Parrot 1.6.0 "half-pie" released! | Test CallSignature PMC | pcc_reapply branch still needs your love! trac.parrot.org/parrot/wiki/Callin...nsOverview
Set by moderator on 12 October 2009.
00:00 jrtayloriv joined
Whiteknight darbelo: yeah, that would be awesome too 00:00
if we have lots of libraries, we can use plumage to install them 00:01
jonathan bignums would be liked by Rakudo.
(people ask if we support them yet now and then)
darbelo Sure, plumage already works for decnum-dynpmcs, but I'd like to have big numbers *in parrot* without fhuge exteranl libraries or post install-dependencies. 00:02
Whiteknight jonathan: we'll have much more then that soon
where "soon" is very relative
darbelo Writing it is a pretty big task, but parrot provides a lot of stuff already. 00:05
All that's left is the actual numerical coding. 00:06
Whiteknight right 00:07
jonathan Whiteknight: BTW, the custom binder for Rakudo that I've been working on is coming on nicely. 00:10
Whiteknight custom binder?
jonathan Whiteknight: Parameter to signature binder. 00:11
darbelo I'm reasobaly convinced that a PMC using RIAs for 'buckets' and not-really-that-clever algorithms could kick decnum-dynpmcs' ass at integer math.
jonathan Whiteknight: So we can bind against Perl 6 signatures directly.
Whiteknight: It'll get even nicer when we have call_sig :-)
:call_sig, I mean.
jonathan is impressed now and then just how many bits of Parrot a language can choose to plug something else in to if it needs to. 00:12
Whiteknight jonathan: I know exactly how to mplement :call_sig, I want to fix tests on pcc_reapply first though 00:13
darbelo Pluggable everything!
jonathan Whiteknight: Oh, sure - and I'm not blocking on it. :-)
Whiteknight: It'll just be nice when it's there, that's all. :-)
OK, sleep time... 00:15
cotto_work clock? 00:20
purl cotto_work: LAX: Tue 5:20pm PDT / CHI: Tue 7:20pm CDT / NYC: Tue 8:20pm EDT / LON: Wed 1:20am BST / BER: Wed 2:20am CEST / IND: Wed 5:50am IST / TOK: Wed 9:20am JST / SYD: Wed 11:20am EST /
chromatic Wouldn't it be nice to unify the names of push_float and set_number_native? 00:24
cotto_work + a lot 00:25
darbelo +Inf
There's a lot of Number/Float confusion in the parrot repo.
cotto_work sounds like we need a deprecation notice 00:26
darbelo We allso need to pick the right color for the bikeshed.
cotto_work hands darbelo a dart 00:27
darbelo throws the dart at the bi-color bikeshed.
bacek_at_work cotto_work, we have it already :) 00:28
cotto_work, trac.parrot.org/parrot/ticket/866
00:28 rhr joined
darbelo =item VTABLE nomenclature. [eligible in 1.5] 00:28
cotto_work So it appears we can do it now. 00:29
darbelo That is one big blanket sir.
chromatic Do I smell a 1.8 milestone of renaming? Yes we can! 00:30
darbelo Rename everything! Break all languages! Take over *the world*! 00:32
cotto_work hulk smash 00:34
chromatic Hm, CallSignature is already the 18th largest PMC. 00:48
darbelo largest?
purl well, largest is picked first
chromatic Largest in terms of disk size. At least it's smaller than Capture, .o file wise.
00:50 payload joined
darbelo Hm. I had never actually looked at PMC on-disk sizes before. 00:50
chromatic Everything builds. Let's pass some tests. 00:52
00:59 rhr joined
chromatic That looks like a success. 01:01
dalek rrot: r41845 | chromatic++ | branches/pcc_optimize_sig (2 files):
[PMC] Added Pcc_cell array and counter to CallSignature PMC.
 01:07
Tene chromatic: you could add the type indicator flag to the Pcc_cell struct... or is there a reason you're going to store it elsewhere? 01:10
cotto_work chromatic, should your new CallSignature still be extending Capture?
dalek rrot: r41846 | chromatic++ | branches/pcc_optimize_sig/t/pmc/callsignature.t:
[t] Corrected vtable entry names in CallSignature tests.
 01:11
rrot: r41847 | chromatic++ | branches/pcc_optimize_sig/src/pmc/callsignature.pmc:
[PMC] Refactored CallSignature PMC to store positional arguments in a type-safe

CallSignature. Note that the current implementation uses tagged pointers to store type information in a memory-cheap way. It also uses linked lists instead of an array to avoid lots of expensive resizing and copying. Judicious profiling will reveal if either one needs tuning or rethinking. This PMC needs more documentation. Anyone reading this is capable of writing/copying it.
treed holy fuck
purl holy fuck is that area dangerous!
treed purl, holy fuck is that crazy
purl ...but holy fuck is that area dangerous!...
treed ...
chromatic It extends Capture because I haven't converted away from the Hash yet. 01:12
treed purl, no, holy fuck is that crazy
purl okay, treed.
treed holy fuck
purl i think holy fuck is that crazy
treed dammit
Tene treed: what's surprising?
purl it has been said that surprising is still surprising
treed The huge commit memo.
cotto_work got it
Tene detailed commit messages are good.
chromatic If you add 500 lines to a PMC, you'd better write a detailed commit message. 01:13
darbelo purl, no, holy fuck is <reply>That's crazy!
purl okay, darbelo.
darbelo treed: is that what you were trying to teach purl?
treed Nah, I was trying to teach purl to say "Holy fuck is that crazy!" 01:14
or s/crazy/$adjective/
although that works, too
Tene chromatic: you can use the lower three bits like that because it's aligned in memory? 01:15
chromatic Yep. 01:16
A compiler stricter than gcc on a platform less lenient than 32-bit x86 ought even to mask off those lower bits when treating it as a pointer, but goodness if that wasn't some weird debugging. 01:17
Whiteknight we start getting all sorts of SIGBUS on weird platforms, and we know who to yell at 01:18
chromatic Hence I decided to embrace Crazy Macro Fun Time, because hey -- C doesn't have type-safe genericity. 01:19
darbelo Heh. During my (thankfully) brief use of SPARC hardware some years back I used to curse at Linux/x86-32 programmers on a daily basis. 01:20
01:21 kid51 joined 01:22 athomason joined
Whiteknight type-safe genericity? that's one of those "feature" thingys the perl people are always talking about 01:22
dukeleto Whiteknight: pong 01:23
darbelo Bah! Features are just a different kind of bug.
dukeleto darbelo: exactly 01:24
Whiteknight dukeleto: I've already got most of my issues sorted out.
got two repos on github now! 01:25
Tene GO WK!
You'll reach the rank of GIT MASTER any minute now!
darbelo Does that mean people can start pulling him? 01:26
darbelo pushes the new master :) 01:27
Whiteknight GIT MASTER? right now I'm shooting for the rank of "GIT doesn't fuck everything up" 01:28
not quite there yet, but I'm learning from my mistakes
dukeleto Whiteknight: learn to love the directed acyclic graph :) 01:29
Whiteknight yeah, i can feel the love 01:30
and on that note, out for the night
later
dalek p-rx: 4d50dbc | pmichaud++ | src/PAST/Compiler-Regex.pir:
Adjust zerowidth handling slightly in enumcharlist.
 01:33
p-rx: 8c37aa5 | pmichaud++ | src/Regex/P6Regex/ (2 files):
Refactor backtracking modifiers, add :ratchet modifier.
shorten dalek's url is at xrl.us/bfsaiw
dalek's url is at xrl.us/bfsai2
Tene pmichaud: is nqp-rx going to fix the zero-width repeated matches bug? 01:35
pmichaud probably
it should be an easy fix.
much easier than in pge, at any rate.
Tene Okay, thanks for the information. 01:36
pmichaud in PGE it's a real pain to figure out the position where a quantification started. In the new engine, it's actually immediately available at the point of the loop.
so it should be easy to say "did we advance, if not, fail" 01:37
01:40 mikehh_ joined 01:45 rhr joined
dalek rrot: r41848 | jkeenan++ | trunk/docs/book/pir/ch04_variables.pod:
Apply patch to documentation submitted in response to posting on list by Vadim Konovalov: trac.parrot.org/parrot/ticket/1104.
