Relying on upstream version of datasketch and stop using HLL 4. (#2936)

We were relying on a fork for:

a bugfix in LIST serialization
a better API exposing a new Coupon type, required for caching coupons.
We also stop using HLL8 in hope to fix
https://datadoghq.atlassian.net/browse/CLOUDPREM-625

Co-authored-by: Paul Masurel <paul.masurel@datadoghq.com>

Cargo.toml

@@ -65,7 +65,7 @@ tantivy-bitpacker = { version = "0.10", path = "./bitpacker" }
 common = { version = "0.11", path = "./common/", package = "tantivy-common" }
 tokenizer-api = { version = "0.7", path = "./tokenizer-api", package = "tantivy-tokenizer-api" }
 sketches-ddsketch = { version = "0.4", features = ["use_serde"] }
-datasketches = { git = "https://github.com/fulmicoton-dd/datasketches-rust", rev = "7635fb8" }
+datasketches = { version = "0.3.0", features = ["hll"] }
 futures-util = { version = "0.3.28", optional = true }
 futures-channel = { version = "0.3.28", optional = true }
 fnv = "1.0.7"
@@ -75,7 +75,7 @@ typetag = "0.2.21"
 winapi = "0.3.9"
 
 [dev-dependencies]
-binggan = "0.16.1"
+binggan = "0.17.0"
 rand = "0.9"
 maplit = "1.0.2"
 matches = "0.1.9"

columnar/Cargo.toml

@@ -23,7 +23,7 @@ downcast-rs = "2.0.1"
 proptest = "1"
 more-asserts = "0.3.1"
 rand = "0.9"
-binggan = "0.16.1"
+binggan = "0.17.0"
 
 [[bench]]
 name = "bench_merge"

common/Cargo.toml

@@ -19,6 +19,6 @@ time = { version = "0.3.47", features = ["serde-well-known"] }
 serde = { version = "1.0.136", features = ["derive"] }
 
 [dev-dependencies]
-binggan = "0.16.1"
+binggan = "0.17.0"
 proptest = "1.0.0"
 rand = "0.9"

src/aggregation/agg_req.rs

@@ -115,6 +115,71 @@ pub fn get_fast_field_names(aggs: &Aggregations) -> HashSet<String> {
     fast_field_names
 }
 
+/// Validates that all fields referenced in the aggregation request exist in the schema
+/// and are configured as fast fields.
+///
+/// This is a convenience function for upfront validation before executing aggregations.
+/// Returns an error if any field doesn't exist or is not a fast field.
+///
+/// Validation is intentionally opt-in rather than baked into aggregation execution: the
+/// default lenient behavior (returning empty results for missing fields) supports
+/// schema evolution and federated queries where the same request runs against segments
+/// or indices with different schemas.
+///
+/// # Example
+/// ```
+/// use tantivy::aggregation::agg_req::{Aggregations, validate_aggregation_fields_exist};
+/// use tantivy::schema::{Schema, FAST};
+/// use tantivy::Index;
+///
+/// # fn main() -> tantivy::Result<()> {
+/// // Create a simple index
+/// let mut schema_builder = Schema::builder();
+/// schema_builder.add_f64_field("price", FAST);
+/// let schema = schema_builder.build();
+/// let index = Index::create_in_ram(schema);
+///
+/// // Parse aggregation request
+/// let agg_req: Aggregations = serde_json::from_str(r#"{
+///     "avg_price": { "avg": { "field": "price" } }
+/// }"#)?;
+///
+/// let reader = index.reader()?;
+/// let searcher = reader.searcher();
+///
+/// // Validate fields before executing
+/// for segment_reader in searcher.segment_readers() {
+///     validate_aggregation_fields_exist(&agg_req, segment_reader)?;
+/// }
+/// # Ok(())
+/// # }
+/// ```
+pub fn validate_aggregation_fields_exist(
+    aggs: &Aggregations,
+    reader: &crate::SegmentReader,
+) -> crate::Result<()> {
+    let field_names = get_fast_field_names(aggs);
+    let schema = reader.schema();
+
+    for field_name in field_names {
+        // Check if the field is either directly in the schema or could be part of a json field
+        // present in the schema, and verify it's a fast field.
+        if let Some((field, _path)) = schema.find_field(&field_name) {
+            let field_type = schema.get_field_entry(field).field_type();
+            if !field_type.is_fast() {
+                return Err(crate::TantivyError::SchemaError(format!(
+                    "Field '{}' is not a fast field. Aggregations require fast fields.",
+                    field_name
+                )));
+            }
+        } else {
+            return Err(crate::TantivyError::FieldNotFound(field_name));
+        }
+    }
+
+    Ok(())
+}
+
 #[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
 /// All aggregation types.
 pub enum AggregationVariants {

