00:22 evalable6 joined 00:57 evalable6 joined 01:08 vendethiel- joined 01:55 ilbot3 joined 02:26 arnsholt joined 03:16 AlexDaniel joined 03:38 evalable6 joined, greppable6 joined, quotable6 joined, nativecallable6 joined, committable6 joined, bloatable6 joined, coverable6 joined, benchable6 joined, releasable6 joined, unicodable6 joined, bisectable6 joined, squashable6 joined, statisfiable6 joined 04:20 nativecallable6 joined, bloatable6 joined, coverable6 joined, quotable6 joined, greppable6 joined, benchable6 joined, unicodable6 joined, bisectable6 joined, evalable6 joined, committable6 joined, releasable6 joined, squashable6 joined, statisfiable6 joined 04:53 quotable6 joined, nativecallable6 joined, benchable6 joined, greppable6 joined, coverable6 joined, bloatable6 joined, evalable6 joined, committable6 joined, bisectable6 joined, unicodable6 joined, releasable6 joined, squashable6 joined 04:54 statisfiable6 joined 05:00 Ven`` joined 05:25 AlexDaniel joined 05:55 brrt joined 05:58 domidumont joined 06:04 bisectable6 joined 06:09 bisectable6 joined 06:13 bisectable6 joined
nwc10 good *, jnthn 06:18
t/spec/S17-supply/syntax.t does seem to complete (is this an off-topic-for-this-channel observation?)
It just takes some time
Files=1213, Tests=75446, 21022 wallclock secs (29.89 usr 6.44 sys + 84927.03 cusr 17246.28 csys = 102209.64 CPU)
06:22 brrt1 joined 06:26 brrt joined
brrt good * #moarvm 06:27
nwc10 good *, brrt
samcv good * brrt, nwc10 06:30
brrt good hi nwc10 samcv 06:31
.tell nine i've been meaning to make all JIT errors nonfatal 06:32
yoleaux brrt: I'll pass your message to nine.
brrt because if they're due to memory corruption, we'll just crash anyhow, and if they're due to unexpected input, that might mean some assumption was invalidated, and a 'graceful degradation' without JIT is better than crashing because no JIT 06:33
Geth MoarVM: 394cdd260f | (Samantha McVey)++ | src/strings/ops.c
Used FSA allocator for jump table in KMP index algorithm

We can save cpu cycles by using FSA, since we just end up freeing it as soon as the function returns. Also, since the loop is duplicated, turn it into a #define so we can use the same code with different functions for accessing the Haystack.
MoarVM: c68ca49655 | (Samantha McVey)++ | 4 files
Download all emoji files and generate UCD for emoji 4+5

Since emoji names are not guaranteed to be unchanged, we now download past emoji versions as well to ensure we don't break compatibility in the future. This isn't expected to cause any issues, just a few extra entries. For now the only thing that may have changed is "man in business suit levitating: skin tone" is now "man in suit levitating: skin tone".
samcv that was annoying. regarding the emoji changing names, but it's fixed now we won't have to worry about it when regenerating 06:37
06:37 domidumont joined 07:00 brrt joined 07:04 zakharyas joined 08:01 bloatable6 joined, quotable6 joined, benchable6 joined, committable6 joined, nativecallable6 joined, releasable6 joined, coverable6 joined, evalable6 joined, greppable6 joined, bisectable6 joined, unicodable6 joined, squashable6 joined, statisfiable6 joined 08:06 leont joined 08:07 brrt joined 08:17 robertle joined 08:39 eater left
lizmat jnthn: this was just on #perl6: <+yoleaux>08:27Z <sena_kun> lizmat: hi! there is a rakudo regression for cro, that gives me "Serialization Error: missing static code ref for closure '' (src/Perl6/World.nqp:2217)". I am not sure how bad it is, but I hope it is not. ping me for a way to reproduce(or better .tell) if it's obscure enough. thanks. 08:43
do you see that also? 08:44
08:52 brrt joined 09:07 evalable6 joined 09:11 zakharyas joined
jnthn lizmat: It started showing up in our Travis, for master only 09:30
(we build against Rakudo 2017.08, 2017.09, and master...only master fails this way)
lizmat so, is it something I did with the BUILDALL refactor?
