Camelia, the Perl 6 bug

IRC log for #parrot, 2011-05-06

Parrot | source cross referenced

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

All times shown according to UTC.

Time Nick Message
00:13 dalek parrot/pdds_restart: d8c425c | Whiteknight++ | docs/pdds/draft/pdd00_pdd.pod:
00:13 dalek parrot/pdds_restart: first draft/rewrite of PDD 0, cut out some older stuff, be more explicit about certain things. Be clear that PDDs are the intended ideal designs
00:13 dalek parrot/pdds_restart: review: https://github.com/parrot/parrot/commit/d8c425c84a
00:18 Coke left #parrot
00:19 Coke joined #parrot
00:21 cotto_work whiteknight: I note that your changes to PDD 0 conflict with the way I'm using PDD 32 in the m0-spec branch.
00:21 cotto_work Do you think it'd be worthwhile to have Parrot Implementation Documents for that kind of information?
00:21 whiteknight we can hardly get our first set of P*D documents working correctly
00:22 cotto_work That's a valid concern.
00:22 whiteknight That might be worthwhile, eventually. I really want to focus on PDDs first
00:22 cotto_work I'll keep doing what I'm doing with the M0 pdd and put it in the right place later.
00:24 perlite left #parrot
00:24 whiteknight the PDD00 thing is still draft. I did more deleting of old details than adding of new details
00:25 whiteknight if the M0 doc still doesn't conform, we'll have to delete more details
00:26 cotto_work My goal in that doc is to say "here's everything you need to implement a conformant M0 vm".
00:27 whiteknight well, that doc may have a slightly different purpose than other PDDs
00:28 cotto_work That's probably true.
00:28 whiteknight most PDDs are only going to talk about the internals of Parrot specifically
00:28 whiteknight the PIR PDD might be in the same boat
00:28 whiteknight but we can re-evaluate PDD00 at any time, of course. I suspect the PDD review project will last quite a long time
00:29 cotto_work It's a good project for pretty much anyone.  It's good for newbies because they're better at exposing implicit assumptions and it's good for people who know the code because they know the code.
00:30 whiteknight yes
00:31 cotto_work http://www.gitalytics.com/user/detail/leto
00:31 Coke left #parrot
00:31 Coke joined #parrot
00:36 darbelo_ left #parrot
00:38 perlite joined #parrot
00:41 whiteknight I think we are going to want to put in a heck of a lot of effort into PDD01
00:42 whiteknight that has the potential to really become our mission statement of sorts
00:46 whiteknight anyway, I'm out for the night
00:47 whiteknight left #parrot
00:48 theory left #parrot
00:58 davidfetter left #parrot
01:01 atrodo whiteknight++
01:02 atrodo cotto_work> gitalytics is really neat
01:03 lucian joined #parrot
01:09 cotto_work atrodo: I just saw it fly by on hacker news.  It has a few raw spots, but there's a lot of potential and realized awesomeness.
01:11 theory joined #parrot
01:13 mtk0 left #parrot
01:20 mtk0 joined #parrot
01:20 sjn left #parrot
01:20 krunen left #parrot
01:20 sjn joined #parrot
01:24 krunen joined #parrot
01:27 lucian left #parrot
01:47 rurban_ joined #parrot
01:49 rurban left #parrot
01:49 rurban_ is now known as rurban
01:51 woosley joined #parrot
02:56 cotto ~~
02:57 atrodo =~
02:57 cotto new lappy is here
02:58 atrodo Hurray!  New toys are fun
03:04 hudnix left #parrot
03:08 cotto Awesome.  Apparently the ssd I bought is too thick.
03:09 cotto stupid first world problems
03:13 cotto I have angries.
03:17 dukeleto sadface
03:27 dukeleto msg plobsing i finally have the gcc40 machined configured to send smoke reports
03:27 aloha OK. I'll deliver the message.