src/aggregation/agg_tests.rs

@@ -1436,3 +1436,46 @@ fn test_aggregation_on_json_object_mixed_numerical_segments() {
         )
     );
 }
+
+#[test]
+fn test_aggregation_field_validation_helper() {
+    // Test the standalone validation helper function for field validation
+    let index = get_test_index_2_segments(false).unwrap();
+    let reader = index.reader().unwrap();
+    let searcher = reader.searcher();
+    let segment_reader = searcher.segment_reader(0);
+
+    // Test with invalid field
+    let agg_req: Aggregations = serde_json::from_str(
+        r#"{
+        "avg_test": {
+            "avg": { "field": "nonexistent_field" }
+        }
+    }"#,
+    )
+    .unwrap();
+
+    let result =
+        crate::aggregation::agg_req::validate_aggregation_fields_exist(&agg_req, segment_reader);
+    assert!(result.is_err());
+    match result {
+        Err(crate::TantivyError::FieldNotFound(field_name)) => {
+            assert_eq!(field_name, "nonexistent_field");
+        }
+        _ => panic!("Expected FieldNotFound error, got: {:?}", result),
+    }
+
+    // Test with valid field
+    let agg_req: Aggregations = serde_json::from_str(
+        r#"{
+        "avg_test": {
+            "avg": { "field": "score" }
+        }
+    }"#,
+    )
+    .unwrap();
+
+    let result =
+        crate::aggregation::agg_req::validate_aggregation_fields_exist(&agg_req, segment_reader);
+    assert!(result.is_ok());
+}

src/aggregation/metric/cardinality.rs

@@ -166,8 +166,12 @@ impl CouponCache {
         let should_use_dense =
             highest_term_ord < 1_000_000u64 || highest_term_ord < num_terms as u64 * 3u64;
         if should_use_dense {
-            let mut coupon_map: Vec<Coupon> = vec![Coupon::EMPTY; highest_term_ord as usize + 1];
-            for (term_ord, coupon) in term_ords.into_iter().zip(coupons.into_iter()) {
+            // We don't really care about the value here. We will populate all the values we will
+            // read anyway.
+            let uninitialized_coupon = Coupon::from_hash(0);
+            let mut coupon_map: Vec<Coupon> =
+                vec![uninitialized_coupon; highest_term_ord as usize + 1];
+            for (term_ord, coupon) in term_ords.into_iter().zip(coupons) {
                 coupon_map[term_ord as usize] = coupon;
             }
             CouponCache::Dense {
@@ -821,7 +825,7 @@ impl<'de> Deserialize<'de> for CardinalityCollector {
 impl CardinalityCollector {
     fn new(salt: u8) -> Self {
         Self {
-            sketch: HllSketch::new(LG_K, HllType::Hll4),
+            sketch: HllSketch::new(LG_K, HllType::Hll8),
             salt,
         }
     }
@@ -852,7 +856,7 @@ impl CardinalityCollector {
         let mut union = HllUnion::new(LG_K);
         union.update(&self.sketch);
         union.update(&right.sketch);
-        self.sketch = union.to_sketch(HllType::Hll4);
+        self.sketch = union.to_sketch(HllType::Hll8);
         Ok(())
     }
 }

src/index/segment_reader.rs

@@ -6,6 +6,7 @@ use common::{ByteCount, HasLen};
 use fnv::FnvHashMap;
 use itertools::Itertools;
 
+use crate::directory::error::OpenReadError;
 use crate::directory::{CompositeFile, FileSlice};
 use crate::error::DataCorruption;
 use crate::fastfield::{intersect_alive_bitsets, AliveBitSet, FacetReader, FastFieldReaders};
@@ -159,12 +160,10 @@ impl SegmentReader {
         let postings_file = segment.open_read(SegmentComponent::Postings)?;
         let postings_composite = CompositeFile::open(&postings_file)?;
 
-        let positions_composite = {
-            if let Ok(positions_file) = segment.open_read(SegmentComponent::Positions) {
-                CompositeFile::open(&positions_file)?
-            } else {
-                CompositeFile::empty()
-            }
+        let positions_composite = match segment.open_read(SegmentComponent::Positions) {
+            Ok(positions_file) => CompositeFile::open(&positions_file)?,
+            Err(OpenReadError::FileDoesNotExist(_)) => CompositeFile::empty(),
+            Err(open_read_error) => return Err(open_read_error.into()),
         };
 
         let schema = segment.schema();

sstable/src/dictionary.rs

@@ -14,11 +14,8 @@ use itertools::Itertools;
 use tantivy_fst::Automaton;
 use tantivy_fst::automaton::AlwaysMatch;
 
-use crate::sstable_index_v3::SSTableIndexV3Empty;
 use crate::streamer::{Streamer, StreamerBuilder};
-use crate::{
-    BlockAddr, DeltaReader, Reader, SSTable, SSTableIndex, SSTableIndexV3, TermOrdinal, VoidSSTable,
-};
+use crate::{BlockAddr, DeltaReader, Reader, SSTable, SSTableIndex, TermOrdinal, VoidSSTable};
 
 /// An SSTable is a sorted map that associates sorted `&[u8]` keys
 /// to any kind of typed values.
@@ -288,33 +285,7 @@ impl<TSSTable: SSTable> Dictionary<TSSTable> {
         let (sstable_slice, index_slice) = main_slice.split(index_offset as usize);
         let sstable_index_bytes = index_slice.read_bytes()?;
 
