Parrot 2.6.0 | parrot.org Log: irclog.perlgeek.de/parrot/today | Nopaste: nopaste.snit.ch:8001 | merge html_cleanup (talk to Coke), merge gc_* branches, fix/replace/optimize hashing
Set by moderator on 10 August 2010.
00:00 Psyche^ joined
dalek kudo: 73a7674 | masak++ | :
<pre>m src/core/Buf.pm

   <pre style='white-space:pre-wrap;width:81ex'>[Buf] added prefix/infix ~^, infix ~& and infix ~|</pre>		 00:00
darbelo chromatic: Looks good to me. For after the release anyway. 00:02
chromatic Right. Want to try that experimental opcode in a branch?
darbelo Sure. I still have some tweaks to unshared_buffers to make. 00:03
So, prbably tomorrow-ish
chromatic Sounds good. 00:04
darbelo Although anything that makes the rakudo build's memory usage less pathological will likely benefit that as well. 00:05
chromatic True. I'm still of mixed feelings about getting rid of shared buffers. 00:06
dalek kudo: bef86ee | masak++ | :
<pre>m t/spectest.data

   <pre style='white-space:pre-wrap;width:81ex'>[t/spectest.data] added S03-operators/buf.t</pre>
darbelo I'm experimenting with the tuning, right now. If I make compacting more aggressive I can build rakudo under 300mb of mem, but it'll take about hour. 00:08
chromatic How fast is your machine? 00:10
darbelo At the branch point, it builds in ~8 minutes, for comparison. 00:11
chromatic Ouch.
jnthn I'll take the extra memory usage over an hour long build. :-) 00:12
darbelo Yeah. Me too.
chromatic If that opcode works though....
jnthn Yes, that should result in a load less string headers. 00:13
darbelo jnthn: OTOH, kid51 can't build at all without some memory usage reductions, which is what made start experimenting along this line.
00:27 jsut joined 00:46 gbacon joined
dalek rrot: r48518 | darbelo++ | branches/unshared_buffers (67 files):
Sync with trunk.
 00:52
01:13 theory joined 01:34 chromatic joined 01:35 Andy joined 01:46 kid51 joined
dalek rrot: r48519 | chromatic++ | branches/substr_eq_at:
Created branch to experiment with substr_eq_at opcode
 01:58
02:05 tcurtis joined 02:10 Chandon joined 02:12 jimk joined 02:24 mj41 joined
dalek rrot: r48520 | darbelo++ | branches/unshared_buffers/src/gc/alloc_resources.c:
Typo fix.
 02:31
purl hmmm... typo fix is in documentation (perlapi), not a comment
luben good morning parrot 02:40
Andy proof please? gist.github.com/526305 02:45
02:48 janus joined
Andy Any comments? 02:50
purl Any comments is appriciated =) or punt!
dafrito is reading :) 02:51
Andy kthx
tcurtis Looks good. No grammar- or typos that I notice. 02:52
dafrito Yeah, I don't see anything glaring 02:54
Perhaps this could read better? "if Perl 6 programmers wanted vim to support Perl 6, they had a perl6.vim file that got passed around"
Andy Reload, I updated that.
I just edited the file 02:55
tcurtis wait... Perl has a "static" keyword now? 02:56
dafrito "because of updates of all the Perl-related support files" Should one of those 'of' be 'for'?
Andy eek, state 02:57
STATE
thanks.
TiMBuS i would like vim to detect filetype based on mimeinfo =/ 02:59
all my perl 6 files are application/perl6
sorear What OS are you on?
dafrito Andy: You could also put a link for "Template Toolkit" if you want
TiMBuS ubuntu linux
Andy TiMBuS: How are you storing mimeinfo?
perlbuzz.com/2010/08/vim-73-supports-perl-6.html published 03:00
thanks
TiMBuS er, im not sure of the specifics Andy. I used associgate to add the magic (in the past I modified a mime.magic file but i cant recall where it was stored...) 03:03
sorear you mean /etc/magic.mime? 03:11
TiMBuS nah, looks like it was /usr/share/mime 03:12
and associgate adds to /usr/local/share/mime
03:17 atrodo joined
dalek rrot: r48521 | chromatic++ | branches/substr_eq_at (7 files):
[str] Added API to find substring in string.