03:31 dukeleto whoa, gitalytics is cool
03:32 jrt4__ joined #parrot
03:45 dalek parrot: ed43c68 | petdance++ | src/dynext.c:
03:45 dalek parrot: consting local vars
03:45 dalek parrot: review: https://github.com/parrot/parrot/commit/ed43c68ed0
03:45 dalek parrot: 7580e82 | petdance++ | src/dynext.c:
03:45 dalek parrot: lib_name cannot be NULL
03:45 dalek parrot: review: https://github.com/parrot/parrot/commit/7580e82ec1
04:11 dalek parrot/m0-prototype: 3dc9be0 | cotto++ | src/m0/m0_interp.pl:
04:11 dalek parrot/m0-prototype: zero all context registers during context initialization
04:11 dalek parrot/m0-prototype: review: https://github.com/parrot/parrot/commit/3dc9be0543
04:11 cotto dukeleto, it seemed like it'd be right up your alley
04:16 dukeleto cotto: yeah, quie spiffy
04:16 dukeleto quite, even
04:30 cotto I got the ssd installed, but I don't think it's ever coming out.
04:32 cotto seems to work fine
04:51 atrodo okay, i give up.  Anyone know how to cast in nqp?
04:52 tadzik cast from what to what?
04:53 atrodo Regex::Match to number
04:53 dalek parrot: 766efa7 | jkeenan++ | Configure.pl:
04:53 dalek parrot: Minor code reformatting.
04:53 dalek parrot: review: https://github.com/parrot/parrot/commit/766efa78a8
04:54 atrodo basically call the get_number that i see in the pir defination of regex::match
04:59 cotto atrodo, prepend +
05:03 cotto note that nqp uses N, not I
05:12 fperrad joined #parrot
05:14 atrodo cotto++ Looks like that was it.  Thanks
05:15 atrodo I'm curious why +0 doesn't work
05:17 atrodo I'm also curious why it was auto-casting sometimes, and not others.  but that's no here nor there at this point
05:19 cotto nqp has its own set of quirks
05:40 dukeleto msg plobsing "make test" passed on your tt1931 branch on debian/powerpc64
05:40 aloha OK. I'll deliver the message.
05:40 dukeleto msg plobsing now that I think about this, your branch needs darwin/ppc testing, not debian/ppc testing. Blarg.
05:40 aloha OK. I'll deliver the message.
05:49 cotto I need a name for my new laptop.
06:01 tcurtis ~~
06:07 moritz cotto: regarding calling to srand(), I think the original reason it was done automatically was to prevent complexity attacks on hashes
06:08 cotto moritz, the hash seed is still initialized from the system clock.
06:08 moritz ah
06:08 cotto and that was indeed the motivation
06:08 tcurtis ooc, does anyone here happen to both live in the Chicago area, and own one of those Captain Crunch whistles that were once used for phone phreaking? And also be willing to lend it to some college students Sunday for a scavenger hunt?
06:09 cotto Half of me wishes it grew up in those days.
06:09 cotto The other half is pretty happy with how things have worked out so far.
06:10 cotto The third half belongs to someone else entirely.
06:22 dalek parrot/m0-prototype: 2cdd5cd | dukeleto++ | / (5 files):
06:22 dalek parrot/m0-prototype: Add another test for invalid M0 and the beginning of version parsing
06:22 dalek parrot/m0-prototype: review: https://github.com/parrot/parrot/commit/2cdd5cd52e
06:24 cotto woot
06:24 cotto dukeleto++
06:26 dalek parrot/m0-prototype: 34e3c86 | dukeleto++ | / (2 files):
06:26 dalek parrot/m0-prototype: Actually parse version and add a test
06:26 dalek parrot/m0-prototype: review: https://github.com/parrot/parrot/commit/34e3c866d0
06:32 mj41 joined #parrot
06:40 dalek parrot/m0-prototype: c4f3c2d | dukeleto++ | src/m0/m0_assembler.pl:
06:40 dalek parrot/m0-prototype: Add a stub for generate_bytecode()
06:40 dalek parrot/m0-prototype: review: https://github.com/parrot/parrot/commit/c4f3c2d2dd
08:07 cotto dukeleto, are you around?
08:07 jrt4__ left #parrot
08:08 cotto Bah.  I need to sleep.