-        let sstable_index = match version {
-            2 => SSTableIndex::V2(
-                crate::sstable_index_v2::SSTableIndex::load(sstable_index_bytes).map_err(|_| {
-                    io::Error::new(io::ErrorKind::InvalidData, "SSTable corruption")
-                })?,
-            ),
-            3 => {
-                let (sstable_index_bytes, mut footerv3_len_bytes) = sstable_index_bytes.rsplit(8);
-                let store_offset = u64::deserialize(&mut footerv3_len_bytes)?;
-                if store_offset != 0 {
-                    SSTableIndex::V3(
-                        SSTableIndexV3::load(sstable_index_bytes, store_offset).map_err(|_| {
-                            io::Error::new(io::ErrorKind::InvalidData, "SSTable corruption")
-                        })?,
-                    )
-                } else {
-                    // if store_offset is zero, there is no index, so we build a pseudo-index
-                    // assuming a single block of sstable covering everything.
-                    SSTableIndex::V3Empty(SSTableIndexV3Empty::load(index_offset as usize))
-                }
-            }
-            _ => {
-                return Err(io::Error::other(format!(
-                    "Unsupported sstable version, expected one of [2, 3], found {version}"
-                )));
-            }
-        };
+        let sstable_index = SSTableIndex::open(version, index_offset, sstable_index_bytes)?;
 
         Ok(Dictionary {
             sstable_slice,
@@ -525,10 +496,15 @@ impl<TSSTable: SSTable> Dictionary<TSSTable> {
 
         // Open the block for the first ordinal.
         let mut bytes = Vec::new();
-        let mut current_block_addr = self.sstable_index.get_block_with_ord(ord);
+        let (mut current_block_addr, block_id) = self.sstable_index.get_and_locate_with_ord(ord);
         let mut current_sstable_delta_reader =
             self.sstable_delta_reader_block(current_block_addr.clone())?;
         let mut current_block_ordinal = current_block_addr.first_ordinal;
+        let mut current_block_end_bound = self
+            .sstable_index
+            .get_block(block_id + 1)
+            .map(|block_addr| block_addr.first_ordinal)
+            .unwrap_or(u64::MAX);
 
         loop {
             // move to the ord inside the current block
@@ -557,17 +533,19 @@ impl<TSSTable: SSTable> Dictionary<TSSTable> {
                 }
             };
 
-            // TODO optimization: it is silly to do a binary search to get the block every single
-            // time.
-            //
-            // Check if block changed for new term_ord
-            let new_block_addr = self.sstable_index.get_block_with_ord(next_ord);
-            if new_block_addr != current_block_addr {
+            if next_ord >= current_block_end_bound {
+                let (new_block_addr, block_id) =
+                    self.sstable_index.get_and_locate_with_ord(next_ord);
                 current_block_addr = new_block_addr;
                 current_block_ordinal = current_block_addr.first_ordinal;
                 current_sstable_delta_reader =
                     self.sstable_delta_reader_block(current_block_addr.clone())?;
                 bytes.clear();
+                current_block_end_bound = self
+                    .sstable_index
+                    .get_block(block_id + 1)
+                    .map(|block_addr| block_addr.first_ordinal)
+                    .unwrap_or(u64::MAX)
             }
             ord = next_ord;
         }