 01:46
01:55 darbelo left 02:12 tetragon joined, hercynium joined
dalek rrot: r41849 | jkeenan++ | branches/auto_libjit (2 files):
1. Following suggestion by Andy Dougherty++, reworked config/auto/libjit.pm

it simply says libjit is not found. 2. Move auto::libjit to just before auto::jit.		 02:12
rrot: r41850 | jkeenan++ | failed to fetch changeset:
Merge library_files branch into trunk. Replace EXTRA_TEST_FILES defined in Makefile with @library_files defined in lib/Parrot/Harness/DefaultTests.pm. Cf.: trac.parrot.org/parrot/ticket/1061.
02:23 rhr joined
dalek TT #1110 created by mikehh++: parrot fails to build with g++ 4.4.1 on Ubuntu 9.10 beta amd64 02:28
02:29 mikehh joined
mikehh bah - dropped my internet connection while I was setting up the ticket in trac - had to login again - fortunately most of the info was in my editor 02:31
kid51: does the libjit stuff work for amd64 or just i386 as before? 02:40
in other words can I test the branch with amd64 and have any significance 02:41
02:41 janus joined
kid51 mikehh: I don't have access to anything 64-bit. So far, I've only tested auto_libjit branch on Linux/i386. 02:42
You can test branch on amd64 and see if you duplicate my 'make' failure.
mikehh well let me give it a go and see what happens :-}
chromatic The pcc_optimize_sig branch could use testing on non-x86 32 bit platforms too. 02:44
kid51 mikehh: I just used your config options on Linux/i386 -- with better results. See TT #1110. 02:46
chromatic: Purpose of pcc_optimize_sig branch is ... ? 02:47
chromatic The pcc_rewiring branch runs one-argument positional function calls 2.5x slower than trunk. 02:49
The purpose of the branch is to remove that speed penalty.
kid51 What is difference between pcc_optimize_sig and pcc_rewiring branches? 02:52
mikehh kid51: I had no problems building with g++ 4.3 - it is just g++ 4.4 that seems to be a problem
Tene kid51: the first is for optimizing the CallSignaure pmc. The second is for implementing and using the callsignature pmc for calling conventions. 02:53
kid51 k. I'll do a Smolder on pcc_optimize_sig on Darwin/PPC. 02:54
Tene kid51: the only test in that branch that matters is t/pmc/callsignature.t 02:55
03:00 plobsing joined
mikehh need to take a break 03:01
03:04 rhr joined
kid51 pcc_optimize_sig branch: Darwin/PPC: t/pmc/callsignature.t PASS 03:12
kid51 must sleep
purl $kid51->sleep(8 * 3600);
03:17 s1n joined
dalek rtcl: 2da712d | coke++ | docs/spectest- (2 files):
Spec test run, reclaim last batch of failing tests.
 03:17
shorten dalek's url is at xrl.us/bfsawi
dalek rrot-plumage: f3eaeb8 | leto++ | :
[t] Add tests for exists() and make the harness keep track of total pass...
 03:23
shorten dalek's url is at xrl.us/bfsaw9
dalek rrot-plumage: 1c1247d | leto++ | :
[t] Add some tests for subst() and refactor the harness a bit
 03:45
shorten dalek's url is at xrl.us/bfsa2e
japhb dukeleto, hope you don't mind I'm just sitting back and watching the commits roll by, rather than commenting on everything -- you seem to be having fun. :-) 03:48
03:55 plobsing_ joined, iblechbot joined
dukeleto japhb: indeed :) 04:01
dalek rrot-plumage: eb31305 | leto++ | :
[t] Add tests for join() and split() and detect invalid plans
 04:02
shorten dalek's url is at xrl.us/bfsa5b
Coke msg allison PGE does NOT build for me. 04:19
purl Message for allison stored.
Coke msg allison (it segfaults on pcc_reapply)
purl Message for allison stored.
dukeleto Coke: do you have a nopaste?
Coke: which platform?
purl I'm running on OS/2 on an Atari, can you help?
dukeleto kicks purl in the face
nopaste "coke" at 72.228.52.192 pasted "useless nopaste for dukeleto++" (1 line) at nopaste.snit.ch/18337 04:20
Coke dukeleto: linux. 04:21
dukeleto Coke: hmmm. can you do a 'make corevm' ?
Coke dukeleto: allison said I didn't have to. =-)
(yes, corevm make completes) 04:22
dukeleto Coke: just wondering
purl somebody said just wondering was the interpreter itself makes any distinction
dukeleto Coke: can you run that command under valgrind? 04:23
Coke dukeleto: fyi, I'm just doing this on feather.
dalek rrot-plumage: 407f8ec | leto++ | :
[t] Report what invalid plans look like and how many files failed, if any
shorten dalek's url is at xrl.us/bfsa69
Coke dukeleto: not easily, due to the sub-makefile. 04:28
04:29 mokurai1 joined
dukeleto Coke: you should be able to do 'valgrind ../../parrot -o PGE.pbc --output-pbc PGE.pir../../parrot ../../runtime/parrot/library/PGE/Perl6Grammar.pir --output=PGE/builtins_gen.pir PGE/builtins.pg' from that directory 04:30
Coke dukeleto: after doing two other steps first, yes. 04:32
(both adjusted for being run out the wrong dir.)
dukeleto who do I have to bribe to get on planet parrot?
Coke me, I guess, although the last time I added someone, it didn't stick. 04:33
treed Did I ever appear at all? 04:36
dukeleto msg Whiteknight if you need help moving matrixy to github, let me know 04:37
purl Message for whiteknight stored.
dukeleto Coke: i don't show up either, i guess
dalek p-rx: b9ea370 | pmichaud++ | src/Regex/P6Regex/Actions.pm:
Factor out creation of regex sub for <nibbler> nodes.
 04:44
p-rx: 3af996e | pmichaud++ | src/Regex/P6Regex/ (2 files):
Add regex parameters to named assertions.
shorten dalek's url is at xrl.us/bfsa9m
p-rx: 9ca78a9 | pmichaud++ | (2 files):
Add <before> builtin subrule.
shorten dalek's url is at xrl.us/bfsa9o
shorten dalek's url is at xrl.us/bfsa9q
Coke dukeleto: it takes some amount of time for it to notice a new feed. 04:51
remind me tomorrow, I'll re-ping ask & robrt.
dukeleto Coke: ok, I will continue to annoy you tomorrow :) 04:54
dalek p-rx: a3bc5aa | pmichaud++ | (2 files):
Add <?> (null) and <!> (fail) assertions.
 04:55
shorten dalek's url is at xrl.us/bfsban
japhb Man, it is a good night for Plumage, between dukeleto++'s work on the test harness/suite and pmichaud++'s work on nqp-rx. Warms my heart, it does. 04:57
05:05 plobsing joined
dalek rrot-plumage: 9606ea6 | leto++ | :
[t] Test harness can now deal with TAP comments
 05:35
shorten dalek's url is at xrl.us/bfsbee
dalek rrot-plumage: 1452fd6 | leto++ | :
[todo] Add some things that would be nice for the test suite to have
 05:40
shorten dalek's url is at xrl.us/bfsbet
05:57 JimmyZ joined 06:01 uniejo joined
chromatic Alrighty, even only that much of CallSignature migrated away from Capture is 14.887% faster. 06:25
We create 28.557% fewer PMCs. 06:26
I want to get to 57.114% fewer PMCs. 06:27
That's not even getting rid of CPointer. 06:29
Tene chromatic: very nice work. 06:37
cotto chromatic, is the next stage to add a Hash* and associated machinery? 06:38
chromatic Yes.