08:08 cotto 'night
08:09 dalek parrot/m0-prototype: f6c45b2 | cotto++ | t/m0/m0_interp.t:
08:09 dalek parrot/m0-prototype: add some interp tests to be unstubbed when the assembler is ready
08:09 dalek parrot/m0-prototype: review: https://github.com/parrot/parrot/commit/f6c45b29aa
08:20 cotto left #parrot
08:22 cotto joined #parrot
08:22 ShaneC joined #parrot
08:23 theory left #parrot
09:01 mtk0 left #parrot
09:08 mtk0 joined #parrot
09:08 ShaneC1 joined #parrot
09:11 utsl left #parrot
09:11 ShaneC left #parrot
09:21 ShaneC joined #parrot
09:22 utsl joined #parrot
09:24 ShaneC1 left #parrot
09:28 moritz what's the state of non-blocking sockets?
09:28 moritz I thought tcurtis or somebody was working on it...
09:30 tadzik there''s a ticket for Select PMC
09:30 tadzik I remember poking cotto about reviewing it again :)
09:39 woosley left #parrot
09:49 rurban_ joined #parrot
09:51 rurban left #parrot
09:52 rurban_ is now known as rurban
09:59 contingencyplan left #parrot
10:06 TiMBuS left #parrot
10:06 tcurtis left #parrot
10:06 KaeseEs left #parrot
10:10 birdwindupbird joined #parrot
10:48 woosley joined #parrot
11:32 bacek moritz, ping
11:32 bacek aloha, humans (btw)
11:33 moritz bacek: pong
11:33 bacek moritz, aloha
11:33 bacek moritz, can you rerun your benchmark on parrot master with patch from http://lists.parrot.org/pipermail​/parrot-dev/2011-May/005837.html ?
11:34 bacek (benchmark from "Rakudo compilation times from 3.0.0 to now")
11:34 moritz bacek: will do
11:34 bacek moritz, thanks!
11:41 KaeseEs joined #parrot
11:41 tcurtis joined #parrot
11:43 darbelo joined #parrot
11:51 mtk0 left #parrot
11:56 mtk joined #parrot
11:59 mtk left #parrot
12:04 mtk joined #parrot
12:07 birdwindupbird left #parrot
12:13 lucian joined #parrot
12:32 Patterner left #parrot
12:32 Psyche^ joined #parrot
12:32 Psyche^ is now known as Patterner
12:36 birdwindupbird joined #parrot
12:42 whiteknight joined #parrot
12:44 whiteknight good morning, #parrot
12:45 tadzik good morning whiteknight
12:45 moritz bacek: 26% speedup with the patch from the mailinglist
12:45 whiteknight 26%? Not bad at all
12:45 whiteknight that definitely puts the numbers into the range we thought they should have been in
12:46 moritz aye
12:50 hudnix joined #parrot
12:51 whiteknight good thing the Rakudo guys caught that regression, that's kind of embarrasing
12:51 pmichaud here are my results
12:52 pmichaud https://github.com/pmichaud/rakbench/blob/ma​ster/results/orange-x86_64-201105051910.txt
12:52 pmichaud moritz: 26% speedup was for ... ?
12:52 moritz pmichaud: the patch that bacek posted to the mailing list
12:53 pmichaud I mean, what was being timed?
12:53 pmichaud core.pm?  spectests?
12:53 moritz core.pm -> perl6 executable
12:53 pmichaud yeah, I only have a 12% speedup.
