OwnedBytes::new(out)).unwrap();
-        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: &OptionalIndex, 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: &OptionalIndex,
-        positions: impl Iterator<Item = u32>,
-    ) -> Option<u32> {
-        let mut dense_idx: Option<u32> = None;
-        for idx in positions {
-            dense_idx = dense_idx.or(codec.rank_if_exists(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) {
-        bench_translate_codec_to_orig_util(0.01f64, 0.005f32, bench);
-    }
-
-    #[bench]
-    fn bench_translate_codec_to_orig_10percent_filled_0comma005percent_hit(bench: &mut Bencher) {
-        bench_translate_codec_to_orig_util(0.1f64, 0.005f32, bench);
-    }
-
-    #[bench]
-    fn bench_translate_codec_to_orig_1percent_filled_10percent_hit(bench: &mut Bencher) {
-        bench_translate_codec_to_orig_util(0.01f64, 10f32, bench);
-    }
-
-    #[bench]
-    fn bench_translate_codec_to_orig_1percent_filled_full_scan(bench: &mut Bencher) {
-        bench_translate_codec_to_orig_util(0.01f64, 100f32, bench);
-    }
-
-    fn bench_translate_codec_to_orig_util(
-        percent_filled: f64,
-        percent_hit: f32,
-        bench: &mut Bencher,
-    ) {
-        let codec = gen_bools(percent_filled);
-        let num_non_nulls = codec.num_non_nulls();
-        let idxs: Vec<u32> = if percent_hit == 100.0f32 {
-            (0..num_non_nulls).collect()
-        } else {
-            n_percent_step_iterator(percent_hit, num_non_nulls).collect()
-        };
-        let mut output = vec![0u32; idxs.len()];
-        bench.iter(|| {
-            output.copy_from_slice(&idxs[..]);
-            codec.select_batch(&mut output);
-        });
-    }
-
-    #[bench]
-    fn bench_translate_codec_to_orig_90percent_filled_0comma005percent_hit(bench: &mut Bencher) {
-        bench_translate_codec_to_orig_util(0.9f64, 0.005, bench);
-    }
-
-    #[bench]
-    fn bench_translate_codec_to_orig_90percent_filled_full_scan(bench: &mut Bencher) {
-        bench_translate_codec_to_orig_util(0.9f64, 100.0f32, bench);
-    }
-}

columnar/src/column_index/serialize.rs

@@ -1,73 +0,0 @@
-use std::io;
-use std::io::Write;
-
-use common::{CountingWriter, OwnedBytes};
-
-use crate::column_index::multivalued_index::serialize_multivalued_index;
-use crate::column_index::optional_index::serialize_optional_index;
-use crate::column_index::{ColumnIndex, SerializableOptionalIndex};
-use crate::column_values::ColumnValues;
-use crate::{Cardinality, RowId};
-
-pub enum SerializableColumnIndex<'a> {
-    Full,
-    Optional(Box<dyn SerializableOptionalIndex<'a> + 'a>),
-    // TODO remove the Arc<dyn> apart from serialization this is not
-    // dynamic at all.
-    Multivalued(Box<dyn ColumnValues<RowId> + 'a>),
-}
-
-impl<'a> SerializableColumnIndex<'a> {
-    pub fn get_cardinality(&self) -> Cardinality {
-        match self {
-            SerializableColumnIndex::Full => Cardinality::Full,
-            SerializableColumnIndex::Optional(_) => Cardinality::Optional,
-            SerializableColumnIndex::Multivalued(_) => Cardinality::Multivalued,
-        }
-    }
-}
-
-pub fn serialize_column_index(
-    column_index: SerializableColumnIndex,
-    output: &mut impl Write,
-) -> io::Result<u32> {
-    let mut output = CountingWriter::wrap(output);
-    let cardinality = column_index.get_cardinality().to_code();
-    output.write_all(&[cardinality])?;
-    match column_index {
-        SerializableColumnIndex::Full => {}
-        SerializableColumnIndex::Optional(optional_index) => {
-            serialize_optional_index(&*optional_index, &mut output)?
-        }
-        SerializableColumnIndex::Multivalued(multivalued_index) => {
-            serialize_multivalued_index(&*multivalued_index, &mut output)?
-        }
-    }
-    let column_index_num_bytes = output.written_bytes() as u32;
-    Ok(column_index_num_bytes)
-}
-
-pub fn open_column_index(mut bytes: OwnedBytes) -> io::Result<ColumnIndex<'static>> {
-    if bytes.is_empty() {
-        return Err(io::Error::new(
-            io::ErrorKind::UnexpectedEof,
-            "Failed to deserialize column index. Empty buffer.",
-        ));
-    }
-    let cardinality_code = bytes[0];
-    let cardinality = Cardinality::try_from_code(cardinality_code)?;
-    bytes.advance(1);
-    match cardinality {
-        Cardinality::Full => Ok(ColumnIndex::Full),
-        Cardinality::Optional => {
-            let optional_index = super::optional_index::open_optional_index(bytes)?;
-            Ok(ColumnIndex::Optional(optional_index))
-        }
-        Cardinality::Multivalued => {
-            let multivalued_index = super::multivalued_index::open_multivalued_index(bytes)?;
-            Ok(ColumnIndex::Multivalued(multivalued_index))
-        }
-    }
-}
-
-// TODO unit tests

columnar/src/column_type_header.rs

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

columnar/src/column_values/bitpacked.rs

@@ -1,115 +0,0 @@
-use std::io::{self, Write};
-
-use common::OwnedBytes;
-use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
-
-use super::serialize::NormalizedHeader;
-use super::{ColumnValues, 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 ColumnValues 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 ColumnValues, 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 ColumnValues) -> 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::column_values::tests::create_and_validate;
-
-    fn create_and_validate_bitpacked_codec(data: &[u64], name: &str) {
-        create_and_validate::<BitpackedCodec>(data, name);
-    }
-
-    #[test]
-    fn test_with_codec_data_sets() {
-        let data_sets = crate::column_values::tests::get_codec_test_datasets();
-        for (mut data, name) in data_sets {
-            create_and_validate_bitpacked_codec(&data, name);
-            data.reverse();
-            create_and_validate::<BitpackedCodec>(&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_bitpacked_codec(&data, "rand");
-            data.reverse();
-            create_and_validate::<BitpackedCodec>(&data, "rand");
-        }
-    }
-}

columnar/src/column_values/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::column_values::line::Line;
-use crate::column_values::serialize::NormalizedHeader;
-use crate::column_values::{ColumnValues, 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: 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 ColumnValues) -> 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 ColumnValues, 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 ColumnValues 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
-    }
-}

columnar/src/column_values/column.rs

@@ -1,376 +0,0 @@
-use std::marker::PhantomData;
-use std::ops::{Range, RangeInclusive};
-
-use tantivy_bitpacker::minmax;
-
-use crate::column_values::monotonic_mapping::StrictlyMonotonicFn;
-
-/// `ColumnValues` provides access to a dense field column.
-///
-/// `Column` are just a wrapper over `ColumnValues` and a `ColumnIndex`.
-pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
-    /// Return the value associated with the given idx.
-    ///
-    /// This accessor should return as fast as possible.
-    ///
-    /// # Panics
-    ///
-    /// May panic if `idx` is greater than the column length.
-    fn get_val(&self, idx: u32) -> T;
-
-    /// Fills an output buffer with the fast field values
-    /// associated with the `DocId` going from
-    /// `start` to `start + output.len()`.
-    ///
-    /// # Panics
-    ///
-    /// Must panic if `start + output.len()` is greater than
-    /// the segment's `maxdoc`.
-    #[inline]
-    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)))
-    }
-}
-
-impl<T: Copy + PartialOrd> ColumnValues<T> for std::sync::Arc<dyn ColumnValues<T>> {
-    fn get_val(&self, idx: u32) -> T {
-        self.as_ref().get_val(idx)
-    }
-
-    fn min_value(&self) -> T {
-        self.as_ref().min_value()
-    }
-
-    fn max_value(&self) -> T {
-        self.as_ref().max_value()
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.as_ref().num_vals()
-    }
-
-    fn iter<'b>(&'b self) -> Box<dyn Iterator<Item = T> + 'b> {
-        self.as_ref().iter()
-    }
-
-    fn get_range(&self, start: u64, output: &mut [T]) {
-        self.as_ref().get_range(start, output)
-    }
-}
-
-impl<'a, C: ColumnValues<T> + ?Sized, T: Copy + PartialOrd> ColumnValues<T> for &'a C {
-    fn get_val(&self, idx: u32) -> T {
-        (*self).get_val(idx)
-    }
-
-    fn min_value(&self) -> T {
-        (*self).min_value()
-    }
-
-    fn max_value(&self) -> T {
-        (*self).max_value()
-    }
-
-    fn num_vals(&self) -> u32 {
-        (*self).num_vals()
-    }
-
-    fn iter<'b>(&'b self) -> Box<dyn Iterator<Item = T> + 'b> {
-        (*self).iter()
-    }
-
-    fn get_range(&self, start: u64, output: &mut [T]) {
-        (*self).get_range(start, output)
-    }
-}
-
-/// VecColumn provides `Column` over a slice.
-pub struct VecColumn<'a, T = u64> {
-    pub(crate) values: &'a [T],
-    pub(crate) min_value: T,
-    pub(crate) max_value: T,
-}
-
-impl<'a, T: Copy + PartialOrd + Send + Sync> ColumnValues<T> for VecColumn<'a, T> {
-    fn get_val(&self, position: u32) -> T {
-        self.values[position as usize]
-    }
-
-    fn iter(&self) -> Box<dyn Iterator<Item = T> + '_> {
-        Box::new(self.values.iter().copied())
-    }
-
-    fn min_value(&self) -> T {
-        self.min_value
-    }
-
-    fn max_value(&self) -> T {
-        self.max_value
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.values.len() as u32
-    }
-
-    fn get_range(&self, start: u64, output: &mut [T]) {
-        output.copy_from_slice(&self.values[start as usize..][..output.len()])
-    }
-}
-
-impl<'a, T: Copy + PartialOrd + Default, V> From<&'a V> for VecColumn<'a, T>
-where V: AsRef<[T]> + ?Sized
-{
-    fn from(values: &'a V) -> Self {
-        let values = values.as_ref();
-        let (min_value, max_value) = minmax(values.iter().copied()).unwrap_or_default();
-        Self {
-            values,
-            min_value,
-            max_value,
-        }
-    }
-}
-
-struct MonotonicMappingColumn<C, T, Input> {
-    from_column: C,
-    monotonic_mapping: T,
-    _phantom: PhantomData<Input>,
-}
-
-/// Creates a view of a column transformed by a strictly monotonic mapping. See
-/// [`StrictlyMonotonicFn`].
-///
-/// E.g. apply a gcd monotonic_mapping([100, 200, 300]) == [1, 2, 3]
-/// monotonic_mapping.mapping() is expected to be injective, and we should always have
-/// monotonic_mapping.inverse(monotonic_mapping.mapping(el)) == el
-///
-/// The inverse of the mapping is required for:
-/// `fn get_positions_for_value_range(&self, range: RangeInclusive<T>) -> Vec<u64> `
-/// The user provides the original value range and we need to monotonic map them in the same way the
-/// serialization does before calling the underlying column.
-///
-/// Note that when opening a codec, the monotonic_mapping should be the inverse of the mapping
-/// during serialization. And therefore the monotonic_mapping_inv when opening is the same as
-/// monotonic_mapping during serialization.
-pub fn monotonic_map_column<C, T, Input, Output>(
-    from_column: C,
-    monotonic_mapping: T,
-) -> impl ColumnValues<Output>
-where
-    C: ColumnValues<Input>,
-    T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
-    Input: PartialOrd + Send + Sync + Clone,
-    Output: PartialOrd + Send + Sync + Clone,
-{
-    MonotonicMappingColumn {
-        from_column,
-        monotonic_mapping,
-        _phantom: PhantomData,
-    }
-}
-
-impl<C, T, Input, Output> ColumnValues<Output> for MonotonicMappingColumn<C, T, Input>
-where
-    C: ColumnValues<Input>,
-    T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
-    Input: PartialOrd + Send + Sync + Clone,
-    Output: PartialOrd + Send + Sync + Clone,
-{
-    #[inline]
-    fn get_val(&self, idx: u32) -> Output {
-        let from_val = self.from_column.get_val(idx);
-        self.monotonic_mapping.mapping(from_val)
-    }
-
-    fn min_value(&self) -> Output {
-        let from_min_value = self.from_column.min_value();
-        self.monotonic_mapping.mapping(from_min_value)
-    }
-
-    fn max_value(&self) -> Output {
-        let from_max_value = self.from_column.max_value();
-        self.monotonic_mapping.mapping(from_max_value)
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.from_column.num_vals()
-    }
-
-    fn iter(&self) -> Box<dyn Iterator<Item = Output> + '_> {
-        Box::new(
-            self.from_column
-                .iter()
-                .map(|el| self.monotonic_mapping.mapping(el)),
-        )
-    }
-
-    fn get_docids_for_value_range(
-        &self,
-        range: RangeInclusive<Output>,
-        doc_id_range: Range<u32>,
-        positions: &mut Vec<u32>,
-    ) {
-        self.from_column.get_docids_for_value_range(
-            self.monotonic_mapping.inverse(range.start().clone())
-                ..=self.monotonic_mapping.inverse(range.end().clone()),
-            doc_id_range,
-            positions,
-        )
-    }
-
-    // We voluntarily do not implement get_range as it yields a regression,
-    // and we do not have any specialized implementation anyway.
-}
-
-/// Wraps an iterator into a `Column`.
-pub struct IterColumn<T>(T);
-
-impl<T> From<T> for IterColumn<T>
-where T: Iterator + Clone + ExactSizeIterator
-{
-    fn from(iter: T) -> Self {
-        IterColumn(iter)
-    }
-}
-
-impl<T> ColumnValues<T::Item> for IterColumn<T>
-where
-    T: Iterator + Clone + ExactSizeIterator + Send + Sync,
-    T::Item: PartialOrd,
-{
-    fn get_val(&self, idx: u32) -> T::Item {
-        self.0.clone().nth(idx as usize).unwrap()
-    }
-
-    fn min_value(&self) -> T::Item {
-        self.0.clone().next().unwrap()
-    }
-
-    fn max_value(&self) -> T::Item {
-        self.0.clone().last().unwrap()
-    }
-
-    fn num_vals(&self) -> u32 {
-        self.0.len() as u32
-    }
-
-    fn iter(&self) -> Box<dyn Iterator<Item = T::Item> + '_> {
-        Box::new(self.0.clone())
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::column_values::monotonic_mapping::{
-        StrictlyMonotonicMappingInverter, StrictlyMonotonicMappingToInternalBaseval,
-        StrictlyMonotonicMappingToInternalGCDBaseval,
-    };
-
-    #[test]
-    fn test_monotonic_mapping() {
-        let vals = &[3u64, 5u64][..];
-        let col = VecColumn::from(vals);
-        let mapped = monotonic_map_column(col, StrictlyMonotonicMappingToInternalBaseval::new(2));
-        assert_eq!(mapped.min_value(), 1u64);
-        assert_eq!(mapped.max_value(), 3u64);
-        assert_eq!(mapped.num_vals(), 2);
-        assert_eq!(mapped.num_vals(), 2);
-        assert_eq!(mapped.get_val(0), 1);
-        assert_eq!(mapped.get_val(1), 3);
-    }
-
-    #[test]
-    fn test_range_as_col() {
-        let col = IterColumn::from(10..100);
-        assert_eq!(col.num_vals(), 90);
-        assert_eq!(col.max_value(), 99);
-    }
-
-    #[test]
-    fn test_monotonic_mapping_iter() {
-        let vals: Vec<u64> = (10..110u64).map(|el| el * 10).collect();
-        let col = VecColumn::from(&vals);
-        let mapped = monotonic_map_column(
-            col,
-            StrictlyMonotonicMappingInverter::from(
-                StrictlyMonotonicMappingToInternalGCDBaseval::new(10, 100),
-            ),
-        );
-        let val_i64s: Vec<u64> = mapped.iter().collect();
-        for i in 0..100 {
-            assert_eq!(val_i64s[i as usize], mapped.get_val(i));
-        }
-    }
-
-    #[test]
-    fn test_monotonic_mapping_get_range() {
-        let vals: Vec<u64> = (0..100u64).map(|el| el * 10).collect();
-        let col = VecColumn::from(&vals);
-        let mapped = monotonic_map_column(
-            col,
-            StrictlyMonotonicMappingInverter::from(
-                StrictlyMonotonicMappingToInternalGCDBaseval::new(10, 0),
-            ),
-        );
-
-        assert_eq!(mapped.min_value(), 0u64);
-        assert_eq!(mapped.max_value(), 9900u64);
-        assert_eq!(mapped.num_vals(), 100);
-        let val_u64s: Vec<u64> = mapped.iter().collect();
-        assert_eq!(val_u64s.len(), 100);
-        for i in 0..100 {
-            assert_eq!(val_u64s[i as usize], mapped.get_val(i));
-            assert_eq!(val_u64s[i as usize], vals[i as usize] * 10);
-        }
-        let mut buf = [0u64; 20];
-        mapped.get_range(7, &mut buf[..]);
-        assert_eq!(&val_u64s[7..][..20], &buf);
-    }
-}

columnar/src/column_values/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/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()))
-    }
-}

columnar/src/column_values/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);
-    }
-}

columnar/src/column_values/compact_space/mod.rs

@@ -1,813 +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::column_values::compact_space::build_compact_space::get_compact_space;
-use crate::column_values::ColumnValues;
-
-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 ColumnValues<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
-    }
-}
-
-// TODO reenable what can be reenabled.
-// #[cfg(test)]
-// mod tests {
-//
-// use super::*;
-// use crate::column::format_version::read_format_version;
-// use crate::column::column_footer::read_null_index_footer;
-// use crate::column::serialize::U128Header;
-// use crate::column::{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: Vec<u32> = get_positions_for_value_range_helper(&decomp, 0..=1, 1..u32::MAX);
-// assert!(positions.is_empty());
-//
-// 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()
-// ),
-// &[3]
-// );
-// assert!(get_positions_for_value_range_helper(
-// &decomp,
-// 99998u128..=99998u128,
-// complete_range.clone()
-// )
-// .is_empty());
-// assert_eq!(
-// &get_positions_for_value_range_helper(
-// &decomp,
-// 333u128..=333u128,
-// complete_range.clone()
-// ),
-// &[8]
-// );
-// assert_eq!(
-// &get_positions_for_value_range_helper(
-// &decomp,
-// 332u128..=333u128,
-// complete_range.clone()
-// ),
-// &[8]
-// );
-// assert_eq!(
-// &get_positions_for_value_range_helper(
-// &decomp,
-// 332u128..=334u128,
-// complete_range.clone()
-// ),
-// &[8]
-// );
-// assert_eq!(
-// &get_positions_for_value_range_helper(
-// &decomp,
-// 333u128..=334u128,
-// complete_range.clone()
-// ),
-// &[8]
-// );
-//
-// assert_eq!(
-// &get_positions_for_value_range_helper(
-// &decomp,
-// 4_000_211_221u128..=5_000_000_000u128,
-// complete_range
-// ),
-// &[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!(
-// &get_positions_for_value_range_helper(&decomp, 0..=5, complete_range.clone())
-// .is_empty(),
-// );
-// assert_eq!(
-// &get_positions_for_value_range_helper(&decomp, 0..=100, complete_range.clone()),
-// &[0]
-// );
-// assert_eq!(
-// &get_positions_for_value_range_helper(&decomp, 0..=105, complete_range),
-// &[0]
-// );
-// }
-//
-// fn get_positions_for_value_range_helper<C: Column<T> + ?Sized, T: PartialOrd>(
-// 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);
-// }
-// }
-// }
-//

columnar/src/column_values/gcd.rs

@@ -1,75 +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::num::NonZeroU64;
-
-    use crate::column_values::gcd::{compute_gcd, find_gcd};
-
-    #[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);
-    }
-}

columnar/src/column_values/line.rs

@@ -1,222 +0,0 @@
-use std::io;
-use std::num::NonZeroU32;
-
-use common::{BinarySerializable, VInt};
-
-use crate::column_values::ColumnValues;
-
-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 ColumnValues) -> 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::column_values::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));
-    }
-}

columnar/src/column_values/linear.rs

@@ -1,230 +0,0 @@
-use std::io::{self, Write};
-
-use common::{BinarySerializable, OwnedBytes};
-use tantivy_bitpacker::{compute_num_bits, BitPacker, BitUnpacker};
-
-use super::line::Line;
-use super::serialize::NormalizedHeader;
-use super::{ColumnValues, 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 ColumnValues 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 ColumnValues, 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 ColumnValues) -> 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::column_values::tests;
-
-    fn create_and_validate(data: &[u64], name: &str) -> Option<(f32, f32)> {
-        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 = tests::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");
-        }
-    }
-}

columnar/src/column_values/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))
-}

columnar/src/column_values/mod.rs

@@ -1,326 +0,0 @@
-#![warn(missing_docs)]
-#![cfg_attr(all(feature = "unstable", test), feature(test))]
-
-//! # `fastfield_codecs`
-//!
-//! - Columnar storage of data for tantivy [`Column`].
-//! - Encode data in different codecs.
-//! - Monotonically map values to u64/u128
-
-#[cfg(test)]
-mod tests;
-
-use std::io;
-use std::io::Write;
-use std::sync::Arc;
-
-use common::{BinarySerializable, OwnedBytes};
-use compact_space::CompactSpaceDecompressor;
-use monotonic_mapping::{
-    StrictlyMonotonicMappingInverter, StrictlyMonotonicMappingToInternal,
-    StrictlyMonotonicMappingToInternalBaseval, StrictlyMonotonicMappingToInternalGCDBaseval,
-};
-use serialize::{Header, U128Header};
-
-mod bitpacked;
-mod blockwise_linear;
-mod compact_space;
-mod line;
-mod linear;
-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};
-use crate::column_values::bitpacked::BitpackedCodec;
-use crate::column_values::blockwise_linear::BlockwiseLinearCodec;
-use crate::column_values::linear::LinearCodec;
-
-#[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_mapped<T: MonotonicallyMappableToU128>(
-    mut bytes: OwnedBytes,
-) -> io::Result<Arc<dyn ColumnValues<T>>> {
-    let header = U128Header::deserialize(&mut bytes)?;
-    assert_eq!(header.codec_type, U128FastFieldCodecType::CompactSpace);
-    let reader = CompactSpaceDecompressor::open(bytes)?;
-
-    let inverted: StrictlyMonotonicMappingInverter<StrictlyMonotonicMappingToInternal<T>> =
-        StrictlyMonotonicMappingToInternal::<T>::new().into();
-    Ok(Arc::new(monotonic_map_column(reader, inverted)))
-}
-
-/// Returns the correct codec reader wrapped in the `Arc` for the data.
-pub fn open_u64_mapped<T: MonotonicallyMappableToU64>(
-    mut bytes: OwnedBytes,
-) -> io::Result<Arc<dyn ColumnValues<T>>> {
-    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>(
-    bytes: OwnedBytes,
-    header: &Header,
-) -> io::Result<Arc<dyn ColumnValues<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.
-pub(crate) 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: ColumnValues<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 ColumnValues, 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 ColumnValues) -> Option<f32>;
-}
-
-#[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::*;
-
-    fn get_data() -> Vec<u64> {
-        let mut rng = StdRng::seed_from_u64(2u64);
-        let mut data: Vec<_> = (100..55000_u64)
-            .map(|num| num + rng.gen::<u8>() as u64)
-            .collect();
-        data.push(99_000);
-        data.insert(1000, 2000);
-        data.insert(2000, 100);
-        data.insert(3000, 4100);
-        data.insert(4000, 100);
-        data.insert(5000, 800);
-        data
-    }
-
-    #[inline(never)]
-    fn value_iter() -> impl Iterator<Item = u64> {
-        0..20_000
-    }
-    fn get_reader_for_bench<Codec: FastFieldCodec>(data: &[u64]) -> Codec::Reader {
-        let mut bytes = Vec::new();
-        let min_value = *data.iter().min().unwrap();
-        let data = data.iter().map(|el| *el - min_value).collect::<Vec<_>>();
-        let col = VecColumn::from(&data);
-        let normalized_header = NormalizedHeader {
-            num_vals: col.num_vals(),
-            max_value: col.max_value(),
-        };
-        Codec::serialize(&VecColumn::from(&data), &mut bytes).unwrap();
-        Codec::open_from_bytes(OwnedBytes::new(bytes), normalized_header).unwrap()
-    }
-    fn bench_get<Codec: FastFieldCodec>(b: &mut Bencher, data: &[u64]) {
-        let col = get_reader_for_bench::<Codec>(data);
-        b.iter(|| {
-            let mut sum = 0u64;
-            for pos in value_iter() {
-                let val = col.get_val(pos as u32);
-                sum = sum.wrapping_add(val);
-            }
-            sum
-        });
-    }
-
-    #[inline(never)]
-    fn bench_get_dynamic_helper(b: &mut Bencher, col: Arc<dyn ColumnValues>) {
-        b.iter(|| {
-            let mut sum = 0u64;
-            for pos in value_iter() {
-                let val = col.get_val(pos as u32);
-                sum = sum.wrapping_add(val);
-            }
-            sum
-        });
-    }
-
-    fn bench_get_dynamic<Codec: FastFieldCodec>(b: &mut Bencher, data: &[u64]) {
-        let col = Arc::new(get_reader_for_bench::<Codec>(data));
-        bench_get_dynamic_helper(b, col);
-    }
-    fn bench_create<Codec: FastFieldCodec>(b: &mut Bencher, data: &[u64]) {
-        let min_value = *data.iter().min().unwrap();
-        let data = data.iter().map(|el| *el - min_value).collect::<Vec<_>>();
-
-        let mut bytes = Vec::new();
-        b.iter(|| {
-            bytes.clear();
-            Codec::serialize(&VecColumn::from(&data), &mut bytes).unwrap();
-        });
-    }
-
-    #[bench]
-    fn bench_fastfield_bitpack_create(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_create::<BitpackedCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_linearinterpol_create(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_create::<LinearCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_multilinearinterpol_create(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_create::<BlockwiseLinearCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_bitpack_get(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_get::<BitpackedCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_bitpack_get_dynamic(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_get_dynamic::<BitpackedCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_linearinterpol_get(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_get::<LinearCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_linearinterpol_get_dynamic(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_get_dynamic::<LinearCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_multilinearinterpol_get(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_get::<BlockwiseLinearCodec>(b, &data);
-    }
-    #[bench]
-    fn bench_fastfield_multilinearinterpol_get_dynamic(b: &mut Bencher) {
-        let data: Vec<_> = get_data();
-        bench_get_dynamic::<BlockwiseLinearCodec>(b, &data);
-    }
-}

columnar/src/column_values/monotonic_mapping.rs

@@ -1,278 +0,0 @@
-use std::marker::PhantomData;
-
-use fastdivide::DividerU64;
-
-use super::MonotonicallyMappableToU128;
-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 {
-    /// 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, Internal> {
-    /// 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;
-}
-
-/// 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>
-{
-    #[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)
-    }
-}
-
-/// 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)
-    }
-}
-
-/// Strictly monotonic mapping with a base value.
-pub(crate) struct StrictlyMonotonicMappingToInternalBaseval {
-    min_value: u64,
-}
-impl StrictlyMonotonicMappingToInternalBaseval {
-    #[inline(always)]
-    pub(crate) fn new(min_value: u64) -> Self {
-        Self { min_value }
-    }
-}
-
-impl<External: MonotonicallyMappableToU64> StrictlyMonotonicFn<External, u64>
-    for StrictlyMonotonicMappingToInternalBaseval
-{
-    #[inline(always)]
-    fn mapping(&self, val: External) -> u64 {
-        External::to_u64(val) - self.min_value
-    }
-
-    #[inline(always)]
-    fn inverse(&self, val: u64) -> External {
-        External::from_u64(self.min_value + val)
-    }
-}
-
-impl MonotonicallyMappableToU64 for u64 {
-    #[inline(always)]
-    fn to_u64(self) -> u64 {
-        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 crate::DateTime {
-    #[inline(always)]
-    fn to_u64(self) -> u64 {
-        common::i64_to_u64(self.timestamp_micros)
-    }
-
-    #[inline(always)]
-    fn from_u64(val: u64) -> Self {
-        crate::DateTime {
-            timestamp_micros: 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
-    }
-}
-
-impl MonotonicallyMappableToU64 for RowId {
-    #[inline(always)]
-    fn to_u64(self) -> u64 {
-        u64::from(self)
-    }
-
-    #[inline(always)]
-    fn from_u64(val: u64) -> RowId {
-        val as RowId
-    }
-}
-
-// 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);
-        // TODO
-        // identity mapping
-        // test_round_trip(&StrictlyMonotonicMappingToInternal::<u128>::new(), 100u128);
-
-        // base value to i64 round trip
-        let mapping = StrictlyMonotonicMappingToInternalBaseval::new(100);
-        test_round_trip::<_, _, u64>(&mapping, 100i64);
-        // base value and gcd to u64 round trip
-        let mapping = StrictlyMonotonicMappingToInternalGCDBaseval::new(10, 100);
-        test_round_trip::<_, _, u64>(&mapping, 100u64);
-    }
-
-    fn test_round_trip<T: StrictlyMonotonicFn<K, L>, K: std::fmt::Debug + Eq + Copy, L>(
-        mapping: &T,
-        test_val: K,
-    ) {
-        assert_eq!(mapping.inverse(mapping.mapping(test_val)), test_val);
-    }
-}

columnar/src/column_values/monotonic_mapping_u128.rs

@@ -1,40 +0,0 @@
-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 {
-    /// 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())
-}

columnar/src/column_values/serialize.rs

@@ -1,320 +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::io;
-use std::num::NonZeroU64;
-
-use common::{BinarySerializable, VInt};
-use log::warn;
-
-use super::bitpacked::BitpackedCodec;
-use super::blockwise_linear::BlockwiseLinearCodec;
-use super::linear::LinearCodec;
-use super::monotonic_mapping::{
-    StrictlyMonotonicFn, StrictlyMonotonicMappingToInternal,
-    StrictlyMonotonicMappingToInternalGCDBaseval,
-};
-use super::{
-    monotonic_map_column, ColumnValues, FastFieldCodec, FastFieldCodecType,
-    MonotonicallyMappableToU64, U128FastFieldCodecType,
-};
-use crate::column_values::compact_space::CompactSpaceCompressor;
-
-/// 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(crate) fn normalize_column<C: ColumnValues>(&self, from_column: C) -> impl ColumnValues {
-        normalize_column(from_column, self.min_value, self.gcd)
-    }
-
-    pub fn compute_header(
-        column: impl ColumnValues<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 = super::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,
-        })
-    }
-}
-
-fn normalize_column<C: ColumnValues>(
-    from_column: C,
-    min_value: u64,
-    gcd: Option<NonZeroU64>,
-) -> impl ColumnValues {
-    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,
-        })
-    }
-}
-
-/// Serializes u128 values with the compact space codec.
-pub fn serialize_column_values_u128<F: Fn() -> I, I: Iterator<Item = u128>>(
-    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)?;
-
-    Ok(())
-}
-
-/// Serializes the column with the codec with the best estimate on the data.
-pub fn serialize_column_values<T: MonotonicallyMappableToU64>(
-    typed_column: impl ColumnValues<T>,
-    codecs: &[FastFieldCodecType],
-    output: &mut impl io::Write,
-) -> 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)?;
-    Ok(())
-}
-
-fn detect_codec(
-    column: impl ColumnValues<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)
-}
-
-pub(crate) fn serialize_given_codec(
-    column: impl ColumnValues<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)?;
-        }
-    }
-    Ok(())
-}
-
-#[cfg(test)]
-pub mod tests {
-    use std::sync::Arc;
-
-    use common::OwnedBytes;
-
-    use super::*;
-    use crate::column_values::{open_u64_mapped, VecColumn};
-
-    const ALL_CODEC_TYPES: [FastFieldCodecType; 3] = [
-        FastFieldCodecType::Bitpacked,
-        FastFieldCodecType::Linear,
-        FastFieldCodecType::BlockwiseLinear,
-    ];
-
-    /// Helper function to serialize a column (autodetect from all codecs) and then open it
-    pub fn serialize_and_load<T: MonotonicallyMappableToU64 + Ord + Default>(
-        column: &[T],
-    ) -> Arc<dyn ColumnValues<T>> {
-        let mut buffer = Vec::new();
-        serialize_column_values(&VecColumn::from(&column), &ALL_CODEC_TYPES, &mut buffer).unwrap();
-        open_u64_mapped(OwnedBytes::new(buffer)).unwrap()
-    }
-    #[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_column_values(&col, &ALL_CODEC_TYPES, &mut buffer).unwrap();
-        // TODO put the header as a footer so that it serves as a padding.
-        // 5 bytes of header, 1 byte of value, 7 bytes of padding.
-        assert_eq!(buffer.len(), 5 + 1);
-    }
-
-    #[test]
-    fn test_fastfield_bool_bit_size_bitwidth_0() {
-        let mut buffer = Vec::new();
-        let col = VecColumn::from(&[true][..]);
-        serialize_column_values(&col, &ALL_CODEC_TYPES, &mut buffer).unwrap();
-        // 5 bytes of header, 0 bytes of value, 7 bytes of padding.
-        assert_eq!(buffer.len(), 5);
-    }
-
-    #[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_column_values(&col, &[FastFieldCodecType::Bitpacked], &mut buffer).unwrap();
-        // Values are stored over 3 bits.
-        assert_eq!(buffer.len(), 7 + (3 * 80 / 8));
-    }
-}

