pugscode.org/ | nopaste: sial.org/pbot/perl6 | pugs: [~] <m oo se> (or rakudo:, kp6:, elf: etc.) (or perl6: for all) | irclog: irc.pugscode.org/
Set by Tene on 29 July 2008.
ruoso Hello! 00:00
00:03 Ontolog left 00:09 justatheory left
Tene Hi! 00:30
00:33 Limbic_Region left 00:43 charsbar_ left, charsbar joined 00:44 charsbar left 00:49 charsbar joined 00:53 meppuru joined 00:58 meppl left 01:08 Alias_ joined
meppuru good night 01:29
01:30 meppuru left
pugs_svn r22268 | ruoso++ | [smop] starting to apply "every value is a list of itself" on array 01:34
01:37 Exodist left 01:47 justatheory joined 02:00 jhorwitz left 02:03 wknight8111 left 02:12 apeiron_ joined 02:20 apeiron left 02:36 cxreg joined
cxreg particle @ spug wants to know who holds the copyright to the test suite. anyone know? 02:36
obra cxreg: that's a difficult and loaded question. 02:43
02:43 apeiron joined 02:50 apeiron_ left 02:51 justatheory left 02:52 sail0r joined 02:53 sail0r left
cxreg obra: that's what we figured :-) 02:54
03:02 elmex_ joined 03:18 elmex left, elmex_ is now known as elmex 03:32 Alias_ left 03:40 z80ASM joined 03:44 Ontolog joined
z80ASM Larry, how's it going? 03:51
03:59 xinming joined, justatheory joined 04:09 alester joined 04:12 sunnavy joined 04:15 ashizawa joined 04:23 BinGOs left 04:24 sunnavy left 04:25 sunnavy joined 04:26 alester left 04:29 BinGOs joined 04:53 ashizawa left 04:54 justatheory left 05:07 bennymack joined 05:17 ashizawa joined 05:20 qwr joined 05:28 xinming left 05:58 Psyche^ joined 06:00 Patterner left, Psyche^ is now known as Patterner 06:09 xinming joined 06:24 xuser left
Ontolog everything under the 'TODO Functions' section of src/gen_builtins.pir is really unimplemented? 06:35
I thought chomp already was implemented
i would like to implement something here, something simple 06:39
any suggestions would be nice
06:48 jferrero joined
Ontolog c'mon nobody wants to hand-hold me through the code?! 06:51
maybe if someone could just give me an explanation of all the things in sub 'infix:=' that would help
i'm not sure which words are PIR native functions and which are defined elsewhere 06:52
for example 'assign' and 'getprop'
moritz_ Ontolog: your chances are better in #parrot (on irc.perl.org) 06:55
Ontolog moritz_: ok i found the answers to those parrot questions
moritz_: about the TODO stuff in src/gen_builtins.pir
moritz_: can I just go about trying to implement some of that stuff? is there some basic stuff I need to know? 06:56
I see all the methods are in the Perl6Object namespace and have some name like .sub 'infix:=' :method
moritz_ Ontolog: gen_builtins.pir is generated from classes/*.pir and builtins/*.pir 06:57
Ontolog: so don't modify that file itself, but rather the files from which it was created
06:57 Alias_ joined
Ontolog haha that's good to know :p 06:57
moritz_ all files starting with gen_ are generated 06:58
(I fell into that trap first as well)
Ontolog moritz_: could you just assign me some basic task that you know shouldn't be too complex?
moritz_: I would just like to jump in and contribute some code
moritz_ Ontolog: in src/builtins/any-str.pir there's the 'comb' method 07:00
Ontolog: it should accept an optional second argument that limits the number of matches
(that's documented in S29)
Ontolog: currently it assumes "as many matches as possible"
Ontolog cool i'll see what i can do 07:01
07:02 mberends joined 07:03 xuser joined 07:08 mberends left, mberends joined 07:11 iblechbot joined 07:19 zamolxes left
Ontolog moritz_: according to the doc it seems that only in the case that capture groups exist in the regex supplied to comb() does there need to be Match objects returns 07:31
If there are captures in the pattern, a list of Match objects (one per match) is returned instead of strings.
moritz_ Ontolog: since we can't return match objects anyway (see RT #55962) we don't yet bother about this detail 07:33
Ontolog: if we *can* return them, the current logic is flawed anyway because truncating s leads to wrong .from and .to values on the match object
Ontolog hmmm... ok so maybe working on this is not so easy after all... anything simpler without blocking bugs? 07:34
moritz_ well, I don't see any blocker for adding a repetition limit 07:35
it should be as simple as adding a second, optional parameter and a counter, and jumping to 'done' when the counter is exceeded 07:36
Ontolog oh i'm sorry i misunderstood what you wanted 07:45
so the $n in the docs
basically implement that?
moritz_ yes 07:46
Ontolog is there a rule about not modifying parameters? or can I just n -= 1 in that loop?
moritz_ on the perl level you are not allowed to modify parameters (unless they are claimed to be 'is rw' in the spec) 07:47
but if you use .param int count in PIR I think that's a copy anyway
Ontolog yes the latter is what i meant 07:49
i'll test it out and see what happens
07:51 zamolxes joined
Ontolog moritz_: ok I got it to work however now I need to make the second argument optional 07:56
moritz_: any idea on that?
moritz_ Ontolog: either add a second multi, or use an :optional argument (preferred) 07:57
Ontolog: see .sub 'index' in the same file for an example
08:00 cotto_w3rk left
Ontolog moritz_: ok got that to work, however i need to know what value optional arguments get set to if they are not supplied 08:01
moritz_ Ontolog: if they are not supplied, the corresponding :opt_flag param is false 08:02
(out of band signalling)++
Ontolog sorry i don't understand :p
what opt_flag param? 08:03
moritz_ you can have something like this:
.param in count :optional
.param int has_count :opt_flag
now if that optional parameter was provided, has_count will be true (aka 1)
08:04 mberends left
Ontolog so how is it associated, by its position in the sub? 08:04
moritz_ yes
it should go immediately after the :optional param
Ontolog i see, so we don't rely on some kind of undef or null value
moritz_ no 08:05
a (parrot) int can't be undef, afaict
Ontolog i see, so in fact we must have this extra flag
moritz_ yes
what you could do is to set count to -1 if it's not provided 08:06
and jump to the end if count == 0
and decrement after each match
08:10 iblechbot left, lolo92 left 08:11 pmurias joined
pmurias ruoso: hi 08:11
Ontolog moritz_: I finished it, so now what do I do? 08:20
moritz_ Ontolog: run 'make fulltest' first 08:21
Ontolog ok
pmurias ruoso: i changed my mind a bit about the context stuff, as it's possible to do stuff like ~(+f()) we need at least a big chunk of the lazy context propagation in addition to the forward context propagation
moritz_ Ontolog: if that's successfull, do a 'svn diff > comb_limit.patch' and send the patch to [email@hidden.address]
Ontolog moritz_: i also wrote a test 08:22
moritz_: just a plain perl 6 script
moritz_ Ontolog: there are already some tests in t/spec/S29-str/comb.t
Ontolog i see
moritz_ Ontolog: it would be great if you could integrate your tests there
Ontolog will do 08:23
moritz_ Ontolog: these tests live in the pugs repository
Ontolog ohhh right
moritz_ Ontolog: do you have a commit bit for that?
