»ö« | perl6.org/ | nopaste: paste.lisp.org/new/perl6 | evalbot usage: 'perl6: say 3;' or rakudo: / pugs: / std: | irclog: irc.pugscode.org/ | UTF-8 is our friend!
Set by moderator on 7 October 2009.
quietfanatic Yeah. It ought to find the right file if nobody's chdir'd 00:00
Tene unless it's in their $PATH ;) 00:01
quietfanatic Oh. 00:02
Of course, I could examine %ENV{PATH}... 00:03
00:03 Limbic_Region joined
quietfanatic ...but that may just be the beginning of a wild goose chase. 00:03
Tene look at Perl's FindBin library for ideas. 00:04
00:07 tylerni7 joined
quietfanatic Another thing: Is there a way to get a file's modification time? 00:12
$file ~~ :M does not work.
I conclude I shall use `stat`
00:16 r0bby joined 00:26 charsbar joined 00:28 envi^office joined
ShaneC does rakudo implement exceptions at all yet? 00:38
Tene ShaneC: Yes, but not extensively.
ShaneC: what do you want to do with them? 00:39
ShaneC just basic try/catch
Tene Sure. Just put a CATCH {...} block in another block. 00:41
00:43 crythias joined 01:16 ihrd joined 01:17 orafu joined 01:18 SmokeMachine joined 01:19 KyleHa joined 01:25 xenoterracide joined 01:35 hello joined 01:39 msmatsko joined 01:46 ihrd joined 01:55 ihrd left
diakopter first post! 02:06
snap out of it
o wait. 02:07
diakopter hides
PerlJam hiding doesn't do much good if you keep making noise 02:08
diakopter heh
sjohnson howdy 02:14
pugs_svn r28658 | lwall++ | [STD] properly check dups inside individual sigs of a multisig 02:15
r28658 | without interering across different sigs, or with only-ness in the body.
PerlJam "interering"? 02:17
:)
diakopter am writing up notes for redoing sprixel's object/type/calling-conventions (the *right* way this time), now that the interpreter engine proves stable for any sort of control flow 02:18
PerlJam btw, diakopter++ for sprixel 02:19
diakopter thanks :) 02:23
PerlJam wonders if the book should mention the other perl6 implementations and give URLs for them 02:39
02:40 ihrd joined 02:45 Confield joined, Confield left, Confield joined 02:48 xinming_ joined 03:03 SmokeMachine joined 03:19 ihrd joined 03:23 Ryan52 left 03:25 [bjoern] left
Patterner "illegal hardware instruction perl6" 03:36
03:50 am0c joined 04:00 ihrd left 04:10 Confield left 04:27 muri joined 04:35 ihrd joined
pugs_svn r28659 | lwall++ | [Cursor] don't delete important bits of the cached settings, duh 04:50
05:12 nbrown_ joined, justatheory joined 05:33 szabgab joined 05:46 snearch joined
moritz_ good morning 06:10
TimToady guten abend 06:13
06:15 ihrd left
sjohnson （´ー｀　） 06:19
07:07 mberends joined 07:08 rfordinal joined 07:10 iblechbot joined
bAUt （　｀ー´） 07:17
07:30 mariuz joined 07:58 Grrrr joined 08:04 simcop2387 left 08:15 |Jedai| joined
pugs_svn r28660 | lwall++ | [STD] parse quasiunquotes correctly 08:18
r28660 | [STD_P5] prefix regex categories with p5 more consistently
r28661 | lwall++ | [rx.t] :Perl5 goes outside of // for now 08:20
jnthn oh morning 08:28
08:31 zloyrusskiy joined 08:33 maja joined
moritz_ \\p/ 08:35
Matt-W \\o/ 08:37
sjohnson \\ö/ 08:42
Matt-W :D 08:46
sjohnson++
08:47 rfordinal left
sjohnson （´ー｀　） 08:47
carlin Unicode art; who needs real conversation anyway? 08:50
jnthn Conversation is srsly overrated. 08:51
08:59 mzedeler joined 09:30 NorwayGeek joined
jnthn (doing stuff at compile time)++ 09:39
(doing said stuff we used to do at runtime in PIR at compile time instead but in NQP)++
09:40 riffraff joined
Matt-W jnthn++ 09:43
09:44 rgrau` joined 09:49 masak joined
masak good day, #perl6. 09:49
moritz_ \\o/
Matt-W oh hai masak 09:50
jnthn's been muttering about doing things at compile time in NQP that we used to do in runtime in PIR. I'm encouraged.
jnthn yayitsmasak 09:51
masak: timezone sync status? ;-)
jnthn is slipping back into late nights and late mornings now his jetlag is wearing off.
09:53 drbean_ joined
jnthn Matt-W: If you're interested, compare what was removed in github.com/rakudo/rakudo/commit/263...1407ff2f5c with what was added in github.com/rakudo/rakudo/commit/c83...cb063b014c 09:53
masak jnthn: today marks the first honest attempt to re-sync in a long time. result: FAIL. but still.
09:54 am0c joined
jnthn o/ am0c :-) 09:55
Matt-W jnthn: does all that stuff you removed only do what the added stuff does?
because that's immensely cool
am0c \\o/ jnthn OH HAI
jnthn Matt-W: Yes 09:57
Matt-W: Well
Matt-W: It did one thing.
That what went in does not do.
Which I'm not sure how to handle yet.
But it pretty much decides all of the stuff at compile time and emits code to build a correct signature. 09:58
Rather than figuring it out at runtime as the PIR did.
I mean, if we've got all the information... :-)
I now make it about half way through the setting, them something explodes when compiling Rat. :-/ 09:59
Matt-W cool 10:02
What's the thing it doesn't do?
10:02 simcop2387 joined
jnthn Consider: 10:02
subset Odd of Int where { $^n % 2 }; 10:03
multi foo(Odd $x) { }
multi foo(Int $x) { }
So we expect foo(3) to call the first of those, not the second.
Matt-W We do
jnthn But for that to happen, we have to take Odd, identify that it's a refinement type, pull out the thingy it's refining (Int) and stick that as the nominal type, and boot the Odd over to the constraints. 10:04
Probably.
Or at least, that's what we were doing before now.
Though out of curiosity...
Matt-W So what you were doing is transforming it into
jnthn rakudo: subset Odd of Int where { $^n % 2 };; say Odd ~~ Int 10:05
p6eval rakudo 30e2cf: OUTPUT«1␤»
jnthn Oh hmm.
Matt-W multi foo(Int $x where { $^n % 2 } )
jnthn Matt-W: Yeah.
moritz_ I think TimToady once mentioned that subset types are notionally inlined
jnthn We may *just* get away with it.