columnar/src/column_values/tests.rs

@@ -1,309 +0,0 @@
-use proptest::prelude::*;
-use proptest::strategy::Strategy;
-use proptest::{prop_oneof, proptest};
-
-use super::bitpacked::BitpackedCodec;
-use super::blockwise_linear::BlockwiseLinearCodec;
-use super::linear::LinearCodec;
-use super::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_column_values(&col, &[Codec::CODEC_TYPE], &mut out).unwrap();
-
-    let actual_compression = out.len() as f32 / (data.len() as f32 * 8.0);
-
-    let reader = super::open_u64_mapped::<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 = (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![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)> =
-            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);
-}
-
-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::column_values::serialize_column_values(
-        &VecColumn::from(&vals),
-        &[codec_type],
-        &mut buffer,
-    )?;
-    let buffer = OwnedBytes::new(buffer);
-    let column = crate::column_values::open_u64_mapped::<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::column_values::serialize_column_values(
-        &VecColumn::from(&vals),
-        &[codec_type],
-        &mut buffer_without_gcd,
-    )?;
-    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::column_values::serialize_column_values(
-        &VecColumn::from(&vals),
-        &[codec_type],
-        &mut buffer,
-    )?;
-    let buffer = OwnedBytes::new(buffer);
-    let column = crate::column_values::open_u64_mapped::<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::column_values::serialize_column_values(
-        &VecColumn::from(&vals),
-        &[codec_type],
-        &mut buffer_without_gcd,
-    )?;
-    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::column_values::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);
-}

columnar/src/columnar/column_type.rs

@@ -1,235 +0,0 @@
-use std::net::Ipv6Addr;
-
-use crate::value::NumericalType;
-use crate::InvalidData;
-
-/// The column type represents the column type and can fit on 6-bits.
-///
-/// - bits[0..3]: Column category type.
-/// - bits[3..6]: Numerical type if necessary.
-#[derive(Hash, Eq, PartialEq, Debug, Clone, Copy)]
-#[repr(u8)]
-pub enum ColumnType {
-    I64 = 0u8,
-    U64 = 1u8,
-    F64 = 2u8,
-    Bytes = 10u8,
-    Str = 14u8,
-    Bool = 18u8,
-    IpAddr = 22u8,
-    DateTime = 26u8,
-}
-
-#[cfg(test)]
-const COLUMN_TYPES: [ColumnType; 8] = [
-    ColumnType::I64,
-    ColumnType::U64,
-    ColumnType::F64,
-    ColumnType::Bytes,
-    ColumnType::Str,
-    ColumnType::Bool,
-    ColumnType::IpAddr,
-    ColumnType::DateTime,
-];
-
-impl ColumnType {
-    pub fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    pub(crate) fn try_from_code(code: u8) -> Result<ColumnType, InvalidData> {
-        use ColumnType::*;
-        match code {
-            0u8 => Ok(I64),
-            1u8 => Ok(U64),
-            2u8 => Ok(F64),
-            10u8 => Ok(Bytes),
-            14u8 => Ok(Str),
-            18u8 => Ok(Bool),
-            22u8 => Ok(IpAddr),
-            26u8 => Ok(Self::DateTime),
-            _ => Err(InvalidData),
-        }
-    }
-}
-
-impl From<NumericalType> for ColumnType {
-    fn from(numerical_type: NumericalType) -> Self {
-        match numerical_type {
-            NumericalType::I64 => ColumnType::I64,
-            NumericalType::U64 => ColumnType::U64,
-            NumericalType::F64 => ColumnType::F64,
-        }
-    }
-}
-
-impl ColumnType {
-    /// get column type category
-    pub(crate) fn column_type_category(self) -> ColumnTypeCategory {
-        match self {
-            ColumnType::I64 | ColumnType::U64 | ColumnType::F64 => ColumnTypeCategory::Numerical,
-            ColumnType::Bytes => ColumnTypeCategory::Bytes,
-            ColumnType::Str => ColumnTypeCategory::Str,
-            ColumnType::Bool => ColumnTypeCategory::Bool,
-            ColumnType::IpAddr => ColumnTypeCategory::IpAddr,
-            ColumnType::DateTime => ColumnTypeCategory::DateTime,
-        }
-    }
-
-    pub fn numerical_type(&self) -> Option<NumericalType> {
-        match self {
-            ColumnType::I64 => Some(NumericalType::I64),
-            ColumnType::U64 => Some(NumericalType::U64),
-            ColumnType::F64 => Some(NumericalType::F64),
-            ColumnType::Bytes
-            | ColumnType::Str
-            | ColumnType::Bool
-            | ColumnType::IpAddr
-            | ColumnType::DateTime => None,
-        }
-    }
-}
-
-// TODO remove if possible
-pub trait HasAssociatedColumnType: 'static + Send + Sync + Copy + PartialOrd {
-    fn column_type() -> ColumnType;
-    fn default_value() -> Self;
-}
-
-impl HasAssociatedColumnType for u64 {
-    fn column_type() -> ColumnType {
-        ColumnType::U64
-    }
-
-    fn default_value() -> Self {
-        0u64
-    }
-}
-
-impl HasAssociatedColumnType for i64 {
-    fn column_type() -> ColumnType {
-        ColumnType::I64
-    }
-
-    fn default_value() -> Self {
-        0i64
-    }
-}
-
-impl HasAssociatedColumnType for f64 {
-    fn column_type() -> ColumnType {
-        ColumnType::F64
-    }
-
-    fn default_value() -> Self {
-        Default::default()
-    }
-}
-
-impl HasAssociatedColumnType for bool {
-    fn column_type() -> ColumnType {
-        ColumnType::Bool
-    }
-    fn default_value() -> Self {
-        Default::default()
-    }
-}
-
-impl HasAssociatedColumnType for crate::DateTime {
-    fn column_type() -> ColumnType {
-        ColumnType::DateTime
-    }
-    fn default_value() -> Self {
-        Default::default()
-    }
-}
-
-impl HasAssociatedColumnType for Ipv6Addr {
-    fn column_type() -> ColumnType {
-        ColumnType::IpAddr
-    }
-
-    fn default_value() -> Self {
-        Ipv6Addr::from([0u8; 16])
-    }
-}
-
-/// Column types are grouped into different categories that
-/// corresponds to the different types of `JsonValue` types.
-///
-/// The columnar writer will apply coercion rules to make sure that
-/// at most one column exist per `ColumnTypeCategory`.
-///
-/// See also [README.md].
-#[derive(Copy, Clone, Ord, PartialOrd, Eq, PartialEq, Hash, Debug)]
-#[repr(u8)]
-pub enum ColumnTypeCategory {
-    Bool,
-    Str,
-    Numerical,
-    DateTime,
-    Bytes,
-    IpAddr,
-}
-
-impl From<ColumnType> for ColumnTypeCategory {
-    fn from(column_type: ColumnType) -> Self {
-        match column_type {
-            ColumnType::I64 => ColumnTypeCategory::Numerical,
-            ColumnType::U64 => ColumnTypeCategory::Numerical,
-            ColumnType::F64 => ColumnTypeCategory::Numerical,
-            ColumnType::Bytes => ColumnTypeCategory::Bytes,
-            ColumnType::Str => ColumnTypeCategory::Str,
-            ColumnType::Bool => ColumnTypeCategory::Bool,
-            ColumnType::IpAddr => ColumnTypeCategory::IpAddr,
-            ColumnType::DateTime => ColumnTypeCategory::DateTime,
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use std::collections::HashSet;
-
-    use super::*;
-    use crate::Cardinality;
-
-    #[test]
-    fn test_column_type_to_code() {
-        let mut column_type_set: HashSet<ColumnType> = HashSet::new();
-        for code in u8::MIN..=u8::MAX {
-            if let Ok(column_type) = ColumnType::try_from_code(code) {
-                assert_eq!(column_type.to_code(), code);
-                assert!(column_type_set.insert(column_type));
-            }
-        }
-        assert_eq!(column_type_set.len(), super::COLUMN_TYPES.len());
-    }
-
-    #[test]
-    fn test_column_category_sort_consistent_with_column_type_sort() {
-        // This is a very important property because we
-        // we need to serialize colunmn in the right order.
-        let mut column_types: Vec<ColumnType> = super::COLUMN_TYPES.iter().copied().collect();
-        column_types.sort_by_key(|col| col.to_code());
-        let column_categories: Vec<ColumnTypeCategory> = column_types
-            .into_iter()
-            .map(ColumnTypeCategory::from)
-            .collect();
-        for (prev, next) in column_categories.iter().zip(column_categories.iter()) {
-            assert!(prev <= next);
-        }
-    }
-
-    #[test]
-    fn test_cardinality_to_code() {
-        let mut num_cardinality = 0;
-        for code in u8::MIN..=u8::MAX {
-            if let Ok(cardinality) = Cardinality::try_from_code(code) {
-                assert_eq!(cardinality.to_code(), code);
-                num_cardinality += 1;
-            }
-        }
-        assert_eq!(num_cardinality, 3);
-    }
-}

columnar/src/columnar/format_version.rs

@@ -1,73 +0,0 @@
-use crate::InvalidData;
-
-pub const VERSION_FOOTER_NUM_BYTES: usize = MAGIC_BYTES.len() + std::mem::size_of::<u32>();
-
-/// We end the file by these 4 bytes just to somewhat identify that
-/// this is indeed a columnar file.
-const MAGIC_BYTES: [u8; 4] = [2, 113, 119, 066];
-
-pub fn footer() -> [u8; VERSION_FOOTER_NUM_BYTES] {
-    let mut footer_bytes = [0u8; VERSION_FOOTER_NUM_BYTES];
-    footer_bytes[0..4].copy_from_slice(&Version::V1.to_bytes());
-    footer_bytes[4..8].copy_from_slice(&MAGIC_BYTES[..]);
-    footer_bytes
-}
-
-pub fn parse_footer(footer_bytes: [u8; VERSION_FOOTER_NUM_BYTES]) -> Result<Version, InvalidData> {
-    if footer_bytes[4..8] != MAGIC_BYTES {
-        return Err(InvalidData);
-    }
-    Version::try_from_bytes(footer_bytes[0..4].try_into().unwrap())
-}
-
-#[derive(Debug, Copy, Clone, Eq, PartialEq)]
-#[repr(u32)]
-pub enum Version {
-    V1 = 1u32,
-}
-
-impl Version {
-    fn to_bytes(&self) -> [u8; 4] {
-        (*self as u32).to_le_bytes()
-    }
-
-    fn try_from_bytes(bytes: [u8; 4]) -> Result<Version, InvalidData> {
-        let code = u32::from_le_bytes(bytes);
-        match code {
-            1u32 => Ok(Version::V1),
-            _ => Err(InvalidData),
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use std::collections::HashSet;
-
-    use super::*;
-
-    #[test]
-    fn test_footer_dserialization() {
-        let parsed_version: Version = parse_footer(footer()).unwrap();
-        assert_eq!(Version::V1, parsed_version);
-    }
-
-    #[test]
-    fn test_version_serialization() {
-        let version_to_tests: Vec<u32> = [0, 1 << 8, 1 << 16, 1 << 24]
-            .iter()
-            .copied()
-            .flat_map(|offset| (0..255).map(move |el| el + offset))
-            .collect();
-        let mut valid_versions: HashSet<u32> = HashSet::default();
-        for &i in &version_to_tests {
-            let version_res = Version::try_from_bytes(i.to_le_bytes());
-            if let Ok(version) = version_res {
-                assert_eq!(version, Version::V1);
-                assert_eq!(version.to_bytes(), i.to_le_bytes());
-                valid_versions.insert(i);
-            }
-        }
-        assert_eq!(valid_versions.len(), 1);
-    }
-}

columnar/src/columnar/merge.rs

@@ -1,176 +0,0 @@
-use std::collections::HashMap;
-use std::io;
-
-use super::column_type::ColumnTypeCategory;
-use crate::columnar::ColumnarReader;
-use crate::dynamic_column::DynamicColumn;
-
-pub enum MergeDocOrder {
-    /// Columnar tables are simply stacked one above the other.
-    /// If the i-th columnar_readers has n_rows_i rows, then
-    /// in the resulting columnar,
-    /// rows [r0..n_row_0) contains the row of columnar_readers[0], in ordder
-    /// rows [n_row_0..n_row_0 + n_row_1 contains the row of columnar_readers[1], in order.
-    /// ..
-    Stack,
-    /// Some more complex mapping, that can interleaves rows from the different readers and
-    /// possibly drop rows.
-    Complex(()),
-}
-
-pub fn merge_columnar(
-    _columnar_readers: &[ColumnarReader],
-    mapping: MergeDocOrder,
-    _output: &mut impl io::Write,
-) -> io::Result<()> {
-    match mapping {
-        MergeDocOrder::Stack => {
-            // implement me :)
-            todo!();
-        }
-        MergeDocOrder::Complex(_) => {
-            // for later
-            todo!();
-        }
-    }
-}
-
-pub fn collect_columns(
-    columnar_readers: &[&ColumnarReader],
-) -> io::Result<HashMap<String, HashMap<ColumnTypeCategory, Vec<DynamicColumn>>>> {
-    // Each column name may have multiple types of column associated.
-    // For merging we are interested in the same column type category since they can be merged.
-    let mut field_name_to_group: HashMap<String, HashMap<ColumnTypeCategory, Vec<DynamicColumn>>> =
-        HashMap::new();
-
-    for columnar_reader in columnar_readers {
-        let column_name_and_handle = columnar_reader.list_columns()?;
-        for (column_name, handle) in column_name_and_handle {
-            let column_type_to_handles = field_name_to_group
-                .entry(column_name.to_string())
-                .or_default();
-
-            let columns = column_type_to_handles
-                .entry(handle.column_type().column_type_category())
-                .or_default();
-            columns.push(handle.open()?);
-        }
-    }
-
-    normalize_columns(&mut field_name_to_group);
-
-    Ok(field_name_to_group)
-}
-
-/// Cast numerical type columns to the same type
-pub(crate) fn normalize_columns(
-    map: &mut HashMap<String, HashMap<ColumnTypeCategory, Vec<DynamicColumn>>>,
-) {
-    for (_field_name, type_category_to_columns) in map.iter_mut() {
-        for (type_category, columns) in type_category_to_columns {
-            if type_category == &ColumnTypeCategory::Numerical {
-                let casted_columns = cast_to_common_numerical_column(&columns);
-                *columns = casted_columns;
-            }
-        }
-    }
-}
-
-/// Receives a list of columns of numerical types (u64, i64, f64)
-///
-/// Returns a list of `DynamicColumn` which are all of the same numerical type
-fn cast_to_common_numerical_column(columns: &[DynamicColumn]) -> Vec<DynamicColumn> {
-    assert!(columns
-        .iter()
-        .all(|column| column.column_type().numerical_type().is_some()));
-    let coerce_to_i64: Vec<_> = columns
-        .iter()
-        .map(|column| column.clone().coerce_to_i64())
-        .collect();
-
-    if coerce_to_i64.iter().all(|column| column.is_some()) {
-        return coerce_to_i64
-            .into_iter()
-            .map(|column| column.unwrap())
-            .collect();
-    }
-
-    let coerce_to_u64: Vec<_> = columns
-        .iter()
-        .map(|column| column.clone().coerce_to_u64())
-        .collect();
-
-    if coerce_to_u64.iter().all(|column| column.is_some()) {
-        return coerce_to_u64
-            .into_iter()
-            .map(|column| column.unwrap())
-            .collect();
-    }
-
-    columns
-        .iter()
-        .map(|column| {
-            column
-                .clone()
-                .coerce_to_f64()
-                .expect("couldn't cast column to f64")
-        })
-        .collect()
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::ColumnarWriter;
-
-    #[test]
-    fn test_column_coercion() {
-        // i64 type
-        let columnar1 = {
-            let mut dataframe_writer = ColumnarWriter::default();
-            dataframe_writer.record_numerical(1u32, "numbers", 1i64);
-            let mut buffer: Vec<u8> = Vec::new();
-            dataframe_writer.serialize(2, &mut buffer).unwrap();
-            ColumnarReader::open(buffer).unwrap()
-        };
-        // u64 type
-        let columnar2 = {
-            let mut dataframe_writer = ColumnarWriter::default();
-            dataframe_writer.record_numerical(1u32, "numbers", u64::MAX - 100);
-            let mut buffer: Vec<u8> = Vec::new();
-            dataframe_writer.serialize(2, &mut buffer).unwrap();
-            ColumnarReader::open(buffer).unwrap()
-        };
-
-        // f64 type
-        let columnar3 = {
-            let mut dataframe_writer = ColumnarWriter::default();
-            dataframe_writer.record_numerical(1u32, "numbers", 30.5);
-            let mut buffer: Vec<u8> = Vec::new();
-            dataframe_writer.serialize(2, &mut buffer).unwrap();
-            ColumnarReader::open(buffer).unwrap()
-        };
-
-        let column_map = collect_columns(&[&columnar1, &columnar2, &columnar3]).unwrap();
-        assert_eq!(column_map.len(), 1);
-        let cat_to_columns = column_map.get("numbers").unwrap();
-        assert_eq!(cat_to_columns.len(), 1);
-
-        let numerical = cat_to_columns.get(&ColumnTypeCategory::Numerical).unwrap();
-        assert!(numerical.iter().all(|column| column.is_f64()));
-
-        let column_map = collect_columns(&[&columnar1, &columnar1]).unwrap();
-        assert_eq!(column_map.len(), 1);
-        let cat_to_columns = column_map.get("numbers").unwrap();
-        assert_eq!(cat_to_columns.len(), 1);
-        let numerical = cat_to_columns.get(&ColumnTypeCategory::Numerical).unwrap();
-        assert!(numerical.iter().all(|column| column.is_i64()));
-
-        let column_map = collect_columns(&[&columnar2, &columnar2]).unwrap();
-        assert_eq!(column_map.len(), 1);
-        let cat_to_columns = column_map.get("numbers").unwrap();
-        assert_eq!(cat_to_columns.len(), 1);
-        let numerical = cat_to_columns.get(&ColumnTypeCategory::Numerical).unwrap();
-        assert!(numerical.iter().all(|column| column.is_u64()));
-    }
-}

columnar/src/columnar/mod.rs

@@ -1,10 +0,0 @@
-mod column_type;
-mod format_version;
-mod merge;
-mod reader;
-mod writer;
-
-pub use column_type::{ColumnType, HasAssociatedColumnType};
-pub use merge::{merge_columnar, MergeDocOrder};
-pub use reader::ColumnarReader;
-pub use writer::ColumnarWriter;

columnar/src/dynamic_column.rs

@@ -1,241 +0,0 @@
-use std::io;
-use std::net::Ipv6Addr;
-use std::sync::Arc;
-
-use common::file_slice::FileSlice;
-use common::{HasLen, OwnedBytes};
-
-use crate::column::{BytesColumn, Column, StrColumn};
-use crate::column_values::{monotonic_map_column, StrictlyMonotonicFn};
-use crate::columnar::ColumnType;
-use crate::{DateTime, NumericalType};
-
-#[derive(Clone)]
-pub enum DynamicColumn {
-    Bool(Column<bool>),
-    I64(Column<i64>),
-    U64(Column<u64>),
-    F64(Column<f64>),
-    IpAddr(Column<Ipv6Addr>),
-    DateTime(Column<DateTime>),
-    Bytes(BytesColumn),
-    Str(StrColumn),
-}
-
-impl DynamicColumn {
-    pub fn column_type(&self) -> ColumnType {
-        match self {
-            DynamicColumn::Bool(_) => ColumnType::Bool,
-            DynamicColumn::I64(_) => ColumnType::I64,
-            DynamicColumn::U64(_) => ColumnType::U64,
-            DynamicColumn::F64(_) => ColumnType::F64,
-            DynamicColumn::IpAddr(_) => ColumnType::IpAddr,
-            DynamicColumn::DateTime(_) => ColumnType::DateTime,
-            DynamicColumn::Bytes(_) => ColumnType::Bytes,
-            DynamicColumn::Str(_) => ColumnType::Str,
-        }
-    }
-
-    pub fn is_numerical(&self) -> bool {
-        self.column_type().numerical_type().is_some()
-    }
-
-    pub fn is_f64(&self) -> bool {
-        self.column_type().numerical_type() == Some(NumericalType::F64)
-    }
-    pub fn is_i64(&self) -> bool {
-        self.column_type().numerical_type() == Some(NumericalType::I64)
-    }
-    pub fn is_u64(&self) -> bool {
-        self.column_type().numerical_type() == Some(NumericalType::U64)
-    }
-
-    pub fn coerce_to_f64(self) -> Option<DynamicColumn> {
-        match self {
-            DynamicColumn::I64(column) => Some(DynamicColumn::F64(Column {
-                idx: column.idx,
-                values: Arc::new(monotonic_map_column(column.values, MapI64ToF64)),
-            })),
-            DynamicColumn::U64(column) => Some(DynamicColumn::F64(Column {
-                idx: column.idx,
-                values: Arc::new(monotonic_map_column(column.values, MapU64ToF64)),
-            })),
-            DynamicColumn::F64(_) => Some(self),
-            _ => None,
-        }
-    }
-    pub fn coerce_to_i64(self) -> Option<DynamicColumn> {
-        match self {
-            DynamicColumn::U64(column) => {
-                if column.max_value() > i64::MAX as u64 {
-                    return None;
-                }
-                Some(DynamicColumn::I64(Column {
-                    idx: column.idx,
-                    values: Arc::new(monotonic_map_column(column.values, MapU64ToI64)),
-                }))
-            }
-            DynamicColumn::I64(_) => Some(self),
-            _ => None,
-        }
-    }
-    pub fn coerce_to_u64(self) -> Option<DynamicColumn> {
-        match self {
-            DynamicColumn::I64(column) => {
-                if column.min_value() < 0 {
-                    return None;
-                }
-                Some(DynamicColumn::U64(Column {
-                    idx: column.idx,
-                    values: Arc::new(monotonic_map_column(column.values, MapI64ToU64)),
-                }))
-            }
-            DynamicColumn::U64(_) => Some(self),
-            _ => None,
-        }
-    }
-}
-
-struct MapI64ToF64;
-impl StrictlyMonotonicFn<i64, f64> for MapI64ToF64 {
-    #[inline(always)]
-    fn mapping(&self, inp: i64) -> f64 {
-        inp as f64
-    }
-    #[inline(always)]
-    fn inverse(&self, out: f64) -> i64 {
-        out as i64
-    }
-}
-
-struct MapU64ToF64;
-impl StrictlyMonotonicFn<u64, f64> for MapU64ToF64 {
-    #[inline(always)]
-    fn mapping(&self, inp: u64) -> f64 {
-        inp as f64
-    }
-    #[inline(always)]
-    fn inverse(&self, out: f64) -> u64 {
-        out as u64
-    }
-}
-
-struct MapU64ToI64;
-impl StrictlyMonotonicFn<u64, i64> for MapU64ToI64 {
-    #[inline(always)]
-    fn mapping(&self, inp: u64) -> i64 {
-        inp as i64
-    }
-    #[inline(always)]
-    fn inverse(&self, out: i64) -> u64 {
-        out as u64
-    }
-}
-
-struct MapI64ToU64;
-impl StrictlyMonotonicFn<i64, u64> for MapI64ToU64 {
-    #[inline(always)]
-    fn mapping(&self, inp: i64) -> u64 {
-        inp as u64
-    }
-    #[inline(always)]
-    fn inverse(&self, out: u64) -> i64 {
-        out as i64
-    }
-}
-
-macro_rules! static_dynamic_conversions {
-    ($typ:ty, $enum_name:ident) => {
-        impl Into<Option<$typ>> for DynamicColumn {
-            fn into(self) -> Option<$typ> {
-                if let DynamicColumn::$enum_name(col) = self {
-                    Some(col)
-                } else {
-                    None
-                }
-            }
-        }
-
-        impl From<$typ> for DynamicColumn {
-            fn from(typed_column: $typ) -> Self {
-                DynamicColumn::$enum_name(typed_column)
-            }
-        }
-    };
-}
-
-static_dynamic_conversions!(Column<bool>, Bool);
-static_dynamic_conversions!(Column<u64>, U64);
-static_dynamic_conversions!(Column<i64>, I64);
-static_dynamic_conversions!(Column<f64>, F64);
-static_dynamic_conversions!(Column<crate::DateTime>, DateTime);
-static_dynamic_conversions!(StrColumn, Str);
-static_dynamic_conversions!(BytesColumn, Bytes);
-static_dynamic_conversions!(Column<Ipv6Addr>, IpAddr);
-
-#[derive(Clone)]
-pub struct DynamicColumnHandle {
-    pub(crate) file_slice: FileSlice,
-    pub(crate) column_type: ColumnType,
-}
-
-impl DynamicColumnHandle {
-    // TODO rename load
-    pub fn open(&self) -> io::Result<DynamicColumn> {
-        let column_bytes: OwnedBytes = self.file_slice.read_bytes()?;
-        self.open_internal(column_bytes)
-    }
-
-    // TODO rename load_async
-    pub async fn open_async(&self) -> io::Result<DynamicColumn> {
-        let column_bytes: OwnedBytes = self.file_slice.read_bytes_async().await?;
-        self.open_internal(column_bytes)
-    }
-
-    /// Returns the `u64` fast field reader reader associated with `fields` of types
-    /// Str, u64, i64, f64, or datetime.
-    ///
-    /// If not, the fastfield reader will returns the u64-value associated with the original
-    /// FastValue.
-    pub fn open_u64_lenient(&self) -> io::Result<Option<Column<u64>>> {
-        let column_bytes = self.file_slice.read_bytes()?;
-        match self.column_type {
-            ColumnType::Str | ColumnType::Bytes => {
-                let column: BytesColumn = crate::column::open_column_bytes(column_bytes)?;
-                Ok(Some(column.term_ord_column))
-            }
-            ColumnType::Bool => Ok(None),
-            ColumnType::IpAddr => Ok(None),
-            ColumnType::I64 | ColumnType::U64 | ColumnType::F64 | ColumnType::DateTime => {
-                let column = crate::column::open_column_u64::<u64>(column_bytes)?;
-                Ok(Some(column))
-            }
-        }
-    }
-
-    fn open_internal(&self, column_bytes: OwnedBytes) -> io::Result<DynamicColumn> {
-        let dynamic_column: DynamicColumn = match self.column_type {
-            ColumnType::Bytes => {
-                crate::column::open_column_bytes::<BytesColumn>(column_bytes)?.into()
-            }
-            ColumnType::Str => crate::column::open_column_bytes::<StrColumn>(column_bytes)?.into(),
-            ColumnType::I64 => crate::column::open_column_u64::<i64>(column_bytes)?.into(),
-            ColumnType::U64 => crate::column::open_column_u64::<u64>(column_bytes)?.into(),
-            ColumnType::F64 => crate::column::open_column_u64::<f64>(column_bytes)?.into(),
-            ColumnType::Bool => crate::column::open_column_u64::<bool>(column_bytes)?.into(),
-            ColumnType::IpAddr => crate::column::open_column_u128::<Ipv6Addr>(column_bytes)?.into(),
-            ColumnType::DateTime => {
-                crate::column::open_column_u64::<crate::DateTime>(column_bytes)?.into()
-            }
-        };
-        Ok(dynamic_column)
-    }
-
-    pub fn num_bytes(&self) -> usize {
-        self.file_slice.len()
-    }
-
-    pub fn column_type(&self) -> ColumnType {
-        self.column_type
-    }
-}

columnar/src/lib.rs

@@ -1,80 +1,89 @@
-#![cfg_attr(all(feature = "unstable", test), feature(test))]
-
-#[cfg(test)]
-#[macro_use]
-extern crate more_asserts;
-
-#[cfg(all(test, feature = "unstable"))]
-extern crate test;
-
-use std::io;
-
-mod column;
-mod column_index;
-mod column_values;
-mod columnar;
+mod column_type_header;
 mod dictionary;
-mod dynamic_column;
+mod reader;
 pub(crate) mod utils;
 mod value;
+mod writer;
 
-pub use column::{BytesColumn, Column, StrColumn};
-pub use column_values::ColumnValues;
-pub use columnar::{
-    merge_columnar, ColumnType, ColumnarReader, ColumnarWriter, HasAssociatedColumnType,
-    MergeDocOrder,
-};
+pub use column_type_header::Cardinality;
+pub use reader::ColumnarReader;
 pub use value::{NumericalType, NumericalValue};
+pub use writer::ColumnarWriter;
 
-pub use self::dynamic_column::{DynamicColumn, DynamicColumnHandle};
-
-pub type RowId = u32;
-
-#[derive(Clone, Copy, PartialOrd, PartialEq, Default, Debug)]
-pub struct DateTime {
-    pub timestamp_micros: i64,
-}
+pub type DocId = u32;
 
 #[derive(Copy, Clone, Debug)]
 pub struct InvalidData;
 
-impl From<InvalidData> for io::Error {
-    fn from(_: InvalidData) -> Self {
-        io::Error::new(io::ErrorKind::InvalidData, "Invalid data")
-    }
-}
-
-/// Enum describing the number of values that can exist per document
-/// (or per row if you will).
-///
-/// The cardinality must fit on 2 bits.
-#[derive(Clone, Copy, Hash, Default, Debug, PartialEq, Eq, PartialOrd, Ord)]
-#[repr(u8)]
-pub enum Cardinality {
-    /// All documents contain exactly one value.
-    /// `Full` is the default for auto-detecting the Cardinality, since it is the most strict.
-    #[default]
-    Full = 0,
-    /// All documents contain at most one value.
-    Optional = 1,
-    /// All documents may contain any number of values.
-    Multivalued = 2,
-}
-
-impl Cardinality {
-    pub(crate) fn to_code(self) -> u8 {
-        self as u8
-    }
-
-    pub(crate) fn try_from_code(code: u8) -> Result<Cardinality, InvalidData> {
-        match code {
-            0 => Ok(Cardinality::Full),
-            1 => Ok(Cardinality::Optional),
-            2 => Ok(Cardinality::Multivalued),
-            _ => Err(InvalidData),
-        }
-    }
-}
-
 #[cfg(test)]
-mod tests;
+mod tests {
+    use std::ops::Range;
+
+    use common::file_slice::FileSlice;
+
+    use crate::column_type_header::{ColumnType, ColumnTypeAndCardinality};
+    use crate::reader::ColumnarReader;
+    use crate::value::NumericalValue;
+    use crate::{Cardinality, ColumnarWriter};
+
+    #[test]
+    fn test_dataframe_writer_bytes() {
+        let mut dataframe_writer = ColumnarWriter::default();
+        dataframe_writer.record_str(1u32, "my_string", "hello");
+        dataframe_writer.record_str(3u32, "my_string", "helloeee");
+        let mut buffer: Vec<u8> = Vec::new();
+        dataframe_writer.serialize(5, &mut buffer).unwrap();
+        let columnar_fileslice = FileSlice::from(buffer);
+        let columnar = ColumnarReader::open(columnar_fileslice).unwrap();
+        assert_eq!(columnar.num_columns(), 1);
+        let cols: Vec<(ColumnTypeAndCardinality, Range<u64>)> =
+            columnar.read_columns("my_string").unwrap();
+        assert_eq!(cols.len(), 1);
+        assert_eq!(cols[0].1, 0..158);
+    }
+
+    #[test]
+    fn test_dataframe_writer_bool() {
+        let mut dataframe_writer = ColumnarWriter::default();
+        dataframe_writer.record_bool(1u32, "bool.value", false);
+        let mut buffer: Vec<u8> = Vec::new();
+        dataframe_writer.serialize(5, &mut buffer).unwrap();
+        let columnar_fileslice = FileSlice::from(buffer);
+        let columnar = ColumnarReader::open(columnar_fileslice).unwrap();
+        assert_eq!(columnar.num_columns(), 1);
+        let cols: Vec<(ColumnTypeAndCardinality, Range<u64>)> =
+            columnar.read_columns("bool.value").unwrap();
+        assert_eq!(cols.len(), 1);
+        assert_eq!(
+            cols[0].0,
+            ColumnTypeAndCardinality {
+                cardinality: Cardinality::Optional,
+                typ: ColumnType::Bool
+            }
+        );
+        assert_eq!(cols[0].1, 0..21);
+    }
+
+    #[test]
+    fn test_dataframe_writer_numerical() {
+        let mut dataframe_writer = ColumnarWriter::default();
+        dataframe_writer.record_numerical(1u32, "srical.value", NumericalValue::U64(12u64));
+        dataframe_writer.record_numerical(2u32, "srical.value", NumericalValue::U64(13u64));
+        dataframe_writer.record_numerical(4u32, "srical.value", NumericalValue::U64(15u64));
+        let mut buffer: Vec<u8> = Vec::new();
+        dataframe_writer.serialize(5, &mut buffer).unwrap();
+        let columnar_fileslice = FileSlice::from(buffer);
+        let columnar = ColumnarReader::open(columnar_fileslice).unwrap();
+        assert_eq!(columnar.num_columns(), 1);
+        let cols: Vec<(ColumnTypeAndCardinality, Range<u64>)> =
+            columnar.read_columns("srical.value").unwrap();
+        assert_eq!(cols.len(), 1);
+        // Right now this 31 bytes are spent as follows
+        //
+        // - header 14 bytes
+        // - vals  8 //< due to padding? could have been 1byte?.
+        // - null footer 6 bytes
+        // - version footer 3 bytes // Should be file-wide
+        assert_eq!(cols[0].1, 0..31);
+    }
+}

columnar/src/reader/mod.rs

@@ -1,11 +1,11 @@
+use std::ops::Range;
 use std::{io, mem};
 
 use common::file_slice::FileSlice;
 use common::BinarySerializable;
 use sstable::{Dictionary, RangeSSTable};
 
