MasterDuke back while pork chops bake. so think this is worth PRing after i clean it up (remove debugging, squash commits, etc.)? 00:07
jnthn github.com/MoarVM/MoarVM/compare/m...dbc21b3R84 # should it not set ssize here also so it knows the size to free? 00:37
MasterDuke: Overall, I think it's worth it. 00:41
MasterDuke jnthn: cool, then i'll work on cleaning it up 00:54
jnthn MasterDuke++ 00:57
'night
timotimo MasterDuke: i assume all these $_ ~ rand.Str strings end up being strands-based? it'd probably be interesting to give jnthn's suggestion a shot to use fsa for the strands arrays 00:58
MasterDuke timotimo: i think i do that in my branch already? github.com/MoarVM/MoarVM/compare/m...9bb347cR47 00:59
timotimo oh, ok! 01:00
02:56 ilbot3 joined 07:02 domidumont joined 07:07 domidumont joined 07:17 geospeck joined 07:53 brrt joined 08:07 brrt joined
nwc10 good *, #moarvm 08:17
09:04 brrt joined
brrt good * #moarvm 09:05
09:07 domidumont joined 09:57 zakharyas joined 10:02 zakharyas joined 10:06 lizmat joined 11:09 AlexDaniel joined
MasterDuke ugh, my branch builds nqp and passes its `make m-test`, and builds rakudo and passes its `make m-test`, but i get a bunch of 'realloc: invalid pointer' and 'free: invalid pointer' during `make m-spectest` 11:18
11:24 brrt joined
MasterDuke oh, i think it's because i originally branched this from my branch that uses the FSA in VMArray 11:37
but i've now rebased it to master
hm, the MVMROOT* macros can be annoying, valgrind points to a line with MVMROOT3, but i assume it's actually something inside 13:44
jnthn Try building with --debug=3
If you didn't already
MasterDuke i have --debug=3 and --valgrind
jnthn Ah
Then, I guess a smarter C compiler is wanted :P 13:45
MasterDuke gcc (Ubuntu 7.2.0-8ubuntu3) 7.2.0
jnthn I've actually no idea what different versions do in this regard, tbh 13:46
I've certainly seen this issue
Oh, it's valgrind doing the pointing
And I guess it's in gdb where --debug=3 helps
I think there's a way to do the valgrind run under gdb and breakpoint its reporting, though 13:47
MasterDuke i guess if i had to i could also convert to manual push/pop for debugging, but i'd really rather not 13:49
timotimo i think c compilers say "macro expansions being considered a single line is a feature!" 13:51
MasterDuke oh, btw, have you had a chance to look at github.com/MoarVM/MoarVM/pull/689 ? (but don't let me distract you from segv hunting)
timotimo so, uhm, time for a perl6 script that rewrites MVMROOT macros for us :)
14:02 zakharyas joined 14:03 zakharyas joined 14:05 zakharyas1 joined
MasterDuke ah, i'm trying to fixed_size_free a string with 0 num_graphs or num_strands 14:11
timotimo ah, i know that problem from somewhere 14:12
MasterDuke fwiw it's num_graphs in this instance 14:13
the line is: MVM_fixed_size_free(tc, tc->instance->fsa, (str->body.storage_type == MVM_STRING_STRAND ? str->body.num_strands : str->body.num_graphs) * type , str->body.storage.any); 14:15
timotimo i wonder how often we generate "the empty string" 14:16
because why else would it try to free something like that? unless we created an empty string somewhere that turned into garbage
MasterDuke it's happening during in/because of MVM_string_split 14:17
hm, changing the 0 to a 1 doesn't fix anything. how does one MVM_fixed_size_free a 0 sized thing? 14:23
jnthn You don't
You can't allocate on in the first place
*one
iirc, anyway
MasterDuke "can't" == don't do that, or "can't" == shouldn't be possible ? 14:24
timotimo malloc has some secret sauce for zero-sized allocations that allows you to free them
jnthn "Don't do that" 14:26
Though when MVM_FSA_DEBUG is set to 1, we could put in sanity checks for that
MasterDuke yep, a couple requests for 0 bytes showed up in MVM_fixed_size_alloc 14:30
ok, at least one is from MVM_string_flip 14:33
rbuffer = MVM_fixed_size_alloc(tc, tc->instance->fsa, sizeof(MVMGrapheme32) * sgraphs);
what's the right behavior? just do `res = (MVMString *)MVM_repr_alloc_init(tc, tc->instance->VMString);`, but don't set any of its fields (e.g., don't do `res->body.storage.blob_32 = rbuffer;`)? 14:38
ilmari timotimo: some mallocs do, but malloc(0) is also allowed to return NULL (and free(NULL) is allowed as well) 14:41
jnthn MasterDuke: That or just detect the case and use the empty_string constant hanging off tc->instance 14:42
MasterDuke ah ha, i thought there was something like that but couldn't come up with the right name 14:43
14:48 lizmat joined
MasterDuke ugh, now i'm not seeing any 0 byte requests in MVM_fixed_size_alloc, but still getting the 0 bytes free 14:50
none in fixed_size_realloc either... 15:00
oh, it's not just 0 bytes to free, to_free is NULL also 15:03
heh, running under valgrind with --debug=3 and --optimize=0 is not fast 15:36
15:42 zakharyas joined 15:56 geospeck joined 18:05 domidumont joined 18:06 brrt joined 18:29 lizmat joined 18:42 committable6 joined 18:44 brrt joined 19:49 quotable6 joined, coverable6 joined, greppable6 joined, bloatable6 joined, nativecallable6 joined, tangible6 joined, benchable6 joined, statisfiable6 joined
dogbert17 hmm, quite silent in here 20:47
nwc10 good *, dogbert17! 20:49
dogbert17 any deadlock experts around? Does the last comment in github.com/rakudo/rakudo/issues/1202 give any clues as what might be going on?
nwc10 sorry, I'm not a deadlock expert
(and I don't know how to flush one out)
dogbert17 good evening nwc10, found any ASAN problems recently?
nwc10 evening! heresey!
(and quite accurate, as I'm likely to go to bed soon) 20:50
no, sadly/usefully not
code quality is too good
the right sort of problem to have
dogbert17 yes indeed
finding bugs in MoarVM is getting way to hard 20:51
21:10 zakharyas joined 21:18 lizmat joined
Geth MoarVM/jit-expr-optimizer: 11 commits pushed by (Bart Wiegmans)++
review: github.com/MoarVM/MoarVM/compare/a...60e747882a		 23:00
23:25 evalable6 joined

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!