may help PCT require far fewer new STRING headers.		 04:26
rrot: r48522 | chromatic++ | branches/substr_eq_at (5 files):
[ops] Added substr_eq_at ISSI op for PCT.
cotto msg khairul minor typo in the op instrumentation page: "If $get_raw is pass and is true" 04:55
purl Message for khairul stored.
dukeleto cotto: pong 04:58
cotto dukeleto, Does it matter if a student keeps working on a project after the pencils down date? 05:00
i.e. would that just mean that all evaluation doesn't take the extra work into consideration? 05:01
chromatic Know any so-called professional programmers who always make their estimates? me neither. 05:02
cotto is surprised he was able to remember his question.
dukeleto cotto: yes, that is fine 05:13
cotto msg khairul The class instrumentation page should probably explain what a call sig is and where to find out more about it.
purl Message for khairul stored.
cotto dukeleto, great. Thanks.
dukeleto cotto: the final evalution should only take into account stuff done before the pencils down. We should encourage all students to keep working well after that :) 05:14
cotto It'll only really matter for the survey anyway. There's very little my student could do at this point to avoid passing. 05:15
dukeleto cotto: sounds good 05:17
cotto msg khairul Other than those suggestions, the docs are looking good. When you put your code on github, make sure the use guide goes with it. 05:23
purl Message for khairul stored.
05:27 jsut_ joined 06:05 uniejo joined 06:34 preflex joined 07:43 fperrad joined
luben chromatic, ping 08:18
chromatic pong 08:19
luben I have made experiments with hashes
added paged allocation for buckets
it is slower here
here is it: luben.spnet.net/gitweb/ 08:20
chromatic I think we should address the indirection of our hashing functions before we make progress on optimizing our hashes.
luben the paged allocation branch is "single-paged" 08:21
chromatic Instead of function pointers for key/value hashing, set the type of a hash for the most common types and hard-code the comparison/hashing functions.
luben yes, this will give us better performace 08:22
chromatic There's too much overhead there so it dwarfs any improvements we make in allocation/reallocation/bucket distribution.
luben in the benchmaks for hashes, the ones that use C macros or C++ templates are fastest (they do not use any indirection) 08:23
chromatic This is one place where I wish we could use C++.
luben yes 08:24
I will be offline for 2 weeks - going on vacation. When I come back I will continue
chromatic Excellent, this would be good for 2.8. 08:25
luben yes
moritz luben++
luben from my branches "hash-iter-macros" is ready/tested
chromatic I'll apply it as soon as 2.7 comes out.
luben If somebody could clone it, because in 7-8 hours I will whitch off the PC here 08:26
Or I could atach a patch to the ticet 08:27
*ticket
chromatic A patch is the easiest.
must sleep now 08:28
thanks for your work
08:48 AndyA joined 09:43 gbacon joined 10:10 JimmyZ joined 10:45 aloha joined 10:56 lucian joined 11:42 NotFound joined
NotFound Hi 11:43
11:54 ruoso joined
jnthn I'm told Rakudo's core.pm compilation is nomming lots and lots of RAM again. Wonder if recent tuning has notched things a little too far along the RAM/CPU trade-off scale. 12:08
12:10 whiteknight joined
whiteknight good morning, #parrot 12:18
szabgab if I am not mistaken parrot also uses the public smolder server
if so, anyone knows what happened to it?
smolder.plusthree.com/ does not seem to respond
NotFound whiteknight: pon 12:25
g
12:29 elmex_ joined 13:11 allison joined 13:19 gbacon joined 13:27 Paul_the_Greek joined
Paul_the_Greek Hey ho! 13:27
purl,messages 13:28
ping cotto 13:29
dalek rrot: r48523 | coke++ | branches/substr_eq_at/src/ops/experimental.ops:
Add a pod description for this opcode.
 13:32
whiteknight NotFound: how do I load a library in winxed? I can't find an example 13:34
13:39 Paul_the_Greek joined
Paul_the_Greek ping anyone 13:39
whiteknight ? 13:40
Paul_the_Greek Just checcking to see if I'm really connected.
whiteknight signs point to yes 13:41
Paul_the_Greek Downloaded XChat onto my wife's laptop. Flakey wifi connection here.
whiteknight: Do you know why PMC attribute groups are allocated separately from other fixed-size blocks? 13:42
whiteknight how differently? I thought they used the fixed-size allocator? 13:43
13:54 Paul_the_Greek joined
Paul_the_Greek whiteknight: Are you still here? 13:54
ping cotto 13:56
ping anyone 14:01
moritz pong
Paul_the_Greek Hey moritz. Just trying to get this chat program to work. Got a flakey wifi connection.
14:03 bubaflub joined
Paul_the_Greek Hey bubaflub. 14:04
purl bubaflub is Bob Kuo, mailto:bobjkuo@gmail.com
bubaflub hello Paul_the_Greek
moritz purl is very noisy
Paul_the_Greek Yes, he is chatty. 14:05
Memory management question, anyone? 14:08
moritz we need to improve it! 14:09
that's how much I know about it :)
Paul_the_Greek I think I can speed up PMC attribute allocation.
Logging off and back on ... 14:12
14:13 Paul_the_Greek joined
Paul_the_Greek Cool! 14:13
moritz Paul_the_Greek: btw you can also use the IRC logs at irclog.perlgeek.de/parrot/today for reading here if your connection is not stable
Paul_the_Greek I'll do that. I just don't want to pop off in the middle of conversations and annoy everyone. 14:14
Be back later. 14:16
dalek kudo: 2f18a49 | pmichaud++ | :
<pre>m src/Perl6/Grammar.pm

