Post closed tickets in your report. | Note: This channel is for our weekly status meetings (Fridays at 13:30 UTC); you probably want #parrot instead. | irclog: irclog.perlgeek.de/
Set by moderator on 12 November 2014.
08:10 Reini joined 08:23 Reini1 joined 08:29 lizmat joined
Reini1 Today I'm well prepared 13:28
Util Reini1: Today I am late; my report in 2 minutes. 13:34
Reini1 git reflog|grep merge 13:36
DONE
* Overall: lot of correctness, some optimizations, and extended precision
* smoke-me/valgrind-tests #1150 CAGE
* smoke-me/mathl-gh1111 PREC
* check auto::infnan for selected floatval, not just double #1146 FIX
* smoke-me/opt-print_n-gh828 OPT
* smoke-me/bignumsub-gh1144 FIX
* smoke-me/gc-gh1136+1143 FIX (but ms2/ms/inf still a bit broken with strings)
* smoke-me/test-cpuarch-gh1142 FIX
* pbc leaks 24byte imcc->globals #1151
* warn in Configure about experimental flags #1148 CAGE
WORK
* rurban/rpa-offset-gh1152 resizablepmcarray optimizations as in nqp/qrpa
just splice is missing. OPT
* smoke-me/embed-gh829 FIX
* analyze performance loss with --gc gms #1141 (20%), see parrot-bench OPT
BLOCKED
* smoke-me/signal-tests + smoke-me/test-capture-gh892 need windows smokes
* smoke-me/vms-gh866 needs a vms smoke
and moritz' ilbot could use better # urls, and detect gh\\d+ 13:38
it still points to old perl6 ticket urls
Util Report: 13:39
# Done:
* rosettacode.org/wiki/Phrase_reversals#Perl_6
* Thanksgiving (U.S. Holiday)
.end
Hello! Who all is in attendance?
Reini1 me 13:40
Util Glad you could make it! Lots of good work this week, I see. 13:41
What are our prospects of getting VMS smoke?
Reini1 yep, even if I burned my foot very badly. have to stay in bed.
Util Yikes? May I ask how? I am imagining a motorcycle exhaust pipe. 13:42
Reini1 VMS smokes? I think I called out to mvorl 13:43
cooking
just hot water. I hope I can keep my leg
joke
And I cannot do windows smokes still. Still no machines. they are expected next friday 13:44
I've got 2 WONTFIX:
smoke-me/null-call-gh1024 and pbc leaks 24byte imcc->globals irc://irc.perl.org:6667/#1151
the null call is illegal usage of :immediate with globals, but I added some more assertions 13:45
the leak is harmless for now I think
Util wishes Reini1 a speedy recovery. 13:46
Reini1 the big leaks are all gone though, and I added proper leak tests for valgrind. It runs through now (1-2 hrs)
Util Great!
Reini1 asan is also fine
Util I don't understand the URL you just referenced: irc://irc.perl.org:6667/#1151
Reini1 ticket id 1151
github.com/parrot/parrot/issues/1151 13:47
Util OK, thanks
Reini1 interesting url expander, my clients fault
so I started finally with the optimizations 13:48
rpa first (from qrpa)
then maybe switch back to ms2, if I can make it stable
and I started fixing the embed and extend interfaces and tests. 13:49
calling conventions later, too tricky for now 13:50
I started looking at autoboxing, but didn't come to a good ticket for chirag or me yet 13:52
Util In the final form of your GC work,
do you plan to have all the GCs (ms2, etc) selectable at runtime, or at compile-time?
(compile-time == compilation of Parrot itself)
Reini1 I keep everything as is
Util OK 13:53
Reini1 for now now only gms makes sense run-time, as the others are all a bit unstable.
but I fixed most issues, just a memory_pool problem for strings is remaining. then we can switch properly at startup time at will 13:54
it's good to have them with —gc for easier testing.
Util Indeed!
Changes to the calling conventions *will* be tricky, but I foresee much opportunity for speed-up there.
Let's discuss it thoroughly, when the time comes to tackle it.
Reini1 I also thought of writing a new libgc b&w (boehm & weiss) interface, which would be safer and faster
as names: b&w, bw or gc 13:55
it's a copying collector, with optional incremental sweeps
and it doesn't need our waste for ptr pools, 2 words for every ptr, as header and a prolog 13:56
that's the rough idea for next year
Util s/weiss/weiser/ ???
Reini1 just out of my head, probably 13:57
yes
Util OK.
Reini1 maybe boehm is the best name
and it's not copying, just simple m&s 13:58
I would love to try out a copying collector also, as it is much better for moar and potion/p2
for big boxes 13:59
the boehm advantage would be less root marking. it finds all used objects automatically from the stack, the dll's and teh used registers 14:01
so it's precise, uses less memory, and can do incremental (less pause times, as in the new ruby) 14:02
the contra would be the license
and I've looked briefly into the new ruby collector and didn't like it that much. it's better than ours, but not better than moar and potion's. 14:03
our biggest GC problem is probably StringBuilder.substr to make a copy 14:04
Util The license does not need to be an issue while only (Parrot) developers are working on the interface. If it (boehm) turns out to be enough better, then when can deal then with how to handle the license issue. 14:05
Reini1 yep. it's run-time changable and configurable 14:06
Util I have another obligation now. 14:07
Sorry to end the meeting before this topic tapered off.
Anything else to discuss? From anyone?
Reini1 no, have to go also 14:08
Util Reini1: If you post any more of your thoughts on GC here for the archive, I will be sure to read it.
Thanks, everyone!
Reini1 bye
Util Meeting adjourned.
Util waves goodbye 14:09
Reini1 when I find more out why our gms is so slow, and when ms2 is fixed I'll post something to the mailing list. my idea is that the cause for gms slowness lead to a faster and less memory hungry boehm. otherwise ms2 is fine also 14:10
14:36 lizmat joined 21:10 Reini joined 23:08 lizmat joined 23:28 Reini left

Powered by Cro and the Raku® Programming Language.

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

Thank you!