01:37 tokuhiro_ joined 01:48 ilbot3 joined 01:53 tokuhiro_ joined 01:59 vendethiel joined 03:27 colomon joined 07:03 lizmat joined 07:05 Ven joined
nwc10 found the reasoning behind all of this interesting: www.snarky.ca/try-to-not-use-the-c-api-directly 07:05
oh my 07:06
www.blogger.com/comment.g?blogID=3...amp;bpli=1 -- ... the non-SSE2 backend for older x86 processors is fully working and can be built from source, which takes almost 7h on a 2.2GHz Athlon XP.
07:08 kjs__ joined
timotimo oh wow 07:13
that's a whole lot of time 07:14
07:16 FROGGS joined 07:37 dalek joined 07:49 zakharyas joined 07:58 psch joined 08:28 brrt joined
brrt nwc10: ain't pypy fun 08:32
:-P
timotimo brrt: anything i can do to help out with your workload? 08:33
brrt how good are you with gdb :-P
no, not really
sorry :-)
thansk for the offer though
what you can do is bug me about it after i've had a few days to mull over the next few things 08:34
timotimo mhm
brrt ok, the one thing i really want done is unbrittling the DynASM rsp case handling 08:36
so i'm kind of thinking on how to do that best 08:37
timotimo right, that could otherwise bonk someone on the head at some point 08:39
brrt possibly, yes
timotimo and it's always a good idea to leave third party projects you've contributed to in the course of adding a feature to your project in a good state 08:40
brrt oh, yes, of course 08:43
but nobody suffers from my wonky dynasm branch but me
so i'm not too worried about that
timotimo yeah, but some day it may want to get merged up stream, right?
we can't be the only ones interested in dynasm for x86_64 08:44
brrt right 08:45
one of the nasty bits about that was that i ran dynasm through an indenter before starting to hack on it
so that will need some undoing 08:46
timotimo oh crap 08:48
find an intern to do that for you %)
brrt well, it's not that bad? just get the patches from post-indentation to final and apply them on the unindented
or, i speak with them about it and ask how they feel about having dynasm sanely indented 08:49
timotimo well, yeah, but the code you'll have added will still be indented wrong?
haha, "sanely"? :)
someone will not like that choice of words
brrt lets just say that it was previously mike-pall-indented
of course not. but i didn't apply an indenter to satisfy my personal whims. it was unreadable 08:50
timotimo :) 08:52
brrt let's just say that brilliance and comparative isolation can be a toxic combination
09:03 FROGGS joined 09:09 kjs__ joined
jnthn returns to the nasty frames issue 10:14
timotimo ah, yes
10:19 ShimmerFairy joined
timotimo i'll run spectests for a patch that uses a Want + WVal's value.WHAT instead of the Want's .returns in that place 10:29
it seems to improve code generated in this case
FROGGS timotimo: what happened to the patch we discussed last night? 10:30
timotimo oh!
that was excellent :)
FROGGS is worth anything?
timotimo wait, what was it for again? :P
FROGGS to remove the calls to infix:<,> I believe 10:31
timotimo ah, yes
i have to check it a bit more again
it seemed good
FROGGS k
timotimo i have no clue why it seemed to not work when i built pretty much the exact same code :D
FROGGS timotimo: well, you didnt :P 10:32
timotimo that must be it
FROGGS you have to put the $list in the want 10:33
timotimo i thought i did that, though
oh, the patch i gave to you wasn't what i tried first
it was some strange in-between state, i fear
FROGGS yes, you tried to recover your modifications somehow 10:35
timotimo huh
a few times i get "no subtests run"
i hope it's just increased debug output causing trouble for the harness
FROGGS well, the odds are not very high though 10:38
10:38 Util joined 10:41 ShimmerFairy joined
timotimo spec tests aren't very happy :| 10:45
yeah, trouble 10:48
i should have used =:= NQPMu instead of just boolifying the type from the start 10:49
because type objects boolify to false 10:50
i want a tool that knows about recent spectest runs and highlights differing results immediately as the results are flowing past 10:51
but the spectest seems much better now 10:53
but perhaps i made my optimization useless and it's only because of that it works :D
11:09 kjs__ joined 11:22 Ven joined
dalek arVM: 113c3b5 | jnthn++ | src/ (6 files):
Fix data race in frame reference count handling.

