Parrot 3.2.0 released | parrot.org | Log: irclog.perlgeek.de/parrot/today | Parrot is accepted for GSoC 2011! Student application deadline is Apr 8
Set by moderator on 27 March 2011.
00:02 dngor left 00:06 whiteknight joined
whiteknight good evening, #parrot 00:12
bubaflub evening whiteknight 00:18
whiteknight hello bubaflub 00:19
00:29 mikehh left
dalek rrot: 6853f88 | petdance++ | include/parrot/compiler.h:
close off unused the way clang likes it
 00:37
00:42 mikehh joined 00:51 mikehh left 00:59 loufoque joined
loufoque hi, i've looked at this document and was astonished www.parrot.org/dev/irc-logs 01:00
oops wrong link
docs.parrot.org/parrot/latest/html/...s.pod.html
normalization forms apply to combining character sequences, not grapheme clusters.
therefore this NFG thing is just weird 01:01
does it store "\\r\\n" as an entry in the table too? 01:02
sorear It doesn't do any of that yet. 01:04
Most of the PDDs are "wishful thinking"
*yet (if ever)
whiteknight NFG was implemented as a GSOC project last year, but never merged 01:05
loufoque: \\r\\n would probably not be in the table
loufoque plus dealing with variable-width encodings does not necessarily require expensive lookaheads, you can just work at the code point or even code unit level and check the match lies on the right boundaries, which is pretty cheap.
01:06 contingencyplan left
loufoque \\r\\n is a single grapheme though 01:06
whiteknight loufoque: I don't remember the details, but each grapheme was treated as a tuple with the character and combiners
The benefit is that we gain linear-time lookups by moving the grapheme complexity to the lookup table 01:07
loufoque if you want to search for "foo" in "foo<some_combining_character> bar foo baz" you could just look for foo, test if it lies on the boundaries you want, and if it doesn't discard the match and keep searching. So you'd find the first foo but discard it, but then you find the second one. This is basically just adding a boundary check (which is constant-time) as part of your comparison criterion. And everything can be done at the byt 01:15
e level.
just an idea, I don't know why it wouldn't be linear anyway
dalek sella: 9fff443 | Whiteknight++ | src/prototype/Protoclass.winxed:
Add in a new mechanism for handling prototypes using a more JavaScript-like model. The basic idea is based on code posted by NotFound++, with a few additions like invokability for sub-like objects
 01:16
01:18 dngor joined
whiteknight loufoque: for details, you probably want to talk to darbelo. He implemented the NFG thing for GSoC last year 01:19
01:20 estrabd left, estrabd joined
plobsing loufoque: I'm not sure of the details, but it seems your scheme would not properly match characters with multiple combiners in different orders 01:28
loufoque plobsing: yes, the strings need to be normalized of course 01:29
or at least canonically ordered 01:30
yes they need to be normalized, be it NFC or NFD
plobsing which requires a preprocessing pass anyways. might as well build a table of characters.
loufoque yes, you can, but the NFG system isn't necessarily necessary, nor is it necessary the best solution, nor is it a silver bullet that solves everything 01:37
whiteknight no normaliztion form solves every problem 01:38
but NFG is an interesting tool to use, especially in a VM which commonly does operations on graphemes 01:39
01:41 whiteknight left
cotto bacek, ping 01:47
01:48 bacek left 01:59 bacek joined
cotto aloha, clock? 02:03
aloha cotto: LAX: Sat, 19:03 PDT / CHI: Sat, 21:03 CDT / NYC: Sat, 22:03 EDT / UTC: Sun, 02:03 UTC / LON: Sun, 03:03 BST / BER: Sun, 04:03 CEST / TOK: Sun, 11:03 JST / SYD: Sun, 12:03 EST
dalek rrot: 49b5e9d | petdance++ | include/parrot/api.h:
Added PARROT_ASSERTS_ON to tell if assertions are on
 02:06
rrot: 9cb8686 | petdance++ | src/gc/gc_gms.c:
quieting unused args.
soh_cah_toa i just wanna make sure...we don't use svn anymore, right? 02:18
cotto soh_cah_toa, not for parrot. The languages repo is still svn. 02:19
soh_cah_toa languages repo? never heard about that before
cotto languages used to live in the main repo. They were kicked out of the nest a while ago. 02:20
soh_cah_toa alright. well, i was just just readin the GSoCersStartHere wiki and it still mentioned svn. i was gonna replace it w/ git but i wanted to make sure first 02:21
cotto go ahead 02:23
soh_cah_toa++
soh_cah_toa alright. btw, how long after i email the contributions cla can i get commit access?
b/c i have some changes i'd like to submit by tomorrow 02:24
cotto soh_cah_toa, you can always fork on github 02:25
we love pull requests
CLAs need to be verified and you need to have someone propose that you get a commit bit at a #ps meeting. 02:26
soh_cah_toa yeah i did that. i'm gonna need some help w/ that tomorrow
bubaflub soh_cah_toa: the pull request? 02:27
soh_cah_toa well maybe i can figure it out now...so i've made a pull request made changes. once i commit, what's my next step?