12:54 pmichaud which is definitely better than the 47% slowdown that 3.3 had
12:54 whiteknight that sin.t test has me befuddled
12:54 whiteknight I suspect strongly that it is not a GC-intensive test
12:54 pmichaud well, bacek's 3.3 patch did improve the timing on that substantially
12:55 pmichaud so gc is a part, but there's another part that is slowing things down somewhere
12:55 whiteknight and if GC got faster overall, that means we must have regressed somewhere else if the timing of that test stayed at 100%
12:55 whiteknight right
12:55 pmichaud there's definitely been some regressions in other parts of parrot, I think
12:55 pmichaud actually, I can prove it -- I can run the tests with gc=ms2 :-P
12:56 pmichaud that will show a huge regression
12:56 whiteknight yeah, we are going to have to track those down
12:57 whiteknight we should be significantly faster now, methinks
12:57 pmichaud I'm getting a new (supposedly very fast) computer later today, so I should be able to run more benchmarks on it
12:58 pmichaud anyway, the commontest stats at the bottom are also revealing -- we've still only improved by 1.1% since january
12:59 pmichaud (I've modified the script to run two trials there instead of just one... but that does add two hours to the time needed to run everything :-| )
12:59 whiteknight We're really feeling the absence of chromatic. He's always been very good keeping up with optimizations and hunting down hotspots
12:59 whiteknight it's not that nobody else does that kind of work, but nobody does it as regularly
13:01 pmichaud I'll build a ms2 version of 3.3 with bacek's patch, just so we can see the slowdown
13:02 bubaflub joined #parrot
13:02 whiteknight There should be a way to specify the intended gc core in pbc_to_exe, to save you the effort
13:02 whiteknight I don't know if there is, but there should be
13:02 bluescreen joined #parrot
13:03 pmichaud that feels like it wants a trac ticket :)
13:03 whiteknight let me double check things before I declare there to be a problem
13:06 darbelo left #parrot
13:08 bluescreen left #parrot
13:08 bluescreen joined #parrot
13:13 UltraDM joined #parrot
13:17 whiteknight no, pbc_to_exe does not have that option
13:17 whiteknight it really should
13:17 whiteknight I think it also really should be rewritten in NQP or Winxed :)
13:18 tadzik is that pir?
13:18 whiteknight yessir
13:19 atrodo winxed++
13:19 atrodo karma winxed
13:19 aloha winxed has karma of 2.
13:19 atrodo karma Winxed
13:19 aloha Winxed has karma of 2.
13:19 atrodo wow, that's it?
13:23 ambs joined #parrot
13:23 plobsing left #parrot
13:23 plobsing joined #parrot
13:24 coke_ karma parrot
13:24 aloha parrot has karma of 7.
13:24 coke_ I suspect that the "import old karma" step never happened. ;)
13:24 coke_ karma Coke
13:24 aloha Coke has karma of 195.
13:24 coke_ that cool refreshing drink
13:29 whiteknight I'm testing a quick patch that should allow pbc_to_exe to take a --gc= option
13:29 whiteknight then we can test Rakudo on different GCs without rebuilding Parrot, only changing the Makefile
13:35 kid51 joined #parrot
13:35 kid51 Trac unavailable with 503 error; have filed bug ticket with OSUOSL
13:37 lateau joined #parrot
13:39 kid51 Looks like Smolder needs a restart as well.
13:39 darbelo joined #parrot
13:46 whiteknight blarg
13:46 whiteknight I haven't contacted OSU yet
13:50 whiteknight ah nevermind. it appears kid51++ already has
13:50 kid51 I'm hoping we haven't been hacked.  The non-functioning of my account yesterday was very surprising.
13:51 kid51 But it's very early here on the Left Coast, so nothing will happen with the ticket for several hours
13:52 whiteknight yeah, that's probably true
13:52 kid51 afk
13:53 darbelo_ joined #parrot
13:53 darbelo left #parrot
14:02 ambs left #parrot
14:03 ambs joined #parrot
14:11 kid51 master w/bacek patch: linux/i386: make fulltest PASS
14:11 dalek parrot: 49f46ec | Whiteknight++ | tools/dev/pbc_to_exe.pir:
14:11 dalek parrot: Add a provisional --gc= option to pbc_to_exe.
14:11 dalek parrot: review: https://github.com/parrot/parrot/commit/49f46ec975
14:12 whiteknight pmichaud: I just committed a new --gc= option to pbc_to_exe. After you update and build with that, you should be able to change the gc used by rakudo with a simple change to the Makefile
14:13 whiteknight no more rebuilding parrot for comparisons
14:13 pmichaud whiteknight: excellent, thanks
14:13 pmichaud although that doesn't really help with the older releases :)
14:13 pmichaud and, at least from my experiments over the past few days, the overall timing is *very* sensitive to the parrot build itself
14:13 whiteknight we could create branches from the old tags and cherry-pick that commit, if we were feeling productive
14:14 pmichaud I prefer to benchmark the actual releases, though, instead of branches
14:14 whiteknight well, this should allow us to narrow down sources of fluctuations
14:14 pmichaud yes, I agree there.
