»ö« Welcome to Perl 6! | perl6.org/ | evalbot usage: 'p6: say 3;' or rakudo:, std:, or /msg camelia p6: ... | irclog: irc.perl6.org | UTF-8 is our friend!
Set by moritz on 25 December 2014.
lue "half stumbles to the place on the floor here the gun is." missing 'w' in 'where' 00:01
00:02 BenGoldberg_ joined
masak lue: fixed. lue++ 00:02
00:03 eternaleye left 00:04 dj_goku joined, dj_goku left, dj_goku joined
lue "the aim is to produce something a Perlish macro system" did you mean "something of a" ? 00:05
masak removing the "something". 00:06
geekosaur or something resembling...
00:06 Ben_Goldberg left
lue masak++ I love how it starts out as perfectly serious, then halfway through instantly breaks down into something silly :) 00:08
masak \o/
timotimo just watched Airplane! with friends
good times
00:10 eternaleye joined
b2gills m: say 5 ~~ Int:D; say 5 ~~ Int:U; # I wonder if this can be easily made to work by adding adverbs to &infix:<~~> 00:13
camelia rakudo-moar e37eb0: OUTPUT«True␤True␤»
masak b2gills: that... seems like an odd way to implement it...
b2gills: what about `my $t = Int:U; say 5 ~~ $t;` ? 00:14
00:22 rurban left
masak b2gills: also, what about all the places where smartmatching is used, but not infix:<~~>? where clauses, when statements...? 00:23
00:34 virtualsue joined 00:43 tgt left
masak 'night, #perl6 01:02
01:02 Ben_Goldberg joined 01:03 eternaleye left
dalek kudo/nom: 285435e | TimToady++ | src/core/Any.pm:
use object hashes for .classify and .categorize
 01:03
kudo/nom: 20c0958 | TimToady++ | src/core/Signature.pm:
internal use of .classify needs object semantics
kudo/nom: 949d82c | TimToady++ | src/ (2 files):
make :{} mean object hash
01:06 BenGoldberg_ left 01:09 eternaleye joined
dalek ecs: 1159122 | TimToady++ | S03-operators.pod:
describe :{} object hash composer
 01:17
01:22 Mso150 joined
dalek kudo/nom: 989bb60 | TimToady++ | src/core/Hash.pm:
make :{} symbol available to rest of setting
 01:24
vendethiel seems useful :D 01:25
TimToady m: say :{}.perl 01:41
camelia rakudo-moar 949d82: OUTPUT«:{}␤»
01:41 virtualsue left
TimToady m: say :{ 0 => 42, 1 => 43 }{0} 01:42
camelia rakudo-moar 949d82: OUTPUT«42␤»
TimToady m: say :{ 0 => 42, 1 => 43 }<0>
camelia rakudo-moar 949d82: OUTPUT«(Any)␤»
vendethiel is "custom indexing" implemented in rakudo? 01:43
TimToady m: say (^100).classify(* % 7).perl
camelia rakudo-moar 949d82: OUTPUT«Hash[Any,Any].new(2 => [2, 9, 16, 23, 30, 37, 44, 51, 58, 65, 72, 79, 86, 93], 5 => [5, 12, 19, 26, 33, 40, 47, 54, 61, 68, 75, 82, 89, 96], 0 => [0, 7, 14, 21, 28, 35, 42, 49, 56, 63, 70, 77, 84, 91, 98], 4 => [4, 11, 18, 25, 32, 39, 46, 53, 60, 67, 74, 8…»
TimToady not yet
hmm 01:44
m: say (^100).classify(* % 7).item.perl
camelia rakudo-moar 949d82: OUTPUT«:{4 => [4, 11, 18, 25, 32, 39, 46, 53, 60, 67, 74, 81, 88, 95], 3 => [3, 10, 17, 24, 31, 38, 45, 52, 59, 66, 73, 80, 87, 94], 0 => [0, 7, 14, 21, 28, 35, 42, 49, 56, 63, 70, 77, 84, 91, 98], 1 => [1, 8, 15, 22, 29, 36, 43, 50, 57, 64, 71, 78, 85, 92, 99], …»
TimToady yeah, has to be item to get :{} form back out of it 01:48
m: .say for (^100).classify(* % 12).sort 01:50
camelia rakudo-moar 949d82: OUTPUT«0 => 0 12 24 36 48 60 72 84 96␤1 => 1 13 25 37 49 61 73 85 97␤2 => 2 14 26 38 50 62 74 86 98␤3 => 3 15 27 39 51 63 75 87 99␤4 => 4 16 28 40 52 64 76 88␤5 => 5 17 29 41 53 65 77 89␤6 => 6 18 30 42 54 66 78 90␤7 => 7 19 31 43 55 67 79 91␤8 =>…»
TimToady \o/ (numeric sort)
01:52 kaare__ left
timotimo i don't really know what an object hash is? 01:57
TimToady a hash of objects rather than strings 02:00
Any --> Any rather than Str --> Any
02:02 BenGoldberg_ joined 02:04 eternaleye left 02:06 Ben_Goldberg left
timotimo oh 02:10
of course 02:11
these are useful very often
02:16 eternaleye joined, tinyblak joined
JimmyZ moritz: Thanks :) 02:39
02:57 kjs_ joined 03:01 Sqirrel left 03:02 Sqirrel joined
dalek ast: b5e1723 | TimToady++ | S32-list/categorize.t:
update to use new :{} notation
 03:02
TimToady forgot to check that one in...
03:03 Ben_Goldberg joined, eternaleye left 03:06 BenGoldberg_ left 03:09 erdic left 03:10 erdic joined 03:17 eternaleye joined 03:18 Patterner joined 03:19 gfldex left 03:22 Psyche^ left 03:43 leont left
novice777 Does anyone knows best(good?) TrueType Unicode font for Windows console with the wide range support of character code tables? 03:48
03:54 kjs_ left
TimToady dunno about windows; on linux I use the font 'Monospace' augmented with Aegean.ttf Aegyptus_B.ttf Aegyptus_R.ttf Akkadian.ttf Analecta.ttf Anatolian.ttf Gardiner.ttf HAN NOM A.ttf HAN NOM B.ttf Maya.ttf Musica.ttf Symbola.ttf Unidings.ttf 03:54
03:58 noganex_ joined 04:01 noganex left 04:02 eternaleye left 04:12 Mso150 left 04:13 eternaleye joined
raydiak iirc dejavu mono has above-average unicode coverage...download is a ttf, iirc that installs just fine in windows 04:15
04:18 telex left 04:20 telex joined
raydiak I think "Monospace" means "use the system's default mono font" usually 04:22
geekosaur yes, it's a font alias and usually mapped to bitstream vera mono or dejavu mono 04:53
04:56 Rounin joined
geekosaur vera sans mono that is 04:56
raydiak panda doesn't work on rakudo more than a few months old since the reporter branch merge 05:00
wonder if rakudobrew build-panda should at least take an optional arg for a commit so you can build older pandas with older rakudos 05:02
05:04 eternaleye left 05:08 KCL_ joined 05:10 eternaleye joined 05:21 kurahaupo1 joined 05:43 eternaleye left, eternaleye joined 05:45 Ben_Goldberg left 05:52 novice777 left 06:03 rindolf joined 06:05 eternaleye left 06:06 eternaleye joined
japhb masak++ # github.com/masak/007/blob/master/README.md 06:10
06:15 Rounin left 06:21 booly-yam-4706 left 06:24 Rounin joined 06:30 tinyblak left, tinyblak joined 06:55 keyle joined 06:56 Akagi201 joined 06:58 keyle left, tinyblak_ joined 07:01 booly-yam-4706 joined, tinyblak left
dalek c: 5be61c2 | skids++ | lib/Type/EnumMap.pod:
Remove old caveat about operator adverbs

  .{}:exists has worked in rakudo for some time now		 07:06
07:12 booly-yam-4706 left 07:15 darutoko joined, Rounin left 07:25 xfix joined 07:39 kaare__ joined
moritz m: enum A <b c>; say A.^mro; say b.^mro 07:50
camelia rakudo-moar 989bb6: OUTPUT«(A) (Int) (Cool) (Any) (Mu)␤(A) (Int) (Cool) (Any) (Mu)␤»
moritz m: enum A <b c>; say A.^roles; say b.^roles 07:51
camelia rakudo-moar 989bb6: OUTPUT«No such method 'roles' for invocant of type 'Perl6::Metamodel::EnumHOW'␤ in block <unit> at /tmp/uGmEMJDYcU:1␤␤»
moritz m: enum A <b c>; say b.^roles
camelia rakudo-moar 989bb6: OUTPUT«No such method 'roles' for invocant of type 'Perl6::Metamodel::EnumHOW'␤ in block <unit> at /tmp/QynMOU3Zr_:1␤␤»
moritz m: enum A <b c>; say $_ ~~ Enumeration for A, b
camelia rakudo-moar 989bb6: OUTPUT«False␤False␤»
07:52 rindolf left 07:54 kurahaupo1 left
moritz masak++ # 007 REAMDE 08:01
08:05 prime left 08:06 prime joined
moritz m: enum A <b c>; say b.pick 08:09
camelia rakudo-moar 989bb6: OUTPUT«b␤»
moritz m: enum A <b c>; say A.pick
camelia rakudo-moar 989bb6: OUTPUT«c␤»
08:23 kurahaupo1 joined
dalek c: ea577db | moritz++ | lib/Type/Supply.pod:
Document Supply.from-list
 08:24
08:34 booly-yam-4706 joined 08:35 vendethiel left 08:36 vendethiel joined
dalek q: bb96233 | moritz++ | answers.md:
Expand a bit on private attributes set from a constructor
 08:38
q: 5aecccd | moritz++ | answers.md:
Explain numierc types, and which ones to use as type constraints
08:45 gfldex joined 08:46 booly-yam-4706 left
dalek q: bb42933 | moritz++ | answers.md:
Linkify some types
 08:49
q: e0a4272 | moritz++ | answers.md:
Link to container docs
 08:52
08:57 kurahaupo1 left 09:14 booly-yam-4706 joined 09:22 Rounin joined 09:24 kurahaupo1 joined, rurban joined 09:26 molaf__ joined 09:30 molaf_ left 09:34 adu joined 09:48 kurahaupo1 left 09:55 Alina-malina left
dalek q: 9c47517 | moritz++ | answers.md:
Mention gradual typing
 09:58
10:01 booly-yam-4706_ joined 10:02 booly-yam-4706 left 10:08 adu left 10:12 Alina-malina joined 10:14 tinyblak_ left, tinyblak joined 10:20 spider-mario joined 10:30 FROGGS joined, anaeem1 joined 10:34 espadrine joined
FROGGS moritz: I get a service timeout for the dalek webhook :o( 10:57
moritz: is this the correct url? hack.p6c.org/dalek?t=freenode,perl6-lwp-gsoc
11:03 jluis_ joined
FROGGS ohh, found it at: github.com/perl6/mu/blob/master/mi...k-push.txt 11:08
moritz \o/ docs++ 11:09
FROGGS indeed :o) 11:11
11:13 anaeem1 left 11:14 denis_boyun_ joined 11:15 anaeem1 joined
jnthn afternoon, #perl6 11:24
FROGGS hi jnthn 11:25
lizmat FROGGS jnthn o/ 11:29
11:30 pmurias joined
FROGGS hi lizmat 11:31
jnthn is well rested and mostly free of cold o/ 11:32
...and still has a BS bug /o\
dalek kudo/nom: 20d5ef3 | lizmat++ | / (4 files):
Introduce Rangeint: simple integer ranges like ^10

Is not used in the core (yet). The idea being that the whole range machinery is not needed for simpel integer ranges like ^10 or 200..500. Bolting on will need to wait until some bugs are fixed.		 11:33
lizmat jnthn FROGGS: need some nqp:: functions related help 11:34
do you see anything obviously wrong in Rangeint.BUILD ?