   <pre style='white-space:pre-wrap;width:81ex'>Fix precedence of numeric bitshift operators; fixes RT #77232 (masak++).</pre>		 14:29
14:45 Andy joined 15:12 chromatic joined
pmichaud chromatic: just sent email about substr_eq_at 15:17
chromatic I'm reading it and trying to understand it before breakfast. 15:19
The "before breakfast" part may be a mistake.
pmichaud if I can help demystify it, let me know. :) 15:28
the key is that it's more like C's strncmp than anything else. 15:30
i.e., only compare up to n characters
where 'n' is implied by the length of the token string argument (i.e., the one we're searching for)
chromatic What you want or what it's doing?
pmichaud what I want 15:31
what it's doing seems to be more like strcmp with an offset
chromatic Right. 15:32
pmichaud perhaps a better way of phrasing what I want is: strncmp with an offset
in C, it'd be strncmp(s1 + offset, s2, strlen(s2)) 15:33
and then for equality I'd look for zero as the return value 15:34
if we want the number of characters to be compared to be explicit in the opcode, that'd be fine with me also
i.e., it could be 15:35
15:35 Paul_the_Greek joined
pmichaud $I0 = strncmp $S0, $S1, offset, length 15:35
chromatic That's what the code does... supposedly.
pmichaud I don't think it honors length
chromatic ascii_compare does when the encodings are the same, as they are here. 15:36
Paul_the_Greek Memory management question?
chromatic go ahead, Paul_the_Greek.
Paul_the_Greek Why is there one set of fixed-size pools for PMC attributes and another set for everything else? 15:37
Is it just for efficiency?
chromatic Good question. That might have been the desire.
pmichaud in the branch, I'm getting
$I0 = substr_eq_at "hello", "he", 0
say $I0
1
if substr_eq_at is returning cmp semantics, that should be zero. 15:38
Paul_the_Greek I think I might be able to speed up allocating PMC attributes.
pmichaud I *think* what it's doing is looking at the 'l' following the "he" and saying "oh, the first string is greater" and thus returning 1. 15:39
i.e., it's not stopping the comparison after the first two chars.
chromatic The return value processing is wrong.
Paul_the_Greek If it returns -1/0/1, why is it called 'substr_eq'? 15:40
pmichaud Paul_the_Greek: noted, it should be 'cmp'
chromatic That... is a different problem.
I believe I've fixed the return value; I'm running the tests to make sure. 15:41
pmichaud chromatic: can you nopaste the diff?
nopaste "chromatic" at 192.168.1.3 pasted "pmichaud: the patch" (14 lines) at nopaste.snit.ch/22830 15:42
chromatic It's not completely right, but it passes your test cases.
pmichaud looking 15:43
Paul_the_Greek Why the compare ret_val < 0? 15:44
pmichaud I have the same question 15:45
seems like one can more usefully do
Paul_the_Greek Does memcmp return something other than -1?
pmichaud return memcmp(lhs->strstart + offset, rhs->strstart, rhs->strlen)
Paul_the_Greek Yeah.
chromatic It was the cheapest, easiest patch. 15:46
Paul_the_Greek Why not just 'return retval;' ? 15:47
pmichaud Paul_the_Greek: because min_len will be wrong in that case
chromatic I was too lazy to type man memcmp.
pmichaud there are a few edge cases that this code doesn't quite handle 15:48
chromatic Right.
Paul_the_Greek But if the only negative value of retval is -1, then the comparison is pointless.
pmichaud oh oh oh oh 15:49
Paul_the_Greek Maybe I don't have memcmp correct in my mind.
pmichaud this whole approach isn't going to quite work
chromatic Right.
pmichaud the ascii_compare function itself doesn't have a usable 'n' limit
chromatic I'm not sure it needs one.
I've been using the size of the smaller string. 15:50
pmichaud that doesn't quite work
because $I0 = substr_eq_at 'ab', 'abc', 0
shouldn't show them as being equal
15:50 tcurtis joined
pmichaud while 15:50
$I0 = substr_eq_at 'abc', 'ab', 0
should
be "equal" 15:51
chromatic Ah, right.
Reusing ascii_compare isn't going to work so well here.
pmichaud right
there really needs to be a strncmp equivalent 15:52
chromatic pmichaud, if you commit these test cases (and the interface you prefer for this), I'll make them pass.
Paul_the_Greek Is that how the regular compare works?
pmichaud chromatic: would an opcode name of "strncmp_at" seem to be appropriate?
chromatic Fine with me, if it works for PCT. 15:53
Paul_the_Greek Once you delimiter the substring, then the compare should work the same.
moritz why the 'n' ?
pmichaud moritz: because I need to limit the length of the comparison
look, here's the use case
moritz pmichaud: ok, I thought the limit was implicit
pmichaud src = "given xyz { when 1 { ... } }"
moritz I know the use case, yes 15:54
pmichaud moritz: oh. I'm thinking maybe it's more general to have the caller supply 'n' instead of it being inferred
moritz wfm
Paul_the_Greek Optional n, as in substr, right?
pmichaud for the case of comparing against a constant string, the length is statically known at compiletime
i.e., instead of
$I0 = strncmp_at $S0, $S1, offset # length is implied as length($S1) 15:55
I think
$I0 = strncmp_at $S0, "when", offset, 4 # length is explicit
Paul_the_Greek Are you using a substring of $1 or $2?
pmichaud $1 15:56
Paul_the_Greek Then shouldn't the offset and length apply to $1?
pmichaud Paul_the_Greek: yes
(although the length should probably apply to both)
Paul_the_Greek: if you're suggesting a different order of arguments, I'd be agreeable
Paul_the_Greek So if the length is missing, it's the tail of $1 starting at offset.
pmichaud no
essentially I need an opcode that does the same as the following in C: 15:57
i = strncmp(s1 + offset, s2, n)
Paul_the_Greek Okay, so it's not a substring of $1, just a tail of $1.
pmichaud one can look at it that way, yes
one can also look at it as being a substring of n characters of $1 and end up with the same results 15:58
Paul_the_Greek You're sure you don't want strcmp(substr($1, offset, n), $2) ?
pmichaud we're just trying to avoid creating the separate substring
strcmp(substr($1, offset, n), $2) is what we do now.
Paul_the_Greek Not if *$1 < n
Sorry, *$2 < n
Oh, this is in addition to that. Got it. 15:59
pmichaud right
currently to do a token comparison, we generate
$S1 = substr src, offset, n
$I0 = iseq $S1, "when"
NotFound whiteknight: the load_bytecode and loadlib predefined functions of winxed are equivallente of the opcodes with the same name. 16:00
pmichaud but the problem with that approach is that we end up creating a new substring for every token comparison
actually, I'll be a bit more precise
currently we generate
$S1 = substr src, offset, 4
Paul_the_Greek So what is wrong with the existing op code?
pmichaud $I0 = iseq $S1, "when" 16:01
Paul_the_Greek This one: strcmp(substr($1, offset, n), $2) is what we do now.
pmichaud that's not an opcode.
that's the equivalent C semantic
Paul_the_Greek Sorry, now I understand.
pmichaud the opcodes are what I just wrote -- substr followed by iseq
Paul_the_Greek Why not make the new opcode do that? 16:02
pmichaud Paul_the_Greek: that's what we're doing
I'm proposing instead:
$I0 = strncmp_at src, "when", offset, 4
with the difference that $I0 returns -1/0/+1 instead of just 0/1 16:03
Paul_the_Greek We're going in circles. Does $4 specify the length of the substring of src, or the length of the comparison?
pmichaud in this case I'd let $4 be the length of the comparison. (more)
chromatic The length of the comparison, because the op knows the lengths of both strings. 16:04
pmichaud for my use cases, it doesn't matter though, since $4 would always be the known length of $2
but for a more general-purpose operator, having it follow strncmp semantics ('n' is the length of the comparison) seems more useful.
Paul_the_Greek But might it be more generally useful if $4 was the length of the substring?
pmichaud Paul_the_Greek: I doubt it. 16:05
and I'm not sure what "length of the substring" means here.
Paul_the_Greek If the tail of $1 is shorter than $4, then $4 cannot be used as the comparison length.
pmichaud why not?
strncmp does it
in that case, assuming the prefixes are equal, we'd get back -1
because the lhs is less than the rhs
Paul_the_Greek Because then you would compare more characters than are in the tail of $4.
pmichaud you stop comparing if $1 is shorter. 16:06
Paul_the_Greek Right, so $4 is not necessarily the comparison length. But if that's the semantics of strncmp, then I'm raising a red herring.
pmichaud basically $4 says "compare at most n characters" 16:07
if either of the source strings doesn't have enough, then that stops the comparision
Paul_the_Greek Limited by the length of both strings, I presume. Sorry for belaboring this.
pmichaud chromatic: since Parrot already has cmp_str, perhaps this should be cmp_str_at ? 16:08
chromatic Makes sense.
pmichaud oh, but it wants an 'n' in there somewhere.
cmp_str_n_at ? cmp_str_at_n ? 16:09