14:15 allison Linus on GC: http://gcc.gnu.org/ml/gcc/2002-08/msg00552.html
14:15 moritz that's... not new at all.
14:17 allison moritz: yup
14:17 allison moritz: it does make sense though, the kernel is a very different use case than a high-level language implementation
14:18 moritz allison: agreed
14:18 pmichaud "Yes, I generalize.  Don't we all?"   Linus++
14:18 moritz oh, another autopun :-)
14:18 moritz where's masak? :-)
14:20 whiteknight I love his work on the kernel, but his opinions about most other computing topics are....off
14:21 whiteknight his rants against C++ for instance are humorous, but ignorant
14:21 moritz his rants against users of centralized version control systems are just stupid and offensive
14:21 whiteknight one good point he did make in that mail about GC was the fact that GC is typically very unfriendly to cache locality
14:22 woosley left #parrot
14:22 whiteknight moritz: yes, those too. We all understand that he made git and he likes git, but that doesn't mean it is the way, the truth, and the life
14:22 whiteknight refcounting, he correctly points out, is probably much better on the cache
14:23 moritz ... and comes with other problems we all know about
14:23 kid51 master w/bacek patch: darwin/ppc: make test PASS
14:23 ambs_ joined #parrot
14:24 ambs left #parrot
14:24 ambs_ is now known as ambs
14:27 whiteknight moritz: are refcount updates any more intrusive than GMS write barriers? It's not a clear-cut victory for either side
14:27 whiteknight plus, I suspect we could come up with gc algorithms which are more friendly on the cache
14:28 moritz whiteknight: I'm not talking about the updates, but about circular data structures
14:38 lateau1 joined #parrot
14:40 sorear one other problem with refcounting is that it requires locking (or atomic operations) in a multi-CPU case
14:41 sorear probably not much of a problem with Linux because it doesn't need to juggle links very often
14:43 lateau left #parrot
14:44 plobsing refcount updates can be done without locking on multi-cpu. atomic increment/decrement are usually (always?) implementable with the hardware's concurrency primatives
14:45 plobsing that would be a lot faster than locking
14:45 kid51 whiteknight: OSU OSL has restarted Trac, which enable me to do a password reset.  No further action needed on your part at this time.  Thanks.
14:46 NotFound The problem I see with a refcounting engine is that in general purpose language or machine, forbiding structrures with circular references is not a realistic option.
14:47 plobsing msg dukeleto I had hoped the problem might be an architecture-only problem. thanks for testing.
14:47 aloha OK. I'll deliver the message.
14:47 kid51 Hmm, maybe I spoke too soon.
14:48 plobsing NotFound: IIUC, python runs a cycle collector periodically to deal with that problem
14:49 NotFound How does that? Stoping the world?
14:49 lucian plobsing: it does
14:49 lucian NotFound: possibly, i'm not sure
14:49 lucian it has very little work to do, though
14:50 lateau1 left #parrot
14:52 plobsing lucian: http://docs.python.org/extending​/extending.html#reference-counts
14:53 UltraDM left #parrot
14:54 kid51 left #parrot
14:54 whiteknight I'm certainly not suggesting we move to refcounting. I'm only saying that Linus has a point about cache locality
14:54 whiteknight And I suspect a GC algorithm could be developed which is much more friendly to the cache
14:55 whiteknight for instance, using pointer prefixes to sort items into bins according to how close together they live, then marking bins at once
14:56 whiteknight if the item is in the current bin, mark it directly. Otherwise, put it in the bin to mark it later
14:57 pmichaud http://gist.github.com/959083   # results of rakbench on 2011.04 with gms and ms2
14:57 lucian whiteknight: i'm not convinced cache locality is the lowest hanging fruit
14:57 plobsing I'm sure the guys at Azul have some solution to the problem.