jnthn Not immediately. 11:36
Other than it's one of the unsubclassable ones 'cus it breaks the normal contract. 11:37
11:37 denis_boyun_ left
lizmat well, it's supposed to be an internal class, to be used only if Range.new sees it's possble to use 11:37
jnthn For these cheating ones it may be better to write them as !BUILD
lizmat ok 11:38
jnthn Ah, then fair enough...but it will leak out due to .WHAT of course.
lizmat $ 6 'my $r = Rangeint.new(0,9); say $r.flat.perl'
This type does not support positional operations
jnthn: yes, but ~~ Range should be True
jnthn $!list := nqp::null();
null doesn't support positional operations...
:P
I wonder if there's a missing guard... 11:39
lizmat so, I want $!list to be populated only when needed
FROGGS --ll-exception will at least point you to the offending line... 11:40
jnthn yeah, I can see...I'm not sure where it's not being done
11:40 booly-yam-4706_ left
jnthn lizmat: How much overhead does the "is this an Int range" stuff actually have? 11:41
I thought we already special-cased this inside of Range.
lizmat not much, really
and there's all the include/exclude min checks 11:42
that Rangeint doesn't have
reification
at_pos wouldn't need to reify, e.g. 11:43
jnthn Yes, your class missed reify
*misses
lizmat nope, it has one
self.flat
jnthn ah, I see
lizmat but I'm not sure that just shouldn't return anything 11:44
especially since we don't need reification for at_pos
jnthn Still, I don't know we should be handling this with type explosion.
Rather than the "stash a closure that knows how to reify" approach we're looking at for the GLR. 11:45
lizmat at startup, already 6 Range objects are created, all that could be handled by this class
jnthn Yes, but at startup 50% of our time is spent deserializing.
Well, 50% of our CPU cycles.
And adding more types/methods doesn't help that, if we're looking at startup cost.
lizmat so you're saying this is a bad idea? 11:46
11:46 rindolf joined
lizmat any slice with a range should become much faster with this as well 11:46
jnthn Yeah,b ut when do we slice integer ranges? 11:47
*but
The really simple iteration cases like "for ^1000 { }" should already be getting re-written by the optimizer to while loops. 11:48
lizmat in the core, in Termporal
jnthn Line?
lizmat 198, 206, 215
but I'm more thinking about ecosystem doing @a[^10] and stuff 11:49
ok, that's for the exception
jnthn Yes, those are for exceptions
For ^10 note that if we were a little smarter we could constant-fold ^10 11:50
But now I see you meant slicing *with* a range, not slicing the range itself.
lizmat ah, yes, sorry for the unclarity
jnthn np
lizmat anyway, doing an at_pos on a range like this wouldn't need reification 11:51
jnthn But don't we .at_pos($_) for @the-param-that-has-the-range ?
lizmat there's quite a lot ^$var in the core 11:52
jnthn multi sub postcircumfix:<[ ]>( \SELF, Positional:D \pos ) is rw { 11:53
nqp::iscont(pos)
?? SELF.at_pos(pos.Int)
!! POSITIONS(SELF,pos).map({ SELF[$_] }).eager.Parcel;
}So the range is passed to POSITIONS
And overload of POSITIONS on Range might be the easier place to get a win? 11:54
lizmat well, that would be the next step: if pos is a Rangint 11:55
jnthn The whole POSITIONS mechanism - introducing a chain of 2 maps once you include the one in the postcircumfix:<[ ]> method - looks not entirely cheap
lizmat I'm open to suggestions :-)
remember, it also needs to handle nested arrays / hashes and attributes 11:56
11:56 tgt joined
lizmat the fact that we need to make so many different MMD candidates is also something that may need some scrutiny 11:56
11:57 kjs_ joined
jnthn Attributes? 11:59
lizmat yup
FROGGS in case I have an xml attribute: <foo:bar xmlns:foo="kungfoo"/> - what should be the .name?
I'd tend to "bar", but I dunno is that's too lax 12:00
jnthn lizmat: Example?
lizmat multi sub postcircumfix:<[ ]>( \SELF, int $pos, :$delete!, *%other ) is rw { 12:01
moritz FROGGS: it could be 'bar' or 'foo:bar'
jnthn lizmat: Adverbs?
lizmat multi sub postcircumfix:<[ ]>( \SELF, int $pos, :$exists!, *%other ) is rw {
multi sub postcircumfix:<[ ]>( \SELF, int $pos, :$kv!, *%other ) is rw {
FROGGS moritz: yes, the question is: "foo:bar" to be more strict by default, or should it be "bar" ?
arnsholt FROGGS: Isn't there an answer to that in the XML standards?
lizmat jnthn: etc etc
FROGGS moritz: and... in case it is "bar", what's the long form? .fq-name? .longname? ... 12:02
jnthn Right, but I see no attributes there...I can only guess you are talking about the adverbs. In which case, yes.
FROGGS arnsholt: no idea
lizmat sorry, yes, adverbs
*sigh* :-(
FROGGS the P5 lib has .name as "foo:bar", and "bar" is .localname
jnthn FROGGS: And presumably .namespace or so for getting the foo? 12:03
FROGGS .ns, aye
there is .ns.prefix (will change it to .ns.name probably), and .ns.uri and some more
jnthn lizmat: On the upside, each of the multi candidates is quite short and isn't doing much in the way of introspection.
arnsholt Yeah, I think that's the correct way
FROGGS arnsholt: with .localname? 12:04
lizmat jnthn: yes, that's why it's that way
otherwise a lot of checks would need to be done inside a single candidate
FROGGS arnsholt: the naming in libxml2 is quite inconsistent and weird btw... 12:05
lizmat jnthn: re code size: Range could become smaller, as no special casing for Int would be needed there anymore
I also wonder whether adding MMD for infix:<..> and friends would help 12:07
FROGGS arnsholt: do you have any resources at that about a good naming scheme?
arnsholt Don't have any good references, sorry
But "local name" sounds familiar 12:08
FROGGS yeah, it sounds sane in some way
12:10 KCL joined 12:11 denis_boyun_ joined
moritz can somebody recommend a book on evidence-based software engineering practices? 12:12
I've come across " Making Software: What Really Works, and Why We Believe It" by Andy Oram and Greg Wilson
it looks interesting, but also pretty heterogenous (many authors) and rather thick 12:13
12:13 KCL_ left
moritz something more condensed would be very welcome 12:13
(unless you all recommend it very highly)
FROGGS I can't recommend anything 12:14
pmurias masak: 007 will only have integers? 12:19
masak good afternoon, #perl6 12:22
pmurias: I imagine so, yes.
I'm glad y'all like the README.md :)
it was fun to write. 12:23
here it is again for those who missed it: github.com/masak/007/blob/master/README.md
pmurias: floats and complex numbers are not really on the critical path to exploring macros ;) 12:25
12:26 Rounin left
arnsholt +1 to that =) 12:27
pmurias adding float support with macros could prove informative ;)
12:31 kjs_ left
masak well, there's a thought there that I keep returning to: "adding support for [data type X], using macros" 12:36
but I usually reach the conclusion that it's not that feasible. and probably the wrong tool for the job.
macros are good for adding syntax, keywords, and influencing control flow.
they're not that good at runtime things, like values and their types. 12:37
12:37 pmurias left
moritz well, maybe for floating point literals 12:40
arnsholt Yeah. You can add pretty much whatever you want with macros (cf. TeX), but it might not always be the best way to go about it (cf. TeX, again =) 12:43
12:44 zakharyas joined
masak moritz: literals should be fine, as long as I implement macro term:<...> 12:45
moritz: the problem is more on the other end: what should such a macro emit? 12:46
12:47 mr-foobar left
arnsholt Same thing as an expression macro, no? 12:47
masak it seems reasonable to reply "sign, mantissa, and exponent, encapsulated in a suitable data structure". in 007, arrays could do that, but then you'd have to somehow distinguish a "normal" array value from one that pretends to encode a float.
arnsholt Oh, that kind of "what should it emit". Never mind me 12:48
masak and (crucially) the whole runtime has to make that distinction, which turns the change into a nastily global one.
hence "not that feasible".
