Perl 6 - the future is here, just unevenly distributed

IRC log for #phasers, 2011-09-06

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

All times shown according to UTC.

Time Nick Message
03:04 samlh joined #phasers
05:13 pmichaud joined #phasers
05:13 cotto_work joined #phasers
05:14 tadzik joined #phasers
15:07 samlh joined #phasers
15:13 sorear DID: Not enough to warrant a pre-report
15:13 moritz same here
15:14 moritz oh wait, I did fix Test::Util on Windows
15:14 jnthn moritz++
15:14 jnthn That made me happy :)
15:15 moritz and I stopped pi.t from disturbing everybody
15:15 moritz and renamed &run to &shell
15:15 moritz end of not-really-report
15:53 samlh_ joined #phasers
16:59 masak joined #phasers
17:34 colomon joined #phasers
18:22 samlh joined #phasers
18:32 mberends joined #phasers
18:50 Util Pre-report:
18:50 Util # Done: gave talk on RosettaCode to Atlanta.pm.
18:50 Util # Plan to do: Nothing this week due to $WORK.
18:50 Util # NOTE: substantial traffic on Parrot mailing list re: abandoning deprecation policy to better accommodate Rakudo development speeds.
18:50 Util EOR
19:00 masak o/
19:00 Util Hello
19:00 jnthn o/
19:00 mberends o/
19:01 tadzik o~
19:01 pmichaud report:  I worked on profiling improvements and locating sources of slowdown, including the source of increased PAST object creation during compilation in nom.
19:01 pmichaud Drafted some messages for the master->ng branch rename.
19:02 pmichaud Worked on the ticket queue stuff a bit.
19:02 felher joined #phasers
19:02 pmichaud ran out of $tuits but will have more tomorrow.
19:02 pmichaud EOR
19:02 mberends pmichaud++
19:02 pmichaud I'm also mulling over what the recent parrot-dev discussion is likely to imply for rakudo and nqp.
19:04 mberends Did:
19:04 mberends * Closer study of per-test benchmarking.  The execution time of the old
19:04 mberends code dominated the shorter tests.  The quickest way for Rakudo to read
19:04 mberends the clock is nqp::p6box_n(pir::time__N).
19:04 mberends * With a modified lib/Test.pm (diff http://pastebin.com/tDqSx0gZ) the
19:04 mberends fastest tests execute in 9, 10 or 11 microseconds on a 2.67GHz i7.
19:04 mberends * Adding a reporting option to tools/test_summary.pl, good results so far
19:04 mberends Plan:
19:04 mberends * More reporting and analysis of benchmark results
19:04 mberends * Add times() into Rakudo for better accuracy
19:04 mberends .EOR
19:05 masak did: submitted grant proposal to TPF. blogged. plan to: push my macros branch. blockers: there's not enough time in the day. EOR
19:06 moritz anybody else wants to report anything?
19:06 jnthn I have one
19:06 tadzik I have a one-liner too :)
19:06 jnthn Over the last week...
19:06 jnthn * Recovered from YAPC, visiting, $dayjob speaking tour
19:06 jnthn * Re-worked boolification, so it's a bunch cheaper
19:06 jnthn * Put the serialization context code-gen on a diet, so we have a load less PAST nodes
19:06 jnthn * Also made us do less work when compiling to run immediately
19:06 jnthn * Fixed bug in statement-modifying for loop
19:07 jnthn * Improved method caching
19:07 jnthn * Tracked down and dealt with many method cache misses; NQP's build now has none, and the Rakudo build has a tiny fraction of what it once had
19:07 jnthn * Identified an issue in Parrot's stack tracing that was inefficeint and likely caused a ton of CPU cache misses, wrote to the parrot-dev list about it and was happy to see whiteknight++ jump on it and patch it
19:07 jnthn * Started sketching out an optimizer in the optimizer branch
19:07 jnthn * Started planning for the months ahead