Previously, it was possible for a reference to a frame to be acquired from the body of an object and for it to die before the frame's ref count was increased. This adds an "acquire frame" operation that will take care of this issue by ensuring the reference count did not go to zero already, thus avoiding the race. To ensure the zeroed count will remain zero after frame death, we arrange to not free the frame's memory until the next safe point if it was ever referenced by an object body; the safe point mechanism can still free it right away in a single-threaded program.
11:54
timotimo that doesn't sound like terribly much overhead for the multi-threaded case 11:55
jnthn It ain't too bad
FROGGS jnthn: wow, such comment 12:09
:o)
timotimo the hairy problems deserve big commit messages
12:15 [Coke] joined 12:16 kjs__ joined
arnsholt Yeah, if the time spent hacking vs. lines of code changed is very weird, long commit messages are definitely in order 12:28
timotimo coders shall now be paid by lines-of-commit-message 12:29
jnthn Sadly, any payment scheme is gameable :) 12:40
timotimo yes 12:43
i mean
git commit -m $(yes)
12:55 Ven joined 13:00 brrt joined 13:12 colomon joined 13:14 brrt joined 13:24 dalek joined 13:40 brrt joined 13:52 Ven joined 14:24 Ven joined 14:25 brrt joined 14:55 kjs__ joined 15:09 tokuhiro_ joined 15:14 Ven joined 16:00 FROGGS[mobile] joined 16:11 tokuhiro_ joined
dalek arVM: 3d63248 | (Pawel Murias)++ | src/strings/ops.c:
Make nqp::findcclass behave as on the jvm.
16:17
arVM: d17524c | (Pawel Murias)++ | src/strings/ops.c:
Make findnotcclass behave as on the jvm.
arVM: ce87c97 | jnthn++ | src/strings/ops.c:
Merge pull request #259 from pmurias/findcclass-fix

Make the nqp::findcclass na nqp::findnotclass behave the same as on the jvm.
16:42 kjs__ joined 16:57 vendethiel joined 17:12 kjs__ joined 18:06 kjs__ joined 18:12 tokuhiro_ joined 18:19 lizmat joined 18:21 FROGGS joined 18:23 tokuhiro_ joined 18:58 kjs__ joined 19:25 tokuhiro_ joined 19:48 lizmat joined 19:57 Peter_R joined 20:05 kjs__ joined 20:09 brrt joined 20:31 colomon joined
brrt oh, have i a wonderful nugget of surprise for y'all 20:37
you know how rsp encoding is weird?
rbp encoding is *also weird* 20:38
timotimo is that so!
brrt but differently
yes
well, bit less weird actually
but still 20:39
timotimo *sigh* 20:40
brrt is going to blog about this for sure
and expose the evil to the world!
timotimo i'm just glad we have you to make it all work out in the end :) 20:42
brrt how long can the end wait? 20:43
timotimo hah
brrt i think i'm going to claim TimToady's defence. doesn't matter if it took N years if N years lie in the past :-) 20:44
and, at some point in the future, it will all be past 20:45
timotimo ouch! :) 20:49
brrt oh, lord, what botched corner of x86 have i opened up now 20:50
apparantly, 0044 is the secret code for 'just use the rsp register damnit' 20:53
octal, that is
and any other way lies madness 20:58
timotimo where lies mundaneness? 20:59
brrt a sane ISA like ehm.... i dunno 21:17
some hypothetical one i presume
yay, i brokened it 21:19
timotimo hm 21:21
explain that to the intel engineers that'll show up at your doorstep every second now
you broke their ISA!
brrt i broke dynasm :-)
intel guys can sleep in peace
psch ^5 for breaking things!
i'm more breaking myself with that ConcurrentModificationException, which means it's bed time 21:22
lizmat good night, psch
timotimo gnite psch 21:23
i hope you'll have more luck tomorrow!
21:24 colomon joined
brrt good night psch 21:24
(ConcurrentModificationException implies java?)
timotimo yup
brrt sometimes feels like throwing all of this to the wind and programming sanely and boringly in java or javascript or ruby or whatever is currently fashionable 21:25
then i renember i'll never be truly sane
timotimo mhh 21:26
21:26 tokuhiro_ joined
brrt the thing is, i can kind of cargo-cult the 'rsp trick', but it's absolutely puzzling why it works 21:29
timotimo mhm
brrt it is not, so to speak, in the table 21:30
21:37 kjs__ joined
brrt hmm 21:37
or maybe it is
timotimo aaaand he's gone :) 21:41
good night and see you tomorrow or so :)
23:28 tokuhiro_ joined