On Friday, August 1, 2025 at 4:39:17 PM UTC-4 Pierre Durand wrote: I think it's related to the fact that b.Loop() disables some optimizations, but I'm not sure.
Pierre, The difference between b.N and b.Loop is a known issue. It is by design, https://groups.google.com/g/golang-nuts/c/Z017c0f59vY/m/AY1-wT0OBAAJ Peter Le vendredi 1 août 2025 à 22:33:56 UTC+2, Pierre Durand a écrit : I can reproduce the issue with this code https://github.com/pierrre/go-libs/blob/31bec3f12a86382924cc0b486c2008db062a14bd/runtimeutil/runtimeutil_test.go#L101-L118 Sorry I can't write a smaller code snippet, because I think it's related to escape analysis, and I don't fully understand it. As you can see, I'm benchmarking the same function WriteFrames() with the old and the new benchmarking methods. When I run the benchmark, here is the result go test -v -run=^$ -bench="^BenchmarkWriteFrames" -benchmem ./runtimeutil goos: linux goarch: amd64 pkg: github.com/pierrre/go-libs/runtimeutil cpu: Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz BenchmarkWriteFrames BenchmarkWriteFrames-12 366825 3091 ns/op 0 B/op 0 allocs/op BenchmarkWriteFramesNew BenchmarkWriteFramesNew-12 328620 3445 ns/op 80 B/op 4 allocs/op PASS ok github.com/pierrre/go-libs/runtimeutil 2.303s Le ven. 1 août 2025 à 22:04, Junyang Shao <shaoj...@google.com> a écrit : > > Hello Pierre, > > Thank you for bringing up this issue. > > May you share the code snippet that triggers this behavior? Thanks. > > On Fri, Aug 1, 2025 at 12:50 PM Pierre Durand <pierre...@gmail.com> wrote: >> >> Hello >> >> I noticed a weird behavior when I'm benchmarking with testing.B.Loop() code that uses iterators . >> The benchmark shows allocations where I start to iterate the iterator, and where I declare variables (before the loop) that are used inside the iterator loop. >> I know that my code is not doing any allocation, so it's strange. >> If I change my benchmark to use the old "range b.N", then it doesn't show this strange behavior. >> If I check allocations with "testing.AllocsPerRun", I don't see any allocation. >> >> Is that a know issue ? Should I open a bug ? >> What should I do ? Use the old benchmarking method ? >> >> Regards >> >> -- >> You received this message because you are subscribed to the Google Groups "golang-nuts" group. >> To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts...@googlegroups.com. >> To view this discussion visit https://groups.google.com/d/msgid/golang-nuts/70456dc1-3380-40fd-951d-e52275bc48a5n%40googlegroups.com . > > > > -- > Thanks, > Junyang -- Pierre Durand -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. To view this discussion visit https://groups.google.com/d/msgid/golang-nuts/f6758be0-7ded-4e08-8854-a2c9dcd385f4n%40googlegroups.com.