I guess the latter -- and the underscores hint at the arguments.
Paul_the_Greek ncmp_str_at ?
chromatic I don't know if we need the n; for PCT's purposes, it's superfluous, right?
16:09 allison joined, bacek joined
pmichaud yeah, PCT doesn't really care. I'll go with cmp_str_at, and we'll just know that the 'n' argument is required for now. 16:10
Paul_the_Greek pmichaud: Thanks for putting up with my questions.
pmichaud or perhaps it should be cmp_substr_at
16:10 aloha joined
pmichaud Paul_the_Greek: thanks for helping us clarify exactly what we want/is meant 16:10
chromatic Does that help write test cases? 16:11
pmichaud yes
I'm writing test cases now.
will commit shortly, then have to run an errand.
szabgab hi, can someone help me understanding the status of languages on Parrot? 16:12
I looked at the link bitbucket.org/allison/pynie/ and Pynie seems to be stuck 5 month ago
allison szabgab: yes, I'm the only Pynie developer, and yes, I haven't had time to work on it lately 16:13
szabgab hi allison
allison szabgab: it's current status is "starting over from scratch"
moritz the curse of being involved in too many projects/activities, I guess
allison szabgab: because it depends on the old NQP
Paul_the_Greek Pynie is? 16:14
moritz huh? most things can be quite easily adopted
allison Python on Parrot
moritz to nqp-rx
szabgab so does that mean the Python community has no much interest / tuits to give a hand?
(we could send them a box of tuits :)
Paul_the_Greek You can buy those?
allison moritz: except that Pynie was a) not a great implementation to start with and b) originally an implementation of 2.5, not 3.x
kthakore szabgab: how do you buy those?
Paul_the_Greek: gosh you beat me 16:15
:p
allison szabgab: the Python community is very interested in it, but not in contributing to it
szabgab Paul_the_Greek: kthakore you have to come to a YAPC to get them :)
allison szabgab: my current thought is to redo Pynie as a Parrot backend for PyPy
szabgab allison: thanks, is there any other language (except Perl 6) that is being actively developed for Parrot? 16:16
allison szabgab: kill two birds with one stone in getting a parser for Python, and getting an active Python community for the implementation
16:16 ruoso joined
szabgab Cardinal last changed in May 16:17
Paul_the_Greek I presume the compiler tools can deal with a whitespace language?
szabgab maybe implement Python as a Perl 6 regex?
allison szabgab: that depends on whether "active" means "activity equal to Perl 6" and whether "language" means only major languages
pmichaud (whitespace language) actually, that ended up being far simpler than I had expected
Paul_the_Greek I'm amazed.
pmichaud it fell almost directly out of the python (2.5) specification
16:18 davidfetter joined
szabgab allison: I don't have hard definition for active and it does not have to be major 16:18
arnsholt szabgab: I'm working on a Prolog, but it's slow going since I have to do other stuff (like working on my degree =) as well
allison szabgab: then yes, there are lots of active languages on Parrot
Paul_the_Greek Speaking of languages, but not HLLs, what is the long-term plan for the PIR assembler? 16:19
szabgab I am interviewing in Hungarian (!) about Perl 6 and they also asked about Parrot
Paul_the_Greek Cool.
kthakore szabgab: I will get there soon ish 16:20
szabgab so I am trying to figure out what to say
kthakore szabgab: next summer I will definately plan for it
ok work
szabgab arnsholt: prolog is not listed on www.parrot.org/languages ? 16:21
16:21 theory joined
arnsholt No, it's relatively recent (I started in May or so) 16:22
So I just haven't gotten around to adding it to the page
Also, it's currently somewhat useless as I'm still working on figuring out how do the implementation (also figuring out the semantics of Prolog beyond what I need to write Prolog programs) 16:23
szabgab kthakore: here isa temporary round tuit: Ź˜ 16:24
Paul_the_Greek That came out as a square tuit on my display.
moritz thinks that prolog compiler writing is less obvious than for procedural languages
arnsholt Yeah, Prolog has its quirks 16:25
szabgab Paul_the_Greek: something about square pegs and round holes? 16:26
arnsholt Thankfully I have a good guide to getting backtracking and such working in a CPS style from a book by Paul Graham (On Lisp) though
Currently I'm fighting a bit with Parrot's continuations though
moritz I guess they fight back :-) 16:27
Paul_the_Greek Once I learn more about Parrot, I might consider implementing my language on it. That would be fun. 16:28
arnsholt Indeed they do. I have an idea how I might be able to fix it though
Just need the time to do it
Paul_the_Greek: I have several ideas of languages Parrot needs =)
Paul_the_Greek I use my own personal language for writing everything for my work. The rest of the world doesn't need it. :D 16:29
szabgab Paul_the_Greek: your langauge? you mean Greek ?
Would be nice to have a spoken language implemnetd on Parrot :) 16:30
Paul_the_Greek Greek is good, but not for programming.
moritz hey, in Greek you can use lambda calculus, beta reductions, ... :-) 16:31
pmichaud chromatic: tests committed as t/op/substr_cmp.t in r48524 16:32
Paul_the_Greek The language would have lots of epsilons, too.
szabgab I see pmichaud is trying to fight our noise :) 16:33
chromatic Thanks, will have them passing later today.
pmichaud chromatic: excellent. I can then quickly test them on nqp and come up with a rough estimate of performance change (if any)
davidfetter szabgab, just the person i'm looking for! :) 16:34
got a minute re: CPAN & SuSE?
Paul_the_Greek pmichaud: Do we have performance tests for various aspects of Parrot? 16:35
pmichaud Paul_the_Greek: nafaik
particle like, benchmarks?
Paul_the_Greek Er, right, the word would be benchmark.
dalek rrot: r48524 | pmichaud++ | branches/substr_eq_at/t/op/substr_cmp.t:
[substr_cmp]: Add tests for proposed/experimental cmp_str_at opcode.
particle alioth, the debian language shootout
Paul_the_Greek Looking up ... 16:36
particle i wonder if someone would write nqp versions of those benchmarks, if that would help
chromatic I still think we need NQP and parsing specific benchmarks.
Paul_the_Greek Do we have a start-up benchmark? 16:37
particle we have a 'time to compile rakudo' benchmark
Paul_the_Greek Oh, that's good. 16:38
If I try to improve PMC attribute allocation, I'll need a benchmark.
pmichaud question: Even with the proposed cmp_str_at opcode, I can also get NQP to optimize single-character comparisons to use the 'ord' opcode. Would 'ord' still be significantly better than cmp_str_at to warrant that optimization?
chromatic darbelo might be able to answer that question. 16:39
Paul_the_Greek, see examples/benchmarks/*pir
I like oofib.pir.
pmichaud I'm guessing "likely yes"
after lunch I think I'll go ahead and put that optimization into nqp and see if it's a significant win
after looking at the generated code for rakudo's parser, I'm thinking it could be very significant 16:40
Paul_the_Greek Is the ord index known to be valid?
Thanks, chromatic.
pmichaud there are a lot of generated 1-character constant string comparisons
afk, lunch 16:41
Paul_the_Greek pmichaud: Check out the number of substr , , 1 in the system.
pmichaud Paul_the_Greek: yes, that's what I was pointing to 16:44
(really gone now)
Paul_the_Greek chromatic: What are the future plans for the PIR assembler? Continue work on IMCC? 16:47
chromatic I'm curious to see how PIRATE turns out.
Paul_the_Greek Is it a new assembler for PIR? 16:48
chromatic Yes. 16:49
davidfetter is the production version scheduled for mid-september?
Paul_the_Greek Who is working on it? 16:50
chromatic bacek has been working on it.
davidfetter say, around the 19th?
Paul_the_Greek Is it much different from IMCC? 16:53
chromatic It's written in NQP. 16:54
Paul_the_Greek Does it have fancy macros or simple ones?
chromatic I don't know that.
16:59 DrForr joined 17:02 brianwisti joined
Paul_the_Greek chromatic: Who is the current memory management expert, if any? 17:03
chromatic A handful of us: me, whiteknight, bacek, whoever else pokes in there. 17:04
Paul_the_Greek Okay, then I'll ask everyone if I have any questions.
17:10 jsut joined 17:20 cotto_work joined
cotto_work ~~ 17:25
17:25 plobsing joined
dalek rrot: r48525 | NotFound++ | trunk/src/pmc/null.pmc:
[cage] reorganize and improve pod of Null PMC
 17:28
dukeleto I am seeing t/library/pg.t fail on trunk on linux 17:35
(datura)(~/svn/parrot)$ git 17:36
git github_post_receive git-receive-pack git-shell git-spread git-upload-archive git-upload-pack
(datura)(~/svn/parrot)$ git
dukeleto facepalms
gist.github.com/527377 is the failing t/library/pg.t output 17:37
dalek rrot: r48526 | NotFound++ | trunk/src/pmc (5 files):
[cage] update headerizing
 17:44
kudo: b647188 | pmichaud++ | :
<pre>m build/PARROT_REVISION

