01:20 FROGGS_ joined 05:57 avuserow joined 06:24 avuserow joined 06:55 lizmat_ joined 06:57 woolfy joined
sergot o/ 07:11
FROGGS o/ 07:33
07:39 woolfy1 joined 07:54 kjs_ joined 08:04 brrt joined
brrt ok, interesting bug by tgt 08:06
it's not sensitive to JIT-ting, so that is a thing
timotimo something about handlers not getting merged properly perhaps? 08:08
brrt ehm.... d'you think? 08:13
i think it has something to do with the magic $_ lexical 08:14
(duh)
i mean to say
i don't know how default {} is actually implemented
timotimo --target=ast and --target=optimize are your friends 08:26
timotimo will be afk for a long bit
brrt oh, yes 08:27
and the spesh log of course :-)
in other, better news, /me finally had an epiphany on how to use the SSA form to aid with register assignment and allocation 08:32
also known as 'why phi nodes are cool'
timotimo sounds good :) 08:49
brrt basically, phi nodes form the constraints from which values can come? i think that's the way i'd like to put it, and i'm not sure that i'm correct about it 08:52
but that means that if you can ensure that both the 'sides' of the phi node use the same register for the same value, you can keep the value in a CPU register 08:53
so that seems like a tractable problem 08:54
has jnthn landed ok, btw?
lismat I think jnthn should be somewhere on the ground now 08:58
his flight left from Sofia 14 hours ago or so
brrt wonders if there is a way we could set spesh logging only for the actual script and not for the compiler 08:59
hmm yeah
lismat even for a lexical scope would be nice
would help in the profiler as well
brrt did you know he had to fly with an actual propeller plane for the last pace of his journey?
lismat wouldn't be surprised.... turboprops are very efficient, though loud 09:00
brrt hmm yeah, but i'm not about to design a DSL for logging options on a monday morning :-)
hmm
i hadn't thought of them as efficient
i thought of them as 'scary'
lismat in the late 80's I used to take a turboprop for the last 200 km from Chicago to Champaign Urbane
brrt you're still in sofia, though, right? 09:01
lismat most of those planes can fly *without* engines
yup, still in Sofia
plane leaves in ~ 18 hours
not sure how many of those we'll be sleeping
brrt or should i say Š”ŠžŠ¤Š˜Š 09:02
:-)
lismat we're with 13 people now in our room... :-)
brrt 13 people? all perl folks?
timotimo brrt: maybe we could introduce ops like speshlogenter and speshlogexit and have them control when spesh should start loggin and when it should stop
(well, the stopping part is probably not that important) 09:03
that way we'll cut a good chunk of the startup specialization
brrt hmm
yeah, i'd still suspect some times you want those logging
timotimo in that case you just don't supply the flag that turns that on
it'd be similar to --profile
brrt virtualbox y u always one release behind the kernel
hmm 09:04
anyway, not designing a DSL for logging on a monday morning just yet
just thinking it'd be cool
timotimo yeah :)
it'd be nice if the spesh log would output at least the first line number annotation 09:05
so that corellating stuff to code would be easier if you have a big pile of speshlog to search through and you get 100 different '' named entries :)
brrt that is not so simple to acquire fwiw
brrt nods
anyway, afk for a bit
timotimo OK 09:06
09:06 brrt left 09:19 kjs_ joined
lismat tell brrt yes, all perl folks 09:23
having a programmatic spesh on/off switch could also be useful once we have official releases 09:25
and it turns out an official release has a bug, and the only way around it would be to switch off spesh
but then the program would run a lot slower
being able to switch spesh off selectively could then be a good workaround for some class of problems
09:28 zakharyas joined
timotimo we were actually talking only about spesh log output so far 09:28
09:40 kjs_ joined 09:45 woolfy joined 09:52 zakharyas1 joined 09:54 zakharyas joined 10:02 ggoebel11111117 joined 10:18 kjs_ left 10:26 Ven joined 10:55 kjs_ joined
kjs_ 6 10:55
11:08 woolfy joined 11:27 Ven joined 12:15 Ven joined 12:28 kjs_ joined 12:31 jnap joined 12:36 lizmat joined 13:20 mj41 joined
mj41 Hi. Very nicely done www.brendangregg.com/Perf/linux_obs..._tools.png ( www.brendangregg.com/linuxperf.html ). 13:23
cognome I don't know if there exists the same for modern versions of Solaris. 13:24
14:24 woosley joined 14:30 flussenc1 joined 14:31 Juerd_ joined, lee__ joined 14:32 bcode joined 14:35 flussence joined, synopsebot joined
[Coke] ... I should ask the guy I know who works on oracle/sun HA stuff if we can get a test login somewhere. :| 14:37
14:49 bcode_ joined 14:51 synopsebot joined 15:17 brrt joined
brrt timotimo: i'm fine with extending programmatic control to speshing 15:19
but... i'm wondering about how to do that exactly 15:22
i'm not sure if perl6 has some standard interface to the compiler
(it should have, if you ask me. python has it.)
lizmat: how'd your talk go? 15:24
same question to jnthn, if he's here
nwc10 hopefully jnthn is in bed, because that would mean that he's finished travelling 15:26
brrt we could also implement a not-speshed trait, i think, which would default to nop on other platforms than moarvm 15:27
yay, new wingolog: wingolog.org/archives/2014/08/25/re...n-in-guile 16:30
16:30 mj41 joined 16:50 zakharyas joined
brrt i'm all alone here and not sufficiently motivated to do real bugfixing :-( 16:53
lizmat hi brrt 16:54
brrt hi brrt :-)
how'd your talk go
lizmat back again from errands and afk and stuff
brrt hi lizmat
oh i'm dumb
lizmat my talk was *way* too shirt
*short rather
so it also was a Q&A for about 20 mins 16:55
and then I let them go for an early lunch
brrt :-)
any idea when the videos will be posted?
lizmat matter of weeks 16:56
brrt i see
did you get to see jnthn's talk?
lizmat yes... 16:57
brrt how was it?
lizmat it was very nice to see it making thumbnails and uploading them all at the same time
brrt oh, yes, i recall now
lizmat although I feel we need a better way for throttling work load
brrt hmm
lizmat perhaps some sugar... 16:58
brrt is the code online somewhere that i could see?
lizmat the slides are not posted yet, afaik
brrt ugh, github has changed again
16:59 tgt joined
brrt i'm with timotimo basically that the spesh log could really use file and line number annotations 17:00
so that will be my first project tonight
lizmat ++brrt 17:01
afk mostly again&
brrt yeah, dinner & 17:02
have fun in sofia :-)
17:03 brrt left 17:05 lizmat_ joined 17:46 brrt joined
timotimo brrt: jnthn actually posted the slides: jnthn.net/papers/2014-yapceu-async.pdf 17:46
brrt ooh very nice 17:49
17:52 Ven joined
cognome jnthn++ 18:05
18:11 Ven joined 18:12 zakharyas joined
[Coke] running under a high load on OS X, t/spec/S03-operators/subscript-adverbs.t died with no TAP output 18:18
runs fine standalone. 18:19
brrt hmm? that's weird 18:29
maybe malloc failed and we died
i can imagine that happening
lizmat suspects .WHICH breakage 18:31
brrt hmm what's .WHICH ? 18:34
oh
i know
lizmat hehe... that what is now partially an address... 18:36
18:45 kjs_ joined
brrt oh, that thingy 18:47
i recall that discussion
that is ... a pain
there may be some nice ways arround it, though
brrt bbiab 18:48
18:49 zakharyas1 joined 19:10 kjs_ joined 19:14 woolfy joined 19:41 Ven joined 19:52 kjs_ joined 20:01 woolfy left 20:08 kjs__ joined, zakharyas joined 20:10 brrt joined 20:17 bcode joined
dalek arVM: a246bbd | (Bart Wiegmans)++ | src/spesh/dump.c:
Add filename to spesh log