cotto have you read help.github.com/pull-requests/ ? 02:31
02:32 dafrito left
soh_cah_toa nopaste, but i think that is what i'm looking for 02:32
cotto lawl
soh_cah_toa agh...did it again
cotto your client might be configured to autocomplete words ending with a comma 02:33
02:33 plobsing left
cotto istr something similar with xchat 02:33
soh_cah_toa yeah
are pull requests any different than branches? 02:34
correct me if i'm wrong but it looks like pull request are specific to github, not git
cotto soh_cah_toa, right 02:35
soh_cah_toa so pull request == branch? 02:36
cotto If you think you'll make more than one pull request, it's good practice to have one branch per pull request
no
02:37 dafrito joined, woosley joined
soh_cah_toa okay i think i got it now...a pull request is like an "almost commit". it show everybody what i'm going to change. then one of you guys makes the actual commit. close? 02:41
cotto you commit and push to a clone of parrot.git, use github to submit a pull request for whatever commits you want pulled, then we'll review and commit/push to parrot/parrot 02:42
soh_cah_toa alright i get it 02:44
cotto great, because I have to take off in a few minutes
soh_cah_toa okay
bubaflub soh_cah_toa: yes, i just merged in a pull request for cardinal earlier today 02:55
soh_cah_toa: basically, you branch, hack away, and then when you do a pull request that's a signal to parrot devs that this is good to merge
soh_cah_toa: a parrot dev can come in and take a look, comment, and merge your commits
soh_cah_toa okay, so when i issue a git commit, it get pushed to the clone i made w/ the pull request? 02:56
Tene soh_cah_toa: "pull request" is a pretty normal git workflow, it's just that github has specific UI around the process. The kernel devs use pull requests, just done through email instead of a web UI. 03:03
soh_cah_toa okay. that's for general changes though, right? if i fix something in a trac ticket, i just submit a patch by attaching it to the ticket. right? 03:06
Tene soh_cah_toa: that works fine too. Simple things aren't much harder to deal with through patches. It's for the complicated stuff involving multiple commits where working with it via git natively is a big win. 03:07
soh_cah_toa okay. b/c i have two things to fix. i have some changes i've made to some docs and i have some changes to fix ticket #236 03:08
should i put everything in one pull request or separate or what? 03:09
ooops, not #236
#1215 i think
bubaflub soh_cah_toa: two separate commits, if you please 03:12
soh_cah_toa: that way it's easier to revert if someone needs to
soh_cah_toa: not saying that we will, but make the commits logical chunks of work
soh_cah_toa i figured since they address two separate issues
bubaflub soh_cah_toa: exactly. they aren't related to each other, can be applied separately
soh_cah_toa right 03:13
bubaflub soh_cah_toa: and the main idea is that we never break the build on the master branch
soh_cah_toa right
bubaflub soh_cah_toa: for bigger work, open up a new branch and hack away on that. when you're done, send a pull request
soh_cah_toa: for small work, patches on tickets work just fine
soh_cah_toa alright, i think i understand 03:14
bubaflub soh_cah_toa: great. and once you get your CLA in, all of this won't matter...
soh_cah_toa: you can just commit directly
Coke finally catches up 03:15
soh_cah_toa oh okay. that's just for people who aren't necessarily apart of the dev team yet who'd like to submit various changes
after the cla, you become a "trusted member" 03:16
cool 03:18
bubaflub soh_cah_toa: bingo
soh_cah_toa i always liked being a lone coder but i'm actuualy finding the whole process of collaborating w/ others to be pretty neat 03:20
i mean, i've read books on git before but i never really had a use for it. now i do and it all makes sense. very effecient and organized. i like it 03:21
bubaflub soh_cah_toa: at work, every commit i make is reviewed by another coder 03:26
soh_cah_toa: it's real nice - learn lots
soh_cah_toa bubaflub: what kind of software do you develop? 03:28
bubaflub: at work, i mean 03:29
bubaflub soh_cah_toa: web development stuff in perl
soh_cah_toa: which is to say all kinds of things - jQuery, HTML, CSS, Catalyst, DBIx::Class
soh_cah_toa bubaflub: wow, i didn't think perl was used much as an enterprise language 03:30
bubaflub: i'm jealous. i wish i had a job writing perl 03:31
03:33 woosley1 joined 03:34 woosley left
bubaflub soh_cah_toa: it ain't as sexy as other hip languages 03:35
soh_cah_toa: but i love it
03:35 krunen left
soh_cah_toa yeah, perl's my little baby 03:36
03:41 Khisanth left
bubaflub goodnight #parrot 03:46
soh_cah_toa see ya
03:46 bubaflub left 03:47 plobsing joined 03:51 Khisanth joined
dalek tracwiki: v6 | soh_cah_toa++ | GSoCersStartHere 03:55
tracwiki: Originally only meant to remove svn references. Ended up editing the whole thing. Added a few links, though I don't think the mailto link are working. No biggie.
