Perl 6 - the future is here, just unevenly distributed

IRC log for #moarvm, 2013-11-24

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

All times shown according to UTC.

Time Nick Message
01:26 BenGoldberg joined #moarvm
01:48 woolfy joined #moarvm
02:16 woolfy joined #moarvm
02:17 rurban1 joined #moarvm
02:20 BenGoldberg joined #moarvm
03:17 Ben_Goldberg joined #moarvm
03:40 ggoebel18 joined #moarvm
03:47 ggoebel19 joined #moarvm
06:18 rurban1 joined #moarvm
06:21 sorear joined #moarvm
08:45 nwc10 jnthn: I left valgrind running, and it finds no problems
08:54 FROGGS[mobile] not too bad
09:39 rurban1 joined #moarvm
10:32 lizmat_ joined #moarvm
12:16 tgt joined #moarvm
12:20 jnthn nwc10: Nice
12:49 nwc10 jnthn: you are having the appropriate amount of fun?
12:50 jnthn nwc10: I'm...uh..packing clothes for a week's trip away
12:50 nwc10 where you are going, is the beer good?
12:52 jnthn Stockholm
12:52 jnthn Yes :)
16:01 jnthn Confirm that I make it to stage optimize here as well
16:02 FROGGS I'd like to know why I die that early with an optimized moar
16:03 FROGGS I mean, I can build nqp just fine, and it dies when compiling Perl6::BOOTSTRAP
16:03 jnthn Oh...I have an optimized one here and make it all the way through the optimize stage
16:03 FROGGS "Heap corruption detected: pointer 0x65b5178 to past fromspace"
16:03 jnthn *to the
16:03 FROGGS ohh
16:04 FROGGS the bt only shows MVM_args_slurpy_positional as an interesting step
16:05 FROGGS jnthn: how long does stage parse take then?
16:05 FROGGS and what is your vm/parrot's stage parse?
16:07 jnthn Forever.... I don't know about the latter number. I know it's a bunch slower than JVM.
16:08 FROGGS I believe it took 460s on my box (unoptimized)
16:26 cognominal joined #moarvm
16:27 jnthn I'm working on getting rid of the massive warnings spewing...
16:31 FROGGS uhh, nice :o)
16:32 * diakopter is curious how long moar takes to parse the setting compared to jvm
16:39 FROGGS ohh
16:39 FROGGS the gc debug output takes 2.5GB >.<
16:39 timotimo sounds good :D
16:39 timotimo oh wow, that's a bunch
16:39 FROGGS good that I have an SSD :o)
16:40 timotimo does it dump out the whole nurseries? :P
16:40 diakopter essentially...
16:40 timotimo er, we do have nurseries, right?
16:41 diakopter in the gcorch branch the debug output is much more granular
16:41 diakopter (controlled)
16:41 FROGGS hmmm, the address from the error message is not in that output :/
16:42 diakopter search case-insensitive?
17:04 rurban1 joined #moarvm
17:05 rurban1 diakopter: urcu is also bsd, cygwin and more. Until recently RCU was patent encumbered, not anymore. LGPL is a problem? You can simply link to it, no problem. Only inlining large parts would be.
17:06 nwc10 jnthn: I get:
17:06 nwc10 Stage start      :   0.000Nominal type check failed for parameter '%spec'; expected Associative but got BOOTHash
17:07 nwc10 oh, well, sort of like that. Clearly the terminal had the munchies
17:08 FROGGS[mobile] nwc10: I believe you are missing a commit
17:08 nwc10 but no 10 PRINT "NYI"; 20 GOTO 10
17:08 nwc10 FROGGS[mobile]: Rakudo is at a6eeb0b159b82f41a2088cb20105f651fdead89b
17:08 nwc10 Implement nominal type checks in binder
17:09 nwc10 everything seems to be current
17:09 FROGGS[mobile] then it might be right, cant check right now
17:11 FROGGS[mobile] right, the BOOTHash msg from sat/fri was a different one
17:14 jnthn nwc10: Yup, me too
17:15 jnthn nwc10: I implemented the missing type check and it caught something we got away with before
17:22 jnthn oh, joy...the train now as a "technical problem"...
