Memory not released or resused under Linux

[freemin7]

When running @btime length($rand(100000000)) repeatedly Julia allocates around 0.8G of memory (almost) each time and doesn't release it even if the system starts swapping. Unlike here the memory is not reused for the next call. #29085 might be related but that doesn't involve interpolation ($ stuff). The issue does not exist for @time length(rand(100000000))

I could reproduce the issue on x86_64 and AARCH 64 in 1.6.1 and 1.7.0-rc2 on two Linux variants. The repeated allocation of 0.8 Gigabytes does not happen under Windows and a GC.gc() call frees the 0.8G too which doesn't happen under Linux.

@btime length(x) setup=(x = rand(100000000)) on Linux behaves nicely under Linux including GC.gc() working.

[KristofferC]

An interpolated value in BenchmarkTools will never be released (based in how BenchmarkTools uses it) . I'm pretty sure there is an issue about this sonmewhere in the BenchmarkTools repo but can't find it now.

[freemin7]

I didn't find anything like that either.

[KristofferC]

It was in a PR JuliaCI/BenchmarkTools.jl#247. Anyway this is not a julia issue but a BenchmarkTools one.