tracwiki: trac.parrot.org/parrot/wiki/GSoCers...ction=diff
tracwiki: v7 | soh_cah_toa++ | GSoCersStartHere 04:11
tracwiki: Fixed mailto links. Also added link to GSoC page which I forgot before. It's overkill, I know.
tracwiki: trac.parrot.org/parrot/wiki/GSoCers...ction=diff
tracwiki: v11 | soh_cah_toa++ | WhereIsIt
tracwiki: Added link to GitHub repository.
tracwiki: trac.parrot.org/parrot/wiki/WhereIs...ction=diff
04:12 mikehh joined 04:14 mikehh left
bacek cotto, pong 04:18
04:23 mikehh joined
dalek tracwiki: v31 | soh_cah_toa++ | NewParrotDeveloperGuide 04:42
tracwiki: Added some links to a few pages which I think are really important to newcomers (like me).
tracwiki: trac.parrot.org/parrot/wiki/NewParr...ction=diff
soh_cah_toa says goodnight to parrot 04:55
04:55 soh_cah_toa left
cotto bacek, do you think you'd have the tuits to mentor the newPOST gsoc project? 05:37
bacek cotto, I hope so. I even added my self to melange's "I would like to mentor this project" list
cotto bacek, great. If you don't I can, but I think you're better qualified. 05:38
bacek I'll mentor it. No worries
cotto I'll take the debugger task then. I saw on the gsoc list that a major determinant of how many slots we get is how many tasks have potential mentors. 05:40
bacek I think we have mentors for all valuable tasks.
cotto I wish melange would let me sort by which tasks have mentors and which ones I've voted on. 05:41
05:43 edu joined, edu is now known as Eduardow_ 05:44 Eduardow left 05:45 Eduardow_ is now known as Eduardow
cotto I'd like to see the parser generator and the java bytecode translator happen. 05:48
bacek I'm not sure about "java bytecode translator". What the point? 05:51
cotto interesting potential for interop 05:52
bacek heh. Can we have "native-to-parrot-languages" interop happened first? 05:53
e.g. rakudo<->cardinal<->particle
cotto touche 05:55
Tene I never looked closely at the java bytecode project, but did they look at jnthn's previous work on the subject? 05:59
cotto I pointed it out in the melange comments
seen bbantha
aloha Sorry, I haven't seen bbantha.
Tene declines to repeat his traditional comments about interop.
cotto That every time you've done work, the rest of us haven't shown much interest? 06:00
Tene cotto: it's not very useful for me to continue to be grumpy over things that happened years ago, and were entirely reasonable. I need to get over it. 06:03
sorear Who cares about interop? 06:11
06:11 Eduardow left
cotto sorear, was that serious? 06:12
My ISP filters out the sarcasm bit. 06:14
06:14 Eduardow joined
bacek sorear, I have another question about Grammars, NQP, Universe and everything. 06:14
sorear I'm not sure whether it counts as "sarcasm" or not
but I know the answer
bacek: go...
bacek sorear, for example. I have something like (foo)(bar). How to distinguish between prefix<()>.circumfix<()> and circumfix<()>.postcircumfix<()>? 06:16
bacek hates C...
sorear bacek: you can't. 06:17
the C grammar is fundamentally ambiguous with casts
I think the sensible thing to do is to check if foo is a valid type 06:18
bacek sorear, sigh... Any ideas about how to workaround this? Or I have to to some "magical checks" after parse?
sorear valid types always start with a type keyword (you know the list) or a typedef name
bacek "foo" can be typedefed. Which mean I need full "cpp"
sorear typedefs aren't part of cpp
#define pie typedef 06:19
pie int cow;
bacek yes. But for opsc it means I have to parse all parrot headers before parsing ops bodies 06:20
which is kinda slow and painful
sorear C sucks like that
ever notice how slow building parrot is? gcc is suffering from the same problem you are
bacek opsc is few orders of magnitude slower than gcc. 06:21
looks like I have to cheat somehow and bring "well known definitions" closer to opsc... 06:22
next question about PASTing: what is best practise for PASTing postcircumfix/postfix without explicitly define them in Grammar? 06:23
e.g. "foo(bar)". is term<identifier>.postcircumfix<()>
cotto bacek, can you restrict the C used by opsc to make parsing headers unnecessary?