Ontolog i heard about that
no... at least not that i remember
let me check my email
moritz_ then give me your email address, and I can get you want
(use a private /msg if you have privacy concerns) 08:24
Ontolog moritz_: I have applied for commit bits it appears in November 2006 :p but i never did anything with it 08:26
moritz_: username is cdavaz, not sure what I set the password to
moritz_: email is [email@hidden.address] 08:27
moritz_ Ontolog: I'm sending the mail again 08:28
Ontolog cool... that test is still running :p
moritz_ heh, you have a commit bit longer than I do ;) 08:29
Ontolog yeah i remember at that time someone encouraged me to contrib to pugs
but i didn't know any haskell really
moritz_ neither did I ;)
Ontolog but now i want to make contribs to rakudo
ahh did you learn Haskell for Pugs? 08:30
moritz_ I tried, and failed
Ontolog haha I did spend some hours learning Haskell at some point but just some basic stuff... completely not intuitive that language :p
seems more geared towards mathematicians
ok the test passed i'll send the patch 08:31
moritz_ you have to re-wire your brain to learn haskell
great
Ontolog ok there is already a test in there like this: is "a bc d".comb(:limit(2)), <a bc>, 'default matcher with supplied limit'; 08:38
I see the 'named parameter' syntax there :limit(2)
however in my patch i didn't do anything about named parameters
should I go back and revise the patch?
moritz_ no, we generally don't handle many named arguments yet :( 08:40
parrot doesn't support arguments that can be both named an positional
at least not in the way we need it
that should be fixed at some point
before that there's no use in trying to support them 08:41
Ontolog i see... hmm so i'm not sure exactly how to add my tests
there are these formatted comments
that seem to tell the test harness to skip certain tests for either pugs or rakudo
moritz_ yes
Ontolog so if we want to skip the test for pugs but execute it on rakudo
what should that comment look like?
moritz_ well, first add the tests 08:42
when they pass, no further comments needed
Ontolog but they will fail on pugs
moritz_ then add a '#?pugs $count skip "todo: Str.comb"' line before them
Ontolog ok 08:43
moritz_ Ontolog: applied patch, thanks for the great work 08:47
rakudo_svn r31204 | moritz++ | [rakudo] implement 2-param form of Str.comb, Ontolog++
r31204 | moritz++ | Patch courtesy by Chris Davaz <cdavaz at gmail dot com>
moritz_ Ontolog: (I changed count -= 1 to 'dec count', but that's only a small stylistic fix)
Ontolog cool 08:49
if I just want to run an individual test from the pugs test suite is there any easy way to do that? 08:52
moritz_ yes, make t/spec/S29-str/comb.t 08:53
or 'echo S29-str/comb.t > t/localtest.data' and then 'make localtest' or 'make localtest_loud'
08:54 barney joined 08:57 masak joined
pugs_svn r22269 | cdavaz++ | Added tests to t/spec/S29-str/comb.t for second parameter of comb 09:04
pmurias moritz_: does dec count have any benefits compared to count -= 1 09:06
?
moritz_ and Ontolog++ again
pmurias: I'm not entirely sure, which is why I called it a "stylistic" fix
pmurias: it's just a bit more ideomatic
Ontolog moritz_: any other easy ones? 09:10
09:11 iblechbot joined
moritz_ Ontolog: there's Str.split(/regex/), but I don't think it's as easy as the previous one 09:12
Ontolog: but it would be *really* helpful to real world development
Ontolog the split method is not implemented at all? 09:13
or just some feature thereof?
moritz_ just Str.split(Str) is impelemented
Ontolog i see
which file? 09:14
classes/Str.pir it seems
moritz_ aye
Ontolog i'll see what i can do
moritz_: so this requires making another sub with the heading.sub 'split' :method :multi('Regex') ? 09:19
moritz_ Ontolog: yes, but I don't think that 'Regex' is the right type
Ontolog: parrot internally stores them as Code or Block objects or so
Ontolog I see, I'll try to find an example of another method that takes a Perl regex as an argument 09:20
moritz_ rakudo: say /a/.WHAT
p6eval rakudo 31204: OUTPUT[Blockā¤] 09:21
Ontolog I see, so it is a "Block" and this is some type that is only known to Parrot, not a Perl type
moritz_ there's another difficulty - there are some type maps between Perl and Parrot land 09:22
literal moritz_: I'd like to help with Rakudo, where could I start?
moritz_ so you can try with 'Block', but I'm not sure if it will work
maybe groups.google.com/group/november-wi...147adaad2f is of interest
lambdabot Title: .split(//) - november-wiki | Google Groups, tinyurl.com/5cqbj8
moritz_ literal: what's your skill set?
literal For one, I don't know C. 09:23
moritz_ no problem
literal I know Perl 5 pretty well and I've been reading the synopses now and then for a while 09:24
moritz_ literal: common helpful task for beginners are 1) writing tests 2) infrastructure 3) writing builtins (requires a bit of PIR knowledge, but not that hard to learn)
literal could you point me to some aspect of Rakudo that's easy to get into?
hm, writing tests is no. 1? aren't there thousands of tests already which Rakudo doesn't pass yet? :P 09:25
moritz_ literal: yes, and tens of thousands to be written
literal I see 09:26
is there some 'chart' or something that shows which parts of the standard need tests?
09:26 baest joined
moritz_ one easy starter is to go through the bug tracker. Many tickets contain small Perl 6 samples that could be turned into test 09:26
literal ok 09:27
moritz_ literal: if you look at the synopsis under perlcabal.org/syn/ you'll see these small links to tests spread all over the specs
lambdabot Title: Official Perl 6 Documentation
moritz_ literal: when there's a gap somewhere with no links for a long time, there's likely a test deficit 09:28
literal I see
and where's the bug tracker?
moritz_ literal: also t/TASKS in the pugs repo contains a list of things that need to be done
rt.perl.org/rt3/
lambdabot Title: Login
literal hm, is a PAUSE id not sufficient for logging in? 09:29
moritz_ only for rt.cpan.org 09:30
literal ok
moritz_ either get a bitcard account, or use the guest interface
09:33 kanru left 09:35 langri joined 09:36 kanru joined
masak moritz_: ping 09:41
langri hi 09:42
masak hi langri
09:44 langri left
moritz_ masak: pong 09:44
masak moritz_: do you have an opinion on rt.perl.org/rt3/Ticket/Display.html?id=58826 ?
lambdabot Title: #58826: Non-rw attributes are not changeable from within a class in Rakudo
masak I feel I have a point, but I don't want to be battling windmills either 09:45
it there's a point to non-rw attrs being writable by _nobody_, I just don't see it
moritz_ wait a sec (distraction in #perlde...) 09:47
masak sure thing.
moritz_ from an implementation POV it makes sense to forbid $.foo = 3 if $.foo is not rw 09:48
09:48 smg left
moritz_ because $.stuff is really just $(self.stuff) 09:49
masak so how'd you set $.foo ?
moritz_ $!foo = stuff
masak :(
can't say I like it
moritz_ me neither 09:50
masak but I also see that my view of things is as arbritary as the one currently implemented, if not more so
and I don't have much in the way of defense of my view, except a strong personal preference...
moritz_ there's a point in using the private accessors inside the class exclusively 09:51
remeber, the attributed is called $!foo, the $.foo is only an accessor
masak yes.
it will also confuse the heck out of beginners
ruoso pmurias, cool :)
moritz_ so if you change your mind about which attributes should be public
you have to change code if you don't use the private accessors
it's ugly, but still loads better than $self->{foo} 09:52
otoh the first example in perlcabal.org/syn/S12.html#Submethods suggests that $.stuff is rw from inside the class 09:53
lambdabot Title: S12
masak moritz_: only under the current regime. were writing to non-rw attrs allowed within the class, there would be no problem in the first place
I sure hope they are.