Matt-W Provided your candidate sorting on wheres is okay, it's probably fine
jnthn But I think we need to really split it up.
My problem is that at the moment, I don't have a way to know - at compile time - if what I have is a refinement type or not. 10:06
Matt-W Although it might lead to LTA error reporting
Ah
Hmm
that's a bit of a problem
moritz_ jnthn: but you can know at CHECK time, no? 10:07
jnthn The role stuff we can figure out easy enough, the fact that the user didn't write a type on something and that we need the default we can also figure out.
moritz_: Actually I think we *can* know at compile time.
moritz_ can subset types be stubbed?
jnthn Hmm. Not sure. 10:08
moritz_: The issue is more than we don't have the infrastructure at the moment to know.
Matt-W And at compile time, is there a problem if the subset type hasn't been declared yet, or is predeclaration enforced in Perl 6 for that sort of thing?
moritz_ well
you can say ::Odd
jnthn Well, there's that too, yeah...what if it's a late-bound lookup. 10:09
moritz_ which means "that's a type, I'll declare that later"
jnthn Careful.
::Odd inside a signature is a type capture.
moritz_ oh
right
jnthn Which is the case that I'm considering here.
I don't think there is a way to build a signature from a not yet declared type.
(I think I consider that a good thing...) 10:10
Matt-W WEll
it would be quite tricky, since you don't knwo what the type is
moritz_ and you can't do any sorting at compile time
Matt-W And I suppose you can't really build it incomplete, and come back later
jnthn Yeah
Matt-W Does that mean you need to be able to build them at runtime as well
jnthn So anyway, I expect by the time we hit the Odd in the signature we know what kind of type it is.
Even if it's stubbed, I highly doubt you can stub something as a class and then define it as a subset type. 10:11
I'd think std says no to those kinda tricks.
Matt-W Well that'd be a name collision wouldn't it
moritz_ std: class A { ... }; class A { } # sanity check
jnthn std: class Foo { ... }; subset Foo of Int;
p6eval std 28658: OUTPUT«[31m===[0mSORRY![31m===[0m␤Illegal redeclaration of symbol GLOBAL::<A> (from line 1) at /tmp/GsE9hZKcli line 1:␤------> [32mclass A { ... }; class A[33m⏏[31m { } # sanity check[0m␤FAILED 00:01 96m␤»
std 28658: OUTPUT«[31m===[0mSORRY![31m===[0m␤Illegal redeclaration of symbol Foo (from line 1) at /tmp/EcYNnXY1jv line 1:␤------> [32mclass Foo { ... }; subset Foo [33m⏏[31mof Int;[0m␤FAILED 00:02 96m␤»
Matt-W 'This name already exists as a class, go away'
10:11 drbean left
moritz_ it doesn't do any sstubbing at all, it seems 10:12
jnthn Oh, I thought that had been put into std.
I musta been mistaken.
moritz_ I thought so too
maybe it got b0rked by some other change 10:13
jnthn Anyway, I rather hope the kind of a type can't change, so I think with the right infrastructure we'll be able to know that at compile time.
Matt-W I don't think the kind of a type *should* be able to change
jnthn For now, I suspect I'll have to check it at the time we build the signature (check).
oh hmm 10:14
We go explodey on our Int multi(...) { ... }
Matt-W is that... a sub called multi
jnthn oh no
I shoulda really copied it
I was getting at, compiling a multi method with the "our Int" bit out front. 10:15
Matt-W ah
jnthn I suspect it may be the call to the trait handler though.
Matt-W so is it the our or the return type
ah
moritz_ I think much of that 'our Int' is bogus, especially on methods
jnthn Yeah
It's probably on the call to trait_mod:<of> 10:16
Matt-W isn't sure yet why you need 'our' methods
jnthn That probably just means I've a bug when building the sig for one of the trait_mod:<of> multi variants.
Matt-W Given what you're doing, that seems plausible 10:17
jnthn It's kinda challening to hack on some of this stuff.
Matt-W I bet
That's why I'm not doing it
jnthn Mostly because you have to compile the setting before you can run the tests. 10:18
And compiling the setting and surviving its init sequence is actually quite hard.
Matt-W Well
it's a test :)
jnthn The trait mods are probably some of the deepest use of multiple dispatch in the entire setting.
moritz_ well, compiling the setting worked, no?
Matt-W A very big, very mean test
jnthn moritz_: Yeah, it does not, it blows up at init time. 10:19
aha
moritz_ you can also try to run some basic tests with the s1 compiler
jnthn figured it.
moritz_ or did we move say/print to the setting?
jnthn moritz_: Yes, I've been playing with the s1 a lot.
moritz_: We moved say/print to the setting
moritz_ oh
jnthn moritz_: However, die ... will throw an exception with the output you desire. ;-) 10:20
moritz_ so no .t file runs without compiling the setting. Ha.
jnthn moritz_: I think we may end up though, having a PIR-based minimal "setting".
That we don't compile in at all other than to the stage 1 compiler.
Yes, correct. That's why this is a pain. :-) 10:21
colomon jnthn: does something in Rat really generate an "our 10:49
Int multi"?
jnthn colomon: Num not Int actually 10:51
line 36
colomon ah, sure
I did that.
must have copied and pasted from somewhere, as I do not usually use "our" or return types. 10:52
jnthn colomon: It's valid Perl 6, just hits a bug.
In the refactor I'm doing.
(So I need to isolate and fix the bug...)
If I didn't hit it here, I'd do so in the spectets anyway. 10:53
colomon understood, and good hunting 10:54
11:13 ihrd joined 11:29 ihrd left
jnthn Initial signs are that changing the way we construct signatures at startup has won us another 10% saving. 11:30
Matt-W yay! 11:31
jnthn++
jnthn Provided fixing up spectests doesn't cause any slow-down, it seems the last few days changes since I started on re-working signatures have taken away 20% of our startup time. 11:32
spinclad sweet! 11:33
jnthn Since we spend plenty of startup doing dispatches to set various things up, I'm hoping that as dispatch gets faster (next steps in what I'm working on over the next week or so) we'll shave yet more off. 11:34
Matt-W :) 11:35
spinclad (burning fat out of the layers of interpretation)++
11:37 ihrd joined
masak all speedups are very welcome indeed. 11:37
ihrd hi there
masak hi ihrd!
ihrd masak: hello 11:38
colomon does someone remember where pmichaud's awesome short card script lives online? I know I found it for someone else a few month's back, but I don't remember where I found it now...
rgrau` colomon: They are in one of his slides IIRC 11:41
colomon yes...
masak colomon: it can also be found in the "Perl Myths" talk on SlideShare.
probably not copyable, though. :/
colomon Ha! 11:42
carlin gist.github.com/153993
colomon that reminded me, I actually have a copy.
locally, I mean.