sstable/src/index/mod.rs

@@ -0,0 +1,319 @@
+pub(crate) mod v2;
+pub(crate) mod v3;
+
+use std::io::{self, Read, Write};
+use std::ops::Range;
+
+use common::{BinarySerializable, FixedSize, OwnedBytes};
+use tantivy_fst::{Automaton, MapBuilder};
+
+use crate::{TermOrdinal, common_prefix_len};
+
+#[derive(Debug, Clone)]
+pub enum SSTableIndex {
+    V2(v2::SSTableIndex),
+    V3(v3::SSTableIndexV3),
+    V3Empty(v3::SSTableIndexV3Empty),
+}
+
+impl SSTableIndex {
+    pub(crate) fn open(
+        version: u32,
+        index_offset: u64,
+        index_bytes: OwnedBytes,
+    ) -> io::Result<Self> {
+        let index = match version {
+            2 => {
+                SSTableIndex::V2(v2::SSTableIndex::load(index_bytes).map_err(|_| {
+                    io::Error::new(io::ErrorKind::InvalidData, "SSTable corruption")
+                })?)
+            }
+            3 => {
+                let (index_bytes, mut footerv3_len_bytes) = index_bytes.rsplit(8);
+                let store_offset = u64::deserialize(&mut footerv3_len_bytes)?;
+                if store_offset != 0 {
+                    SSTableIndex::V3(v3::SSTableIndexV3::load(index_bytes, store_offset).map_err(
+                        |_| io::Error::new(io::ErrorKind::InvalidData, "SSTable corruption"),
+                    )?)
+                } else {
+                    // if store_offset is zero, there is no index, so we build a pseudo-index
+                    // assuming a single block of sstable covering everything.
+                    SSTableIndex::V3Empty(v3::SSTableIndexV3Empty::load(index_offset as usize))
+                }
+            }
+            _ => {
+                return Err(io::Error::other(format!(
+                    "Unsupported sstable version, expected one of [2, 3], found {version}"
+                )));
+            }
+        };
+        Ok(index)
+    }
+
+    /// Get the [`BlockAddr`] of the requested block.
+    pub(crate) fn get_block(&self, block_id: u64) -> Option<BlockAddr> {
+        match self {
+            SSTableIndex::V2(v2_index) => v2_index.get_block(block_id as usize),
+            SSTableIndex::V3(v3_index) => v3_index.get_block(block_id),
+            SSTableIndex::V3Empty(v3_empty) => v3_empty.get_block(block_id),
+        }
+    }
+
+    /// Get the block id of the block that would contain `key`.
+    ///
+    /// Returns None if `key` is lexicographically after the last key recorded.
+    pub(crate) fn locate_with_key(&self, key: &[u8]) -> Option<u64> {
+        match self {
+            SSTableIndex::V2(v2_index) => v2_index.locate_with_key(key).map(|i| i as u64),
+            SSTableIndex::V3(v3_index) => v3_index.locate_with_key(key),
+            SSTableIndex::V3Empty(v3_empty) => v3_empty.locate_with_key(key),
+        }
+    }
+
+    /// Get the [`BlockAddr`] of the block that would contain `key`.
+    ///
+    /// Returns None if `key` is lexicographically after the last key recorded.
+    pub fn get_block_with_key(&self, key: &[u8]) -> Option<BlockAddr> {
+        match self {
+            SSTableIndex::V2(v2_index) => v2_index.get_block_with_key(key),
+            SSTableIndex::V3(v3_index) => v3_index.get_block_with_key(key),
+            SSTableIndex::V3Empty(v3_empty) => v3_empty.get_block_with_key(key),
+        }
+    }
+
+    pub(crate) fn locate_with_ord(&self, ord: TermOrdinal) -> u64 {
+        match self {
+            SSTableIndex::V2(v2_index) => v2_index.locate_with_ord(ord) as u64,
+            SSTableIndex::V3(v3_index) => v3_index.locate_with_ord(ord),
+            SSTableIndex::V3Empty(v3_empty) => v3_empty.locate_with_ord(ord),
+        }
+    }
+
+    /// Get the [`BlockAddr`] of the block containing the `ord`-th term.
+    pub(crate) fn get_block_with_ord(&self, ord: TermOrdinal) -> BlockAddr {
+        match self {
+            SSTableIndex::V2(v2_index) => v2_index.get_block_with_ord(ord),
+            SSTableIndex::V3(v3_index) => v3_index.get_block_with_ord(ord),
+            SSTableIndex::V3Empty(v3_empty) => v3_empty.get_block_with_ord(ord),
+        }
+    }
+
+    pub(crate) fn get_and_locate_with_ord(&self, ord: TermOrdinal) -> (BlockAddr, u64) {
+        match self {
+            SSTableIndex::V2(v2_index) => v2_index.get_and_locate_with_ord(ord),
+            SSTableIndex::V3(v3_index) => v3_index.get_and_locate_with_ord(ord),
+            SSTableIndex::V3Empty(v3_empty) => v3_empty.get_and_locate_with_ord(ord),
+        }
+    }
+
+    pub fn get_block_for_automaton<'a>(
+        &'a self,
+        automaton: &'a impl Automaton,
+    ) -> impl Iterator<Item = (u64, BlockAddr)> + 'a {
+        match self {
+            SSTableIndex::V2(v2_index) => {
+                BlockIter::V2(v2_index.get_block_for_automaton(automaton))
+            }
+            SSTableIndex::V3(v3_index) => {
+                BlockIter::V3(v3_index.get_block_for_automaton(automaton))
+            }
+            SSTableIndex::V3Empty(v3_empty) => {
+                BlockIter::V3Empty(std::iter::once((0, v3_empty.block_addr.clone())))
+            }
+        }
+    }
+}
+
+enum BlockIter<V2, V3, T> {
+    V2(V2),
+    V3(V3),
+    V3Empty(std::iter::Once<T>),
+}
+
+impl<V2: Iterator<Item = T>, V3: Iterator<Item = T>, T> Iterator for BlockIter<V2, V3, T> {
+    type Item = T;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        match self {
+            BlockIter::V2(v2) => v2.next(),
+            BlockIter::V3(v3) => v3.next(),
+            BlockIter::V3Empty(once) => once.next(),
+        }
+    }
+}
+
+#[derive(Clone, Eq, PartialEq, Debug)]
+pub struct BlockAddr {
+    pub first_ordinal: u64,
+    pub byte_range: Range<usize>,