-use crate::columnar::{format_version, ColumnType};
-use crate::dynamic_column::DynamicColumnHandle;
+use crate::column_type_header::ColumnTypeAndCardinality;
 
 fn io_invalid_data(msg: String) -> io::Error {
     io::Error::new(io::ErrorKind::InvalidData, msg)
@@ -26,14 +26,9 @@ impl ColumnarReader {
     }
 
     fn open_inner(file_slice: FileSlice) -> io::Result<ColumnarReader> {
-        let (file_slice_without_sstable_len, footer_slice) = file_slice
-            .split_from_end(mem::size_of::<u64>() + format_version::VERSION_FOOTER_NUM_BYTES);
-        let footer_bytes = footer_slice.read_bytes()?;
-        let (mut sstable_len_bytes, version_footer_bytes) =
-            footer_bytes.rsplit(format_version::VERSION_FOOTER_NUM_BYTES);
-        let version_footer_bytes: [u8; format_version::VERSION_FOOTER_NUM_BYTES] =
-            version_footer_bytes.as_slice().try_into().unwrap();
-        let _version = format_version::parse_footer(version_footer_bytes)?;
+        let (file_slice_without_sstable_len, sstable_len_bytes) =
+            file_slice.split_from_end(mem::size_of::<u64>());
+        let mut sstable_len_bytes = sstable_len_bytes.read_bytes()?;
         let sstable_len = u64::deserialize(&mut sstable_len_bytes)?;
         let (column_data, sstable) =
             file_slice_without_sstable_len.split_from_end(sstable_len as usize);
@@ -44,27 +39,26 @@ impl ColumnarReader {
         })
     }
 
-    // TODO Add unit tests
-    pub fn list_columns(&self) -> io::Result<Vec<(String, DynamicColumnHandle)>> {
+    // TODO fix ugly API
+    pub fn list_columns(
+        &self,
+    ) -> io::Result<Vec<(String, ColumnTypeAndCardinality, Range<u64>, u64)>> {
         let mut stream = self.column_dictionary.stream()?;
         let mut results = Vec::new();
         while stream.advance() {
             let key_bytes: &[u8] = stream.key();
             let column_code: u8 = key_bytes.last().cloned().unwrap();
-            let column_type: ColumnType = ColumnType::try_from_code(column_code)
+            let column_type_and_cardinality = ColumnTypeAndCardinality::try_from_code(column_code)
                 .map_err(|_| io_invalid_data(format!("Unknown column code `{column_code}`")))?;
             let range = stream.value().clone();
-            let column_name =
-                // The last two bytes are respectively the 0u8 separator and the column_type.
-                String::from_utf8_lossy(&key_bytes[..key_bytes.len() - 2]).to_string();
-            let file_slice = self
-                .column_data
-                .slice(range.start as usize..range.end as usize);
-            let column_handle = DynamicColumnHandle {
-                file_slice,
-                column_type,
-            };
-            results.push((column_name, column_handle));
+            let column_name = String::from_utf8_lossy(&key_bytes[..key_bytes.len() - 1]);
+            let range_len = range.end - range.start;
+            results.push((
+                column_name.to_string(),
+                column_type_and_cardinality,
+                range,
+                range_len,
+            ));
         }
         Ok(results)
     }
@@ -74,7 +68,10 @@ 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>> {
+    pub fn read_columns(
+        &self,
+        column_name: &str,
+    ) -> io::Result<Vec<(ColumnTypeAndCardinality, Range<u64>)>> {
         // Each column is a associated to a given `column_key`,
         // that starts by `column_name\0column_header`.
         //
@@ -83,8 +80,6 @@ impl ColumnarReader {
         //
         // This is in turn equivalent to searching for the range
         // `[column_name,\0`..column_name\1)`.
-
-        // TODO can we get some more generic `prefix(..)` logic in the dictioanry.
         let mut start_key = column_name.to_string();
         start_key.push('\0');
         let mut end_key = column_name.to_string();
@@ -100,17 +95,10 @@ impl ColumnarReader {
             let key_bytes: &[u8] = stream.key();
             assert!(key_bytes.starts_with(start_key.as_bytes()));
             let column_code: u8 = key_bytes.last().cloned().unwrap();
-            let column_type = ColumnType::try_from_code(column_code)
+            let column_type_and_cardinality = ColumnTypeAndCardinality::try_from_code(column_code)
                 .map_err(|_| io_invalid_data(format!("Unknown column code `{column_code}`")))?;
             let range = stream.value().clone();
-            let file_slice = self
-                .column_data
-                .slice(range.start as usize..range.end as usize);
-            let dynamic_column_handle = DynamicColumnHandle {
-                file_slice,
-                column_type,
-            };
-            results.push(dynamic_column_handle);
+            results.push((column_type_and_cardinality, range));
         }
         Ok(results)
     }

columnar/src/tests.rs

@@ -1,212 +0,0 @@
-use std::net::Ipv6Addr;
-
-use crate::column_values::MonotonicallyMappableToU128;
-use crate::columnar::ColumnType;
-use crate::dynamic_column::{DynamicColumn, DynamicColumnHandle};
-use crate::value::NumericalValue;
-use crate::{Cardinality, ColumnarReader, ColumnarWriter};
-
-#[test]
-fn test_dataframe_writer_str() {
-    let mut dataframe_writer = ColumnarWriter::default();
-    dataframe_writer.record_str(1u32, "my_string", "hello");
-    dataframe_writer.record_str(3u32, "my_string", "helloeee");
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer.serialize(5, &mut buffer).unwrap();
-    let columnar = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar.num_columns(), 1);
-    let cols: Vec<DynamicColumnHandle> = columnar.read_columns("my_string").unwrap();
-    assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 158);
-}
-
-#[test]
-fn test_dataframe_writer_bytes() {
-    let mut dataframe_writer = ColumnarWriter::default();
-    dataframe_writer.record_bytes(1u32, "my_string", b"hello");
-    dataframe_writer.record_bytes(3u32, "my_string", b"helloeee");
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer.serialize(5, &mut buffer).unwrap();
-    let columnar = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar.num_columns(), 1);
-    let cols: Vec<DynamicColumnHandle> = columnar.read_columns("my_string").unwrap();
-    assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 158);
-}
-
-#[test]
-fn test_dataframe_writer_bool() {
-    let mut dataframe_writer = ColumnarWriter::default();
-    dataframe_writer.record_bool(1u32, "bool.value", false);
-    dataframe_writer.record_bool(3u32, "bool.value", true);
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer.serialize(5, &mut buffer).unwrap();
-    let columnar = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar.num_columns(), 1);
-    let cols: Vec<DynamicColumnHandle> = columnar.read_columns("bool.value").unwrap();
-    assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 22);
-    assert_eq!(cols[0].column_type(), ColumnType::Bool);
-    let dyn_bool_col = cols[0].open().unwrap();
-    let DynamicColumn::Bool(bool_col) = dyn_bool_col else { panic!(); };
-    let vals: Vec<Option<bool>> = (0..5).map(|row_id| bool_col.first(row_id)).collect();
-    assert_eq!(&vals, &[None, Some(false), None, Some(true), None,]);
-}
-
-#[test]
-fn test_dataframe_writer_u64_multivalued() {
-    let mut dataframe_writer = ColumnarWriter::default();
-    dataframe_writer.record_numerical(2u32, "divisor", 2u64);
-    dataframe_writer.record_numerical(3u32, "divisor", 3u64);
-    dataframe_writer.record_numerical(4u32, "divisor", 2u64);
-    dataframe_writer.record_numerical(5u32, "divisor", 5u64);
-    dataframe_writer.record_numerical(6u32, "divisor", 2u64);
-    dataframe_writer.record_numerical(6u32, "divisor", 3u64);
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer.serialize(7, &mut buffer).unwrap();
-    let columnar = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar.num_columns(), 1);
-    let cols: Vec<DynamicColumnHandle> = columnar.read_columns("divisor").unwrap();
-    assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 29);
-    let dyn_i64_col = cols[0].open().unwrap();
-    let DynamicColumn::I64(divisor_col) = dyn_i64_col else { panic!(); };
-    assert_eq!(
-        divisor_col.get_cardinality(),
-        crate::Cardinality::Multivalued
-    );
-    assert_eq!(divisor_col.num_rows(), 7);
-}
-
-#[test]
-fn test_dataframe_writer_ip_addr() {
-    let mut dataframe_writer = ColumnarWriter::default();
-    dataframe_writer.record_ip_addr(1, "ip_addr", Ipv6Addr::from_u128(1001));
-    dataframe_writer.record_ip_addr(3, "ip_addr", Ipv6Addr::from_u128(1050));
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer.serialize(5, &mut buffer).unwrap();
-    let columnar = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar.num_columns(), 1);
-    let cols: Vec<DynamicColumnHandle> = columnar.read_columns("ip_addr").unwrap();
-    assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 42);
-    assert_eq!(cols[0].column_type(), ColumnType::IpAddr);
-    let dyn_bool_col = cols[0].open().unwrap();
-    let DynamicColumn::IpAddr(ip_col) = dyn_bool_col else { panic!(); };
-    let vals: Vec<Option<Ipv6Addr>> = (0..5).map(|row_id| ip_col.first(row_id)).collect();
-    assert_eq!(
-        &vals,
-        &[
-            None,
-            Some(Ipv6Addr::from_u128(1001)),
-            None,
-            Some(Ipv6Addr::from_u128(1050)),
-            None,
-        ]
-    );
-}
-
-#[test]
-fn test_dataframe_writer_numerical() {
-    let mut dataframe_writer = ColumnarWriter::default();
-    dataframe_writer.record_numerical(1u32, "srical.value", NumericalValue::U64(12u64));
-    dataframe_writer.record_numerical(2u32, "srical.value", NumericalValue::U64(13u64));
-    dataframe_writer.record_numerical(4u32, "srical.value", NumericalValue::U64(15u64));
-    let mut buffer: Vec<u8> = Vec::new();
-    dataframe_writer.serialize(6, &mut buffer).unwrap();
-    let columnar = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar.num_columns(), 1);
-    let cols: Vec<DynamicColumnHandle> = columnar.read_columns("srical.value").unwrap();
-    assert_eq!(cols.len(), 1);
-    // Right now this 31 bytes are spent as follows
-    //
-    // - header 14 bytes
-    // - vals  8 //< due to padding? could have been 1byte?.
-    // - null footer 6 bytes
-    assert_eq!(cols[0].num_bytes(), 33);
-    let column = cols[0].open().unwrap();
-    let DynamicColumn::I64(column_i64) = column else { panic!(); };
-    assert_eq!(column_i64.idx.get_cardinality(), Cardinality::Optional);
-    assert_eq!(column_i64.first(0), None);
-    assert_eq!(column_i64.first(1), Some(12i64));
-    assert_eq!(column_i64.first(2), Some(13i64));
-    assert_eq!(column_i64.first(3), None);
-    assert_eq!(column_i64.first(4), Some(15i64));
-    assert_eq!(column_i64.first(5), None);
-    assert_eq!(column_i64.first(6), None); //< we can change the spec for that one.
-}
-
-#[test]
-fn test_dictionary_encoded_str() {
-    let mut buffer = Vec::new();
-    let mut columnar_writer = ColumnarWriter::default();
-    columnar_writer.record_str(1, "my.column", "a");
-    columnar_writer.record_str(3, "my.column", "c");
-    columnar_writer.record_str(3, "my.column2", "different_column!");
-    columnar_writer.record_str(4, "my.column", "b");
-    columnar_writer.serialize(5, &mut buffer).unwrap();
-    let columnar_reader = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar_reader.num_columns(), 2);
-    let col_handles = columnar_reader.read_columns("my.column").unwrap();
-    assert_eq!(col_handles.len(), 1);
-    let DynamicColumn::Str(str_col) = col_handles[0].open().unwrap() else  { panic!(); };
-    let index: Vec<Option<u64>> = (0..5).map(|row_id| str_col.ords().first(row_id)).collect();
-    assert_eq!(index, &[None, Some(0), None, Some(2), Some(1)]);
-    assert_eq!(str_col.num_rows(), 5);
-    let mut term_buffer = String::new();
-    let term_ords = str_col.ords();
-    assert_eq!(term_ords.first(0), None);
-    assert_eq!(term_ords.first(1), Some(0));
-    str_col.ord_to_str(0u64, &mut term_buffer).unwrap();
-    assert_eq!(term_buffer, "a");
-    assert_eq!(term_ords.first(2), None);
-    assert_eq!(term_ords.first(3), Some(2));
-    str_col.ord_to_str(2u64, &mut term_buffer).unwrap();
-    assert_eq!(term_buffer, "c");
-    assert_eq!(term_ords.first(4), Some(1));
-    str_col.ord_to_str(1u64, &mut term_buffer).unwrap();
-    assert_eq!(term_buffer, "b");
-}
-
-#[test]
-fn test_dictionary_encoded_bytes() {
-    let mut buffer = Vec::new();
-    let mut columnar_writer = ColumnarWriter::default();
-    columnar_writer.record_bytes(1, "my.column", b"a");
-    columnar_writer.record_bytes(3, "my.column", b"c");
-    columnar_writer.record_bytes(3, "my.column2", b"different_column!");
-    columnar_writer.record_bytes(4, "my.column", b"b");
-    columnar_writer.serialize(5, &mut buffer).unwrap();
-    let columnar_reader = ColumnarReader::open(buffer).unwrap();
-    assert_eq!(columnar_reader.num_columns(), 2);
-    let col_handles = columnar_reader.read_columns("my.column").unwrap();
-    assert_eq!(col_handles.len(), 1);
-    let DynamicColumn::Bytes(bytes_col) = col_handles[0].open().unwrap() else  { panic!(); };
-    let index: Vec<Option<u64>> = (0..5)
-        .map(|row_id| bytes_col.ords().first(row_id))
-        .collect();
-    assert_eq!(index, &[None, Some(0), None, Some(2), Some(1)]);
-    assert_eq!(bytes_col.num_rows(), 5);
-    let mut term_buffer = Vec::new();
-    let term_ords = bytes_col.ords();
-    assert_eq!(term_ords.first(0), None);
-    assert_eq!(term_ords.first(1), Some(0));
-    bytes_col
-        .dictionary
-        .ord_to_term(0u64, &mut term_buffer)
-        .unwrap();
-    assert_eq!(term_buffer, b"a");
-    assert_eq!(term_ords.first(2), None);
-    assert_eq!(term_ords.first(3), Some(2));
-    bytes_col
-        .dictionary
-        .ord_to_term(2u64, &mut term_buffer)
-        .unwrap();
-    assert_eq!(term_buffer, b"c");
-    assert_eq!(term_ords.first(4), Some(1));
-    bytes_col
-        .dictionary
-        .ord_to_term(1u64, &mut term_buffer)
-        .unwrap();
-    assert_eq!(term_buffer, b"b");
-}

columnar/src/value.rs

@@ -1,22 +1,12 @@
 use crate::InvalidData;
 
-#[derive(Copy, Clone, PartialEq, Debug)]
+#[derive(Copy, Clone, Debug, PartialEq)]
 pub enum NumericalValue {
     I64(i64),
     U64(u64),
     F64(f64),
 }
 
-impl NumericalValue {
-    pub fn numerical_type(&self) -> NumericalType {
-        match self {
-            NumericalValue::I64(_) => NumericalType::I64,
-            NumericalValue::U64(_) => NumericalType::U64,
-            NumericalValue::F64(_) => NumericalType::F64,
-        }
-    }
-}
-
 impl From<u64> for NumericalValue {
     fn from(val: u64) -> NumericalValue {
         NumericalValue::U64(val)
@@ -35,6 +25,18 @@ impl From<f64> for NumericalValue {
     }
 }
 
+impl NumericalValue {
+    pub fn numerical_type(&self) -> NumericalType {
+        match self {
+            NumericalValue::F64(_) => NumericalType::F64,
+            NumericalValue::I64(_) => NumericalType::I64,
+            NumericalValue::U64(_) => NumericalType::U64,
+        }
+    }
+}
+
+impl Eq for NumericalValue {}
+
 #[derive(Clone, Copy, Debug, Default, Hash, Eq, PartialEq)]
 #[repr(u8)]
 pub enum NumericalType {
@@ -104,13 +106,6 @@ impl Coerce for f64 {
     }
 }
 
-impl Coerce for crate::DateTime {
-    fn coerce(value: NumericalValue) -> Self {
-        let timestamp_micros = i64::coerce(value);
-        crate::DateTime { timestamp_micros }
-    }
-}
-
 #[cfg(test)]
 mod tests {
     use super::NumericalType;

columnar/src/writer/column_operation.rs

@@ -1,9 +1,7 @@
-use std::net::Ipv6Addr;
-
 use crate::dictionary::UnorderedId;
 use crate::utils::{place_bits, pop_first_byte, select_bits};
 use crate::value::NumericalValue;
-use crate::{InvalidData, NumericalType, RowId};
+use crate::{DocId, InvalidData, NumericalType};
 
 /// When we build a columnar dataframe, we first just group
 /// all mutations per column, and appends them in append-only buffer
@@ -15,7 +13,7 @@ use crate::{InvalidData, NumericalType, RowId};
 /// We represents all of these operations as `ColumnOperation`.
 #[derive(Eq, PartialEq, Debug, Clone, Copy)]
 pub(super) enum ColumnOperation<T> {
-    NewDoc(RowId),
+    NewDoc(DocId),
     Value(T),
 }
 
@@ -27,12 +25,12 @@ struct ColumnOperationMetadata {
 
 impl ColumnOperationMetadata {
     fn to_code(self) -> u8 {
-        place_bits::<0, 6>(self.len) | place_bits::<6, 8>(self.op_type.to_code())
+        place_bits::<0, 4>(self.len) | place_bits::<4, 8>(self.op_type.to_code())
     }
 
     fn try_from_code(code: u8) -> Result<Self, InvalidData> {
-        let len = select_bits::<0, 6>(code);
-        let typ_code = select_bits::<6, 8>(code);
+        let len = select_bits::<0, 4>(code);
+        let typ_code = select_bits::<4, 8>(code);
         let column_type = ColumnOperationType::try_from_code(typ_code)?;
         Ok(ColumnOperationMetadata {
             op_type: column_type,
@@ -144,21 +142,9 @@ impl SymbolValue for bool {
     }
 }
 
-impl SymbolValue for Ipv6Addr {
-    fn serialize(self, buffer: &mut [u8]) -> u8 {
-        buffer[0..16].copy_from_slice(&self.octets());
-        16
-    }
-
-    fn deserialize(bytes: &[u8]) -> Self {
-        let octets: [u8; 16] = bytes[0..16].try_into().unwrap();
-        Ipv6Addr::from(octets)
-    }
-}
-
 #[derive(Default)]
 struct MiniBuffer {
-    pub bytes: [u8; 17],
+    pub bytes: [u8; 10],
     pub len: u8,
 }
 

columnar/src/writer/column_writers.rs

@@ -2,9 +2,9 @@ use std::cmp::Ordering;
 
 use stacker::{ExpUnrolledLinkedList, MemoryArena};
 
-use crate::columnar::writer::column_operation::{ColumnOperation, SymbolValue};
 use crate::dictionary::{DictionaryBuilder, UnorderedId};
-use crate::{Cardinality, NumericalType, NumericalValue, RowId};
+use crate::writer::column_operation::{ColumnOperation, SymbolValue};
+use crate::{Cardinality, DocId, NumericalType, NumericalValue};
 
 #[derive(Copy, Clone, Debug, Eq, PartialEq)]
 #[repr(u8)]
@@ -53,7 +53,7 @@ impl ColumnWriter {
     ///
     /// This function will also update the cardinality of the column
     /// if necessary.
-    pub(super) fn record<S: SymbolValue>(&mut self, doc: RowId, value: S, arena: &mut MemoryArena) {
+    pub(super) fn record<S: SymbolValue>(&mut self, doc: DocId, value: S, arena: &mut MemoryArena) {
         // Difference between `doc` and the last doc.
         match delta_with_last_doc(self.last_doc_opt, doc) {
             DocumentStep::Same => {
@@ -77,7 +77,7 @@ impl ColumnWriter {
     // The overall number of docs in the column is necessary to
     // deal with the case where the all docs contain 1 value, except some documents
     // at the end of the column.
-    pub(crate) fn get_cardinality(&self, num_docs: RowId) -> Cardinality {
+    pub(crate) fn get_cardinality(&self, num_docs: DocId) -> Cardinality {
         match delta_with_last_doc(self.last_doc_opt, num_docs) {
             DocumentStep::Same | DocumentStep::Next => self.cardinality,
             DocumentStep::Skipped => self.cardinality.max(Cardinality::Optional),
@@ -102,29 +102,18 @@ pub(crate) struct NumericalColumnWriter {
     column_writer: ColumnWriter,
 }
 
-impl NumericalColumnWriter {
-    pub fn force_numerical_type(&mut self, numerical_type: NumericalType) {
-        assert!(self
-            .compatible_numerical_types
-            .is_type_accepted(numerical_type));
-        self.compatible_numerical_types = CompatibleNumericalTypes::StaticType(numerical_type);
-    }
-}
-
 /// State used to store what types are still acceptable
 /// after having seen a set of numerical values.
 #[derive(Clone, Copy)]
-enum CompatibleNumericalTypes {
-    Dynamic {
-        all_values_within_i64_range: bool,
-        all_values_within_u64_range: bool,
-    },
-    StaticType(NumericalType),
+struct CompatibleNumericalTypes {
+    all_values_within_i64_range: bool,
+    all_values_within_u64_range: bool,
+    // f64 is always acceptable.
 }
 
 impl Default for CompatibleNumericalTypes {
     fn default() -> CompatibleNumericalTypes {
-        CompatibleNumericalTypes::Dynamic {
+        CompatibleNumericalTypes {
             all_values_within_i64_range: true,
             all_values_within_u64_range: true,
         }
@@ -132,59 +121,36 @@ impl Default for CompatibleNumericalTypes {
 }
 
 impl CompatibleNumericalTypes {
-    fn is_type_accepted(&self, numerical_type: NumericalType) -> bool {
-        match self {
-            CompatibleNumericalTypes::Dynamic {
-                all_values_within_i64_range,
-                all_values_within_u64_range,
-            } => match numerical_type {
-                NumericalType::I64 => *all_values_within_i64_range,
-                NumericalType::U64 => *all_values_within_u64_range,
-                NumericalType::F64 => true,
-            },
-            CompatibleNumericalTypes::StaticType(static_numerical_type) => {
-                *static_numerical_type == numerical_type
-            }
-        }
-    }
-
     fn accept_value(&mut self, numerical_value: NumericalValue) {
-        match self {
-            CompatibleNumericalTypes::Dynamic {
-                all_values_within_i64_range,
-                all_values_within_u64_range,
-            } => match numerical_value {
-                NumericalValue::I64(val_i64) => {
-                    let value_within_u64_range = val_i64 >= 0i64;
-                    *all_values_within_u64_range &= value_within_u64_range;
-                }
-                NumericalValue::U64(val_u64) => {
-                    let value_within_i64_range = val_u64 < i64::MAX as u64;
-                    *all_values_within_i64_range &= value_within_i64_range;
-                }
-                NumericalValue::F64(_) => {
-                    *all_values_within_i64_range = false;
-                    *all_values_within_u64_range = false;
-                }
-            },
-            CompatibleNumericalTypes::StaticType(typ) => {
-                assert_eq!(numerical_value.numerical_type(), *typ);
+        match numerical_value {
+            NumericalValue::I64(val_i64) => {
+                let value_within_u64_range = val_i64 >= 0i64;
+                self.all_values_within_u64_range &= value_within_u64_range;
+            }
+            NumericalValue::U64(val_u64) => {
+                let value_within_i64_range = val_u64 < i64::MAX as u64;
+                self.all_values_within_i64_range &= value_within_i64_range;
+            }
+            NumericalValue::F64(_) => {
+                self.all_values_within_i64_range = false;
+                self.all_values_within_u64_range = false;
             }
         }
     }
 
     pub fn to_numerical_type(self) -> NumericalType {
-        for numerical_type in [NumericalType::I64, NumericalType::U64] {
-            if self.is_type_accepted(numerical_type) {
-                return numerical_type;
-            }
+        if self.all_values_within_i64_range {
+            NumericalType::I64
+        } else if self.all_values_within_u64_range {
+            NumericalType::U64
+        } else {
+            NumericalType::F64
         }
-        NumericalType::F64
     }
 }
 
 impl NumericalColumnWriter {
-    pub fn column_type_and_cardinality(&self, num_docs: RowId) -> (NumericalType, Cardinality) {
+    pub fn column_type_and_cardinality(&self, num_docs: DocId) -> (NumericalType, Cardinality) {
         let numerical_type = self.compatible_numerical_types.to_numerical_type();
         let cardinality = self.column_writer.get_cardinality(num_docs);
         (numerical_type, cardinality)
@@ -192,7 +158,7 @@ impl NumericalColumnWriter {
 
     pub fn record_numerical_value(
         &mut self,
-        doc: RowId,
+        doc: DocId,
         value: NumericalValue,
         arena: &mut MemoryArena,
     ) {
@@ -209,15 +175,15 @@ impl NumericalColumnWriter {
     }
 }
 
-#[derive(Copy, Clone)]
-pub(crate) struct StrOrBytesColumnWriter {
+#[derive(Copy, Clone, Default)]
+pub(crate) struct StrColumnWriter {
     pub(crate) dictionary_id: u32,
     pub(crate) column_writer: ColumnWriter,
 }
 
-impl StrOrBytesColumnWriter {
-    pub(crate) fn with_dictionary_id(dictionary_id: u32) -> StrOrBytesColumnWriter {
-        StrOrBytesColumnWriter {
+impl StrColumnWriter {
+    pub(crate) fn with_dictionary_id(dictionary_id: u32) -> StrColumnWriter {
+        StrColumnWriter {
             dictionary_id,
             column_writer: Default::default(),
         }
@@ -225,7 +191,7 @@ impl StrOrBytesColumnWriter {
 
     pub(crate) fn record_bytes(
         &mut self,
-        doc: RowId,
+        doc: DocId,
         bytes: &[u8],
         dictionaries: &mut [DictionaryBuilder],
         arena: &mut MemoryArena,
@@ -296,27 +262,4 @@ mod tests {
         test_column_writer_coercion_aux(&[1i64.into(), 1u64.into()], NumericalType::I64);
         test_column_writer_coercion_aux(&[u64::MAX.into(), (-1i64).into()], NumericalType::F64);
     }
-
-    #[test]
-    #[should_panic]
-    fn test_compatible_numerical_types_static_incompatible_type() {
-        let mut compatible_numerical_types =
-            CompatibleNumericalTypes::StaticType(NumericalType::U64);
-        compatible_numerical_types.accept_value(NumericalValue::I64(1i64));
-    }
-
-    #[test]
-    fn test_compatible_numerical_types_static_different_type_forbidden() {
-        let mut compatible_numerical_types =
-            CompatibleNumericalTypes::StaticType(NumericalType::U64);
-        compatible_numerical_types.accept_value(NumericalValue::U64(u64::MAX));
-    }
-
-    #[test]
-    fn test_compatible_numerical_types_static() {
-        for typ in [NumericalType::I64, NumericalType::I64, NumericalType::F64] {
-            let compatible_numerical_types = CompatibleNumericalTypes::StaticType(typ);
-            assert_eq!(compatible_numerical_types.to_numerical_type(), typ);
-        }
-    }
 }

columnar/src/writer/mod.rs

@@ -4,36 +4,30 @@ mod serializer;
 mod value_index;
 
 use std::io;
-use std::net::Ipv6Addr;
 
 use column_operation::ColumnOperation;
 use common::CountingWriter;
+use fastfield_codecs::serialize::ValueIndexInfo;
+use fastfield_codecs::{Column, MonotonicallyMappableToU64, VecColumn};
 use serializer::ColumnarSerializer;
 use stacker::{Addr, ArenaHashMap, MemoryArena};
 
-use crate::column_index::SerializableColumnIndex;
-use crate::column_values::{
-    ColumnValues, MonotonicallyMappableToU128, MonotonicallyMappableToU64, VecColumn,
-};
-use crate::columnar::column_type::{ColumnType, ColumnTypeCategory};
-use crate::columnar::writer::column_writers::{
-    ColumnWriter, NumericalColumnWriter, StrOrBytesColumnWriter,
-};
-use crate::columnar::writer::value_index::{IndexBuilder, PreallocatedIndexBuilders};
+use crate::column_type_header::{ColumnType, ColumnTypeAndCardinality, ColumnTypeCategory};
 use crate::dictionary::{DictionaryBuilder, TermIdMapping, UnorderedId};
 use crate::value::{Coerce, NumericalType, NumericalValue};
-use crate::{Cardinality, RowId};
+use crate::writer::column_writers::{ColumnWriter, NumericalColumnWriter, StrColumnWriter};
+use crate::writer::value_index::{IndexBuilder, SpareIndexBuilders};
+use crate::{Cardinality, DocId};
 
 /// This is a set of buffers that are used to temporarily write the values into before passing them
 /// to the fast field codecs.
 #[derive(Default)]
 struct SpareBuffers {
-    value_index_builders: PreallocatedIndexBuilders,
+    value_index_builders: SpareIndexBuilders,
     i64_values: Vec<i64>,
     u64_values: Vec<u64>,
     f64_values: Vec<f64>,
     bool_values: Vec<bool>,
-    ip_addr_values: Vec<Ipv6Addr>,
 }
 
 /// Makes it possible to create a new columnar.
@@ -51,11 +45,8 @@ struct SpareBuffers {
 /// ```
 pub struct ColumnarWriter {
     numerical_field_hash_map: ArenaHashMap,
-    datetime_field_hash_map: ArenaHashMap,
     bool_field_hash_map: ArenaHashMap,
-    ip_addr_field_hash_map: ArenaHashMap,
     bytes_field_hash_map: ArenaHashMap,
-    str_field_hash_map: ArenaHashMap,
     arena: MemoryArena,
     // Dictionaries used to store dictionary-encoded values.
     dictionaries: Vec<DictionaryBuilder>,
@@ -67,10 +58,7 @@ impl Default for ColumnarWriter {
         ColumnarWriter {
             numerical_field_hash_map: ArenaHashMap::new(10_000),
             bool_field_hash_map: ArenaHashMap::new(10_000),
-            ip_addr_field_hash_map: ArenaHashMap::new(10_000),
             bytes_field_hash_map: ArenaHashMap::new(10_000),
-            str_field_hash_map: ArenaHashMap::new(10_000),
-            datetime_field_hash_map: ArenaHashMap::new(10_000),
             dictionaries: Vec::new(),
             arena: MemoryArena::default(),
             buffers: SpareBuffers::default(),
@@ -78,115 +66,20 @@ impl Default for ColumnarWriter {
     }
 }
 