17:23 timotimo more time to hack before you arrive ;)
17:45 nwc10 jnthn: the "technical problem" affects movement? Or screws the lights, the food and the Internet too?
17:45 nwc10 is the gravity on the blink yet?
17:45 timotimo don't blink
17:45 timotimo don't even blink
17:45 timotimo blink and you're dead
17:50 jnthn nwc10: Movement, for a while. We're going againnow....
17:51 dalek MoarVM: 50a6529 | jonathan++ | src/core/interp.c:
17:51 dalek MoarVM: Bring iscont inline with other backends.
17:51 dalek MoarVM: review: https://github.com/MoarVM/MoarVM/commit/50a6529278
17:52 FROGGS diakopter: no, it already was a case-insensitive search
17:55 jnthn detrain &
18:31 colomon joined #moarvm
20:20 jnap joined #moarvm
21:43 jnthn OK, got local commits that recover us to the point we were at before, but with the sig binder actually enforcing type checks...
21:45 FROGGS jnthn++ # :o)
21:45 jnthn I've also got a lead on what's making setting compilation so utterly slow.
21:45 FROGGS ohh cool!
21:45 FROGGS that is way more interesting :D
21:45 FROGGS do tell
21:45 jnthn (mind-blowing number of call frames hanging around
21:46 FROGGS ohh
21:46 jnthn And I can guess where from, *and* I think if we can fix it we might well get a win on all backends.
21:46 diakopter why's that
21:46 FROGGS could that case the "Heap corruption detected: pointer 0x65b5178 to past fromspace"?
21:46 jnthn No, that sounds like a plain old bug...
21:47 tadzik how boaring
21:47 FROGGS awesome
21:47 diakopter roar
21:47 FROGGS gah
21:47 FROGGS how goating
21:47 diakopter boaning
21:48 jnthn I love how talking about goating makes diakopter thing about boaning :P
21:48 jnthn *think
21:49 diakopter gloating
21:50 FROGGS now you're just woaffling :P
21:52 diakopter jnthn: lemme guess: past stub clones
21:52 jnthn no, something grammar related
21:53 diakopter I thiught that was grammar related
21:53 FROGGS thought* # that was grammar related
21:53 FROGGS sort of
21:54 diakopter hee
21:54 FROGGS damn it... how should I sleep now when there might be a win on all backends out there??
21:55 tadzik don't sleep, let's not sleep together
21:55 tadzik ...that sounds a bit bad
21:55 dalek MoarVM: 4867b27 | jonathan++ | src/core/hll. (2 files):
21:55 dalek MoarVM: Support HLL array/hash/code mappings.
21:55 dalek MoarVM: review: https://github.com/MoarVM/MoarVM/commit/4867b27698
21:55 FROGGS been there, done that :o)
21:57 diakopter tadzik: but, 《cry》
21:59 diakopter well that didn't work out very well
21:59 diakopter (now on a real keyboard)
22:12 cognominal joined #moarvm
22:35 jnthn OK, my prototype of the change makes CORE.setting compilation time on Moar about 2/3 of what it was before...and seems to shave some space off
22:36 jnthn I'm going to push it to a branch in NQP and leave testing on other platforms to @other
22:36 timotimo yays!
22:36 jnthn Also it warrants a Rakudo spectest run on it too
22:37 FROGGS I'll do that and report after $sleep
22:42 benabik What change makes CORE take 2/3rds the time?
22:42 timotimo https://github.com/perl6/nqp/commit/3b280649c2
22:42 jnthn benabik: Only tried on Moar, and only stage parse.
22:43 diakopter jnthn: so what's keeping that cursor hanging around
22:44 jnthn diakopter: Well, cursors stay around because match objects point to them.
22:44 jnthn diakopter: But there's no reason cursors we'll never backtrack into need to keep pointing at other things.
22:44 diakopter even failed cursors?
22:44 diakopter er, failed matches
22:45 jnthn Yeah, but those tend not to stick around long anyway