+}
+
+impl BlockAddr {
+    fn to_block_start(&self) -> BlockStartAddr {
+        BlockStartAddr {
+            first_ordinal: self.first_ordinal,
+            byte_range_start: self.byte_range.start,
+        }
+    }
+}
+
+#[derive(Debug, Clone, PartialEq, Eq)]
+struct BlockStartAddr {
+    first_ordinal: u64,
+    byte_range_start: usize,
+}
+
+impl BlockStartAddr {
+    fn to_block_addr(&self, byte_range_end: usize) -> BlockAddr {
+        BlockAddr {
+            first_ordinal: self.first_ordinal,
+            byte_range: self.byte_range_start..byte_range_end,
+        }
+    }
+}
+
+#[derive(Debug, Clone)]
+pub(crate) struct BlockMeta {
+    /// Any byte string that is lexicographically greater or equal to
+    /// the last key in the block,
+    /// and yet strictly smaller than the first key in the next block.
+    pub last_key_or_greater: Vec<u8>,
+    pub block_addr: BlockAddr,
+}
+
+impl BinarySerializable for BlockStartAddr {
+    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
+        let start = self.byte_range_start as u64;
+        start.serialize(writer)?;
+        self.first_ordinal.serialize(writer)
+    }
+
+    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
+        let byte_range_start = u64::deserialize(reader)? as usize;
+        let first_ordinal = u64::deserialize(reader)?;
+        Ok(BlockStartAddr {
+            first_ordinal,
+            byte_range_start,
+        })
+    }
+
+    // Provided method
+    fn num_bytes(&self) -> u64 {
+        BlockStartAddr::SIZE_IN_BYTES as u64
+    }
+}
+
+impl FixedSize for BlockStartAddr {
+    const SIZE_IN_BYTES: usize = 2 * u64::SIZE_IN_BYTES;
+}
+
+/// Given that left < right,
+/// mutates `left into a shorter byte string left'` that
+/// matches `left <= left' < right`.
+fn find_shorter_str_in_between(left: &mut Vec<u8>, right: &[u8]) {
+    assert!(&left[..] < right);
+    let common_len = common_prefix_len(left, right);
+    if left.len() == common_len {
+        return;
+    }
+    // It is possible to do one character shorter in some case,
+    // but it is not worth the extra complexity
+    for pos in (common_len + 1)..left.len() {
+        if left[pos] != u8::MAX {
+            left[pos] += 1;
+            left.truncate(pos + 1);
+            return;
+        }
+    }
+}
+
+#[derive(Default)]
+pub struct SSTableIndexBuilder {
+    blocks: Vec<BlockMeta>,
+}
+
+impl SSTableIndexBuilder {
+    /// In order to make the index as light as possible, we
+    /// try to find a shorter alternative to the last key of the last block
+    /// that is still smaller than the next key.
+    pub(crate) fn shorten_last_block_key_given_next_key(&mut self, next_key: &[u8]) {
+        if let Some(last_block) = self.blocks.last_mut() {
+            find_shorter_str_in_between(&mut last_block.last_key_or_greater, next_key);
+        }
+    }
+
+    pub fn add_block(&mut self, last_key: &[u8], byte_range: Range<usize>, first_ordinal: u64) {
+        self.blocks.push(BlockMeta {
+            last_key_or_greater: last_key.to_vec(),
+            block_addr: BlockAddr {
+                byte_range,
+                first_ordinal,
+            },
+        })
+    }
+
+    pub fn serialize<W: std::io::Write>(&self, wrt: W) -> io::Result<u64> {
+        if self.blocks.len() <= 1 {
+            return Ok(0);
+        }
+        let counting_writer = common::CountingWriter::wrap(wrt);
+        let mut map_builder = MapBuilder::new(counting_writer).map_err(fst_error_to_io_error)?;
+        for (i, block) in self.blocks.iter().enumerate() {
+            map_builder
+                .insert(&block.last_key_or_greater, i as u64)
+                .map_err(fst_error_to_io_error)?;
+        }
+        let counting_writer = map_builder.into_inner().map_err(fst_error_to_io_error)?;
+        let written_bytes = counting_writer.written_bytes();
+        let mut wrt = counting_writer.finish();
+
+        let mut block_store_writer = v3::BlockAddrStoreWriter::new();
+        for block in &self.blocks {
+            block_store_writer.write_block_meta(block.block_addr.clone())?;
+        }
+        block_store_writer.serialize(&mut wrt)?;
+
+        Ok(written_bytes)
+    }
+}
+
+fn fst_error_to_io_error(error: tantivy_fst::Error) -> io::Error {
+    match error {
+        tantivy_fst::Error::Fst(fst_error) => io::Error::other(fst_error),
+        tantivy_fst::Error::Io(ioerror) => ioerror,
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    #[track_caller]
+    fn test_find_shorter_str_in_between_aux(left: &[u8], right: &[u8]) {
+        let mut left_buf = left.to_vec();
+        super::find_shorter_str_in_between(&mut left_buf, right);
+        assert!(left_buf.len() <= left.len());
+        assert!(left <= &left_buf);
+        assert!(&left_buf[..] < right);
+    }
+
+    #[test]
+    fn test_find_shorter_str_in_between() {
+        test_find_shorter_str_in_between_aux(b"", b"hello");
+        test_find_shorter_str_in_between_aux(b"abc", b"abcd");
+        test_find_shorter_str_in_between_aux(b"abcd", b"abd");
+        test_find_shorter_str_in_between_aux(&[0, 0, 0], &[1]);
+        test_find_shorter_str_in_between_aux(&[0, 0, 0], &[0, 0, 1]);
+        test_find_shorter_str_in_between_aux(&[0, 0, 255, 255, 255, 0u8], &[0, 1]);
+    }
+
+    use proptest::prelude::*;
+
+    proptest! {
+        #![proptest_config(ProptestConfig::with_cases(100))]
+        #[test]
+        fn test_proptest_find_shorter_str(left in any::<Vec<u8>>(), right in any::<Vec<u8>>()) {
+            if left < right {
+                test_find_shorter_str_in_between_aux(&left, &right);
+            }
+        }
+    }
+}