   <pre style='white-space:pre-wrap;width:81ex'>Bump PARROT_REVISION in preparation for Parrot release.</pre>		 17:56
17:57 cotto_work joined
tcurtis cannot get installed parrot to work lately. 17:57
17:57 gbacon joined
tcurtis I keep getting packfile version errors. Even nuking my previous parrot install didn't help. 17:57
nopaste "NotFound" at 192.168.1.3 pasted "PLA multiply example in winxed" (28 lines) at nopaste.snit.ch/22832 18:02
NotFound msg whiteknight nopaste.snit.ch/22832 PLA multiply example in winxed 18:03
purl Message for whiteknight stored.
plobsing tcurtis: PBC_COMPAT was bumped, you have to remake any and all .pbc files to run them on a new parrot. 18:04
tcurtis plobsing: would nuking my parrot installation (rm -rf $PREFIX/lib/parrot) and realcleaning not do that? 18:05
plobsing it depends. For example, if you're trying to use non-core pbc files (eg: rakudo), they would not be remade by default. 18:07
but I would expect it to work for core.
cotto_work reconfig should work 18:08
18:08 s1n joined
tcurtis Even parrot with no args and parrot_config do it for me. I'll try again tonight after GSoC pencils down and see if I can figure out what [I am doing || is going] wrong. 18:09
plobsing try a clean checkout perhaps? 18:10
tcurtis will do.
18:14 AndyA joined 18:16 Paul_the_Greek joined 18:17 kurahaupo joined
pmichaud odd datapoint.... switching nqp-rx to use 'ord' instead of 'substr' for single-character literal comparisons did not appear to significantly improve rakudo compile time. 18:27
18:27 AndyA joined 18:31 ruoso joined
whiteknight Notfound++ 18:31
dalek ee-optimization: f270ec3 | tcurtis++ | :
<pre>m docs/Tree/Optimizer.pod