Well, the next step is to add the elements entry that wouldn't let the other version build, but yes. 06:39
cotto chromatic, would it make sense to do some/all of this work in Capture? 06:44
allison cotto: it's pretty special-purpose to the CallSignature 06:45
that is, it's *typed*, which Capture isn't
chromatic What kind of keyed (hashlike) access is necessary? 06:50
allison set keyed, get keyed, and iterate over keys 06:51
(which can just be "give me a list of keys")
really, getting a list of keys would be better than the current way of iterating over the hash 06:52
chromatic I'm not sure which vtable entry is best for "give me a list of keys". 06:53
allison chromatic: me neither 06:56
we can provide a method for PIR access, but that's too slow for access during the call 06:57
chromatic Iteration it is!
allison a function like we use on context will work for C access
Parrot_get_signature_named
chromatic: get_iter would also work, but we need iterators for both positionals and named 06:58
chromatic: how about get_attr_str?
chromatic I had all of the tests passing on pcc_reapply without making a positional iterator. 06:59
allison chromatic: we're using that for several other PMC pieces of data inside CallSignature already
06:59 AndyA joined
chromatic I didn't add any special iteration support, I mean. 07:00
allison did you have 'elements' returning the number of positional arguments?
that's what the code currently relies on 07:01
('elements' on the CallSignature returns the number of positional arguments)
chromatic Yes. 07:02
allison for named, the code relies on being able to fetch the internal hash and iterate over it at a very low and very yuckky level
chromatic Tests incoming now.
allison but, with a list of keys, that could change to something much cleaner, more like the iteration for positional args
chromatic I'm going to play the expedience card and make it work with what's there now. 07:04
I have no objection to cleaner in the near future.
dalek rrot: r41851 | chromatic++ | branches/pcc_optimize_sig (2 files):
[PMC] Added elements vtable entry to CallSignature PMC as well as tests.
 07:05
chromatic allison, do you know which vtable entries in the Capture PMC iterate over named arguments? 07:06
allison chromatic: none 07:07
07:07 fperrad joined
allison Capture PMC provides no support for iterating over named arguments 07:08
(that's why the code fetches the hash from the CallSignature and iterates over it directly)
chromatic Through the hash() method there? 07:09
07:09 iblechbot joined
allison chromatic: through get_attr_str, for "named" 07:09
chromatic: I've avoided calling methods from inside CallSignature
(speed mostly, but also bootstrapping problem) 07:10
dalek tracwiki: v15 | kurahaupo++ | ArrayTasklist
tracwiki: trac.parrot.org/parrot/wiki/ArrayT...ction=diff
shorten dalek's url is at xrl.us/bfsbkc
allison chromatic: got to run to class
chromatic Hm.
That interface is likely to change, because *yergh*. 07:11
allison chromatic: yes please
chromatic: worked with what we had, but not pretty 07:12
chromatic: a get_attr_str "named_keys" would be lovely
chromatic: even if it's stored as a complex string that's split on access 07:13
chromatic I'm going to return a FixedStringArray because it's simple and... I've read *The Practice of Programming*. 07:18
Goodness, I even asked Brian Kernighan about the relevant portion I have in mind.
07:19 payload joined
chromatic We're going to pay O(n) for named access because n < 10 almost almost always here, and I refuse to believe that O(1) here, hash-wise with all of its GCables and iteration and iterator keys, is any cheaper than O(n) for n <= 12. 07:20
07:24 donaldh joined 07:40 donaldh left 07:42 silug joined 08:00 allison joined 08:05 szabgab joined
chromatic Let me fix GC in the CallSignature PMC and update the 'named' attribute returns with a FixedStringArray, and it's all yours, allison. 08:10
allison chromatic: awesome!
purl awesome is, like, a window manager or at awesome.naquadah.org or awesome
allison chromatic: free to merge into the pcc_reapply branch?
chromatic You'll want to test it first. 08:11
'named' doesn't return a Hash anymore, so your custom iteration there can disappear into something somewhat saner.
allison chromatic: aye, and will need to do some code integration
chromatic I'm not *quite* certain it's ready to merge, but it's worth inspecting that way.
allison chromatic: okay, I'll try it out
chromatic: if I find anything that's missing, I'll add it as a test in the pcc_optimize_sig branch 08:12
chromatic Give me about 10 more minutes.
I'm fixing VTABLE_mark now and VTABLE_destroy in a moment and then a couple of minutes to make 'named' work. 08:13
Otherwise you get a thousand lines of code out of me in a day.
allison chromatic: it'll be about 30 minutes before I can get to it
chromatic Good.
Checking in now. 08:35
There are a few interface bits where we might consider exceptions. 08:39
The keyed tests could use some extension as well.
dalek rrot: r41852 | chromatic++ | branches/pcc_optimize_sig (2 files):
[PMC] Gave CallSignature PMC its own hash component and removed its inheritance

Added tests for keyed access. Fixed mark() and destroy() vtable entries for named arguments. Made 'named' attribute access of CallSignature return a FixedStringArray of all named arguments.
chromatic There's also no documentation for any of the VTABLE entries I added. 08:40
allison Can add later today 08:42
(get it working first)
time?
purl time is 08:34:26 2009 and (did you mean "clock"?) or flowing like a river or the fire in which we burn
allison clock?
purl allison: LAX: Wed 1:42am PDT / CHI: Wed 3:42am CDT / NYC: Wed 4:42am EDT / LON: Wed 9:42am BST / BER: Wed 10:42am CEST / IND: Wed 2:12pm IST / TOK: Wed 5:42pm JST / SYD: Wed 7:42pm EST /
chromatic src/pmc/callsignature.pmc | 914 ++++++++++++++++++++++++++++++++++++++++++++-- 08:43
t/pmc/callsignature.t | 247 ++++++++++++
2 files changed, 1121 insertions(+), 40 deletions(-)
allison chromatic: figure you're about to drop off, will catch up with you tomorrow
chromatic: the changes are looking good 08:44
chromatic: thanks!
chromatic You should be able to drop in the single PMC file as a replacement and see what happens.
08:44 mokurai1 left
allison yup, I'll just copy it over 08:44
and work on the code integration locally 08:45
chromatic Your VTABLE_get_attr_str(interp, call_obj, "named") will have to change. Otherwise, as far as I know, it's a drop in replacement.
allison will check it all in when I get it fitting smoothly 08:46
chromatic Now I'm out. West side.
allison catch you later
allison moves class rooms 08:50
09:02 allison joined 09:16 allison joined 09:50 bacek joined 10:18 payload joined 10:35 AndyA joined
bacek o hai 10:59
dalek kudo: 3eceb87 | moritz++ | docs/ChangeLog:
[docs] update ChangeLog
shorten dalek's url is at xrl.us/bfsb35
11:06 uniejo joined, kid51 joined 11:31 masak joined 11:33 HG` joined 11:49 whiteknight joined
whiteknight good morning #parrot 11:52
masak squawks happily 11:54
11:57 bluescreen joined
whiteknight purl msg dukeleto thanks for the offer. I actually got it loaded onto github last night. I'm sure I'll need plenty of other help later though! 11:58
purl Message for dukeleto stored.
12:19 tetragon joined, dukeleto joined
dukeleto seems like feather.perl6.nl is down, anybody heard anything? 12:19
12:37 JimmyZ joined 12:46 preflex joined 12:55 payload joined 13:03 Coke joined
Coke feather is unreachable. 13:03
msg allison I didn't use buildframes=0, and feather is now unreachable, so I can't verify. 13:04
purl Message for allison stored.
13:04 ruoso joined
dukeleto Coke: yes, feather is down right now. i emailed juerd 13:04
Coke I seem to get a lot of admin mail about stef@payrard.net being removed from mailing lists. 13:08
stef?
purl stef is not you or a purveyor of useless factoids or a mere rumour! or not a mere rumour or cognominal
13:09 allison joined 13:22 iblechbot joined 13:31 dalek joined
Coke ah, right, I stopped working on osx because I needed an installed parrot. :| 13:32
13:32 jonathan joined 13:33 pmichaud joined
Coke anyone /installing/ on OS X intel? 13:34
13:41 dukeleto_ joined
dukeleto_ feather is back up 13:41
cognominal Coke, I am so sorry to not have unsubribed my old email address from these mailings list 13:57
13:57 PerlJam joined
Coke cognominal: oh, no worries, apparently it happens automatically. =-) 13:57
14:01 PacoLinux joined
dalek TT #1111 created by coke++: cannot build on OSX/intel with --optimize 14:03
TT #49 closed by coke++: install email2trac 14:06
TT #526 closed by coke++: create parrot-devel macport 14:09
14:11 Psyche^ joined 14:14 coke_ joined
dukelet0 coke_: seems like that is a 10.4 issue 14:15
Coke dukelet0: quite possible. 14:18
care to buy me a new imac? =-)
dalek rrot: r41854 | allison++ | branches/pcc_optimize_sig/t/pmc/callsignature.t:
[pcc] Adding tests for 'exists' opcodes, needed in pcc_reapply branch.
