retupmoca github.com/MoarVM/MoarVM/pull/90 please :) 00:11
00:20 woolfy left
masak I think I dare merge that one. 02:59
it looks like an unproblematic improvement.
retupmoca: tests still pass after that one?
dalek arVM: 69dbec5 | (Andrew Egeler)++ | src/io/fileops.c:
When readline/linenoise see EOF, mark handle eof

This lets you use ^D to exit the rakudo/nqp REPL.		 03:00
arVM: 5393d34 | (Carl Mäsak)++ | src/io/fileops.c:
Merge pull request #90 from retupmoca/master

When readline/linenoise see EOF, mark handle eof
masak went ahead and merged 03:01
retupmoca masak++ 03:21
o.O apparently my patch is causing a failure in S06-other/main-usage.t - I'd assumed that was unrelated, and just one of the tests that always fail 04:37
but if I run that test myself with a .pl file, everything works
retupmoca is confused 04:38
but either way, retupmoca--
I don't understand these function signatures (for is_run). It looks to me like main-usage.t is wrong (line 157 should be moved after line 160, for example) 04:50
doing that makes the test pass, and actually matches the is_run parameters to my eyes, but I'll let someone smarter than me take a look at it first 04:51
I don't understand why the args can come before the expected values, because is_run has %expected as positional, and args is a slurpy hash 04:54
ok, now it's passing sometimes, but not all the time 05:08
I give up. I don't even know if MVM_file_readline_interactive_fh is being called during these tests 05:13
pretty sure my patch caused this weirdness, though
retupmoca goes to bed
07:13 Ven_ joined 07:28 zakharyas joined
masak I'll let someone else have a look at it too, rather than just going ahead and knee-jerk reverting. 07:30
hehe, guess this puts us in the "forgiveness" part of the workflow... :) 07:31
07:37 lizmat joined
nwc10 jnthn: are you the last person left writing PIR? In that, most the Parrot activity I've seen in recent months seems to be in other parts of the codebase (or their build system) 07:47
07:47 woolfy joined
jnthn I'm rather doubtful the patch's code path is invoked in the test in question. 08:15
08:20 woolfy left
masak oh phew 08:24
:)
it seemed an innocuous patch to apply.
jnthn nwc10: Well, I was writing code to generate PIR... :) 08:25
nwc10: I don't really plan to repeat the experience if I can avoid it ;)
nwc10 you don't say
timotimo with the removal of the scanning for the callsite's arg names i seem to be able to build an nqp 08:29
and a rakudo 08:34
very nice.
(not using it in the specializer so far, but i probably *could* do that next up 08:37
which would at least give me a bit of a "yay" factor to this project a bit earlier to keep me going :P ) 08:42
jnthn You know you'll have to get interning to handle callsites first too, yes? :) 08:53
ah, callsites with nameds I mean
timotimo i do?
the specializer will only ever touch interned callsites?
jnthn Yes. 08:54
timotimo ah
jnthn That's really important.
timotimo i guess i'll have to do that then :)
that also means it'll never touch flattened stuff, eh?
jnthn Correct
Which is sane. 08:55
For at least the medium-term.
timotimo aye
in the long term we may be able to do some cool stuff with inlining there
jnthn yeah
timotimo for example, we should be able to generate some really tight code for foo(|%bar) given sub foo(*%bar) 08:56
jnthn Yeah, exactly
timotimo hm, i'm wondering why we can't do that earlier yet 08:57
well, for subs it's much easier of course
the piece of code that lowers binding could do that probably
jnthn umm
Only if inlining.
timotimo oh
jnthn And that happens in the optimizer.
timotimo right, the receiving part needs to know about it too
jnthn It'll want to be Moar-level otherwise. 08:58
Yes.
timotimo i need to hold back a lot today, wrist's kind of hurtin' 09:01
