diff --git a/milli/src/heed_codec/roaring_bitmap/cbo_roaring_bitmap_codec.rs b/milli/src/heed_codec/roaring_bitmap/cbo_roaring_bitmap_codec.rs index 79b52695e..117da1308 100644 --- a/milli/src/heed_codec/roaring_bitmap/cbo_roaring_bitmap_codec.rs +++ b/milli/src/heed_codec/roaring_bitmap/cbo_roaring_bitmap_codec.rs @@ -6,6 +6,7 @@ use byteorder::{NativeEndian, ReadBytesExt, WriteBytesExt}; use roaring::RoaringBitmap; use crate::heed_codec::BytesDecodeOwned; +use crate::update::del_add::{DelAdd, KvReaderDelAdd}; /// This is the limit where using a byteorder became less size efficient /// than using a direct roaring encoding, it is also the point where we are able @@ -99,6 +100,28 @@ impl CboRoaringBitmapCodec { Ok(()) } + + /// Merges a DelAdd delta into a CboRoaringBitmap. + pub fn merge_deladd_into( + deladd: KvReaderDelAdd<'_>, + previous: &[u8], + buffer: &mut Vec, + ) -> io::Result<()> { + // Deserialize the bitmap that is already there + let mut previous = Self::deserialize_from(previous)?; + + // Remove integers we no more want in the previous bitmap + if let Some(value) = deladd.get(DelAdd::Deletion) { + previous -= Self::deserialize_from(value)?; + } + + // Insert the new integers we want in the previous bitmap + if let Some(value) = deladd.get(DelAdd::Addition) { + previous |= Self::deserialize_from(value)?; + } + + previous.serialize_into(buffer) + } } impl heed::BytesDecode<'_> for CboRoaringBitmapCodec { diff --git a/milli/src/update/index_documents/typed_chunk.rs b/milli/src/update/index_documents/typed_chunk.rs index e0e2ff1ec..faeee944f 100644 --- a/milli/src/update/index_documents/typed_chunk.rs +++ b/milli/src/update/index_documents/typed_chunk.rs @@ -134,7 +134,7 @@ pub(crate) fn write_typed_chunk_into_index( wtxn, index_is_empty, deladd_serialize_add_side, - merge_cbo_roaring_bitmaps, + merge_deladd_cbo_roaring_bitmaps, )?; is_merged_database = true; } @@ -153,7 +153,7 @@ pub(crate) fn write_typed_chunk_into_index( wtxn, index_is_empty, deladd_serialize_add_side, - merge_cbo_roaring_bitmaps, + merge_deladd_cbo_roaring_bitmaps, )?; let exact_word_docids_iter = unsafe { as_cloneable_grenad(&exact_word_docids_reader) }?; @@ -163,7 +163,7 @@ pub(crate) fn write_typed_chunk_into_index( wtxn, index_is_empty, deladd_serialize_add_side, - merge_cbo_roaring_bitmaps, + merge_deladd_cbo_roaring_bitmaps, )?; let word_fid_docids_iter = unsafe { as_cloneable_grenad(&word_fid_docids_reader) }?; @@ -173,7 +173,7 @@ pub(crate) fn write_typed_chunk_into_index( wtxn, index_is_empty, deladd_serialize_add_side, - merge_cbo_roaring_bitmaps, + merge_deladd_cbo_roaring_bitmaps, )?; // create fst from word docids @@ -195,7 +195,7 @@ pub(crate) fn write_typed_chunk_into_index( wtxn, index_is_empty, deladd_serialize_add_side, - merge_cbo_roaring_bitmaps, + merge_deladd_cbo_roaring_bitmaps, )?; is_merged_database = true; } @@ -216,7 +216,7 @@ pub(crate) fn write_typed_chunk_into_index( wtxn, index_is_empty, deladd_serialize_add_side, - merge_cbo_roaring_bitmaps, + merge_deladd_cbo_roaring_bitmaps, )?; is_merged_database = true; } @@ -227,7 +227,7 @@ pub(crate) fn write_typed_chunk_into_index( wtxn, index_is_empty, deladd_serialize_add_side, - merge_cbo_roaring_bitmaps, + merge_deladd_cbo_roaring_bitmaps, )?; is_merged_database = true; } @@ -238,7 +238,7 @@ pub(crate) fn write_typed_chunk_into_index( wtxn, index_is_empty, deladd_serialize_add_side, - merge_cbo_roaring_bitmaps, + merge_deladd_cbo_roaring_bitmaps, )?; is_merged_database = true; } @@ -249,7 +249,7 @@ pub(crate) fn write_typed_chunk_into_index( wtxn, index_is_empty, deladd_serialize_add_side, - merge_cbo_roaring_bitmaps, + merge_deladd_cbo_roaring_bitmaps, )?; is_merged_database = true; } @@ -388,17 +388,6 @@ fn merge_word_docids_reader_into_fst( Ok(builder.into_set()) } -fn merge_cbo_roaring_bitmaps( - new_value: &[u8], - db_value: &[u8], - buffer: &mut Vec, -) -> Result<()> { - Ok(CboRoaringBitmapCodec::merge_into( - &[Cow::Borrowed(db_value), Cow::Borrowed(new_value)], - buffer, - )?) -} - /// A function that extracts and returns the Add side of a DelAdd obkv. /// This is useful when there are no previous value in the database and /// therefore we don't need to do a diff with what's already there. @@ -409,6 +398,22 @@ fn deladd_serialize_add_side<'a>(obkv: &'a [u8], _buffer: &mut Vec) -> Resul Ok(KvReaderDelAdd::new(obkv).get(DelAdd::Addition).unwrap_or_default()) } +/// A function that merges a DelAdd of bitmao into an already existing bitmap. +/// +/// The first argument is the DelAdd obkv of CboRoaringBitmaps and +/// the second one is the CboRoaringBitmap to merge into. +fn merge_deladd_cbo_roaring_bitmaps( + deladd_obkv: &[u8], + previous: &[u8], + buffer: &mut Vec, +) -> Result<()> { + Ok(CboRoaringBitmapCodec::merge_deladd_into( + KvReaderDelAdd::new(deladd_obkv), + previous, + buffer, + )?) +} + /// Write provided entries in database using serialize_value function. /// merge_values function is used if an entry already exist in the database. fn write_entries_into_database(