Perl 6 - the future is here, just unevenly distributed

IRC log for #moarvm, 2014-10-24

| Channels | #moarvm index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
00:12 daxim joined #moarvm
02:48 colomon joined #moarvm
04:46 ggoebel111111110 joined #moarvm
05:50 avuserow joined #moarvm
06:13 garyaj joined #moarvm
07:00 PerlJam joined #moarvm
07:00 retupmoca joined #moarvm
07:00 odc joined #moarvm
07:00 leedo joined #moarvm
07:00 nine_ joined #moarvm
07:10 btyler_ joined #moarvm
07:32 nine_ joined #moarvm
07:32 leedo joined #moarvm
07:32 retupmoca joined #moarvm
07:32 btyler_ joined #moarvm
07:37 PerlJam joined #moarvm
07:37 odc joined #moarvm
07:37 FROGGS joined #moarvm
07:40 retupmoca joined #moarvm
07:40 leedo joined #moarvm
07:40 nine_ joined #moarvm
08:33 zakharyas joined #moarvm
09:06 kjs_ joined #moarvm
09:48 brrt joined #moarvm
09:49 brrt please somebody remind me to blog for moar one of these days (after the weekend)
09:53 avuserow joined #moarvm
09:53 ggoebel111111110 joined #moarvm
09:53 japhb joined #moarvm
09:53 cxreg joined #moarvm
09:53 lue joined #moarvm
09:56 retupmoca joined #moarvm
09:56 leedo joined #moarvm
09:56 nine_ joined #moarvm
09:56 cxreg joined #moarvm
09:58 nebuchadnezzar joined #moarvm
09:58 [Coke] joined #moarvm
12:12 JimmyZ joined #moarvm
12:30 JimmyZ masak: ping?
12:34 masak JimmyZ: pong
12:35 kjs_ joined #moarvm
13:49 [Coke] seen in today's moar non-jit run:
13:49 [Coke] S32-hash/delete-adverb.t...........................moar(34011,0x7fff7bce2310) malloc: *** error for object 0x7f8954e56840: pointer being freed was not allocated
13:49 [Coke] *** set a breakpoint in malloc_error_break to debug
13:49 [Coke] sh: line 1: 34011 Abort trap: 6           nice -20 ./perl6-m t/spec/S32-hash/del
13:49 [Coke] OSX.
13:56 [Coke] Looks like that's the only fail on non-jit, but it cost 131 tests.
14:15 brrt [Coke] - are there jit-specific fails?
14:15 brrt i kind of need to know about them :-)
14:15 brrt also [Coke]++ for testing
14:15 timotimo brrt: how much have you looked at nativecall_invoke yet?
14:16 timotimo i'm asking just because i'd like to begin turning native call invocations into jitted stuff by using the jgb_append_call_c function
14:17 brrt i have not looked at it at all
14:17 timotimo OK
14:17 brrt where... is that stuff
14:17 brrt :-)
14:17 timotimo so i'm not going to be trampling on previous work without knowing :)
14:17 brrt is that a new op?
14:17 timotimo nativecall.c is in core/ i think
14:17 brrt oh
14:17 brrt i'll look
14:18 timotimo it's called nativecallinvoke actually
14:18 timotimo i think it'll be enough if i just look at it and ask specific questions directly, so you can focus on other stuff
14:23 [Coke] brrt: the JIT-specific failures are here when they are posted:
14:23 [Coke] https://github.com/coke/perl6-roast-data/blob/master/log/rakudo.moar-jit_summary.out
14:23 brrt thanks
14:23 brrt timotimo: good idea :-)
14:24 [Coke] today? no failures on JIT. non-JIT is the problem.
14:24 brrt [Coke] - this is on which platform?
14:24 brrt hmm
14:24 brrt seems like flappiness
14:24 brrt JIT is rather unlikely to do something right that was done wrong by the interpreter
14:24 [Coke] (but yesterday, S32-hash/delete-adverb.t had the failure non-jit had today; sounds like a flapper)
14:24 brrt yeah i agree
14:24 [Coke] brrt: right now is OS X. historically, it was linux.
14:25 timotimo brrt: have you considered building fast-paths for big integer arithmetic in the cases where both operands are stored as "small big ints"?
14:25 * timotimo is AFK for a bit
14:25 brrt considered, yes, but i don't know if you can prove that they will always be fast paths
14:25 brrt and you're going to end up testing-and-jumping for the case they are not
14:26 brrt i can do that, and it will probably help a bit, but I don't know if the JIT is the right level for it
14:26 brrt (of course the JIT should do it one day or another)
14:26 brrt s/day/way/
14:27 brrt I'm going to see if i can repeat those tests. we don't nearly have enough os-x based devs :-)