-#[inline]
-fn mutate_or_create_column<V, TMutator>(
-    arena_hash_map: &mut ArenaHashMap,
-    column_name: &str,
-    updater: TMutator,
-) where
-    V: Copy + 'static,
-    TMutator: FnMut(Option<V>) -> V,
-{
-    assert!(
-        !column_name.as_bytes().contains(&0u8),
-        "key may not contain the 0 byte"
-    );
-    arena_hash_map.mutate_or_create(column_name.as_bytes(), updater);
-}
-
 impl ColumnarWriter {
-    pub fn mem_usage(&self) -> usize {
-        // TODO add dictionary builders.
-        self.arena.mem_usage()
-            + self.numerical_field_hash_map.mem_usage()
-            + self.bool_field_hash_map.mem_usage()
-            + self.bytes_field_hash_map.mem_usage()
-            + self.str_field_hash_map.mem_usage()
-            + self.ip_addr_field_hash_map.mem_usage()
-            + self.datetime_field_hash_map.mem_usage()
-    }
-
-    pub fn record_column_type(&mut self, column_name: &str, column_type: ColumnType) {
-        match column_type {
-            ColumnType::Str | ColumnType::Bytes => {
-                let (hash_map, dictionaries) = (
-                    if column_type == ColumnType::Str {
-                        &mut self.str_field_hash_map
-                    } else {
-                        &mut self.bytes_field_hash_map
-                    },
-                    &mut self.dictionaries,
-                );
-                mutate_or_create_column(
-                    hash_map,
-                    column_name,
-                    |column_opt: Option<StrOrBytesColumnWriter>| {
-                        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)
-                        }
-                    },
-                );
-            }
-            ColumnType::Bool => {
-                mutate_or_create_column(
-                    &mut self.bool_field_hash_map,
-                    column_name,
-                    |column_opt: Option<ColumnWriter>| column_opt.unwrap_or_default(),
-                );
-            }
-            ColumnType::DateTime => {
-                mutate_or_create_column(
-                    &mut self.datetime_field_hash_map,
-                    column_name,
-                    |column_opt: Option<ColumnWriter>| column_opt.unwrap_or_default(),
-                );
-            }
-            ColumnType::I64 | ColumnType::F64 | ColumnType::U64 => {
-                let numerical_type = column_type.numerical_type().unwrap();
-                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
-                    },
-                );
-            }
-            ColumnType::IpAddr => mutate_or_create_column(
-                &mut self.ip_addr_field_hash_map,
-                column_name,
-                |column_opt: Option<ColumnWriter>| column_opt.unwrap_or_default(),
-            ),
-        }
-    }
-
-    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,
+        doc: DocId,
         column_name: &str,
         numerical_value: T,
     ) {
+        assert!(
+            !column_name.as_bytes().contains(&0u8),
+            "key may not contain the 0 byte"
+        );
         let (hash_map, arena) = (&mut self.numerical_field_hash_map, &mut self.arena);
-        mutate_or_create_column(
-            hash_map,
-            column_name,
+        hash_map.mutate_or_create(
+            column_name.as_bytes(),
             |column_opt: Option<NumericalColumnWriter>| {
                 let mut column: NumericalColumnWriter = column_opt.unwrap_or_default();
                 column.record_numerical_value(doc, numerical_value.into(), arena);
@@ -195,62 +88,23 @@ impl ColumnarWriter {
         );
     }
 
-    pub fn record_ip_addr(&mut self, doc: RowId, column_name: &str, ip_addr: Ipv6Addr) {
+    pub fn record_bool(&mut self, doc: DocId, column_name: &str, val: bool) {
         assert!(
             !column_name.as_bytes().contains(&0u8),
             "key may not contain the 0 byte"
         );
-        let (hash_map, arena) = (&mut self.ip_addr_field_hash_map, &mut self.arena);
+        let (hash_map, arena) = (&mut self.bool_field_hash_map, &mut self.arena);
         hash_map.mutate_or_create(
             column_name.as_bytes(),
             |column_opt: Option<ColumnWriter>| {
                 let mut column: ColumnWriter = column_opt.unwrap_or_default();
-                column.record(doc, ip_addr, arena);
+                column.record(doc, val, arena);
                 column
             },
         );
     }
 
-    pub fn record_bool(&mut self, doc: RowId, column_name: &str, val: bool) {
-        let (hash_map, arena) = (&mut self.bool_field_hash_map, &mut self.arena);
-        mutate_or_create_column(hash_map, column_name, |column_opt: Option<ColumnWriter>| {
-            let mut column: ColumnWriter = column_opt.unwrap_or_default();
-            column.record(doc, val, arena);
-            column
-        });
-    }
-
-    pub fn record_datetime(&mut self, doc: RowId, column_name: &str, datetime: crate::DateTime) {
-        let (hash_map, arena) = (&mut self.datetime_field_hash_map, &mut self.arena);
-        mutate_or_create_column(hash_map, column_name, |column_opt: Option<ColumnWriter>| {
-            let mut column: ColumnWriter = column_opt.unwrap_or_default();
-            column.record(doc, NumericalValue::I64(datetime.timestamp_micros), arena);
-            column
-        });
-    }
-
-    pub fn record_str(&mut self, doc: RowId, column_name: &str, value: &str) {
-        let (hash_map, arena, dictionaries) = (
-            &mut self.str_field_hash_map,
-            &mut self.arena,
-            &mut self.dictionaries,
-        );
-        hash_map.mutate_or_create(
-            column_name.as_bytes(),
-            |column_opt: Option<StrOrBytesColumnWriter>| {
-                let mut column: StrOrBytesColumnWriter = column_opt.unwrap_or_else(|| {
-                    // Each column has its own dictionary
-                    let dictionary_id = dictionaries.len() as u32;
-                    dictionaries.push(DictionaryBuilder::default());
-                    StrOrBytesColumnWriter::with_dictionary_id(dictionary_id)
-                });
-                column.record_bytes(doc, value.as_bytes(), dictionaries, arena);
-                column
-            },
-        );
-    }
-
-    pub fn record_bytes(&mut self, doc: RowId, column_name: &str, value: &[u8]) {
+    pub fn record_str(&mut self, doc: DocId, column_name: &str, value: &str) {
         assert!(
             !column_name.as_bytes().contains(&0u8),
             "key may not contain the 0 byte"
@@ -262,61 +116,50 @@ impl ColumnarWriter {
         );
         hash_map.mutate_or_create(
             column_name.as_bytes(),
-            |column_opt: Option<StrOrBytesColumnWriter>| {
-                let mut column: StrOrBytesColumnWriter = column_opt.unwrap_or_else(|| {
+            |column_opt: Option<StrColumnWriter>| {
+                let mut column: StrColumnWriter = column_opt.unwrap_or_else(|| {
                     // Each column has its own dictionary
                     let dictionary_id = dictionaries.len() as u32;
                     dictionaries.push(DictionaryBuilder::default());
-                    StrOrBytesColumnWriter::with_dictionary_id(dictionary_id)
+                    StrColumnWriter::with_dictionary_id(dictionary_id)
                 });
-                column.record_bytes(doc, value, dictionaries, arena);
+                column.record_bytes(doc, value.as_bytes(), dictionaries, arena);
                 column
             },
         );
     }
-    pub fn serialize(&mut self, num_docs: RowId, wrt: &mut dyn io::Write) -> io::Result<()> {
+
+    pub fn serialize(&mut self, num_docs: DocId, wrt: &mut dyn io::Write) -> io::Result<()> {
         let mut serializer = ColumnarSerializer::new(wrt);
-        let mut columns: Vec<(&[u8], ColumnTypeCategory, Addr)> = self
+        let mut field_columns: Vec<(&[u8], ColumnTypeCategory, Addr)> = self
             .numerical_field_hash_map
             .iter()
-            .map(|(column_name, addr, _)| (column_name, ColumnTypeCategory::Numerical, addr))
+            .map(|(term, addr, _)| (term, ColumnTypeCategory::Numerical, addr))
             .collect();
-        columns.extend(
+        field_columns.extend(
             self.bytes_field_hash_map
                 .iter()
-                .map(|(term, addr, _)| (term, ColumnTypeCategory::Bytes, addr)),
+                .map(|(term, addr, _)| (term, ColumnTypeCategory::Str, addr)),
         );
-        columns.extend(
-            self.str_field_hash_map
-                .iter()
-                .map(|(column_name, addr, _)| (column_name, ColumnTypeCategory::Str, addr)),
-        );
-        columns.extend(
+        field_columns.extend(
             self.bool_field_hash_map
                 .iter()
-                .map(|(column_name, addr, _)| (column_name, ColumnTypeCategory::Bool, addr)),
+                .map(|(term, addr, _)| (term, ColumnTypeCategory::Bool, addr)),
         );
-        columns.extend(
-            self.ip_addr_field_hash_map
-                .iter()
-                .map(|(column_name, addr, _)| (column_name, ColumnTypeCategory::IpAddr, addr)),
-        );
-        columns.extend(
-            self.datetime_field_hash_map
-                .iter()
-                .map(|(column_name, addr, _)| (column_name, ColumnTypeCategory::DateTime, addr)),
-        );
-        columns.sort_unstable_by_key(|(column_name, col_type, _)| (*column_name, *col_type));
-
+        field_columns.sort_unstable_by_key(|(column_name, col_type, _)| (*column_name, *col_type));
         let (arena, buffers, dictionaries) = (&self.arena, &mut self.buffers, &self.dictionaries);
         let mut symbol_byte_buffer: Vec<u8> = Vec::new();
-        for (column_name, column_type, addr) in columns {
-            match column_type {
+        for (column_name, bytes_or_numerical, addr) in field_columns {
+            match bytes_or_numerical {
                 ColumnTypeCategory::Bool => {
                     let column_writer: ColumnWriter = self.bool_field_hash_map.read(addr);
                     let cardinality = column_writer.get_cardinality(num_docs);
+                    let column_type_and_cardinality = ColumnTypeAndCardinality {
+                        cardinality,
+                        typ: ColumnType::Bool,
+                    };
                     let mut column_serializer =
-                        serializer.serialize_column(column_name, ColumnType::Bool);
+                        serializer.serialize_column(column_name, column_type_and_cardinality);
                     serialize_bool_column(
                         cardinality,
                         num_docs,
@@ -325,32 +168,18 @@ impl ColumnarWriter {
                         &mut column_serializer,
                     )?;
                 }
-                ColumnTypeCategory::IpAddr => {
-                    let column_writer: ColumnWriter = self.ip_addr_field_hash_map.read(addr);
-                    let cardinality = column_writer.get_cardinality(num_docs);
-                    let mut column_serializer =
-                        serializer.serialize_column(column_name, ColumnType::IpAddr);
-                    serialize_ip_addr_column(
-                        cardinality,
-                        num_docs,
-                        column_writer.operation_iterator(arena, &mut symbol_byte_buffer),
-                        buffers,
-                        &mut column_serializer,
-                    )?;
-                }
-                ColumnTypeCategory::Bytes | ColumnTypeCategory::Str => {
-                    let (column_type, str_column_writer): (ColumnType, StrOrBytesColumnWriter) =
-                        if column_type == ColumnTypeCategory::Bytes {
-                            (ColumnType::Bytes, self.bytes_field_hash_map.read(addr))
-                        } else {
-                            (ColumnType::Str, self.str_field_hash_map.read(addr))
-                        };
+                ColumnTypeCategory::Str => {
+                    let str_column_writer: StrColumnWriter = self.bytes_field_hash_map.read(addr);
                     let dictionary_builder =
                         &dictionaries[str_column_writer.dictionary_id as usize];
                     let cardinality = str_column_writer.column_writer.get_cardinality(num_docs);
+                    let column_type_and_cardinality = ColumnTypeAndCardinality {
+                        cardinality,
+                        typ: ColumnType::Bytes,
+                    };
                     let mut column_serializer =
-                        serializer.serialize_column(column_name, column_type);
-                    serialize_bytes_or_str_column(
+                        serializer.serialize_column(column_name, column_type_and_cardinality);
+                    serialize_bytes_column(
                         cardinality,
                         num_docs,
                         dictionary_builder,
@@ -364,8 +193,12 @@ impl ColumnarWriter {
                         self.numerical_field_hash_map.read(addr);
                     let (numerical_type, cardinality) =
                         numerical_column_writer.column_type_and_cardinality(num_docs);
+                    let column_type_and_cardinality = ColumnTypeAndCardinality {
+                        cardinality,
+                        typ: ColumnType::Numerical(numerical_type),
+                    };
                     let mut column_serializer =
-                        serializer.serialize_column(column_name, ColumnType::from(numerical_type));
+                        serializer.serialize_column(column_name, column_type_and_cardinality);
                     serialize_numerical_column(
                         cardinality,
                         num_docs,
@@ -375,20 +208,6 @@ impl ColumnarWriter {
                         &mut column_serializer,
                     )?;
                 }
-                ColumnTypeCategory::DateTime => {
-                    let column_writer: ColumnWriter = self.datetime_field_hash_map.read(addr);
-                    let cardinality = column_writer.get_cardinality(num_docs);
-                    let mut column_serializer =
-                        serializer.serialize_column(column_name, ColumnType::DateTime);
-                    serialize_numerical_column(
-                        cardinality,
-                        num_docs,
-                        NumericalType::I64,
-                        column_writer.operation_iterator(arena, &mut symbol_byte_buffer),
-                        buffers,
-                        &mut column_serializer,
-                    )?;
-                }
             };
         }
         serializer.finalize()?;
@@ -396,9 +215,9 @@ impl ColumnarWriter {
     }
 }
 
-fn serialize_bytes_or_str_column(
+fn serialize_bytes_column(
     cardinality: Cardinality,
-    num_docs: RowId,
+    num_docs: DocId,
     dictionary_builder: &DictionaryBuilder,
     operation_it: impl Iterator<Item = ColumnOperation<UnorderedId>>,
     buffers: &mut SpareBuffers,
@@ -423,7 +242,7 @@ fn serialize_bytes_or_str_column(
             ColumnOperation::NewDoc(doc) => ColumnOperation::NewDoc(doc),
         }
     });
-    send_to_serialize_column_mappable_to_u64(
+    serialize_column(
         operation_iterator,
         cardinality,
         num_docs,
@@ -437,7 +256,7 @@ fn serialize_bytes_or_str_column(
 
 fn serialize_numerical_column(
     cardinality: Cardinality,
-    num_docs: RowId,
+    num_docs: DocId,
     numerical_type: NumericalType,
     op_iterator: impl Iterator<Item = ColumnOperation<NumericalValue>>,
     buffers: &mut SpareBuffers,
@@ -452,7 +271,7 @@ fn serialize_numerical_column(
     } = buffers;
     match numerical_type {
         NumericalType::I64 => {
-            send_to_serialize_column_mappable_to_u64(
+            serialize_column(
                 coerce_numerical_symbol::<i64>(op_iterator),
                 cardinality,
                 num_docs,
@@ -462,7 +281,7 @@ fn serialize_numerical_column(
             )?;
         }
         NumericalType::U64 => {
-            send_to_serialize_column_mappable_to_u64(
+            serialize_column(
                 coerce_numerical_symbol::<u64>(op_iterator),
                 cardinality,
                 num_docs,
@@ -472,7 +291,7 @@ fn serialize_numerical_column(
             )?;
         }
         NumericalType::F64 => {
-            send_to_serialize_column_mappable_to_u64(
+            serialize_column(
                 coerce_numerical_symbol::<f64>(op_iterator),
                 cardinality,
                 num_docs,
@@ -487,7 +306,7 @@ fn serialize_numerical_column(
 
 fn serialize_bool_column(
     cardinality: Cardinality,
-    num_docs: RowId,
+    num_docs: DocId,
     column_operations_it: impl Iterator<Item = ColumnOperation<bool>>,
     buffers: &mut SpareBuffers,
     wrt: &mut impl io::Write,
@@ -497,7 +316,7 @@ fn serialize_bool_column(
         bool_values,
         ..
     } = buffers;
-    send_to_serialize_column_mappable_to_u64(
+    serialize_column(
         column_operations_it,
         cardinality,
         num_docs,
@@ -508,116 +327,56 @@ fn serialize_bool_column(
     Ok(())
 }
 
-fn serialize_ip_addr_column(
-    cardinality: Cardinality,
-    num_docs: RowId,
-    column_operations_it: impl Iterator<Item = ColumnOperation<Ipv6Addr>>,
-    buffers: &mut SpareBuffers,
-    wrt: &mut impl io::Write,
-) -> io::Result<()> {
-    let SpareBuffers {
-        value_index_builders,
-        ip_addr_values,
-        ..
-    } = buffers;
-    send_to_serialize_column_mappable_to_u128(
-        column_operations_it,
-        cardinality,
-        num_docs,
-        value_index_builders,
-        ip_addr_values,
-        wrt,
-    )?;
-    Ok(())
-}
-
-fn send_to_serialize_column_mappable_to_u128<
-    T: Copy + std::fmt::Debug + Send + Sync + MonotonicallyMappableToU128 + PartialOrd,
->(
-    op_iterator: impl Iterator<Item = ColumnOperation<T>>,
-    cardinality: Cardinality,
-    num_docs: RowId,
-    value_index_builders: &mut PreallocatedIndexBuilders,
-    values: &mut Vec<T>,
-    mut wrt: impl io::Write,
-) -> io::Result<()>
-where
-    for<'a> VecColumn<'a, T>: ColumnValues<T>,
-{
-    values.clear();
-    // TODO: split index and values
-    let serializable_column_index = match cardinality {
-        Cardinality::Full => {
-            consume_operation_iterator(
-                op_iterator,
-                value_index_builders.borrow_required_index_builder(),
-                values,
-            );
-            SerializableColumnIndex::Full
-        }
-        Cardinality::Optional => {
-            let optional_index_builder = value_index_builders.borrow_optional_index_builder();
-            consume_operation_iterator(op_iterator, optional_index_builder, values);
-            let optional_index = optional_index_builder.finish(num_docs);
-            SerializableColumnIndex::Optional(Box::new(optional_index))
-        }
-        Cardinality::Multivalued => {
-            let multivalued_index_builder = value_index_builders.borrow_multivalued_index_builder();
-            consume_operation_iterator(op_iterator, multivalued_index_builder, values);
-            let multivalued_index = multivalued_index_builder.finish(num_docs);
-            SerializableColumnIndex::Multivalued(Box::new(multivalued_index))
-        }
-    };
-    crate::column::serialize_column_mappable_to_u128(
-        serializable_column_index,
-        || values.iter().cloned(),
-        values.len() as u32,
-        &mut wrt,
-    )?;
-    Ok(())
-}
-
-fn send_to_serialize_column_mappable_to_u64<
+fn serialize_column<
     T: Copy + Default + std::fmt::Debug + Send + Sync + MonotonicallyMappableToU64 + PartialOrd,
 >(
     op_iterator: impl Iterator<Item = ColumnOperation<T>>,
     cardinality: Cardinality,
-    num_docs: RowId,
-    value_index_builders: &mut PreallocatedIndexBuilders,
+    num_docs: DocId,
+    value_index_builders: &mut SpareIndexBuilders,
     values: &mut Vec<T>,
     mut wrt: impl io::Write,
 ) -> io::Result<()>
 where
-    for<'a> VecColumn<'a, T>: ColumnValues<T>,
+    for<'a> VecColumn<'a, T>: Column<T>,
 {
     values.clear();
-    let serializable_column_index = match cardinality {
-        Cardinality::Full => {
+    match cardinality {
+        Cardinality::Required => {
             consume_operation_iterator(
                 op_iterator,
                 value_index_builders.borrow_required_index_builder(),
                 values,
             );
-            SerializableColumnIndex::Full
+            fastfield_codecs::serialize(
+                VecColumn::from(&values[..]),
+                &mut wrt,
+                &fastfield_codecs::ALL_CODEC_TYPES[..],
+            )?;
         }
         Cardinality::Optional => {
             let optional_index_builder = value_index_builders.borrow_optional_index_builder();
             consume_operation_iterator(op_iterator, optional_index_builder, values);
             let optional_index = optional_index_builder.finish(num_docs);
-            SerializableColumnIndex::Optional(Box::new(optional_index))
+            fastfield_codecs::serialize::serialize_new(
+                ValueIndexInfo::SingleValue(Box::new(optional_index)),
+                VecColumn::from(&values[..]),
+                &mut wrt,
+                &fastfield_codecs::ALL_CODEC_TYPES[..],
+            )?;
         }
         Cardinality::Multivalued => {
             let multivalued_index_builder = value_index_builders.borrow_multivalued_index_builder();
             consume_operation_iterator(op_iterator, multivalued_index_builder, values);
             let multivalued_index = multivalued_index_builder.finish(num_docs);
-            SerializableColumnIndex::Multivalued(Box::new(multivalued_index))
+            fastfield_codecs::serialize::serialize_new(
+                ValueIndexInfo::MultiValue(Box::new(multivalued_index)),
+                VecColumn::from(&values[..]),
+                &mut wrt,
+                &fastfield_codecs::ALL_CODEC_TYPES[..],
+            )?;
         }
-    };
-    crate::column::serialize_column_mappable_to_u64(
-        serializable_column_index,
-        &VecColumn::from(&values[..]),
-        &mut wrt,
-    )?;
+    }
     Ok(())
 }
 
@@ -641,7 +400,7 @@ fn consume_operation_iterator<T: std::fmt::Debug, TIndexBuilder: IndexBuilder>(
     for symbol in operation_iterator {
         match symbol {
             ColumnOperation::NewDoc(doc) => {
-                index_builder.record_row(doc);
+                index_builder.record_doc(doc);
             }
             ColumnOperation::Value(value) => {
                 index_builder.record_value();
@@ -653,10 +412,12 @@ fn consume_operation_iterator<T: std::fmt::Debug, TIndexBuilder: IndexBuilder>(
 
 #[cfg(test)]
 mod tests {
+    use column_operation::ColumnOperation;
     use stacker::MemoryArena;
 
-    use crate::columnar::writer::column_operation::ColumnOperation;
-    use crate::{Cardinality, NumericalValue};
+    use super::*;
+    use crate::value::NumericalValue;
+    use crate::Cardinality;
 
     #[test]
     fn test_column_writer_required_simple() {
@@ -665,7 +426,7 @@ mod tests {
         column_writer.record(0u32, NumericalValue::from(14i64), &mut arena);
         column_writer.record(1u32, NumericalValue::from(15i64), &mut arena);
         column_writer.record(2u32, NumericalValue::from(-16i64), &mut arena);
-        assert_eq!(column_writer.get_cardinality(3), Cardinality::Full);
+        assert_eq!(column_writer.get_cardinality(3), Cardinality::Required);
         let mut buffer = Vec::new();
         let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
             .operation_iterator(&mut arena, &mut buffer)

columnar/src/writer/serializer.rs

@@ -5,7 +5,7 @@ use common::CountingWriter;
 use sstable::value::RangeValueWriter;
 use sstable::RangeSSTable;
 
-use crate::columnar::ColumnType;
+use crate::column_type_header::ColumnTypeAndCardinality;
 
 pub struct ColumnarSerializer<W: io::Write> {
     wrt: CountingWriter<W>,
@@ -15,11 +15,15 @@ pub struct ColumnarSerializer<W: io::Write> {
 
 /// Returns a key consisting of the concatenation of the key and the column_type_and_cardinality
 /// code.
-fn prepare_key(key: &[u8], column_type: ColumnType, buffer: &mut Vec<u8>) {
+fn prepare_key(
+    key: &[u8],
+    column_type_cardinality: ColumnTypeAndCardinality,
+    buffer: &mut Vec<u8>,
+) {
     buffer.clear();
     buffer.extend_from_slice(key);
     buffer.push(0u8);
-    buffer.push(column_type.to_code());
+    buffer.push(column_type_cardinality.to_code());
 }
 
 impl<W: io::Write> ColumnarSerializer<W> {
@@ -36,10 +40,14 @@ impl<W: io::Write> ColumnarSerializer<W> {
     pub fn serialize_column<'a>(
         &'a mut self,
         column_name: &[u8],
-        column_type: ColumnType,
+        column_type_cardinality: ColumnTypeAndCardinality,
     ) -> impl io::Write + 'a {
         let start_offset = self.wrt.written_bytes();
-        prepare_key(column_name, column_type, &mut self.prepare_key_buffer);
+        prepare_key(
+            column_name,
+            column_type_cardinality,
+            &mut self.prepare_key_buffer,
+        );
         ColumnSerializer {
             columnar_serializer: self,
             start_offset,
@@ -51,9 +59,6 @@ impl<W: io::Write> ColumnarSerializer<W> {
         let sstable_num_bytes: u64 = sstable_bytes.len() as u64;
         self.wrt.write_all(&sstable_bytes)?;
         self.wrt.write_all(&sstable_num_bytes.to_le_bytes()[..])?;
-        self.wrt
-            .write_all(&super::super::format_version::footer())?;
-        self.wrt.flush()?;
         Ok(())
     }
 }
@@ -92,15 +97,20 @@ impl<'a, W: io::Write> io::Write for ColumnSerializer<'a, W> {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use crate::columnar::column_type::ColumnType;
+    use crate::column_type_header::ColumnType;
+    use crate::Cardinality;
 
     #[test]
     fn test_prepare_key_bytes() {
         let mut buffer: Vec<u8> = b"somegarbage".to_vec();
-        prepare_key(b"root\0child", ColumnType::Str, &mut buffer);
+        let column_type_and_cardinality = ColumnTypeAndCardinality {
+            typ: ColumnType::Bytes,
+            cardinality: Cardinality::Optional,
+        };
+        prepare_key(b"root\0child", column_type_and_cardinality, &mut buffer);
         assert_eq!(buffer.len(), 12);
         assert_eq!(&buffer[..10], b"root\0child");
         assert_eq!(buffer[10], 0u8);
-        assert_eq!(buffer[11], ColumnType::Str.to_code());
+        assert_eq!(buffer[11], column_type_and_cardinality.to_code());
     }
 }

columnar/src/writer/value_index.rs

@@ -1,6 +1,6 @@
-use crate::column_index::SerializableOptionalIndex;
-use crate::column_values::{ColumnValues, VecColumn};
-use crate::RowId;
+use fastfield_codecs::serialize::{MultiValueIndexInfo, SingleValueIndexInfo};
+
+use crate::DocId;
 
 /// The `IndexBuilder` interprets a sequence of
 /// calls of the form:
@@ -10,52 +10,56 @@ use crate::RowId;
 /// It has different implementation depending on whether the
 /// cardinality is required, optional, or multivalued.
 pub(crate) trait IndexBuilder {
-    fn record_row(&mut self, doc: RowId);
+    fn record_doc(&mut self, doc: DocId);
     #[inline]
     fn record_value(&mut self) {}
 }
 
-/// The FullIndexBuilder does nothing.
+/// The RequiredIndexBuilder does nothing.
 #[derive(Default)]
-pub struct FullIndexBuilder;
+pub struct RequiredIndexBuilder;
 
-impl IndexBuilder for FullIndexBuilder {
+impl IndexBuilder for RequiredIndexBuilder {
     #[inline(always)]
-    fn record_row(&mut self, _doc: RowId) {}
+    fn record_doc(&mut self, _doc: DocId) {}
 }
 
 #[derive(Default)]
 pub struct OptionalIndexBuilder {
-    docs: Vec<RowId>,
+    docs: Vec<DocId>,
 }
 
 struct SingleValueArrayIndex<'a> {
-    // RowIds with a value, in a strictly increasing order
-    row_ids: &'a [RowId],
-    num_rows: RowId,
+    // DocIds with a value. DocIds are strictly increasing
+    docs: &'a [DocId],
+    num_docs: DocId,
 }
 
-impl<'a> SerializableOptionalIndex<'a> for SingleValueArrayIndex<'a> {
-    fn num_rows(&self) -> RowId {
-        self.num_rows
+impl<'a> SingleValueIndexInfo for SingleValueArrayIndex<'a> {
+    fn num_vals(&self) -> u32 {
+        self.num_docs as u32
     }
 
-    fn non_null_rows(&self) -> Box<dyn Iterator<Item = RowId> + 'a> {
-        Box::new(self.row_ids.iter().copied())
+    fn num_non_nulls(&self) -> u32 {
+        self.docs.len() as u32
+    }
+
+    fn iter(&self) -> Box<dyn Iterator<Item = u32> + '_> {
+        Box::new(self.docs.iter().copied())
     }
 }
 
 impl OptionalIndexBuilder {
-    pub fn finish<'a>(&'a mut self, num_rows: RowId) -> impl SerializableOptionalIndex + 'a {
+    pub fn finish(&mut self, num_docs: DocId) -> impl SingleValueIndexInfo + '_ {
         debug_assert!(self
             .docs
             .last()
             .copied()
-            .map(|last_doc| last_doc < num_rows)
+            .map(|last_doc| last_doc < num_docs)
             .unwrap_or(true));
         SingleValueArrayIndex {
-            row_ids: &self.docs[..],
-            num_rows,
+            docs: &self.docs[..],
+            num_docs,
         }
     }
 
@@ -66,7 +70,7 @@ impl OptionalIndexBuilder {
 
 impl IndexBuilder for OptionalIndexBuilder {
     #[inline(always)]
-    fn record_row(&mut self, doc: RowId) {
+    fn record_doc(&mut self, doc: DocId) {
         debug_assert!(self
             .docs
             .last()
@@ -79,32 +83,53 @@ impl IndexBuilder for OptionalIndexBuilder {
 
 #[derive(Default)]
 pub struct MultivaluedIndexBuilder {
-    start_offsets: Vec<RowId>,
+    // TODO should we switch to `start_offset`?
+    // contains the num values so far for each `DocId`.
+    end_offsets: Vec<DocId>,
     total_num_vals_seen: u32,
 }
 
+pub struct MultivaluedValueArrayIndex<'a> {
+    end_offsets: &'a [DocId],
+}
+
+impl<'a> MultiValueIndexInfo for MultivaluedValueArrayIndex<'a> {
+    fn num_docs(&self) -> u32 {
+        self.end_offsets.len() as u32
+    }
+
+    fn num_vals(&self) -> u32 {
+        self.end_offsets.last().copied().unwrap_or(0u32)
+    }
+
+    fn iter(&self) -> Box<dyn Iterator<Item = u32> + '_> {
+        if self.end_offsets.is_empty() {
+            return Box::new(std::iter::empty());
+        }
+        let n = self.end_offsets.len();
+        Box::new(std::iter::once(0u32).chain(self.end_offsets[..n - 1].iter().copied()))
+    }
+}
+
 impl MultivaluedIndexBuilder {
-    pub fn finish(&mut self, num_docs: RowId) -> impl ColumnValues<u32> + '_ {
-        self.start_offsets
-            .resize(num_docs as usize + 1, self.total_num_vals_seen);
-        VecColumn {
-            values: &&self.start_offsets[..],
-            min_value: 0,
-            max_value: self.start_offsets.last().copied().unwrap_or(0),
+    pub fn finish(&mut self, num_docs: DocId) -> impl MultiValueIndexInfo + '_ {
+        self.end_offsets
+            .resize(num_docs as usize, self.total_num_vals_seen);
+        MultivaluedValueArrayIndex {
+            end_offsets: &self.end_offsets[..],
         }
     }
 
     fn reset(&mut self) {
-        self.start_offsets.clear();
-        self.start_offsets.push(0u32);
+        self.end_offsets.clear();
         self.total_num_vals_seen = 0;
     }
 }
 
 impl IndexBuilder for MultivaluedIndexBuilder {
-    fn record_row(&mut self, row_id: RowId) {
-        self.start_offsets
-            .resize(row_id as usize + 1, self.total_num_vals_seen);
+    fn record_doc(&mut self, doc: DocId) {
+        self.end_offsets
+            .resize(doc as usize, self.total_num_vals_seen);
     }
 
     fn record_value(&mut self) {
@@ -115,14 +140,14 @@ impl IndexBuilder for MultivaluedIndexBuilder {
 /// The `SpareIndexBuilders` is there to avoid allocating a
 /// new index builder for every single column.
 #[derive(Default)]
-pub struct PreallocatedIndexBuilders {
-    required_index_builder: FullIndexBuilder,
+pub struct SpareIndexBuilders {
+    required_index_builder: RequiredIndexBuilder,
     optional_index_builder: OptionalIndexBuilder,
     multivalued_index_builder: MultivaluedIndexBuilder,
 }
 
-impl PreallocatedIndexBuilders {
-    pub fn borrow_required_index_builder(&mut self) -> &mut FullIndexBuilder {
+impl SpareIndexBuilders {
+    pub fn borrow_required_index_builder(&mut self) -> &mut RequiredIndexBuilder {
         &mut self.required_index_builder
     }
 
