Closed vchuravy closed 9 months ago
@vchuravy I think that's precisely what I addressed with #318. And also the reason why I think BenchmarkTools.jl v2.0 is long overdue. But I would appreciate a discussion around it, cause that would mean a breaking release for a very central package.
@vchuravy can you confirm that the fix works for you?
closing since the example works on master
The docs state:
So looking at the example:
The docs also state:
So why is this a problem? PkgBenchmarks.jl calls
tune!
on every benchmarkable, thus changingevals
leading to false results when benchmarking mutating functions.Not sure who maintains PkgBenchmarks these days but cc: @KristofferC @shashi @DilumAluthge
Potential solution would be to add a new parameter
tuneable
that does indeed skips tuning in all instances.