14:57 pmichaud click the "raw" link for an easier-to-read version
14:57 whiteknight lucian: I didn't say it was. I'm just saying that if it's a problem, I'm sure there are solutions for it
14:58 whiteknight lucian: I'm positive that our GC can be tuned and optimized pretty aggressively as it is, without worrying about changing the algorithm
14:58 lucian whiteknight: yeah. also, i'm not convinced refcounting is actually better with cache locality. i've yet to see conclusive evidence (perhaps i've missed it)
14:58 lucian that too
14:59 whiteknight lucian: it makes good logical sense. When you're using the object and have it in memory, you twiddle the refcount which is in the same place
14:59 whiteknight lucian: as opposed to GC, where you mark items that you are not currently using and do not have in cache
15:00 lucian whiteknight: sure, but you do it all the time, for all sorts of objects, in all sorts of places
15:00 lucian GC happens at one point, and the GC's own code gets cached just fine
15:00 whiteknight lucian: that doesn't make it right. Plus, GC doesn't need to be adding to a problem just because everybody else does it
15:00 plobsing refcounting has advantages and disadvantages. another refcount advantage is on non-memory resource handles, which don't map well to GC
15:00 * lucian nods at plobsing++
15:00 whiteknight GC is one of those things that should be low-level, tucked away, and fast
15:01 plobsing whiteknight: the problem with refcount collocated with the object is with COW forking
15:01 plobsing because twiddling the refcount marks the page dirty
15:02 mj41 left #parrot
15:02 whiteknight We don't currently use COW for pmcs, and likely won't in the future
15:02 whiteknight COW is a bear to deal with in a threaded context
15:02 plobsing parrot doesn't, but your kernel does
15:02 * dukeleto thinks our trac and website got borked because smolder ate all the memory
15:02 dukeleto smolder--
15:03 plobsing COW (and its effect on their zygote-forking approach) is why google went to such great lengths to separate their GC information from their objects
15:04 pmichaud what machine/os information would be good to capture in these benchmark reports?
15:04 moritz architecture, register width, OS, gcc version (IMHO)
15:05 pmichaud architecture == uname -i ?
15:05 TiMBuS joined #parrot
15:07 benabik left #parrot
15:11 whiteknight brb. firefox fail
15:11 whiteknight left #parrot
15:12 pmichaud flussence++ on #perl6 says "lscpu", so I'll use that.  :)
15:24 davidfetter joined #parrot
15:26 dukeleto msg cotto i added the PDS to the parrot calendar
15:26 aloha OK. I'll deliver the message.
15:26 hercynium joined #parrot
15:33 simcop2387_ joined #parrot
15:35 simcop2387 left #parrot
15:35 simcop2387_ is now known as simcop2387
15:46 theory joined #parrot
16:04 whiteknight joined #parrot
16:11 benabik joined #parrot
16:17 contingencyplan joined #parrot
16:20 lucian left #parrot
16:23 theory left #parrot
16:28 dalek parrot/m0-prototype: 237f12a | dukeleto++ | / (8 files):
16:28 dalek parrot/m0-prototype: Merge remote branch 'origin/m0-prototype' into m0-prototype
16:28 dalek parrot/m0-prototype:
16:28 dalek parrot/m0-prototype: Conflicts:
16:28 dalek parrot/m0-prototype: src/m0/m0_assembler.pl
16:28 dalek parrot/m0-prototype: review: https://github.com/parrot/parrot/commit/237f12a9b4
16:32 apt415 left #parrot
16:34 whiteknight interesting presntation about GC ane exceptions in Java: http://www.slideshare.net/eplawless/exception-sa​fety-and-garbage-collection-and-some-other-stuff
16:36 NotFound I'd like to see parrot running on this: http://www.raspberrypi.org/
16:38 whiteknight yes, that would be fun
16:42 whiteknight "If PHP gets something right before your language does, you should reassess your life goals"
16:43 whiteknight that's a quote I might keep in my heart forever
16:43 cotto Who said that?
16:43 whiteknight that's on that link I posted above
16:43 NotFound I don't think beating php at anything must be a language design principle.
16:43 whiteknight bottom of slide 16
16:44 cotto perfect
16:48 mtk left #parrot
16:50 mtk joined #parrot
16:57 birdwindupbird left #parrot
17:11 cotto_work ~~
17:21 dodathome joined #parrot
17:25 whiteknight left #parrot
17:41 whiteknight joined #parrot
17:49 rurban_ joined #parrot
17:51 rurban left #parrot
17:52 rurban_ is now known as rurban
18:46 cotto_work It's quiet.