jnthn lizmat: It feels like the most likely candidate 09:31
I didn't dig into it myself yet
lizmat and this is occurring with HEAD in rakudo? 09:32
I removed the $/ hack, and that fixed it for dogbert
jnthn Apparently so
lizmat :-(
09:45 evalable6 joined
timotimo aye, get it locally too :| 09:58
10:05 evalable6 joined
timotimo so the code it points at is the $compiler_thunk inside fininsh_code_object 10:05
i'm not sure at all how all of this works :) 10:06
i didn't even know code has a @compstuff attribute 10:07
interesting. i made the buildplan compiling stuff disabled and i still get the error in question 10:15
timotimo bisects a little bit 10:16
lizmat ok, I found the cause of the failure in S32-str/sprintf.t 10:21
somehow we wind up with a BOOTInt and a BOOTNum in the attributes of X::Str::Sprintf::Directives::Count 10:22
timotimo incriminatingly checking out the commit right before the very beginning of the buildall work makes it work again 10:30
now building the commit before $/ was added to the compiler services 10:31
10:39 lizmat joined
timotimo the commit that adds $/ gets me a different error, which is the missing serialization thingie for MVMContext 10:41
ah, i didn't pay enough attention
yeah, removing the $/ gets me back the missing code ref problem 10:50
10:50 bartolin joined
timotimo i'd have to modify all commits in between to not use $/ in the compilation services 10:50
makes me wonder, though. what does Cro::HTTP do that rakudo and spectests don't 10:53
bartolin a tiny bit off-topic, but since i've just spotted the discussion about a serialization error: since commit 9837687d93 the jvm build fails during stage jast with a similiar error: "Serialization Error: could not locate static code ref for closure" 10:56
lizmat hmmm... the block is added to the SC, but maybe the method isn't ?? 11:01
11:06 evalable6 joined
timotimo i'm not really familiar with the ins and outs of serialization problem debugging 11:10
lizmat neither :-( 11:12
timotimo i'm glad that your buildall work has gotten so much further than mine :) 11:15
lizmat too 11:16
it was quite a steep learning curve
but I think I learned a lot from it
11:47 dalek joined, SourceBaby joined 11:56 domidumont joined
MasterDuke timotimo: btw, by backing out a bunch of the changes i made (and then rediscovering that problem with not supporting FSA_SIZE_DEBUG in realloc anymore) my branch now dies in stage parse 11:59
timotimo which changes were you backing out? i assume the ones that use FSA for the chars buffers? 12:00
MasterDuke yeah, pretty much everything in decode_stream.c and Decoder.c 12:01
timotimo right
lizmat jnthn: re not having a compiler at runtime: getting one would be as easy as nqp::getcomp('perl6') no? 12:04
timotimo the nativecall stuff already did it like that 12:05
but you don't have the compilation services
brrt by the way, do we have something pack-like in perl6 12:07
lizmat nqp::p6bindattrinvres(nqp::create(Rakudo::Compiler::Services),Rakudo::Compiler::Services,'$!compiler'),nqp::getcomp('perl6') ???
brrt by the by the way, i honestly don't care about the language name discussion, but can't we just AB test it 12:08
lizmat brrt: there's experimental pack/unpack, and there's a module PackUnpack in the ecosystem
which badly needs some love
brrt i don't like not having a consistent name for the language
aha, got it
lizmat timotimo: I think the main problem is not having a $*W at runtime :-) 12:09
brrt do we want to improve on pack, i'm thinking
MasterDuke timotimo: btw, would i be correct in assuming that don't need to use the MVM_fixed_size_*_at_safepoint functions for these MVMString changes, since they aren't being modified concurrently (unlike the VMArray change)?
timotimo yeah, that should work without a safepoint 12:10
it should always be safe to free stuff inside of gc_free 12:11
MasterDuke brrt: isn't the AB test kind of happening right now with the inconsistent naming? whichever ends up getting used the most wins?