@@ -144,22 +169,22 @@ mod tests {
     #[test]
     fn test_optional_value_index_builder() {
         let mut opt_value_index_builder = OptionalIndexBuilder::default();
-        opt_value_index_builder.record_row(0u32);
+        opt_value_index_builder.record_doc(0u32);
         opt_value_index_builder.record_value();
         assert_eq!(
             &opt_value_index_builder
                 .finish(1u32)
-                .non_null_rows()
+                .iter()
                 .collect::<Vec<u32>>(),
             &[0]
         );
         opt_value_index_builder.reset();
-        opt_value_index_builder.record_row(1u32);
+        opt_value_index_builder.record_doc(1u32);
         opt_value_index_builder.record_value();
         assert_eq!(
             &opt_value_index_builder
                 .finish(2u32)
-                .non_null_rows()
+                .iter()
                 .collect::<Vec<u32>>(),
             &[1]
         );
@@ -168,20 +193,20 @@ mod tests {
     #[test]
     fn test_multivalued_value_index_builder() {
         let mut multivalued_value_index_builder = MultivaluedIndexBuilder::default();
-        multivalued_value_index_builder.record_row(1u32);
+        multivalued_value_index_builder.record_doc(1u32);
         multivalued_value_index_builder.record_value();
         multivalued_value_index_builder.record_value();
-        multivalued_value_index_builder.record_row(2u32);
+        multivalued_value_index_builder.record_doc(2u32);
         multivalued_value_index_builder.record_value();
         assert_eq!(
             multivalued_value_index_builder
                 .finish(4u32)
                 .iter()
                 .collect::<Vec<u32>>(),
-            vec![0, 0, 2, 3, 3]
+            vec![0, 0, 2, 3]
         );
         multivalued_value_index_builder.reset();
-        multivalued_value_index_builder.record_row(2u32);
+        multivalued_value_index_builder.record_doc(2u32);
         multivalued_value_index_builder.record_value();
         multivalued_value_index_builder.record_value();
         assert_eq!(
@@ -189,7 +214,7 @@ mod tests {
                 .finish(4u32)
                 .iter()
                 .collect::<Vec<u32>>(),
-            vec![0, 0, 0, 2, 2]
+            vec![0, 0, 0, 2]
         );
     }
 }

examples/aggregation.rs

@@ -27,7 +27,7 @@ fn main() -> tantivy::Result<()> {
     let score_fieldtype =
         crate::schema::NumericOptions::default().set_fast(Cardinality::SingleValue);
     let highscore_field = schema_builder.add_f64_field("highscore", score_fieldtype.clone());
-    let price_field = schema_builder.add_f64_field("price", score_fieldtype);
+    let price_field = schema_builder.add_f64_field("price", score_fieldtype.clone());
 
     let schema = schema_builder.build();
 
@@ -112,7 +112,7 @@ fn main() -> tantivy::Result<()> {
                 ],
                 ..Default::default()
             }),
-            sub_aggregation: sub_agg_req_1,
+            sub_aggregation: sub_agg_req_1.clone(),
         }),
     )]
     .into_iter()
@@ -123,7 +123,7 @@ fn main() -> tantivy::Result<()> {
     let searcher = reader.searcher();
     let agg_res: AggregationResults = searcher.search(&term_query, &collector).unwrap();
 
-    let res: Value = serde_json::to_value(agg_res)?;
+    let res: Value = serde_json::to_value(&agg_res)?;
     println!("{}", serde_json::to_string_pretty(&res)?);
 
     Ok(())

examples/custom_collector.rs

@@ -14,7 +14,7 @@ use fastfield_codecs::Column;
 // Importing tantivy...
 use tantivy::collector::{Collector, SegmentCollector};
 use tantivy::query::QueryParser;
-use tantivy::schema::{Schema, FAST, INDEXED, TEXT};
+use tantivy::schema::{Field, Schema, FAST, INDEXED, TEXT};
 use tantivy::{doc, Index, Score, SegmentReader};
 
 #[derive(Default)]
@@ -52,11 +52,11 @@ impl Stats {
 }
 
 struct StatsCollector {
-    field: String,
+    field: Field,
 }
 
 impl StatsCollector {
-    fn with_field(field: String) -> StatsCollector {
+    fn with_field(field: Field) -> StatsCollector {
         StatsCollector { field }
     }
 }
@@ -73,7 +73,7 @@ impl Collector for StatsCollector {
         _segment_local_id: u32,
         segment_reader: &SegmentReader,
     ) -> tantivy::Result<StatsSegmentCollector> {
-        let fast_field_reader = segment_reader.fast_fields().u64(&self.field)?;
+        let fast_field_reader = segment_reader.fast_fields().u64(self.field)?;
         Ok(StatsSegmentCollector {
             fast_field_reader,
             stats: Stats::default(),
@@ -171,9 +171,7 @@ fn main() -> tantivy::Result<()> {
 
     // here we want to get a hit on the 'ken' in Frankenstein
     let query = query_parser.parse_query("broom")?;
-    if let Some(stats) =
-        searcher.search(&query, &StatsCollector::with_field("price".to_string()))?
-    {
+    if let Some(stats) = searcher.search(&query, &StatsCollector::with_field(price))? {
         println!("count: {}", stats.count());
         println!("mean: {}", stats.mean());
         println!("standard deviation: {}", stats.standard_deviation());

examples/integer_range_search.rs

@@ -27,7 +27,7 @@ fn main() -> Result<()> {
     reader.reload()?;
     let searcher = reader.searcher();
     // The end is excluded i.e. here we are searching up to 1969
-    let docs_in_the_sixties = RangeQuery::new_u64("year".to_string(), 1960..1970);
+    let docs_in_the_sixties = RangeQuery::new_u64(year_field, 1960..1970);
     // Uses a Count collector to sum the total number of docs in the range
     let num_60s_books = searcher.search(&docs_in_the_sixties, &Count)?;
     assert_eq!(num_60s_books, 10);

examples/ip_field.rs

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

examples/warmer.rs

@@ -4,7 +4,7 @@ use std::sync::{Arc, RwLock, Weak};
 
 use tantivy::collector::TopDocs;
 use tantivy::query::QueryParser;
-use tantivy::schema::{Schema, FAST, TEXT};
+use tantivy::schema::{Field, Schema, FAST, TEXT};
 use tantivy::{
     doc, DocAddress, DocId, Index, IndexReader, Opstamp, Searcher, SearcherGeneration, SegmentId,
     SegmentReader, Warmer,
@@ -25,13 +25,13 @@ pub trait PriceFetcher: Send + Sync + 'static {
 }
 
 struct DynamicPriceColumn {
-    field: String,
+    field: Field,
     price_cache: RwLock<HashMap<(SegmentId, Option<Opstamp>), Arc<Vec<Price>>>>,
     price_fetcher: Box<dyn PriceFetcher>,
 }
 
 impl DynamicPriceColumn {
-    pub fn with_product_id_field<T: PriceFetcher>(field: String, price_fetcher: T) -> Self {
+    pub fn with_product_id_field<T: PriceFetcher>(field: Field, price_fetcher: T) -> Self {
         DynamicPriceColumn {
             field,
             price_cache: Default::default(),
@@ -48,7 +48,7 @@ impl Warmer for DynamicPriceColumn {
     fn warm(&self, searcher: &Searcher) -> tantivy::Result<()> {
         for segment in searcher.segment_readers() {
             let key = (segment.segment_id(), segment.delete_opstamp());
-            let product_id_reader = segment.fast_fields().u64(&self.field)?;
+            let product_id_reader = segment.fast_fields().u64(self.field)?;
             let product_ids: Vec<ProductId> = segment
                 .doc_ids_alive()
                 .map(|doc| product_id_reader.get_val(doc))
@@ -123,7 +123,7 @@ fn main() -> tantivy::Result<()> {
 
     let price_table = ExternalPriceTable::default();
     let price_dynamic_column = Arc::new(DynamicPriceColumn::with_product_id_field(
-        "product_id".to_string(),
+        product_id,
         price_table.clone(),
     ));
     price_table.update_price(OLIVE_OIL, 12);

fastfield_codecs/src/column.rs

@@ -35,6 +35,10 @@ pub trait Column<T: PartialOrd + Debug = u64>: Send + Sync {
     /// Get the positions of values which are in the provided value range.
     ///
     /// Note that position == docid for single value fast fields
+    ///
+    /// # Truncation
+    /// `DateTime` has a truncation setting. This function should get passed the truncated values
+    /// to avoid unexpected results.
     #[inline]
     fn get_docids_for_value_range(
         &self,
@@ -137,7 +141,8 @@ impl<'a, T: Copy + PartialOrd + Send + Sync + Debug> Column<T> for VecColumn<'a,
 }
 
 impl<'a, T: Copy + PartialOrd + Default, V> From<&'a V> for VecColumn<'a, T>
-where V: AsRef<[T]> + ?Sized
+where
+    V: AsRef<[T]> + ?Sized,
 {
     fn from(values: &'a V) -> Self {
         let values = values.as_ref();
@@ -248,7 +253,8 @@ where
 pub struct IterColumn<T>(T);
 
 impl<T> From<T> for IterColumn<T>
-where T: Iterator + Clone + ExactSizeIterator
+where
+    T: Iterator + Clone + ExactSizeIterator,
 {
     fn from(iter: T) -> Self {
         IterColumn(iter)

fastfield_codecs/src/serialize.rs

@@ -402,8 +402,8 @@ mod tests {
         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);
+        // 5 bytes of header, 1 byte of value, 7 bytes of padding.
+        assert_eq!(buffer.len(), 3 + 5 + 8 + 4 + 2);
     }
 
     #[test]
@@ -411,8 +411,8 @@ mod tests {
         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);
+        // 5 bytes of header, 0 bytes of value, 7 bytes of padding.
+        assert_eq!(buffer.len(), 3 + 5 + 7 + 4 + 2);
     }
 
     #[test]
@@ -422,6 +422,6 @@ mod tests {
         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);
+        assert_eq!(buffer.len(), 3 + 7 + (3 * 80 / 8) + 7 + 4 + 2);
     }
 }

run-tests.sh

@@ -0,0 +1,2 @@
+#!/bin/bash
+cargo test

src/aggregation/agg_req.rs

@@ -51,10 +51,7 @@ use serde::{Deserialize, Serialize};
 
 pub use super::bucket::RangeAggregation;
 use super::bucket::{HistogramAggregation, TermsAggregation};
-use super::metric::{
-    AverageAggregation, CountAggregation, MaxAggregation, MinAggregation, StatsAggregation,
-    SumAggregation,
-};
+use super::metric::{AverageAggregation, StatsAggregation};
 use super::VecWithNames;
 
 /// The top-level aggregation request structure, which contains [`Aggregation`] and their user
@@ -240,38 +237,20 @@ impl BucketAggregationType {
 /// called multi-value numeric metrics aggregation.
 #[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
 pub enum MetricAggregation {
-    /// Computes the average of the extracted values.
+    /// Calculates the average.
     #[serde(rename = "avg")]
     Average(AverageAggregation),
-    /// Counts the number of extracted values.
-    #[serde(rename = "value_count")]
-    Count(CountAggregation),
-    /// Finds the maximum value.
-    #[serde(rename = "max")]
-    Max(MaxAggregation),
-    /// Finds the minimum value.
-    #[serde(rename = "min")]
-    Min(MinAggregation),
-    /// Computes a collection of statistics (`min`, `max`, `sum`, `count`, and `avg`) over the
-    /// extracted values.
+    /// Calculates stats sum, average, min, max, standard_deviation on a field.
     #[serde(rename = "stats")]
     Stats(StatsAggregation),
-    /// Computes the sum of the extracted values.
-    #[serde(rename = "sum")]
-    Sum(SumAggregation),
 }
 
 impl MetricAggregation {
     fn get_fast_field_names(&self, fast_field_names: &mut HashSet<String>) {
-        let fast_field_name = match self {
-            MetricAggregation::Average(avg) => avg.field_name(),
-            MetricAggregation::Count(count) => count.field_name(),
-            MetricAggregation::Max(max) => max.field_name(),
-            MetricAggregation::Min(min) => min.field_name(),
-            MetricAggregation::Stats(stats) => stats.field_name(),
-            MetricAggregation::Sum(sum) => sum.field_name(),
+        match self {
+            MetricAggregation::Average(avg) => fast_field_names.insert(avg.field.to_string()),
+            MetricAggregation::Stats(stats) => fast_field_names.insert(stats.field.to_string()),
         };
-        fast_field_names.insert(fast_field_name.to_string());
     }
 }
 
@@ -279,38 +258,6 @@ impl MetricAggregation {
 mod tests {
     use super::*;
 
-    #[test]
-    fn test_metric_aggregations_deser() {
-        let agg_req_json = r#"{
-            "price_avg": { "avg": { "field": "price" } },
-            "price_count": { "value_count": { "field": "price" } },
-            "price_max": { "max": { "field": "price" } },
-            "price_min": { "min": { "field": "price" } },
-            "price_stats": { "stats": { "field": "price" } },
-            "price_sum": { "sum": { "field": "price" } }
-        }"#;
-        let agg_req: Aggregations = serde_json::from_str(agg_req_json).unwrap();
-
-        assert!(
-            matches!(agg_req.get("price_avg").unwrap(), Aggregation::Metric(MetricAggregation::Average(avg)) if avg.field == "price")
-        );
-        assert!(
-            matches!(agg_req.get("price_count").unwrap(), Aggregation::Metric(MetricAggregation::Count(count)) if count.field == "price")
-        );
-        assert!(
-            matches!(agg_req.get("price_max").unwrap(), Aggregation::Metric(MetricAggregation::Max(max)) if max.field == "price")
-        );
-        assert!(
-            matches!(agg_req.get("price_min").unwrap(), Aggregation::Metric(MetricAggregation::Min(min)) if min.field == "price")
-        );
-        assert!(
-            matches!(agg_req.get("price_stats").unwrap(), Aggregation::Metric(MetricAggregation::Stats(stats)) if stats.field == "price")
-        );
-        assert!(
-            matches!(agg_req.get("price_sum").unwrap(), Aggregation::Metric(MetricAggregation::Sum(sum)) if sum.field == "price")
-        );
-    }
-
     #[test]
     fn serialize_to_json_test() {
         let agg_req1: Aggregations = vec![(

src/aggregation/agg_req_with_accessor.rs

@@ -8,10 +8,7 @@ use fastfield_codecs::Column;
 
 use super::agg_req::{Aggregation, Aggregations, BucketAggregationType, MetricAggregation};
 use super::bucket::{HistogramAggregation, RangeAggregation, TermsAggregation};
-use super::metric::{
-    AverageAggregation, CountAggregation, MaxAggregation, MinAggregation, StatsAggregation,
-    SumAggregation,
-};
+use super::metric::{AverageAggregation, StatsAggregation};
 use super::segment_agg_result::BucketCount;
 use super::VecWithNames;
 use crate::fastfield::{type_and_cardinality, MultiValuedFastFieldReader};
@@ -94,7 +91,10 @@ impl BucketAggregationWithAccessor {
             BucketAggregationType::Terms(TermsAggregation {
                 field: field_name, ..
             }) => {
-                let field = reader.schema().get_field(field_name)?;
+                let field = reader
+                    .schema()
+                    .get_field(field_name)
+                    .ok_or_else(|| TantivyError::FieldNotFound(field_name.to_string()))?;
                 inverted_index = Some(reader.inverted_index(field)?);
                 get_ff_reader_and_validate(reader, field_name, Cardinality::MultiValues)?
             }
@@ -134,11 +134,7 @@ impl MetricAggregationWithAccessor {
     ) -> crate::Result<MetricAggregationWithAccessor> {
         match &metric {
             MetricAggregation::Average(AverageAggregation { field: field_name })
-            | MetricAggregation::Count(CountAggregation { field: field_name })
-            | MetricAggregation::Max(MaxAggregation { field: field_name })
-            | MetricAggregation::Min(MinAggregation { field: field_name })
-            | MetricAggregation::Stats(StatsAggregation { field: field_name })
-            | MetricAggregation::Sum(SumAggregation { field: field_name }) => {
+            | MetricAggregation::Stats(StatsAggregation { field: field_name }) => {
                 let (accessor, field_type) =
                     get_ff_reader_and_validate(reader, field_name, Cardinality::SingleValue)?;
 
@@ -192,7 +188,10 @@ fn get_ff_reader_and_validate(
     field_name: &str,
     cardinality: Cardinality,
 ) -> crate::Result<(FastFieldAccessor, Type)> {
-    let field = reader.schema().get_field(field_name)?;
+    let field = reader
+        .schema()
+        .get_field(field_name)
+        .ok_or_else(|| TantivyError::FieldNotFound(field_name.to_string()))?;
     let field_type = reader.schema().get_field_entry(field).field_type();
 
     if let Some((_ff_type, field_cardinality)) = type_and_cardinality(field_type) {
@@ -212,10 +211,10 @@ fn get_ff_reader_and_validate(
     let ff_fields = reader.fast_fields();
     match cardinality {
         Cardinality::SingleValue => ff_fields
-            .u64_lenient(field_name)
+            .u64_lenient(field)
             .map(|field| (FastFieldAccessor::Single(field), field_type.value_type())),
         Cardinality::MultiValues => ff_fields
-            .u64s_lenient(field_name)
+            .u64s_lenient(field)
             .map(|field| (FastFieldAccessor::Multi(field), field_type.value_type())),
     }
 }

src/aggregation/agg_result.rs

@@ -30,7 +30,7 @@ impl AggregationResults {
         } else {
             // Validation is be done during request parsing, so we can't reach this state.
             Err(TantivyError::InternalError(format!(
-                "Can't find aggregation {:?} in sub-aggregations",
+                "Can't find aggregation {:?} in sub_aggregations",
                 name
             )))
         }
@@ -70,51 +70,27 @@ impl AggregationResult {
 pub enum MetricResult {
     /// Average metric result.
     Average(SingleMetricResult),
-    /// Count metric result.
-    Count(SingleMetricResult),
-    /// Max metric result.
-    Max(SingleMetricResult),
-    /// Min metric result.
-    Min(SingleMetricResult),
     /// Stats metric result.
     Stats(Stats),
-    /// Sum metric result.
-    Sum(SingleMetricResult),
 }
 
 impl MetricResult {
     fn get_value(&self, agg_property: &str) -> crate::Result<Option<f64>> {
         match self {
             MetricResult::Average(avg) => Ok(avg.value),
-            MetricResult::Count(count) => Ok(count.value),
-            MetricResult::Max(max) => Ok(max.value),
-            MetricResult::Min(min) => Ok(min.value),
             MetricResult::Stats(stats) => stats.get_value(agg_property),
-            MetricResult::Sum(sum) => Ok(sum.value),
         }
     }
 }
 impl From<IntermediateMetricResult> for MetricResult {
     fn from(metric: IntermediateMetricResult) -> Self {
         match metric {
-            IntermediateMetricResult::Average(intermediate_avg) => {
-                MetricResult::Average(intermediate_avg.finalize().into())
-            }
-            IntermediateMetricResult::Count(intermediate_count) => {
-                MetricResult::Count(intermediate_count.finalize().into())
-            }
-            IntermediateMetricResult::Max(intermediate_max) => {
-                MetricResult::Max(intermediate_max.finalize().into())
-            }
-            IntermediateMetricResult::Min(intermediate_min) => {
-                MetricResult::Min(intermediate_min.finalize().into())
+            IntermediateMetricResult::Average(avg_data) => {
+                MetricResult::Average(avg_data.finalize().into())
             }
             IntermediateMetricResult::Stats(intermediate_stats) => {
                 MetricResult::Stats(intermediate_stats.finalize())
             }
-            IntermediateMetricResult::Sum(intermediate_sum) => {
-                MetricResult::Sum(intermediate_sum.finalize().into())
-            }
         }
     }
 }
@@ -124,13 +100,13 @@ impl From<IntermediateMetricResult> for MetricResult {
 #[serde(untagged)]
 pub enum BucketResult {
     /// This is the range entry for a bucket, which contains a key, count, from, to, and optionally
-    /// sub-aggregations.
+    /// sub_aggregations.
     Range {
         /// The range buckets sorted by range.
         buckets: BucketEntries<RangeBucketEntry>,
     },
     /// This is the histogram entry for a bucket, which contains a key, count, and optionally
-    /// sub-aggregations.
+    /// sub_aggregations.
     Histogram {
         /// The buckets.
         ///
@@ -175,7 +151,7 @@ pub enum BucketEntries<T> {
 }
 
 /// This is the default entry for a bucket, which contains a key, count, and optionally
-/// sub-aggregations.
+/// sub_aggregations.
 ///
 /// # JSON Format
 /// ```json
@@ -225,7 +201,7 @@ impl GetDocCount for BucketEntry {
 }
 
 /// This is the range entry for a bucket, which contains a key, count, and optionally
-/// sub-aggregations.
+/// sub_aggregations.
 ///
 /// # JSON Format
 /// ```json
@@ -261,7 +237,7 @@ pub struct RangeBucketEntry {
     /// Number of documents in the bucket.
     pub doc_count: u64,
     #[serde(flatten)]
-    /// Sub-aggregations in this bucket.
+    /// sub-aggregations in this bucket.
     pub sub_aggregation: AggregationResults,
     /// The from range of the bucket. Equals `f64::MIN` when `None`.
     #[serde(skip_serializing_if = "Option::is_none")]

src/aggregation/bucket/histogram/histogram.rs

@@ -548,7 +548,9 @@ pub(crate) fn intermediate_histogram_buckets_to_final_buckets(
     };
 
     // If we have a date type on the histogram buckets, we add the `key_as_string` field as rfc339
-    let field = schema.get_field(&histogram_req.field)?;
+    let field = schema
+        .get_field(&histogram_req.field)
+        .ok_or_else(|| TantivyError::FieldNotFound(histogram_req.field.to_string()))?;
     if schema.get_field_entry(field).field_type().is_date() {
         for bucket in buckets.iter_mut() {
             if let crate::aggregation::Key::F64(val) = bucket.key {
@@ -1364,6 +1366,7 @@ mod tests {
                     "min": Value::Null,
                     "max": Value::Null,
                     "avg": Value::Null,
+                    "standard_deviation": Value::Null,
                 }
             })
         );

src/aggregation/intermediate_agg_result.rs

@@ -17,15 +17,13 @@ use super::bucket::{
     cut_off_buckets, get_agg_name_and_property, intermediate_histogram_buckets_to_final_buckets,
     GetDocCount, Order, OrderTarget, SegmentHistogramBucketEntry, TermsAggregation,
 };
-use super::metric::{
-    IntermediateAverage, IntermediateCount, IntermediateMax, IntermediateMin, IntermediateStats,
-    IntermediateSum,
-};
+use super::metric::{IntermediateAverage, IntermediateStats};
 use super::segment_agg_result::SegmentMetricResultCollector;
 use super::{format_date, Key, SerializedKey, VecWithNames};
 use crate::aggregation::agg_result::{AggregationResults, BucketEntries, BucketEntry};
 use crate::aggregation::bucket::TermsAggregationInternal;
 use crate::schema::Schema;
+use crate::TantivyError;
 
 /// Contains the intermediate aggregation result, which is optimized to be merged with other
 /// intermediate results.
@@ -206,43 +204,21 @@ pub enum IntermediateAggregationResult {
 /// Holds the intermediate data for metric results
 #[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
 pub enum IntermediateMetricResult {
-    /// Intermediate average result.
+    /// Average containing intermediate average data result
     Average(IntermediateAverage),
-    /// Intermediate count result.
-    Count(IntermediateCount),
-    /// Intermediate max result.
-    Max(IntermediateMax),
-    /// Intermediate min result.
-    Min(IntermediateMin),
-    /// Intermediate stats result.
+    /// AverageData variant
     Stats(IntermediateStats),
-    /// Intermediate sum result.
-    Sum(IntermediateSum),
 }
 
 impl From<SegmentMetricResultCollector> for IntermediateMetricResult {
     fn from(tree: SegmentMetricResultCollector) -> Self {
         match tree {
-            SegmentMetricResultCollector::Stats(collector) => match collector.collecting_for {
-                super::metric::SegmentStatsType::Average => IntermediateMetricResult::Average(
-                    IntermediateAverage::from_collector(collector),
-                ),
-                super::metric::SegmentStatsType::Count => {
-                    IntermediateMetricResult::Count(IntermediateCount::from_collector(collector))
-                }
-                super::metric::SegmentStatsType::Max => {
-                    IntermediateMetricResult::Max(IntermediateMax::from_collector(collector))
-                }
-                super::metric::SegmentStatsType::Min => {
-                    IntermediateMetricResult::Min(IntermediateMin::from_collector(collector))
-                }
-                super::metric::SegmentStatsType::Stats => {
-                    IntermediateMetricResult::Stats(collector.stats)
-                }
-                super::metric::SegmentStatsType::Sum => {
-                    IntermediateMetricResult::Sum(IntermediateSum::from_collector(collector))
-                }
-            },
+            SegmentMetricResultCollector::Average(collector) => {
+                IntermediateMetricResult::Average(IntermediateAverage::from_collector(collector))
+            }
+            SegmentMetricResultCollector::Stats(collector) => {
+                IntermediateMetricResult::Stats(collector.stats)
+            }
         }
     }
 }
@@ -253,36 +229,18 @@ impl IntermediateMetricResult {
             MetricAggregation::Average(_) => {
                 IntermediateMetricResult::Average(IntermediateAverage::default())
             }
-            MetricAggregation::Count(_) => {
-                IntermediateMetricResult::Count(IntermediateCount::default())
-            }
-            MetricAggregation::Max(_) => IntermediateMetricResult::Max(IntermediateMax::default()),
-            MetricAggregation::Min(_) => IntermediateMetricResult::Min(IntermediateMin::default()),
             MetricAggregation::Stats(_) => {
                 IntermediateMetricResult::Stats(IntermediateStats::default())
             }
-            MetricAggregation::Sum(_) => IntermediateMetricResult::Sum(IntermediateSum::default()),
         }
     }
     fn merge_fruits(&mut self, other: IntermediateMetricResult) {
         match (self, other) {
             (
-                IntermediateMetricResult::Average(avg_left),
-                IntermediateMetricResult::Average(avg_right),
+                IntermediateMetricResult::Average(avg_data_left),
+                IntermediateMetricResult::Average(avg_data_right),
             ) => {
-                avg_left.merge_fruits(avg_right);
-            }
-            (
-                IntermediateMetricResult::Count(count_left),
-                IntermediateMetricResult::Count(count_right),
-            ) => {
-                count_left.merge_fruits(count_right);
-            }
-            (IntermediateMetricResult::Max(max_left), IntermediateMetricResult::Max(max_right)) => {
-                max_left.merge_fruits(max_right);
-            }
-            (IntermediateMetricResult::Min(min_left), IntermediateMetricResult::Min(min_right)) => {
-                min_left.merge_fruits(min_right);
+                avg_data_left.merge_fruits(avg_data_right);
             }
             (
                 IntermediateMetricResult::Stats(stats_left),
@@ -290,9 +248,6 @@ impl IntermediateMetricResult {
             ) => {
                 stats_left.merge_fruits(stats_right);
             }
-            (IntermediateMetricResult::Sum(sum_left), IntermediateMetricResult::Sum(sum_right)) => {
-                sum_left.merge_fruits(sum_right);
-            }
             _ => {
                 panic!("incompatible fruit types in tree");
             }
@@ -657,7 +612,9 @@ impl IntermediateRangeBucketEntry {
 
         // If we have a date type on the histogram buckets, we add the `key_as_string` field as
         // rfc339
-        let field = schema.get_field(&range_req.field)?;
+        let field = schema
+            .get_field(&range_req.field)
+            .ok_or_else(|| TantivyError::FieldNotFound(range_req.field.to_string()))?;
         if schema.get_field_entry(field).field_type().is_date() {
             if let Some(val) = range_bucket_entry.to {
                 let key_as_string = format_date(val as i64)?;

src/aggregation/metric/average.rs

@@ -1,58 +1,114 @@
 use std::fmt::Debug;
 
+use fastfield_codecs::Column;
 use serde::{Deserialize, Serialize};
 
-use super::{IntermediateStats, SegmentStatsCollector};
+use crate::aggregation::f64_from_fastfield_u64;
+use crate::schema::Type;
+use crate::DocId;
 
+#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
 /// A single-value metric aggregation that computes the average of numeric values that are
 /// extracted from the aggregated documents.
+/// Supported field types are u64, i64, and f64.
 /// See [super::SingleMetricResult] for return value.
 ///
 /// # JSON Format
 /// ```json
 /// {
 ///     "avg": {
-///         "field": "score"
+///         "field": "score",
 ///     }
 /// }
 /// ```
-#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
 pub struct AverageAggregation {
-    /// The field name to compute the average on.
+    /// The field name to compute the stats on.
     pub field: String,
 }
-
 impl AverageAggregation {
-    /// Creates a new [`AverageAggregation`] instance from a field name.
+    /// Create new AverageAggregation from a field.
     pub fn from_field_name(field_name: String) -> Self {
-        Self { field: field_name }
+        AverageAggregation { field: field_name }
     }
-    /// Returns the field name the aggregation is computed on.
+    /// Return the field name.
     pub fn field_name(&self) -> &str {
         &self.field
     }
 }
 