 14:19
allison msg chromatic The code looks good. Slurpy hashes are using the 'exists' vtable function, so need that. I added tests for it, just in case you're up before I get back in an hour or so. 14:25
purl Message for chromatic stored.
Coke rant: I shouldn't have to have all this boilerplate make crap. 14:41
(in partcl)
Coke will dynamically generate it, Coke supposes.
15:07 payload joined
cotto_work hio 15:14
whiteknight hello cotto_work 15:18
Coke hio 15:25
cotto_work Does someone want to add exists to chromatic's CallSignature so I can stop being tempted to? 15:26
nm. It looks like allison just forgot to update the test plan. 15:27
or not
cotto_work hopes this isn't indicative of how his day will go. 15:28
15:30 davidfetter joined
dukelet0 reset his bioclock and got up at 5am this morning. yowzer! 15:31
Coke is there a gitsquash for something git-only, rather than git-svn ? 15:38
(I can do the interactive, but it would be nice to have it strip out things that are already pushed upstream from consideration) 15:39
dalek rtcl: a4b859a | coke++ | config/makefiles/ops.in:
Don't build in one dir and then copy to another dir.

already complicated rules to accurately reflect how dynext depends on src/ops)		 15:45
Coke (git rebase -i origin ?)
shorten dalek's url is at xrl.us/bfsc5h
dukelet0 Coke: you probably want git rebase -i master 15:46
Coke: or git rebase -i origin/master (which should be the same)
moritz not the same 15:47
origin (or origin/master) sounds much saner than master
15:54 mikehh joined
Coke much easier than with git svn. =-) 15:57
mikehh trunk - pre/post-config, smoke (#29008) PASS, fulltest FAIL at r41854 (see TT #1102) - Ubuntu 9.10 (beta) amd64 15:59
fulltest (testf, testg and testS) (same tests FAIL, same results)
t/compilers/imcc/syn/macro.t - Failed tests: 2-4
t/compilers/imcc/syn/regressions.t - Failed test: 7
rest of fulltest passes
16:01 darbelo joined
dukelet0 moritz: if you are working off of the remote 'origin', then master and origin/master are the same, no ? 16:03
moritz: duh, no they are not
moritz dukelet0: not if you have local commits that you haven't pushed yet
dukelet0 moritz: the local 'master' could be ahead of origin/master
moritz: yeah, I realized that as I typed it. I blame me being sick :)
moritz dukelet0: then get well soon ;-) 16:04
dukelet0 moritz: doing my best to, thanks
mikehh pcc_reapply - make smoke (#29009) at r41854 - 10,691 ok, 17 failed, 262 todo, 560 skipped and 0 unexpectedly succeeded - Ubuntu 9.10 beta amd64 16:18
16:23 bluescreen joined
dukelet0 mikehh: down to 17 failing tests, nice! 16:31
16:40 bluescreen joined
mikehh dukeleto - 3 or 4 are missed because t/library/coroutine.t - test died after test#2 - bad plan You planned 6 tests but ran 2. 16:43
don't know if it failed in test 3 or after exiting test 2
16:50 mokurai joined 16:59 chromatic joined
dukelet0 mikehh: still, that is a lot better than where we were 1 or 2 weeks ago 17:08
whiteknight much better 17:10
nopaste "mikehh" at 81.149.189.7 pasted "pcc_reapply branch - summary of failures from make fulltest at r41854" (47 lines) at nopaste.snit.ch/18341 17:16
17:18 fperrad_ joined 17:21 redbrain joined 17:45 joeri joined
Coke seen pmichaud ? 17:53
purl pmichaud was last seen on #parrot 16 hours, 16 minutes and 36 seconds ago, saying: so it should be easy to say "did we advance, if not, fail"
jonathan wonders if that was about regex engine or life outlook 17:58
moritz lol
chromatic I hope regex. I've been forgetting to take regular continuations.
Coke looks forward to when we can do that. 17:59
jonathan ooh, chromatic :-) 18:00
chromatic: You did a change ot the internals of CallSig PMC, yes?
Coke was just wondering if pmichaud had had any time to work on porting tcl to PCT, but I seriously doubt it. =-)
jonathan chromatic: How are the nameds stored within that?
chromatic: I got the impression from what you wrote, not as a hash, but more flattened? 18:01
chromatic They're still stored as a hash right now. 18:02
jonathan chromatic: Ah, OK. 18:03
chromatic: That can work for me too.
chromatic There are weird semantics to work out doing anything else, so this was the minimal change.
jonathan chromatic: nod 18:04
chromatic I changed what the 'named' attribute returns for a CallSig though. It used to return the Hash PMC directly. Now it returns a FixedIntegerArray of all of the known named names.
jonathan chromatic: FIA?
purl rumour has it FIA is at ctx->constants[interp->current_args + 1L]->u.key
jonathan Not FixedStringArray?
chromatic I'm sorry, yes. FSA.
jonathan ah, phew :-)
chromatic: That's nice, it saves me needing an iterator.
Or does it use one internally?
Coke message pmichaud in nqp-rx, wouldn't it be faster to compare the single position against an ord, rather than doing a single-char substring? 18:05
purl Message for pmichaud stored.
jonathan (That is, is there an FSA internally, or one is creted for me?)
Coke msg pmichaud (looking at the pir generated for the simple 'abcde*f' example - though no doubt that would start to get wonky for unicode)
purl Message for pmichaud stored.
chromatic It's created for you. You can destroy it if you want. 18:06
jonathan chromatic: I probably don't need to do that, was just curious it it was meaning yet another PMC created. 18:07
The fewer PMCs we've gotta create to bind a signature, the better.
If there are no named args, what do I get?
Empty array, or Null PMC? 18:08
chromatic Good question.
purl Yeah, it is. I'm stumped.
18:08 mikehh joined
pmichaud Coke: which part of nqp-rx are you referring to? 18:08
chromatic You get PMCNULL, jonathan.
Coke ./p6regex --target=pir > abcde*f 18:09
chromatic I'm not sure if that's right, but that's how I handled it right now.
pmichaud Coke: are you referring to the 'e' and 'f' tokens?
that for single-character strings an ord might be quicker?
Coke yes.
pmichaud it's a good thought. Do we know that ord comparisons are faster than substring + string compare? 18:10
and is it a frequent enough operation to make a difference?
Coke ISTR having tested that at some point in the past, but having someone do a benchmark would be instructive. (not you or me.)
well, in that sample it is. =-)
pmichaud I mean, how often do we do single-character comparisons? ;-) 18:11
18:11 KatrinaTheLamia joined, Util joined
Coke just something to keep in mind in later optimizations. 18:12
pmichaud sure.
a bigger optimization will be quantifications of things like \\w, \\s, etc.
PGE optimized those into single opcodes, nqp-rx doesn't do that yet but I'm planning to enable it in the future.
jonathan chromatic: I approve of PMCNULL there.
chromatic I'm sure ord is faster than substr + strcmp. 18:13
jonathan chromatic: Because it'd kinda suck to create a PMC to then check is empty and throw away straight off.
Coke pmichaud: in the generated PIR, what's the .lex for?
(same example from the STATUS)
pmichaud the cursor is lexically available as $¢ .
part of the S05 specification.
just like the match object will be available as $/ 18:14
it doesn't matter in this particular example, but when we have nested closures it will be important
Coke k
also, 'make test' fails for me, presumably because there's a (wrong) path to parrot in the .t shebangs. 18:15
pmichaud yes, probably.