bacek postcircumfix<()> will create PAST::Op(<call>). How "term<identifier>" can find that EXPR has postcircumfix part? 06:24
cotto, " I have to cheat somehow and bring "well known definitions" closer to opsc" 06:25
cotto how clever of me
bacek sorear, any thoughts on "postcircumifix pasting"? 06:27
sorear bacek: I'm not sure what you're getting at. This is what EXPR is for. 06:28
bacek sorear, I have to inverse order of evaluation. "foo(bar)" should produce PAST::Op(:name<foo>, PAST::Var(:name<bar>)). 06:29
"(bar)" will create "PAST::Op(:type<call>, PAST::Var(:name<bar>))". 06:30
method term:identifier (for "foo" part) have to check is there is postcircumfix exists and use it.
or, let's rephrase - when I have to create PAST::Op(:type<call>) node? 06:31
06:42 theory left
sorear bacek: think about the function pointer case 07:02
bacek sorear, yes... My final goal is properly parse/past foo->bar->baz(quux)
e.g. VTABLE_foo(interp, pmc) is actually pmc->vtable->foo(interp, pmc) 07:04
sorear bacek: in the general case "foo" needs to be parsed as an expression 07:06
because it might be "int (*foo)();"
in which case you want to generate a call with args (foo) and (bar)
bacek sorear, yes. But for now I want simple case done first.
e.g. foo->bar() should produce something meaningful 07:07
with postfix< -> >, postcircumfix<()>
looks like I have to overridde "method EXPR" to handle postfixish things automatically 07:10
07:10 fperrad joined
benabik I should be very asleep but the CAPTCHA image on parrot.org's drupal reg is broken: error on line 1 at column 14: String not started expecting ' or " 07:12
Erm. User registration page: www.parrot.org/user/register 07:14
I'm gonna pass out now though.
cotto awesome 07:20
by which I mean the opposite of awesome
benabik, it works fine for me 07:22
I had to submit a form without the captcha once, but it appeared after that 07:23
either way, you probably don't need a parrot.org account unless you're planning on having a blog there. 07:24
07:30 Themeruta joined, NotFound left 07:50 lateau joined, Themeruta is now known as NotFound 08:09 mj41 joined 08:43 contingencyplan joined 08:47 Eduardow left 09:46 lucian joined 10:09 lucian_ joined 10:11 lucian__ joined 10:12 lucian left 10:15 lucian_ left 10:39 ligne left 10:49 sirmacik left 11:01 whiteknight joined
whiteknight good morning, #parrot 11:08
dalek nxed: r944 | NotFound++ | trunk/winxedst0.cpp:
some refactor in stage 0 Modifier classes to make them more like its stage 1
 11:10
11:27 lucian__ left 11:34 gbacon joined 11:40 mtk left
moritz gets some rakudo segfaults on newest parrot 11:43
$ ./perl6 t/spec/S03-sequence/arity-2-or-more.t
1..19
ok 1 - arity-2 Fibonacci
Segmentation fault
nopaste "moritz" at 192.168.1.3 pasted "segfault" (28 lines) at nopaste.snit.ch/39787 11:45
11:45 lucian_ joined 11:48 mtk joined 11:51 lucian_ left 11:55 Patterner left, Psyche^ joined 11:56 Psyche^ is now known as Patterner
dalek nxed: r945 | NotFound++ | trunk/winxedst0.cpp:
use an Argument class instead or raw Expr in ArgumentList in stage 0
 11:56
11:56 kid51 joined 12:03 lucian joined
dalek tracwiki: v8 | jkeenan++ | GSoCersStartHere 12:12
tracwiki: trac.parrot.org/parrot/wiki/GSoCers...ction=diff
tracwiki: v9 | jkeenan++ | GSoCersStartHere
tracwiki: trac.parrot.org/parrot/wiki/GSoCers...ction=diff
12:27 lucian_ joined 12:30 lucian left
whiteknight moritz: interesting 12:38
moritz: is that a recent issue?
I mean, is it from a change today, or earlier?
12:46 gbacon left
moritz whiteknight: I don't know 12:55
I get excessive memory usage while trying to bisect parrot (and run rakudo on top of it) 12:56
ie enough to OOM with my ulimit -v of 1.5G 12:57
NotFound moritz: Can you try winth a non optimized build, to get a more detailed backtrace? 12:58
dalek nxed: r946 | NotFound++ | trunk/winxedst0.cpp:
argument modifiers in stage 0
moritz NotFound: I'm bisecting first, but will do that too
whiteknight shit, I hope thats not related to the imcc_compreg_pmc merge 12:59
dalek nxed: r947 | NotFound++ | trunk/winxed (2 files):
now that stage 0 allow modifiers, use them in the non installed driver, and fix
 13:04
rrot: 99786be | Whiteknight++ | / (4 files):
remove some cargo-cult nonsense in the packfile api. Since the packfile PMCs are GC-protected, use them more and don't pass raw PackFile* pointers back from the compiler
 13:06
rrot: 9ec9ca8 | Whiteknight++ | src/ (3 files):
fixup the IMCCompiler.compile method. compile method and invoke vtable return an Eval PMC. compile_file returns a PtrObj PMC for a PackFile*. this situation isn't ideal, but we need other fixes in the system before we can saneify it
nxed: r948 | NotFound++ | trunk/winxedst1.winxed:
remove the drivers helper methods in the compreg'ed compiler object that are no
 13:14
nxed: r949 | NotFound++ | trunk/pir/winxed_compiler.pir:
update installable compiler
moritz can somebody please run git describe --tags on HEAD and tell me the result?
