BenchmarkTools.jl
BenchmarkTools.jl copied to clipboard
Trial display assumes the max GC comes from the sample with the max time
If I followed the breadcrumbs here right, the displayed max GC time comes from the maximum as computed here:
https://github.com/JuliaCI/BenchmarkTools.jl/blob/7eebf01f7b9e469ee41e83e38170857fb33a3da3/src/trials.jl#L116-L119
Which isn't really the maximum GC time. Since it's displayed independently, this can be surprising.
Cf. https://discourse.julialang.org/t/time-and-benchmark-give-different-results/66374/3
In BenchmarkTools GC time is always printed as a ratio. That's why we do gcratio(maximum(trial))
.
So yes the fact that we say the "maximum GC" is 0.0%
is confusing
See discussion in #250 as well.