moritz well, just add (primitive) custom types to the language
arnsholt Well, you could always ignore that particular aspect of the problem 12:50
moritz round to the nearest integer :-)
arnsholt In Perl 6 a new literal would for example create an instance of a class, but 007 can very well choose to ignore problems of typing 12:51
New datatypes simply being lists under the hood is very much in the Lisp tradition, after all =)
12:52 Rounin joined
moritz how does lisp folks distinguish the various data types that are all implemented as lists? 12:56
arnsholt They don't, necessarily 12:57
My thought is that 007 can just ignore typing concerns like that. In Perl 6 the existing type system should be fine, I think 12:58
13:01 zakharyas left 13:03 denis_boyun_ left
moritz news.ycombinator.com/item?id=8906331 another discussion about the "Why Perl Didn't Win" article from last year 13:03
13:08 kjs_ joined 13:09 Ben_Goldberg joined 13:10 Ben_Goldberg left 13:11 Ben_Goldberg joined
rurban moritz: Just an old rehash from the typical perl6 killed perl5 mindset 13:17
not a really original and educated post 13:18
13:21 booly-yam-4706_ joined
woolfy moritz: what rurban said, plus, please don't read that shit, it is what makes me depressed every so often, and makes me unfriend people on Facebook and ignore people on irc... 13:23
moritz: I will not ignore you, just be clear!
moritz woolfy: now I'm glad 13:24
jnthn Well, it seems the thing I ran into with the mixin stuff has been discovered by lizmat++ before... 13:31
irclog.perlgeek.de/perl6/2014-06-26#i_8936466 and irclog.perlgeek.de/perl6/2013-08-30#i_7523993
Took me multiple hours to get to the bottom of what I think is going on... 13:32
And now I want lunch. But at least we might get a fix for a more general issue out of this, even if it's blocking the stuff I actually want to be doing. :)
lizmat that's always good :-)
jnthn Aye. Will see if I can fix it post-eating. :) 13:33
13:33 denis_boyun_ joined
rurban t/spectest.data seems to be not adjusted yet: Missing test file: t/spec/S32-list/uniq.t 13:35
woolfy: It's not shit. that's the bad reality out there. we need to educate the people 13:36
JimmyZ we will keep silent on that 13:37
woolfy rurban: it becomes shit when Perl-people start believing that they are working in a dead or dying language, and start iterating it to other Perl-people, utterly de-motivating developers on the way...
JimmyZ *hopes we
lizmat rurban: you need a more recent rakudo then 13:38
it *is* in there now
moritz it's unique.t now
lizmat if you want a quick fix, symlink it locally to unique.t
rurban I've got the new test, saw it. just got this warning also 13:41
13:44 leont joined
moritz lizmat: S17-lowlevel/lock.t is commented out in t/spectest.data -- is that as a flapper? 13:48
lizmat yes
13:54 virtualsue joined
masak arnsholt: you opened my eyes a bit. if I just give up and accept that (a) my float is an array, (b) if people want to misuse it, they can -- then I can basically have macros implement floats. 13:55
even better, that works for any data type. I could even build a MOP that way. 13:56
vendethiel *g*
arnsholt Welcome to Lisp! ;)
masak attempts to implement custom operators, which are all but switched on in 007 already 13:57
how does one say "commit to this rule" again? was that `::` ? 13:59
m: grammar G { rule foo { OH :: HAI } }; say "alive" 14:01
camelia rakudo-moar 20d5ef: OUTPUT«===SORRY!=== Error while compiling /tmp/SkWOK0AkRH␤:: not yet implemented␤at /tmp/SkWOK0AkRH:1␤------> grammar G { rule foo { OH ::⏏ HAI } }; say "alive"␤»
masak oh, you've got to be *kidding* me.
masak adds a bunch of red cells to perl6.org/compilers/features 14:02
n: grammar G { rule foo { OH :: HAI } }; say "alive"
camelia niecza v24-109-g48a8de3: OUTPUT«(signal ABRT)Stacktrace:␤␤ at <unknown> <0xffffffff>␤ at (wrapper managed-to-native) object.__icall_wrapper_mono_gc_alloc_string (intptr,intptr,int) <0xffffffff>␤ at (wrapper alloc) object.AllocString (intptr,int) <0xffffffff>␤ at string…»
masak Niecza is feeling a bit under the weather. but I bet it implements `::`. 14:03
14:07 beastd joined
dalek atures: a807c94 | masak++ | features.json:
[features.json] add "::"

Currently NYI in Rakudo (though I think we had it before, sometime pre-ng).		 14:08
nwc10 Double colon LTA? stopper 14:09
masak nwc10: see evals above. 14:10
oh, "LTA". hm.
"LTM", hold on :)
it would be nice if we could stop LTA sometimes...
jnthn masak: You can use {} to terminate LTM.
dalek atures: 06ae482 | masak++ | features.json:
[features.json] s/LTA/LTM/ typo

  nwc10++		 14:11
masak jnthn: thanks. I'll try that.
jnthn: is the behavior *exactly* identical? can we use the `{}` semantics to implement `::` ?
jnthn masak: I don't think it's identical, no, it's just that they are equally good at LTM termination and that's what, afaik, STD relies on it for. 14:12
lizmat m: say 1.5 ~~ 1^..^2 # this is a bit of a WAT to me
camelia rakudo-moar 20d5ef: OUTPUT«True␤»
FROGGS no, it differs wrt backtracking IIRC
jnthn masak: Which is largely why it didn't get implemented yet, I guess; there's a workaround that's usually good enough. :) 14:13
lizmat: Why?
masak lizmat: then we have different understandings of what 1^..^2 means.
jnthn m: say 1 ~~ 1^..^2
camelia rakudo-moar 20d5ef: OUTPUT«False␤»
jnthn m: say 1 ~~ 1..^2
camelia rakudo-moar 20d5ef: OUTPUT«True␤»
masak lizmat: does it help if you think of it as an intercal on the real line?
jnthn Looks like it does what I want.
*I'd
masak interval*
lizmat I guess I'm thinking too much int :-) 14:14
jnthn
.oO( COME FROM the_real_line )
masak lizmat: with the ^ thingies excluding *only* one real number.
lizmat yeah, gotcha
masak jnthn: my reason for adding it to the compiler comparison table is *purely* to shame us all into implementing it. the fact that we once had the feature only makes it more depressing that we don't now. 14:15
14:15 eternaleye left
jnthn masak: I don't remember us having it, tbh. 14:17
masak: It hasn't been there so long as I've been aware of the regex engine internals, at least.
nwc10 masak: so I guessed right that there was someting LTA with the "LTA"? :-) 14:23
14:26 spider-mario left 14:29 tinyblak_ joined 14:32 anaeem1 left 14:33 tinyblak left
FROGGS masak: btw, I was thinking about implementing ::, :::, ::::, <cut> and <commit>... but after reading the spec carefully I shuddered 14:33
my regex engine backtracking foo is not good enough
14:37 anaeem1_ joined
masak "fu". 14:53
jnthn "fu2!" :P 14:55
jnthn has sorted out his flights for FOSDEM o/ 14:56
14:56 denis_boyun_ left
lizmat ;-) 14:56
14:56 kurahaupo joined
jnthn Arrive (well, plane lands) at 1pm on the 30th 14:56
lizmat well in time for delirium tremens 14:58
jnthn ;)
Only if I finished writing my talk! 14:59
lizmat :-)
15:00 BenGoldberg joined 15:02 zakharyas joined
masak .oO( sorry about slides 26..51 being all shaky-like. that's as far as I got before the delirium tremens ) 15:05
15:07 kjs_ left
masak m: grammar G { regex TOP { <foo> }; proto rule foo {*}; rule foo:one { sub {} foo }; rule foo:two { \w+ \w+ } }; say ?G.parse("sub bar") 15:08
camelia rakudo-moar 20d5ef: OUTPUT«True␤»
itz delirium tremens? that's a cafe in Belgium isn't it? :)
yoleaux 17 Jan 2015 20:24Z <lizmat> itz: what happened to szabgab's post on pl6anet.org ?
masak if the `{}` commits the parse to foo:one, shouldn't the above parse fail?
itz lizmat: it came back
masak itz: it's also a beer. www.ratebeer.com/beer/delirium-tremens/1039/
15:14 kurahaupo left
lizmat itz: I'm not seeing any of them? 15:15
jnthn masak: I didn't say it commits, I said it terminates LTM. 15:16
masak: As in, ends, the declarative prefix
lizmat itz: Delirium is the bar where people hang out the night before FOSDEM
masak jnthn: hm. ok.
how do I make a rule commit at some point, and fail the whole parse if the rule fails?
lizmat itz: deliriumcafe.be 15:17
jnthn The Perl 6 grammar usually does something like 'sub' [ ...what we want... || <.panic: 'OH NOES'> ] }
masak ok, that works, I guess.
jnthn Well, it also means you get a good error. :)
masak right. 15:18
15:21 rurban left
dalek kudo/6pe-mop: f21a64c | jnthn++ | src/Perl6/ (2 files):
Keep compiler callbacks in place during optimizer.

The optimizer may trigger compilation of things that it calls - in the setting because it defines many built-ins, but also in normal code that uses, for example, 'is pure'. This prevents us from removing fixup infrastructure too early. A pre-requisite for fixing problems with multis that get run during compilation when a subclass has derived the proto.		 15:22
15:24 brrt joined
brrt (i suppose most of you have seen the 'why perl didn't win' article now) 15:24
does it make sense to even write and post a counterargument? 15:25
lizmat brrt: it's old news
but feel free to :-)
brrt :-)
15:25 rurban joined
moritz brrt: it's from Aug 2014 15:25
brrt o really?
itz lizmat: I'm seeing szabgab's feed now .. I assume his server was down at some point yesterday since it wasn't displayed then
moritz brrt: just now on hackernews again
itz I've found avoiding perl postings on mainstream tech sites a good idea particularly if it's "old news" from last year 15:26
brrt i'll brb. the weakness of that argument is that it focuses only on the programmers, as if they're making the decisions 15:27
:-)
15:28 denis_boyun_ joined, kurahaupo joined 15:31 brrt left 15:34 denis_boyun_ left
lizmat jnthn: what's the 2nd parameter to p6parcel for ? 15:36
ENODOC :-(
jnthn lizmat: "Replace nulls with this value", iirc
lizmat ah, ok
15:38 denis_boyun joined
lizmat m: class A is Range { has int $.a }; nqp::getattr(A.new(0,10),A,q/$!a/) # sort of expect this to work 15:45
camelia rakudo-moar 20d5ef: OUTPUT«No such method 'BUILD' for invocant of type 'A'␤ in method new at src/gen/m-CORE.setting:8278␤ in block <unit> at /tmp/f5MACDNJqR:1␤␤»
lizmat oops 15:46
15:46 BenGoldberg left
jnthn Yes, that's the danger of non-standard new, BUILD, etc. 15:46
lizmat no, locally I get another error 15:47
P6opaque: no such attribute '$!a'
JimmyZ I think somewhere is using = for nqp:create, it want := 15:49
jnthn Hm, mixin caching helps a bit. 100,000 mixins benchmark now runs in a tenth of the time it used to. :)
lizmat :)
JimmyZ: hmmm.... 15:51
15:52 denis_boyun left 15:54 grondilu joined
jnthn m: say 13866276 - 12781785 15:56
camelia rakudo-moar 20d5ef: OUTPUT«1084491␤»
jnthn Wow, 1MB off CORE.setting from the mix-in re-use too
Turns out we generated a LOAD of them.