carlin++
ihrd roles-related question 11:43
rakudo: role R { method foo { ... } }; class C {}; (C.new does R).^methods(:local)>>.name>>.say;
p6eval rakudo 30e2cf: ( no output )
ihrd why I do not got 'foo' here?
colomon X~, that's what I was looking for. \\o/
jnthn ihrd: Ah, I've run into that one before.
masak ihrd: I agree that you should. I don't know why you don't.
jnthn ihrd: I think it's a bug.
ihrd: Suspect "does" is somehow not generating creating an updated meta-class or something. 11:44
ihrd jnthn, masak: thank you very mach :)
masak ihrd: will you submit it to RT, or shall I?
ihrd masak: yes!
masak great.
any(ihrd, masak) will submit it. :) 11:45
ihrd masak: :) I mean I like to submit this one
masak: I can`t do much for perl6 now, so I happy to do this bit 11:46
masak ihrd++
colomon rakudo: say ((1, 2, 3) X+ (5, 4, 3)).perl 11:47
p6eval rakudo 30e2cf: OUTPUT«[6, 5, 4, 7, 6, 5, 8, 7, 6]␤»
ihrd jnthn: can I use you suggestions here in the bug report? 11:48
jnthn Sure
12:06 envi^home joined 12:08 abra joined 12:09 takadonet joined
takadonet morning all 12:09
ihrd syntax 'of' NYI in Rakudo? 12:10
I mean something like '$o does R of Some'?
masak ihrd: there's R[Some] 12:11
takadonet: \\o 12:12
jnthn std: $x does Positional of Int 12:13
p6eval std 28658: OUTPUT«[31m===[0mSORRY![31m===[0m␤Two terms in a row at /tmp/2CW3O8VXuf line 1:␤------> [32m$x does Positional [33m⏏[31mof Int[0m␤ expecting any of:␤ bracketed infix␤ infix stopper␤ standard stopper␤ statement modifier loop␤ terminator␤Other potential difficulties:␤
..Variable $…
jnthn aye
std: $x does Positional[Int]
p6eval std 28658: OUTPUT«Potential difficulties:␤ Variable $x is not predeclared at /tmp/yHcZ4m20r0 line 1:␤------> [32m$x[33m⏏[31m does Positional[Int][0m␤ok 00:01 99m␤»
ihrd jnthn: I just read last part of you talk 12:15
where you say about Can of Beer
masak: yes, R[Some] works 12:16
jnthn rakudo: class Beer { }; role Can[::Contents] { }; my Can of Beer $starobrno;
p6eval rakudo 30e2cf: ( no output )
jnthn ihrd: You probably saw that.
ihrd: That syntax is valid in declarations.
But I think not just as a noun anywhere.
ihrd jhnth: but not with `does`... *confused* 12:17
jnthn ihrd: does is just an operator
ihrd: It's RHS is parsed no differently to any other operator.
erm, Its 12:18
.oO( must make less grammar mistakes )
12:22 lmc joined
jnthn rakudo: say 2 * 69.72 12:22
p6eval rakudo 30e2cf: OUTPUT«139.44␤»
ihrd jnthn: when you say 'my Pack of Can of Beer ...' you mean Pack is parametrize role with argument Can wich is parametrized role too? 12:27
masak ihrd: pravda. 12:28
masak cheats again with Cyrillic. bad masak! 12:29
ihrd masak: it is nice ;)
masak typing with Cyrillic layout is slow, because I don't know where the keys are at. :/ 12:30
правда 12:31
ihrd OMG multiple variants of Roles! roles multi dispatch, I missed this one in spec.
ha! a lot of people who use single layout think we use to keyboards 12:32
s/to/two/;
and wondering how fast we swap it 12:33
zloyrusskiy ihrd, привет 12:34
masak are roles multi by default?
ihrd: are those the same people who think that the Chinese must have HUGE keyboards?
rakudo: role Foo[Int] { method bar() { "I only do Ints" } }; role Foo[Str] { method bar() { "I only do Strs" } }; say .bar for Foo[Int].new, Foo[Str].new
p6eval rakudo 30e2cf: OUTPUT«I only do Ints␤I only do Strs␤»
jnthn masak: yes 12:37
oh noes! I haz Slovak class...I wonder if I can still even speak this language...
masak: Chinese keyboards are crazy, so many thousands of keys... 12:38
...and the fact that to fit them in such a small space, they have to be really small, and then you have a thing like a pointy chop-stick to press them, and you gotta be really accurate...
...it's amazing! 12:39
masak jnthn: and you have to push the key from the right direction to get the right tone... it's a wonder they manage it.
jnthn Yeah, I know.
jnthn bbl, slovak 12:40
12:56 [particle] joined 13:02 NorwayGeek joined 13:05 [particle] joined
PerlJam good $localtime all 13:12
ihrd Perljam: hi
moritz_ oh hai
PerlJam++ # nice preface
PerlJam It's a start.
moritz_ just learned a new word from reading a scientific paper: to epitomize. I like it. 13:13
PerlJam moritz_: btw, I have told you that I see your name every where I go? There's a car place here called Moritz that puts its name on all of the vehicles it sells, so as I'm driving around town and glance at the cars around me, I'll invariably see your name :) 13:16
ihrd jhnthn: does traits example in your talk works?
moritz_ PerlJam: I deny the existence of any conspiracy related to that. 13:18
ihrd I have problem with multi trait_mod:<is>...
rakudo: multi tarit_mod<is>() { ... }
p6eval rakudo 30e2cf: OUTPUT«Malformed routine definition at line 2, near "tarit_mod<"␤in Main (src/gen_setting.pm:146)␤»
PerlJam ihrd: install mod_speling in your brain and/or fingers :-) 13:19
ihrd Perljam: you are right, I need one
rakudo: multi trait_mod:<is> (Container $r, doc, $text) {...} 13:21
p6eval rakudo 30e2cf: OUTPUT«Malformed routine definition at line 2, near "trait_mod:"␤in Main (src/gen_setting.pm:146)␤»
ihrd rakudo: multi trait_mod:<is> () {...}
p6eval rakudo 30e2cf: ( no output )
ihrd rakudo: role doc {}; multi trait_mod:<is>(Container $c, doc, $text) { $c does doc($text) } 13:23
p6eval rakudo 30e2cf: OUTPUT«Malformed routine definition at line 2, near "trait_mod:"␤in Main (src/gen_setting.pm:146)␤»
ihrd it is exactly what I see in talk :( 13:24
masak ihrd: 'doc' is missing a sigil. 13:26
oh wait. 13:27
ihrd ::?
masak perhaps not.
ihrd this example have many confused parts for me
PerlJam ihrd: I could have sworn it worked at one point, but rakudo was very picky about the exact syntax. 13:28
ihrd how should work does role($foo)?