09:53 hanekomu joined
pmurias ruoso: isn't the lazy context propagation the same thing as coersions? 09:53
ruoso it can be seen as the same thing, yes 09:54
masak it's like this: we have to litter the attr declarations with "is rw" specifiers, and it feels _pointless_.
09:55 xiaoyafeng left
masak if those were needed when we actually needed to set attrs from the outside, it would be bearable 09:55
but now they are needed as soon as we plan to give an attr a value, which is, um, all the time
hence all attrs get "is rw"
very non-Perlish
pmurias masak: why are you opposed to using $!foo? 09:57
moritz_ masak: that's mostly because .new() doesn't set attributes, right?
masak pmurias: because I _declared_ $.foo
moritz_: I guess
moritz_: so they're writable through constructors?
why?
literal man, I hope Perl 8 won't have trigils
masak literal: :) 09:58
moritz_ masak: because they need to be
smtms the armaggeddon is planned just after Perl 7 is relesed
masak trigils would be for AOP, I guess
moritz_ masak: otherwise $obj.perl simply can't work in any meaningful way
ruoso breakfast &
masak I see how my problems would be solved if I'd just bite the bullet and use $!foo in my methods. it's just that... it's one more thing to keep in mind, when $.foo could easily have worked 10:00
pmurias masak: consider the case when i change has $.foo into has $!foo;method foo {$!foo*2}
masak hm 10:01
10:01 baest left, baest joined
masak pmurias: at least then I wouldn't expect setting $.foo to work, no 10:02
pmurias fixing a (nokia pc suite)-- automatisation app& 10:05
masak it's a question of DWIM, I think. if I write `$.foo = 5;`, chances are that I really mean that, regardless of whether I wrote `has $.foo` or `has $.foo is rw` before. 10:06
moritz_ masak: it's the old question: do we/you want to sacrifice consitency for convenience?
masak: I suggest you take it to p6l 10:07
masak yes.
just thought I'd start here :)
now I know there's a chance I'll be shot down, because I represent convenience, not consistency :)
I didn't know that before
moritz_ masak: the nice thing about being shot down from p6l is that you recover rather quickly ;) 10:09
yes, starting here is good but in this case I think we'll benefit from the broader audience 10:10
masak writing mail now
10:12 ashizawa left 10:17 Alias__ joined 10:18 Alias_ left
pmurias masak: i can apply the same thought process to $.foo = 5 from outside the class 10:19
masak pmurias: you mean, you could argue for the opinion that it should be writable from the outside as well? please do. 10:20
I don't see it.
I think it's different.
pmurias if i want to write to a ro attribute i want to do it so perl 6 should DWIM and do it 10:21
masak ah, but then you're crossing a class boundary 10:24
not your call anymore, the class gets to decide
that's the key -- when I'm writing stuff _within_ the class, I'm (by definition) the author of the class
it's my call when to write to accessors
a call from the outside does not have the same luxury 10:25
it's some sort of principle in OO
"encapsulation", perhaps? or "information hiding"? I dunno.
lunch &
moritz_ rakudo: class A { has $.b }; my A $x .= new(b => 3); say $x.b 10:29
p6eval rakudo 31204: OUTPUT[3ā¤]
10:38 Alias__ left
pmurias masak: has $.foo;method bar {$.foo = 4} does violate some sort of "information hiding" too, as you assume that $.foo is defined in your class (it could have mean inherited etc.) is a rw one underneath 10:38
and the benefit of this is that you will be able to use the wrong twigil and Perl 6 will just silently assume you wanted to use the right one 10:39
the "authorship" of a given component in information hiding is not really relevant 10:41
10:41 ruoso left 10:43 zamolxes left
masak pmurias: you have a point. 10:51
11:06 masak left 11:15 Ontolog left 11:19 smg joined
pugs_svn r22270 | pmurias++ | [smop] IO RI converted to the RI DSL 11:27
11:36 Ontolog joined 11:41 BinGOs left, BinGOs joined 11:42 BinGOs left 11:47 zamolxes joined 11:53 BinGOs joined 11:57 masak joined 12:01 hanekomu left, hanekomu joined 12:02 abra joined 12:14 mncharity joined 12:16 hanekomu left, hanekomu joined 12:23 sri_work joined
mncharity moritz: masak: Thanks for the perl.net.au/wiki/Elf suggestions. It's been updated. 12:30
lambdabot mncharity: You have 2 new messages. '/msg lambdabot @messages' to read them.
Title: Elf - PerlNet
moritz_ mncharity: np, great
mncharity moritz_: re STD_blue, beyond what the page says, my hope is for the simplicity of using STD/gimme5 directly, with TimToady fixing the bugs which turn up, and one of {TT, or someone else, adding Match tagging, or being able to live without it}. 12:33
masak mncharity: glad to help
moritz_ mncharity: is there a compelling reason not to use the YAML dump from STD? 12:35
12:36 alester joined
mncharity masak: there are some low-hanging fruit if you wanted something to work on. Like adding arguments and types to the bare names in docs/p6types. rhr++ 12:36
moritz_: re YAML, dealing with yaml turned out slow and buggy. Slow as in something vaguely like a 50% or 100% hit on total parsing time. Buggy as in such amusements as a the value of a hash field mutating into its field name for particular original field values. I was surprised by how bad it all was. 12:39
My current model is that yaml doesn't really get exercised for large data sets. Common case seems to be small config files.
moritz_ mncharity: ok, interesting
mncharity: another thing... is there a single "make test" somewhere that gives a simple yes/no answer to the question "did I break something"? 12:40
mncharity: if not, I feel that's essential in our TDD Perl 6 world
pmurias mncharity: hi 12:41
masak mncharity: I'll consider it. right now, November gives all the joy I can accept and takes all the free time I have. :) 12:42
pmurias mncharity: what does the using gimme5 directly mean?
back to eating soup& 12:43
mncharity re make test, hmm, interesting. it currently comes in two parts, make rebuild, and make test. make test runs t/, so takes some time. it's also not quite turnkey, since it doesn't sync your t/ with the revision used to generate the baseline file run-tests.result.
hi pmurias
masak: re November joy, lol. understood. :) 12:44
pmurias: re gimme5 directly, -I src/perl6 use STD; 12:45
that uses gimme5, yes? 12:46
moritz_ mncharity: the fact thet it takes more than 2 lines to explain in IRC means it's too complicated right now :/
mncharity lol. good point. ok, will tool. 12:47
moritz_ mncharity: I'd imagine something like rakudo's spectest_regression, which runs those tests that are known to pass
mncharity: that also reduces run time
pmurias mncharity: it uses gimme5 indirectly as gimme5 is used to compile STD 12:48
12:49 alester left
mncharity pmurias: ah, ok. 12:50
12:51 araujo joined
pmurias re branching STD i i don't think it's even vaguely a goodish idea as TimToady does a very good job with STD, and we need to have *one* official grammar 12:53
moritz_ aye
mncharity moritz_: re just run passing tests... hmm... 12:55
moritz_ mncharity: if you want to know if additional tests pass, you can still use another test target
12:57 alester joined
mncharity possible reasons for branching STD: to fix things which seem to be bugs (like the current(?) absence of identifier info in sub calls); to replace | with || (there being no existent fast-enough implementation of | , and none written in p6); to add Match node labeling (so AST->IR doesn't have to guess what the Match are); and 12:59
to decouple TT STD.pm development from elf development (so T can proceed full speed without worrying about elf, and the 13:00
to decouple TT STD.pm development from elf development (so TT can proceed full speed without worrying about elf, and elf isn't destablized by those changes). 13:01
*Matches are
moritz_ do you think that STD_red (as an implicit branch) was successful? 13:02
13:04 meppl joined
mncharity moritz_: re tests... the thing is, at least when passing % is low, a problematic "unexpected thing" may as often(?) manifest as failing tests starting to pass. maybe. anyway, no reason it can't be a test target. 13:04
originally thought it hard, thus my hesitation, but if not, customer rules.