   <pre style='white-space:pre-wrap;width:81ex'>Fix typo.</pre>		 18:32
tcurtis O.o
dalek ee-optimization: 198314c | tcurtis++ | :
<pre>+ docs/Tree/Optimizer/Pass.pod

   <pre style='white-space:pre-wrap;width:81ex'>Add docs for Tree::Optimizer::Pass. Last commit before I tag pencils down.</pre>		 18:38
whiteknight tcurtis++
html--
Paul_the_Greek pmichaud: Weird. 18:39
pmichaud Paul_the_Greek: I'm guessing it's overwhelmed by other factors 18:40
tcurtis wonders how the HTML is getting in there.
pmichaud tcurtis: I'm thinking github may have changed the format of their rss feed
and that dalek hasn't caught up
Paul_the_Greek pmichaud: Did you time this against the system that shares substrings or one that does not?
whiteknight you know, github does have service hooks for IRC. We don't need dalek, we could ask github to automatically report commits to here
pmichaud Paul_the_Greek: ? 18:41
cotto_work but does it do karma?
pmichaud I timed it against parrot trunk, substring version versus ord version
Paul_the_Greek pmichaud: I don't know whether the trunk still has shared substrings or not.
pmichaud Paul_the_Greek: I don't know either.... but since nqp-rx (and rakudo) always tend to work against parrot trunk, that's my basis for comparison 18:42
Paul_the_Greek It has to allocate a string header one way or the other.
pmichaud ord requires allocation of a string header? 18:43
that would be..... odd
Paul_the_Greek No, substr.
pmichaud right
chromatic If the comparison is against a string literal, that string literal will always be a constant string literal in the bytecode.
Paul_the_Greek Sorry to be a pest: Can someone tell me where DPOINTER is defined? 18:45
pmichaud the patch I made changes, e.g.
$S11 = substr target, pos, 1 18:46
plobsing Paul_the_Greek: include/parrot/config.h
pmichaud ne $S11, '#', fail
to
$I11 = ord target, pos
ne $I11, 35, fail
Paul_the_Greek plobsing: Thanks. That must be generated; not available in online browser. 18:47
plobsing Paul_the_Greek: 'make tags-{vi,emacs}' makes it really easy to find these things 18:48
Paul_the_Greek plobsing: I'm on my wife's laptop at the beach, so can only look online. Easy pie at home. 18:49
That's why I apologized for being a pest. ;)
dalek rrot: r48527 | NotFound++ | trunk/t/op/string.t:
fix and improve substr oob tests
 18:50
rrot: r48528 | plobsing++ | trunk/src/packout.c:
eliminate useless var
rrot: r48529 | NotFound++ | trunk/t/op/string.t:
test for substr null string
chromatic pmichaud, five of your new tests still fail, but I should have them passing in a few minutes. 18:55
pmichaud chromatic: excellent 18:56
tcurtis created pencils-down tags in both tree-optimizations repo and simple-optimizations repo.
Paul_the_Greek pmichaud: That's a testament to the speed of substr if your optimization doesn't help much. 18:57
pmichaud Paul_the_Greek: I think it's more that there are other factors that swamp the difference. It's also very possible/likely that LTM is avoiding a lot of unnecessary checks.
Paul_the_Greek Well, it's been a tough day so far, so time for a nap. 18:59
cotto_work tcurtis: not a bad idea 19:01
dalek rrot: r48530 | chromatic++ | branches/substr_eq_at (5 files):
[ops] Renamed substr_eq_at op to cmp_str_at.
 19:07
rrot: r48531 | chromatic++ | branches/substr_eq_at (4 files):
[str] Added length to Parrot_str_compare_offset.
chromatic pmichaud, r48532 should work for you now. 19:14
pmichaud chromatic++ # will try it shortly
same branch?
chromatic Yes. 19:15
bubaflub hey GSoC-ers, pencils-down time has already passed and the final student evaluations are on Melange now 19:18
dalek p-rx: e91bc31 | pmichaud++ | :
<pre>m build/PARROT_REVISION

   <pre style='white-space:pre-wrap;width:81ex'>Bump PARROT_REVISION to be closer to 2.7.0 release.</pre>		 19:19
p-rx: 90e0bcd | pmichaud++ | :
<pre>m src/PAST/Compiler-Regex.pir

   <pre style='white-space:pre-wrap;width:81ex'>Refactor case-insensitive literals in preparation for other improvements.</pre>
p-rx: 6ab1f66 | pmichaud++ | :
<pre>m src/PAST/Compiler-Regex.pir

   <pre style='white-space:pre-wrap;width:81ex'>Optimize single-character constant literal tests to use 'ord' instead of 'substr'.</pre>
p-rx: 2a8a5a9 | pmichaud++ | :
<pre>m src/stage0/HLL-s0.pir

m src/stage0/P6Regex-s0.pir m src/stage0/Regex-s0.pir
  </pre>
   <pre style='white-space:pre-wrap;width:81ex'>Update bootstrap.</pre>
rrot: r48532 | chromatic++ | branches/substr_eq_at/src (2 files):
[str] Fixed cmp_str_at opcode.

encodings as well as unsanitized length parameters, but all of Patrick's tests pass now.		 19:24
cotto_work a cookie to anyone who can get the trac github plugin working 19:39
dalek rrot: r48533 | chromatic++ | branches/substr_eq_at (2 files):
[str] Sanitized Parrot_str_compare_offset.

Additional tests welcome.		 19:41
19:43 kurahaupo joined
pmichaud chromatic: I added some more (failing) tests for cmp_str_at -- there are some problems when comparing ascii vs. unicode operands 19:46
oh, I'll have to merge with r48533.... just a sec 19:47
chromatic That means string iteration, but I can write that loop. It's easy.
pmichaud committed, r48534 19:52
dalek rrot: r48534 | pmichaud++ | branches/substr_eq_at/t/op/substr_cmp.t:
[str]: Added some (failing) tests for experimental cmp_str_at with mixed unicode+ascii operands.
 19:58
chromatic Fix coming in 30 seconds, pmichaud. 20:04
pmichaud chromatic: okay! 20:05
chromatic r48535 should do it. 20:06
dalek rrot: r48535 | chromatic++ | branches/substr_eq_at (2 files):
[str] Enabled Unicode in Parrot_str_compare_offset.