I haven't quite figured out how to handle that for installed parrots yet.
and at the moment there's only one .t :)
chromatic jonathan, even at 1:15 am I handled that case correctly. I'm getting good at PMCs. 18:16
jonathan chromatic++
:-)
moritz can't you just use prove --exec $(PARROT) t/file.t
in Makefile.in
pmichaud that would likely work, yes.
moritz and ignore the shebang line
pmichaud doesn't help much for running prove from the command line, though.
jonathan chromatic: OK, once pcc_reapply lands and we have :call_sig, I'll be able to quite quickly switch to that, I expect.
chromatic I think so too. It's pretty straightforward. 18:17
Coke no, but that would fix make test, which is nice. (and .t is slightly better if you assume an installed parrot if you just use #! parrot
pmichaud agreed
I'll undoubtedly make moritz++'s change 18:18
18:18 msmatsko joined
jonathan chromatic: Have you assessed performance of pcc_reapply at all? I'm curious what kind of hit - if any - we'll take (or if it'll be a win). 18:18
chromatic I haven't tried with the full CallSig migration. 18:19
jonathan OK.
chromatic Still inheriting from Capture and using the Hash component (at least two unnecessary PMCs created for each CallSig), my version was 15% faster.
I think the migrated version will create some 57% fewer PMCs, though. 18:20
jonathan My resig branch is making calls and binding params 2-4 times faster at the moment than before I started.
chromatic My goal is to double the speed of the branch as currently exists. That gets us within some tuning of trunk.
jonathan Ooh, that'll help. :-)
Ah, OK, so we're lagging trunk ATM.
But cleaner. 18:21
purl cleaner is probably with OO (e.g. the planet one thingy)
pmichaud jonathan: yes, by about 2x
jonathan Ouch.
But OK.
That said, if the cost is in the binding, and not in the creation of a :call_sig...
Erm, of a CallSignature, I mean.
pmichaud then Win
jonathan And Rakudo is just using the CallSignature
Then it's a win.
pmichaud right
chromatic Remember that no one's done any optimization on the branch either. 18:22
jonathan chromatic: Aye.
chromatic: Same on my one too, really. 18:23
chromatic: Other than optimizations that just come through the overall algorithm/process being more efficient.
So I figure between potential optimizations in Parrot and potential ones in the new Rakudo binder, we've lots of space for win.
Coke potentially related question - how is perl6 going to throw errors about arg mismatches?
chromatic slushballs -- no gravel. 18:24
jonathan Coke: Exception.
purl exception is A reasonable response to an unreasonable situation.
Coke right now in partcl, I basically declare everything as taking a single slurp parameter and then manipulate it myself.
jonathan Coke: Rakudo too. :-)
Coke (rather than being able to use parrot's exception handling)
jonathan Coke: Apart from named slurpy and normal slurpy 18:25
Coke: And if there's a problem unpacking it, then I just throw an exception.
allison chromatic: PMCNULL is good for no named arguments (cheap)
jonathan Woo agreement!
On pcc_reapply and Rakudo - I suspect Rakudo's C bits will probably need some patching. 18:28
18:28 AndyA joined
allison chromatic: I just got settled in for a night's work, do you want me to go ahead and implement 'exists' on CallSignature? 18:29
jonathan However, that patching will be rather different on the resig branch.
allison chromatic: looks like that's all I'm missing for a merge into pcc_reapply
jonathan Just in case anyone goes ahead and makes the effort on master Rakudo.
pmichaud jonathan: I think pcc_reapply won't be landing in parrot trunk before the release. 18:30
allison chromatic: I can work around it, if I know the various get_*_keyed_* operations return PMCNULL when there is no corresponding argument
chromatic: (which might be cheaper than checking exists a then doing a get anyway
jonathan pmichaud: Aye. I was more aiming this at anyone who wanted to test Rakudo against pcc_reapply.
chromatic I can add the exists, if you give me 30.
allison cool
jonathan Like sooner than then.
pmichaud jonathan: right 18:31
allison finishes eating
jonathan Just attempting to minimize waste of resources. :-)
pmichaud parrot tends to return PMCNULL for any sort of "non-existent element of aggregate" fetch
e.g., from Hash or *Array
allison pmichaud: some of the old structures still return Undef, so never sure 18:32
pmichaud yes, I'd like to standardize on the PMCNULL behavior
allison Undef should probably go away at some point
jonathan PMCNULL++ 18:33
chromatic STRINGNULL++
jonathan bbiab - food
pmichaud in many cases it's easier to just do the fetch and check for PMCNULL than it is to do an explicit existence check beforehand
allison pmichaud: it's a good feature to have, even if we don't use it in this particular block of code 18:34
Coke allison: (undef go away), hey, we already removed None...
allison pmchaud: (the CallSignature on trunk inherited exists from Capture)
Coke: definitely progress :)
Tene we OBVIOUSLY need to change all uses of Undef to exceptions. 18:43
jonathan: have you seen the build failures for rakudo against pcc_reapply? 18:46
jonathan: nopaste.snit.ch/18329
dukelet0 Tene: those look fun 18:53
Tene dukelet0: cardinal and my scheme compiler both compile and pass their tests appropriately on the pcc branch. 18:54
dukelet0 Tene: that is very cool
Tene: i wonder if Coke has tried partcl on the pcc_reapply branch
18:57 kthakore joined
kthakore chromatic: where is this C stuff!!!! 18:57
chromatic src/pmc/callsignature.pmc
kthakore PMC = C ?
chromatic preprocessed C, yes.
jonathan Tene: The good news: those ones will go away when pcc_reapply rebases against trunk
Uh, "rebases" ;-)
Tene: The bad news: even after that, all you'll get is different errors. :-) 18:58
kthakore chromatic: um sorry I am new to parrot where is the source ? it comes with rakudo right?
jonathan (But those ones will actually apply to pcc_reapply ;-))
kthakore win2
Tene kthakore: svn.parrot.org/parrot/branches/pcc...imize_sig/
kthakore Tene: thanks
jonathan kthakore: Parrot lives in a separate repository to Rakudo.
kthakore jonathan: ah ok
jonathan kthakore: Rakudo checks it out when you do Configure.pl with --gen-parrot. 18:59
Tene kthakore: the rakudo configure script has an ... what jonathan said.
kthakore Tene: lol
ok
thank you
chromatic: btw major major XS rewrite of SDL has begun. Drop by #sdl some time. I would like to get your thoughts 19:00
chromatic I browsed your discussion with nothingmuch. I think he's on the right track. I always wanted to make separate .xs files for each subsystem.
19:03 msmatsko joined
kthakore chromatic: thats what we are doing me an acme. ur1.ca/dltk 19:05
chromatic: we broke it down to be SDL::* is only bindings to C nothing more nothing less. Because before we where doing to many things with SDL::* 19:06
cotto_work Tene, it looks like the branch needs PARROT_MAX_ARGS bumped. That's what I'd expect is causing most of those warnings. 19:09
include/parrot/op.h 19:10
Tene What should I bump it to?
chromatic 16?
purl i guess 16 is such an easy number to multiple or divide by in my head
cotto_work 16, like in trunk
Tene lemme try it. 19:11
cotto_work permission granted ;)
it'll probably still fail, but less noisily 19:12
Tene cotto_work: nope, still got those warnings. 19:14
cotto_work it was worth a shot
Tene should I bump it anyway, to match trunk?
cotto_work wouldn't hurt, as iirc rakudo needs it 19:15
dalek rrot: r41855 | tene++ | branches/pcc_reapply/include/parrot/op.h:
[pcc] Bump PARROT_MAX_ARGS to match trunk.