pmurias mncharity: TT = ?
mncharity TimToady
13:05 charsbar_ joined
mncharity moritz_: re "do you think that STD_red (as an implicit branch) was successful?", oh, definitely. Well, the big picture question is whether anything was gained by being able to write elf over the last months, rather than waiting until now. Err, no. Or rather, yes. gimme5 is still rather slow, and STD_red may have helped to push STD development along. As for the role of branching, 13:07
pmurias mncharity: you should fix the bugs directly in STD, there is no need for Match node labeling as you now what is it from what you got it from
13:08 nipotan_ joined
mncharity branching permitting fixing tens of bugs old versions of STD (without which elf couldn't work); permitted using || for | (without which elf couldn't exist (working | is recent), or run fast (still no fast | )); permitted tagging Matches with the rule and branch they came from (making the Match AST -> IR node mapping code vastly simpler and more maintainable). 13:09
13:10 charsbar left, charsbar joined, charsbar_ left
mncharity re tagging Matches, for comparison, there's the older attempt to use rakudo as front-end, which took the "guess" approach. really really messy. 13:10
13:11 charsbar left
pmurias when i was doing pixie i didn't have to guess anything 13:11
mncharity re old rakudo front-end, though it appears not easy to find... hmm... 13:13
13:13 jan_ left
mncharity pmurias: you tweaked STD.pm and applied {*} tagging? 13:13
13:14 jan_ joined
moritz_ mncharity: rakudo is rather clos to allow builtins written in Perl 6 (modules can already be compiled), and one of the next milestones is to implement LTM in PGE, then bolt STD.pm onto rakudo 13:14
mncharity: when that's actually done I see no big advantage of elf over rakudo any more 13:15
pmurias i overwrote the {*} handling in STD so i could add hooks at runtime and set the result for Matches as they were created by the grammar
moritz_: elf is perl 5 base which should allow good perl 5 integration
moritz_ pmurias: point taken 13:16
moritz_ has to leave no, bbl
mncharity svn.pugscode.org/pugs/misc/rakudo/e...o_ast.data
lambdabot tinyurl.com/5lhhe7
pmurias mncharity: and i wrote a script which allowed me to write {*} handlers *as* if i was filling the {*} blocks 13:17
mncharity re close to allow builtins written in Perl 6, nifty. though... well, my fuzzy recollection is it was also believed to be close a month+ ago. nice to know it really is now. ;)
yay on module compilation
Ontolog in which files are Regexp objects defined?
mncharity re LTM, neat. perhaps do a p6 reference implementation before getting down to PIR? 13:18
moritz_ dunno, that's pmichaud's design work, I'm mostly observing 13:19
have to really run now &
pmurias mncharity: but when TimToady removed all the rule foo {... {*} } {*} blocks pixie broke
Ontolog in src/classes I can't find either Rule.pir or Regex.pir or anything of the like 13:20
mncharity moritz_: re "when that's actually done I see no big advantage of elf over rakudo any more", possibilities include faster (or at least was, and a CL backend may provide another 10x); much more of the compiler written in p6;
13:21 z80ASM left
pmurias mncharity: rakudo could be made to emit CL 13:22
mncharity everything directly controllable (ie, nothing blocks on parrot bugs, though obviously, some new desired approach to say a p5 backend my block on p5 bugs, but that's a different, less problematic issue);
s/my/may/
Ontolog man dividing the development effort among several competing implementations suck 13:23
none of the impl's have enough developers
mncharity pmurias: re "pixie broke", :-( I was about to say, re {*}, that maybe we would have to copy pixie's approach. drat. then... well, yes, STD.pm may need massage.
we'll see. 13:24
re "rakudo could be made to emit CL", err, good luck. most of rakudo/parrot is written in C and PIR. 13:25
pmurias mncharity: "pixie broke" it should be possible to fix pixie, it was that i moved on to other things
mncharity re fix, right.
pmurias if rakudo has working multi methods it should be possible to write a CL emitter in Perl 6 13:26
mncharity I'm still hoping to avoid the entire issue. Just need to bootstrap the elf code and STD.pm itself. Then it is off the critical path. Elf is fully bootstrapped, and it becomes a "improve the compiler any way you want" issue. Question is whether it cause a maintainability meltdown nightmare. 13:27
pmurias for all of the language using PCT
Ontolog what is CL?
pmurias common lisp
Ontolog i see
pmurias * languages
entire issues = ? 13:28
Ontolog anyone on the Regex thing? I'm trying to find out how to manipulate Regex (or is it Rule?) objects within PIR
taking a look at the Regex class would be nice but I can't seem to find it :(
13:29 charsbar joined
masak Ontolog: as jonathan said, regexes are of typ Sub right now 13:29
mncharity re CL emitter, yes, that would let you use the rakudo parser and AST aka IR. But you would still need assorted elf p6 code, and assorted elf p6 code as yet unwritten.
pmurias assorted elf p6 being the runtime support 13:30
Ontolog masak: Sub.pir contains almost nothing 13:31
masak: which files defines the methods on a regex?
mncharity while "rakudo" is written in p6 and PIR, the compiler is rakudo+parrot, so being able to compile to non-parrot doesn't get you very far without a lot of parrot and pir replacement effort. which is what elf is.
masak Ontolog: which methods on a regex do you mean_
?
Ontolog masak: well I always see this match = regex.ACCEPTS(something) 13:32
masak Ontolog: I suspect that regexes are created (as subs) which then call into PGE
Ontolog: yes, regexes are basically that
Ontolog 'call into PGE' what does that mean?
masak Ontolog: PGE is a Perl 6 grammar engine
it parses Perl 6 for Rakudo when compiling
but it also handles regexes for Perl 6 programs in Rakudo 13:33
Ontolog ahhh
masak oops, sorry PGE == Parrot Grammar Engine
Ontolog ohh
how about the Match object
same deal? 13:34
pmurias mncharity: very far = bootstrap
masak probably directly created by PGE, or a wrapper of same
Ontolog so I have to look into this 'PGE' to see what a Regexp and a Match do?
pmurias mncharity: ?
Ontolog where is the pge?
pmurias compilers/pge
Ontolog ok
mncharity moritz_: re "when that's actually done I see no big advantage of elf over rakudo any more", possibilities include faster (or at least was, and a CL backend may provide another 10x); much more of the compiler written in p6 (so can use the many interesting backends out there); everything at hand (no blocking on parrot bugs). Downsides too of course, but those are potential advantages. 13:35
masak pmichaud++ # thanks for PGE! it makes Rakudo so much nicer to work with
pmurias PGE makes rakudo possible
masak that, too :)
Ontolog haha I see! there are things about Perl6 outside of the languages/perl6 dir 13:36
yes Regex and Match are there
I need to implement Str.split(/regex)
masak Ontolog: no, probably not
Ontolog: go with Str.split(Sub) as jonathan said
Ontolog yes i mean that 13:37
i mean i need to implement the version of split that takes a regex
so any hints on how to get the position information from the Match?
masak Ontolog: if you want inspiration, check out .subst that I implemented a while ago
it's in any-str, I think
Ontolog yes i am looking at that
masak Ontolog: you are? cool!
pmurias mncharity: the main issue with elf for me is that rather then doing one thing good properly, we have a complete set of not good enough components
Ontolog masak: I see you use the from and to 13:38
masak aye
mncharity pmurias: re very far, regex engine for p5 (as for elf), p6 (may get done for the CL backend; needed for others); and, well, "can now compile modules" and "soon can do builtins" means everything else, not currently written in p6, would have to be so. no?