13:33 frederico joined
masak ihrd++ # "does" is somehow not generating creating an updated meta-class or something 13:38
ihrd I gonna look at LolDispatch, it use trait_aluxiliary:<is>... 13:41
13:49 abra joined
ihrd actually LolDispatch do not work too, with same issue 13:50
"No applicable candidates found to dispatch to for 'trait_mod:is'" 13:51
13:53 SmokeMachine joined
masak that sounds familiar. 13:54
ihrd: maybe look at the spectests, and see how it's done there?
13:56 Ananth joined
pmichaud good morning, #perl6 14:02
PerlJam morning Pm
ihrd masak: woow. it works
masak moning, pm.
ihrd: :)
ihrd example is broken, or mb, it is a little bit modern 14:03
masak it's possible there have been spec shifts in the meantime, yes. 14:04
14:10 rfordinal joined
ihrd I should sleep now, bye 14:11
14:11 ihrd left 14:12 hanekomu joined, abra joined 14:17 KyleHa joined, Psyche^ joined
jnthn back 14:19
14:19 rfordinal left
masak jnthn: so, what's on the agenda? more signature refactors? 14:19
jnthn Fixing up spectest brekage fallout from this refactor first. 14:20
After that, I'm ready to start on the new binder. 14:21
I'm really not sure how long that is going to take me to get good enough.
Should allow us to close a bunch of tickets though.
Once it's done.
(like, the binding nameds to positionals one.) 14:22
moritz_ including a very old one by diakopter++ :-)
PerlJam jnthn: stop jabbering to us and get back to coding! ;)
moritz_ that's the one I was thinking of, yes
masak ooh, binding-nameds-to-positionals! \\o/
jnthn PerlJam: I'm enjoying jabbering in a langauge I speak well for a bit, after spending the last hour and a half jabbering in one that I speak badly. :-P
masak I have a new favorite problem ticket, by the way: rt.perl.org/rt3/Ticket/Display.html?id=69438 14:23
pmichaud names-to-positionals.... what's the approach there?
moritz_ when that's done we have to go through the setting and see if the names of the parameters agree in setting and spec
masak it's always possible to work around it, but it's often a bit of extra work.
pmichaud since we don't have :callsig yet ?
masak I think I'll take a look at it myself to try to figure out what the cause is.
jnthn pmichaud: Debating that a bit still, but one of: 14:24
masak moritz_: and if the names in the spec could be improved.
moritz_ masak: aye
14:24 abra_ joined
jnthn 1) Take slurpies and process those instead of callsig for now. Easy migration path then. 14:24
14:24 justatheory joined
jnthn 2) Just do what Perl6MultiSub does to get the args. But that is actually further away from what :callsig will give us. 14:25
And also only works if we're invoked from PIR.
pmichaud I'm fine with either of those.
jnthn Actually I'm really surprised we've never hit an issue where we try to invoke a Perl 6 multi from anywhere other than PIR. We've got very lucky there.
PerlJam jnthn: or we've constructed the maze just so we don't fall into the traps :) 14:26
jnthn pmichaud: I'm still pondering exactly how I want a lot of this to look though.
pmichaud: Most specifically, when do we bind. 14:27
pmichaud jnthn: ack
jnthn pmichaud: There's a certain level of temptation to do it inside invoke.
pmichaud I think I like them separate
I'm still in favor of a "bind_params" dynop 14:28
jnthn That's the obvious way, but I'm a tad "hmm" on it. :-)
pmichaud okay.
jnthn I mean.
The problem I have is...
We really need a way to be able to potentially bind a signature 14:29
But being able to fall back out if we fail to.
And have that cheap.
I think we have to create a full on lexical context for that anyway though.
But if we don't fall back into the runloop until after we've bound, the control flow gets a bit easier to manage. 14:30
Well 14:31
Maybe it doesn't actually matter.
Anywhere can invoke the ret cont to make it as if we never called in there.
The fun is in the multiple dispatch though.
We don't want to re-do a bunch of checks.
pmichaud seems like we want a continuation 14:32
jnthn It only gets tricky in that if we are invoked in a single rather than a multi dispatch, we still need to do all the checks.
And depending on the nature of the multi-dispatch, we may either
1) Know that we can just find and be done with it or 14:33
2) Have already done a full bind because we needed to know that we could
s/find/bind/
Yes, continuations may well be the answer here.
pmichaud iwbni there's an easy way for the binding operation to know if it's single or multi dispatch 14:34
jnthn Indeed, I'm still trying to work out a clean way to do that.
pmichaud i hate to suggest it, but perhaps a flag on the sub pmc?
jnthn Since contexts are PMCs now, though...
No, *not* the sub.
Imagine a bunch of recursive calls.
Some multi, some not. 14:35
Or in a threaded context.
14:35 ruoso joined
pmichaud but wouldn't we have to complete binding into a call before we can recurse? 14:35
jnthn It's about the current invocation rather than every invocation. So I think the flag wants to go on the context PMC.
Not if something in one of the constraints calls it, which would yes be contrived. 14:36
pmichaud ah
jnthn But anyway, contexts are PMCs, we can flag them.
pmichaud sure, flagging the context works nicely also
er, wait
hmmm
flagging the caller context or the newly created context? I guess caller.
jnthn Newly created, surely?
As in, the current invocation. 14:37
pmichaud ...how do you flag it before it exists?
jnthn By the time we get to binding, it has to exist.
And if we have already started binding, it already exists.
pmichaud I'm missing a step, just a sec
jnthn Binding needs the lexpad.
pmichaud the dispatcher calls invoke on a sub
that creates a new context 14:38
when we get back from invoke, aren't we already in the new sub's context ?
jnthn Right.
Yes.
Oh wait
Which caller did you mean?
:-)
pmichaud I don't know which caller I meant
jnthn Heh, I don't know which you meant either. :-) 14:39
pmichaud I'm just wondering when/if the dispatcher gets control after calling the invoke vtable
jnthn It'd depend.
There's two cases.
pmichaud so that it could flag the newly created context
because prior to the invoke vtable, the new context doesn't exist
if the dispatcher doesn't regain control, it doesn't have an opportunity to flag the context to say "this is multidispatch"
jnthn Wait, there's three cases. Let me enumerate them.
1) A single dispatch. The multi-dispatcher hasn't been invoked at all. The binder needs to do all checks and binding work. 14:40
In this case, nothing has been flagged up at all. No flags set = do everything.