NotFound moritz: RELEASE_3_2_0-344-g9ec9ca8 13:15
kid51 $ git describe --tags 13:16
RELEASE_3_2_0-334-gabe179c
NotFound kid51: My HEAD is more head than yours :P
moritz thanks 13:18
13:19 lateau left
moritz somehow my bisecting takes me to RELEASE_3_0_0-605-g19f0540 which makes no sense at all 13:19
becasue I took 3.0.0 as the good and HEAD as the bad rev
and another copy of parrot which I used for the unoptimized build was on an old branch :/ 13:21
kid51 moritz: I had a situation a few months ago where 'git bisect' took me to a revision that was *before* the older of the two revisions from which I was starting. 13:23
So I had to conduct the bisecting entirely manually.
Is that's what
's happening with you?
NotFound Bisecting non lineal structures can be hard. 13:24
13:27 lucian_ left 13:30 lateau joined
dalek rrot: 0f0dc31 | Whiteknight++ | src/pmc/imccompiler.pmc:
comment out method stubs in IMCCompiler which have not been implemented yet
 13:44
13:48 lateau left
moritz kid51: seems like 13:57
kid51: I had to skip one revision, guess it didn't like it
without --optimize, rakudo build fails for me 14:02
$ /home/moritz/tmp/rakudo/parrot_install/bin/parrot src/gen/perl6.pbc --target=pr \\ src/gen/core.pm > src/gen/core.pir
PackFile_unpack: Buffer length 0 is shorter than PACKFILE_HEADER_BYTES 18.
make: *** [perl6.pbc] Error 1
RELEASE_3_2_0-330-gae4f8e1 is bad 14:04
dalek nxed: r950 | NotFound++ | trunk/winxedst (2 files):
refactor common parts of Return and Yield statements into a base class in stage
 14:06
moritz on, first it segfaults during build (uoptimized), then when I restart it complains about an empty packfile 14:12
14:24 ambs joined 14:25 JimmyZ joined 14:47 kid51 left 14:58 particle1 is now known as particle 15:06 plobsing_ joined 15:08 JimmyZ left 15:09 theory joined 15:10 plobsing left 15:16 he left, he joined
dalek nxed: r951 | NotFound++ | trunk/winxedst0.cpp:
refactor function parameters in stage 0
 15:18
15:23 he left, he joined 15:24 mikehh left 15:29 woosley1 left 15:49 Eduardow joined
dalek rth: eb5a59d | plobsing++ | forth/forth.pir:
Exception isn't array-based
 15:54
16:11 mikehh joined 16:26 mtk left
NotFound The rakudo problem seems to be caused by a GC error: a ResizablePMCArray data pointer is NULL. 16:42
Looks like it's been prematurely collected.
whiteknight damnit, that's the same kind of error I was seeing in the imcc_compreg_pmc branch 16:44
I thought I had all those instances fixed
moritz NotFound: great that you found it. My attempts to bisect it have repeatedly killed several of my desktop applications (due to OOM)
whiteknight well, there was actually only one bug that I thought was fixed
NotFound moritz: I used the same backtrace you posted before, but in a non optimized build. 16:45
Uh, no, not the same backtrace, the same core dump during the build. 16:49
whiteknight starts building rakudo with a fresh parrot 16:50
NotFound I'm thinking about using a PMC flag in debug builds, if set emit a diagnostic message when collected. 16:53
16:54 ligne joined
whiteknight hmm...I got a segfault during the rakudo build. I'm trying again without threads 16:54
NotFound whiteknight: the problem I've found is in that segfault. 16:55
whiteknight okay, that FIA has a NULL ->data 16:57
and that is a problem
NotFound FIA? Mine was in RPA. 16:59
16:59 ambs left, ambs joined
whiteknight Mine is an FIA that's coming out of the constants table while saving bytecode to file 17:01
which indicates to me that a packfile is being collected somewhere
NotFound Mine is the todo attribute in imageiosize.pmc 17:02
Also while saving bytecode.
