Welcome to the main channel on the development of MoarVM, a virtual machine for NQP and Rakudo (moarvm.org). This channel is being logged for historical purposes.
Set by lizmat on 24 May 2021.
00:29 japhb joined 01:16 japhb left 08:53 linkable6 left 09:28 japhb joined 10:06 nine left, nine joined 10:20 notable6 left, notable6 joined 10:49 shareable6 left, shareable6 joined 11:01 unicodable6 left 11:07 evalable6 left 12:53 shareable6 left
Geth MoarVM: MasterDuke17++ created pull request #1942:
Update Ubuntu images for CI
15:53
16:16 tellable6 left 16:24 releasable6 left 17:07 TheAthlete joined 17:28 TheAthlete left, notable6 left, notable6 joined 20:14 MasterDuke joined
MasterDuke . 20:14
github.com/MoarVM/MoarVM/pull/1942 is interesting. we could just not use ubuntu 22.04 for the clang+dynasm jobs, but that doesn't seem good 20:19
we could install a different clang
we could switch to libffi by default, or just remove dynasm completely 20:24
oops, not use *24.04* 20:26
i'm sort of surprised nobody ran into this 20:33
japhb Heh, I just commented something similar on the PR.
From my point of view if two different stable versions of clang are producing actual different results (not just an extra warning or some such), that would indicate we've got an undefined behavior bug somewhere in that code 20:34
MasterDuke or ubuntu 24.04 shipped with a buggy default version of clang 20:35
which isn't impossible, clang 18-20 or something like that severely regressed in computed goto performance 20:36
which is a problem for us and python 20:37
japhb Seems like it would have been patched by now, assuming those ubuntu images get package updates
Ouch (computed goto performance)
So you're saying the problem may not be a small bug, but a complete change in behavior for the stable series? Ugh. 20:38
MasterDuke well, i don't really know. just the one test failed... 20:40
[Coke]: unrelated, but i guess the mimalloc bump will have to be reverted, i haven't heard yet from my question about when they're going to do a next release 20:49
afk 20:50
japhb It's just one test sure, but that test is computing a completely different result than expected; it's not a matter of a performance dip, an extra warning, or even a compile error. It's just getting the wrong result, which is pretty much the worst of those behaviors. 20:51
20:54 MasterDuke left