22:45 jnthn Especially as <foo>? is now Nil
22:46 jnthn There's something that makes me uneasy though...the fix I did shouldn't have helped as much as it did, or so goes my gut feeling
22:46 diakopter right, that's my gut feeling too
22:46 diakopter which I was trying to express by those questions :)
22:47 jnthn Or at least, if may help other backends proportionally do how it helped Moar.
22:47 jnthn Which is *fine*. :)
22:47 jnthn But still leaves us with some stuff to hunt down :)
22:47 diakopter it hints that there are plenty larger gains out there
22:47 jnthn Aye.
22:48 diakopter "they're not inefficiencies! they're potential programmer rewards!"
22:48 jnthn Well, one advantage I have with Moar is I understand it near enough fully and can do quick instrumentation hacks if I need to go exploring...
22:49 jnthn Ah...yeah, something needs looking into
22:50 jnthn The inter-generational roots list is huge.
22:51 diakopter oo that's a potential big gain :)
22:51 jnthn Well, I once knew we had algorithmic todo there, but I did that a while back. But yeah, that shouldn't happen. Will need to work out why.
22:51 timotimo jnthn: see #perl6, on parrot i get no big speed improvement
22:52 timotimo like 4% perhaps
22:53 jnthn timotimo: OK. What about memory, ooc?
22:53 timotimo i'll have to measure that more carefully
22:53 timotimo i'm doing the spectests ATM
22:53 timotimo should i measure the memory usage first?
22:56 jnthn Well, whatever order you wish :)
22:56 jnthn Something odd is that we seem to box an insane number of ints
22:57 FROGGS moar stage parse is 318.597s, was 468s
22:58 diakopter win.
22:58 jnthn Nice
22:58 jnthn Well, "nice"
22:58 FROGGS :o)
22:58 diakopter that's still about 318 more seconds than I'd hope it to be eventually
22:58 timotimo oh wow.
22:58 timotimo you have some high ambitions
22:58 FROGGS diakopter: that is an unoptimized moar
22:59 diakopter timotimo: like you wouldn't even believe
23:00 FROGGS it only takes a few seconds (to explode) using an optimized moar though
23:00 timotimo oh yikes
23:00 benabik Erm.  Exploding bad.
23:01 timotimo so the parrot spectests are completely clean except the one failure froggs also has had for ages
23:01 FROGGS benabik: but it explodes fast :o)
23:01 FROGGS and that is what we want after all
23:01 benabik Slow explosions are just burning.  ;-)
23:02 jnthn OK, sleep time for me :)
23:02 jnthn 'night
23:03 FROGGS gnight jnthn
23:04 diakopter FROGGS: wait, asplode?
23:04 FROGGS diakopter: yes
23:04 diakopter boo
23:05 FROGGS a gc bug as it seems
23:05 FROGGS missing MVMROOT probably
23:05 diakopter I'm curious if the gcorch branch fixes it (once the gcorch branch is fixed to work at all really)
23:08 timotimo yeah, when is that going to hit master? :)
23:08 benabik Presumably sometime after it starts working.
23:08 diakopter <shrug>
23:08 timotimo i guess that's fair
23:09 diakopter I'll have to kickban dalek temporarily so jnthn doesn't notice the merge ;)
23:13 diakopter that was a joke, future jnthn
23:16 timotimo i wonder what big problems moarvm has for parsing performance
23:21 FROGGS I'll compile tomorrow on windows to see how fast/slow an optimized moar does it
23:23 timotimo should i do JVM today?
23:24 FROGGS I did, 54s, but I dunno how long it took before
23:24 FROGGS currently spectesting both
23:26 FROGGS no difference in p's stage parse, as you mentioned
23:28 FROGGS gnight
23:28 timotimo ginte :)
23:50 BenGoldberg joined #moarvm
23:56 FROGGS joined #moarvm

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