18:46 benabik Too quiet.
18:47 ZOMBIES joined #parrot
18:48 plobsing 3 quiet
18:48 ZOMBIES BRAAAIIIIINNNSSSSS
18:48 ZOMBIES left #parrot
18:48 whiteknight it's depressing that a zombie looking for brains wouldn't hang out in #parrot for longer :(
18:49 cotto_work It could be anyone.  Mibbit's nice like that.
18:50 plobsing I like to think they got what they wanted and left
18:50 plobsing mission accomplished #parrot!
19:15 theory joined #parrot
19:20 dukeleto cotto_work: new jitterbug URL is http://jitterbug.leto.net
19:20 jevin joined #parrot
19:20 dukeleto jevin: welcome to #parrot
19:24 davidfetter left #parrot
19:27 ambs left #parrot
19:33 ambs joined #parrot
20:06 cotto_work dukeleto: what's wrong with the Rakudo build?
20:07 cotto_work Your jitterbug install doesn't show any successful builds.
20:17 tcurtis ~~
20:22 whiteknight hello tcurtis
20:24 davidfetter joined #parrot
20:25 ShaneC left #parrot
20:32 bluescreen left #parrot
20:33 dodathome left #parrot
20:39 ShaneC joined #parrot
20:41 plobsing an interesting data point for the 3.0-3.1 regression: valgrind shows the reverse. 3.0 ran 6,548,083,721 instructions vs 3.1's 6,068,114,582 on the Ωη benchmark which exhibits similar behaviour to rakudo. by valgrind, 3.1 is 7.3% faster than 3.0.
20:43 benabik plobsing: Did it spend more time in I/O?  Swap?  Just more expensive instructions?
20:43 plobsing benabik: that program doesn't really do any IO (besides slurping in a file) and I have swap dissabled
20:44 benabik plobsing: So the CPU just does less work in more time.  Boo.
20:44 plobsing I could understand cache playing a part in a disparity between valgrind results and wallclock timings, but this is just too big for that
20:46 whiteknight that's half a billion fewer instructions
20:46 plobsing yeah, like I said, it's huge
20:48 plobsing I'm firing up cachegrind to see if my cache hunch is accurate
20:49 benabik whiteknight: Modern processors execute something like 20 billion instructions per second.
20:50 plobsing benabik: only if they're running full-out. that rarely happens.
20:51 benabik plobsing: Yes, but a half billion still goes by pretty quick.
20:52 bacek left #parrot
20:55 jsut joined #parrot
20:58 benabik left #parrot
20:59 sjn left #parrot
20:59 dolmen joined #parrot
21:00 jsut_ left #parrot
21:00 whiteknight benabik: in modern computers, the CPU is rarely the bottleneck. If each of those half-billion instructions case a hit to memory or disk (even worse) that can take a hell of a long time
21:00 whiteknight I'm sure most of those aren't so naughty, but it can be worse than the CPU throughput numbers indicate
21:02 whiteknight left #parrot
21:02 bacek joined #parrot
21:05 ShaneC left #parrot
21:07 dukeleto cotto_work: jitterbug is messed up with respect to the rakudo test suite, nothing is wrong with rakudo
21:08 Tene "You can take my deterministic resource management when my cold dead hand goes out of scope." -- GC exceptions presentation
21:09 dukeleto lulz
21:11 cotto_work I don't know who eplawless is, but I want to hear him or her speak.
21:11 cotto_work http://www.ericlawless.com/
21:17 * cotto_work makes a note for future conferences
21:22 plobsing the only large cache difference I can see is a 1% increase in the branch mispredict rate, but that is partly due to a decreased amount of branching done
21:24 ambs left #parrot
21:25 hercynium left #parrot
21:29 cotto_work dukeleto: is jitterbug good enough that we could replace smolder with it?
21:31 Tene cotto_work: that's the eventual plan, at least.
21:34 cotto_work Is someone with a trac admin bit logged in?
21:35 cotto_work my password suddenly doesn't work
21:37 dukeleto cotto_work: i think all our admin accounts were lost or our trac instance got hacked. not quite sure
21:37 cotto_work That's not good.