sstable/src/index/v2.rs

@@ -77,6 +77,13 @@ impl SSTableIndex {
         self.get_block(self.locate_with_ord(ord)).unwrap()
     }
 
+    pub(crate) fn get_and_locate_with_ord(&self, ord: TermOrdinal) -> (BlockAddr, u64) {
+        let location = self.locate_with_ord(ord);
+        // locate_with_ord always returns an index within range
+        let block_addr = self.get_block(location).unwrap();
+        (block_addr, location as u64)
+    }
+
     pub(crate) fn get_block_for_automaton<'a>(
         &'a self,
         automaton: &'a impl Automaton,

sstable/src/index/v3.rs

@@ -1,106 +1,14 @@
 use std::io::{self, Read, Write};
-use std::ops::Range;
 use std::sync::Arc;
 
 use common::{BinarySerializable, FixedSize, OwnedBytes};
 use tantivy_bitpacker::{BitPacker, compute_num_bits};
 use tantivy_fst::raw::Fst;
-use tantivy_fst::{Automaton, IntoStreamer, Map, MapBuilder, Streamer};
+use tantivy_fst::{Automaton, IntoStreamer, Map, Streamer};
 
+use super::{BlockAddr, BlockStartAddr};
 use crate::block_match_automaton::can_block_match_automaton;
-use crate::{SSTableDataCorruption, TermOrdinal, common_prefix_len};
-
-#[derive(Debug, Clone)]
-pub enum SSTableIndex {
-    V2(crate::sstable_index_v2::SSTableIndex),
-    V3(SSTableIndexV3),
-    V3Empty(SSTableIndexV3Empty),
-}
-
-impl SSTableIndex {
-    /// Get the [`BlockAddr`] of the requested block.
-    pub(crate) fn get_block(&self, block_id: u64) -> Option<BlockAddr> {
-        match self {
-            SSTableIndex::V2(v2_index) => v2_index.get_block(block_id as usize),
-            SSTableIndex::V3(v3_index) => v3_index.get_block(block_id),
-            SSTableIndex::V3Empty(v3_empty) => v3_empty.get_block(block_id),
-        }
-    }
-
-    /// Get the block id of the block that would contain `key`.
-    ///
-    /// Returns None if `key` is lexicographically after the last key recorded.
-    pub(crate) fn locate_with_key(&self, key: &[u8]) -> Option<u64> {
-        match self {
-            SSTableIndex::V2(v2_index) => v2_index.locate_with_key(key).map(|i| i as u64),
-            SSTableIndex::V3(v3_index) => v3_index.locate_with_key(key),
-            SSTableIndex::V3Empty(v3_empty) => v3_empty.locate_with_key(key),
-        }
-    }
-
-    /// Get the [`BlockAddr`] of the block that would contain `key`.
-    ///
-    /// Returns None if `key` is lexicographically after the last key recorded.
-    pub fn get_block_with_key(&self, key: &[u8]) -> Option<BlockAddr> {
-        match self {
-            SSTableIndex::V2(v2_index) => v2_index.get_block_with_key(key),
-            SSTableIndex::V3(v3_index) => v3_index.get_block_with_key(key),
-            SSTableIndex::V3Empty(v3_empty) => v3_empty.get_block_with_key(key),
-        }
-    }
-
-    pub(crate) fn locate_with_ord(&self, ord: TermOrdinal) -> u64 {
-        match self {
-            SSTableIndex::V2(v2_index) => v2_index.locate_with_ord(ord) as u64,
-            SSTableIndex::V3(v3_index) => v3_index.locate_with_ord(ord),
-            SSTableIndex::V3Empty(v3_empty) => v3_empty.locate_with_ord(ord),
-        }
-    }
-
-    /// Get the [`BlockAddr`] of the block containing the `ord`-th term.
-    pub(crate) fn get_block_with_ord(&self, ord: TermOrdinal) -> BlockAddr {
-        match self {
-            SSTableIndex::V2(v2_index) => v2_index.get_block_with_ord(ord),
-            SSTableIndex::V3(v3_index) => v3_index.get_block_with_ord(ord),
-            SSTableIndex::V3Empty(v3_empty) => v3_empty.get_block_with_ord(ord),
-        }
-    }
-
-    pub fn get_block_for_automaton<'a>(
-        &'a self,
-        automaton: &'a impl Automaton,
-    ) -> impl Iterator<Item = (u64, BlockAddr)> + 'a {
-        match self {
-            SSTableIndex::V2(v2_index) => {
-                BlockIter::V2(v2_index.get_block_for_automaton(automaton))
-            }
-            SSTableIndex::V3(v3_index) => {
-                BlockIter::V3(v3_index.get_block_for_automaton(automaton))
-            }
-            SSTableIndex::V3Empty(v3_empty) => {
-                BlockIter::V3Empty(std::iter::once((0, v3_empty.block_addr.clone())))
-            }
-        }
-    }
-}
-
-enum BlockIter<V2, V3, T> {
-    V2(V2),
-    V3(V3),
-    V3Empty(std::iter::Once<T>),
-}
-
-impl<V2: Iterator<Item = T>, V3: Iterator<Item = T>, T> Iterator for BlockIter<V2, V3, T> {
-    type Item = T;
-
-    fn next(&mut self) -> Option<Self::Item> {
-        match self {
-            BlockIter::V2(v2) => v2.next(),
-            BlockIter::V3(v3) => v3.next(),
-            BlockIter::V3Empty(once) => once.next(),
-        }
-    }
-}
+use crate::{SSTableDataCorruption, TermOrdinal};
 
 #[derive(Debug, Clone)]
 pub struct SSTableIndexV3 {
@@ -160,6 +68,11 @@ impl SSTableIndexV3 {
         self.block_addr_store.binary_search_ord(ord).1
     }
 
+    pub(crate) fn get_and_locate_with_ord(&self, ord: TermOrdinal) -> (BlockAddr, u64) {
+        let (location, block_addr) = self.block_addr_store.binary_search_ord(ord);
+        (block_addr, location)
+    }
+
     pub(crate) fn get_block_for_automaton<'a>(
         &'a self,
         automaton: &'a impl Automaton,
@@ -216,7 +129,7 @@ impl<A: Automaton> Iterator for GetBlockForAutomaton<'_, A> {
 