14:41 kent\\n joined
jnthn 2) A multi-dispatch where the dispatch cache on the nominal types made a decision for us. We didn't actually bind anything, but we know that we only have positional arguments and that we don't need to do any type checks on them, just bind them, because we know from the type-based cache that it's fine. 14:42
In this case, the multi-disptacher's invoke vtable method after deciding this is the candidate to invoke calls VTABLE_invoke on the chosen candidate. At this point, we get an opportunity to set a flag saying "you've got a simple positional unchecked binding to do, go do it fast"
pmichaud okay. 14:43
jnthn 3) A multi-dispatch where we have constraints or named parameters. In this case, we need to actually try doing the binding to see if it will work out.
pmichaud don't we still need to do the remaining type checks, though, in case they have any side effects?
jnthn The key phrase was "nominal types" - let me expand on that...
These are the types that we cache based on already.
They are a simple "isa" or "does" check. 14:44
There's no side-effect.
If there's constraints, case (2) does not apply.
pmichaud even if the nominal types were sufficient to make a decision, we're not in case (2) if any of the checks weren't strictly nominal ?
jnthn Correct. 14:45
pmichaud then case #2 should read: a multi dispatch wehre all of the type checks are nominal
jnthn We already have a bunch of stuff in Perl6MultiSub that says "is it OK to cache this"
Correct.
pmichaud okay
jnthn Case 3 is the interesting one in a copule of ways.
pmichaud the piece I was missing was that multisub's invoke gets control after doing the sub invoke. 14:46
yes, I'd suggest setting a flag on the context there
jnthn - We need to actually try to bind. That means we need to invoke, because we need a lexpad. The question is if we then enter the runloop or not.
Or really, "can I get hold of the call sig without doing so".
pmichaud if the multisub sets a flag on the called context 14:47
jnthn We'd then set a flag saying to the bind_params dynop "oh, already bound, you has no work, go relax, has a beer"
pmichaud then the binding opcode can check for that flag, and throw an exception or otherwise get control back to the multidispatcher
jnthn Well, we do exceptions today. But erm, I didn't figure out how on earth to catch exceptiosn in C. :-) 14:48
pmichaud right
might not be easily doable
jnthn Which is why I'm wary of re-entering the runloop.
I think if it was easily doable, I'd have found a way by now. :-)
TimToady surely Parrot catches exceptions in C :)
14:48 zaphar_ps joined
Matt-W You'd think so... 14:48
jnthn TimToady: Well, yeah, but returning to the right point in C is more fun. :-)
TimToady yes, thinking about how Perl 5 does it, I can see the difficulty 14:49
jnthn pmichaud: Anyway, the other thing I am guessing is that when you do an invoke, it creates a return continuation.
TimToady if Parrot uses setjmp/longjmp
or something like it 14:50
jnthn pmichaud: In theory then, if we fail to bind, then we can just invoke that to clear up, and try binding the next thing.
moritz_ is it a "I want to sleep tonight, don't tell me" thing?
jnthn TimToady: yes, it uses those.
pmichaud: But this means control never falls back into the Parrot runloop.
Well
It does for constraints.
But they're quite clean enter/exit.
pmichaud I don't understand "clear up" 14:51
TimToady wonders how to mix longjmp with continuations and will probbly not sleep tonight :)
jnthn pmichaud: Invoke creates a new context and makes it the current one.
pmichaud okay.
jnthn pmichaud: We need the lexpad available in that when trying to do the binding.
pmichaud yes, I understand that invoke creates the lexpad needed for binding.
jnthn pmichaud: However, if we fail to bind, we need to throw away that context.
Because we're not going to use it. 14:52
pmichaud s/throw away the context/return to the caller context/
jnthn Invoking the return continuation stashed within the context PMC will do that.
pmichaud: Exactly.
pmichaud we don't have to explicitly "throw it away", gc will do that.
jnthn Yes, bad phrasing on my part.
pmichaud we just need to get back to the caller context
jnthn Anyway, what I'm getting at is it won't be degree level gymnastics, just calling vtable methods. :-) 14:53
pmichaud hmmmm
jnthn So we won't be fiddling too deeply.
pmichaud how evil would it be for the binding opcode to actually put us back in the outer context if binding fails?
instead of throwing an exception or something that has to be caught?
jnthn pmichaud: Let me back up a bit.
My plan for this case (3) was that we would *not* enter the runloop and call the binding opcode. 14:54
We call VTABLE_invoke, and then enter the binder directly.
pmichaud from within the called sub's VTABLE_invoke?
jnthn No
pmichaud from the multi's vtable_invoke, then? 14:55
jnthn Yes
pmichaud okay
jnthn Well, something called from...
but yes
pmichaud I think I'd be okay with that
jnthn I think it's relatively clean.
pmichaud (need more precision about _which_ invoke is doing what :-)
jnthn And makes me *very* happy that contexts are now PMCs so we actually can flag this.
In a cheap and easy way.
pmichaud yes, that was an important change. I'm surprised we didn't take a bigger speed hit.
jnthn Me too.
Anyway, I expect that we're going to win a bunch of startup too. 14:56
pmichaud which I've been taking to be an important lesson (more)
we keep doing things that I think ought to be big speed hits but aren't, thus invalidating all of our concerns about optimizing around them
jnthn Because trait_mod goes on a non-nominal dispatch, so we have to do a bindability check.
Well, trait_mod:<is> does.
And at the moment those are really, really not cheap. 14:57
pmichaud: I may try sig construction via methods, once I've got it working via the ops, just to see what the real win is, at some pint.
*point
PerlJam jnthn: beer on the brain? ;) 14:58
pmichaud okay, that'd be good.
jnthn PerlJam: Not enough beer, I suspect.
pmichaud it'd be useful to know what the relative costs are there.
jnthn pmichaud: Yeah. The thing is...
...I think that something other than that is what is really swamping startup time at the moment.
KyleHa I'm almost always surprised by where performance problems actually are vs. where I thought they'd be.
jnthn And I rather suspect it may be that the trait mod invocations are costly. 14:59
masak KyleHa: that's why the "premature optimization" meme exists in the first place. :)
PerlJam sounds like we're back to mentioning that we need better profiling tools
KyleHa masak++ : Agreed.
jnthn PerlJam: Thing is, we have some now. ;-) 15:00
PerlJam: Just need to use them. :-)
PerlJam perhaps profiling should be integrated in the same way that testing is
Though with testing, you run the tests for the results right then, but with profiling you're going to want to look at history more. 15:01
Didn't Alias mention something like this WRT benchmarking perl a few months ago? 15:02
jnthn pmichaud: Anyway, seems I now have a way forward that you're not going to disagree with too strongly. :-)
Or maybe even like. :-)
pmichaud I don't envision a better way forward anytime soon :) 15:03
jnthn OK, let's try this one then. 15:04
Man is this going to be "fun" to implement. :-)
But at least it's not a regex engine.