Because of using them to add precedence info. 15:57
15:57 booly-yam-4706_ left 15:58 leont left
jnthn And a bit more than 3MB off the process base memory. 15:58
And some spectest fails. Aww. 16:00
arnsholt jnthn++ # smo-pe 16:01
dalek kudo/6pe-mop: aa14b2a | jnthn++ | src/Perl6/World.nqp:
Track and update clones of uncompiled coderefs.

This happens if a proto method in a base class is cloned as part of deriving a proto for a subclass, and afterwards the proto in the base class was invoked before the compilation process was over. The lack of update led to a problem locating the appropriate static code object during serialization.		 16:02
kudo/6pe-mop: 4c52df6 | jnthn++ | src/Perl6/Metamodel/ (2 files):
Implement per-class mixin cache.

Uses the 6model parametric extensions to handle the interning of them. We get some spectest failures out of this, though they may be showing up an existing, somewhat hiding, bug. On the upside: 100,000 mixins benchmark runs in 10% of the time, CORE.setting shrinks by 1MB, and perl6 -e "loop { }" has over 3MB knocked off it too.		 16:06
jnthn Hm, curious. 16:13
role R { }; class C { }; for ^100000 { C.new.^mixin(R) } # runs in 0.93s, including startup
role R { }; class C { }; for ^100000 { C.new does R }; # 4.78s 16:14
timotimo "does" causes global deopt, but ^mixin doesn't?
jnthn What does does do?
timotimo: Nope.
16:14 xfix left
jnthn It's the VM-level changetype op that triggers it. 16:14
timotimo mhm 16:15
is 6pe-mop at spectest success again? :)
jnthn No.
But it's not going to be a merge candidate for a while. 16:16
timotimo oh
OK
jnthn Since 6pe will need to be up to scratch on Moar and JVM, and at least not regression-inducing on Parrot. 16:17
Plus the mixin cache thing is just the easiest thing I could do to start giving it some exercise. 16:18
Other things I'll be doing with it include coercion types, typed array/hash improvements, and it's also one ingredient for native arrays. 16:21
And hopefully fixing various role-related pre-comp bugs along the way. :) 16:22
jnthn away for a bit
timotimo ah, yes of course 16:23
i'm only interested in the moar aspect at the moment :P
because i'm also interested in what can possibly be achieved with spesh
16:27 pmurias joined
pmurias jnthn: is 6pe something that I should be concered about for nqp-js? 16:27
16:30 KCL left 16:33 jluis_ left 16:37 KCL_ joined 16:38 mr-foobar joined 16:47 virtualsue_ joined 16:49 virtualsue left, virtualsue_ is now known as virtualsue
psch hi #perl6 \o 16:49
16:50 telex left, salv0 left
psch FROGGS: i had had a look at the composetype op a bit ago, but i don't quite see how to work that into what i have 16:52
16:52 telex joined
FROGGS well, I don't know what you have :o) 16:52
psch github.com/peschwa/rakudo/blob/jav...JavaHOW.pm 16:54
although that's probably the wrong-ish aproach in at least one other aspect - namely the new op near the bottom
should probably do that similar to how we do other types, i.e. declare them in BOOTSTRAP
i've tried throwing more MOP-roles at it and that didn't help... :P 16:55
arnsholt The composetype op is basically to send information from Perl 6 land to the REPR instance. Not sure if that's what you need or not 16:57
psch basically, what i'm trying is to write the JavaHOW in Perl6 instead of NQP, so i can access other Perl6-level types to automatically get accessors to native fields
and do this automatically: gist.github.com/peschwa/0f8830ba3c6a2eb852ba 16:58
for which the first step is having JavaHOW in the CORE.setting instead of inside the Metamodel, if i understood jnthn++ correctly 16:59
masak "What does does do?" is a screwey sentence. :)
arnsholt Use-mention is always a bit weird =) 17:00
psch: Right. Not really composetype, I agree 17:01
I'm not really familiar with what is required for a HOW in Perl 6, but probably a liberal application of nqp::ops and a certain amount of care in whether the objects you're working with are NQP or Perl 6 objects 17:02
masak arnsholt: speaking of use-mention: twitter.com/wokingwriter/status/55...7844271105 (#autopun)
vendethiel psch: "@methods[+@methods] =" isn't there a .push? 17:04
psch vendethiel: yes, there is, but that's a VMArray iirc 17:05
vendethiel makes sense, noted
FROGGS psch: look at this example: github.com/perl6/nqp/blob/master/t...pes.t#L115
arnsholt masak: Oooh, I love it!
FROGGS psch: forget about the serialization bit... but line 132 is rather important I bet
psch FROGGS: so compose attaches the Method to the type and makes it invokable? 17:06
(among other things)
FROGGS psch: I think so.... you can test what happens if you comment out that line and runt the test 17:07
arnsholt Not quite, I think
The call to HOW.compose isn't the composetype 17:08
FROGGS ohh
arnsholt Grep for composetype in src/Perl6/Metamodel for examples
FROGGS I thought so and that's why I mentioned composetype in the first place
arnsholt Also: github.com/perl6/nqp/blob/master/d...l.markdown 17:09
psch that's another thing, JavaHOW.nqp has "method is_composed($obj) { 1 }", and .^add_method works
FROGGS (on an unrelated note, XML::LibXML now has working xpaths...)
psch simply running the same code in CORE.setting instead of from the Metamodel gives me the mentioned "no such postcircumfix:<( )> for type 'Method'"
17:09 salv0 joined
arnsholt Those are the things the standard 6model REPRs understand. You are of course free to add new things to your own REPR 17:09
psch which also means there's no .invoke
*on Method 17:10
17:15 booly-yam-4706_ joined
psch i now wonder if i actually was justified in writing add_method myself in the JavaHOW instead of does'ing MethodContainer... currently running build will tell i suppose 17:21
lizmat
.oO( what does does and but do? )
 17:27
masak: ^^^
moritz m: say Submethod ~~ Method 17:28
camelia rakudo-moar 20d5ef: OUTPUT«False␤»
17:29 booly-yam-4706_ left
masak just saw whiteknight.github.io/2015/01/15/pa...heend.html 17:31
haven't read it yet.
moritz m: class A { submethod x() { say 42 } }; say A.^can("x").perl
camelia rakudo-moar 20d5ef: OUTPUT«()␤»
moritz is that correct?
masak m: class A { submethod x { say 42 } }; A.new.x
camelia rakudo-moar 20d5ef: OUTPUT«42␤» 17:32
masak moritz: I'd say "no", since you can call it.
vendethiel should "can" take another parameter, for private method, or for walking the inheritance chain?
17:32 anaeem1_ left
masak is that a separate question? submethods are not private. 17:33
17:33 anaeem1 joined
vendethiel that's an idea 17:33
masak if you ask me, it makes little sense to ask .can questions about the non-public interface of an object.
moritz vendethiel: there are ways to walk private methods through the MOP if you really want/need to 17:34
like .^find_private_method($name) and .^private_method_table 17:35
walking the MRO makes little sense for that, because they aren't inherited
17:35 spider-mario joined 17:37 kjs_ joined, anaeem1 left 17:40 mvuets joined 17:43 rmgk is now known as Guest85303, rmgk_ joined, Guest85303 left, rmgk_ is now known as rmgk 17:46 Vlavv_ joined 17:48 anaeem1_ joined 17:50 Rounin left
dalek c: 91e654e | moritz++ | type-graph.txt:
type graph: s:g/"Perl6::Metamodel"/Metamodel/
 18:00
18:06 anaeem1_ left, booly-yam-4706_ joined
dalek c: 2c651c3 | moritz++ | lib/Type/Metamodel/ClassHOW.pod:
more s:g/"Perl6::Metamodel"/Metamodel/
 18:06
18:06 anaeem1 joined 18:07 anaeem1 left 18:08 anaeem1 joined
jnthn pmurias: (6pe) Yes, but you may want to wait for the dust to settle a bit more before porting it. 18:08
18:11 booly-yam-4706_ left, pnu left 18:12 araujo joined, araujo left, araujo joined, clkao_ left, pnu joined, booly-yam-4366 joined, clkao joined
jnthn About 18:13
m: class A { submethod x() { say 42 } }; say A.^can("x").perl
camelia rakudo-moar 20d5ef: OUTPUT«()␤»
jnthn Pretty sure that one has to be a bug.
I can easily imagine how it'd happen too :)
RT it; it's likely a very easy fix.
cooking &
masak RTs it
did Parrot "kick [Rakudo] out" of the Parrot repo? that's not quite how I remember it. 18:16
lizmat context?
ah, I see it 18:17
18:19 Sqirrel left
masak reading the postmortem of a Parrot developer is a little bit like reading something from a mirror world. 18:20
because I'm so squarely placed in the Perl 6 side of things.
18:21 regreg joined
masak as Perl 6 people we can say "oh well, things might not always have evolved as quickly or as successfully as we hoped/planned -- but at least we're not Parrot" 18:21
lizmat++ # "Edsel" meme, "everything to everybody" 18:22
18:22 Sir_Ragnarok left
masak another thought: many in the Parrot project saw the benefit that could have been gained by re-hosting on top of 6model. they didn't have the skills to invent 6model (or similar) in the first place, but they saw the advantages of it. but even as they wanted to move to it, *they were unable to do so* -- due to some unknown mix of technical debt and lack of tuits. 18:25
18:25 booly-yam-4366 left
masak I also note that WhiteKnight's Parrot postmortem has basically the same dark shade of grey background as does Dan's Parrot postmortem :) 18:26
apparently, that's the background to do it in.