jnthn Yes, take care of those... 09:02
timotimo i wonder if we should use the new rc-forest-fire in p6bench 09:09
now the p6 version has a different algorithm than the p5 version and the nqp version 09:10
so that would probably want ported
jnthn timotimo: Well, I'd fix the .bless(*) in it
timotimo: But if you change the algo then it makes it worthless to compare with historical data.
timotimo yes
jnthn I'd leave it. 09:11
timotimo we could have rc-forest-fire-2
jnthn True :)
timotimo the bless(*) is only called once during the whole program, though
doesn't mean i won't fix it 09:12
just means it doesn't impact performance
jnthn Oh, I thought it was hotter than that
Fair enough
Still nice to kill the dep warning.
timotimo yes 09:13
warnings don't show during test runs, so i didn't notice it yet
jnthn Oh, I ran it manually here to get a profile. :) 09:14
10:12 woosley left 10:40 brrt joined 11:34 brrt joined
retupmoca I get to keep my ^D handling! \o/ (0100 local and lack of sleep do not make for a debug session that makes any sense :P) 11:45
JimmyZ brrt timotimo : irclog.perlgeek.de/perl6/2014-04-10#i_8567933 # I think you may be interested too. 11:54
brrt will read 12:01
it is, on the other hand, not very convincing 12:20
(i.e. the speed boost)
12:24 lee_ joined
brrt is in the process of trying out new wm's for the netbook i'm working on 12:46
anybody has a good idea?
tadzik i3 13:13
brrt didn't work too well just now 13:44
i'm now using lxde 13:45
which is ok i guess
lee_ i always liked ion3, but it is abandoned now 13:54
13:59 btyler joined 14:11 jnap joined 14:23 jnap1 joined 14:54 odc joined
dalek arVM/named_to_positional: b50df7a | (Timo Paulssen)++ | src/ (2 files):
less debug spam
 14:56
arVM/named_to_positional: 27fd8cd | (Timo Paulssen)++ | src/gc/roots.c:
only need to scan arg_name for the flattened case
15:35 daxim_ joined
japhb timotimo: Just finished reading the backlog here (but am a couple days back in #perl6, so if you discussed this there, I apologize) -- what is "the new rc-forest-fire" that you are suggesting? 15:55
jnthn japhb: Somebody (TimToady?) worked out an alternative forest fire impl that runs faster.
timotimo Larry wall improved the forest fire code by a whole lot on Rosetta code 15:56
jnthn japhb: Good for RC, but better to keep the current one in the benchmark suite, imo
japhb: Though maybe add second impl as an extra benchmark
timotimo The secret was to introduce a new state for the trees, which indicates that nearby tree is burning, and then treating the state array as mutable 15:57
dalek arVM: f7ddc7e | jnthn++ | src/core/interp.c:
Set hint for invokespec and multispec.

Means the attribute lookups go a good amount faster.		 15:58
japhb timotimo, jnthn: Oh interesting. OK, will his optimizations make sense translated to NQP and Perl 5 as well? Because if so, I agree with making a new set of "rc-forest-fire-opt" (or somesuch) that compares matching algorithm. 16:04
16:06 dalek joined
timotimo I think it will for NQP and most probably also for Perl 5 16:06
17:44 zakharyas joined 18:17 colomon joined
retupmoca jnthn, timotimo: I think I found a spesh bug 21:03
perl6 t/spec/S06-other/main-usage.rakudo.moar # fails one test
MVM_SPESH_DISABLE=1 perl6 t/spec/S06-other/main-usage.rakudo.moar # all pass
it still doesn't happen exactly the same way every time (I had one run with 2 failing tests), but I have yet to see a failure with MVM_SPESH_DISABLE 21:07
jnthn nwc10++ mentioned to me he got ASAN trips in some spectests; it wouldn't surprise me if they're related 21:08
(I've got a stack trace also, to study.)
22:32 lizmat joined 22:36 woolfy joined 23:02 woolfy1 joined 23:47 woolfy1 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!