;-)
TimToady it bothers me that you're talking about applying traits at binding time 15:05
jnthn is at this point *very* happy with the current devision of labor.
TimToady: Erm.
TimToady: We weren't.
TimToady: Well, I wasn't...
TimToady okay, need more coffee
jnthn TimToady: I was talking about the fact that applying traits uses multiple dispatch.
TimToady: And the way it makes use of multiple dispatch is currently a way that is rather slow in Rakudo.
TimToady 'kay, but that's still just compile time, so not in anyone's tight loop 15:06
except maybe the testor's :)
jnthn No, it's not in a tight loop, but it affects startup time.
pmichaud for the setting it's at "load type", though.
*"load time"
i.e., it's a cost incurred at startup
jnthn We currently run 450 test files, or there abouts. 15:07
Before I started working on things a couple of days back, for me Rakudo had a startup time of just over a second.
If we can pull that down to 0.5s, then I save over three minutes on a spectest run. 15:08
masak \\o/ 15:09
TimToady eventually we'll just find a way to mmap the (presumably immutable) setting structure into memory
jnthn Then with pmichaud++'s parsing speed improvements, we can parse the tests a bunch faster.
pmichaud we haven't found a good way to do that yet.
jnthn Then we have more time to go to the pub. 15:10
pmichaud (mmap)
TimToady std just pulls it in with Storable
'course, you're pulling a lot more in
jnthn erm, I mean, then we can test faster and implement new features faster...
15:10 infrared joined
pmichaud basically doing something like mmap means we need a linker 15:10
TimToady but a setting can be linked ahead of time 15:11
pmichaud so that we can tie the references in the setting structure to wherever things happened to be loaded
I mean for the things that the setting refers to that aren't part of the setting
jnthn It's not impossible, but it's not on a Rakudo * timescale either.
TimToady just thinking about how it might play out 15:12
waiting for the blood to hit my coffeestream
15:15 nihiliad joined 15:18 drbean joined, NorwayGeek joined 15:21 ejs joined
jnthn digs in to fixing the various spectest failures in the switch over to the new signature representation. 15:22
TimToady looks at why class stubs don't work 15:23
15:23 drbean_ joined
diakopter is just sayin' 15:24
15:32 crythias joined 15:33 mberends joined 15:34 iblechbot joined
TimToady .öÖ(King Diakopter the Just Sayin') 15:37
crythias howdy 15:38
masak crythias: oh hai there. 15:42
crythias Hee :) 15:48
Convicted of the crime of trying to split light into many colors, he found his eureka moment when he went to prism. 15:50
jnthn Woo. My fails list has been trimmed down muchly by two fixes. :-) 15:51
jnthn groans
crythias: that's *terrible*
masak throws crythias a punny
TimToady I had classified you as Chaotic Good, but now I'm not so sure... 15:52
crythias I'll try my saving throw.
pmichaud wonders about his classification.
crythias When you put it into perspective, a rectangle is a trapezoid. 15:53
TimToady and a trapezoid is a rectangle, sometimes
masak sorry, you just crack me up. srsly.
crythias is so chaotic that he throws two 1d20. The first determines the saving throw requirement of the second. 15:55
pugs_svn r28662 | lwall++ | [Cursor] remove setting storage bug installed yesterday 15:56
r28662 | [STD] check for yada on class declarations too
15:59 Woodi joined, meppl joined 16:02 cdarroch joined
TimToady std: class A {...}; class A {} 16:02
p6eval std 28658: OUTPUT«[31m===[0mSORRY![31m===[0m␤Illegal redeclaration of symbol GLOBAL::<A> (from line 1) at /tmp/i8GCnQO0py line 1:␤------> [32mclass A {...}; class A[33m⏏[31m {}[0m␤FAILED 00:01 96m␤»
TimToady I actually broke the build-from-scratch at 28659, TimToady--
std: 28662 16:06
p6eval std 28658: OUTPUT«ok 00:01 94m␤»
diakopter std: class A {...}; class A {} 16:09
p6eval std 28658: OUTPUT«[31m===[0mSORRY![31m===[0m␤Illegal redeclaration of symbol GLOBAL::<A> (from line 1) at /tmp/zEshZy3YD4 line 1:␤------> [32mclass A {...}; class A[33m⏏[31m {}[0m␤FAILED 00:02 96m␤»
diakopter so p6eval needs a make clean?
TimToady I think it does that eventually via cron 16:10
jnthn rakudo: &kv.candidates>>.signature>>.perl>>.say
p6eval rakudo 30e2cf: OUTPUT«:(Object *@values)␤:(Mapping self)␤»
TimToady std: 28662 16:20
p6eval std 28662: OUTPUT«ok 00:01 94m␤»
TimToady std: class A {...}; class A {}
p6eval std 28662: OUTPUT«ok 00:01 96m␤»
TimToady was just never implemented on classes
std: class A {...}; subset A of Int where 0..9; 16:21
p6eval std 28662: OUTPUT«ok 00:02 98m␤»
TimToady std: class A {}; subset A of Int where 0..9;
p6eval std 28662: OUTPUT«[31m===[0mSORRY![31m===[0m␤Illegal redeclaration of symbol A (from line 1) at /tmp/zLd8o8NBoz line 1:␤------> [32mclass A {}; subset A [33m⏏[31mof Int where 0..9;[0m␤FAILED 00:02 96m␤»
TimToady std: package A {...}; subset A of Int where 0..9; 16:23
p6eval std 28662: OUTPUT«ok 00:01 98m␤»
jnthn Eww. 16:24
TimToady std: subset A where {...}; subset A of Int where 0..9;
p6eval std 28662: OUTPUT«[31m===[0mSORRY![31m===[0m␤Illegal redeclaration of symbol A (from line 1) at /tmp/P2nUUnWErb line 1:␤------> [32msubset A where {...}; subset A [33m⏏[31mof Int where 0..9;[0m␤FAILED 00:02 97m␤»
TimToady maybe I should make that one work
std: package A {...}; constant A = 0x10; 16:25
p6eval std 28662: OUTPUT«ok 00:01 98m␤»
pmichaud for those who are interested in playing with the nqp-rx development: github.com/perl6/nqp-rx/blob/master/STATUS describes where things stand at the moment
TimToady all typenames and constant names are created equal, as far as STD is concerned
std: package A {}; constant A = 0x10; 16:26
p6eval std 28662: OUTPUT«[31m===[0mSORRY![31m===[0m␤Illegal redeclaration of symbol A (from line 1) at /tmp/uCedAf7d0o line 1:␤------> [32mpackage A {}; constant A [33m⏏[31m= 0x10;[0m␤ expecting any of:␤ constant_declarator␤ declarator␤ multi_declarator␤ typename␤FAILED 00:01 96m␤»
TimToady std: subset A; subset A of Int where 0..9; 16:27
p6eval std 28662: OUTPUT«[31m===[0mSORRY![31m===[0m␤Illegal redeclaration of symbol A (from line 1) at /tmp/mIFfCs2GCa line 1:␤------> [32msubset A; subset A [33m⏏[31mof Int where 0..9;[0m␤FAILED 00:01 95m␤»
TimToady std: proto subset A; subset A of Int where 0..9;
p6eval std 28662: OUTPUT«ok 00:01 98m␤»
TimToady heh
jnthn std: multi subset A of Int where 0..9; 16:30
p6eval std 28662: OUTPUT«ok 00:01 98m␤»
jnthn std: multi subset A of Int where 0..9; multi subset A of Str where 'A'..'Z'
p6eval std 28662: OUTPUT«ok 00:02 98m␤»
jnthn heh heh
TimToady perhaps not for 6.0.0 16:31
but one could view it as an | type 16:32
slavik2 jnthn: what is that? Perl understands SQL like things? 16:33
jnthn slavik2: heh, I was more seeing if std would accept it :-)
slavik2 oh ... 16:34
jnthn slavik2: I'm not sure what it'd mean semantically if it did though.