This allows us to correlate spesh frames with source code hopefully a bit easier		 20:49
arVM: 2966c01 | (Bart Wiegmans)++ | src/spesh/dump.c:
Don't leak memory when dumping the filename info
 20:54
20:55 brrt left
tgt m: Int but False 21:11
camelia rakudo-moar c793d8: OUTPUTĀ«(signal )Ā»
21:43 jnap joined
lizmat tgt: confirmed locally 21:53
r: Int but False 21:54
camelia ( no output )
..rakudo-moar c793d8: OUTPUTĀ«(signal )Ā»
tgt m: Int.clone()
camelia rakudo-moar c793d8: OUTPUTĀ«(signal )Ā»
lizmat $ MVM_SPESH_DISABLE=1 6 'Int but False' 21:55
Segmentation fault: 11
so appears to not be spesh related
tgt Where are nqp:: ops implemented? Specifically looking for nqp::clone 22:02
22:02 daxim_ joined
hoelzro tgt: you can find the mappings in the nqp repo 22:02
src/vm/moar/QASTMoarMappings or something
hoelzro looks
22:02 kjs_ joined
hoelzro src/vm/moar/QAST/QASTOperationsMAST.nqp 22:03
QAST::MASTOperations.add_core_moarop_mapping('clone', 'clone', :decont(0));
so it binds directly to the Moar clone instruction (which it pretty typical)
22:05 tgt_ joined 22:06 ggoebel11111118 joined
hoelzro and Moar ops are impld' in src/core/interp.c 22:06
clone starts on line 2558
tgt_ Thanks hoelzro 22:07
hoelzro np
tgt_ So does that correspond to MVM_OP_clone? Or some other clone? 22:08
hoelzro I believe that's it
I've just gotten used to searching for OP(clone) in src/core/interp.c
tgt_ Well, that's not something I can understand without spending some time with moarvm. 22:16
22:16 btyler joined
hoelzro it takes a bit of time, yes =) 22:17
but that's where you start
tgt_ :)
Any suggestion about where to start? Are there docs? 22:18
hoelzro I think there are some in the repo
I'm a beginner with Moar myself
tgt_ Yup, found them. Thanks. 22:26
Is there a way to get perl6-m to print a stacktrace?
hoelzro tgt_: a native stack trace, or a Moar one? 22:27
tgt_ Uh. What do you mean? 22:28
22:32 tgt joined

Powered by Cro and the RakuĀ® Programming Language.

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

Thank you!