21:37 dukeleto cotto_work: one difference between smolder and jitterbug is that jitterbug is a dedicated client where smolder allows anybody to submit reports. It is a slightly different model
21:38 dukeleto cotto_work: yeah, it's not good.
21:38 dukeleto cotto_work: kid51++ noticed first, he sent a ticket to OSUOSL already
21:38 cotto_work ok
21:38 cotto_work kid51++
21:40 darbelo joined #parrot
21:40 darbelo_ left #parrot
21:42 cotto_work It's times like this that I'm glad I use a generated password and a password manager.
21:43 cotto_work At least the drupal site is fine.
21:44 dukeleto cotto_work: yep, me too
21:45 cotto_work dukeleto: did you try resetting your password?
21:46 darbelo left #parrot
21:48 dukeleto cotto_work: i haven't touched trac yet, just saw some emails fly by
21:51 darbelo joined #parrot
21:52 darbelo_ joined #parrot
21:52 darbelo left #parrot
22:01 dukeleto cotto_work: the OSL peeps seem to think it was a full disk a
22:01 dukeleto cotto_work: does your old password work now?
22:02 cotto_work dukeleto: no joy
22:09 dalek parrot/m0-prototype: 71edc13 | dukeleto++ | / (2 files):
22:09 dalek parrot/m0-prototype: Add a test for detecting an invalid M0 version
22:09 dalek parrot/m0-prototype: review: https://github.com/parrot/parrot/commit/71edc137d7
22:09 dalek parrot/m0-prototype: 0137f68 | dukeleto++ | / (4 files):
22:09 dalek parrot/m0-prototype: Add test for invalid bytecode and add invalid version m0
22:09 dalek parrot/m0-prototype: review: https://github.com/parrot/parrot/commit/0137f68d8d
22:09 dalek parrot/m0-prototype: 02b432b | dukeleto++ | / (2 files):
22:09 dalek parrot/m0-prototype: Parse op table and add a test
22:09 dalek parrot/m0-prototype: review: https://github.com/parrot/parrot/commit/02b432bb77
22:27 dalek parrot/m0-prototype: c544f51 | cotto++ | t/m0/m0 (2 files):
22:27 dalek parrot/m0-prototype: rename a test for better clarity and tab completion
22:27 dalek parrot/m0-prototype: review: https://github.com/parrot/parrot/commit/c544f51bc8
22:32 darbelo_ ~~
22:58 fperrad left #parrot
22:59 darbelo_ left #parrot
23:07 bubaflub left #parrot
23:15 mtk left #parrot
23:16 wagle http://trac.parrot.org/parrot/wiki/LoritoRoadmap doesnt say what lorito is?
23:17 dukeleto wagle: http://trac.parrot.org/parrot/wiki/Lorito
23:18 PerlJam maybe occurences of "Lorito" in that first one should link to the second one  :)
23:19 wagle yeah, something like that..  at least one occurance..  8)
23:21 dalek winxed: r975 | NotFound++ | trunk/winxedst0.cpp:
23:21 dalek winxed: fix an addparent misindentation in stage 0
23:21 dalek winxed: review: http://code.google.com/p/w​inxed/source/detail?r=975
23:23 * wagle tries to do it himself..  waits..  waits
23:31 kid51 joined #parrot
23:33 wagle apparently trac.parrot.org doesnt want to let me in..  i just sit and wait to login
23:33 kid51 wagle: You are not alone ... either in the universe or your inability to login.
23:34 wagle 8)
23:36 kid51 Just updated OSUOSL on the situation
23:37 kid51 The wiki is viewable but not log-in-able hence not editable.
23:40 kid51 At least git has not been touched by these troubles.
23:41 cotto_work nope.  git++ and github++
23:42 kid51 Oh, that's right, our repository is NOT at parrot.org
23:44 wagle oh look!  trac woke up!
23:44 kid51 Huh? Can you login?  I cannot.
23:45 wagle it sent me two new passwords..  neither work
23:46 kid51 This morning, after it (briefly) came back to life, it sent me a new password .... which no longer works.
23:46 kid51 So I doubt we're cured.
23:52 wagle wheee

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

Parrot | source cross referenced