TimToady: well, perhaps that, yes...
Not 6.0.0. :-)
TimToady subset A of Any where Int & 0..9 | Str & 'A'..'Z' 16:35
so maybe trivial
fsdo trivial... 16:36
anyway, we can declare names without necessarily knowing the exact type 16:37
but they should be known by CHECK time
I should make STD check for unrequited stubs 16:38
well, maybe not
optional check for when you think you're done maybe 16:39
jnthn Oh gah, something in the setting is using the (currently not spec'd and likely something we want to change) signature introspection interface. 16:41
moritz_ the series operator, yes 16:42
jnthn moritz_: Aye.
moritz_ feel free to regress on that if necessary
jnthn I'm pondering that we may want to have something more object-y 16:43
So instead of a hash of data, we have a bunch of descriptors.
I knew we'd have to re-do this interface, I just hadn't realized that somebody relied on it.
Anyway, it's not a lot of work to re-do it now.
moritz_ apologizes 16:44
jnthn moritz_: No, no worries. :-)
It's one of those "expected to have to do it at some point, and now is no biggy".
Ignoring series and arity.t, which also calls .params, I've got 5 spectests that have some issues now. 16:45
That's not bad to say we've completely changed the internal representation of signatures, and shuffled a load of stuff we used to deal with at runtime to be compile time decisions in with it. 16:46
16:47 stephenlb joined
moritz_ pmichaud: is it intentional that [ab] produces two separate atoms in nqp-rx? 16:51
table tennis match, bbiafh&
16:52 ewilhelm left 16:53 payload joined 16:57 NorwayGeek|Away joined 16:59 ejs joined 17:00 icwiener joined 17:08 arthur-_ joined 17:09 alfieANDfred joined
colomon wishes he could easily test his C++ code using Perl 6... 17:13
int data [] = { -5, 0, 3, 72 }; 17:17
for (unsigned int i = 0; i < 4; i++)
pmichaud moritz_: if you mean it produces two separate atoms on the parse, then yes.
colomon versus
for (-5, 0, 3, 72) -> $i
pmichaud moritz_: however, in the resulting PAST tree you'll see that it's combined back to a single atom :)
moritz_: and in STD.pm each individual character ends up being its own atom 17:18
17:22 REPLeffect joined
TimToady colomon: the parens aren't necessary 17:25
std: for -5, 0, 3, 72 -> $i {...} 17:26
p6eval std 28662: OUTPUT«ok 00:02 105m␤»
colomon rakudo: for -5, 0, 3, 72 -> $i { say $i }
p6eval rakudo 30e2cf: OUTPUT«-5␤0␤3␤72␤»
colomon TimToady: that is so nice it makes me want to cry....
My perl 6 version which was oodles better than the C++ version still wasn't as good as Perl 6 can get.... 17:27
17:36 Chillance joined
TimToady rakudo: for <-5 0 3 72> { .say } 17:38
p6eval rakudo 30e2cf: OUTPUT«-5␤0␤3␤72␤»
17:38 crythias joined
TimToady bugs me that that doesn't turn into List of Int 17:39
jnthn A typed one? 17:41
Or just that they're integers? 17:42
erm, Ints.
mberends rakudo:<-5 0 3 72>>>.say
TimToady : <
jnthn
.oO( eyes, legs and no body? )
 17:43
mberends rakudo: <-5 0 3 72>>>.say 17:44
p6eval rakudo 30e2cf: OUTPUT«-5␤0␤3␤72␤»
TimToady it ought to randomize those just to keep people honest
s/those/the order/ 17:45
colomon : rakudo: for <-5 0 3 72> { say .WHAT }
TimToady ^ that's strange 17:46
PerlJam leading colon+space == confused bot 17:47
rakudo: for <-5 0 3 72> { say .WHAT } 17:48
p6eval rakudo 30e2cf: OUTPUT«Str()␤Str()␤Str()␤Str()␤»
TimToady only in random order :)
jnthn oh argh 17:49
jnthn has made himself a fascinating bug... 17:50
PerlJam TimToady: so ... >> is random only by virtue of asynchronicity, right?
TimToady it's random by virtue of promising that you don't care about the order 17:51
which is what "hyper" means
ruoso .oO( maybe we could accept that if the hyper is NYI as hyper, it should really randomize )
lambdabot ruoso: You have 1 new message. '/msg lambdabot @messages' to read it.
jnthn (note, order of the calls, not order of the results)
TimToady hyper for <-5 0 3 72> { .say } would make similar promises 17:52
PerlJam jnthn: right.
TimToady .say is side effect
jnthn Indeed.
TimToady so depends on order of calls, not order of results
PerlJam I think I would want what ruoso just said to remind myself of the fact. 17:53
TimToady I wonder if we should define a .yaml to go with .perl
17:53 abra joined
TimToady PerlJam: that's what I was suggesting above 17:54
ruoso in a completely different matter... I was wondering yesterday if we could somehow have auto-threading that really uses fork instead of threads...
PerlJam TimToady: yes ... my brain is a little slow right now because I'm eating. For some reason I can't eat and think at the same time :)
ruoso which would basically mean some time of iterator that knows how to serialize and deserialize the objects that are sent from/to the iterator
jnthn ruoso: Sounds like a pragma. 17:55
use hyperfork;
;-)
TimToady why would it be a pragma if it's the default?
jnthn Why would it be the default?
ruoso .oO( ouch ) 17:56
TimToady to randomize hypers?
jnthn TimToady: No, I was talking about the forking idea.
