This is an automated email from the ASF dual-hosted git repository.
tisonkun pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/datasketches-rust.git
The following commit(s) were added to refs/heads/main by this push:
new d375eca docs: refine unreleased changelog entries (#345)
d375eca is described below
commit d375eca34f8c1eec3622f9bf407a92a4fdb10fc1
Author: tison <[email protected]>
AuthorDate: Thu Nov 12 00:56:56 2026 +0800
docs: refine unreleased changelog entries (#345)
---
CHANGELOG.md | 32 ++++++++++++++++----------------
1 file changed, 16 insertions(+), 16 deletions(-)
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 53d8fb6..a4c114f 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,26 +4,26 @@ All significant changes to this project will be documented in
this file.
## Unreleased
+### Breaking changes
+
+* `CountMinSketch::new` and `CountMinSketch::with_seed` now create serial
version 2 sketches, require `num_hashes` in `1..=127`, and emit images that
require a version 2 reader. Version 2 enables cross-language interchange when
using `i64` counters and the documented input encodings, while existing version
1 images remain readable, updatable, and serializable as version 1.
`CountMinSketch::merge` rejects mixing version 1 and version 2 sketches;
upgrade readers before writers and rebuild [...]
+
### New features
-* Use serial version 2 for new `CountMinSketch` instances, enabling
cross-language interchange with the documented input encodings. New sketches
support `1..=127` rows and require a version 2 reader. Existing version 1 data
remains readable and updatable, but must be rebuilt from source items to merge
with new sketches.
-* Add `CpcUnion::serialize`, which produces the same bytes as
`to_sketch().serialize()` and can be faster.
-* Add `HllUnion::serialize` to serialize union results directly.
-* Add `ThetaUnion::serialize` and `TupleUnion::serialize` to serialize ordered
or unordered union results directly in their compact binary formats. Tuple
summaries are serialized without cloning, avoiding per-entry allocations for
heap-backed summaries.
-* Add `XorFilter` and `XorFilterBuilder` behind the `xor` feature for
immutable membership queries on values or precomputed hashes. Filters support
8-, 16-, and 32-bit fingerprints and can be serialized for later queries.
+* Add `CpcUnion::serialize`, `HllUnion::serialize`, `ThetaUnion::serialize`,
and `TupleUnion::serialize` to encode union results without calling
`to_sketch`. Direct serialization avoids intermediate sketch allocations for
`HllUnion` and low-cardinality `CpcUnion` inputs, writes an `HllType::Hll8`
image at the union's effective `lg_k` for `HllUnion`, and encodes `TupleUnion`
summaries without requiring `Clone` or cloning heap-backed summaries.
+* Add `XorFilter`, `XorFilterBuilder`, and `XorFilterType` behind the `xor`
feature for immutable membership queries on values or precomputed hashes.
Filters support 8-, 16-, and 32-bit fingerprints and can be serialized for
later queries.
-### Improvements
+### Performance improvements
-* `FrequentItemsSketch` updates are faster for streams of distinct items while
the map is growing.
-* `HllUnion` is faster when adopting HLL8 inputs without downsampling or
merging multiple dense inputs before querying or serializing the result.
-* `KllSketch::rank` is faster for repeated queries on compacted sketches.
-* Repeated scalar `TDigest::rank` queries are faster, including small digests
and data with repeated values.
-* `TDigest` and `TDigestMut` CDF and PMF batches are faster for many split
points; batches with sparse split points keep the original binary-search
behavior.
-* `BloomFilter::insert` is faster for streams of new items when multiple
hashes are configured.
-* Compressed `CompactThetaSketch` serialization and deserialization are faster.
-* Tuple sketch serialization and deserialization are faster.
-* `ThetaANotB` and `TupleANotB` are faster when either input is unordered.
-* Compacting Theta and Tuple sketches and producing ordered union,
intersection, and A-not-B results are faster.
+* Speed up `BloomFilter::insert` when adding new items to filters configured
with multiple hash functions; single-hash, repeated-item, and saturated-filter
workloads see smaller gains.
+* Reduce `FrequentItemsSketch` update cost on streams of distinct items while
growing toward `max_map_size`, particularly for keys such as strings or byte
slices whose comparisons cost more than integers; updates once at capacity are
unchanged.
+* Speed up `HllUnion::update` when adopting a high-cardinality `HllType::Hll8`
sketch within `lg_max_k` or merging multiple high-cardinality sketches before
querying or serializing the union; workloads that query after every merge are
roughly unchanged.
+* Speed up `KllSketch::rank` in estimation mode while keeping exact-mode query
latency unchanged.
+* Speed up `TDigest` and `TDigestMut` queries for dense `cdf` and `pmf`
split-point batches and scalar `rank` lookups, particularly on digests with
repeated values, with roughly unchanged performance on sparse split-point
batches.
+* Speed up `CompactThetaSketch::serialize_compressed` and
`CompactThetaSketch::deserialize` for compressed images while preserving the
serialized format.
+* Speed up `CompactTupleSketch` and `TupleUnion` serialization and
`CompactTupleSketch` deserialization for primitive summaries and custom
`TupleSummaryValue` implementations built on `SketchBytes` and `SketchSlice`.
+* Speed up `ThetaSketch::compact` and `TupleSketch::compact` on populated
sketches, along with ordered result generation across `ThetaUnion`,
`TupleUnion`, `ThetaIntersection`, `TupleIntersection`, `ThetaANotB`, and
`TupleANotB` for larger retained entry counts; sketches with very few retained
entries are unchanged.
+* Speed up `ThetaANotB::compute` and `TupleANotB::compute` when either input
is unordered, while avoiding intermediate allocations when subtracting a
mutable `ThetaSketch` or `TupleSketch` directly; pairs of ordered compact
inputs are unchanged.
## v0.6.0 (2026-10-29)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]