18:27 Akagi201_ joined
timotimo i'm not sure i remember dan's parrot postmortem 18:28
18:28 Akagi201 left
skids That's just the background for the site. I actually wish liht text on dark was more popular I find it better on my eyes. 18:31
timotimo skids: i recommend "hacker vision" chrome plugin 18:32
skids Maybe I'll give that a spin. Certainly we are long past the day when you can just set browser colors and expect anything functional to happen. 18:33
18:33 FROGGS_ joined, grondilu left
masak skids: I know it's the background for the site. I still thought it was a weird coincidence. 18:33
skids: maybe Parrot bloggers like that background color.
18:35 kurahaupo left, booly-yam-4366 joined
masak "What I do know is that we *did have the good ideas*, we just weren’t able to implement them correctly or quickly enough." -- I think this sums it up. all of it. 18:35
18:36 FROGGS left
masak I said something similar back in 2011. irclog.perlgeek.de/perl6/2011-08-23#i_4316480 18:37
I *think* M0 was their 6model effort? don't recall exactly. it had something to do with a very small core.
vendethiel masak: I actually first thought it was the same post :P (I never read whiteknight's) 18:40
lizmat timotimo: dan's post mortem was about PONIE: www.sidhe.org/~dan/blog/archives/000435.html
or not? hmmm.. apparently it *was* about parrot 18:41
masak it was about Parrot.
18:42 Sqirrel joined
lizmat this was the PONIE one, by Jesse Vincent: www.nntp.perl.org/group/perl.ponie....sg487.html 18:42
masak by the way... assume the Parrot people at some (surely distant) moment wishes to announce the end of the Parrot project. how *do* they do that, given that the project has that name and, yes, is based on that sketch? it'd have to be a helluva convincing announcement. 18:43
TimToady they obviously would have to do it on April 1st
pity we probably can't get to beta by then... 18:44
pmurias rurban is still working on parrot...
skids I'm sure that was just a funny hypothetical. 18:45
TimToady well, we're not really into shooting our wounded, more like "them's as falls behind, gets left behind" 18:46
rurban Mo or lorito is a llvm-like bytecode project, small core, based on the assumption that parrot is fast and c not. ha! 18:47
And the parrot people who destroyed it already left, So they cannot announce anymore that they failed. 18:49
But there still good stuff in it, which can be revived. Esp. threads are good. Better than moar, the rest is better in moar
18:49 adu joined
masak rurban: I keep seeing absolutely zero evidence that Parrot threads are better in any way at anything. 18:52
timotimo i'm looking forward to a demonstration :)
rurban masak: maybe you should about it then? 18:53
the demos are out there since 2012, showing linear, lockfree performance. show me that in another vm.
esp for arrays and hashes 18:56
adu "about it"?
rurban: is "about" a verb? 18:57
rurban read about it
adu ah
rurban "Those who cannot read need to walk" 18:58
adu writing is good too
masak rurban: can you link me to some Perl 6 code running on Parrot showing off this nice threading? 18:59
rurban: I'll happily look at any URL you give me.
rurban Perl6 does not yet support parrot threading 19:00
And I don't know enough nqp to write it by myself 19:01
adu blog.klocwork.com/user-documentatio...better-fm/
rurban The easy to read part is this one: perl6advent.wordpress.com/2012/12/...t-threads/ 19:03
with winxed and pir samples
masak rurban: there is *no* Perl 6 code making use of Parrot's allegedly awesome threading!? 19:04
rurban you are the perl 6 coders. I'm just maintaining parrot 19:05
masak rurban: all this time since 2012, no-one has bothered to show off the fantastically impressive threading *on Parrot* by making it work on Perl 6?
rurban nobody cared
and I doubt that perl6 has enough competence in this field to see the bigger picture 19:06
there was a famous flamefest initiated by diakopter when we added native threads
masak what language is threading awesome in, then? PIR?
timotimo froggs tried at one point to make something work with parrot threading, was it hypers and math or something?
19:07 booly-yam-4366 left
rurban and the original author never defended his work, so it stayed 19:07
19:07 anaeem1 left
rurban in parrot of course, not comparable to moars' locked, conservative threading. moar locks on everything 19:07
adu rurban: I care
rurban so add it
adu I'm busy working on slangs 19:08
pmurias masak: from an outsider perspective a lot of the parrot folk loved pir
19:08 anaeem1_ joined
rurban but it's a bit tricky, because you need to track the owner and schedule writer tasks for all updates, if you are not the owner. only owner may write 19:09
all clients who want to write need to do it via little updaters
so it's basically lockfree but not wait free. 19:10
19:12 booly-yam-4366 joined 19:14 zakharyas left
timotimo so it's basically asynchronous? 19:14
masak rurban: "I am the best chess player in the world. it's just that I don't want to play with other chess players." 19:15
pmurias isn't it a crude for of software transactional memory 19:17
masak rurban: if Parrot has super-awesome threading, and Rakudo is arguably the only client that ever gained any foothold to speak of on Parrot, and Rakudo can't exploit Parrot's threading model, then Parrot effectively doesn't have super-awesome threading.
19:21 kjs_ left
skids Is it that Rakudo "can't" or that it "hasn't"? 19:22
masak the latter, surely.
19:22 booly-yam-4366 left
lizmat jnthn: I'm trying to establish whether the ^100 -> while $i optimization is being done 19:22
but apparently it doesn't even enter "optimize_for_range" in Optimizer.nqp ?
pmurias masak: pir is also a parrot client 19:23
FROGGS_ my parrot threading tests in nqp resulted in a single cpu core use... because lexical access is only possible from the main thread
rurban because nqp uses its own broken lexpad implementation
FROGGS_ you can use more cores, but that only works for operations that can be done without lexicals (and adding two bigints uses the lexical &infix:<+>) 19:24
masak pmurias: from my perspective, PIR seems to me to be pessimal for both human authoring and code generation. it might be the single biggest concrete mistake that got perpetuated as a component of Parrot.
pmurias: and no, I do not consider it a client. I consider it technical debt.
timotimo lizmat: --target=ast and --target=optimize should be a good entry point to figuring out if/whether an optimization is happening
though i'd say a tool that can diff trees like that more "prettily" or "domain-knowledgingly" would be helpful 19:25
pmurias rurban: have you looked at STM (software transactional memory)?
19:27 booly-yam-4366 joined
rurban we looked into it some decades ago, yes. 19:27
lizmat timotimo: if I do an "nqp::say" in optimize_for_range, it should show up when optimizing, no ?
rurban it's documented in the threads pdd somewhere
timotimo it should, aye 19:28
lizmat it doesn't :-(
timotimo but that requires a full core setting recompile, which takes about 5 minutes. oh wait, we have 2015 now :)
lizmat yeah, <1minute now
$ time perl6 -e 'for ^5000000 { 1 }' # with optimizing: real0m3.521s 19:29
$ time perl6 --optimize=off -e 'for ^5000000 { 1 }' # without: real0m4.739s 19:30
so it *is* optimizing, but apparently not where I think it is :-(
pmurias rurban: it seems what you are proposing is a subset of STM 19:31
rurban: if I remember correctly it adresses one of the best cases for STM 19:34
it should have excelent peformance
rurban pmurias: I'm proposing nothing. parrot has excellent threads and nqp does not use it. Not my problem really. 19:35
all the other things are better in moar, but I will catch up sooner or later
pmurias rurban: do you think parrot threads are enough for all uses? 19:37
lizmat timotimo: ok, it looks to me that if I disable SPESH, there is no difference between using --profile or not 19:38
*--optimize=off
19:38 regreg left, Sqirrel left
lizmat which leads me to think that the difference we see is purely caused by SPESH/JIT 19:38
19:40 brrt joined 19:44 Mso150 joined 19:48 davercc`` left 19:49 adu left, Bucciarati left 19:51 rindolf left
masak I think my personal conclusion from this discussion is that Parrot people and Perl 6 people are no longer actively communicating across the two communities. nor does there seem to exist much interest in such communication. for some time now -- maybe two-three years. 19:51
FROGGS_ that meets my perception 19:54
masak if Parrot's only edge over other established VMs out there is threading, and Rakudo doesn't use it and hasn't since it was built in 2012, then... it's because the communities are no longer in an ongoing relationship.
Perl 6/Rakudo is busy courting MoarVM and (to a lesser extent) the JVM.
Parrot is busy doing who-knows-what. 19:55
skids The impression I get is that there is a hole where an nqp/parrot-focused dev needs to be.
19:55 Bucciarati joined
FROGGS_ solving existing github issues 19:55
19:55 Sqirrel joined 19:56 LLamaRider joined
masak read that as "rearranging deck chairs on the Titanic" 19:57
FROGGS_ one problem (for me) is the development turn-around for perl6-p is quite huge... it takes a while to recompile nqp-p + rakudo+p, not to mention parrot
I mean, you can recompile moarvm without needing to touch nqp or rakudo... you can play with the .java files in nqp and you can test perl6-j just like that... 19:59
nothing like that works for parrot/nqp-p
moritz rurban, masak: iirc somebody did try to use parrot threads with rakudo (maybe FROGGS_++?), and failed, because access to lexicals from threads caused segfaults, or something
no idea where that idea went 20:00
FROGGS_ moritz: yes, and when it did not segfault it was limited to one core...
masak moritz: ok, that's like half of a data point, then. I remain unimpressed by Parrot's threads.
FROGGS_ and since nobody helped me (from neither side), I left it at that state - sleeping in a github issue
moritz masak: not trying to prove anything, except that there *was* at least one attempt to use parrot threads from rakudo 20:01
masak FROGGS_: was the github issue ever resolved?
moritz: one attempt, which didn't prove Parrot's threads' superiority, but dug up some segfaults instead. 20:02
moritz and the "nqp uses its own broken lexpads" argument is moot, because parrot's lexpads didn't provide the necessary functionality, so there was a *need* to reinvent them
masak: yes, we're in violent agreement here
20:02 brrt left
pmurias the moarvm and parrot threads would have different apis 20:03
even if Rakudo used the parrot ones
moritz pmurias: that's fine; the JVM threads have another different API, presumably
pmurias moritz: why? 20:04
moritz pmurias: why what?
skids Well, that depends on exactly why nqp lexpads were broken. Is it just bitrot?