Ontolog masak: ok I have a better idea of what to do now i'll give it a shot tomorrow, so don't go implementing it first :p wait for me to try
masak Ontolog: no promises :)
(actually, I won't have time to implement it, so you're safe) 13:39
Ontolog haha cool
mncharity pmurias: re "rather then doing one thing good properly, we have a complete set of not good enough components", good observation. the
masak Ontolog: but don't get too comfortable! I might find some free time!
Ontolog i'm trying to make some contributions and maybe earn a commit bit
masak Ontolog: sounds great! 13:40
Rakudo needs people aspiring for commit rights 13:41
Ontolog yeah i'm lucky i have some time right now, work isn't too busy ;p
mncharity it seems very unlikely that any p6 code part of any existing implementation is anything but throwaway. to be thoroughly changed as p6 gets closer. perhaps some portion of rakudo's p6 parser and actions is more stable(?). certainly I'd never write a p6 compiler with callbacks rather than multis. so
so the question is, how to get there from here. when we have no usable p6 compiler, and no prelude written in p6. 13:42
how to get to xmass.
masak mncharity: btw, I think it's spelled "X-mas"
but I'm no native, so don't take my word for it 13:43
mncharity tnx
masak also, "X-mas" might have more US/commercialism connotations than "Christmas" 13:44
again, don't take my word for it :)
mncharity how to get to X-mass. kp6 you know. pugs until resurrection is dead, and even then will be mostly hs rather than p6 - migrating the prelude from hs to p6 was a work (struggle) in progress back when pugs wound down. elf is better than redsix. and rakudo you know. 13:46
masak it's not dead, it's sleeping! 13:47
mncharity elf still seems to me likely the fastest path to a usable p6 compiler. unquestionably to a bootstrapped one.
pmurias i did a few commits on it recently! ;)
(on pugs)
masak it's pinin' for the fjords 13:48
13:48 xiaoyafeng joined
mncharity re sleeping, indeed. and I look forward to morning. but... that just gets us back to where we were in 2005? modulo a much clearer and fleshed out p6 spec. 13:48
:)
pmurias mncharity: STD.pm is not throwaway 13:49
masak mncharity: Pugs, stalled in late 2006, but point taken 13:50
s/,//
moritz_ (kinda re) 13:52
mncharity basically, elf seems something vaguely like <1 FT person-year away from pugs-like t/ passing. perhaps just a couple of months. perhaps parallelizable down to 1 calendar month. but... the concept of writing a p6 compiler from scratch, more or less by myself, has very no appeal. even though that seems to be happening slowwwly, on an incremental installment plan.
moritz_ modulo a much larger test suite, one half of which is implementation independent
mncharity hell, if a good lisp hacker and one other wizzy person turned up, I suggest a hackathon and see if we could get to pugsian t/ in a long weekend. 13:53
masak mind the brain! you don't want to do that... :P 13:54
13:54 charsbar left
masak that said, it sounds incredibly cool 13:54
13:56 charsbar joined
mncharity pmurias: re "STD.pm is not throwaway", current STD very little resembles STD of a month or few ago. while "it now works" is great, it's not clear to me that is equivalent to "it is now unlikely to change before X-mass". 13:56
masak mncharity: you're still one "s" too many on that "X-mas" word. 13:57
mncharity ah well.
masak :)
13:57 charsbar left
pmurias mncharity: there is a difference between throwaway and "will evolve" 13:57
mncharity :) have to head out.
re difference between throwaway and "will evolve", 13:58
moritz_ volatile != throw-away, yes
pmurias gimme5 is throwaway
avar also nothing uses STD.pm in a self-hosting way so it might need lots of changes for that use
pmurias it will certainly need fixing the produced Match
13:59 charsbar joined
moritz_ avar: possible, but not very likely. Rakudo's parser is modeled after earlier version of STD.pm, and didn't hit major usage obstacles 13:59
pmurias major usage obstacles != hordes of tiny bugs 14:00
moritz_ tiny bugs usually don't require large refactors 14:01
mncharity pmurias: I think of "throwaway" as "long-term maintenance is not an issue", "code exists solely to fulfill short-term goals", "ideas play as much role as source", "one can't point to a line and say with confidence 'this line will be in the release version'", "one can point to lots and lots of lines and say with confidence 'this wont be'", ... 14:02
14:04 charsbar left
abra hi all 14:05
14:06 charsbar joined
mncharity pmurias: so back to "good thing properly", one certainly could, from scratch, write "if you had full p6, this is how you would write a p6 compiler". and "given that, here's how you would write a p6/CL/whatever backend". and you would be unable to run or test it until rakudo or elf or pugs was finished. 14:06
moritz_ hi abra :)
mncharity at some point rakudo will start being able to run elf. but if one wants p6 to be bootstrapped, or non-parrot backends, that doesn't remove the need for something like elf. though obviously it would alter the details of how you went about it. 14:08
moritz_ aye
masak mncharity++
Ontolog Can I tmpstr = new 'Perl6Str' and then later tmpstr = subst self, start_pos, end_pos ? 14:09
pmurias pluggin a non-parrot backend into parrot and replacing pir with perl6 when possible could also give a bootstrapped Perl 6 compile
Ontolog subst should return a Parrot String yes?
pmurias * plugin in
Ontolog but should I only be putting a Perl6Str in there?
pmurias s/parrot/PCT
* compiler
moritz_ Ontolog: that should work because Perl6Str is a subtype of String 14:10
Ontolog: re "should I?", in first approximation everything that works is fair game ;) 14:11
Ontolog haha i just worry that we might need a special subst that returns a Perl6Str with some special stuff that a normal String doesn't have
mncharity pmurias: I'm not sure I understand the last suggestion. 14:12
masak rakudo: class A { grammar B { token greeting { HAI }; } }; say ?("OH HAI" ~~ B::greeting)
p6eval rakudo 31210: OUTPUT[1ā¤]
moritz_ Ontolog: once HLL type maps are in place that shouldn't be a problem
rakudo: class A { grammar B { token greeting { HAI }; } }; say ?("OH HAI" ~~ &B::greeting) 14:13
p6eval rakudo 31210: OUTPUT[1ā¤]
abra sorry for my english :) I try to build pugs from source, but it's failed :( Somebody can help me? rafb.net/p/G74g6D28.html
lambdabot Title: Nopaste - build pugs failed ((
masak will putting a class/grammar inside a class reduce its visibility in any way?
pmurias mncharity: what i'm suggesting is that it should be possible to morph rakudo into a Perl 6 only compiler emitting CL
14:14 z80ASM joined
moritz_ abra: how new or old is your copy of pugs, and which version of GHC6 do you use? 14:14
Ontolog how should I test if match.'next'() is successful or not? 14:15
mncharity have to go. pmurias: much appreciate the thoughts. perhaps the Elf page could use info on "possible long-term roles of elf", and "nature of the elf bootstrap". re morph,
abra moritz_, The Glorious Glasgow Haskell Compilation System, version 6.8.2
pmichaud (named parameters) -- why is it not worth supporting named parameters yet?