brrt there's no metrics to it though
i'm going to have to squint pretty hard before i'm going to see 'putting random stuff out there' as a controlled experiment 12:12
MasterDuke timotimo: if you're curious, this has the backtrack i'm seeing right now gist.github.com/MasterDuke17/f5796...38b0b23750 12:14
oh, and my branch has been force pushed to a bunch
brrt that came out harsher than i meant it
if i'm being realistic though, the whole fact that we have the discussion means that perl6 is losing, or has lost, some power as a 'rallying flag' 12:15
which is probably due to the continuing bad reputation of perl in the wider developer community
MasterDuke brrt: i'd say you're correct though, it hasn't been controlled in any way
lizmat jnthn: isn't this a race condition? 12:23
nqp::setmethcacheauth($obj, 0);
%!cache := {};
12:32 domidumont joined
timotimo MasterDuke: merge my FSA valgrind error support branch and try again under valgrind 12:39
MasterDuke heh, just started it (gonna take a while to run) and commented on your PR 12:40
started it with your branch merged
timotimo cool, lemme look 12:41
ah good point
Geth MoarVM/fsa_valgrind_error_support: 235f6147dd | (Timo Paulssen)++ | src/core/fixedsizealloc.c
use %lu instead of %d, put in missing output in one case

MasterDuke ugh. stage parse in valgrind is slow... 12:48
and pretty sure i complied with --optimize=0, that's not gonna help 12:49
timotimo hah 12:51
MasterDuke timotimo: no extra output from valgrind 13:01
timotimo oh
did you configure with --valgrind? 13:02
MasterDuke arg, just saw that
gonna recompile
timotimo that step is fast at least :) 13:03
MasterDuke yup, stage parse running again 13:04
timotimo++ gist updated 13:15
timotimo ok, so the normalizer uses malloc 13:16
this is really helpful, isn't it! <3
MasterDuke yep!
timotimo now i just need to find a good place to put "search for things pointing at this" into moarvm :) 13:23
bbl 13:35
13:37 leedo joined 13:41 domidumont joined 14:00 domidumont joined 15:04 brrt joined 15:24 brrt1 joined 15:25 leont joined 15:57 leont joined 16:09 committable6 joined
MasterDuke timotimo: wow, `make m-test` actually passes, though `make m-spectest` has many many segvs 16:33
timotimo well, it's improvement!
16:47 leont joined 17:02 AlexDaniel joined 17:23 domidumont joined 17:39 robertle joined 17:42 AlexDaniel joined
Geth MoarVM: coypoop++ created pull request #715:
Correct typo in comment
MoarVM: fc4285ab87 | coypoop++ (committed by timo) | src/profiler/instrument.c
Correct typo in comment
MasterDuke timotimo: gist updated with another valgrind log, but i'm not following what's happening. afk for a bit, but any comments appreciated 19:10
timotimo MasterDuke: the normalizer code that you use the fsa for there doesn't fill up the buffer completely, so the number it passes as num_graphs in the string is smaller than the actual number of graphemes in the buffer, so later the number passed for freeing is wrong 20:08
you'd probably best realloc at the end
20:23 evalable6 joined 20:37 brrt joined 20:55 zakharyas joined 21:24 quotable6 joined, unicodable6 joined, bisectable6 joined, benchable6 joined, bloatable6 joined, evalable6 joined, greppable6 joined, nativecallable6 joined, releasable6 joined, coverable6 joined, squashable6 joined, statisfiable6 joined 21:39 evalable6 joined
lizmat and another Perl 6 Weekly hits the Net: p6weekly.wordpress.com/2017/10/02/...e-granted/ 22:30
22:35 bisectable6 joined 22:54 bisectable6 joined
timotimo nice, lizmat++ 22:57
23:04 bisectable6 joined, quotable6 joined, nativecallable6 joined, evalable6 joined, bloatable6 joined, greppable6 joined, releasable6 joined, benchable6 joined, bisectable6 joined, coverable6 joined, unicodable6 joined, committable6 joined, squashable6 joined, statisfiable6 joined 23:50 evalable6 joined