-/// Intermediate result of the average aggregation that can be combined with other intermediate
-/// results.
+#[derive(Clone, PartialEq)]
+pub(crate) struct SegmentAverageCollector {
+    pub data: IntermediateAverage,
+    field_type: Type,
+}
+
+impl Debug for SegmentAverageCollector {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("AverageCollector")
+            .field("data", &self.data)
+            .finish()
+    }
+}
+
+impl SegmentAverageCollector {
+    pub fn from_req(field_type: Type) -> Self {
+        Self {
+            field_type,
+            data: Default::default(),
+        }
+    }
+    pub(crate) fn collect_block(&mut self, doc: &[DocId], field: &dyn Column<u64>) {
+        let mut iter = doc.chunks_exact(4);
+        for docs in iter.by_ref() {
+            let val1 = field.get_val(docs[0]);
+            let val2 = field.get_val(docs[1]);
+            let val3 = field.get_val(docs[2]);
+            let val4 = field.get_val(docs[3]);
+            let val1 = f64_from_fastfield_u64(val1, &self.field_type);
+            let val2 = f64_from_fastfield_u64(val2, &self.field_type);
+            let val3 = f64_from_fastfield_u64(val3, &self.field_type);
+            let val4 = f64_from_fastfield_u64(val4, &self.field_type);
+            self.data.collect(val1);
+            self.data.collect(val2);
+            self.data.collect(val3);
+            self.data.collect(val4);
+        }
+        for &doc in iter.remainder() {
+            let val = field.get_val(doc);
+            let val = f64_from_fastfield_u64(val, &self.field_type);
+            self.data.collect(val);
+        }
+    }
+}
+
+/// Contains mergeable version of average data.
 #[derive(Default, Clone, Debug, PartialEq, Serialize, Deserialize)]
 pub struct IntermediateAverage {
-    stats: IntermediateStats,
+    pub(crate) sum: f64,
+    pub(crate) doc_count: u64,
 }
 
 impl IntermediateAverage {
-    /// Creates a new [`IntermediateAverage`] instance from a [`SegmentStatsCollector`].
-    pub(crate) fn from_collector(collector: SegmentStatsCollector) -> Self {
-        Self {
-            stats: collector.stats,
+    pub(crate) fn from_collector(collector: SegmentAverageCollector) -> Self {
+        collector.data
+    }
+
+    /// Merge average data into this instance.
+    pub fn merge_fruits(&mut self, other: IntermediateAverage) {
+        self.sum += other.sum;
+        self.doc_count += other.doc_count;
+    }
+    /// compute final result
+    pub fn finalize(&self) -> Option<f64> {
+        if self.doc_count == 0 {
+            None
+        } else {
+            Some(self.sum / self.doc_count as f64)
         }
     }
-    /// Merges the other intermediate result into self.
-    pub fn merge_fruits(&mut self, other: IntermediateAverage) {
-        self.stats.merge_fruits(other.stats);
-    }
-    /// Computes the final average value.
-    pub fn finalize(&self) -> Option<f64> {
-        self.stats.finalize().avg
+    #[inline]
+    fn collect(&mut self, val: f64) {
+        self.doc_count += 1;
+        self.sum += val;
     }
 }

src/aggregation/metric/count.rs

@@ -1,58 +0,0 @@
-use std::fmt::Debug;
-
-use serde::{Deserialize, Serialize};
-
-use super::{IntermediateStats, SegmentStatsCollector};
-
-/// A single-value metric aggregation that counts the number of values that are
-/// extracted from the aggregated documents.
-/// See [super::SingleMetricResult] for return value.
-///
-/// # JSON Format
-/// ```json
-/// {
-///     "value_count": {
-///         "field": "score"
-///     }
-/// }
-/// ```
-#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
-pub struct CountAggregation {
-    /// The field name to compute the minimum on.
-    pub field: String,
-}
-
-impl CountAggregation {
-    /// Creates a new [`CountAggregation`] instance from a field name.
-    pub fn from_field_name(field_name: String) -> Self {
-        Self { field: field_name }
-    }
-    /// Returns the field name the aggregation is computed on.
-    pub fn field_name(&self) -> &str {
-        &self.field
-    }
-}
-
-/// Intermediate result of the count aggregation that can be combined with other intermediate
-/// results.
-#[derive(Default, Clone, Debug, PartialEq, Serialize, Deserialize)]
-pub struct IntermediateCount {
-    stats: IntermediateStats,
-}
-
-impl IntermediateCount {
-    /// Creates a new [`IntermediateCount`] instance from a [`SegmentStatsCollector`].
-    pub(crate) fn from_collector(collector: SegmentStatsCollector) -> Self {
-        Self {
-            stats: collector.stats,
-        }
-    }
-    /// Merges the other intermediate result into self.
-    pub fn merge_fruits(&mut self, other: IntermediateCount) {
-        self.stats.merge_fruits(other.stats);
-    }
-    /// Computes the final minimum value.
-    pub fn finalize(&self) -> Option<f64> {
-        Some(self.stats.finalize().count as f64)
-    }
-}

src/aggregation/metric/max.rs

@@ -1,58 +0,0 @@
-use std::fmt::Debug;
-
-use serde::{Deserialize, Serialize};
-
-use super::{IntermediateStats, SegmentStatsCollector};
-
-/// A single-value metric aggregation that computes the maximum of numeric values that are
-/// extracted from the aggregated documents.
-/// See [super::SingleMetricResult] for return value.
-///
-/// # JSON Format
-/// ```json
-/// {
-///     "max": {
-///         "field": "score"
-///     }
-/// }
-/// ```
-#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
-pub struct MaxAggregation {
-    /// The field name to compute the maximum on.
-    pub field: String,
-}
-
-impl MaxAggregation {
-    /// Creates a new [`MaxAggregation`] instance from a field name.
-    pub fn from_field_name(field_name: String) -> Self {
-        Self { field: field_name }
-    }
-    /// Returns the field name the aggregation is computed on.
-    pub fn field_name(&self) -> &str {
-        &self.field
-    }
-}
-
-/// Intermediate result of the maximum aggregation that can be combined with other intermediate
-/// results.
-#[derive(Default, Clone, Debug, PartialEq, Serialize, Deserialize)]
-pub struct IntermediateMax {
-    stats: IntermediateStats,
-}
-
-impl IntermediateMax {
-    /// Creates a new [`IntermediateMax`] instance from a [`SegmentStatsCollector`].
-    pub(crate) fn from_collector(collector: SegmentStatsCollector) -> Self {
-        Self {
-            stats: collector.stats,
-        }
-    }
-    /// Merges the other intermediate result into self.
-    pub fn merge_fruits(&mut self, other: IntermediateMax) {
-        self.stats.merge_fruits(other.stats);
-    }
-    /// Computes the final maximum value.
-    pub fn finalize(&self) -> Option<f64> {
-        self.stats.finalize().max
-    }
-}

src/aggregation/metric/min.rs

@@ -1,58 +0,0 @@
-use std::fmt::Debug;
-
-use serde::{Deserialize, Serialize};
-
-use super::{IntermediateStats, SegmentStatsCollector};
-
-/// A single-value metric aggregation that computes the minimum of numeric values that are
-/// extracted from the aggregated documents.
-/// See [super::SingleMetricResult] for return value.
-///
-/// # JSON Format
-/// ```json
-/// {
-///     "min": {
-///         "field": "score"
-///     }
-/// }
-/// ```
-#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
-pub struct MinAggregation {
-    /// The field name to compute the minimum on.
-    pub field: String,
-}
-
-impl MinAggregation {
-    /// Creates a new [`MinAggregation`] instance from a field name.
-    pub fn from_field_name(field_name: String) -> Self {
-        Self { field: field_name }
-    }
-    /// Returns the field name the aggregation is computed on.
-    pub fn field_name(&self) -> &str {
-        &self.field
-    }
-}
-
-/// Intermediate result of the minimum aggregation that can be combined with other intermediate
-/// results.
-#[derive(Default, Clone, Debug, PartialEq, Serialize, Deserialize)]
-pub struct IntermediateMin {
-    stats: IntermediateStats,
-}
-
-impl IntermediateMin {
-    /// Creates a new [`IntermediateMin`] instance from a [`SegmentStatsCollector`].
-    pub(crate) fn from_collector(collector: SegmentStatsCollector) -> Self {
-        Self {
-            stats: collector.stats,
-        }
-    }
-    /// Merges the other intermediate result into self.
-    pub fn merge_fruits(&mut self, other: IntermediateMin) {
-        self.stats.merge_fruits(other.stats);
-    }
-    /// Computes the final minimum value.
-    pub fn finalize(&self) -> Option<f64> {
-        self.stats.finalize().min
-    }
-}

src/aggregation/metric/mod.rs

@@ -3,18 +3,10 @@
 //! The aggregations in this family compute metrics, see [super::agg_req::MetricAggregation] for
 //! details.
 mod average;
-mod count;
-mod max;
-mod min;
 mod stats;
-mod sum;
 pub use average::*;
-pub use count::*;
-pub use max::*;
-pub use min::*;
 use serde::{Deserialize, Serialize};
 pub use stats::*;
-pub use sum::*;
 
 /// Single-metric aggregations use this common result structure.
 ///
@@ -36,61 +28,3 @@ impl From<Option<f64>> for SingleMetricResult {
         Self { value }
     }
 }
-
-#[cfg(test)]
-mod tests {
-    use crate::aggregation::agg_req::Aggregations;
-    use crate::aggregation::agg_result::AggregationResults;
-    use crate::aggregation::AggregationCollector;
-    use crate::query::AllQuery;
-    use crate::schema::{Cardinality, 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 = 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();
-
-        for i in 0..3 {
-            index_writer
-                .add_document(doc!(
-                    field => i as f64,
-                ))
-                .unwrap();
-        }
-        index_writer.commit().unwrap();
-
-        for i in 3..6 {
-            index_writer
-                .add_document(doc!(
-                    field => i as f64,
-                ))
-                .unwrap();
-        }
-        index_writer.commit().unwrap();
-
-        let aggregations_json = r#"{
-            "price_avg": { "avg": { "field": "price" } },
-            "price_count": { "value_count": { "field": "price" } },
-            "price_max": { "max": { "field": "price" } },
-            "price_min": { "min": { "field": "price" } },
-            "price_stats": { "stats": { "field": "price" } },
-            "price_sum": { "sum": { "field": "price" } }
-        }"#;
-        let aggregations: Aggregations = serde_json::from_str(aggregations_json).unwrap();
-        let collector = AggregationCollector::from_aggs(aggregations, None, index.schema());
-        let reader = index.reader().unwrap();
-        let searcher = reader.searcher();
-        let aggregations_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
-        let aggregations_res_json = serde_json::to_value(aggregations_res).unwrap();
-
-        assert_eq!(aggregations_res_json["price_avg"]["value"], 2.5);
-        assert_eq!(aggregations_res_json["price_count"]["value"], 6.0);
-        assert_eq!(aggregations_res_json["price_max"]["value"], 5.0);
-        assert_eq!(aggregations_res_json["price_min"]["value"], 0.0);
-        assert_eq!(aggregations_res_json["price_sum"]["value"], 15.0);
-    }
-}

src/aggregation/metric/stats.rs

@@ -5,15 +5,16 @@ use crate::aggregation::f64_from_fastfield_u64;
 use crate::schema::Type;
 use crate::{DocId, TantivyError};
 
-/// A multi-value metric aggregation that computes a collection of statistics on numeric values that
-/// are extracted from the aggregated documents.
+/// A multi-value metric aggregation that computes stats of numeric values that are
+/// extracted from the aggregated documents.
+/// Supported field types are `u64`, `i64`, and `f64`.
 /// See [`Stats`] for returned statistics.
 ///
 /// # JSON Format
 /// ```json
 /// {
 ///     "stats": {
-///         "field": "score"
+///         "field": "score",
 ///     }
 ///  }
 /// ```
@@ -25,11 +26,11 @@ pub struct StatsAggregation {
 }
 
 impl StatsAggregation {
-    /// Creates a new [`StatsAggregation`] instance from a field name.
+    /// Create new StatsAggregation from a field.
     pub fn from_field_name(field_name: String) -> Self {
         StatsAggregation { field: field_name }
     }
-    /// Returns the field name the aggregation is computed on.
+    /// Return the field name.
     pub fn field_name(&self) -> &str {
         &self.field
     }
@@ -39,14 +40,16 @@ impl StatsAggregation {
 #[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
 pub struct Stats {
     /// The number of documents.
-    pub count: u64,
+    pub count: usize,
     /// The sum of the fast field values.
     pub sum: f64,
+    /// The standard deviation of the fast field values. `None` for count == 0.
+    pub standard_deviation: Option<f64>,
     /// The min value of the fast field values.
     pub min: Option<f64>,
     /// The max value of the fast field values.
     pub max: Option<f64>,
-    /// The average of the fast field values. `None` if count equals zero.
+    /// The average of the values. `None` for count == 0.
     pub avg: Option<f64>,
 }
 
@@ -55,36 +58,33 @@ impl Stats {
         match agg_property {
             "count" => Ok(Some(self.count as f64)),
             "sum" => Ok(Some(self.sum)),
+            "standard_deviation" => Ok(self.standard_deviation),
             "min" => Ok(self.min),
             "max" => Ok(self.max),
             "avg" => Ok(self.avg),
             _ => Err(TantivyError::InvalidArgument(format!(
-                "Unknown property {} on stats metric aggregation",
+                "unknown property {} on stats metric aggregation",
                 agg_property
             ))),
         }
     }
 }
 
-/// Intermediate result of the stats aggregation that can be combined with other intermediate
-/// results.
+/// `IntermediateStats` contains the mergeable version for stats.
 #[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
 pub struct IntermediateStats {
-    /// The number of extracted values.
-    count: u64,
-    /// The sum of the extracted values.
+    count: usize,
     sum: f64,
-    /// The min value.
+    squared_sum: f64,
     min: f64,
-    /// The max value.
     max: f64,
 }
-
 impl Default for IntermediateStats {
     fn default() -> Self {
         Self {
             count: 0,
             sum: 0.0,
+            squared_sum: 0.0,
             min: f64::MAX,
             max: f64::MIN,
         }
@@ -92,15 +92,33 @@ impl Default for IntermediateStats {
 }
 
 impl IntermediateStats {
-    /// Merges the other stats intermediate result into self.
+    pub(crate) fn avg(&self) -> Option<f64> {
+        if self.count == 0 {
+            None
+        } else {
+            Some(self.sum / (self.count as f64))
+        }
+    }
+
+    fn square_mean(&self) -> f64 {
+        self.squared_sum / (self.count as f64)
+    }
+
+    pub(crate) fn standard_deviation(&self) -> Option<f64> {
+        self.avg()
+            .map(|average| (self.square_mean() - average * average).sqrt())
+    }
+
+    /// Merge data from other stats into this instance.
     pub fn merge_fruits(&mut self, other: IntermediateStats) {
         self.count += other.count;
         self.sum += other.sum;
+        self.squared_sum += other.squared_sum;
         self.min = self.min.min(other.min);
         self.max = self.max.max(other.max);
     }
 
-    /// Computes the final stats value.
+    /// compute final resultimprove_docs
     pub fn finalize(&self) -> Stats {
         let min = if self.count == 0 {
             None
@@ -112,17 +130,13 @@ impl IntermediateStats {
         } else {
             Some(self.max)
         };
-        let avg = if self.count == 0 {
-            None
-        } else {
-            Some(self.sum / (self.count as f64))
-        };
         Stats {
             count: self.count,
             sum: self.sum,
+            standard_deviation: self.standard_deviation(),
             min,
             max,
-            avg,
+            avg: self.avg(),
         }
     }
 
@@ -130,33 +144,22 @@ impl IntermediateStats {
     fn collect(&mut self, value: f64) {
         self.count += 1;
         self.sum += value;
+        self.squared_sum += value * value;
         self.min = self.min.min(value);
         self.max = self.max.max(value);
     }
 }
 
-#[derive(Clone, Debug, PartialEq)]
-pub(crate) enum SegmentStatsType {
-    Average,
-    Count,
-    Max,
-    Min,
-    Stats,
-    Sum,
-}
-
 #[derive(Clone, Debug, PartialEq)]
 pub(crate) struct SegmentStatsCollector {
-    field_type: Type,
-    pub(crate) collecting_for: SegmentStatsType,
     pub(crate) stats: IntermediateStats,
+    field_type: Type,
 }
 
 impl SegmentStatsCollector {
-    pub fn from_req(field_type: Type, collecting_for: SegmentStatsType) -> Self {
+    pub fn from_req(field_type: Type) -> Self {
         Self {
             field_type,
-            collecting_for,
             stats: IntermediateStats::default(),
         }
     }
@@ -233,6 +236,7 @@ mod tests {
                 "count": 0,
                 "max": Value::Null,
                 "min": Value::Null,
+                "standard_deviation": Value::Null,
                 "sum": 0.0
             })
         );
@@ -309,6 +313,7 @@ mod tests {
                 "count": 7,
                 "max": 44.0,
                 "min": 1.0,
+                "standard_deviation": 13.65313748796613,
                 "sum": 85.0
             })
         );
@@ -320,6 +325,7 @@ mod tests {
                 "count": 7,
                 "max": 44.0,
                 "min": 1.0,
+                "standard_deviation": 13.65313748796613,
                 "sum": 85.0
             })
         );
@@ -331,6 +337,7 @@ mod tests {
                 "count": 7,
                 "max": 44.5,
                 "min": 1.0,
+                "standard_deviation": 13.819905785437443,
                 "sum": 85.5
             })
         );
@@ -342,6 +349,7 @@ mod tests {
                 "count": 3,
                 "max": 14.0,
                 "min": 7.0,
+                "standard_deviation": 2.867441755680877,
                 "sum": 32.0
             })
         );
@@ -353,6 +361,7 @@ mod tests {
                 "count": 0,
                 "max": serde_json::Value::Null,
                 "min": serde_json::Value::Null,
+                "standard_deviation": serde_json::Value::Null,
                 "sum": 0.0,
             })
         );

src/aggregation/metric/sum.rs

@@ -1,58 +0,0 @@
-use std::fmt::Debug;
-
-use serde::{Deserialize, Serialize};
-
-use super::{IntermediateStats, SegmentStatsCollector};
-
-/// A single-value metric aggregation that sums up numeric values that are
-/// extracted from the aggregated documents.
-/// See [super::SingleMetricResult] for return value.
-///
-/// # JSON Format
-/// ```json
-/// {
-///     "sum": {
-///         "field": "score"
-///     }
-/// }
-/// ```
-#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
-pub struct SumAggregation {
-    /// The field name to compute the minimum on.
-    pub field: String,
-}
-
-impl SumAggregation {
-    /// Creates a new [`SumAggregation`] instance from a field name.
-    pub fn from_field_name(field_name: String) -> Self {
-        Self { field: field_name }
-    }
-    /// Returns the field name the aggregation is computed on.
-    pub fn field_name(&self) -> &str {
-        &self.field
-    }
-}
-
-/// Intermediate result of the minimum aggregation that can be combined with other intermediate
-/// results.
-#[derive(Default, Clone, Debug, PartialEq, Serialize, Deserialize)]
-pub struct IntermediateSum {
-    stats: IntermediateStats,
-}
-
-impl IntermediateSum {
-    /// Creates a new [`IntermediateSum`] instance from a [`SegmentStatsCollector`].
-    pub(crate) fn from_collector(collector: SegmentStatsCollector) -> Self {
-        Self {
-            stats: collector.stats,
-        }
-    }
-    /// Merges the other intermediate result into self.
-    pub fn merge_fruits(&mut self, other: IntermediateSum) {
-        self.stats.merge_fruits(other.stats);
-    }
-    /// Computes the final minimum value.
-    pub fn finalize(&self) -> Option<f64> {
-        Some(self.stats.finalize().sum)
-    }
-}

src/aggregation/mod.rs

@@ -1,5 +1,6 @@
 //! # Aggregations
 //!
+//!
 //! An aggregation summarizes your data as statistics on buckets or metrics.
 //!
 //! Aggregations can provide answer to questions like:
@@ -40,10 +41,6 @@
 //! - [Metric](metric)
 //!     - [Average](metric::AverageAggregation)
 //!     - [Stats](metric::StatsAggregation)
-//!     - [Min](metric::MinAggregation)
-//!     - [Max](metric::MaxAggregation)
-//!     - [Sum](metric::SumAggregation)
-//!     - [Count](metric::CountAggregation)
 //!
 //! # Example
 //! Compute the average metric, by building [`agg_req::Aggregations`], which is built from an
@@ -78,7 +75,7 @@
 //! }
 //! ```
 //! # Example JSON
-//! Requests are compatible with the elasticsearch JSON request format.
+//! Requests are compatible with the elasticsearch json request format.
 //!
 //! ```
 //! use tantivy::aggregation::agg_req::Aggregations;
@@ -219,8 +216,8 @@ impl<T: Clone> VecWithNames<T> {
     fn from_entries(mut entries: Vec<(String, T)>) -> Self {
         // Sort to ensure order of elements match across multiple instances
         entries.sort_by(|left, right| left.0.cmp(&right.0));
-        let mut data = Vec::with_capacity(entries.len());
-        let mut data_names = Vec::with_capacity(entries.len());
+        let mut data = vec![];
+        let mut data_names = vec![];
         for entry in entries {
             data_names.push(entry.0);
             data.push(entry.1);

src/aggregation/segment_agg_result.rs

@@ -15,8 +15,7 @@ use super::bucket::{SegmentHistogramCollector, SegmentRangeCollector, SegmentTer
 use super::collector::MAX_BUCKET_COUNT;
 use super::intermediate_agg_result::{IntermediateAggregationResults, IntermediateBucketResult};
 use super::metric::{
-    AverageAggregation, CountAggregation, MaxAggregation, MinAggregation, SegmentStatsCollector,
-    SegmentStatsType, StatsAggregation, SumAggregation,
+    AverageAggregation, SegmentAverageCollector, SegmentStatsCollector, StatsAggregation,
 };
 use super::VecWithNames;
 use crate::aggregation::agg_req::BucketAggregationType;
@@ -164,46 +163,30 @@ impl SegmentAggregationResultsCollector {
 
 #[derive(Clone, Debug, PartialEq)]
 pub(crate) enum SegmentMetricResultCollector {
+    Average(SegmentAverageCollector),
     Stats(SegmentStatsCollector),
 }
 
 impl SegmentMetricResultCollector {
     pub fn from_req_and_validate(req: &MetricAggregationWithAccessor) -> crate::Result<Self> {
         match &req.metric {
-            MetricAggregation::Average(AverageAggregation { .. }) => {
-                Ok(SegmentMetricResultCollector::Stats(
-                    SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Average),
+            MetricAggregation::Average(AverageAggregation { field: _ }) => {
+                Ok(SegmentMetricResultCollector::Average(
+                    SegmentAverageCollector::from_req(req.field_type),
                 ))
             }
-            MetricAggregation::Count(CountAggregation { .. }) => {
+            MetricAggregation::Stats(StatsAggregation { field: _ }) => {
                 Ok(SegmentMetricResultCollector::Stats(
-                    SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Count),
-                ))
-            }
-            MetricAggregation::Max(MaxAggregation { .. }) => {
-                Ok(SegmentMetricResultCollector::Stats(
-                    SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Max),
-                ))
-            }
-            MetricAggregation::Min(MinAggregation { .. }) => {
-                Ok(SegmentMetricResultCollector::Stats(
-                    SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Min),
-                ))
-            }
-            MetricAggregation::Stats(StatsAggregation { .. }) => {
-                Ok(SegmentMetricResultCollector::Stats(
-                    SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Stats),
-                ))
-            }
-            MetricAggregation::Sum(SumAggregation { .. }) => {
-                Ok(SegmentMetricResultCollector::Stats(
-                    SegmentStatsCollector::from_req(req.field_type, SegmentStatsType::Sum),
+                    SegmentStatsCollector::from_req(req.field_type),
                 ))
             }
         }
     }
     pub(crate) fn collect_block(&mut self, doc: &[DocId], metric: &MetricAggregationWithAccessor) {
         match self {
+            SegmentMetricResultCollector::Average(avg_collector) => {
+                avg_collector.collect_block(doc, &*metric.accessor);
+            }
             SegmentMetricResultCollector::Stats(stats_collector) => {
                 stats_collector.collect_block(doc, &*metric.accessor);
             }

src/collector/filter_collector_wrapper.rs

@@ -130,7 +130,7 @@ where
 
         let fast_field_reader = segment_reader
             .fast_fields()
-            .typed_fast_field_reader(schema.get_field_name(self.field))?;
+            .typed_fast_field_reader(self.field)?;
 
         let segment_collector = self
             .collector

src/collector/histogram_collector.rs

@@ -5,7 +5,7 @@ use fastfield_codecs::Column;
 
 use crate::collector::{Collector, SegmentCollector};
 use crate::fastfield::FastValue;
-use crate::schema::Type;
+use crate::schema::{Field, Type};
 use crate::{DocId, Score};
 
 /// Histogram builds an histogram of the values of a fastfield for the
@@ -28,7 +28,7 @@ pub struct HistogramCollector {
     min_value: u64,
     num_buckets: usize,
     divider: DividerU64,
-    field: String,
+    field: Field,
 }
 
 impl HistogramCollector {
@@ -46,7 +46,7 @@ impl HistogramCollector {
     /// # Disclaimer
     /// This function panics if the field given is of type f64.
     pub fn new<TFastValue: FastValue>(
-        field: String,
+        field: Field,
         min_value: TFastValue,
         bucket_width: u64,
         num_buckets: usize,
@@ -112,7 +112,7 @@ 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 ff_reader = segment.fast_fields().u64_lenient(self.field)?;
         Ok(SegmentHistogramCollector {
             histogram_computer: HistogramComputer {
                 counts: vec![0; self.num_buckets],
@@ -211,13 +211,13 @@ mod tests {
     #[test]
     fn test_no_segments() -> crate::Result<()> {
         let mut schema_builder = Schema::builder();
-        schema_builder.add_u64_field("val_field", FAST);
+        let val_field = schema_builder.add_u64_field("val_field", FAST);
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
         let reader = index.reader()?;
         let searcher = reader.searcher();
         let all_query = AllQuery;
-        let histogram_collector = HistogramCollector::new("val_field".to_string(), 0u64, 2, 5);
+        let histogram_collector = HistogramCollector::new(val_field, 0u64, 2, 5);
         let histogram = searcher.search(&all_query, &histogram_collector)?;
         assert_eq!(histogram, vec![0; 5]);
         Ok(())
@@ -238,8 +238,7 @@ mod tests {
         let reader = index.reader()?;
         let searcher = reader.searcher();
         let all_query = AllQuery;
-        let histogram_collector =
-            HistogramCollector::new("val_field".to_string(), -20i64, 10u64, 4);
+        let histogram_collector = HistogramCollector::new(val_field, -20i64, 10u64, 4);
         let histogram = searcher.search(&all_query, &histogram_collector)?;
         assert_eq!(histogram, vec![1, 1, 0, 1]);
         Ok(())
@@ -263,8 +262,7 @@ mod tests {
         let reader = index.reader()?;
         let searcher = reader.searcher();
         let all_query = AllQuery;
-        let histogram_collector =
-            HistogramCollector::new("val_field".to_string(), -20i64, 10u64, 4);
+        let histogram_collector = HistogramCollector::new(val_field, -20i64, 10u64, 4);
         let histogram = searcher.search(&all_query, &histogram_collector)?;
         assert_eq!(histogram, vec![1, 1, 0, 1]);
         Ok(())
@@ -287,7 +285,7 @@ mod tests {
         let searcher = reader.searcher();
         let all_query = AllQuery;
         let week_histogram_collector = HistogramCollector::new(
-            "date_field".to_string(),
+            date_field,
             DateTime::from_primitive(
                 Date::from_calendar_date(1980, Month::January, 1)?.with_hms(0, 0, 0)?,
             ),

src/collector/tests.rs

@@ -155,7 +155,7 @@ impl SegmentCollector for TestSegmentCollector {
 ///
 /// This collector is mainly useful for tests.
 pub struct FastFieldTestCollector {
-    field: String,
+    field: Field,
 }
 
 pub struct FastFieldSegmentCollector {
@@ -164,7 +164,7 @@ pub struct FastFieldSegmentCollector {
 }
 
 impl FastFieldTestCollector {
-    pub fn for_field(field: String) -> FastFieldTestCollector {
+    pub fn for_field(field: Field) -> FastFieldTestCollector {
         FastFieldTestCollector { field }
     }
 }
@@ -180,7 +180,7 @@ impl Collector for FastFieldTestCollector {
     ) -> crate::Result<FastFieldSegmentCollector> {
         let reader = segment_reader
             .fast_fields()
-            .u64(&self.field)
+            .u64(self.field)
             .expect("Requested field is not a fast field.");
         Ok(FastFieldSegmentCollector {
             vals: Vec::new(),
@@ -238,9 +238,7 @@ impl Collector for BytesFastFieldTestCollector {
         _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))?;
+        let reader = segment_reader.fast_fields().bytes(self.field)?;
         Ok(BytesFastFieldSegmentCollector {
             vals: Vec::new(),
             reader,

src/collector/top_score_collector.rs

@@ -156,7 +156,7 @@ impl CustomScorer<u64> for ScorerByField {
         // 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))?;
+            .typed_fast_field_reader(self.field)?;
         Ok(ScorerByFastFieldReader { ff_reader })
     }
 }
@@ -454,7 +454,7 @@ impl TopDocs {
     ///             // In our case, we will get a reader for the popularity
     ///             // fast field.
     ///             let popularity_reader =
-    ///                 segment_reader.fast_fields().u64("popularity").unwrap();
+    ///                 segment_reader.fast_fields().u64(popularity).unwrap();
     ///
     ///             // We can now define our actual scoring function
     ///             move |doc: DocId, original_score: Score| {
@@ -561,9 +561,9 @@ impl TopDocs {
     ///             // Note that this is implemented by using a `(u64, u64)`
     ///             // as a score.
     ///             let popularity_reader =
-    ///                 segment_reader.fast_fields().u64("popularity").unwrap();
+    ///                 segment_reader.fast_fields().u64(popularity).unwrap();
     ///             let boosted_reader =
-    ///                 segment_reader.fast_fields().u64("boosted").unwrap();
+    ///                 segment_reader.fast_fields().u64(boosted).unwrap();
     ///
     ///             // We can now define our actual scoring function
     ///             move |doc: DocId| {

src/core/index.rs

@@ -231,7 +231,7 @@ impl IndexBuilder {
     fn validate(&self) -> crate::Result<()> {
         if let Some(schema) = self.schema.as_ref() {
             if let Some(sort_by_field) = self.index_settings.sort_by_field.as_ref() {
-                let schema_field = schema.get_field(&sort_by_field.field).map_err(|_| {
+                let schema_field = schema.get_field(&sort_by_field.field).ok_or_else(|| {
                     TantivyError::InvalidArgument(format!(
                         "Field to sort index {} not found in schema",
                         sort_by_field.field

src/core/segment_reader.rs

@@ -95,8 +95,7 @@ impl SegmentReader {
 
         match field_entry.field_type() {
             FieldType::Facet(_) => {
-                let term_ords_reader =
-                    self.fast_fields().u64s(self.schema.get_field_name(field))?;
+                let term_ords_reader = self.fast_fields().u64s(field)?;
                 let termdict = self
                     .termdict_composite
                     .open_read(field)

src/fastfield/bytes/mod.rs

@@ -25,7 +25,7 @@ mod tests {
         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();
+        let bytes_reader = segment_reader.fast_fields().bytes(bytes_field).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]);
@@ -109,7 +109,8 @@ mod tests {
         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();
+        let field = searcher.schema().get_field("string_bytes").unwrap();
+        let fast_field_reader = fast_fields.bytes(field).unwrap();
         assert_eq!(fast_field_reader.get_bytes(0u32), b"tantivy");
         Ok(())
     }

src/fastfield/mod.rs

@@ -226,7 +226,7 @@ mod tests {
             serializer.close().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 27);
+        assert_eq!(file.len(), 34);
         let composite_file = CompositeFile::open(&file)?;
         let fast_field_bytes = composite_file.open_read(*FIELD).unwrap().read_bytes()?;
         let fast_field_reader = open::<u64>(fast_field_bytes)?;
@@ -275,7 +275,7 @@ mod tests {
             serializer.close()?;
         }
         let file = directory.open_read(path)?;
-        assert_eq!(file.len(), 55);
+        assert_eq!(file.len(), 62);
         {
             let fast_fields_composite = CompositeFile::open(&file)?;
             let data = fast_fields_composite
@@ -316,7 +316,7 @@ mod tests {
             serializer.close().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 28);
+        assert_eq!(file.len(), 35);
         {
             let fast_fields_composite = CompositeFile::open(&file).unwrap();
             let data = fast_fields_composite
@@ -355,7 +355,7 @@ mod tests {
             serializer.close().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 80042);
+        assert_eq!(file.len(), 80049);
         {
             let fast_fields_composite = CompositeFile::open(&file)?;
             let data = fast_fields_composite
@@ -397,7 +397,7 @@ mod tests {
             serializer.close().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 42_usize);
+        assert_eq!(file.len(), 49_usize);
 
         {
             let fast_fields_composite = CompositeFile::open(&file)?;
@@ -583,7 +583,7 @@ mod tests {
             assert_eq!(searcher.segment_readers().len(), 1);
             let segment_reader = searcher.segment_reader(0);
             let fast_fields = segment_reader.fast_fields();
-            let text_fast_field = fast_fields.u64s("text").unwrap();
+            let text_fast_field = fast_fields.u64s(text_field).unwrap();
 
             assert_eq!(
                 get_vals_for_docs(&text_fast_field, 0..5),
@@ -622,7 +622,7 @@ mod tests {
             assert_eq!(searcher.segment_readers().len(), 2);
             let segment_reader = searcher.segment_reader(1);
             let fast_fields = segment_reader.fast_fields();
-            let text_fast_field = fast_fields.u64s("text").unwrap();
+            let text_fast_field = fast_fields.u64s(text_field).unwrap();
 
             assert_eq!(get_vals_for_docs(&text_fast_field, 0..3), vec![0, 1, 0]);
         }
@@ -638,7 +638,7 @@ mod tests {
         let searcher = reader.searcher();
         let segment_reader = searcher.segment_reader(0);
         let fast_fields = segment_reader.fast_fields();
-        let text_fast_field = fast_fields.u64s("text").unwrap();
+        let text_fast_field = fast_fields.u64s(text_field).unwrap();
 
         assert_eq!(
             get_vals_for_docs(&text_fast_field, 0..8),
@@ -681,7 +681,7 @@ mod tests {
             assert_eq!(searcher.segment_readers().len(), 1);
             let segment_reader = searcher.segment_reader(0);
             let fast_fields = segment_reader.fast_fields();
-            let text_fast_field = fast_fields.u64s("text").unwrap();
+            let text_fast_field = fast_fields.u64s(text_field).unwrap();
 
             assert_eq!(get_vals_for_docs(&text_fast_field, 0..6), vec![1, 0, 0, 2]);
 
@@ -712,7 +712,7 @@ mod tests {
             assert_eq!(searcher.segment_readers().len(), 2);
             let segment_reader = searcher.segment_reader(1);
             let fast_fields = segment_reader.fast_fields();
-            let text_fast_field = fast_fields.u64s("text").unwrap();
+            let text_fast_field = fast_fields.u64s(text_field).unwrap();
 
             assert_eq!(get_vals_for_docs(&text_fast_field, 0..2), vec![0, 1]);
         }
@@ -728,7 +728,7 @@ mod tests {
         let searcher = reader.searcher();
         let segment_reader = searcher.segment_reader(0);
         let fast_fields = segment_reader.fast_fields();
-        let text_fast_field = fast_fields.u64s("text").unwrap();
+        let text_fast_field = fast_fields.u64s(text_field).unwrap();
 
         assert_eq!(
             get_vals_for_docs(&text_fast_field, 0..9),
@@ -773,8 +773,8 @@ mod tests {
         assert_eq!(searcher.segment_readers().len(), 1);
         let segment_reader = searcher.segment_reader(0);
         let fast_fields = segment_reader.fast_fields();
-        let date_fast_field = fast_fields.date("date").unwrap();
-        let dates_fast_field = fast_fields.dates("multi_date").unwrap();
+        let date_fast_field = fast_fields.date(date_field).unwrap();
+        let dates_fast_field = fast_fields.dates(multi_date_field).unwrap();
         let mut dates = vec![];
         {
             assert_eq!(date_fast_field.get_val(0).into_timestamp_micros(), 1i64);
@@ -836,7 +836,7 @@ mod tests {
             serializer.close().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 26);
+        assert_eq!(file.len(), 33);
         let composite_file = CompositeFile::open(&file)?;
         let data = composite_file.open_read(field).unwrap().read_bytes()?;
         let fast_field_reader = open::<bool>(data)?;
@@ -874,7 +874,7 @@ mod tests {
             serializer.close().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 38);
+        assert_eq!(file.len(), 45);
         let composite_file = CompositeFile::open(&file)?;
         let data = composite_file.open_read(field).unwrap().read_bytes()?;
         let fast_field_reader = open::<bool>(data)?;
@@ -906,7 +906,7 @@ mod tests {
         }
         let file = directory.open_read(path).unwrap();
         let composite_file = CompositeFile::open(&file)?;
-        assert_eq!(file.len(), 25);
+        assert_eq!(file.len(), 32);
         let data = composite_file.open_read(field).unwrap().read_bytes()?;
         let fast_field_reader = open::<bool>(data)?;
         assert_eq!(fast_field_reader.get_val(0), false);
@@ -940,10 +940,10 @@ mod tests {
     pub fn test_gcd_date() -> crate::Result<()> {
         let size_prec_sec =
             test_gcd_date_with_codec(FastFieldCodecType::Bitpacked, DatePrecision::Seconds)?;
-        assert_eq!(size_prec_sec, 5 + 4 + 21 + (1_000 * 13) / 8); // 13 bits per val = ceil(log_2(number of seconds in 2hours);
+        assert_eq!(size_prec_sec, 5 + 4 + 28 + (1_000 * 13) / 8); // 13 bits per val = ceil(log_2(number of seconds in 2hours);
         let size_prec_micro =
             test_gcd_date_with_codec(FastFieldCodecType::Bitpacked, DatePrecision::Microseconds)?;
-        assert_eq!(size_prec_micro, 5 + 4 + 19 + (1_000 * 33) / 8); // 33 bits per val = ceil(log_2(number of microsecsseconds in 2hours);
+        assert_eq!(size_prec_micro, 5 + 4 + 26 + (1_000 * 33) / 8); // 33 bits per val = ceil(log_2(number of microsecsseconds in 2hours);
         Ok(())
     }
 