Ideally the STRING parameters can go const again with some cleverness.		 20:14
20:22 kurahaupo joined 20:23 hercynium joined 20:25 davidfetter joined 20:28 kurahaupo joined
pmichaud chromatic: with cmp_str_at I seem to get no significant speedup or slowdown in building Rakudo's core.pir (which is the biggest/meanest test I can throw at it for now). I'm not sure why. 20:53
I think I will need to retrace my steps and make sure I'm measuring what I think I'm measuring.
I can tell that Rakudo is definitely using cmp_str_at.... just don't see any significant speedup as a result. 20:54
chromatic That's strange, unless the memory cost of mixed encoding iteration is significant.
pmichaud There might be fewer gcables created.... but I'm not suite sure how to measure that.
moritz do you know how of it is called?
pmichaud moritz: cmp_str_at gets called a lot -- basically anytime rakudo wants to match a keyword longer than one character 20:55
s/rakudo/the parser/
20:55 whiteknight joined
moritz which is... for all operators and terms, right? 20:55
pmichaud and statements
purl statements is not recommended.
cotto_work forget statements
purl cotto_work: I forgot statements
pmichaud purl, forget statements
purl pmichaud, I didn't have anything matching statements
chromatic How difficult is it to have PCT emit matching encodings for the substrings to compare? 20:56
pmichaud at the time the parser is compiled, it doesn't know the encoding of the source
chromatic Perl 6 programs tend to use fixed-8 though, right? 20:57
pmichaud iso_8859_1, generally
because of Ā« and Ā»
and any codepoint outside of latin-1 means that the source is utf8 (more) 20:58
chromatic As a quick test, can you emit those substrings as iso-8859-1 encodings?
pmichaud I might be able to do it 20:59
oh, that won't help
because core.pm has Ā« and Ā» in it, it's definitely going to be parsed as utf8
oh, wait, or latin-1 21:00
jnthn Do we actually have it as utf8 while parsing?
pmichaud just a sec, let me see if it falls into latin 1
jnthn Do we have a fixed width encoding that we can just always explode stuff into?
pmichaud jnthn: PCT tries to recode source as ascii or iso-8859-1 if it can. If it cannot, it's pretty much stuck with utf8.
jnthn Oh. :-(
pmichaud jnthn: no fixed width encoding is reliable enough yet, and it requires ICU.
let me see if core.pm at least ends up as iso-8859-1. 21:01
jnthn Ugh, OK
nopaste "chromatic" at 192.168.1.3 pasted "pmichaud, jnthn: this isomorphism appears to work" (14 lines) at nopaste.snit.ch/22835 21:05
jnthn That looks...scary.
pmichaud surely you mean iso88591 there instead of utf8 ?
ummm, wait 21:06
under what circumstances would we get into the ascii compare function if the lhs is utf8 encoded?
I could see using memcmp if lhs is ascii and rhs is iso88591 21:07
chromatic Yeah, I have lhs and rhs backwards there.
|| rhs->encoding == Parrot_fixed_8_encoding_ptr) { 21:09
That's much saner.
pmichaud agreed, I feel better about that one.
it wouldn't be correct if there are two charsets using fixed_8, though.
chromatic This is me not pleased with the Cartesian product of our encodings and character sets. 21:10
pmichaud seems like it'd be better to check the rhs for iso-8859-1 charset 21:11
instead of checking the encodings
(or perhaps check both)
there are (were) a few other places in the strings code that tended to take a shortcut like that 21:12
okay, I converted the parser to use iso-8859-1:"const string" instead of just "const string" 21:13
let's see if it gives any improvement
chromatic Checking the rhs charset for iso-8859-1 works too.
pmichaud nope, still no significant change. 21:15
less than 0.5 second out of 160
that seems very strange. 21:16
chromatic It does. 21:17
purl stays quiet
jnthn purl: you should do that much more often. :-)
purl jnthn: excuse me?
21:23 tadzik joined
chromatic If I had Callgrind output from before and after the changes, I'd have a better sense of what's happening. 21:23
A sandwich wouldn't hurt either though.
pmichaud yeah, I was thinking something similar.
I'll push my changes into branches and then we can perhaps track things down from there.
21:26 Paul_the_Greek joined
pmichaud tries a basic test with rakudo, has a serious "wtf?" moment 21:31
Paul_the_Greek What's up? 21:32
21:32 bubaflub joined
pmichaud okay 21:32
21:32 lucian joined
pmichaud found it -- forgot an option. 21:33
I'm convinced that when Rakudo is parsing core.pm to generate core.pir, it's using iso-8859-1 as the source encoding.
chromatic darbelo should put his pencil down soon, so perhaps he can look in on this with us.
pmichaud well, I'm going to need a break shortly -- and I'd like to get the Nil fixes into rakudo before the release 21:34
also, looking at this code convinces me that I can improve nqp-rx a fair bit even without the opcode, so I'd like to get those improvements in before the parrot release
chromatic If your code is available in branches, we can play with that.
pmichaud so I may move to that for the rest of the evening and pick up with cmp_str_at again tomorrow
chromatic Fine by me. What's "a fair bit" in your estimation? 21:35
pmichaud don't know.... but I can avoid quite a lot of unnecessary method calls
I can optimize the debugging calls out of the hot codepaths
the nqp-rx version that makes use of the cmp_str_at opcode is the 'cmp_str_at' branch of the nqp-rx repo 21:36
you have to build the parrot branch separately, then configure nqp-rx with --parrot-config=... to get it to use it
oh
I can also push my changes to the parrot branch 21:37
just a sec
that'd be easier
r48536 21:39
just use that version of Parrot to build Rakudo
(or any other code that you wish to test)
chromatic Thanks. 21:41
pmichaud afk for a short while
21:44 Administrator_ joined 21:51 Paul_the_Greek joined 21:52 seatek joined 21:54 davidfetter joined
dalek rrot: r48536 | pmichaud++ | branches/substr_eq_at/ext/nqp-rx/src/stage0 (4 files):
[nqp-rx]: Push nqp-rx that uses cmp_str_at for substring comparisons.
 21:54
22:08 darbelo joined 22:18 tcurtis joined
Paul_the_Greek Here's a question: When I get back home after vacation, what's the best way to update my local copy? 22:30
cotto_work Paul_the_Greek: did you see that I fixed the conflicts and applied your patch?>
mikehh Paul_the_Greek: as in svn update 22:31
Paul_the_Greek cotto! You made my vacation.I was dreading sorting it all out when I returned. I thanked you in the ticket.
cotto_work What else do you have going, and are those things in separate checkouts? 22:32
Paul_the_Greek I have two further tickets with patches pending. Would you like the links?
They are much simpler. 22:33
cotto_work I can find them.
Are they all you have?
Paul_the_Greek New math opcodes and cleanup of a lexical function.
cotto_work The reason I ask is because it might be simpler to clobber all local changes once those patches are committed. 22:34
(mainly to avoid dealing with conflicts from the pobj.h changes)