FROGGS_ masak: no, not resolved: github.com/parrot/parrot/issues/889 20:05
pmurias moritz: why does Rakudo on JVM expose threading differently than on MoarVM?
dalek kudo/nom: b76f601 | lizmat++ | src/Perl6/Optimizer.nqp:
Fix broken for ^1000 -> while $i < 1000 opt
moritz skids: I see no evidence that they are, except rurban's comment about them being broken in context of parrot threads
pmurias: it doesn't. JVM and MoarVM presumably have different thread APIs
pmurias: but nqp + rakudo unify them 20:06
pmurias but it won't be possible on Parrot threads
FROGGS_ pmurias: why not?
moritz pmurias: which is why I meant it's fine that parrot has another different API - as long as it's powerful enough for rakudo + nqp to unify it with the others
masak FROGGS_: that github issue is like a parody of the broken communication between Perl 6 and Parrot.
FROGGS_ masak: :o)
moritz m: say nqp::null().perl 20:07
camelia rakudo-moar 20d5ef: OUTPUT«Cannot call method 'perl' on a null object␤ in block <unit> at /tmp/uYMkPpdZBq:1␤␤»
pmurias FROGGS_: I can't imagine how to unify their api
20:07 rmgk left
masak "If we can't compute in parallel when lexicals are involved, then I'm not sure how usable it is." 20:07
pmurias FROGGS_: they parrot threads are a stripped down STM
masak when *lexicals* are involved. 20:08
well, good thing not much of my everyday code uses *lexicals*.
dalek kudo/nom: 6279e12 | moritz++ | t/spectest.data:
Run S16-io/print.t
masak rurban: yeah, it's not looking good for Parrot threads. :( 20:09
20:09 kjs_ joined
moritz masak: be careful with sarcasm on the Internet 20:09
pmurias is not sure what mocking rurban is trying to accomplish
20:10 rmgk joined
FROGGS_ masak: well, you should in general try to avoid accessing lexicals, or rather avoid to share variables between threads... 20:10
20:10 zakharyas joined
FROGGS_ masak: the problem is more that our routines are lexicals too 20:10
dalek c: 397e0c0 | moritz++ | lib/Type/Metamodel/PrivateMethodContainer.pod:
Document Metamodel::PrivateMethodContainer
 20:11
20:11 darutoko left
moritz let's ask a different question: what advantages does rakudo-parrot offer over rakudo-moar *right now*? 20:12
lizmat www.youtube.com/watch?v=Iu7vySQbgXI
FROGGS_ lizmat: :o) 20:13
skids So OOC, if OSthread M owns X, OSthread A tells M to write X, OSthread B tells M to write X, and thread A reads X, how does A know to use B's value for X and desist using the value it wrote, under the "lock-free" model? 20:16
20:16 brrt joined 20:23 denis_boyun_ joined
FROGGS_ skids: good question 20:26
lizmat skids: how does OSthread M "own" X ??? 20:28
skids Just on a surface level the parrot model has advantages with 2-thread sharing, and possibly a downside of more bookeeping vs locking.
lizmat skids: or was that about the parrot model?
skids lizmat: according to the parrot model only one OS thread has write access to any particular var.
lizmat ah, ok
20:34 denis_boyun_ left, adu joined 20:36 denis_boyun_ joined
masak pmurias: I would be delighted if Parrot threads were useful enough to be measurable against anything else. unfortunately, they seem several steps away from that. this makes me sad, especially if it's true as several people claim that threads is the only remaining edge Parrot has on other VMs. 20:38
pmurias masak: they are usable from winxed 20:40
masak moritz: thanks for the warning about sarcasm. that was very helpful. 20:43
pmurias: does winxed have a community where I could discuss their experience of the Parrot threads? 20:44
20:44 adu left
pmurias masak: you can ask the stm community how they feel about not using any reads in their transactions 20:45
masak: parrot threads are a stm subset
so we could sort of fake using them on top of pugs 20:46
20:47 anaeem1_ left 20:52 denis_boyun_ left, LLamaRider left
FROGGS_ huh "Streaming decode NYI for encoding 4" 20:54
I tried to: "example/utf-16-2.xml".IO.slurp( :enc<utf-16> )
:o( 20:55
21:00 Mso150_o joined 21:04 Mso150 left 21:09 jakesyl joined 21:11 Mso150_o left, zakharyas left, Mso150 joined
masak I think the consensus in the Perl 6 universe is that (a) STM is an interesting solution to apply in some cases, a bit like Prolog is, but it's not universal, and (b) where the composability of STM really shines is with fully immutable runtime models, which figure heavily in languages like Haskell, but basically not at all in Perl 6. 21:12
at least that's what I've come to understand. I picked up (a) during the Pugs days, and (b) in the past few years.
21:13 KCL_ left 21:14 Ovid joined, Ovid is now known as Guest12460 21:15 Guest12460 is now known as CurtisOvidPoe
CurtisOvidPoe m: class Point { has $.x; has $.y; multi method set(:$x = $.x) { $!x = $x }; multi method set(:$y = $.y) { $!y = $y } }; my $p = Point.new( x => 5, y => 2 ); $p.set( y => -3 ); say $p.y; 21:15
camelia rakudo-moar b76f60: OUTPUT«2␤»
CurtisOvidPoe I’ve created a point object, trying to use named multi methods and tried to set y to -3, but it still says that it’s 2. 21:16
masak CurtisOvidPoe: named params are optional by default.
CurtisOvidPoe: try adding !
also, you could get away with writing just `:$!x!`, etc 21:17
CurtisOvidPoe masak: I don’t understand that comment.
If my code is wrong, how is it wrong?
masak m: class Point { has $.x; has $.y; multi method set(:$x! = $.x) { $!x = $x }; multi method set(:$y! = $.y) { $!y = $y } }; my $p = Point.new( x => 5, y => 2 ); $p.set( y => -3 ); say $p.y
camelia rakudo-moar b76f60: OUTPUT«===SORRY!===␤Cannot put default on required parameter $x␤at /tmp/xzGarl9Sr8:1␤------> .x; has $.y; multi method set(:$x! = $.x⏏) { $!x = $x }; multi method set(:$y! = ␤Cannot put default on required parameter …»
masak m: class Point { has $.x; has $.y; multi method set(:$x!) { $!x = $x }; multi method set(:$y!) { $!y = $y } }; my $p = Point.new( x => 5, y => 2 ); $p.set( y => -3 ); say $p.y
camelia rakudo-moar b76f60: OUTPUT«-3␤»
masak there. 21:18
CurtisOvidPoe: hope my comment makes sense now.
m: class Point { has $.x; has $.y; multi method set(:$!x!) {}; multi method set(:$!y!) {} }; my $p = Point.new( x => 5, y => 2 ); $p.set( y => -3 ); say $p.y
camelia rakudo-moar b76f60: OUTPUT«-3␤»
masak and that was my other comment.
FROGGS_ which works but might be kinda awkward :o)
masak the `:$!x!` pill looks kind confusing at first, yes. 21:19
jnthn FROGGS_: It means what it says; there's an NYI.
masak one has to mentally disambiguate the first `!` (private attr) from the second (mandatory parameter).
FROGGS_ jnthn: yes, but... 21:20
CurtisOvidPoe masak: I agree that my code doesn’t make sense, but nonetheless, it doesn’t appear to be dispatching properly.
FROGGS_ I don't want to ó.ò
moritz or
jnthn FROGGS_: Then write a MoarVM ticket and I'll look at it in the next days.
moritz method set(:$!x, :$!y) { }
masak CurtisOvidPoe: unless it's clear already, you also fell prey to several matching candidates picking the first one in text order.
FROGGS_ jnthn: no, I won't distract you from doing 6pe :o)
masak CurtisOvidPoe: also, what moritz++ just said.
moritz m: class Point { has ($.x, $.y); method set(:$!x, :$!y) { } }; my $p = Point.new( x => 5, y => 2; $p.set( y => -3 ); say $p.y 21:21
camelia rakudo-moar b76f60: OUTPUT«===SORRY!=== Error while compiling /tmp/4O83QGu8LO␤Unable to parse expression in argument list; couldn't find final ')' ␤at /tmp/4O83QGu8LO:1␤------> > 5, y => 2; $p.set( y => -3 ); say $p.y⏏<EOL>␤ expectin…»
CurtisOvidPoe masak: actually, I didn’t fall prey to several candidates picking the first one because *neither* value is changed.
m: class Point { has $.x; has $.y; multi method set(:$x = $.x) { $!x = $x }; multi method set(:$y = $.y) { $!y = $y } }; my $p = Point.new( x => 5, y => 2); $p.set( y => -3 ); say $p.y;say $p.x;
camelia rakudo-moar b76f60: OUTPUT«2␤5␤»
masak CurtisOvidPoe: that's because it dispatched to the first one.
CurtisOvidPoe: which doesn't care about $.y
CurtisOvidPoe No, as you can see from the above, $p.x isn’t changed, either. 21:22
masak I didn't say it was.
I just said it dispatched to the first one.
which doesn't care about $.y, and defaults $x to $.x
so you've successfully changed $.x to $.x
FROGGS_ CurtisOvidPoe: is this about chowing multi methods or about setting attributes? 21:23
CurtisOvidPoe So that means that multi foo(:$x) can take multiple named arguments and ignores those which do not show up?
FROGGS_ CurtisOvidPoe: yes
CurtisOvidPoe (ignores those which don’t match $x)
masak CurtisOvidPoe: yes, that's what "not mandatory" means.
moritz m: class Point { has ($.x, $.y); method set(:$!x, :$!y) { } }; my $p = Point.new( x => 5, y => 2); $p.set( y => -3 ); say $p.y
camelia rakudo-moar b76f60: OUTPUT«-3␤»
jnthn To be clear, the 3 things going on here are 1) Methods happily accept any unexpected named parameters for reasons of interface consistency, 2) named parameters are optional by default, 3) named parameters don't affect candidate sorting, but required ones can serve as a constraint.
FROGGS_ there is an implicit *%_ on methods 21:24
masak jnthn++
21:24 denis_boyun_ joined
CurtisOvidPoe Thanks, jnthn! I expect that this ties in with the multi method ambiguity issue and is going to cause confusion. 21:24
moritz design.perl6.org/S12.html#Interface_Consistency
masak I think we all hate interface consistency with some degree of passion -- but no-one seems keen on arguing against it either, as it makes some amount of sense.