 19:16
moritz perl6.ops:631: error: ‘struct parrot_interp_t’ has no member named ‘returns_signature’ 19:24
that's what rakudo build says when trying to build against latest pcc_reapply 19:25
dukelet0 do we have any tests for PARROT_MAX_ARGS ?
allison moritz: does it have an op that's peeking directly at subroutine returns?
moritz: perhaps something to do with 'wants'?
jonathan allison: Oh, sure it is. :-) 19:26
allison so, that'll just need to be updated to pull the call signature object instead
jonathan allison: We muddle into the calling conventions in a couple of places.
allison: It'll certainly have to change to work on pcc_reapply.
allison 'returns_signature' is no more 19:27
jonathan allison: Aye.
allison: That's fine, I saw this lot coming. Things should get a *lot* neater too. :-)
Tene jonathan: would it be possible for you to make a rakudo branch that's updated to use pcc_reapply? 19:29
allison Tene: that could help us make sure there aren't unexpected problems
Tene jonathan: I'd really like to know that rakudo can work against pcc-reapply, or if not, find and fix the problems.
jonathan Tene: I will, but given (a) pcc_reapply won't land before 1.7 and (b) I've got my hands rather full with the resig changes, which heavily influence this, I don't think it's wroth doing for a couple of days yet. 19:30
Tene Okay, thanks.
jonathan Tene: Once resig lands in trunk, I'll work on a branch to get it working against pcc_reapply though.
erm, lands in master :-) 19:31
19:42 cconstantine joined
kthakore um 19:43
parrot fails to compile
oops no rakudo does
mutablevar.c:99: warning: implicit declaration of function 19:44
make: *** [src/pmc/perl6_group.so] Error 1make: *** [src/pmc/perl6_group.so] Error 1
19:50 kthakore joined 19:51 mokurai joined 19:53 AndyA joined 20:05 bacek joined
pmichaud phone 20:06
kthakore it seems to work now 20:15
pmichaud: my phone is not ringing :P stop trying to trick me 20:16
chromatic allison, I have a question about the 'exists' tests for CallSig. 20:38
20:40 joeri left
Tene kthakore: mail 20:41
kthakore Tene: whut? I have no mail why you lie ... 20:42
Tene hehehehehe 20:43
chromatic kthakore: pants
kthakore chromatic: why you trying to get my pants off?!!
dalek rrot: r41856 | chromatic++ | branches/pcc_optimize_sig (2 files):
[PMC] Added exists_keyed() and exists_keyed_int() VTABLE entries to

check for boolean return values, not any specific value.
jonathan
.oO( must remember to parse pants like an American, not like the Brit I am )
chromatic Now you'll always wonder how I meant it.
Tene chromatic: what's your question about callsig exists tests?
kthakore jonathan: how does a Brit parse pants? 20:46
chromatic Tene, I modified the tests; instead of checking for truthiness returned from exists_keyed or exists_keyed_str, they checked for specific values of truthiness.
jonathan kthakore: In British English, pants is what you wear underneath what Americans call pants. ;-) 20:47
kthakore jonathan: that just made it worst
worse
Tene I don't know of any reason pcc would want specific values back from exists.
kthakore jonathan: I thought thats what underpants was
moritz AE pants = BE trousers, no? 20:48
jonathan moritz: right
chromatic I don't either. I checked the docs for Capture and they said truthiness, so I changed the tests.
jonathan kthakore: In BE we just use "pants" to mean those, normally. :-)
Tene chromatic: quick review of pcc diff shows no non-truthy uses of exists. 20:49
kthakore hi moritz
moritz \\o/
kthakore moritz: be careful people try to get into your pants here 20:50
chromatic Then the new CallSig PMC on pcc_optimize_sig should work on the pcc_reapply branch.
Tene orly? Lemme try it.
chromatic I'm trying it too.
kthakore ok back to #sdl for hacking ...
allison chromatic: you had a question? 20:51
moritz kthakore: no worries, I'm here for for more than 2.5 years already ;-)
chromatic Any reason exists_keyed and exists_keyed_str should return anything other than true or false for CallSignature?
kthakore moritz: so you are already pants free and just now getting to be open and proud about it
allison chromatic: a simple 1/0 is fine, yes 20:52
moritz kthakore: it makes people happy to be proud of what they do - but that pants were your invention ;-)
jonathan Oh boy, what have I started. :-)
cotto_work You threw a spanner in the works. 20:53
Tene I got "no exists_keyed_str in callsig... realcleaning and rebuilding. 20:54
jonathan It just fell out of me hand and into 'em, sir. Honest.
kthakore moritz: Behold I invented PANTS! Bow down to my Awesomeness
chromatic Tene, the new CallSig only exists on the pcc_optimize_sig branch. It hasn't merged yet. 20:55
kthakore jonathan: seriously I was going to leave too
Tene chromatic: I copied it from that branch to my local branch.
git checkout pcc_optimize_sig src/pmc/callsignature.pmc
chromatic Okay.
allison Tene: yes, chromatic is adding the 'exists' feature now 20:56
Tene ah, I missed that.
kthakore j #p5p 20:58
Tene if I implement exists_keyed_str, I now get a segfault in parrot_hash_get_idx 21:02
chromatic I already implemented that on the branch. 21:03
21:04 fperrad joined
Tene chromatic: I see exists_keyed and exists_keyed_int, but not exists_keyed_str 21:05
chromatic Ah, right.
kthakore bye guys gtg hom
home
chromatic You have to play the hash_key games that set_integer_keyed_str does. 21:06
nopaste "tene" at 97.117.62.135 pasted "eks for c" (22 lines) at nopaste.snit.ch/18344
21:06 allison joined
chromatic That should do it, Tene. 21:06
nopaste "tene" at 97.117.62.135 pasted "the next error I get is a segfault because of this code:" (4 lines) at nopaste.snit.ch/18345 21:07
allison Tene: I've already fixed that 21:08
Tene: will check it in when I merge chromatic's code in
Tene Okay. Nevermind, then.
allison chromatic: my irc client dropped off for a bit there
chromatic: ready to copy over? 21:09
21:09 Austin joined
allison Tene: (that was the old low-level hash iteration, which isn't necessary any more with the nice new implementation) 21:09
chromatic It's ready to merge, yes. 21:10
You should run the tests too, but I have confidence.
allison okay, trying now
Tene goes back to real work. 21:11
21:15 payload joined
allison chromatic: works fine in pcc_optimize_sig, but still getting "exists_keyed_str() not implemented in class 'CallSignature'" in pcc_reapply 21:23
chromatic Alright, I'll make a test and such there.
allison chromatic: checking to make sure copy was clean 21:24
chromatic I know how to make that work. I need food first though.
Just for fun, here's a 2.59% optimization on pcc_reapply.
I have another similar optimization in the works.
allison chromatic: the weird thing is, I copied over the tests (passing on pcc_optimize_sig) and the two I added are failing on pcc_reapply 21:25
chromatic That is weird for sure.
dalek rrot: r41857 | chromatic++ | branches/pcc_reapply/t/src/extend.t:
[t] Untodoed test for calling MultiSub from C (r41828).
 21:26
allison chromatic: and diff -u says the two files are identical
rrot: r41858 | chromatic++ | branches/pcc_reapply/src/call/args.c:
[PCC] Switched from VTABLE calls with expensive STRING comparisons to direct

Parrot_pcc_build_sig_object_returns_from_op(). This improves the fib.pir benchmark by 2.858%.
chromatic Yay, Blazers tickets!
allison chromatic: excellent!
purl EGG-see-lent!
chromatic Preseason and exhibition yes, but free! 21:27
allison, it looks like all of the string appending and such to add the return signature is also expensive. 21:28
allison chromatic: to add the string signature from an op call? 21:29
chromatic Yes.
allison chromatic: in theory, we could do away with the string signature if we have the FIA for the signature
chromatic: but, it'd be nice to get rid of the FIA too 21:30
chromatic: perhaps an integer flag on each array element?
integer flags, I should say
chromatic It looks like all of the string manipulation in Parrot_pcc_build_sig_object_returns_from_op() adds some 7-8% overhead.
"sheldor is afk" 21:31
allison chromatic: okay, something to look into
21:32 zerhash joined 21:33 Whiteknight joined
Austin What would cause parrot to just "stop", or at least appear to? 21:33
(Of course, with mad cpu usage...)
Tene Austin: an infinite loop, of course. 21:34
Austin Sure. But I'm running with trace 1, and if there's a loop, it's internal.
Tene yeah
Austin As a hint, I'm trying to return from a sub that just updated the namespace it was called from. 21:35
Hey. Tene, was it you that had the export/import code for parrot?
Tene Austin: yes. 21:37
Whiteknight a sub is changing the namespace it's located in?
