Patch the CBO roaring encoders to use the main branch

Cargo.lock

@@ -5168,8 +5168,7 @@ checksum = "3582f63211428f83597b51b2ddb88e2a91a9d52d12831f9d08f5e624e8977422"
 [[package]]
 name = "roaring"
 version = "0.10.12"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "19e8d2cfa184d94d0726d650a9f4a1be7f9b76ac9fdb954219878dc00c1c1e7b"
+source = "git+https://github.com/RoaringBitmap/roaring-rs.git#6535a822358fce546eb021da8b02d52e4906fe7a"
 dependencies = [
  "bytemuck",
  "byteorder",

Cargo.toml

@@ -49,3 +49,6 @@ opt-level = 3
 opt-level = 3
 [profile.dev.package.roaring]
 opt-level = 3
+
+[patch.crates-io]
+roaring = { git = "https://github.com/RoaringBitmap/roaring-rs.git" }

crates/milli/src/heed_codec/roaring_bitmap/bo_roaring_bitmap_codec.rs

@@ -43,8 +43,10 @@ impl heed::BytesEncode<'_> for BoRoaringBitmapCodec {
     type EItem = RoaringBitmap;
 
     fn bytes_encode(item: &Self::EItem) -> Result<Cow<'_, [u8]>, BoxedError> {
+        let mut item = item.clone();
+        item.optimize();
         let mut out = Vec::new();
-        BoRoaringBitmapCodec::serialize_into(item, &mut out);
+        BoRoaringBitmapCodec::serialize_into(&item, &mut out);
         Ok(Cow::Owned(out))
     }
 }

crates/milli/src/heed_codec/roaring_bitmap/cbo_roaring_bitmap_codec.rs

@@ -12,7 +12,7 @@ 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
 /// to determine the encoding used only by using the array of bytes length.
-pub const THRESHOLD: usize = 7;
+pub const THRESHOLD: usize = 3;
 
 /// A conditionnal codec that either use the RoaringBitmap
 /// or a lighter ByteOrder en/decoding method.
@@ -177,8 +177,10 @@ impl heed::BytesEncode<'_> for CboRoaringBitmapCodec {
     type EItem = RoaringBitmap;
 
     fn bytes_encode(item: &Self::EItem) -> Result<Cow<'_, [u8]>, BoxedError> {
-        let mut vec = Vec::with_capacity(Self::serialized_size(item));
-        Self::serialize_into_vec(item, &mut vec);
+        let mut item = item.clone();
+        item.optimize();
+        let mut vec = Vec::with_capacity(Self::serialized_size(&item));
+        Self::serialize_into_vec(&item, &mut vec);
         Ok(Cow::Owned(vec))
     }
 }

crates/milli/src/heed_codec/roaring_bitmap/roaring_bitmap_codec.rs

@@ -27,6 +27,8 @@ impl heed::BytesEncode<'_> for RoaringBitmapCodec {
     type EItem = RoaringBitmap;
 
     fn bytes_encode(item: &Self::EItem) -> Result<Cow<'_, [u8]>, BoxedError> {
+        let mut item = item.clone();
+        item.optimize();
         let mut bytes = Vec::with_capacity(item.serialized_size());
         item.serialize_into(&mut bytes)?;
         Ok(Cow::Owned(bytes))