19:07 jnthn In the coming week...
19:07 jnthn * Focus on stuff we need for nom-based release
19:07 jnthn * Start detailed planning for bounded serialization
19:07 jnthn * Try to finalize my current grant work
19:07 jnthn EOR
19:08 mberends jnthn++
19:09 jnthn (Side note: the optimizer branch is *not* aimed at being merged any time before the initial nom-based release.)
19:10 masak jnthn++
19:10 tadzik did: measured growing rakudo performance, nom and optimizer branches; plan: pass the exam; eor
19:10 masak good luck!
19:10 tadzik thanks, I'll need that :)
19:10 jnthn tadzik++ # Rakudo measurer in chief
19:11 pmichaud jnthn++ tadzik++ masak++
19:11 jnthn BTW, it was *very* -Ofun to me to be working on improvements and having somebody quickly tell me whether they'd worked.
19:11 pmichaud and even though he's not here....
19:11 pmichaud mls++ mls++
19:11 masak tadzik++ jnthn++ # the optimal duo
19:11 jnthn mls++ indeed
19:12 tadzik everyone++
19:12 jnthn :)
19:13 mberends does curing the hang in S02-builtin_data_types/instants-and-durations.t on 32 bit deserve its own mention in the punch list in NOMMAP?
19:14 jnthn yes
19:14 masak yes
19:14 jnthn oh, it's 32-bit?
19:14 pmichaud depends on the reason for the hang
19:15 pmichaud it might just deserve a regression and rt ticket.
19:15 masak is it "the eternal now"?
19:15 jnthn Well, it breaks "no"
19:15 jnthn er
19:15 mberends it hangs only on 32 bit
19:15 jnthn ...darn, I'll get famous for that word :P
19:15 jnthn It breaks "*now*"
19:15 jnthn mberends: 32-bit on any platform?
19:15 masak jnthn: NO
19:16 jnthn :P
19:16 mberends jnthn: I know only about Linux 32-bit atm
19:16 jnthn mberends: Well, it hangs on 32-bit windows.
19:16 jnthn Or at least, for me
19:17 jnthn Seems like a plausible correlation.
19:17 pmichaud what test is hanging, exactly?
19:17 jnthn pmichaud: Calling "now" hangs.
19:17 pmichaud that's pretty significant.
19:17 tadzik I think it's the fact that some int doesn't fit in 32 bits
19:17 pmichaud so yes, NOMMAP entry.
19:17 jnthn Yes, I think so too.
19:17 tadzik istr mls said that
19:17 pmichaud I wonder if it's bigint/libgmp related.
19:18 pmichaud fixing the int operations to autopromote to num would likely fix it.
19:18 pmichaud I've been meaning to do that but haven't gotten $tuit yet.
19:19 jnthn That may well be the case.
19:19 mberends can "* fix huge memory leak in simple while loop" also be removed from NOMMAP?
19:19 pmichaud is the leak gone?
19:20 mberends I've heard rumors to that effect
19:20 moritz that one is gone, yes
19:20 moritz mls++
19:20 masak \o/
19:20 jnthn I thought moritz++ removed that earlier today, after he bumped PARROT_REVISION?
19:20 moritz jnthn++
19:20 PerlJam It seemed gone as of this morning.
19:20 pmichaud did anyone try the "other" memory leak?
19:20 pmichaud (and do we care?)
19:20 jnthn Which "other" one? :)
19:21 moritz pmichaud: btw I've observed that the performance of iterating '1 for ^$int' grows nonlinearly for $int >= 20_000
19:21 pmichaud while 1 { nqp::say(pir::set__SP($foo)) }
19:21 jnthn Worth re-checking
19:22 pmichaud i.e.,  iiuc the bool fixes avoid the leak as opposed to fixing it.
19:22 pmichaud (by rerouting the get_bool vtable)
19:22 pmichaud but other vtables might still have a leak.
19:22 moritz pmichaud: that doesn't seem to leak on my system with newest nom
19:23 pmichaud I'm building newest nom now... will report back.
19:23 moritz pmichaud: there's also been an invocation-related memleak fix in parrot
19:23 pmichaud that could've been the source/fix
19:24 pmichaud (cost of iterating ^$int)  -- yeah, I suspect it's nonlinear with garbage collection
