Post closed tickets in your report. | Note: This channel is for our weekly status meetings (Tuesdays at 19:30 UTC); you probably want #parrot instead. | irclog: irclog.perlgeek.de/
Set by moderator on 19 August 2012.
00:46 benabik joined 07:52 lucian joined 13:32 bluescreen joined 13:55 benabik joined 14:04 benabik_ joined 15:07 benabik joined 15:08 benabik_ joined 15:36 benabik joined 15:43 wagle joined 15:44 cotto joined 15:45 contingencyplan joined, rurban joined 15:47 sorear joined, slavorg joined, Tene joined 16:25 bluescreen joined 16:34 bluescreen joined 16:42 bluescreen joined 17:20 benabik joined 17:57 benabik joined
rurban now or in one hour? 18:32
18:39 lucian joined 19:29 lucian joined 19:36 tadzik joined
rurban I'll wait for whiteknight 19:37
no ps today? 19:56
cotto no whiteknight, I guess 19:57
rurban I've got 30 lines to post 20:00
cotto go for it 20:01
rurban DONE:
* merged several minor branches and fixed tickets.
* bumped PBC_COMPAT to 12.1 for the new pmc/os, and need 13.0 for
the new PBC format in native_pbc.
* had a lot of parrot performance discussions with the rakudo folks.
parrot performance is the only problem, not bugs, not features.
TODO:
* replace nqp-rx with nqp-p6 (started as git submodule)
* started working on missing or wrong number converters in native_pbc,
now that the tests are stable.
* started looking into replacing parrot with a llvm/vmkit based
vm with natural types and calling convention for nqp. esp. no
continuations and a cdecl/fastcall ABI. libraries from java or .NET.
* sorear hates the lorito design. why should this be faster at all?
easier jittable yes, but the calling convention still sucks and jitting
this mess would make it only slower, not faster.
* lack of management: who made all these crazy parrot decisions?
My summary is: "premature pessimation is evil" 20:02
e.g. pbc broken by design on purpose.
nqp switched to a different ffi, because libffi did not work on windows
for them. I do not understand why this could have happened. libffi works
better than dyncall on windows for me.
* performance:
* esp. our calling convention looks incredibly stupid and slow, methods go via
Parrot_ext_call really? Parrot_ext_call as the only generic calling function
is a joke also. Parse signature strings at run-time. Every simple ffi does better.
* create a new object for each function/method call? not long ago even two.
BLOCKS:
* threads look a bit racy here and there, but adougherty fixed the
biggest blocker, the sleep deadlock. looks mergable.
* native_pbc is ready and needs votes to get merged. replaced
the parrot version triple for opslib with the PBC version.
* without parrot performance soon rakudo will switch to something else.
me also.
benabik We should consider moving #ps. Judging by attendance, this isn't a great time for people anymore. I know I'm in class. 20:03
Coke we also just don't have many developers anymore. rurban and whiteknight, now that GSOC is over. 20:09
I'm happy to kibbitz, but I'm not contributing code at this point. 20:10
benabik I'm sticking around, although my output will be lessened.
Coke regarding your question about who made these decision decisions -- it's been 11 years, and a lot of groups have come and gone. 20:12
rurban it was more rhetorical 20:16
but someone should overlook the efforts - whiteknight seems to be the pm - and talk to the rakudo folks. 20:17
their lorito and calling convention opinions
but our threads look really good and fast. 20:18
20:18 pmichaud joined
benabik Doesn't rakudo ignore our calling conventions for the most part? 20:18
rurban I just have to bing the default number to the run-time num of cpus.
sure, because ours is too stupid 20:19
I just have to bind the default number of threads to the run-time num of cpus 20:20
The GC also looks good.
Coke our GC has issues, to be sure. 20:21
rurban Only with --leak-test on destruction. 20:22
nci and threads seem to work fine.
Coke no, with the amount of memory allocated depending on what parrot thinks is available, the inability to say "only use this much memory and no more.." 20:23
sorear o/ pmichaud
Coke (the actual collection itself I know of no issues with) 20:24
pmichaud I think Coke++ is referring to github.com/parrot/parrot/issues/795, perhaps among other things 20:47
rurban Oh, no libsigsegv, and no malloc-fail check 21:11
Fixed now all pbc number converters 21:31
21:32 tadzik left 21:33 pmichaud left 21:43 benabik joined 21:48 benabik joined, lucian joined 23:06 whiteknight joined