moritz_ Ontolog: either it returns a value which you can test for success, or it mutates match (I don't know) 14:16
Ontolog moritz_: it returns () in every case
moritz_ abra: how did you get pugs? via svn?
abra: if so, which revision do you use?
abra moritz_, via svn
pmurias mncharity: i'm willing to help with the CL backend as it would pave the way for a smop one
pmichaud: were did you get the not worth to support named parameters stuff from? 14:17
moritz_ abra: could you please nopaste the output of svn info? 14:18
mncharity awesome, help would be great. my next on_sbcl step is to get class and method emitting written. assistance then on writing Array and Hash and etc primitives, part p6 and part CL (like PrimitivesP5.pm), would be most nifty. 14:19
pmichaud pmurias: sorry, I was reading scrollback and didn't realize I was reading scrollback
just a sec
08:37 <Ontolog> ok there is already a test in there like this: is "a bc d".comb(:limit(2)), <a bc>, 'default matcher with supplied limit'; 14:20
08:38 <Ontolog> I see the 'named parameter' syntax there :limit(2)
08:38 <Ontolog> however in my patch i didn't do anything about named parameters
08:38 <Ontolog> should I go back and revise the patch?
08:40 <moritz_> no, we generally don't handle many named arguments yet :(
moritz_ pmichaud: that was meant wrt to named versions of positional arguments 14:21
pmichaud is what I was referring to.
moritz_ not named arguments in general
pmichaud right, named versions of positionals we don't support. But :limit(2) isn't a positional there :-)
ohhhh, I see, it's a positional in the prototype
never mind, thanks for clarifying
abra moritz_, paste.org.ru/?6621un 14:22
pmichaud I must've been up too late last night doing a release or something... :-|
moritz_ abra: you're using an old repository that's not updated anymore
abra: try svn co svn.pugscode.org/pugs/
lambdabot Title: Revision 22270: /
pmurias mncharity: help with the CL PCT/rakudo one :(
moritz_ abra: and then build that one instead
abra moritz_, ok, tnx 14:23
mncharity pmurias: if I may rephrase your earlier argument, once rakudo can be build using p6 (regardless of whether or not it is), and is usable for large programs, then one could write p6 bootstrap compiler parts on rakudo instead of elf, and p5/CL emitters too, rather than on elf. And you would probably want to use the rakudo IR in that case. so
oh
re :(, ah well. :/
rakudo_svn r31211 | pmichaud++ | [rakudo]: spectest-progress.csv update: 165 files, 3377 passing tests 14:26
pmurias as writting elf emitters is mostly fighting with the **** ast 14:27
mncharity: if i replace the pir emitter stuff with perl 6 why would you prefer going the elf root? 14:29
mncharity I will be most interested in watching how development of a non-parrot rakudo emitter goes. The first one ever I suspect. That rx tests didn't get written because rakudo IO wasn't working(?), makes me wonder how it will work out short-term. 14:30
moritz_ mncharity: well, I wrote them later on 14:31
pmurias mncharity: there is a partial lol code one
moritz_ mncharity: and TimToady hacked them together as inline tests later on
mncharity hmm, we should try to find someone to clean up the rx tests then?
14:31 nothingmuch left
mncharity inline tests? 14:31
pmurias mncharity: i'll take a walk, and then check if an emitter in perl6 would be possible& 14:32
moritz_ mncharity: the old ones were read from a file
14:32 pmurias left
moritz_ mncharity: the new ones have the regexes in the test file 14:32
mncharity: t/spec/S05-mass/rx.t iirc
14:32 nothingmuch joined
mncharity pmurias: re why go elf root, you might try writing a little p5 emitter in rakudo as an experiment. if it works out, awesome. if not, the "why prefer elf?" reduces to the old "wait for rakudo to work, or pursue other options" question. 14:33
lol, you had the same idea. I'm quite curious how it works out. I've no clear idea. 14:34
well, I strongly suspect how it will go. but would be most delighted to change opinion. 14:35
moritz_: re rx.t, ah, awesome. that will come in handy. 14:36
Ontolog Perl6Str is a String, but why does somePerl6Str = substr self, start_pos, end_pos create an error about no such opcode? $S0 = substr self, start_pos, end_pos works fine 14:37
moritz_ mncharity: that's what caused the big leap on the charts on rakudo.de/ :) 14:38
lambdabot Title: Rakudo - A Perl 6 Compiler
mncharity lol :)
yes, I always loved the % bump one gets from getting the rx tests working. :)
masak what happened at those two red dates? 14:39
how do we avoid it happening again? :)
avar wrt elf you might want to re-use some of my work on getting kp6 to cpan to get a wider audience:) 14:40
masak kp6 is on CPAN? where?
moritz_ masak: build failures 14:41
masak moritz_: ok
moritz_ masak: how to avoid it? dunno, perhaps an svn bot that allows commits to trunk only when at least a few tests pass?
masak so one shouldn't do what I do: submit first and run tests later :)
moritz_ masak: but I don't think it's worth the trouble; somebody broke it, and it's fixed a few hours later. If UTC midnight is inbetween, the graph shows 14:42
masak: it's not pretty, but not harmful either
masak true 14:43
just feels like the bad PR far exceeds the actual harm :)
it's realistic, yes, but that's not always a desirable thing
moritz_ masak: for PR reasons I can invent an algorithm to only plot data values that we like ;) 14:44
masak go right ahead :)
mncharity avar: re CPAN, yes, but... pick a random p6 feature, and rakudo is vastly more likely to handle it than elf. elf is "collect a set of features needed for p6 compiler development, and go on from there". so it's not really appropriate at the moment for a general p6 audience. 14:45
moritz_ next if $fail > $pass;
masak is all for it
avar masak: kp6 used to be on cpan but I deleted it from there 14:46
masak oh, you can do that?
avar I did a bunch of changes to it to make it behave like a proper CPAN package, then fglock broke cpanness again so the cpan version was out of date
masak avar: from what I see in actions.pm in Rakudo, kp6 is very usable
avar It's usable, it just doesn't work as a CPAN distro last I checked 14:47
masak has been thinking of releasing November on CPAN 14:48
but right now there is little sense in doing that, and only extra work 14:49
avar I'm not going to tell you (elf & nov authors) how to maintain your packages but I find it very useful to be able to do `make disttest' on code even if I don't plan to release it since then I know everything is working properly outside my scm working directory. 14:50
It was certainly helpful with kp6 where the build system was some elaborate mix of shellscripts 14:51
masak avar: thanks, I'll remember that
mncharity elf's only excuse for existing is demonstrably being able to write a p6 compiler in p6 (narrow dialect) _now_, rather than waiting until rakudo supports it. If one is confident that will be soon, there's still no excuse.
Or if there turns out to be no interest in writing a p6 compiler now, in which case the ability is largely uninteresting. 14:52
avar: re disttest, good idea. 14:53
oh, right, I was writing wizzier make test... distracted... maybe this evening. 14:54
14:54 Front_slash joined
mncharity really have to go. have fun all & 14:54
14:54 mncharity left, alester left 14:57 alester joined 14:58 hercynium joined 15:03 Front_slash left
Ontolog This code: $test.split(/X/); always calls .sub 'split' :method :multi('String') even though I have defined .sub 'split' :method :multi('Sub') 15:04
What is the bifff?!??!
is /X/ a String?
masak no.