17:03 ambs_ joined, ambs left, ambs_ is now known as ambs
whiteknight I think I have it fixed 17:23
a little bit of a hack
at this point we need to tread water until we have properly GCable packfiles
...no, nevermind. Not fixed 17:24
17:25 ambs_ joined, ambs left, ambs_ is now known as ambs
whiteknight the packfile PMCs are all being GC registered to prevent premature collection, so that's not the problem 17:27
cotto ~ 17:33
dalek rrot: e53b5c5 | mikehh++ | src/packfile/api.c:
fix codetest failure - add missing ASSERT_ARGS()
 17:42
rrot: 8869eca | mikehh++ | src/pmc/imccompiler.pmc:
fix codetest failure - add missing documentation
17:46 dodathome joined
dalek rrot: 0573725 | mikehh++ | include/parrot/compiler.h:
fix codetest failure - wrap macro argument
 17:52
18:02 he left 18:03 he joined 18:08 ambs left, bubaflub joined, ambs joined 18:11 ShaneC left 18:13 petdance joined 18:28 ShaneC joined 18:29 hudnix left 18:36 mj41 left
whiteknight NotFound: ping 18:47
NotFound whiteknight: pong 18:49
whiteknight NotFound: this rakudo bug is definitely GC related. If I run parrot with -G I don't see segfault
NotFound: can you send me a copy of the backtrace you are seeing?
NotFound whiteknight: the bizarre thing is that I don't see any call to destroy on that pmc. 18:50
whiteknight really? that is weird
cotto which bug? 18:51
NotFound And more strange, supposedly in a debug build data should be set to 0xdeadbeef, not to NULL.
nopaste "NotFound" at 192.168.1.3 pasted "Backtrace of segfault during rakudo build" (31 lines) at nopaste.snit.ch/39930 18:53
whiteknight cotto: rakudo is segfaulting in random ways
cotto great
NotFound whiteknight: the line numbers in imageiosize are a bit shifted because of my debug statements.
whiteknight ok 18:54
okay, no PackFile PMCs are ever being destroyed
they are all properly registered like I thought
cotto: and it's dealing with packfiles, at least in the versions NotFound and I are seeing. I have to suspect the imcc_compreg_pmc merge 18:55
cotto that's the only big thing to happen recently 18:56
NotFound Uhhhh..... I've inserted PARROT_ASSERT(todo->data); in ImageIOSize init_pmc... and it fails. 18:57
whiteknight ...wat?
that's not possible 18:58
I mean, it's possibe
but damnit
NotFound Something might be wrong in some macro. 19:00
whiteknight this bug is absurd 19:06
on my machine, PMC constant #2596 / 2831 is corrupted 19:07
but the first 2595 pmcs freeze just fine
what's that command to randomize memory layout? 19:10
cotto you mean hash seed?
whiteknight no, chromatic had a command that you could use to cause linux to randomize the memory space
so the next program run wouldn't use the same pages in the same order
19:10 lucian joined
cotto I didn't know about that. 19:11
NotFound whiteknight: the push_string vtables in the imageio PMC look suspicious.
benabik whiteknight: gcc.gnu.org/wiki/Randomization ? 19:12
NotFound whiteknight: looks like they don't return appropiately in some branches.
whiteknight NotFound: okay, that's weird
benabik whiteknight: Sorry, that _dis_ables randomization, sorry. 19:13
NotFound The ones that say 'TODO should really be: PANIC'
whiteknight NotFound: which PMC? imageiosize?
NotFound whiteknight: and imageiofreeze
19:16 he left
whiteknight yes, these vtables do look...weird 19:16
NotFound Sometimes I think that strict structured programming, only one function entty and only one function exit points, is not such a bad thing. 19:17
mikehh t/examples/pod.t fails - see TT #2088 19:18
all other tests PASS (pre/post-config, make corevm/make coretest, smoke (#14321) fulltest) at 3_2_0-348-g0573725 - Ubuntu 11.04 beta i386 (gcc --optimize --gc-gms)
NotFound Those uses of return just to avoid increse indentations are weird.
plobsing_ NotFound: what is odd about those?
19:19 he joined
NotFound Dont't know well how they are supposed to work. 19:20
Coke I prefer fail-early, myself. 19:21
(rather than single-exit) 19:22
think of them as asserts.
plobsing_ exactly as they do. if we are in the context of a packfile, lookup the string index. if the string index is found, push it and return; otherwise push '-1' and do the context-less string freeze
NotFound Coke: maybe, but in a lot of places they are not failures.
plobsing_ I'll do single-return, but only if liberal use of goto is acceptable. 19:23
NotFound plobsing_: then I don't understant the TODO: PANIC comments. 19:24
plobsing_ NotFound: ideally, the string table should be populated with all the strings the PMCs will want. we hit that code when the string table does not contain the pushed string. 19:25
19:25 petdance left
plobsing_ and so we need to do an inefficient in-line string serialization 19:25
whiteknight plobsing_: I don't know if you've been following along. Rakudo is randomly segfaulting in places related to packfile serialization 19:26
NotFound is seeing segfaults in one place, I'm seeing them in another 19:27
plobsing_ the PANIC would identify this as undesirable behaviour. unfortunately, IMCC does this more than enough times to make me give up for the time being.
notably, :anon :immediate subs work this way.
