Diogo Netto
Diogo Netto
> As mentioned above, promotion will happen between full collection and the marking pass has to discover any new old->new generated because of this. I can only see old->young references...
Superseded by https://github.com/JuliaLang/julia/pull/47292.
I think it's not ready to merge/review yet. It's still causing regressions on some of the GCBenchmarks and I'm diagnosing that (will post some scaling plots in a sec). I...
Speedup/slowdowns on GC times relative to the master commit from which it branched vs nthreads (1, 2, 4, 8, 16):   
Superseded by https://github.com/JuliaLang/julia/pull/48600.
Segfault [here](https://buildkite.com/julialang/julia-master/builds/33747#018daeb2-95b3-457f-b62b-7284000d893e) seems related to the PR.
We maintain a side byte-map that holds information about the state of the page (GC_PAGE_ALLOCATED , GC_PAGE_LAZILY_FREED, etc). If page_metadata is returning NULL, it likely means that the byte in...
The part which removes the fast path in which we check `if (n_threads_marking == 0)` outside of the lock is relevant to 1.10. We should backport it.
`Serialization` doesn't seem happy with this PR...