CurtisOvidPoe In short, $p.set( y => -2 ) being a no-op is very confusing.
ab5tract FROGGS_: that's a useful note to make, but not entirely fitting with CurtisOvidPoe's talk which aims at mere mortals 21:25
21:25 rurban left
ab5tract CurtisOvidPoe: sure, but then again so is the mechanism through which you have set it up 21:25
masak CurtisOvidPoe: your original code is not very idiomatic.
CurtisOvidPoe Agreed!
masak CurtisOvidPoe: so while I agree that it being a no-op is very confusing, it's not a path that I would expect people to follow and not a trap I would expect people to land in. 21:26
FROGGS_ m: class Point { has $.x is rw; has $.y is rw }; my $p = Point.new( x => 5, y => 2 ); $p.y = -3; say $p.y;
camelia rakudo-moar b76f60: OUTPUT«-3␤»
FROGGS_ that's idiomatic but does not show MMD of course
ab5tract MMD is not necessary where named parameters are concerned
masak CurtisOvidPoe: moritz++' suggestion is more idiomatic.
and shorter.
CurtisOvidPoe Confised the case of Perl 5: new programmers usually don’t write “idiomatic” Perl code.
masak *nod*
CurtisOvidPoe moritz’s suggestion is pretty much what I already have in my talk :)
masak \o/ 21:27
jnthn Also, why a set method at all?
skids When I was first learning P6 that (named parm dispatch) did manage to snag me once. It's just one of those things to be pointed out to beginners
ab5tract jnthn: excellent point
jnthn "is rw" is the (not hard to understand) way to get a public accessor
FROGGS_ is what I showed, aye
CurtisOvidPoe jnthn: I don’t have a set() method the final Point class I show. I was using one as an example of “you really don’t need this after all” 21:28
And I throw a subset on $.x and $.y to make it clear we can safely expose them.
FROGGS_ m: class Point { has $.x is rw; has $.y is rw }; my $p = Point.new( :5x, :2y ); $p.y = -3; say $p # :o)
camelia rakudo-moar b76f60: OUTPUT«Point.new(x => 5, y => -3)␤»
ab5tract CurtisOvidPoe: for MMD (and doing more than necessary) you could create PosX and PosY classes 21:29
FROGGS_ (or roles)
ab5tract .... :5x ?
jnthn CurtisOvidPoe: Ah, OK, then fair enough.
ab5tract do we really need more ways to frame a pair?
FROGGS_ ab5tract: is meant when you have units.... $date.add( :24hours )
ab5tract ah
ok 21:30
i guess :.
:/
jnthn Well, the reason we got it is because :2nd, :3rd, etc. as a regex modifier.
FROGGS_ jnthn: but it is awesome for datetime math :o)
jnthn FROGGS_: Opinions vary ;)
FROGGS_ jnthn: that's your opinion :P 21:31
jnthn implemented it and remembers the mixed reactions of "way cool" and "omfg WHY" :)
FROGGS_ yeah
"Every DWIM comes with a WAT"
ab5tract i assume that is only ever possible with numeric values? 21:32
pmurias masak: I haven't used STM at all so I don't have any opinions about it, I just read the rurban blog post and it reminded me of STM
FROGGS_ ab5tract: aye
masak pmurias: url?
ab5tract i was reading the backlog on the train 21:33
pmurias masak: the advent one
pmurias hasn't figured out how to copy & paste stuff into an xterm without a mouse
ab5tract a data point i had was at YAPC::EU 2012, after the guy who implemented it (name escapes) how well it was working with P6
masak ok, I might be able to find it. hold on.
pmurias (it's in the irc log)
ab5tract he simply looked at me like i was insane and replied "don't care" 21:34
masak got it. perl6advent.wordpress.com/2012/12/...t-threads/
FROGGS_ ab5tract: what are you talking about?
masak ab5tract: maybe "Nat Tucks"? 21:35
ab5tract FROGGS_: parrot threads
FROGGS_ ab5tract: ahh
masak ab5tract: that's what the p6advent post says.
geekosaur pmurias, blog.ssokolow.com/archives/2011/05/...rms-paste/ ?
FROGGS_ nine++ implemented it and gave that talk
ab5tract^^
masak or it was nine, yes.
ab5tract ok
wish that didn't make things even more confusing 21:36
CurtisOvidPoe m: say .3.^methods
camelia rakudo-moar 6279e1: OUTPUT«Rat FatRat new nude Num floor ceiling Int Bridge base succ pred norm narrow abs sign conj sqrt rand sin asin cos acos tan atan atan2 sec asec cosec acosec cotan acotan sinh asinh cosh acosh tanh atanh sech asech cosech acosech cotanh acotanh round unpolar …»
CurtisOvidPoe m: say .3.HOW.methods
camelia rakudo-moar 6279e1: OUTPUT«Too few positionals passed; expected 2 arguments but got 1␤ in any methods at src/gen/m-Metamodel.nqp:453␤ in block <unit> at /tmp/dI37LKjaUr:1␤␤»
21:36 kurahaupo joined
pmurias masak: "wait" from the blog post is just atomic (thread_has_ended_flag == 1) {} 21:37
CurtisOvidPoe ^ and HOW aren’t the same thing?
jnthn CurtisOvidPoe: No
masak m: say .3.HOW.methods(.3)
camelia rakudo-moar 6279e1: OUTPUT«Rat FatRat new nude Num floor ceiling Int Bridge base succ pred norm narrow abs sign conj sqrt rand sin asin cos acos tan atan atan2 sec asec cosec acosec cotan acotan sinh asinh cosh acosh tanh atanh sech asech cosech acosech cotanh acotanh round unpolar …»
jnthn CurtisOvidPoe: Meta-methods take the object being operated on as a first argument
pmurias masak: and each parrot thread starts a transaction for each write
jnthn a.^foo(b) is a.HOW.foo(a, b) but without re-evaluating a
vendethiel "Every DWIM comes with a WAT" <- this is so well said, FROGGS_
CurtisOvidPoe jnthn: cheers!
pmurias masak: and the transactions don't contain any reads so they always succeed 21:38
moritz guesses it's time to write perlmop.pod instead of j API documentation for the MOP 21:39
jnthn moritz: I suspect the latter is more useful with the former :)
masak I think both are useful. it's just, I would look for the former first. 21:40
ab5tract agrees 21:41
jnthn Right, that's a better way of saying what I was trying to :)
moritz it's also that writing the API documentation gives man anoverview over the available MOP functionality
CurtisOvidPoe Night all.
jnthn 'night
masak jnthn: oh you said *with* the former. I read *than* the former. sorry :)
'night, Ovid
21:41 denis_boyun_ left
jnthn masak: I wondered why you were seeming disagreeable :P 21:42
jnthn goes hunting a basic block eliminator bug...
masak jnthn: I seem to do that a lot tonight ;)
rawr!
FROGGS_ *g*
moritz basic blog alligator bug!
masak seen on Twitter: "Someone once said I was fun and easy-going. I laughed so hard I accidantally ate them." 21:43
ab5tract moritz: my miscomprehension matrix is missing, can i have it back please? :)
pmurias masak: the more I think about the parrot threads the more crazy they seem
masak pmurias: whee!
moritz ab5tract: sorry, I already diagnolized it
masak pmurias: crazy *awesome*, right?
"The solution for these problems implemented in hybrid threads is to sidestep them altogether by disallowing write access to shared variables." -- you what? 21:47
that's not a solution. that's avoiding the problem. 21:48
vendethiel I strongly believe that's a solution :P
moritz
.oO( you communicate between threads over avian carriers)
vendethiel to a different problem, through
which happens to fix gracefully both...
masak vendethiel: heh :)
if I am reading this post right, it's saying that if I have a program with an array `@results`, then I am allowed to fork 100 threads to compute results in parallel -- as long as I don't push the results to the `@results` array from any of the threads. 21:50
dalek c: bebee61 | moritz++ | lib/Type/Submethod.pod:
document type Submethod
masak "In other words, all data is owned by the thread creating it and only the owner may write to it. Other threads have only read access." -- yeah, I'm sorry, that's not awesome. 21:51
that sucks.
21:52 raiph joined
lizmat is looking at a segfault and decides to sleep on it 21:52
ab5tract masak: your disagreeable threads are showing quite plainly today ;)
masak ab5tract: I'll bet.
I take full responsibility. I clearly should be doing something fun and constructive instead. 21:53
I don't even care that much about Parrot.
ab5tract that said, it was really interesting to see the linear performance increases from the (admittedly arbitrary) mandelbrot test
jnthn lizmat: Hah, you've just made me wnat a pillow with SEGV written on it :P 21:54
ab5tract and in practice, i haven't gotten any shared array code working in moar threads either
which is what i understand we need Supply.act for 21:55
masak ab5tract: ooh, I should totally try that.
std: our enum day of int <Sun Mon Tue Wed>
camelia std f9b7f55: OUTPUT«ok 00:00 135m␤»
masak m: our enum day of int <Sun Mon Tue Wed>
camelia rakudo-moar 6279e1: OUTPUT«===SORRY!=== Error while compiling /tmp/QEVPGELusO␤Method 'set_of' not found for invocant of class 'Perl6::Metamodel::EnumHOW'␤at /tmp/QEVPGELusO:1␤------> ␤» 21:56
masak NYI?
ab5tract cannot tell if masak is being sarcastic
masak ab5tract: I'm not.
ab5tract: trying it now.
ab5tract i have about zero experience with concurrent programming 21:57
jnthn If you're going to have multiple threads producing results to a common consumption point, you either want a Channel or a Supply, or sometimes just to solve the problem more functionally. :)
FROGGS_ jnthn: you can get such a pillow if that helps :o)
jnthn FROGGS_: I'm not sure it'd *help*... :) 21:58
FROGGS_ *g*
masak m: my @results; for (1..20).pick(*) -> $n { $*SCHEDULER.cue: { @results.push($n) } }; say @results.perl; say @results.elems
camelia rakudo-moar 6279e1: OUTPUT«Could not spawn thread: errorcode -1␤»
masak hm. this works locally.