@@ -1014,7 +1014,7 @@ mod tests {
         let reader = index.reader().unwrap();
         let searcher = reader.searcher();
         let segment = &searcher.segment_readers()[0];
-        let field = segment.fast_fields().u64("url_norm_hash").unwrap();
+        let field = segment.fast_fields().u64(num_field).unwrap();
 
         let numbers = vec![100, 200, 300];
         let test_range = |range: RangeInclusive<u64>| {
@@ -1063,7 +1063,7 @@ mod tests {
         let reader = index.reader().unwrap();
         let searcher = reader.searcher();
         let segment = &searcher.segment_readers()[0];
-        let field = segment.fast_fields().u64("url_norm_hash").unwrap();
+        let field = segment.fast_fields().u64(num_field).unwrap();
 
         let numbers = vec![1000, 1001, 1003];
         let test_range = |range: RangeInclusive<u64>| {

src/fastfield/multivalued/mod.rs

@@ -52,7 +52,7 @@ mod tests {
         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")?;
+        let multi_value_reader = segment_reader.fast_fields().u64s(field)?;
         {
             multi_value_reader.get_vals(2, &mut vals);
             assert_eq!(&vals, &[4u64]);
@@ -229,7 +229,7 @@ mod tests {
         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();
+        let multi_value_reader = segment_reader.fast_fields().i64s(field).unwrap();
         multi_value_reader.get_vals(2, &mut vals);
         assert_eq!(&vals, &[-4i64]);
         multi_value_reader.get_vals(0, &mut vals);
@@ -261,7 +261,7 @@ mod tests {
         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();
+        let multi_value_reader = segment_reader.fast_fields().bools(bool_field).unwrap();
         multi_value_reader.get_vals(2, &mut vals);
         assert_eq!(&vals, &[false]);
         multi_value_reader.get_vals(0, &mut vals);

src/fastfield/multivalued/reader.rs

@@ -159,7 +159,7 @@ mod tests {
         let searcher = reader.searcher();
         let reader = searcher.segment_reader(0);
 
-        let date_ff_reader = reader.fast_fields().dates("multi_date_field").unwrap();
+        let date_ff_reader = reader.fast_fields().dates(date_field).unwrap();
         let mut docids = vec![];
         date_ff_reader.get_docids_for_value_range(
             DateTime::from_utc(first_time_stamp)..=DateTime::from_utc(two_secs_ahead),
@@ -173,7 +173,7 @@ mod tests {
 
         assert_eq!(
             count_multiples(RangeQuery::new_date(
-                "multi_date_field".to_string(),
+                date_field,
                 DateTime::from_utc(first_time_stamp)..DateTime::from_utc(two_secs_ahead)
             )),
             1
@@ -226,7 +226,7 @@ mod tests {
         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 date_ff_reader = reader.fast_fields().dates(date_field).unwrap();
         let mut docids = vec![];
         date_ff_reader.get_docids_for_value_range(
             DateTime::from_utc(first_time_stamp)..=DateTime::from_utc(two_secs_ahead),
@@ -240,7 +240,7 @@ mod tests {
 
         assert_eq!(
             count_multiples(RangeQuery::new_date(
-                "multi_date_field".to_string(),
+                date_field,
                 DateTime::from_utc(first_time_stamp)..DateTime::from_utc(two_secs_ahead)
             )),
             2
@@ -324,7 +324,7 @@ mod tests {
         index_writer.commit()?;
         let searcher = index.reader()?.searcher();
         let segment_reader = searcher.segment_reader(0);
-        let field_reader = segment_reader.fast_fields().i64s("items")?;
+        let field_reader = segment_reader.fast_fields().i64s(item_field)?;
 
         assert_eq!(field_reader.min_value(), -2);
         assert_eq!(field_reader.max_value(), 6);

src/fastfield/readers.rs

@@ -114,11 +114,9 @@ impl FastFieldReaders {
 
     pub(crate) fn typed_fast_field_reader_with_idx<TFastValue: FastValue>(
         &self,
-        field_name: &str,
+        field: Field,
         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)?;
@@ -127,37 +125,32 @@ impl FastFieldReaders {
 
     pub(crate) fn typed_fast_field_reader<TFastValue: FastValue>(
         &self,
-        field_name: &str,
+        field: Field,
     ) -> crate::Result<Arc<dyn Column<TFastValue>>> {
-        self.typed_fast_field_reader_with_idx(field_name, 0)
+        self.typed_fast_field_reader_with_idx(field, 0)
     }
 
     pub(crate) fn typed_fast_field_multi_reader<TFastValue: FastValue>(
         &self,
-        field_name: &str,
+        field: Field,
     ) -> 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)?;
+        let idx_reader = self.typed_fast_field_reader(field)?;
+        let vals_reader = self.typed_fast_field_reader_with_idx(field, 1)?;
         Ok(MultiValuedFastFieldReader::open(idx_reader, vals_reader))
     }
 
     /// 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: Field) -> crate::Result<Arc<dyn Column<u64>>> {
+        self.check_type(field, FastType::U64, Cardinality::SingleValue)?;
+        self.typed_fast_field_reader(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)?;
+    pub fn ip_addr(&self, field: Field) -> crate::Result<Arc<dyn Column<Ipv6Addr>>> {
         self.check_type(field, FastType::U128, Cardinality::SingleValue)?;
         let bytes = self.fast_field_data(field, 0)?.read_bytes()?;
         Ok(open_u128::<Ipv6Addr>(bytes)?)
@@ -166,13 +159,9 @@ impl FastFieldReaders {
     /// Returns the `ip` fast field reader reader associated to `field`.
     ///
     /// If `field` is not a u128 fast field, this method returns an Error.
-    pub fn ip_addrs(
-        &self,
-        field_name: &str,
-    ) -> crate::Result<MultiValuedFastFieldReader<Ipv6Addr>> {
-        let field = self.schema.get_field(field_name)?;
+    pub fn ip_addrs(&self, field: Field) -> crate::Result<MultiValuedFastFieldReader<Ipv6Addr>> {
         self.check_type(field, FastType::U128, Cardinality::MultiValues)?;
-        let idx_reader: Arc<dyn Column<u64>> = self.typed_fast_field_reader(field_name)?;
+        let idx_reader: Arc<dyn Column<u64>> = self.typed_fast_field_reader(field)?;
 
         let bytes = self.fast_field_data(field, 1)?.read_bytes()?;
         let vals_reader = open_u128::<Ipv6Addr>(bytes)?;
@@ -183,8 +172,7 @@ impl FastFieldReaders {
     /// 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)?;
+    pub(crate) fn u128(&self, field: Field) -> crate::Result<Arc<dyn Column<u128>>> {
         self.check_type(field, FastType::U128, Cardinality::SingleValue)?;
         let bytes = self.fast_field_data(field, 0)?.read_bytes()?;
         Ok(open_u128::<u128>(bytes)?)
@@ -193,11 +181,9 @@ impl FastFieldReaders {
     /// Returns the `u128` multi-valued fast field reader reader associated to `field`.
     ///
     /// If `field` is not a u128 multi-valued fast field, this method returns an Error.
-    pub fn u128s(&self, field_name: &str) -> crate::Result<MultiValuedFastFieldReader<u128>> {
-        let field = self.schema.get_field(field_name)?;
+    pub fn u128s(&self, field: Field) -> crate::Result<MultiValuedFastFieldReader<u128>> {
         self.check_type(field, FastType::U128, Cardinality::MultiValues)?;
-        let idx_reader: Arc<dyn Column<u64>> =
-            self.typed_fast_field_reader(self.schema.get_field_name(field))?;
+        let idx_reader: Arc<dyn Column<u64>> = self.typed_fast_field_reader(field)?;
 
         let bytes = self.fast_field_data(field, 1)?.read_bytes()?;
         let vals_reader = open_u128::<u128>(bytes)?;
@@ -210,88 +196,80 @@ impl FastFieldReaders {
     ///
     /// 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: Field) -> crate::Result<Arc<dyn Column<u64>>> {
+        self.typed_fast_field_reader(field)
     }
 
     /// 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)?;
+    pub fn i64(&self, field: Field) -> crate::Result<Arc<dyn Column<i64>>> {
         self.check_type(field, FastType::I64, Cardinality::SingleValue)?;
-        self.typed_fast_field_reader(self.schema.get_field_name(field))
+        self.typed_fast_field_reader(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)?;
+    pub fn date(&self, field: Field) -> crate::Result<Arc<dyn Column<DateTime>>> {
         self.check_type(field, FastType::Date, Cardinality::SingleValue)?;
-        self.typed_fast_field_reader(field_name)
+        self.typed_fast_field_reader(field)
     }
 
     /// 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)?;
+    pub fn f64(&self, field: Field) -> crate::Result<Arc<dyn Column<f64>>> {
         self.check_type(field, FastType::F64, Cardinality::SingleValue)?;
-        self.typed_fast_field_reader(field_name)
+        self.typed_fast_field_reader(field)
     }
 
     /// 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)?;
+    pub fn bool(&self, field: Field) -> crate::Result<Arc<dyn Column<bool>>> {
         self.check_type(field, FastType::Bool, Cardinality::SingleValue)?;
-        self.typed_fast_field_reader(field_name)
+        self.typed_fast_field_reader(field)
     }
 
     /// 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)?;
+    pub fn u64s(&self, field: Field) -> crate::Result<MultiValuedFastFieldReader<u64>> {
         self.check_type(field, FastType::U64, Cardinality::MultiValues)?;
-        self.typed_fast_field_multi_reader(field_name)
+        self.typed_fast_field_multi_reader(field)
     }
 
     /// 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)
+    pub fn u64s_lenient(&self, field: Field) -> crate::Result<MultiValuedFastFieldReader<u64>> {
+        self.typed_fast_field_multi_reader(field)
     }
 
     /// 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)?;
+    pub fn i64s(&self, field: Field) -> crate::Result<MultiValuedFastFieldReader<i64>> {
         self.check_type(field, FastType::I64, Cardinality::MultiValues)?;
-        self.typed_fast_field_multi_reader(self.schema.get_field_name(field))
+        self.typed_fast_field_multi_reader(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)?;
+    pub fn f64s(&self, field: Field) -> crate::Result<MultiValuedFastFieldReader<f64>> {
         self.check_type(field, FastType::F64, Cardinality::MultiValues)?;
-        self.typed_fast_field_multi_reader(self.schema.get_field_name(field))
+        self.typed_fast_field_multi_reader(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)?;
+    pub fn bools(&self, field: Field) -> crate::Result<MultiValuedFastFieldReader<bool>> {
         self.check_type(field, FastType::Bool, Cardinality::MultiValues)?;
-        self.typed_fast_field_multi_reader(self.schema.get_field_name(field))
+        self.typed_fast_field_multi_reader(field)
     }
 
     /// Returns a `time::OffsetDateTime` multi-valued fast field reader reader associated with
@@ -299,17 +277,15 @@ impl FastFieldReaders {
     ///
     /// 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)?;
+    pub fn dates(&self, field: Field) -> crate::Result<MultiValuedFastFieldReader<DateTime>> {
         self.check_type(field, FastType::Date, Cardinality::MultiValues)?;
-        self.typed_fast_field_multi_reader(self.schema.get_field_name(field))
+        self.typed_fast_field_multi_reader(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)?;
+    pub fn bytes(&self, field: Field) -> crate::Result<BytesFastFieldReader> {
         let field_entry = self.schema.get_field_entry(field);
         if let FieldType::Bytes(bytes_option) = field_entry.field_type() {
             if !bytes_option.is_fast() {

src/indexer/demuxer.rs

@@ -0,0 +1,322 @@
+use common::BitSet;
+use itertools::Itertools;
+
+use crate::fastfield::AliveBitSet;
+use crate::{merge_filtered_segments, Directory, Index, IndexSettings, Segment, SegmentOrdinal};
+/// DemuxMapping can be used to reorganize data from multiple segments.
+///
+/// DemuxMapping is useful in a multitenant settings, in which each document might actually belong
+/// to a different tenant. It allows to reorganize documents as follows:
+///
+/// e.g. if you have two tenant ids TENANT_A and TENANT_B and two segments with
+/// the documents (simplified)
+/// Seg 1 [TENANT_A, TENANT_B]
+/// Seg 2 [TENANT_A, TENANT_B]
+///
+/// You may want to group your documents to
+/// Seg 1 [TENANT_A, TENANT_A]
+/// Seg 2 [TENANT_B, TENANT_B]
+///
+/// Demuxing is the tool for that.
+/// Semantically you can define a mapping from [old segment ordinal, old doc_id] -> [new segment
+/// ordinal].
+#[derive(Debug, Default)]
+pub struct DemuxMapping {
+    /// [index old segment ordinal] -> [index doc_id] = new segment ordinal
+    mapping: Vec<DocIdToSegmentOrdinal>,
+}
+
+/// DocIdToSegmentOrdinal maps from doc_id within a segment to the new segment ordinal for demuxing.
+///
+/// For every source segment there is a `DocIdToSegmentOrdinal` to distribute its doc_ids.
+#[derive(Debug, Default)]
+pub struct DocIdToSegmentOrdinal {
+    doc_id_index_to_segment_ord: Vec<SegmentOrdinal>,
+}
+
+impl DocIdToSegmentOrdinal {
+    /// Creates a new DocIdToSegmentOrdinal with size of num_doc_ids.
+    /// Initially all doc_ids point to segment ordinal 0 and need to be set
+    /// the via `set` method.
+    pub fn with_max_doc(max_doc: usize) -> Self {
+        DocIdToSegmentOrdinal {
+            doc_id_index_to_segment_ord: vec![0; max_doc],
+        }
+    }
+
+    /// Returns the number of documents in this mapping.
+    /// It should be equal to the `max_doc` of the segment it targets.
+    pub fn max_doc(&self) -> u32 {
+        self.doc_id_index_to_segment_ord.len() as u32
+    }
+
+    /// Associates a doc_id with an output `SegmentOrdinal`.
+    pub fn set(&mut self, doc_id: u32, segment_ord: SegmentOrdinal) {
+        self.doc_id_index_to_segment_ord[doc_id as usize] = segment_ord;
+    }
+
+    /// Iterates over the new SegmentOrdinal in the order of the doc_id.
+    pub fn iter(&self) -> impl Iterator<Item = SegmentOrdinal> + '_ {
+        self.doc_id_index_to_segment_ord.iter().cloned()
+    }
+}
+
+impl DemuxMapping {
+    /// Adds a DocIdToSegmentOrdinal. The order of the pus calls
+    /// defines the old segment ordinal. e.g. first push = ordinal 0.
+    pub fn add(&mut self, segment_mapping: DocIdToSegmentOrdinal) {
+        self.mapping.push(segment_mapping);
+    }
+
+    /// Returns the old number of segments.
+    pub fn get_old_num_segments(&self) -> usize {
+        self.mapping.len()
+    }
+}
+
+fn docs_for_segment_ord(
+    doc_id_to_segment_ord: &DocIdToSegmentOrdinal,
+    target_segment_ord: SegmentOrdinal,
+) -> AliveBitSet {
+    let mut bitset = BitSet::with_max_value(doc_id_to_segment_ord.max_doc());
+    for doc_id in doc_id_to_segment_ord
+        .iter()
+        .enumerate()
+        .filter(|(_doc_id, new_segment_ord)| *new_segment_ord == target_segment_ord)
+        .map(|(doc_id, _)| doc_id)
+    {
+        // add document if segment ordinal = target segment ordinal
+        bitset.insert(doc_id as u32);
+    }
+    AliveBitSet::from_bitset(&bitset)
+}
+
+fn get_alive_bitsets(
+    demux_mapping: &DemuxMapping,
+    target_segment_ord: SegmentOrdinal,
+) -> Vec<AliveBitSet> {
+    demux_mapping
+        .mapping
+        .iter()
+        .map(|doc_id_to_segment_ord| {
+            docs_for_segment_ord(doc_id_to_segment_ord, target_segment_ord)
+        })
+        .collect_vec()
+}
+
+/// Demux the segments according to `demux_mapping`. See `DemuxMapping`.
+/// The number of output_directories need to match max new segment ordinal from `demux_mapping`.
+///
+/// The ordinal of `segments` need to match the ordinals provided in `demux_mapping`.
+pub fn demux(
+    segments: &[Segment],
+    demux_mapping: &DemuxMapping,
+    target_settings: IndexSettings,
+    output_directories: Vec<Box<dyn Directory>>,
+) -> crate::Result<Vec<Index>> {
+    let mut indices = vec![];
+    for (target_segment_ord, output_directory) in output_directories.into_iter().enumerate() {
+        let alive_bitset = get_alive_bitsets(demux_mapping, target_segment_ord as u32)
+            .into_iter()
+            .map(Some)
+            .collect_vec();
+        let index = merge_filtered_segments(
+            segments,
+            target_settings.clone(),
+            alive_bitset,
+            output_directory,
+        )?;
+        indices.push(index);
+    }
+    Ok(indices)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::collector::TopDocs;
+    use crate::directory::RamDirectory;
+    use crate::query::QueryParser;
+    use crate::schema::{Schema, TEXT};
+    use crate::{DocAddress, Term};
+
+    #[test]
+    fn test_demux_map_to_alive_bitset() {
+        let max_value = 2;
+        let mut demux_mapping = DemuxMapping::default();
+        // segment ordinal 0 mapping
+        let mut doc_id_to_segment = DocIdToSegmentOrdinal::with_max_doc(max_value);
+        doc_id_to_segment.set(0, 1);
+        doc_id_to_segment.set(1, 0);
+        demux_mapping.add(doc_id_to_segment);
+
+        // segment ordinal 1 mapping
+        let mut doc_id_to_segment = DocIdToSegmentOrdinal::with_max_doc(max_value);
+        doc_id_to_segment.set(0, 1);
+        doc_id_to_segment.set(1, 1);
+        demux_mapping.add(doc_id_to_segment);
+        {
+            let bit_sets_for_demuxing_to_segment_ord_0 = get_alive_bitsets(&demux_mapping, 0);
+
+            assert_eq!(
+                bit_sets_for_demuxing_to_segment_ord_0[0].is_deleted(0),
+                true
+            );
+            assert_eq!(
+                bit_sets_for_demuxing_to_segment_ord_0[0].is_deleted(1),
+                false
+            );
+            assert_eq!(
+                bit_sets_for_demuxing_to_segment_ord_0[1].is_deleted(0),
+                true
+            );
+            assert_eq!(
+                bit_sets_for_demuxing_to_segment_ord_0[1].is_deleted(1),
+                true
+            );
+        }
+
+        {
+            let bit_sets_for_demuxing_to_segment_ord_1 = get_alive_bitsets(&demux_mapping, 1);
+
+            assert_eq!(
+                bit_sets_for_demuxing_to_segment_ord_1[0].is_deleted(0),
+                false
+            );
+            assert_eq!(
+                bit_sets_for_demuxing_to_segment_ord_1[0].is_deleted(1),
+                true
+            );
+            assert_eq!(
+                bit_sets_for_demuxing_to_segment_ord_1[1].is_deleted(0),
+                false
+            );
+            assert_eq!(
+                bit_sets_for_demuxing_to_segment_ord_1[1].is_deleted(1),
+                false
+            );
+        }
+    }
+
+    #[test]
+    fn test_demux_segments() -> crate::Result<()> {
+        let first_index = {
+            let mut schema_builder = Schema::builder();
+            let text_field = schema_builder.add_text_field("text", TEXT);
+            let index = Index::create_in_ram(schema_builder.build());
+            let mut index_writer = index.writer_for_tests()?;
+            index_writer.add_document(doc!(text_field=>"texto1"))?;
+            index_writer.add_document(doc!(text_field=>"texto2"))?;
+            index_writer.commit()?;
+            index
+        };
+
+        let second_index = {
+            let mut schema_builder = Schema::builder();
+            let text_field = schema_builder.add_text_field("text", TEXT);
+            let index = Index::create_in_ram(schema_builder.build());
+            let mut index_writer = index.writer_for_tests()?;
+            index_writer.add_document(doc!(text_field=>"texto3"))?;
+            index_writer.add_document(doc!(text_field=>"texto4"))?;
+            index_writer.delete_term(Term::from_field_text(text_field, "4"));
+
+            index_writer.commit()?;
+            index
+        };
+
+        let mut segments: Vec<Segment> = Vec::new();
+        segments.extend(first_index.searchable_segments()?);
+        segments.extend(second_index.searchable_segments()?);
+
+        let target_settings = first_index.settings().clone();
+
+        let mut demux_mapping = DemuxMapping::default();
+        {
+            let max_value = 2;
+            // segment ordinal 0 mapping
+            let mut doc_id_to_segment = DocIdToSegmentOrdinal::with_max_doc(max_value);
+            doc_id_to_segment.set(0, 1);
+            doc_id_to_segment.set(1, 0);
+            demux_mapping.add(doc_id_to_segment);
+
+            // segment ordinal 1 mapping
+            let mut doc_id_to_segment = DocIdToSegmentOrdinal::with_max_doc(max_value);
+            doc_id_to_segment.set(0, 1);
+            doc_id_to_segment.set(1, 1);
+            demux_mapping.add(doc_id_to_segment);
+        }
+        assert_eq!(demux_mapping.get_old_num_segments(), 2);
+
+        let demuxed_indices = demux(
+            &segments,
+            &demux_mapping,
+            target_settings,
+            vec![
+                Box::<RamDirectory>::default(),
+                Box::<RamDirectory>::default(),
+            ],
+        )?;
+
+        {
+            let index = &demuxed_indices[0];
+
+            let segments = index.searchable_segments()?;
+            assert_eq!(segments.len(), 1);
+
+            let segment_metas = segments[0].meta();
+            assert_eq!(segment_metas.num_deleted_docs(), 0);
+            assert_eq!(segment_metas.num_docs(), 1);
+
+            let searcher = index.reader().unwrap().searcher();
+            {
+                let text_field = index.schema().get_field("text").unwrap();
+
+                let do_search = |term: &str| {
+                    let query = QueryParser::for_index(index, vec![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<_>>()
+                };
+
+                assert_eq!(do_search("texto1"), vec![] as Vec<u32>);
+                assert_eq!(do_search("texto2"), vec![0]);
+            }
+        }
+
+        {
+            let index = &demuxed_indices[1];
+
+            let segments = index.searchable_segments()?;
+            assert_eq!(segments.len(), 1);
+
+            let segment_metas = segments[0].meta();
+            assert_eq!(segment_metas.num_deleted_docs(), 0);
+            assert_eq!(segment_metas.num_docs(), 3);
+
+            let searcher = index.reader().unwrap().searcher();
+            {
+                let text_field = index.schema().get_field("text").unwrap();
+
+                let do_search = |term: &str| {
+                    let query = QueryParser::for_index(index, vec![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<_>>()
+                };
+
+                assert_eq!(do_search("texto1"), vec![0]);
+                assert_eq!(do_search("texto2"), vec![] as Vec<u32>);
+                assert_eq!(do_search("texto3"), vec![1]);
+                assert_eq!(do_search("texto4"), vec![2]);
+            }
+        }
+
+        Ok(())
+    }
+}

src/indexer/doc_id_mapping.rs

@@ -99,7 +99,7 @@ pub(crate) fn expect_field_id_for_sort_field(
     schema: &Schema,
     sort_by_field: &IndexSortByField,
 ) -> crate::Result<Field> {
-    schema.get_field(&sort_by_field.field).map_err(|_| {
+    schema.get_field(&sort_by_field.field).ok_or_else(|| {
         TantivyError::InvalidArgument(format!(
             "field to sort index by not found: {:?}",
             sort_by_field.field
@@ -462,14 +462,15 @@ mod tests_indexsorting {
         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 my_number = index.schema().get_field("my_number").unwrap();
 
-        let fast_field = fast_fields.u64("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 multifield = fast_fields.u64s("multi_numbers").unwrap();
+        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]);

src/indexer/index_writer.rs

@@ -1465,7 +1465,7 @@ mod tests {
         let segment_reader = searcher.segment_reader(0);
         assert_eq!(segment_reader.num_docs(), 8);
         assert_eq!(segment_reader.max_doc(), 10);
-        let fast_field_reader = segment_reader.fast_fields().u64("id")?;
+        let fast_field_reader = segment_reader.fast_fields().u64(id_field)?;
         let in_order_alive_ids: Vec<u64> = segment_reader
             .doc_ids_alive()
             .map(|doc| fast_field_reader.get_val(doc))
@@ -1526,7 +1526,7 @@ mod tests {
         let segment_reader = searcher.segment_reader(0);
         assert_eq!(segment_reader.num_docs(), 8);
         assert_eq!(segment_reader.max_doc(), 10);
-        let fast_field_reader = segment_reader.fast_fields().u64("id")?;
+        let fast_field_reader = segment_reader.fast_fields().u64(id_field)?;
         let in_order_alive_ids: Vec<u64> = segment_reader
             .doc_ids_alive()
             .map(|doc| fast_field_reader.get_val(doc))
@@ -1778,7 +1778,7 @@ mod tests {
             .segment_readers()
             .iter()
             .flat_map(|segment_reader| {
-                let ff_reader = segment_reader.fast_fields().u64("id").unwrap();
+                let ff_reader = segment_reader.fast_fields().u64(id_field).unwrap();
                 segment_reader
                     .doc_ids_alive()
                     .map(move |doc| ff_reader.get_val(doc))
@@ -1789,7 +1789,7 @@ mod tests {
             .segment_readers()
             .iter()
             .flat_map(|segment_reader| {
-                let ff_reader = segment_reader.fast_fields().u64("id").unwrap();
+                let ff_reader = segment_reader.fast_fields().u64(id_field).unwrap();
                 segment_reader
                     .doc_ids_alive()
                     .map(move |doc| ff_reader.get_val(doc))
@@ -1804,7 +1804,7 @@ mod tests {
         let mut all_ips = Vec::new();
         let mut num_ips = 0;
         for segment_reader in searcher.segment_readers().iter() {
-            let ip_reader = segment_reader.fast_fields().ip_addrs("ips").unwrap();
+            let ip_reader = segment_reader.fast_fields().ip_addrs(ips_field).unwrap();
             for doc in segment_reader.doc_ids_alive() {
                 let mut vals = vec![];
                 ip_reader.get_vals(doc, &mut vals);
@@ -1851,7 +1851,7 @@ mod tests {
                 .segment_readers()
                 .iter()
                 .map(|segment_reader| {
-                    let ff_reader = segment_reader.fast_fields().ip_addrs("ips").unwrap();
+                    let ff_reader = segment_reader.fast_fields().ip_addrs(ips_field).unwrap();
                     ff_reader.get_index_reader().num_docs() as usize
                 })
                 .sum();
@@ -1863,7 +1863,7 @@ mod tests {
             .segment_readers()
             .iter()
             .flat_map(|segment_reader| {
-                let ff_reader = segment_reader.fast_fields().ip_addr("ip").unwrap();
+                let ff_reader = segment_reader.fast_fields().ip_addr(ip_field).unwrap();
                 segment_reader.doc_ids_alive().flat_map(move |doc| {
                     let val = ff_reader.get_val(doc);
                     if val == Ipv6Addr::from_u128(0) {
@@ -1902,7 +1902,7 @@ mod tests {
             .segment_readers()
             .iter()
             .flat_map(|segment_reader| {
-                let ff_reader = segment_reader.fast_fields().ip_addrs("ips").unwrap();
+                let ff_reader = segment_reader.fast_fields().ip_addrs(ips_field).unwrap();
                 segment_reader.doc_ids_alive().flat_map(move |doc| {
                     let mut vals = vec![];
                     ff_reader.get_vals(doc, &mut vals);
@@ -1914,9 +1914,9 @@ mod tests {
 
         // multivalue fast field tests
         for segment_reader in searcher.segment_readers().iter() {
-            let id_reader = segment_reader.fast_fields().u64("id").unwrap();
-            let ff_reader = segment_reader.fast_fields().u64s("multi_numbers").unwrap();
-            let bool_ff_reader = segment_reader.fast_fields().bools("multi_bools").unwrap();
+            let id_reader = segment_reader.fast_fields().u64(id_field).unwrap();
+            let ff_reader = segment_reader.fast_fields().u64s(multi_numbers).unwrap();
+            let bool_ff_reader = segment_reader.fast_fields().bools(multi_bools).unwrap();
             for doc in segment_reader.doc_ids_alive() {
                 let mut vals = vec![];
                 ff_reader.get_vals(doc, &mut vals);
@@ -2109,7 +2109,7 @@ mod tests {
         // test facets
         for segment_reader in searcher.segment_readers().iter() {
             let mut facet_reader = segment_reader.facet_reader(facet_field).unwrap();
-            let ff_reader = segment_reader.fast_fields().u64("id").unwrap();
+            let ff_reader = segment_reader.fast_fields().u64(id_field).unwrap();
             for doc_id in segment_reader.doc_ids_alive() {
                 let mut facet_ords = Vec::new();
                 facet_reader.facet_ords(doc_id, &mut facet_ords);

src/indexer/merger.rs

@@ -16,7 +16,7 @@ use crate::fastfield::{
     MultiValueIndex, MultiValuedFastFieldReader,
 };
 use crate::fieldnorm::{FieldNormReader, FieldNormReaders, FieldNormsSerializer, FieldNormsWriter};
-use crate::indexer::doc_id_mapping::SegmentDocIdMapping;
+use crate::indexer::doc_id_mapping::{expect_field_id_for_sort_field, SegmentDocIdMapping};
 use crate::indexer::sorted_doc_id_column::RemappedDocIdColumn;
 use crate::indexer::sorted_doc_id_multivalue_column::RemappedDocIdMultiValueColumn;
 use crate::indexer::SegmentSerializer;
@@ -335,10 +335,8 @@ impl IndexMerger {
             .readers
             .iter()
             .map(|segment_reader| {
-                let ff_reader: MultiValuedFastFieldReader<u128> = segment_reader
-                    .fast_fields()
-                    .u128s(self.schema.get_field_name(field))
-                    .expect(
+                let ff_reader: MultiValuedFastFieldReader<u128> =
+                    segment_reader.fast_fields().u128s(field).expect(
                         "Failed to find index for multivalued field. This is a bug in tantivy, \
                          please report.",
                     );
@@ -403,13 +401,10 @@ impl IndexMerger {
             .readers
             .iter()
             .map(|reader| {
-                let u128_reader: Arc<dyn Column<u128>> = reader
-                    .fast_fields()
-                    .u128(self.schema.get_field_name(field))
-                    .expect(
-                        "Failed to find a reader for single fast field. This is a tantivy bug and \
-                         it should never happen.",
-                    );
+                let u128_reader: Arc<dyn Column<u128>> = reader.fast_fields().u128(field).expect(
+                    "Failed to find a reader for single fast field. This is a tantivy bug and it \
+                     should never happen.",
+                );
                 u128_reader
             })
             .collect::<Vec<_>>();
@@ -436,11 +431,7 @@ impl IndexMerger {
         fast_field_serializer: &mut CompositeFastFieldSerializer,
         doc_id_mapping: &SegmentDocIdMapping,
     ) -> crate::Result<()> {
-        let fast_field_accessor = RemappedDocIdColumn::new(
-            &self.readers,
-            doc_id_mapping,
-            self.schema.get_field_name(field),
-        );
+        let fast_field_accessor = RemappedDocIdColumn::new(&self.readers, doc_id_mapping, field);
         fast_field_serializer.create_auto_detect_u64_fast_field(field, fast_field_accessor)?;
 