15:04 Front_slash joined
Ontolog so what do you think is happening? 15:04
masak I don't know. 15:05
suggest you lean heavily on .subst, which works.
15:06 TJCRI joined
Ontolog huh? 15:07
what is the :lex you have there?
masak hm, it had something to do with $/ working correctly 15:08
15:08 justatheory joined
masak don't remember the exact details, but I could find the relevant #parrot discussion if you want 15:09
Ontolog that's ok i got it to work by copying your :multi
masak hehe 15:10
Ontolog set it to (_,'Sub')
so _ holds self?
is that it?
I am really confused about :multi
I sent an email to perl6-compiler
did you see that?
masak Ontolog: I don't know. I just work here.
Ontolog: yes, I saw
good email
many interesting questions
Ontolog lol
thanks
masak hope someone answers
Ontolog yeah me too
masak Ontolog: oh, pmichaud++ did 15:12
Ontolog yes i'm reading =)
masak re 'so _ holds self?': apparently yes :) 15:13
moritz_ that's unexpected (at least to me) 15:15
masak not altogether unreasonable, though
ever since the dawn of OO, the invocant has been considered the first argument to methods 15:16
moritz_ there's no reason to make it user visible, though
Ontolog but self is defined implicitly
so is there a need for this?
anyway there are a few places in the code that need changing 15:17
because they don't follow the rules about :multi
15:17 zamolxes left
masak Ontolog: do you have an example? 15:18
Ontolog yeah, all moritz stuff ;-)
and the patch i submitted today 15:19
masak :)
moritz_ if it were broken, it wouldn't have worked
Ontolog not exactly! maybe something will break later on because of it 15:20
moritz_ that fact that the tests pass mean that at least one of us understood something wrong
Ontolog in any case we should be sure to get it right, no?
certainly
moritz_ any<parrot moritz Ontolog pmichaud> probably ;)
15:21 alester left
Ontolog well i'm sure it's all over the place 15:21
but looking at pmichauds email
he seems to be pretty clear about how it should work
and i have come across problems because i did it wrong
anyway, i'll fix and test again
masak Ontolog: it would be interesting if you could provide a short piece of code that breaks because of what you say is wrong being wrong 15:22
Ontolog the example i just gave with my own code 15:25
This code: $test.split(/X/); always calls .sub 'split' :method :multi('String') even though I have defined .sub 'split' :method :multi('Sub') 15:26
pmichaud because :multi('String') restricts the type of the invocant
15:26 cotto_work joined
pmichaud you probably want :multi(_,'String') and :multi(_,'Sub') 15:26
Ontolog yes 15:27
i understand how it works now, i'll fix it in places where its wrong and do it right in the future
anyway time to go to bed (+8 gmt here) 15:28
see you all later
masak Ontolog: sleep well 15:30
dream sweet MMD dreams :)
moritz_ ;) 15:31
15:46 Front_slash left 15:47 Front_slash joined 15:49 xiaoyafeng left 15:52 Exodist joined 15:58 barney left 16:00 Front_slash left 16:02 nipotan_ left 16:21 masak left 16:33 clintongormley joined 16:36 hanekomu left 16:39 Exodist left 16:40 Exodist joined 16:44 jferrero left 16:50 alester joined 16:57 z80ASM left 17:10 rindolf joined 17:13 hanekomu joined 17:56 pmurias joined 17:59 Lorn left
pmurias pmichaud: is the lolcode emitter viewable somewhere? i'm looking how i could my own emitter in 18:00
18:00 Lorn joined 18:02 apeiron_ joined 18:06 zamolxes joined
pmichaud the HLLCompiler needs some refactoring before its workable, but I can post the emitter 18:07
and the point of it was to make a humorous lightning talk, not necessary as an example of how emitters should be written :-) 18:08
I do plan to add it to the parrot repo, but I need the HLLCompiler changes first 18:09
www.pmichaud.com/perl6/lolpast.pir
pmurias so a custom emitter would need a custom HLLCompiler?
dinner&
pmichaud not after the refactor
the HLLCompiler just needs a redesign -- the design that is there now isn't as flexible as the one I originally had in mind 18:10
18:11 apeiron left
moritz_ ouch, pugs' Test.pm has an is_deeply function that boils down to sub is_deeply($got, $expected, $desc) { is($got.perl, $expected.perl, $desc) } 18:13
[particle] sounds like eqv_deeply 18:15
"eqv_deeply" (a non-existant test function name)
but perhaps more descriptive. heck, i suppose "eqv" alone would do it 18:16
moritz_ actually infix:<eqv> is quite similar, yes
but if eqv internally uses .perl, it has to enforce ordering of hashes
pmurias pmichaud: i'll wait some more
moritz_ which normal .perl doesn't
now I'm wondering what to do 18:17
it seems wrong to bring that ugly hack to rakudo's Test.pm
a list_eq might be more accurate, and implementable in rakudo 18:18
eqv would really rock, but I don't see that anywhere on the horizon 18:19
otoh if .perl supported an optional argument to serialize in a canonical order... 18:20
pmurias moritz_: .perl should propably sort the keys if it is optimised for human readability rather than for speed 18:24
[particle] like .perl:serialize<lexicographic>
.perl:serialize<insertion>
pmurias [particle]: insertion is hard 18:25
18:26 ruoso joined
[particle] that's not my problem, it's up to the implementors ;) 18:27
moritz_ [particle]: then you have to become an implementor ;)
pmurias [particle]: if they implement Hash'es as linked lists it will be your problem ;) 18:28
[particle] parrot's Hash is like perl 5's, in that it stores in insertion order 18:30
moritz_ perl 5's hashes don't store in insertion order
they store in quasi-random order 18:31
18:34 apeiron joined
ruoso Hello! 18:36
moritz_ hi ruoso;) 18:37
ruoso :)
18:38 iblechbot left 18:41 apeiron_ left 18:55 omichael joined
pmurias ruoso: hi 18:56
ruoso hi pmurias
pmurias shower& 19:03
ruoso: any ideas for stuff to put in the ROADMAP? 19:04
ruoso Well.. the current roadmap is valid 19:06
I'd like to get rid of the leaks
but I'm taking a break from that, and got back to implementing map
pmurias the leaks are a bit annoying, a whole truck-loads of garbage to follow 1..1\nok\n 19:07
* truck-load
i read and thought about tracing gc a bit, we would have to write a stop-the-world one as concurrent ones are very complicated 19:10
19:11 jferrero joined
[particle] if your gc is pluggable, then that's probably ok 19:11
especially if it's switchable during runtime
moritz_ you could always emit code for another vm that already does GC 19:13
or use a C level garbage collector - no idea how good or bad they are 19:14
[particle] llvm?
moritz_ [particle]: llvm doesn't do garbage collection
[particle] orly? i thought they had a lib for that
moritz_ maybe, I just know it's not a "core" feature 19:15
speaking of llvm - I mailed the llvm dev list about #line support in clang
they said it's being worked on 19:16
[particle] great
icc also has funky #line handling, but only with continued lines and macros iirc
foo(SOME_MACRO(bar), \
baz); 19:17
moritz_ llvm.org/bugs/show_bug.cgi?id=2788
lambdabot Title: Bug 2788 ā€“ Support #line and # [0-9]+ directives
ruoso pmurias, [particle], I'm a bit ignorant about how to implement trace gc... is it possible to have trace gc on only part of the objects? 19:19
or is it all-or-nothing?
