NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
Go modules, #2977 by repository stars
Latest stable
v1.2.2
Last release 16 days ago
30 Oct 2026
Ships unpredictably
gaps range from 8 days to 3.3 years
Nearly every release is documented
notes for 12 of 13 stable releases
Nothing withdrawn
no release was ever pulled
14 years old
47 releases · first in 2013
Drag across the ruler to compare, or tap a release to open it.
Since v1.2.0, the one shot sums (Sum32, Sum64, Sum128, and their seeded and string versions) moved the input to the heap when called from another pack
Since v1.2.0, the one shot sums (Sum32, Sum64, Sum128, and their seeded
and string versions) moved the input to the heap when called from
another package, so hashing a stack buffer allocated; reported by
@ncruces in #15. The input now stays on the stack, and summing a stack
buffer is about twice as fast. The []byte sums cost up to 0.3ns more
per call.
A hasher from New32, New64, or New128 created and used in one function
now stays on the stack, and so does a buffer passed to its Write.
New64 and SeedNew64 now return 8 from Size, the length of their Sum,
rather than 16. Every hash value is unchanged, but an HMAC over New64
now reports a Size of 8, so an HKDF over New64 with a nil salt derives
a different key.
Streaming 64 and 128 bit writes now mix four blocks per loop iteration, from @antmat in #14 . Large chunked inputs hash about 7 percent faster; writes
Your coding agent can read these notes before it upgrades. Set up the MCP server →
This removes amd64 assembly for 64 bit and 128 bit, similar to how assembly for 32 bit was removed a few releases ago; it's no longer that impactful,
This removes amd64 assembly for 64 bit and 128 bit, similar to how
assembly for 32 bit was removed a few releases ago; it's no longer that
impactful, especially on Go 1.27. See the README's benchmark section for
a comparison of asm, Go 1.26, and Go 1.27.
This also adds support for hash.Cloner because why not.
The module indicates a minimum of 1.26 even though 1.25 is the technical
minimum for hash.Cloner - benchmarks show the 1.25 compiler is way worse
for pure Go code, and 1.26 is the minimum supported Go version right now
anyway, so it's worth bumping.
Apparently gccgo was not liking the Go assembly, so we just add a build tag to default to pure Go in that case.
Apparently gccgo was not liking the Go assembly, so we just add a build
tag to default to pure Go in that case.
Nothing published for this version
Unsafe slice to string conversation apparently could cause an allocation -- it was in tests, but only when converting a slice that was actually an 819
Unsafe slice to string conversation apparently could cause an allocation
-- it was in tests, but only when converting a slice that was actually
an 8192 byte array in the benchmark itself.
We can just duplicate the logic to drop the allocation.
This was tested by deleting the 128 bit assembly functions, and as it
turns out, hand rolled assembly is still a bit faster.
Nothing published for this version
Thank you @aykevl for making this more portable.
Thank you @aykevl for making this more portable.
Nothing published for this version
Nothing published for this version
v1.1.5: addresses issue pointed out in golang/go#40701
v1.1.5: addresses issue pointed out in golang/go#40701
Switches from BP to R10
Switches from BP to R10
non-amd64 had a missing import, which was a silly mistake. I've tested with and without the amd64 specializing for this release.
non-amd64 had a missing import, which was a silly mistake. I've tested
with and without the amd64 specializing for this release.
Following golang/go#19367, this re-introduces
reflect at the expense of a 0.4ns hit to ensure that the code follows
all unsafe rules.
Nothing published for this version
v1.1.2: minor performance enhancement and removal of reflect
v1.1.2: minor performance enhancement and removal of reflect
Should be a minor perf bump compared to 1.1.0 for people that use the full New128 hasher, which is uncommon.
Should be a minor perf bump compared to 1.1.0 for people that use the
full New128 hasher, which is uncommon.
Nothing published for this version
Has fixes around the unsafe code that -d=checkptr caught in 1.14, and then code changes to regain speed lost from those fixes.
Has fixes around the unsafe code that -d=checkptr caught in 1.14,
and then code changes to regain speed lost from those fixes.
The speed changes are largely a wash, if only a little bit slower.
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
One column per quarter.