Austin I borrowed some of that, but I had to hack it some. I needed to just import an already-loaded namespace, not load any bytecode before hand. I was just shortening names in nqp. 21:38
Whiteknight, yes. I've got a sub that creates and compiles a set of multimethod trampolines. As a result, when you call it, it adds one entry to the calling namespace. 21:39
Tene Austin: if you're up for it, the 'parrot' language could use a "have I loaded this bytecode?" cache.
Austin Tene: maybe so. But mine was just "this is already loaded. bring the names over here, so I don't have to qualify them." 21:40
allison chromatic: so, apparently, there's no way to call 'exists_keyed_str' from PIR
chromatic: there's no op version that calls it
Whiteknight smells a new opcode a'cookin'
dukelet0 loves me some deep fried opcodes 21:41
allison Whiteknight: well, the PIR syntax is "$I0 = exists $P1[$I2]" or "$I0 = exists $P1[$S2]", which create an integer key or string (regular) key 21:42
the keys are PMCs
Tene Austin: there's a method on the Namespace PMC to do that.
Austin Tene: Typing "close::Compiler::Debug::Note(...)" was 'teh suck' 21:43
Tene allison: I posted a patch to add that.
nopaste.snit.ch/18344
allison Tene: to add "exists $P1, $S2"?
Austin Yeah. It needed some snazz, which you had put in the exporter stuff.
allison Tene: cool, adding now 21:44
Tene I didn't want to commit in the middle of you and c working on it. 21:48
allison Tene: that patch is good 21:49
Tene: now I've got Segfault in mark_hash at callsignature.pmc:316
Tene That code looks a little weird. You'd think that the b increment should be in the loop. 21:51
chromatic Hm, it shold.
Tene allison: does that fix it for you? 21:53
allison Tene: a return at the top gave me a quick fix, will move the b increment inside the loop 21:54
chromatic Any idea how to test the exists_keyed_str variant?
allison chromatic: apparently there is no way to do it from PIR 21:55
chromatic I don't see one either. Pity.
allison yes, it is 21:56
21:56 darbelo joined
allison chromatic: okay, I've got it to the point where the shiny new CallSignature is only failing 1 more test than the old CallSignature, so I'm checking it in 21:59
21:59 cconstantine joined
allison (we can fix that one test after) 21:59
chromatic Hang on, I fixed a bug.
r41859 would be good to include. 22:00
allison chromatic: commit already made, but can copy the fix over 22:02
dalek rrot: r41859 | chromatic++ | branches/pcc_optimize_sig/src/pmc/callsignature.pmc:
[PMC] Fixed hash marking in CallSignature PMC, with credit to Tene for
rrot: r41860 | allison++ | branches/pcc_reapply (3 files):
[pcc] Merge in the fast new internals of CallSignature PMC from

tests.
Tene allison: I'm checking in the speedup part of r41859 to pcc_reapply. 22:05
dalek rrot: r41861 | tene++ | branches/pcc_reapply/src/pmc/callsignature.pmc:
[pcc] Minor speedup copied from r41859
allison ah, somebody beat me to it 22:06
(I had already applied moving b->next into the loop earlier
Tene I'm going to time a test run in r41855, and then with current.
chromatic Looks like about a 20% improvement to me. 22:11
jonathan Whee. 22:12
That's pretty good.
22:12 Limbic_Region joined
chromatic We need more, but I'm chipping away. 22:13
22:13 mokurai left
dalek rrot: r41862 | allison++ | branches/pcc_reapply/src/pmc/callsignature.pmc:
[pcc] Add a missing vtable function to CallSignature and reclaim the one extra
 22:15
purl well, failing test is with rev. 38923
allison purl: forget failing test 22:16
purl allison: I forgot failing test
22:19 jrtayloriv joined
nopaste "tene" at 97.117.62.135 pasted "old callsig vs new callsig" (17 lines) at nopaste.snit.ch/18346 22:21
chromatic About 20% then. 22:23
Hm, actually closer to 13% there.
Tene well, that's all of "make test" 22:24
chromatic Right.
22:25 patspam joined
Tene my pcc-parrot right now takes 1.065s on oofib.pir, while trunk-parrot takes 0.465s 22:26
so we're down around the range of 2x vs trunk
22:27 Whiteknight joined
dalek rrot: r41863 | allison++ | branches/pcc_reapply/src/pmc/callsignature.pmc:
[pcc] Reordering a few CallSignature vtable functions to match the grouping of
 22:28
rrot: r41864 | chromatic++ | branches/pcc_reapply/src/call/args.c:
[PCC] Rearranged tests for exceptional condition in

No functional changes.		 22:48
rrot: r41865 | chromatic++ | branches/pcc_reapply/src/call/args.c:
[PCC] Swapped a VTABLE call with a macro to set CallSignature's arg_flags in

the fib.pir benchmark.
darbelo japhb: ping
dukeleto Coke: planet parrot still has no love for me
the fact that smolder reports expire kind of irks me 22:49
darbelo diskpace is cheap, but not *that* cheap. 22:50
dukeleto darbelo: yes, but IIRC, that is not configurable in smolder. perhaps it is now, i haven't looked at the source in a while 22:51
Tene dukeleto: I can always make more for you if you'd like.
dukeleto Tene: more reports? that doesn't help when a random person submits a smolder report from an exotic platform and then it just goes away after an unspecified amount of time 22:52
Tene dukeleto: oh, so you want *useful* feedback.
chromatic Hm, I removed all of the string_sig from Parrot_pcc_build_sig_object_returns_from_op and all (expected) tests still pass. 22:54
mikehh got to re-boot bbiab 22:55
dukeleto Tene: only sometimes :) 22:56
allison chromatic: as I said, we can actually eliminate the string sig once we have the FIA sig
chromatic: the string sig is there for introspection, but could be built on request from the FIA 22:57
chromatic There's a 10.093% speed improvement from removing it. 22:58
allison (the early code iterated over string sigs for arg/param handling, but the current code iterates over FIA instead
chromatic Nothing I can tell inspects the return sig.
allison chromatic: excellent, ditch it
japhb darbelo, pong 22:59
allison chromatic: introspection is a user-accessible feature of the PMC
chromatic: it's the get_string vtable function
darbelo japhb: How smart should plumage be about dependencies? And how smart is it now? 23:00
chromatic Sure, I can't remove the signature altogether.
I only removed appending the return signature.
allison chromatic: but, you can avoid building it until it's needed
chromatic The next big win is getting rid of all of those CPointers.
... but I have a train to catch. 23:01
japhb darbelo: currently, you can hear the wind whistling through the cavernous space where its brain ought to be.
dalek rrot: r41866 | chromatic++ | branches/pcc_reapply/src/call/args.c:
[PCC] Removed return signature string handling from

produces a 10.093% performance improvement on the fib.pir benchmark.
darbelo I kinda expected that :)
allison chromatic: if you save 10% with the returns signature, can you save another 10% with the args signature?
japhb In the future: as smart as we can possibly make it, but iterating up from current. In other words, I don't want to get all architecture astronaut on the dependency design. (more) 23:02
dukeleto chromatic: looks like I should run a fib.pir benchmark across released parrot versions
allison chromatic: catch the train, will catch you tomorrow
japhb It's important to have something working ASAP, because people (including me) want to port over proto's projects. (more)
darbelo proto?
purl proto is to start at the lowest and use the first available. The client should be configurable. or github.com/masak/proto/tree/master
darbelo oh, that. 23:03
japhb And from there, we can keep adding smarts and iterating on the dependency handling pretty much as long as we want.