19:27 he left
whiteknight I can't rule out that the segfaults are GC-related, and just happen to occur in packfile serialization code 19:28
plobsing_ whiteknight: I've read the backlog and am playing around with it myself. The problem goes away if you use --gc gms.
whiteknight oh, does it? That's interesting 19:29
I know it goes away with -G
oh yes, --gc gms works for me too
moritz plobsing_: not confirmed. I always use --gc gms, and it still segfaults for me
mikehh NotFound: ping 19:36
NotFound mikehh: pong 19:37
mikehh NotFound: tried to build winxed - make: *** No rule to make target `stage2', needed by `all'. Stop. 19:38
I usually do something like cd winxed, make clean, svn update, set path to installed parrot, make all, make test1/2/3 then run some example tests 19:40
failed at the make all stage 19:41
NotFound mikehh: try make, without all 19:44
mikehh NotFound: 'k
whiteknight --gc=gms allows me to build rakudo without issue
NotFound mikehh: yes, is a bug, forgot to change the dependeces of that target in a recent reorganization of the Makefile. 19:45
moritz whiteknight: it builds for me too (optimized), but segfaults at a test (see backlog) 19:46
benabik Are failures in extend_vtable.t known? I get tests 2, sometimes 3, 8, 11 & 13 failing. Saw a trac ticket for extend_vtable in g++, but I'm using gcc. 19:47
(Just to add confusion to everyone backtracing rakudo... :-/ ) 19:48
dalek nxed: r952 | NotFound++ | trunk/Makefile:
fix 'all' Makefile target, mikehh++
 19:53
19:54 lucian left
mikehh NotFound: 'k that works ok - managed to build and test with make, just re-built 19:57
benabik But I also get sporadic segfaults on rakudo. (Parrot master: --optimize --gc=gms --debugging=0 --ccflags="-DPARROT_GC_VALIDATE") 20:01
plobsing_ PARROT_ASSERT() appears to do nothing. It used to assert() when compiled without optimize. When did this change? 20:19
and was it intentional
?
20:20 soh_cah_toa joined, dodathome left
whiteknight I thought PARROT_ASSERT was always a no-op with --optimize 20:21
NotFound plobsing_: I've used it a few hours ago.
plobsing_ whiteknight: it is. I'm not compiling with optimize. Just straight 'perl Configure.pl'. Yet the disassembly doesn't lie. 20:22
and parrot_config shows NDEBUG being used. 20:23
NotFound Note that now it tries to emit a backtrace, and then the process can segfault in a completely unrelated position.
whiteknight plobsing_: that's weird. in an unoptimized build PARROT_ASSERT should do something
plobsing_ can someone confirm this? if this is happening, it is a *huge* problem. 20:25
NotFound plobsing_: I've told you, I unconfirm it.
plobsing_ odd. I wonder what is going wrong with mine then. 20:26
NotFound Maybe something not rebuilt because of lack of some dependence? 20:27
plobsing_ nah. I've realcleaned and retried a couple of times. 20:28
trying now with a fresh checkout.
NotFound plobsing_: In what file is the problem? 20:29
whiteknight with --gc gms I can't reproduce any segfaults in build, test or spectest for rakudo
I'll have to take moritz's word for it that it's still a problem 20:31
plobsing_ NotFound: I'm throwing asserts left, right, and center. none of them seem to have effect.
NotFound Odd
whiteknight plobsing_: as a sanity check, are you running the parrot you are building and not an installed one? 20:32
plobsing_ never install parrot
whiteknight then you are a better man than I
benabik I also get segfaults with gms 20:33
whiteknight okay. So it's all random
plobsing_ the inconsistency may be due to GC being sensitive to the amount of memory available on the system. 20:36
whiteknight yeah, that's what I figure too 20:40
memory size and memory layout
moritz is on 64bit, fwiw 20:41
whiteknight me too
I'm going to throw on a ulimit, see if I can get a new failure mode 20:42
plobsing_ fresh checkout; perl Configure.pl; make corevm => PARROT_ASSERT() compiled away
yep, ccflags contains NDEBUG here too
whiteknight that's a bug 20:43
and a serious concern
plobsing_: what compiler are you on? 20:44
NotFound Maybe is imported from perl config?
whiteknight (not that it should make any difference)
plobsing_ gcc 4.5.2, linux 2.6.37, 64-bit 20:45
moritz lemme guess... we have no tests for assertions being triggered in debug builds. 20:46
plobsing_ NotFound: looks like you may be right on that count. the only place Configure.pl adds NDEBUG does not get called.
NotFound moritz: if we add them, we'll get reports about segfaults during tests. 20:47
moritz NotFound: do assertion failures cause segfaults?
whiteknight SIGABORT 20:48
NotFound moritz: systems ususally doesn't make distictions between segafults and deliberated aborts.