 #[derive(Debug, Clone)]
 pub struct SSTableIndexV3Empty {
-    block_addr: BlockAddr,
+    pub block_addr: BlockAddr,
 }
 
 impl SSTableIndexV3Empty {
@@ -230,8 +143,8 @@ impl SSTableIndexV3Empty {
     }
 
     /// Get the [`BlockAddr`] of the requested block.
-    pub(crate) fn get_block(&self, _block_id: u64) -> Option<BlockAddr> {
-        Some(self.block_addr.clone())
+    pub(crate) fn get_block(&self, block_id: u64) -> Option<BlockAddr> {
+        (block_id == 0).then(|| self.block_addr.clone())
     }
 
     /// Get the block id of the block that would contain `key`.
@@ -256,146 +169,9 @@ impl SSTableIndexV3Empty {
     pub(crate) fn get_block_with_ord(&self, _ord: TermOrdinal) -> BlockAddr {
         self.block_addr.clone()
     }
-}
-#[derive(Clone, Eq, PartialEq, Debug)]
-pub struct BlockAddr {
-    pub first_ordinal: u64,
-    pub byte_range: Range<usize>,
-}
 
-impl BlockAddr {
-    fn to_block_start(&self) -> BlockStartAddr {
-        BlockStartAddr {
-            first_ordinal: self.first_ordinal,
-            byte_range_start: self.byte_range.start,
-        }
-    }
-}
-
-#[derive(Debug, Clone, PartialEq, Eq)]
-struct BlockStartAddr {
-    first_ordinal: u64,
-    byte_range_start: usize,
-}
-
-impl BlockStartAddr {
-    fn to_block_addr(&self, byte_range_end: usize) -> BlockAddr {
-        BlockAddr {
-            first_ordinal: self.first_ordinal,
-            byte_range: self.byte_range_start..byte_range_end,
-        }
-    }
-}
-
-#[derive(Debug, Clone)]
-pub(crate) struct BlockMeta {
-    /// Any byte string that is lexicographically greater or equal to
-    /// the last key in the block,
-    /// and yet strictly smaller than the first key in the next block.
-    pub last_key_or_greater: Vec<u8>,
-    pub block_addr: BlockAddr,
-}
-
-impl BinarySerializable for BlockStartAddr {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        let start = self.byte_range_start as u64;
-        start.serialize(writer)?;
-        self.first_ordinal.serialize(writer)
-    }
-
-    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
-        let byte_range_start = u64::deserialize(reader)? as usize;
-        let first_ordinal = u64::deserialize(reader)?;
-        Ok(BlockStartAddr {
-            first_ordinal,
-            byte_range_start,
-        })
-    }
-
-    // Provided method
-    fn num_bytes(&self) -> u64 {
-        BlockStartAddr::SIZE_IN_BYTES as u64
-    }
-}
-
-impl FixedSize for BlockStartAddr {
-    const SIZE_IN_BYTES: usize = 2 * u64::SIZE_IN_BYTES;
-}
-
-/// Given that left < right,
-/// mutates `left into a shorter byte string left'` that
-/// matches `left <= left' < right`.
-fn find_shorter_str_in_between(left: &mut Vec<u8>, right: &[u8]) {
-    assert!(&left[..] < right);
-    let common_len = common_prefix_len(left, right);
-    if left.len() == common_len {
-        return;
-    }
-    // It is possible to do one character shorter in some case,
-    // but it is not worth the extra complexity
-    for pos in (common_len + 1)..left.len() {
-        if left[pos] != u8::MAX {
-            left[pos] += 1;
-            left.truncate(pos + 1);
-            return;
-        }
-    }
-}
-
-#[derive(Default)]
-pub struct SSTableIndexBuilder {
-    blocks: Vec<BlockMeta>,
-}
-
-impl SSTableIndexBuilder {
-    /// In order to make the index as light as possible, we
-    /// try to find a shorter alternative to the last key of the last block
-    /// that is still smaller than the next key.
-    pub(crate) fn shorten_last_block_key_given_next_key(&mut self, next_key: &[u8]) {
-        if let Some(last_block) = self.blocks.last_mut() {
-            find_shorter_str_in_between(&mut last_block.last_key_or_greater, next_key);
-        }
-    }
-
-    pub fn add_block(&mut self, last_key: &[u8], byte_range: Range<usize>, first_ordinal: u64) {
-        self.blocks.push(BlockMeta {
-            last_key_or_greater: last_key.to_vec(),
-            block_addr: BlockAddr {
-                byte_range,
-                first_ordinal,
-            },
-        })
-    }
-
-    pub fn serialize<W: std::io::Write>(&self, wrt: W) -> io::Result<u64> {
-        if self.blocks.len() <= 1 {
-            return Ok(0);
-        }
-        let counting_writer = common::CountingWriter::wrap(wrt);
-        let mut map_builder = MapBuilder::new(counting_writer).map_err(fst_error_to_io_error)?;
-        for (i, block) in self.blocks.iter().enumerate() {
-            map_builder
-                .insert(&block.last_key_or_greater, i as u64)
-                .map_err(fst_error_to_io_error)?;
-        }
-        let counting_writer = map_builder.into_inner().map_err(fst_error_to_io_error)?;
-        let written_bytes = counting_writer.written_bytes();
-        let mut wrt = counting_writer.finish();
-
-        let mut block_store_writer = BlockAddrStoreWriter::new();
-        for block in &self.blocks {
-            block_store_writer.write_block_meta(block.block_addr.clone())?;
-        }
-        block_store_writer.serialize(&mut wrt)?;
-
-        Ok(written_bytes)
-    }
-}
-
-fn fst_error_to_io_error(error: tantivy_fst::Error) -> io::Error {
-    match error {
-        tantivy_fst::Error::Fst(fst_error) => io::Error::other(fst_error),
-        tantivy_fst::Error::Io(ioerror) => ioerror,
+    pub(crate) fn get_and_locate_with_ord(&self, _ord: TermOrdinal) -> (BlockAddr, u64) {
+        (self.block_addr.clone(), 0)
     }
 }
 