19:19 avar joined, kanru left 19:20 avar left
[particle] i suppose you could trace a portion of the objects, but i haven't seen it done 19:20
19:22 kanru joined
pmurias ruoso: it should be possible to do it only on part of the object withought any problems 19:25
* objects 19:26
19:26 avar joined
ruoso I suppose we need to add something like SMOP_RELEASE and SMOP_REFERENCE for that to work... 19:26
and stablish a policy around it
as there is for REFERENCE and RELEASE
pmurias i was thinking of SMOP_MARK 19:27
ruoso that would recursively call MARK on every object it references, right? 19:28
19:28 avar left
pmurias yes 19:28
ruoso and store a bit somewhere, saying it is marked, obviously 19:29
pmurias yes
ruoso so the SMOP_MARK for an object that doesn't use trace gc would simply call mark on the objects it references... 19:30
and not store anything anywhere...
meaning...
pmurias mark on refcounting objects would be a noop
ruoso but that no-op still need to call mark on the objects it references
doesn't it? 19:31
19:31 avar joined, zamolxes left
pmurias the scheme i planned is that *all* objects will have a refcount, and the tracing gc would reference an object when it registers it and release it when it can now longer find it using MARK 19:32
19:33 iblechbot joined
pmurias ruoso: makes sense? 19:33
ruoso why do you need the refcount at all? I mean... if you have the trace, it replaces the refcounting entirely 19:34
19:34 mberends joined
ruoso an alternative gc would mean an alternative to smop_lowlevel.sm0p+smop_lowlevel.h 19:35
which means a different way of allocating an object and a different way of managing the object..
pmurias if i have an object which uses is governed by the gc and i pass it to refcounting code i can't free it unless the refcount falls to zero
ruoso pmurias, that would be solved by the "no-op SMOP_MARK still calling mark on the referred objects" 19:36
even if it doesn't store anywhere the mark bit
pmurias why use refcounting at all then? 19:37
ruoso well, because it's already there...
and I still like the determinism refcount gc provides 19:38
trace gc needs a lot of work to be efficient...
refcount generate bugs... but after you eliminate the bugs, it's performance is quite linear 19:39
pmurias shower for real&
19:44 rindolf left 19:45 exodist_ joined
TimToady but any p6 program that relies on that determinism is erroneous 19:45
19:46 abra left 19:48 Exodist left
TimToady the future of computing is multicore and numa. both of those require us to install a certain amount of non-determinacy into the language or it won't scale 19:48
pmurias } 19:51
void smop_s1p_code_mold_init() {
TimToady and I strongly suspect that ref counting doesn't mix well with the future. can you imagine trying to keep reference counts on web pages?
19:52 lisppaste3 left
TimToady the future will tend instead toward prioritized probabalistic GC 19:52
19:53 lisppaste3 joined
pmurias TimToady: $! requires some determinism 19:54
19:56 clintongormley left, kanru left
moritz_ what's the connection between GC and $!? 19:57
pmurias $! requires semi deterministic destruction (it's specced to trigger a gc run on every scope exit) 19:58
moritz_ uhm, where? 19:59
pmurias if you forget about $! withought checking it, it throws an exception
S04:937
moritz_ thanks 20:00
pmurias which is very usefull in itself but one of the tricky bitd
* bits 20:01
ruoso: you can only free an object once it's refcount is equal to zero and the tracing gc can't find it, so if all objects are traceable the refcounts can't do anything possibly usefull 20:03
moritz_ there are partial refcouting schemes, but I don't think they're particularly successful 20:04
pmurias moritz_: pointers
please
moritz_ ? 20:05
pmurias for partial refcounting 20:06
TimToady $! is lexically scoped, hence scope exit is not a problem
moritz_ pmurias: dunno, it's long time since I read about them 20:07
pmurias moritz_: do you mean something like in python where the gc removes circural references? 20:08
moritz_ pmurias: yes
ruoso TimToady, while I conceptually agree with that (expect determinism to be erroneous), pragmatically there is a lot of uses for determinism in day-by-day programming 20:09
and most machines are still only dual core... 20:10
TimToady to be sure, which is why I've been thinking about what "semi-eager" means
20:11 zamolxes joined
pmurias if the cores get too plentifull there's always haskell ;) 20:11
ruoso and in my experience, trace gc can lead to massive memory use... 20:12
but my experience was with java 1.4
and that gc is indeed horrible
they probably made it better since then... 20:13
buu I'm in a java class right now =[ 20:14
ruoso buu, I know your pain ;) 20:15
buu It really is painful. It's full of obscure syntax and magic conventions and excessive verbose crap 20:16
20:16 hanekomu left
pmurias ruoso: re trace gc needing a lot of work to be efficient i feel that with the amount of time wasted fighting memory leaks we would have wrote a good one ;) 20:18
20:18 cathyal joined
pmurias ruoso: why do some files in src start with smop_ and others don't? 20:19
ruoso maybe... the problem is that I don't have the skills to work on an efficient gc... there's a lack of cs books in my shelf
pmurias, plain inconsistency... ;)
ruoso have sc books instead.. (social sciences...) 20:20
ruoso . o O ( oops... that would be ss books... the joke wasn't even valid.. sigh...
pmurias ruoso: you studied social sciences? 20:22
ruoso yes... I'm still missing some credits to have the actual diploma... but I had already even presented my final monography
pmurias why did you choose social sciences instead of CS? 20:23
ruoso I looked at the list of disciplines of cs and said... too much math 20:24
then I went to a college where there are only 64 hours of math in the entire courese 20:25
statistics applied to social sciences... which, here in Brazil, mostly means no statistics at all
we almost don't use quantitative analysis 20:26
20:26 eternaleye left
ruoso I've actually never used that in any research as a social scientist... 20:26
pmurias should i code a hybrid refcounting/tracing gc or do we switch to a pure gc one? 20:28
ruoso I'm afraid that a bad trace gc could make smop painfully slow and memory hungry in the very near future 20:30
20:31 eternaleye joined
ruoso pmurias, but if you are confident it will work, then just go for it 20:31
pmurias ruoso: i don't think it will end up being *both* memory hungry and painfully slow ;)
ruoso heh..
anyway... I don't actually have the skills to make that decision.. I feel that refcount gc is easier to deal with (even if generates more bugs and requires more debugging) and gets a sane performance and memory usage... 20:33
but that's is just a feeling...
not supported by any concrete information...
so if you really think trace gc will be better, just do it... 20:34
pmurias i'll just code a brutish hybrid one for now 20:35
ruoso maybe a graph builder to help debugging would be good enough
I've just realised 20:38
if we change the SMOP_RELEASE macro to print where it's being called it will already be a huge help... 20:39
ruoso need to reboot for a new kernel version... 20:44
ruoso later &
20:44 ruoso left 20:49 alester left 20:50 mberends left 20:54 cathyal left 20:56 ispy_ joined 21:00 jferrero left, jferrero joined, pmurias left 21:49 araujo left 21:50 iblechbot left 21:52 TJCRI left 22:07 justatheory left
pugs_svn r22271 | moritz++ | [t/spec] updated a fudge message 22:11
22:35 Auzon joined 22:41 Limbic_Region joined 22:49 exodist_ left 22:58 elmex left 23:01 jferrero left 23:03 omichael left
s1n @seen pmichaud 23:05
lambdabot pmichaud is in #perl6. I last heard pmichaud speak 4h 55m 36s ago.
23:10 kanru joined 23:23 justatheory joined 23:24 grindz joined 23:29 grindz left 23:40 kst```` left, kst```` joined 23:44 alester joined 23:51 ruoso joined

Powered by Cro and the RakuĀ® Programming Language.

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

Thank you!