19:26 pmichaud there may be a reference hanging around that is keeping the full (iterated) range alive
19:30 moritz pmichaud: seems it's leaking very, very slowly
19:30 mberends dinner & # can't say nom because of ambiguity - don't name a branch 'dinner' ;)
19:30 moritz (like 1MB per minute or so)
19:31 pmichaud the "while 1 {...}" I gave above?
19:31 moritz the onewith nqp::say
19:31 pmichaud that's probably another leak
19:31 jnthn sounds like
19:31 PerlJam moritz: I've got it running in a loop right now and it seems to be occilating rather than leaking.
19:31 pmichaud the set__SP version I had before leaked incredibly fast -- faster than the while 1 { }  leak
19:32 jnthn ooh, I wonder what "dinner" could be an acronym for... ;)
19:32 pmichaud there's already "dynner"  :-P
19:32 moritz pmichaud: do you redirect output?
19:32 moritz sorry, meant PerlJam
19:32 PerlJam moritz: aye, >/dev/null
19:33 pmichaud I think I was able to get the leak even without the output;
19:33 pmichaud while 1 { pir::set__SP(1); 1; }  # also leaked before
19:33 pmichaud looking
19:33 PerlJam moritz: here's the mem usage for each of the last several minutes: 151384, 158580, 151384, 158580, 151384, 157552
19:34 PerlJam (my loop is while sleep 60; do ...; done)
19:34 PerlJam (i mean the outer loop that's checkign the mem usage)
19:35 pmichaud this is the one I had that leaked really fast as of 2011.08.18:   while True { pir::set__SP(1); Nil }
19:35 pmichaud and we knew that it wasn't the "while True" part that leaked.  :)
19:36 moritz pmichaud: all of those either stopped leaking, or are sleeping very, very slowly
19:37 jnthn sleeping? :)
19:37 moritz leaking
19:37 moritz guess I was thinking of :-)
19:38 masak Freudian parent slip ;)
19:38 moritz aye :-)
19:38 PerlJam oh ,the occilations have bumped up a little bit .. 159608, 152412, 159740, 153572
19:39 PerlJam those are the 12th minute to the 15 minute of execution
19:40 moritz might just be memory fragmentation
19:41 PerlJam maybe
19:43 PerlJam yeah, in the last few minutes it ratchetted back down a little bit.
19:50 PerlJam pmichaud, jnthn: one of you might want to hang out on #parrotsketch and participate in the discussion
19:59 pmichaud PerlJam: I'm a little hesitant to participate in the discussion.  My reactions to the change are a little mixed.
20:00 pmichaud but I will hang out to see what's being said.
20:04 PerlJam pmichaud: for me, if this latest round of changes doesn't turn into some momentum, I say Parrot is doomed.
20:07 * moritz should offer to test rakudo on parrot blead and report failures -- which is what he does anyway
20:08 PerlJam moritz: see #parrotsketch  :)
20:20 * PerlJam crosses his fingers for Rakudo+Parrot
20:25 TimToady it would appear that sphere is no longer leaking \o/
20:25 masak \o/
20:36 TimToady mind, it still runs about 50 times faster under niecza... :D
20:36 jnthn sphere?
20:42 PerlJam #ps seems to be getting less productive over time (as far as Rakudo + Parrot working in harmony)
20:43 TimToady jnthn: http://rosettacode.org/wiki/Draw_a_sphere#Perl_6 suitably modified for lack of MAIN
20:44 tadzik is there a lack of MAIN?
20:44 TimToady rakudo: sub MAIN { say "hi" }
20:44 tadzik I remeber fixing it a while ago
20:44 TimToady er
20:44 TimToady wrong channel
20:45 tadzik oh, I thought it broke
20:45 TimToady I guess it is there
21:01 colomon TimToady: I'm starting to think we may need a PPM / PGM / whatever library for p6.  ;)
22:01 samlh joined #phasers
22:55 [Coke] joined #phasers

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