Like Perl, we want Plumage to eventually feel like an extension of your will, not something that annoys the hell out of you (like, say, yum). 23:04
dukeleto japhb: so what are the next steps for plumage?
japhb dukeleto, glad you asked, I was just about to start talking about that. ;-)
For me, the two main avenues for immediate work are 1) basic dependency handling -- what we just talked about. 2) starting work on the metadata server. 23:05
I have already gotten a donation of server space/cycles to prototype the latter. 23:06
darbelo I just got a simple dep-tracking use case.
japhb (moritz++ for that)
darbelo, do tell. Maybe it will be a good starting point.
darbelo Whiteknight moved matrixy (his MATLAB on parrot) to here github.com/Whiteknight/matrixy/
japhb darbelo, yeah, saw his post (well, *a* post on the subject -- he does so many I don't know if I've read whatever is his latest post :-) 23:07
darbelo and he's looking to rip out and make standalone the linear algebra libs that matrixy needs.
japhb nodnod
darbelo (github.com/Whiteknight/parrot-linear-algebra)
japhb I'd like to do the same with OpenGL, but I was leaving off on that for a while because OpenGL is an excellent test for the NCI system 23:08
OK, so we want to just start supporting that simple one-level dependency?
darbelo So, we have 'matrixy' depends-on 'parrot-linear-algebra' 23:09
Which is about as straighforward as it gets, but once that works we can build up.
japhb All right. Can you commit metadata files for both of those, and I'll make it a priority to hash out a design for handling it properly?
darbelo Looking into it now. Will probably need some work on the language itself before it reaches 'buildable'. 23:11
japhb Sure ... but is the library buildable against install-dev parrot? 23:12
Even if matrixy itself fails to build, we need the dependency to successfully build in order to test the dep handler properly 23:13
darbelo Not right now, from what I can see.
japhb Got cycles to fix that? ;-)
darbelo Diving in as we speak.
dukeleto did someone say linear algebra? 23:14
darbelo :)
dukeleto gets excited
japhb awesome
darbelo dukeleto: github.com/Whiteknight/parrot-linear-algebra
dukeleto darbelo: just saw that!
23:14 dolmen joined
darbelo fortran_helpers.pir 23:14
japhb Now I won't have to write all that when I finally pop a few yaks off my yak shaving stack and get back to 3D fun
dukeleto i talked with particle last year about efficient linear algebra routines for parrot
23:15 plobsing joined
darbelo really wants to see all sorts of numerical stuff on parrot. 23:16
dukeleto like, we need an efficient matrix PMC. and if you want to do 'real' linear algebra, efficient PMCs for real and complex matrices
darbelo: want to help me hack on GSL bindings for Parrot? I promise it will be fun... 23:17
darbelo Sure, where is it?
dukeleto darbelo: in my mind, right now :) But I could have a github repo up in no-time ... 23:18
darbelo: i guess it is conceptually similar to decnum-pmcs, except it is going to be more wrapping library functions instead of creating PMCs. perhaps you have some tuits that you can loan me in this area 23:20
japhb dukeleto, just as long as you provide a method to view the matrices as buffers, so that we can hand the buffer pointer off to other libraries ... like, say, OpenGL ....
dukeleto darbelo: GSL is a pure C library. I want to access it's functions from Parrot, perhaps as PMCs, perhaps not. what is the best plan of action to start?
darbelo: for example, when I wrote Math::GSL ( perl interface to GSL) i started off just getting the special function subsystem working, which has reasonably simple function call signatures 23:21
Tene dukeleto: write a PIR library wrapping the functions. 23:22
dukeleto darbelo: i don't think that subsystem fits well as a PMC. i just want to use NCI to call functions C functions from PIR
Tene: that is what I am going after, I guess. can you suggest the simplest example of a PIR library that wraps a C library?
Tene github.com/tene/parrot-elementary/ is my example 23:23
darbelo dukeleto: I agree, from what little I know about GSL it looks like a pir wrapper is the way to go.
Whiteknight dukeleto: I want to add an efficient matrix PMC
that's one of the goals of the parrot-linear-algebra package
dukeleto darbelo: gsl does have some subsystems that would work as PMCs, but most just need wrapping
Tene dukeleto: that library also has some wrapper classes around the obejcts that library makes that may or may not be relevant to you. 23:24
darbelo dukeleto: also see NotFound's mysql wrapper.
parrot-mysql?
dukeleto Whiteknight: i talked about that with particle at the gsoc mentor summit last year. i didn't have the parrot tuits back then to act on what he told me, but now I think I do
Whiteknight: i am all about numerics on parrot, so consider me 'all-in' on your parrot-linear-algebra and matrixy projects. 23:25
darbelo purl: parrot-mysql is code.google.com/p/parrot-mysql/
purl OK, darbelo.
Tene If all you need to do is manage object points to pass to C functions, you don't need a PMC, just a PIR class.
s/points/pointers/
dukeleto Tene: each gsl subsystem has slightly different needs. there about 45 subsystems
japhb yikes 23:26
dukeleto Tene: parrot-mysql looks like it could be what I need to look at, thanks!
Whiteknight dukeleto: awesome! I've got commit bits galore if you want one
it's going to be a while before either of those projects build or anything, but will be valuable
Tene I remember parrot-mysql as doing rather more than just wrapping the C functions.
dukeleto Whiteknight: yeah, i would definitely like to have commit bits on your github repos
Tene but I don't trust tha tmemory too well.
dukeleto Whiteknight: does either project have a test suite yet? 23:27
dukeleto loves me some test suites
darbelo Whiteknight: matrixy's Configure.pl fails horribly for me, looks like it's assuming a home in the parrot build tree.
dukeleto writing the plumage test suite in NQP was really quite fun
darbelo is collecting commit bits this days. 23:28
dukeleto japhb: i don't quite know what to work on next on plumage. let me know what needs doin'
japhb: i am still thinking about how to properly mock out the source code repos so that I can test each plumage command without hitting 'live' servers. or should we just not care and hit live servers in our test suite? something rubs me the wrong way about that 23:30
network issues => failing tests => angry dukeleto
japhb dukeleto, let me think on both of those. Have to finish a thought at $day_job first.
It sounds like I have requests to think about 1) dep handling, 2) testing mocked source repos, 3) more tasks to share. About sum it up, darbelo and dukeleto? 23:31
dukeleto japhb: currently i am leaning towards changing the URLs in metadata in the test suite, so that it check out/clones from a file:/// protocal repo that lives in the test suite 23:32
Whiteknight Matrixy doesn't build
dukeleto japhb: does Plumage::Downloader support file:/// ?
Whiteknight 1) it doesn't work against an installed parrot. That needs to be fixed
japhb dukeleto, I don't think so, I think it's http-only.
Whiteknight 2) It requires CBLAS and CLAPACK libraries, the bindings to which I moved into parrot-linear-algebra as a separate project 23:33
dukeleto japhb: actually, that shouldn't matter. svn and git will be dealing with the URLS, not plumage, correct?
Whiteknight so if anybody knows how to make a language build against an installed parrot, I would love to hear about it
Tene dukeleto: Plumage::Downloader *should* support all supported url types. It currently doesn't. 23:34
darbelo Whiteknight: I can probably get you building against installed parrot, depending on how attached you are to your current build infrastructure.
dukeleto Whiteknight: when you figure that out, please write a blog post about it. many others are in your situation :)
japhb dukeleto, and I think we will probably do something similar. But I'm thinking shipping the moral equivalent .tar.gz of small git and svn repos, so that we can extract them locally and use real git and svn commands to access them.
allison Whiteknight: it's mainly just a matter of getting your paths right
Whiteknight: specifically, not assuming the build directory paths
japhb dukeleto, how are you going to test dep handling? hmmm, I need to tank on this.
dukeleto japhb: yes, that is what I was thinking. we can use essentially empty svn and git repos. we just need to test that they are checked out properly 23:35
allison Whiteknight: Tcl is a good example
dukeleto japhb: that question requires beer
Whiteknight allison: okay, awesome
dukeleto, darbelo: you're both "collaborators" on both projects 23:36
I think that's the same as a commit bit
dukeleto i am fast approachin the need to write Test::MockObject for Parrot . i hear one of the parrot core committers knows a thing or two about it...
Whiteknight: yes, it is. thanks!
Whiteknight dukeleto: I actually wanted to start a mock framework!
dukeleto Whiteknight: we are like two peas in a pod or something today!
Whiteknight for serious 23:37
dukeleto scary, even
Whiteknight should be relatively easy. Create a PMC type that overloads find_method in some way 23:38
hard part will be unpacking arguments, but with :call_sig that won't even be so hard 23:39
23:43 mokurai joined
Whiteknight of course, maybe I'm underestimating. I've never made a mock object framework before 23:46

Powered by Cro and the Raku® Programming Language.

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

Thank you!