Paul_the_Greek I have other local changes in the works. Those would be clobbered, too. Easiest to save the files off to the side and then merge in my changes?
For example, I have some new tests. 22:35
cotto_work you can revert individual files too
Don't spend your vacation worrying about it. We'll get you sorted out if you want help. 22:36
Paul_the_Greek Oh, I'm a happy camper. You dealt with the one I was dreading.
++Cotto++
cotto_work I'm glad to have it in the past.
Paul_the_Greek From now on, if I comment the crap out of a file, I'll ask for it to be committed immediately. 22:37
cotto_work and I'll be less of a slacker
Paul_the_Greek In the process, I think I found a way to speed up PMC attribute allocation. I'll take a look when I return.
Hey, you can't look at everything immediately. 22:38
22:39 ash_ joined
Paul_the_Greek My first task will be to figure out why PMC attributes are allocated separately from other fixed-size objects. I think it's just an efficiency thing. 22:40
cotto_work whiteknight, bacek or chromatic might be able to answer that question 22:41
Paul_the_Greek chromatic thought it was efficiency, too.
There is an odd bit of hacking to select the PMC attribute vector index that I think we can eliminate. 22:42
22:44 allison joined 22:45 kurahaupo joined 22:47 lucian joined 22:49 dngor_ joined
Paul_the_Greek After a tough day, time for ice cream. Thanks again, cotto. 22:49
cotto_work thanks for the documentation 22:50
23:00 workbench joined 23:03 kid51 joined
kid51 msg darbelo Thanks for noting my concerns about our inability to compile Rakudo on smaller machines. 23:04
purl Message for darbelo stored.
kid51 NotFound++ for providing docs for two .pmc files. 23:05
mikehh yes excellent! 23:08
purl Smithers, release the hounds!
23:09 ruoso joined
mikehh All tests PASS (pre/post-config, make corevm/make coretest, test, fulltest) at r48536 - Ubuntu 10.04 amd64 (gcc with --optimize) 23:10
kid51 mikehh: Now if we can get all the *other* folks to do their part on the .pmc documentation :-) 23:11
mikehh kid51: :-} - doin' fulltest runs gettin' ready for the release, then I will think of our next major documentation project... 23:13
kid51: have you done a PPC fulltest recently? 23:15
In fact I would appreciate fulkltest results from anyone here, especially on some different platforms, maybe some MSWin ones 23:17
23:18 aloha joined
kid51 mikehh: Across five timezones, you read my mind. It's in progress -- hopefully done within an hour. 23:22
Anyone available to glance at trac.parrot.org/parrot/changeset/48492: Paul_the_Greek's documentation additions? 23:23
cotto applied patch but requested additional eyes; looks fine to me; I'd like to close ticket.
23:24 dngor_ joined
cotto_work I'd like eyes familiar with the GC code who can verify that the comments are accurate. 23:25
kid51 cotto_work: Perhaps you should CC: one of our many GC experts on that ticket! ;-) 23:28
perhaps whiteknight? 23:29
cotto_work I'll see if I can find someone to bug about it.
seen whiteknight 23:30
purl whiteknight was last seen on #parrot 4 hours, 49 minutes and 4 seconds ago, saying: you know, github does have service hooks for IRC. We don't need dalek, we could ask github to automatically report commits to here
cotto_work whiteknight, ping
mikehh looks ok to me, but you probably need a perusal by chromatic, bacek or our esteemed chief architect
23:34 Paul_the_Greek joined
kid51 msg whiteknight I have cc-ed myself on TT #1744, but don't have any particular insight. Will try to look at it more closely later in week. 23:34
purl Message for whiteknight stored.
Paul_the_Greek What is 2.7 named? 23:35
cotto_work We'll find out soon. 23:36
Paul_the_Greek How are they named?
23:36 bacek joined
cotto_work It's determined entirely by the release manager. 23:36
Paul_the_Greek Aha. That's fun. 23:37
23:37 aloha joined
kid51 Paul_the_Greek: All related to parrots, the birds. If you have nothing better to do on your vacation, you can assemble a comprehensive list of parrot species/breeds for future releases. :-) 23:38
Paul_the_Greek How long have you worked on Parrot, cotto?
cotto_work something liek 2.5 years
*like
mikehh I have been tending towards Australian King. a parrot that resides where our chief coding robot resides 23:39
Paul_the_Greek Have we had the Macaw release yet?
cotto_work docs/parrothist.pod lists past releases 23:40
Paul_the_Greek Does anyone do Parrot as their $work?
bacek appears with flames and smokes 23:41
kid51 No, but my hunch is that there are those among us who manage to do Parrot *during* $work ...
kid51 is not among them :-(
Paul_the_Greek "Release formerly known as Prince" ? 23:42
bacek cotto, r48492 looks good. You can close ticket :)
cotto_work bacek++
Paul_the_Greek I work for myself, so $work blurs with @play. 23:43
cotto_work done
mikehh bacek: do you have a parrot you want to honour?
bacek mikehh, nope.
Paul_the_Greek A Parrot Looks at 40? 23:44
cotto_work PROTIP: it doesn't have to be a real parrot
bacek breakfast and dayjob time.
See ya
Paul_the_Greek Where is bacek? 23:45
purl bacek is THE MANIAC or some sort of magical coding robot or probably not mailto:pmichaud@pobox.com or mailto:thebacek@gmail.com or blazing new trails
cotto_work sydney iirc 23:46
dalek TT #1605 closed by cotto++: Documentation for buffers in include/parrot/pobj.h is outdated.
TT #1605: trac.parrot.org/parrot/ticket/1605
Paul_the_Greek Man, the idea of distributed participation is no joke.
mikehh dat is why we is gonna move to git probably :-} 23:47
Paul_the_Greek Wait, he closed my ticket before I was done. 23:48
cotto_work That's not one of the many reasons I've heard so far.
Paul_the_Greek: tickets are a renewable resource. You can open a new one (or reopen the old one).
Paul_the_Greek I read the Git book over the past few days. It looks like it will be pretty cool. 23:49
mikehh Paul_the_Greek: Documentation is *never* done, it is a WIP
Paul_the_Greek I'll reopen it when I return with my brand spankin' new Memory Internals doc.
There do seem to be no lack o' tockets. 23:50
cotto_work And you can get karma for working on any one of them.
mikehh well for opening and closing them :-} 23:51
cotto_work and if someone say Paul_the_Greek++
Paul_the_Greek I'm really a ++x kind of guy, rather than an x++ guy. 23:52
I like to be fetched after I'm incremented.
cotto_work purl isn't
Paul_the_Greek purl++
cotto_work purl-- 23:53
purl cotto_work: what?
cotto_work (she cheats)
plobsing dec purl
Paul_the_Greek I wonder what it says about a person whether they use ++x or x++ ?
purl -= 1 23:54
purl Paul_the_Greek: sorry...
sorear "the" Git book?
la la purl--
Paul_the_Greek Yes, I got a book on Git and read it. Reasonably clear about most things.
Very clever with the 160-bit hash.

Powered by Cro and the RakuĀ® Programming Language.

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

Thank you!