Is an abnormal termination in both cases. 20:49
And even if the system logs enough information to make the distinction, people don't. 20:50
plobsing_ urg. my perl5's Config.pm doesn't have any entry with NDEBUG in it. 20:57
where is this NDEBUG comming from?
NotFound Are we importing flags from the pkg-config of some library? 20:58
plobsing_ is there a way to trace through the config step-by-step to figure out which step adds the NDEBUG? 20:59
whiteknight that's a kid51 question 21:01
plobsing_ uninstalling llvm fixed the problem 21:05
21:06 sirmacik joined 21:07 mikehh left
NotFound Weird. 21:08
dalek TT #2089 created by plobsing++: LLVM cflags prevent debug builds 21:16
TT #2089: trac.parrot.org/parrot/ticket/2089
21:17 perlite_ joined 21:19 fperrad left 21:20 perlite left, perlite_ is now known as perlite
whiteknight this rakudo bug is flummoxing me 21:21
I'm going to have to take a break for dinner 21:22
21:22 tcurtis left, athomason left 21:23 bacek left 21:25 TiMBuS left 21:26 TiMBuS joined
moritz this time t/spec/S06-operator-overloading/sub.rakudo segfaulted for me 21:29
21:42 ambs left 21:45 PacoLinux left 21:48 tcurtis joined, athomaso1 joined 21:50 bacek joined
dalek nxed: r953 | NotFound++ | trunk/winxedst1.winxed:
refactor function modifier list into a derived class
 21:58
22:12 mikehh joined 22:15 ligne left 22:28 mikehh left 22:32 bubaflub left 23:01 dafrito left
whiteknight it looks like PackFile_destroy is getting called a few hundred times, all of them with the same pf pointer 23:16
I can't tell if that's sane or not
benabik If it's not ref-counted, that doesn't sound sane. 23:17
whiteknight does not appear to be ref-counted. At least, not in that function
maybe it's extremely weird coincidence that we are deleteing and then reallocating PackFiles with the same address 23:18
plobsing_ if we're using malloc() to do the job, and the allocation occurs right after the deletion, it isn't a huge surprise - the free()d space should be at the start of the free list 23:19
sorear whiteknight: the fixed-size pools tend to reuse memory blocks
benabik Yeah, what they said.
sorear how many other objects does Parrot use that are the exact same size as struct PackFile?
whiteknight I'm not saying it's impossible, but I saw several dozen calls to PackFile_destroy in a row with the exact same pf pointer 23:20
a few with the same address is reasonable, but several dozen, and possibly over a hundred?
at that point, either something weird is happening, or we are using the world's worst algorithm 23:21
and the packfile structures appear to be using malloc/free, not the fixed-size allocator
plobsing_ what function is calling these destroys?
benabik I'd think that even if that was the case then we shouldn't be creating and destroying that many PackFile structs.
plobsing_ ah. I know the source. If ImageIOThaw is invoked on a string (not from packfile), it needs to create a dummy packfile. These get destroyed with the ImageIOThaw object. 23:24
now as to why we're creating hundreds of ImageIOThaw objects from strings... beats me. 23:26
never mind, some of them are ImageIOSize PMCs 23:28
whiteknight Parrot_ImageIOThaw_destroy 23:34
so that actually makes sense. They would be reusing pointers
plobsing_ and I'm about to reduce your 100 packfiles down to 1 23:35
dalek rrot: 5d9afa0 | plobsing++ | src/pmc/imageiosize.pmc:
eliminate unused 'pf' attribute from ImageIOSize
 23:36
23:37 sigue left
cotto plobsing++ 23:37
23:39 sigue joined
whiteknight okay, this is interesting. I pulled plobsings last commit and tried again. I don't get a segfault,but now I get this error : "PackFile_pack segment 'DIRECTORY' used size 561260 but reported 565536" 23:40
so, is that better or worse?
plobsing_ I'mma go with worse. rebuild? 23:41
whiteknight all of parrot? 23:42
plobsing_ yeah. not sure why that would be required, but give it a shot.
23:43 bubaflub joined
whiteknight rebuilt parrot and rakudo. Same error 23:47
PackFile_pack segment 'DIRECTORY' used size 561260 but reported 565536 23:48
23:49 hudnix joined, kid51 joined
whiteknight something weird is going on 23:49
plobsing_ whiteknight: what are you building to see that? 23:50
whiteknight /usr/local/bin/parrot -o src/gen/perl6.pbc src/Perl6/Compiler.pir
it's in the rakudo build sequence
plobsing_ yes, that's the command that I am debugging too. I see no error there. 23:51
well, other than the segfault
whiteknight my checkouts for both are clean 23:54
this whole situation is weird. There is something majorly wrong going on 23:55
plobsing_ We're seeing corrupted memory, so it is possible that is causing runaway seriailization

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!