14:30 brrt clang-is-so-damn-slow
14:32 lizmat joined #moarvm
14:48 timotimo brrt: yeah, i expect that you'd have to test-and-branch
14:49 brrt hmmm
14:49 brrt i'll be able to work on this in a week or so (hopefully)
14:50 timotimo neat (even though arguments would be more pressing or something like that)
15:02 brrt :-)
15:02 brrt yeah i know i know
15:02 brrt :-)
15:02 timotimo sorry for nagging you about that all the time :P
15:03 odc left #moarvm
15:04 kjs_ re Mac OS devs: is there a “press this once” button to run tests? I’m happy to test now and again, but no time to develop
15:04 timotimo a simple script could be built using rakudobrew for building a fresh rakudo and then run the spec tests
15:05 brrt aye
15:05 brrt rakudobrew would be a good place to put that
15:05 timotimo ah, yes
15:05 timotimo rakudobrew test moar
15:05 timotimo OSLT
15:06 brrt get to it
15:06 timotimo and then we could have a reporter
15:06 brrt or 'well volunteered'
15:06 brrt :-)
15:06 brrt i'm going to weekend
15:06 timotimo haha
15:06 timotimo have fun brrt
15:25 * TimToady wonders if it would make more sense in some cases to just attempt the op and then branch on overflow, if it can be detected
15:26 timotimo we'd still have to check if the storage type is a "proper" bigint already or if it's a smallbigint
15:35 JimmyZ Isn't life simpler if there is Int class and BigInt Class :P
15:37 TimToady not for the user
15:37 JimmyZ aye
15:38 timotimo wouldn't it be easier to have a regular string that does the stringy stuff and a unicode string type that you use when you want to do unicody stuff? :P
15:38 * timotimo AFK
15:39 TimToady in any case, the design of P6 assumes that people will write int64 or so when they want raw speed
15:41 * JimmyZ agrees it
15:41 TimToady or have some pragma that says "use Int int64;"
15:41 JimmyZ constant Int = int64 :P
15:41 TimToady we'll need jnthn's N refactor first
15:42 TimToady "constant Int" will not necessarily switch the policy of using big integers for intermediate results
15:43 JimmyZ I saw constant utf8 = Blob[...] in the core setting
15:44 TimToady though I suppose we could say that if you redefined Int to be a native type, it magically assumes that you want that for intermediate results as well
15:45 TimToady but it might be wise to have a separate declaration for such semantic bending
15:46 TimToady then there's the folks who will want "constant Int = num"
15:46 JimmyZ yeah, I saw constant buf8 = Buf[uint8]; there
15:47 TimToady .oO(constant Int = complex)
15:47 JimmyZ :P
15:49 TimToady some people may actually think they want constant Num = Complex
15:50 TimToady though it's probably sufficient to redefine sqrt to always return complex
15:50 [Coke] we need something like smolder to record test runs.
15:51 * [Coke] is reminded to get back to trying to get spectest runs into jenkins.
15:53 JimmyZ Good night ...
15:54 lue joined #moarvm
16:04 Guest1419 joined #moarvm
16:26 FROGGS joined #moarvm
16:55 colomon joined #moarvm
17:36 kjs_ joined #moarvm
17:51 FROGGS joined #moarvm
21:26 woolfy1 joined #moarvm
21:27 lizmat_ joined #moarvm
22:43 japhb joined #moarvm
22:43 avuserow joined #moarvm
22:46 dalek MoarVM/closefhi: 8d9f84a | (Tobias Leich)++ | / (14 files):
22:46 dalek MoarVM/closefhi: implement nqp::closefhi which retuns the exit code of a pipe
22:46 dalek MoarVM/closefhi:
22:46 dalek MoarVM/closefhi: nqp::openpipe returns a file handle one can read from, but since such an
22:46 dalek MoarVM/closefhi: opened pipe is actually a spawned process we also need its exit code to
22:46 dalek MoarVM/closefhi: know if it run cleanly. Use nqp::closefhi on such a file handle if you
22:46 dalek MoarVM/closefhi: are interested in the exit code, nqp::closefh if you are not.
22:46 dalek MoarVM/closefhi: review: https://github.com/MoarVM/MoarVM/commit/8d9f84a818
22:48 ggoebel111111110 joined #moarvm

| Channels | #moarvm index | Today | | Search | Google Search | Plain-Text | summary