v0.8.0

• Add Dump and Loader for serializing/deserializing Bloom filters
• Minor tweaks

v0.7.2

• SyncFilter implementation decoupled from Filter, to reduce the size of binaries linking in SyncFilter
• Other minor changes

v0.7.1

• Bulk operations (Union, Intersection) are now faster, especially on arm64.
• Documentation fixes.
• Benchmarks updated.

v0.7.0

• Filter.AddAtomic has been replaced by a new type, SyncFilter, which implements a concurrency-safe lock-free Bloom filter.
• The new example program bloomstat can be used to determine optimal sizes for Bloom filters.
• Some corner cases in Optimize and FPRate now trigger panics instead of infinite loops.

v0.6.0

The new method Empty reports whether a Filter is empty. The new method Fill fills it, meaning Has will return true for any key. AddAtomic has been optimized for speed.

v0.5.1

Fixed a test failure in an example.

v0.5.0

Blobloom is now more robust to weak hash functions. The documentation contains more extensive advice for picking a hash function.

0.4.0

The new method Filter.Intersect computes the intersection of two equally-sized Bloom filters.

0.3.0

The methods Filter.{Add2,AddAtomic2,Has2} are gone. Use Filter.{Add,AddAtomic,Has} instead and combine pairs of 32-bit inputs h1, h2 into (h1<<32) | h2 if needed.

The Filter.Union method is now faster on amd64. Benchmarks show a twofold speedup.

0.2.1

The false positive rate estimation code was revamped and should now produce accurate results even for Bloom filters that are filled far beyond their capacity.