MasterDuke17

Any difference running with `MVM_(JIT|SPESH)_DISABLE=1`?

> Hm, is it me or does only the fast path care about `lengthu` and the one with the codepoint iterator ignores it? I just tested with this further change:...

> @MasterDuke17 Indeed, because the units of `lengthu` are graphemes, but we iterate codepoints. It seems that actually none of the other encoders really pay attention to `length` either, though,...

@nwc10 it looks like the fails are all on Windows (MSVC + MinGW). I think the only Windows-specific bit is https://github.com/MoarVM/MoarVM/pull/1685/files#diff-e4b5afce4611608fd85a5a5579c1a585e34a2b43573725201841976980f84f6dR9-R17, any idea why we'd have a problem with it?

Here's where heaptrack says the memory is being allocated (for 30k so it wouldn't get killed). ``` Welcome to Rakudo™ v2021.12-118-g888f05d25. Implementing the Raku® Programming Language v6.d. Built on MoarVM...

I guess mimalloc doesn't support 10.6. Until we check for that when running Configure.pl, you can just build with `--no-mimalloc` (i.e, `perl Configure.pl --no-mimalloc` for MoarVM, if you're having Rakudo...

Then it looks like MoarVM won't build on 10.6 without some work done. 10.7 will probably be fine if you can upgrade.

@dogbert17 got a different backtrace with ASAN https://gist.github.com/dogbert17/b318650ee80576d79f54b22d88bf3b7b

I did an strace of `make -j12` and the only syscall that's on the blocked list linked above is `clone` (called 370 times (same with `make -j1`)).

Try adding `--full-cleanup`, which will hopefully get rid of those leaks.