Memory not freed when using Multi-threading

[felixhorger]

Sorry to spill the beans, but I found another issue with GC and multithreading.

Something similar was already flagged in #31923 and closed.

It is reproducible on my windows (11 Pro, Version 10.0.22621, Build 22621) with julia installed today from https://julialang.org/downloads/ using the winget cmd (nice btw :) ).

Julia Version 1.10.0
Commit 3120989f39 (2023-12-25 18:01 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: 8 × Intel(R) Core(TM) i7-8665U CPU @ 1.90GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, skylake)
  Threads: 5 on 8 virtual cores
Environment:
  JULIA_NUM_THREADS = 4
  JULIA_REVISE_POLL = 1

(I also tested this on my unofficial linux Debian build v1.9.4 if that helps, same problem)

Here is my MWE:

# Multi-threaded copy
f(a) = begin
        b = similar(a)
        Threads.@threads for i in eachindex(a)
                b[i] = a[i]
        end
        return b
end

a = zeros(300_000_000)
a = f(a)
GC.gc()
# This results in twice the memory, unexpected!?

a = nothing
GC.gc()
# no memory is freed

varinfo()
# Also check that occupied memory of a is zero

# Is memory just kept by julia/glibc/? for later use?
a = zeros(300_000_000)
GC.gc()
# Now three times the size of a is in memory

# What happens if f(a) is run again?
a = f(a)
GC.gc()
# Memory goes back to 2x size of a

# Now restart, and try the same without Threads.@threads in f()
# I find that memory behaves as expected, i.e. is freed.

Let me know if you need further info.
Thank you!

[vtjnash]

Duplicate of #40626

[kpamnany]

This should also be fixed by #57544 I think?