jnthn Odd
ab5tract but when i read/heard that constraint, it didn't seem like a big deal at all because my limited uynderstanding was "yeah, you can't push onto a common array from threads anyway"
masak maybe some restriction put on camelia?
jnthn Maybe 21:59
masak anyway, I got twenty elements at home.
hobbs masak: yeah, a process/thread rlimit wouldn't be surprising for an IRC bot :)
ab5tract at least without some sort of arbiter, as jnthn points out with Channel or Supply
masak if this way is dangerous, then I seem to have been lucky.
jnthn masak: You shouldn't get it to SEGV, but it's NOT a good idea
masak I'd be happy to learn how to do the above code with a Channel or Supply.
ab5tract masak: why not use start? 22:00
pmurias masak: you are free to fork off 100 threads which modify @results in parallel
masak: their modifications will trickle down to your thread in random order 22:01
jnthn m: my @results = await do for (1..20).pick(*) -> $n { start { $n } }; say @results;
camelia rakudo-moar 6279e1: OUTPUT«Could not spawn thread: errorcode -1␤»
jnthn Yeah, must be some restriction.
pmurias masak: and you can call wait for all the modifications to trickle down
jnthn m: Thread.start({ sleep 1 })
camelia ( no output )
jnthn m: Thread.start({ sleep 1 }) xx 10
camelia ( no output )
jnthn m: Thread.start({ sleep 1 }) xx 20
camelia rakudo-moar 6279e1: OUTPUT«Could not spawn thread: errorcode -1␤»
jnthn m: Thread.start({ sleep 1 }) xx 16
camelia rakudo-moar 6279e1: OUTPUT«Could not spawn thread: errorcode -1␤»
ab5tract poor camelia 22:02
jnthn 16 is the maximum thread poo size by the dfault, iirc
ab5tract :)
jnthn uh, pool
FROGGS_ hehe
+1
jnthn As in, Rakudo won't spawn more than 16
But Camelia won't handle that it seems
masak pmurias: ok, that sounds saner than how I understood it.
ab5tract jnthn: is maximum thread pool size a compile time or run time switch for rakudo? 22:03
jnthn ab5tract: If you change $*SCHEDULER you can control it at runtime.
ab5tract ok 22:04
jnthn ab5tract: In the future it will look at the hardware you have and factor in throughput etc.
ab5tract sweet!
the concurrency bits are one thing i would really love to see both ironed out *and* fully documented for 6.0 22:05
masak +1
ab5tract jnthn: i can still make moar segfault by calling say
jnthn masak: A better example of the restriction: you have a monitor or actor (using OO::Monitor or OO::Actor modules I hacked up). Under the Moar/JVM model you have in Perl 6, any thread can modify them, but an ordering (one at a time) is imposed on the method calls, giving safety.
masak jnthn: cool. 22:06
jnthn masak: So your concurrent web app can process requests that work against those objects over a bunch of threads, and whichever thread wants can modify it - but only one at a time.
masak jnthn: are there any guarantees given about reads not seeing intermediate states?
jnthn masak: As I understand the Parrot model, all changes to an object have to go to the thread that creates it, meaning if that thread is busy doing something else, the first has to block. 22:07
masak: (intermediate states) monitors/actors are object with state, and you can only be inside a method call on a particular instance in one thread at a time. 22:08
masak m: my enum CoinFace <Heads Tails>; say CoinFace.mapping # NYI?
camelia rakudo-moar 6279e1: OUTPUT«No such method 'mapping' for invocant of type 'CoinFace'␤ in block <unit> at /tmp/5dauX1OCWv:1␤␤»
masak jnthn: ah; got it.
jnthn masak: So, provided you don't do something crazy with your OO design, no :)
masak: Ain't that method called .enums 22:09
masak oh, possible. I'm reading an old spec commit.
jnthn m: my enum CoinFace <Heads Tails>; say CoinFace.enums
camelia rakudo-moar 6279e1: OUTPUT«Heads => 0, Tails => 1␤»
masak m: my enum CoinFace <Heads Tails>; say CoinFace.enums.^name 22:10
camelia rakudo-moar 6279e1: OUTPUT«Hash␤»
masak heh -- the old spec commit had the type as "PairValSet" :P
jnthn Such enterprisey :P
m: say &die.WHAT 22:12
camelia rakudo-moar 6279e1: OUTPUT«(Sub+{<anon>})␤»
jnthn m: say &die.^roles
camelia rakudo-moar 6279e1: OUTPUT«(<anon>) (Callable)␤»
jnthn m: say &die ~~ Callable
camelia rakudo-moar 6279e1: OUTPUT«False␤»
jnthn m: say nqp::istype(&die, Callable)
camelia rakudo-moar 6279e1: OUTPUT«0␤»
masak guesses jnthn is fixing some RT ticket or other 22:13
jnthn m: &die.^compose; say nqp::istype(&die, Callable)
camelia rakudo-moar 6279e1: OUTPUT«1␤»
jnthn masak: Well, it probably exists.
FROGGS_ m: say nqp::istype(&fail, Callable)
camelia rakudo-moar 6279e1: OUTPUT«1␤»
jnthn masak: It's just that what once was an obscure bug that showed up in the case of some setting built-ins is now a bug causing spectest fails in my branch that caches mixins. 22:14
m: class A { ... }; class A does Rational { ... }; class A { method omg() { } }; say A.omg; say A.^roles; say A ~~ Rational 22:15
camelia rakudo-moar 6279e1: OUTPUT«===SORRY!===␤None of the parametric role variants for 'Rational' matched the arguments supplied.␤Cannot call ''; none of these signatures match:␤:(::$?CLASS ::::?CLASS $, ::NuT $, ::DeT $)␤»
jnthn oh... :)
m: class A { ... }; class A does Positional { ... }; class A { method omg() { } }; say A.omg; say A.^roles; say A ~~ Rational
camelia rakudo-moar 6279e1: OUTPUT«Nil␤(Positional)␤False␤»
jnthn m: class A { ... }; class A does Positional { ... }; class A { method omg() { } }; say A.omg; say A.^roles; say A ~~ Positional
camelia rakudo-moar 6279e1: OUTPUT«Nil␤(Positional)␤True␤»
jnthn OK, good...it won't be a horrid fix :)
22:19 Mso150 left 22:22 mvuets left, kjs_ left
jnthn Aww, nah, it will be. 22:23
Guess that can be my problem to sleep on tonight :) 22:24
22:25 leont joined
masak embroiders "horrid fix" on a pillow for jnthn 22:25
jnthn m: package Foo { }; sub x(Foo) { } 22:26
camelia rakudo-moar 6279e1: OUTPUT«===SORRY!=== Error while compiling /tmp/oSrLMcGBow␤Foo cannot be used as a nominal type on a parameter␤at /tmp/oSrLMcGBow:1␤------> package Foo { }; sub x(Foo⏏) { }␤»
jnthn m: package Foo { }; my Foo $x = 42;
camelia rakudo-moar 6279e1: OUTPUT«Type check failed in assignment to '$x'; expected 'Foo' but got 'Int'␤ in block <unit> at /tmp/5PVuLGRbHg:1␤␤»
22:40 tgt left
dalek kudo/6pe-mop: e6a6677 | jnthn++ | src/Perl6/Actions.nqp:
Only allow appropriate types on variables.
 22:42
22:42 FROGGS_ left 22:43 espadrine left
timotimo lizmat++ # fixing the for -> while opt 22:53
masak I was thinking earlier today -- would it be feasible/worthwhile to have a special test suite for optimizations? maybe something that mocked certain parts of the compiler stack to be able to check whether an optimization triggered? 22:58
23:05 beastd left
tadzik hello hello! 23:06
yoleaux 17 Jan 2015 19:36Z <raydiak> tadzik: noticed I can't "rakudobrew switch moar-2014.12" :) github.com/tadzik/rakudobrew/pull/25
tadzik oh :)
raydiak++ colomon
colomon++
pmurias masak: it seems like something that would be worthwhile, especially if we introduce better and more complex optimalisations 23:09
masak: the hard part is how to test for correctness without hardcoding resulting qast? 23:10
masak pmurias: that's what the mocks could do. there could be a mock for the `for --> while` optimization; the test would check that it set a bit somewhere, indicating "the optimization triggered". 23:11
23:18 telex left 23:20 telex joined 23:24 spider-mario left 23:25 Ben_Goldberg left, raiph left
pmurias masak: that sort of makes hardcoding qast input and output seem a good option :/ 23:25
23:25 Ben_Goldberg joined
timotimo i wonder how much faster spesh_diff would run if i turned it from a line-by-line thing into a grammar instead that does a whole bunch of work in the actions 23:25
that could also very well make the source code simpler, as i don't have to keep an "accumulator" object around that gets stashed into a storage hash and replaced by a new empty one at a somewhat arbitrary looking point 23:26
masak pmurias: I don't see why you think so. hardcoding qast output seems brittle, if what we care about is that an optimization is done for a certain input.
timotimo it's also very hard to compare 1:1 qast outputs 23:27
as soon as something else changes, your matcher for "success of optimization" vs "failure to optimize" must be looked at again
jnthn I think that already each optimization is separated out into a method that does the transform
So there's analysis methods, and then if they decide to do something they call a transform
23:28 Ben_Goldberg left
jnthn So if you can intercept those calls (like by subclassing) and record them, you have a way to test it. 23:28
23:28 Ben_Goldberg joined 23:30 KCL_ joined 23:31 Ben_Goldberg is now known as BenGoldberg
jnthn Time for some rest...'night 23:32
masak 'night, jnthn
23:33 brrt left
BenGoldberg On the subject of creating a test for whether an optimization is being done, how about compiling with optimizations turned off, then again with them turned on, and see if there's *any* differences between the qasts? 23:33
masak <timotimo> it's also very hard to compare 1:1 qast outputs 23:34
23:35 kjs_ joined, tangentstorm joined
BenGoldberg But at least it's future proof, so if something else changes (some new optimization is done), the test can still say, "something was optimized", even though it might not know precisely which optmization triggered. 23:36
masak right, but you're assuming that any difference between two QAST output is significant, which it may not be. 23:37
23:38 kjs_ left 23:39 gfldex left 23:50 Alina-malina left 23:56 Alina-malina joined 23:58 pmurias left

Powered by Cro and the Raku® Programming Language.

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

Thank you!