         Ok(())
@@ -473,8 +464,8 @@ impl IndexMerger {
         reader: &SegmentReader,
         sort_by_field: &IndexSortByField,
     ) -> crate::Result<Arc<dyn Column>> {
-        reader.schema().get_field(&sort_by_field.field)?;
-        let value_accessor = reader.fast_fields().u64_lenient(&sort_by_field.field)?;
+        let field_id = expect_field_id_for_sort_field(reader.schema(), sort_by_field)?; // for now expect fastfield, but not strictly required
+        let value_accessor = reader.fast_fields().u64_lenient(field_id)?;
         Ok(value_accessor)
     }
     /// Collecting value_accessors into a vec to bind the lifetime.
@@ -578,7 +569,7 @@ impl IndexMerger {
             .map(|reader| {
                 let u64s_reader: MultiValuedFastFieldReader<u64> = reader
                     .fast_fields()
-                    .typed_fast_field_multi_reader::<u64>(self.schema.get_field_name(field))
+                    .typed_fast_field_multi_reader::<u64>(field)
                     .expect(
                         "Failed to find index for multivalued field. This is a bug in tantivy, \
                          please report.",
@@ -622,7 +613,7 @@ impl IndexMerger {
             .map(|reader| {
                 let ff_reader: MultiValuedFastFieldReader<u64> = reader
                     .fast_fields()
-                    .u64s(self.schema.get_field_name(field))
+                    .u64s(field)
                     .expect("Could not find multivalued u64 fast value reader.");
                 ff_reader
             })
@@ -693,11 +684,8 @@ impl IndexMerger {
 
         self.write_multi_value_fast_field_idx(field, fast_field_serializer, doc_id_mapping)?;
 
-        let fastfield_accessor = RemappedDocIdMultiValueColumn::new(
-            &self.readers,
-            doc_id_mapping,
-            self.schema.get_field_name(field),
-        );
+        let fastfield_accessor =
+            RemappedDocIdMultiValueColumn::new(&self.readers, doc_id_mapping, field);
         fast_field_serializer.create_auto_detect_u64_fast_field_with_idx_and_codecs(
             field,
             fastfield_accessor,
@@ -718,13 +706,10 @@ impl IndexMerger {
             .readers
             .iter()
             .map(|reader| {
-                let bytes_reader = reader
-                    .fast_fields()
-                    .bytes(self.schema.get_field_name(field))
-                    .expect(
-                        "Failed to find index for bytes field. This is a bug in tantivy, please \
-                         report.",
-                    );
+                let bytes_reader = reader.fast_fields().bytes(field).expect(
+                    "Failed to find index for bytes field. This is a bug in tantivy, please \
+                     report.",
+                );
                 (reader, bytes_reader)
             })
             .collect::<Vec<_>>();
@@ -1221,10 +1206,7 @@ mod tests {
             {
                 let get_fast_vals = |terms: Vec<Term>| {
                     let query = BooleanQuery::new_multiterms_query(terms);
-                    searcher.search(
-                        &query,
-                        &FastFieldTestCollector::for_field("score".to_string()),
-                    )
+                    searcher.search(&query, &FastFieldTestCollector::for_field(score_field))
                 };
                 let get_fast_vals_bytes = |terms: Vec<Term>| {
                     let query = BooleanQuery::new_multiterms_query(terms);
@@ -1262,7 +1244,7 @@ mod tests {
         let mut index_writer = index.writer_for_tests()?;
         let reader = index.reader().unwrap();
         let search_term = |searcher: &Searcher, term: Term| {
-            let collector = FastFieldTestCollector::for_field("score".to_string());
+            let collector = FastFieldTestCollector::for_field(score_field);
             let bytes_collector = BytesFastFieldTestCollector::for_field(bytes_score_field);
             let term_query = TermQuery::new(term, IndexRecordOption::Basic);
             searcher
@@ -1384,7 +1366,7 @@ mod tests {
             let score_field_reader = searcher
                 .segment_reader(0)
                 .fast_fields()
-                .u64("score")
+                .u64(score_field)
                 .unwrap();
             assert_eq!(score_field_reader.min_value(), 4000);
             assert_eq!(score_field_reader.max_value(), 7000);
@@ -1392,7 +1374,7 @@ mod tests {
             let score_field_reader = searcher
                 .segment_reader(1)
                 .fast_fields()
-                .u64("score")
+                .u64(score_field)
                 .unwrap();
             assert_eq!(score_field_reader.min_value(), 1);
             assert_eq!(score_field_reader.max_value(), 3);
@@ -1438,7 +1420,7 @@ mod tests {
             let score_field_reader = searcher
                 .segment_reader(0)
                 .fast_fields()
-                .u64("score")
+                .u64(score_field)
                 .unwrap();
             assert_eq!(score_field_reader.min_value(), 3);
             assert_eq!(score_field_reader.max_value(), 7000);
@@ -1485,7 +1467,7 @@ mod tests {
             let score_field_reader = searcher
                 .segment_reader(0)
                 .fast_fields()
-                .u64("score")
+                .u64(score_field)
                 .unwrap();
             assert_eq!(score_field_reader.min_value(), 3);
             assert_eq!(score_field_reader.max_value(), 7000);
@@ -1532,7 +1514,7 @@ mod tests {
             let score_field_reader = searcher
                 .segment_reader(0)
                 .fast_fields()
-                .u64("score")
+                .u64(score_field)
                 .unwrap();
             assert_eq!(score_field_reader.min_value(), 6000);
             assert_eq!(score_field_reader.max_value(), 7000);
@@ -1854,7 +1836,7 @@ mod tests {
 
         {
             let segment = searcher.segment_reader(0u32);
-            let ff_reader = segment.fast_fields().u64s("intvals").unwrap();
+            let ff_reader = segment.fast_fields().u64s(int_field).unwrap();
 
             ff_reader.get_vals(0, &mut vals);
             assert_eq!(&vals, &[1, 2]);
@@ -1880,7 +1862,7 @@ mod tests {
 
         {
             let segment = searcher.segment_reader(1u32);
-            let ff_reader = segment.fast_fields().u64s("intvals").unwrap();
+            let ff_reader = segment.fast_fields().u64s(int_field).unwrap();
             ff_reader.get_vals(0, &mut vals);
             assert_eq!(&vals, &[28, 27]);
 
@@ -1890,7 +1872,7 @@ mod tests {
 
         {
             let segment = searcher.segment_reader(2u32);
-            let ff_reader = segment.fast_fields().u64s("intvals").unwrap();
+            let ff_reader = segment.fast_fields().u64s(int_field).unwrap();
             ff_reader.get_vals(0, &mut vals);
             assert_eq!(&vals, &[20]);
         }
@@ -1907,7 +1889,7 @@ mod tests {
         {
             let searcher = reader.searcher();
             let segment = searcher.segment_reader(0u32);
-            let ff_reader = segment.fast_fields().u64s("intvals").unwrap();
+            let ff_reader = segment.fast_fields().u64s(int_field).unwrap();
 
             ff_reader.get_vals(0, &mut vals);
             assert_eq!(&vals, &[1, 2]);

src/indexer/merger_sorted_index_test.rs

@@ -185,7 +185,7 @@ mod tests {
         let segment_reader = searcher.segment_readers().last().unwrap();
 
         let fast_fields = segment_reader.fast_fields();
-        let fast_field = fast_fields.u64("intval").unwrap();
+        let fast_field = fast_fields.u64(int_field).unwrap();
         assert_eq!(fast_field.get_val(5), 1u64);
         assert_eq!(fast_field.get_val(4), 2u64);
         assert_eq!(fast_field.get_val(3), 3u64);
@@ -364,13 +364,15 @@ mod tests {
         .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();
+        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);
@@ -384,7 +386,7 @@ mod tests {
             vals
         };
         let fast_fields = segment_reader.fast_fields();
-        let fast_field = fast_fields.u64s("multi_numbers").unwrap();
+        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]);
@@ -392,7 +394,7 @@ mod tests {
         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();
+        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]);
@@ -525,6 +527,7 @@ mod bench_sorted_index_merge {
             order: Order::Desc,
         };
         let index = create_index(Some(sort_by_field.clone()));
+        let field = index.schema().get_field("intval").unwrap();
         let segments = index.searchable_segments().unwrap();
         let merger: IndexMerger =
             IndexMerger::open(index.schema(), index.settings().clone(), &segments[..])?;
@@ -532,10 +535,8 @@ mod bench_sorted_index_merge {
         b.iter(|| {
             let sorted_doc_ids = doc_id_mapping.iter_old_doc_addrs().map(|doc_addr| {
                 let reader = &merger.readers[doc_addr.segment_ord as usize];
-                let u64_reader: Arc<dyn Column<u64>> = reader
-                    .fast_fields()
-                    .typed_fast_field_reader("intval")
-                    .expect(
+                let u64_reader: Arc<dyn Column<u64>> =
+                    reader.fast_fields().typed_fast_field_reader(field).expect(
                         "Failed to find a reader for single fast field. This is a tantivy bug and \
                          it should never happen.",
                     );

src/indexer/mod.rs

@@ -1,5 +1,6 @@
 pub mod delete_queue;
 
+pub mod demuxer;
 pub mod doc_id_mapping;
 mod doc_opstamp_mapping;
 mod flat_map_with_buffer;

src/indexer/sorted_doc_id_column.rs

@@ -4,6 +4,7 @@ use fastfield_codecs::Column;
 use itertools::Itertools;
 
 use crate::indexer::doc_id_mapping::SegmentDocIdMapping;
+use crate::schema::Field;
 use crate::SegmentReader;
 
 pub(crate) struct RemappedDocIdColumn<'a> {
@@ -40,7 +41,7 @@ impl<'a> RemappedDocIdColumn<'a> {
     pub(crate) fn new(
         readers: &'a [SegmentReader],
         doc_id_mapping: &'a SegmentDocIdMapping,
-        field: &str,
+        field: Field,
     ) -> Self {
         let (min_value, max_value) = readers
             .iter()

src/indexer/sorted_doc_id_multivalue_column.rs

@@ -5,6 +5,7 @@ use fastfield_codecs::Column;
 use super::flat_map_with_buffer::FlatMapWithBufferIter;
 use crate::fastfield::{MultiValueIndex, MultiValuedFastFieldReader};
 use crate::indexer::doc_id_mapping::SegmentDocIdMapping;
+use crate::schema::Field;
 use crate::{DocAddress, SegmentReader};
 
 pub(crate) struct RemappedDocIdMultiValueColumn<'a> {
@@ -19,7 +20,7 @@ impl<'a> RemappedDocIdMultiValueColumn<'a> {
     pub(crate) fn new(
         readers: &'a [SegmentReader],
         doc_id_mapping: &'a SegmentDocIdMapping,
-        field: &str,
+        field: Field,
     ) -> Self {
         // Our values are bitpacked and we need to know what should be
         // our bitwidth and our minimum value before serializing any values.

src/lib.rs

@@ -299,6 +299,7 @@ pub use crate::core::{
     SegmentReader, SingleSegmentIndexWriter,
 };
 pub use crate::directory::Directory;
+pub use crate::indexer::demuxer::*;
 pub use crate::indexer::operation::UserOperation;
 pub use crate::indexer::{merge_filtered_segments, merge_indices, IndexWriter, PreparedCommit};
 pub use crate::postings::Postings;
@@ -994,8 +995,8 @@ pub mod tests {
         let fast_field_unsigned = schema_builder.add_u64_field("unsigned", FAST);
         let fast_field_signed = schema_builder.add_i64_field("signed", FAST);
         let fast_field_float = schema_builder.add_f64_field("float", FAST);
-        schema_builder.add_text_field("text", TEXT);
-        schema_builder.add_u64_field("stored_int", STORED);
+        let text_field = schema_builder.add_text_field("text", TEXT);
+        let stored_int_field = schema_builder.add_u64_field("stored_int", STORED);
         let schema = schema_builder.build();
 
         let index = Index::create_in_ram(schema);
@@ -1010,37 +1011,37 @@ pub mod tests {
         let searcher = reader.searcher();
         let segment_reader: &SegmentReader = searcher.segment_reader(0);
         {
-            let fast_field_reader_res = segment_reader.fast_fields().u64("text");
+            let fast_field_reader_res = segment_reader.fast_fields().u64(text_field);
             assert!(fast_field_reader_res.is_err());
         }
         {
-            let fast_field_reader_opt = segment_reader.fast_fields().u64("stored_int");
+            let fast_field_reader_opt = segment_reader.fast_fields().u64(stored_int_field);
             assert!(fast_field_reader_opt.is_err());
         }
         {
-            let fast_field_reader_opt = segment_reader.fast_fields().u64("signed");
+            let fast_field_reader_opt = segment_reader.fast_fields().u64(fast_field_signed);
             assert!(fast_field_reader_opt.is_err());
         }
         {
-            let fast_field_reader_opt = segment_reader.fast_fields().u64("float");
+            let fast_field_reader_opt = segment_reader.fast_fields().u64(fast_field_float);
             assert!(fast_field_reader_opt.is_err());
         }
         {
-            let fast_field_reader_opt = segment_reader.fast_fields().u64("unsigned");
+            let fast_field_reader_opt = segment_reader.fast_fields().u64(fast_field_unsigned);
             assert!(fast_field_reader_opt.is_ok());
             let fast_field_reader = fast_field_reader_opt.unwrap();
             assert_eq!(fast_field_reader.get_val(0), 4u64)
         }
 
         {
-            let fast_field_reader_res = segment_reader.fast_fields().i64("signed");
+            let fast_field_reader_res = segment_reader.fast_fields().i64(fast_field_signed);
             assert!(fast_field_reader_res.is_ok());
             let fast_field_reader = fast_field_reader_res.unwrap();
             assert_eq!(fast_field_reader.get_val(0), 4i64)
         }
 
         {
-            let fast_field_reader_res = segment_reader.fast_fields().f64("float");
+            let fast_field_reader_res = segment_reader.fast_fields().f64(fast_field_float);
             assert!(fast_field_reader_res.is_ok());
             let fast_field_reader = fast_field_reader_res.unwrap();
             assert_eq!(fast_field_reader.get_val(0), 4f64)

src/query/query_parser/logical_ast.rs

@@ -10,7 +10,7 @@ pub enum LogicalLiteral {
     Term(Term),
     Phrase(Vec<(usize, Term)>, u32),
     Range {
-        field: String,
+        field: Field,
         value_type: Type,
         lower: Bound<Term>,
         upper: Bound<Term>,

src/query/query_parser/query_parser.rs

@@ -672,7 +672,7 @@ impl QueryParser {
                 let field_entry = self.schema.get_field_entry(field);
                 let value_type = field_entry.field_type().value_type();
                 let logical_ast = LogicalAst::Leaf(Box::new(LogicalLiteral::Range {
-                    field: self.schema.get_field_name(field).to_string(),
+                    field,
                     value_type,
                     lower: self.resolve_bound(field, json_path, &lower)?,
                     upper: self.resolve_bound(field, json_path, &upper)?,
@@ -964,7 +964,7 @@ mod test {
         let query = make_query_parser().parse_query("title:[A TO B]").unwrap();
         assert_eq!(
             format!("{:?}", query),
-            "RangeQuery { field: \"title\", value_type: Str, left_bound: Included([97]), \
+            "RangeQuery { field: Field(0), value_type: Str, left_bound: Included([97]), \
              right_bound: Included([98]) }"
         );
     }

src/query/range_query/range_query.rs

@@ -67,7 +67,7 @@ pub(crate) fn map_bound<TFrom, TTo, Transform: Fn(&TFrom) -> TTo>(
 ///
 /// let reader = index.reader()?;
 /// let searcher = reader.searcher();
-/// let docs_in_the_sixties = RangeQuery::new_u64("year".to_string(), 1960..1970);
+/// let docs_in_the_sixties = RangeQuery::new_u64(year_field, 1960..1970);
 /// let num_60s_books = searcher.search(&docs_in_the_sixties, &Count)?;
 /// assert_eq!(num_60s_books, 2285);
 /// Ok(())
@@ -76,7 +76,7 @@ pub(crate) fn map_bound<TFrom, TTo, Transform: Fn(&TFrom) -> TTo>(
 /// ```
 #[derive(Clone, Debug)]
 pub struct RangeQuery {
-    field: String,
+    field: Field,
     value_type: Type,
     left_bound: Bound<Vec<u8>>,
     right_bound: Bound<Vec<u8>>,
@@ -88,12 +88,15 @@ impl RangeQuery {
     /// If the value type is not correct, something may go terribly wrong when
     /// the `Weight` object is created.
     pub fn new_term_bounds(
-        field: String,
+        field: Field,
         value_type: Type,
         left_bound: &Bound<Term>,
         right_bound: &Bound<Term>,
     ) -> RangeQuery {
-        let verify_and_unwrap_term = |val: &Term| val.value_bytes().to_owned();
+        let verify_and_unwrap_term = |val: &Term| {
+            assert_eq!(field, val.field());
+            val.value_bytes().to_owned()
+        };
         RangeQuery {
             field,
             value_type,
@@ -106,7 +109,7 @@ impl RangeQuery {
     ///
     /// If the field is not of the type `i64`, tantivy
     /// will panic when the `Weight` object is created.
-    pub fn new_i64(field: String, range: Range<i64>) -> RangeQuery {
+    pub fn new_i64(field: Field, range: Range<i64>) -> RangeQuery {
         RangeQuery::new_i64_bounds(
             field,
             Bound::Included(range.start),
@@ -122,15 +125,11 @@ impl RangeQuery {
     /// If the field is not of the type `i64`, tantivy
     /// will panic when the `Weight` object is created.
     pub fn new_i64_bounds(
-        field: String,
+        field: Field,
         left_bound: Bound<i64>,
         right_bound: Bound<i64>,
     ) -> RangeQuery {
-        let make_term_val = |val: &i64| {
-            Term::from_field_i64(Field::from_field_id(0), *val)
-                .value_bytes()
-                .to_owned()
-        };
+        let make_term_val = |val: &i64| Term::from_field_i64(field, *val).value_bytes().to_owned();
         RangeQuery {
             field,
             value_type: Type::I64,
@@ -143,7 +142,7 @@ impl RangeQuery {
     ///
     /// If the field is not of the type `f64`, tantivy
     /// will panic when the `Weight` object is created.
-    pub fn new_f64(field: String, range: Range<f64>) -> RangeQuery {
+    pub fn new_f64(field: Field, range: Range<f64>) -> RangeQuery {
         RangeQuery::new_f64_bounds(
             field,
             Bound::Included(range.start),
@@ -159,15 +158,11 @@ impl RangeQuery {
     /// If the field is not of the type `f64`, tantivy
     /// will panic when the `Weight` object is created.
     pub fn new_f64_bounds(
-        field: String,
+        field: Field,
         left_bound: Bound<f64>,
         right_bound: Bound<f64>,
     ) -> RangeQuery {
-        let make_term_val = |val: &f64| {
-            Term::from_field_f64(Field::from_field_id(0), *val)
-                .value_bytes()
-                .to_owned()
-        };
+        let make_term_val = |val: &f64| Term::from_field_f64(field, *val).value_bytes().to_owned();
         RangeQuery {
             field,
             value_type: Type::F64,
@@ -184,15 +179,11 @@ impl RangeQuery {
     /// If the field is not of the type `u64`, tantivy
     /// will panic when the `Weight` object is created.
     pub fn new_u64_bounds(
-        field: String,
+        field: Field,
         left_bound: Bound<u64>,
         right_bound: Bound<u64>,
     ) -> RangeQuery {
-        let make_term_val = |val: &u64| {
-            Term::from_field_u64(Field::from_field_id(0), *val)
-                .value_bytes()
-                .to_owned()
-        };
+        let make_term_val = |val: &u64| Term::from_field_u64(field, *val).value_bytes().to_owned();
         RangeQuery {
             field,
             value_type: Type::U64,
@@ -205,7 +196,7 @@ impl RangeQuery {
     ///
     /// If the field is not of the type `u64`, tantivy
     /// will panic when the `Weight` object is created.
-    pub fn new_u64(field: String, range: Range<u64>) -> RangeQuery {
+    pub fn new_u64(field: Field, range: Range<u64>) -> RangeQuery {
         RangeQuery::new_u64_bounds(
             field,
             Bound::Included(range.start),
@@ -221,15 +212,12 @@ impl RangeQuery {
     /// If the field is not of the type `date`, tantivy
     /// will panic when the `Weight` object is created.
     pub fn new_date_bounds(
-        field: String,
+        field: Field,
         left_bound: Bound<DateTime>,
         right_bound: Bound<DateTime>,
     ) -> RangeQuery {
-        let make_term_val = |val: &DateTime| {
-            Term::from_field_date(Field::from_field_id(0), *val)
-                .value_bytes()
-                .to_owned()
-        };
+        let make_term_val =
+            |val: &DateTime| Term::from_field_date(field, *val).value_bytes().to_owned();
         RangeQuery {
             field,
             value_type: Type::Date,
@@ -242,7 +230,7 @@ impl RangeQuery {
     ///
     /// If the field is not of the type `date`, tantivy
     /// will panic when the `Weight` object is created.
-    pub fn new_date(field: String, range: Range<DateTime>) -> RangeQuery {
+    pub fn new_date(field: Field, range: Range<DateTime>) -> RangeQuery {
         RangeQuery::new_date_bounds(
             field,
             Bound::Included(range.start),
@@ -257,7 +245,7 @@ impl RangeQuery {
     ///
     /// If the field is not of the type `Str`, tantivy
     /// will panic when the `Weight` object is created.
-    pub fn new_str_bounds(field: String, left: Bound<&str>, right: Bound<&str>) -> RangeQuery {
+    pub fn new_str_bounds(field: Field, left: Bound<&str>, right: Bound<&str>) -> RangeQuery {
         let make_term_val = |val: &&str| val.as_bytes().to_vec();
         RangeQuery {
             field,
@@ -271,7 +259,7 @@ impl RangeQuery {
     ///
     /// If the field is not of the type `Str`, tantivy
     /// will panic when the `Weight` object is created.
-    pub fn new_str(field: String, range: Range<&str>) -> RangeQuery {
+    pub fn new_str(field: Field, range: Range<&str>) -> RangeQuery {
         RangeQuery::new_str_bounds(
             field,
             Bound::Included(range.start),
@@ -280,8 +268,22 @@ impl RangeQuery {
     }
 
     /// Field to search over
-    pub fn field(&self) -> &str {
-        &self.field
+    pub fn field(&self) -> Field {
+        self.field
+    }
+
+    /// Lower bound of range
+    pub fn left_bound(&self) -> Bound<Term> {
+        map_bound(&self.left_bound, &|bytes| {
+            Term::from_field_bytes(self.field, bytes)
+        })
+    }
+
+    /// Upper bound of range
+    pub fn right_bound(&self) -> Bound<Term> {
+        map_bound(&self.right_bound, &|bytes| {
+            Term::from_field_bytes(self.field, bytes)
+        })
     }
 }
 
@@ -305,9 +307,7 @@ pub(crate) fn maps_to_u64_fastfield(typ: Type) -> bool {
 impl Query for RangeQuery {
     fn weight(&self, enable_scoring: EnableScoring<'_>) -> crate::Result<Box<dyn Weight>> {
         let schema = enable_scoring.schema();
-        let field_type = schema
-            .get_field_entry(schema.get_field(&self.field)?)
-            .field_type();
+        let field_type = schema.get_field_entry(self.field).field_type();
         let value_type = field_type.value_type();
         if value_type != self.value_type {
             let err_msg = format!(
@@ -320,7 +320,7 @@ impl Query for RangeQuery {
         if field_type.is_fast() && is_type_valid_for_fastfield_range_query(self.value_type) {
             if field_type.is_ip_addr() {
                 Ok(Box::new(IPFastFieldRangeWeight::new(
-                    self.field.to_string(),
+                    self.field,
                     &self.left_bound,
                     &self.right_bound,
                 )))
@@ -335,14 +335,14 @@ impl Query for RangeQuery {
                 let left_bound = map_bound(&self.left_bound, &parse_from_bytes);
                 let right_bound = map_bound(&self.right_bound, &parse_from_bytes);
                 Ok(Box::new(FastFieldRangeWeight::new(
-                    self.field.to_string(),
+                    self.field,
                     left_bound,
                     right_bound,
                 )))
             }
         } else {
             Ok(Box::new(RangeWeight {
-                field: self.field.to_string(),
+                field: self.field,
                 left_bound: self.left_bound.clone(),
                 right_bound: self.right_bound.clone(),
             }))
@@ -351,7 +351,7 @@ impl Query for RangeQuery {
 }
 
 pub struct RangeWeight {
-    field: String,
+    field: Field,
     left_bound: Bound<Vec<u8>>,
     right_bound: Bound<Vec<u8>>,
 }
@@ -379,7 +379,7 @@ impl Weight for RangeWeight {
         let max_doc = reader.max_doc();
         let mut doc_bitset = BitSet::with_max_value(max_doc);
 
-        let inverted_index = reader.inverted_index(reader.schema().get_field(&self.field)?)?;
+        let inverted_index = reader.inverted_index(self.field)?;
         let term_dict = inverted_index.terms();
         let mut term_range = self.term_range(term_dict)?;
         while term_range.advance() {
@@ -443,7 +443,7 @@ mod tests {
         let reader = index.reader()?;
         let searcher = reader.searcher();
 
-        let docs_in_the_sixties = RangeQuery::new_u64("year".to_string(), 1960u64..1970u64);
+        let docs_in_the_sixties = RangeQuery::new_u64(year_field, 1960u64..1970u64);
 
         // ... or `1960..=1969` if inclusive range is enabled.
         let count = searcher.search(&docs_in_the_sixties, &Count)?;
@@ -481,13 +481,10 @@ mod tests {
         let count_multiples =
             |range_query: RangeQuery| searcher.search(&range_query, &Count).unwrap();
 
-        assert_eq!(
-            count_multiples(RangeQuery::new_i64("intfield".to_string(), 10..11)),
-            9
-        );
+        assert_eq!(count_multiples(RangeQuery::new_i64(int_field, 10..11)), 9);
         assert_eq!(
             count_multiples(RangeQuery::new_i64_bounds(
-                "intfield".to_string(),
+                int_field,
                 Bound::Included(10),
                 Bound::Included(11)
             )),
@@ -495,7 +492,7 @@ mod tests {
         );
         assert_eq!(
             count_multiples(RangeQuery::new_i64_bounds(
-                "intfield".to_string(),
+                int_field,
                 Bound::Excluded(9),
                 Bound::Included(10)
             )),
@@ -503,7 +500,7 @@ mod tests {
         );
         assert_eq!(
             count_multiples(RangeQuery::new_i64_bounds(
-                "intfield".to_string(),
+                int_field,
                 Bound::Included(9),
                 Bound::Unbounded
             )),
@@ -543,12 +540,12 @@ mod tests {
             |range_query: RangeQuery| searcher.search(&range_query, &Count).unwrap();
 
         assert_eq!(
-            count_multiples(RangeQuery::new_f64("floatfield".to_string(), 10.0..11.0)),
+            count_multiples(RangeQuery::new_f64(float_field, 10.0..11.0)),
             9
         );
         assert_eq!(
             count_multiples(RangeQuery::new_f64_bounds(
-                "floatfield".to_string(),
+                float_field,
                 Bound::Included(10.0),
                 Bound::Included(11.0)
             )),
@@ -556,7 +553,7 @@ mod tests {
         );
         assert_eq!(
             count_multiples(RangeQuery::new_f64_bounds(
-                "floatfield".to_string(),
+                float_field,
                 Bound::Excluded(9.0),
                 Bound::Included(10.0)
             )),
@@ -564,7 +561,7 @@ mod tests {
         );
         assert_eq!(
             count_multiples(RangeQuery::new_f64_bounds(
-                "floatfield".to_string(),
+                float_field,
                 Bound::Included(9.0),
                 Bound::Unbounded
             )),

src/query/range_query/range_query_ip_fastfield.rs

@@ -11,18 +11,18 @@ use fastfield_codecs::MonotonicallyMappableToU128;
 use super::fast_field_range_query::{FastFieldCardinality, RangeDocSet};
 use super::range_query::map_bound;
 use crate::query::{ConstScorer, Explanation, Scorer, Weight};
-use crate::schema::Cardinality;
+use crate::schema::{Cardinality, Field};
 use crate::{DocId, DocSet, Score, SegmentReader, TantivyError};
 
 /// `IPFastFieldRangeWeight` uses the ip address fast field to execute range queries.
 pub struct IPFastFieldRangeWeight {
-    field: String,
+    field: Field,
     left_bound: Bound<Ipv6Addr>,
     right_bound: Bound<Ipv6Addr>,
 }
 
 impl IPFastFieldRangeWeight {
-    pub fn new(field: String, left_bound: &Bound<Vec<u8>>, right_bound: &Bound<Vec<u8>>) -> Self {
+    pub fn new(field: Field, left_bound: &Bound<Vec<u8>>, right_bound: &Bound<Vec<u8>>) -> Self {
         let parse_ip_from_bytes = |data: &Vec<u8>| {
             let ip_u128: u128 =
                 u128::from_be(BinarySerializable::deserialize(&mut &data[..]).unwrap());
@@ -40,13 +40,10 @@ impl IPFastFieldRangeWeight {
 
 impl Weight for IPFastFieldRangeWeight {
     fn scorer(&self, reader: &SegmentReader, boost: Score) -> crate::Result<Box<dyn Scorer>> {
-        let field_type = reader
-            .schema()
-            .get_field_entry(reader.schema().get_field(&self.field)?)
-            .field_type();
+        let field_type = reader.schema().get_field_entry(self.field).field_type();
         match field_type.fastfield_cardinality().unwrap() {
             Cardinality::SingleValue => {
-                let ip_addr_fast_field = reader.fast_fields().ip_addr(&self.field)?;
+                let ip_addr_fast_field = reader.fast_fields().ip_addr(self.field)?;
                 let value_range = bound_to_value_range(
                     &self.left_bound,
                     &self.right_bound,
@@ -60,7 +57,7 @@ impl Weight for IPFastFieldRangeWeight {
                 Ok(Box::new(ConstScorer::new(docset, boost)))
             }
             Cardinality::MultiValues => {
-                let ip_addr_fast_field = reader.fast_fields().ip_addrs(&self.field)?;
+                let ip_addr_fast_field = reader.fast_fields().ip_addrs(self.field)?;
                 let value_range = bound_to_value_range(
                     &self.left_bound,
                     &self.right_bound,

src/query/range_query/range_query_u64_fastfield.rs

@@ -9,18 +9,18 @@ use fastfield_codecs::MonotonicallyMappableToU64;
 use super::fast_field_range_query::{FastFieldCardinality, RangeDocSet};
 use super::range_query::map_bound;
 use crate::query::{ConstScorer, Explanation, Scorer, Weight};
-use crate::schema::Cardinality;
+use crate::schema::{Cardinality, Field};
 use crate::{DocId, DocSet, Score, SegmentReader, TantivyError};
 
 /// `FastFieldRangeWeight` uses the fast field to execute range queries.
 pub struct FastFieldRangeWeight {
-    field: String,
+    field: Field,
     left_bound: Bound<u64>,
     right_bound: Bound<u64>,
 }
 
 impl FastFieldRangeWeight {
-    pub fn new(field: String, left_bound: Bound<u64>, right_bound: Bound<u64>) -> Self {
+    pub fn new(field: Field, left_bound: Bound<u64>, right_bound: Bound<u64>) -> Self {
         let left_bound = map_bound(&left_bound, &|val| *val);
         let right_bound = map_bound(&right_bound, &|val| *val);
         Self {
@@ -33,13 +33,10 @@ impl FastFieldRangeWeight {
 
 impl Weight for FastFieldRangeWeight {
     fn scorer(&self, reader: &SegmentReader, boost: Score) -> crate::Result<Box<dyn Scorer>> {
-        let field_type = reader
-            .schema()
-            .get_field_entry(reader.schema().get_field(&self.field)?)
-            .field_type();
+        let field_type = reader.schema().get_field_entry(self.field).field_type();
         match field_type.fastfield_cardinality().unwrap() {
             Cardinality::SingleValue => {
-                let fast_field = reader.fast_fields().u64_lenient(&self.field)?;
+                let fast_field = reader.fast_fields().u64_lenient(self.field)?;
                 let value_range = bound_to_value_range(
                     &self.left_bound,
                     &self.right_bound,
@@ -51,7 +48,7 @@ impl Weight for FastFieldRangeWeight {
                 Ok(Box::new(ConstScorer::new(docset, boost)))
             }
             Cardinality::MultiValues => {
-                let fast_field = reader.fast_fields().u64s_lenient(&self.field)?;
+                let fast_field = reader.fast_fields().u64s_lenient(self.field)?;
                 let value_range = bound_to_value_range(
                     &self.left_bound,
                     &self.right_bound,