@@ -647,14 +423,14 @@ fn binary_search(max: u64, cmp_fn: impl Fn(u64) -> std::cmp::Ordering) -> Result
     Err(left)
 }
 
-struct BlockAddrStoreWriter {
+pub(crate) struct BlockAddrStoreWriter {
     buffer_block_metas: Vec<u8>,
     buffer_addrs: Vec<u8>,
     block_addrs: Vec<BlockAddr>,
 }
 
 impl BlockAddrStoreWriter {
-    fn new() -> Self {
+    pub(crate) fn new() -> Self {
         BlockAddrStoreWriter {
             buffer_block_metas: Vec::new(),
             buffer_addrs: Vec::new(),
@@ -662,7 +438,7 @@ impl BlockAddrStoreWriter {
         }
     }
 
-    fn flush_block(&mut self) -> io::Result<()> {
+    pub(crate) fn flush_block(&mut self) -> io::Result<()> {
         if self.block_addrs.is_empty() {
             return Ok(());
         }
@@ -741,7 +517,7 @@ impl BlockAddrStoreWriter {
         Ok(())
     }
 
-    fn write_block_meta(&mut self, block_addr: BlockAddr) -> io::Result<()> {
+    pub(crate) fn write_block_meta(&mut self, block_addr: BlockAddr) -> io::Result<()> {
         self.block_addrs.push(block_addr);
         if self.block_addrs.len() >= STORE_BLOCK_LEN {
             self.flush_block()?;
@@ -749,7 +525,7 @@ impl BlockAddrStoreWriter {
         Ok(())
     }
 
-    fn serialize<W: std::io::Write>(&mut self, wrt: &mut W) -> io::Result<()> {
+    pub(crate) fn serialize<W: std::io::Write>(&mut self, wrt: &mut W) -> io::Result<()> {
         self.flush_block()?;
         let len = self.buffer_block_metas.len() as u64;
         len.serialize(wrt)?;
@@ -824,8 +600,9 @@ mod tests {
     use common::OwnedBytes;
 
     use super::*;
-    use crate::SSTableDataCorruption;
     use crate::block_match_automaton::tests::EqBuffer;
+    use crate::index::BlockMeta;
+    use crate::{SSTableDataCorruption, SSTableIndexBuilder};
 
     #[test]
     fn test_sstable_index() {
@@ -874,36 +651,7 @@ mod tests {
         assert!(matches!(data_corruption_err, SSTableDataCorruption));
     }
 
-    #[track_caller]
-    fn test_find_shorter_str_in_between_aux(left: &[u8], right: &[u8]) {
-        let mut left_buf = left.to_vec();
-        super::find_shorter_str_in_between(&mut left_buf, right);
-        assert!(left_buf.len() <= left.len());
-        assert!(left <= &left_buf);
-        assert!(&left_buf[..] < right);
-    }
-
-    #[test]
-    fn test_find_shorter_str_in_between() {
-        test_find_shorter_str_in_between_aux(b"", b"hello");
-        test_find_shorter_str_in_between_aux(b"abc", b"abcd");
-        test_find_shorter_str_in_between_aux(b"abcd", b"abd");
-        test_find_shorter_str_in_between_aux(&[0, 0, 0], &[1]);
-        test_find_shorter_str_in_between_aux(&[0, 0, 0], &[0, 0, 1]);
-        test_find_shorter_str_in_between_aux(&[0, 0, 255, 255, 255, 0u8], &[0, 1]);
-    }
-
-    use proptest::prelude::*;
-
-    proptest! {
-        #![proptest_config(ProptestConfig::with_cases(100))]
-        #[test]
-        fn test_proptest_find_shorter_str(left in any::<Vec<u8>>(), right in any::<Vec<u8>>()) {
-            if left < right {
-                test_find_shorter_str_in_between_aux(&left, &right);
-            }
-        }
-    }
+    //    use proptest::prelude::*;
 
     #[test]
     fn test_find_best_slop() {

sstable/src/lib.rs

@@ -47,9 +47,8 @@ pub mod merge;
 mod streamer;
 pub mod value;
 
-mod sstable_index_v3;
-pub use sstable_index_v3::{BlockAddr, SSTableIndex, SSTableIndexBuilder, SSTableIndexV3};
-mod sstable_index_v2;
+mod index;
+pub use index::{BlockAddr, SSTableIndex, SSTableIndexBuilder};
 pub(crate) mod vint;
 pub use dictionary::{Dictionary, TermOrdHit};
 pub use streamer::{Streamer, StreamerBuilder};

stacker/Cargo.toml

@@ -27,7 +27,7 @@ rand = "0.9"
 zipf = "7.0.0"
 rustc-hash = "2.1.0"
 proptest = "1.2.0"
-binggan = { version = "0.16.1" }
+binggan = { version = "0.17.0" }
 rand_distr = "0.5"
 
 [features]