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.
00:00 reportable6 left 00:11 TempIRCLogger left, TempIRCLogger joined 00:18 shareable6 joined, quotable6 joined, greppable6 joined 00:19 notable6 joined, linkable6 joined 00:20 benchable6 joined 01:20 releasable6 joined 01:22 frost left 02:08 ggoebel left 02:20 committable6 joined 02:35 drakonis joined 02:50 ggoebel joined 04:01 reportable6 joined 04:33 frost joined 06:00 reportable6 left 08:48 squashable6 left, squashable6 joined 09:05 ggoebel left 09:25 Techcable left, Techcable joined 10:01 reportable6 joined 11:01 reportable6 left, coverable6 left, bloatable6 left, nativecallable6 left, bisectable6 left, statisfiable6 left, releasable6 left, tellable6 left, linkable6 left, unicodable6 left, squashable6 left, evalable6 left, quotable6 left, notable6 left, benchable6 left, shareable6 left, sourceable6 left, greppable6 left, committable6 left 11:02 reportable6 joined, bloatable6 joined 11:03 greppable6 joined 11:04 quotable6 joined, benchable6 joined, bisectable6 joined 12:00 reportable6 left, reportable6 joined 12:02 squashable6 joined 12:03 releasable6 joined, linkable6 joined 12:04 nativecallable6 joined 12:15 Geth left, Geth joined 12:19 Geth left 12:20 Geth joined 12:37 ggoebel joined 13:02 notable6 joined 13:03 statisfiable6 joined 13:04 evalable6 joined 13:58 ggoebel left 13:59 ggoebel joined
MasterDuke [Coke]: what are the modules you use in your script that fails with mimalloc? 14:33
[Coke] some combination of File::Directory::Tree, File::Find, ile::Temp, HTTP::UserAgent, JSON::Tiny, Text::CSV 14:48
*File::Temp
14:55 frost left
MasterDuke may have found something in nativecall-related code. or might not be thinking straight, not sure how i could have missed it before 14:56
[Coke] Happy to test something if you throw it in a branch 14:59
pretty sure HTTP::UserAgent has some nativecall in there for SSL
15:04 unicodable6 joined, shareable6 joined
MasterDuke arg, now getting github errors when trying to push 15:04
well, here's the patch until github fixes itself pastebin.com/5pcuNjEe 15:08
timo do libraries use glibc malloc and try to free() something nativecall allocated for them with mimalloc-malloc? 15:48
16:01 committable6 joined 16:03 sourceable6 joined
MasterDuke yep 16:10
Geth MoarVM: MasterDuke17++ created pull request #1686:
Correctly allocate/free CStrs when using mimalloc
16:11
16:26 MasterDuke left 16:33 MasterDuke joined 17:03 coverable6 joined
lizmat MasterDuke: should that be in the 2022.03 release ? 17:39
[Coke] I need to test it on my windows issue, but that's probably not a blocker if it looks correct otherwise. 17:42
nine Yes, we want that fix 17:43
MasterDuke: why make the fix conditional on MVM_USE_MIMALLOC? I'd say that using MVM_malloc/MVM_free has never been correct there and just worked by virtue of those being wrappers for malloc/free. 17:44
18:00 reportable6 left
[Coke] MasterDuke: going through rebuild now, will report out in a bit. 18:58
19:03 linkable6 left 19:05 linkable6 joined
[Coke] MasterDuke: it's gotten *much* further using mimalloc+your path 19:24
Patch
nice find!
Yup, we're good 19:26
Nicholas "good" - is this like my COVID test result - "NOT DETECTED" ? :-/ 19:29
(unrelated to COVID-or-not, still rather too tired/busy to review things) 19:30
[Coke] yes, I can't reproduce my very difficult to golf error with his patch applied. 19:44
(oof, it's still running, up to 840MB so far...) 19:48
so this seems slower. 19:51
Finally did end, but got up to 1G of memory - which to be fair I'd not been checking previously. 19:52
20:01 reportable6 joined
MasterDuke nine: just to minimize the changes compared to before mimalloc. if we want to remove the use of MVM_* in these sorts of cases i think it'd make sense to do that as one larger commit across the different nativecall reprs 20:03
[Coke]: i'm kind of surprised if you find using mimalloc slower. pretty much everything i've tested has been faster with it, especially so on windows 20:04
japhb Perhaps [Coke] meant "slower memory growth"? 20:06
20:21 sena_kun left 20:23 sena_kun joined
[Coke] no, I meant the script seemed to be taking longer to run 20:30
MasterDuke that's unexpected, but i don't think my patch would cause any noticeable change. have you found using mimalloc generally slower? 20:32
but CI is green and it fixes your issue, so i'm going to merge in a couple minutes unless there are any objections 20:39
[Coke] MasterDuke: I haven't been *using* mimalloc because it was broken. 20:40
so I can't really answer that.
MasterDuke ah 20:41
[Coke] I run these scripts daily or so: If it's really slower, I'll try to do a comparison this weekend when I can set it to run and not worry about it.
lizmat MasterDuke: when merged, will you bump also ? 20:52
MasterDuke sure 20:53
Geth MoarVM: 764fff9b86 | (Daniel Green)++ | src/6model/reprs/CStr.c
Correctly allocate/free CStrs when using mimalloc
21:05
MoarVM: c935e681b8 | MasterDuke17++ (committed using GitHub Web editor) | src/6model/reprs/CStr.c
Merge pull request #1686 from MasterDuke17/more_nativecall_mimalloc_fixes
23:20 linkable6 left, linkable6 joined 23:52 ggoebel left