TimToady or to reorder randomized hypers?
ah, missed that
jnthn pmichaud: ping
ruoso maybe it would not be about serialization 17:57
TimToady .= kid_transport_mode()
17:57 riffraff joined
ruoso but maybe it would 17:58
ruoso thinking about objects pointing to file descriptors
17:58 abra joined
ruoso I think the trick would be to somehow detect that the object it is receiving was alive before the fork, and then somehow access that object 17:59
instead of creating a new one 18:00
ruoso feels that he's getting closer to Erlang
or maybe some objects "belong" to a specific process 18:01
and others are "clonned" to other processes
and for the objects that belong to a different process, talking to it would require sending a message to the owner process
PerlJam ruoso: if you can import any good ideas from Erlang into Perl 6, that would be great :) 18:02
ruoso the point is that I don't like threads 18:03
but forking usually makes it really hard to share data
Erlang model is really sane in that respect 18:04
ok... I think I had an idea... 18:05
18:05 abra joined
ruoso the process always store the current "version" of the memory... which only needs to be incremented after a "fork", since it marks that the new child won't have access to the memory that is allocated after that 18:06
then we have two types of object
the Imutable values, which store the "version" at which they were created
and the mutable values, which store their "owner process" 18:07
when an iterator sends an object from one process to another 18:08
pmichaud jnthn: pong
ruoso it checks
18:08 rblasch joined
ruoso 1) if it is immutable, it will check if the target process have a link to this value 18:09
jnthn pmichaud: Don't worry, I think I found the answer.
pmichaud jnthn: okay, glad I have less to worry about. :)
jnthn pmichaud: I was using :inline(...) # yeah, bad me, I know 18:10
18:10 omega joined
jnthn And, er, wondered why the code genreated for '%named' was going all so wrong. :-) 18:10
Anyway, using signatured pirop now...if only I can write the signature I want. :-)
18:10 omega left 18:11 abra joined
pmichaud you should be able to 18:11
which op?
jnthn set_signature_elem ;-)
yes, got it :-) 18:12
:pirop('set_signature_elem vPisiPPPP'),
pmichaud yup
PCT fwt
*ftw
jnthn pct++ pmichaud++
OK, down to three test files to worry about. 18:13
Plus two after that to spec and implement a new signature introspection interface for.
18:16 Woody2143 left
jnthn \\o/ 18:23
Just one now. And it's the one file I expected to have problems. :-)
18:23 abra joined
PerlJam (reality matching expectations)++ 18:23
jnthn is *really* happy that he decided to do this in small steps. 18:24
Of course, the next step is the *really* big one. 18:25
(switching to a new binder)
But knowing that signatures are in the format the new binder will expect, plus having everything else using that already, will be a big help. 18:26
18:28 krishna_ joined
jnthn OK, it's nom time. 18:29
18:30 lumi joined 18:34 abra joined
pugs_svn r28663 | fglock++ | [mp6] Javascript - 'token' statement works 18:40
18:42 i-blis joined
i-blis hello folks 18:42
does anyone know: is pugs dead?
TimToady it's hibernating 18:43
i-blis long and cold winter? 18:44
TimToady more like one of those crawly things that waits until the once-every-few-years rainstorm 18:45
KyleHa That's patience. 18:46
i-blis i see
TimToady and also impatience :)
obra_ There's definitely also some laziness and hubris in there ;) 18:47
i-blis :)
TimToady the problem with pugs is that it's bus-number was too low, and kinda got bussed
but the moment someone picks it up and starts playing with it again, I suspect it will liven up again 18:48
18:48 lumi joined
TimToady we just need to find that special person who can't make up their mind whether they like Haskell or Perl 6 better :) 18:48
i-blis sad in a way. haskell has been in my way these last months
TimToady "in my way" usually has negative connotations; did you mean it that way? 18:49
explain yourself, and give three examples :) 18:50
18:51 |theJackal| joined
i-blis bad english. i meant, i encountered haskell because of high-order style in perl, 've been playing with it 18:51
TimToady cool 18:52
i-blis very cool language. plays well with perl in my opinion. (many ways to do it)
TimToady nobody here would be upset if you woke pugs back up
i-blis to heavy for me, I guess 18:53
TimToady that's kinda why the bus-number stayed at one, alas
colomon oooo.... if The Book could contain an example of something Higher-Order Perl-ish, but in p6, that would be awesome.
i-blis i am an average perl hacker with less than one year of functional programming
which Book? Erdös's? :) 18:54
TimToady We are trying to steal the promethean fire from functional programming without stealing the procrustean bed.
i-blis ho I see
it seemed to me that p6 has done a lot in that direction 18:55
to permit clearer high-order perl 18:56
TimToady mostly a matter of keeping mutable/immutable distinctions clear, and making it easy to express things without gratuitous sequentiality 18:57
i-blis what do you mean by _gratuitous_ sequentiality? 18:58
TimToady like having to increment an index variable when you could just talk about a list of values as a whole
i-blis perl is already good at that, isn't it? 18:59
TimToady or, to take a misfeature of most FP languages, having to talk about a list in terms of head and tail
i-blis i see your point now 19:00
what needed p5 i think was a better parameter handling
and p6 brings that
this is almost enough for perl-ish fp 19:01
TimToady yes, we want to be able to spec pattern matching with sigs 19:02
i-blis streams are easy to implement any way why bother with "gratuitous sequentiality" as you called it
this is a very important feature imho 19:03
TimToady but yes, Olin Shivers described Perl 5 as one of the successful FP languages at HOPL3 19:05
i-blis i read about that 19:06
TimToady I've also liked Guy Steele's recent talk at ICFP, research.sun.com/projects/plrg/Publ...Steele.pdf 19:07
i-blis this was indeed the point that dominus tried to make a couple of years before hopl3 even 19:08
TimToady indeed, he didn't just try, but succeeded :)
i-blis :)
yes, certainly
thnaks for the link 19:09
TimToady which is one of the reasons that so many things he had to hack around in Perl 5 are just built-in to Perl 6
crythias Xistence is Supple. Let's Re-Engineer Perl. 19:12
TimToady already had both xs and lrep 19:13
crythias yes, but this is perl six backwards.
TimToady not unless you capitalize the I
someone might capitalize the E instead.... 19:14
crythias fine, fine: Xistence is Supple. Let's Re-Engineer Perl
what?
I cap..
grr
Xistence Is Supple. Let's Re-Engineer Perl. 19:15
I just don't understand. It's like I'm talking to a ...
sorry.
TimToady Life is a bitch, and it's not your bitch.
crythias indeed. 19:16
[particle] damn The Man. 19:17
TimToady In my experience, people who are bad to the bone usually have pretty good bones to compensate. 19:18
i-blis @TimToady, the presentation of Steele is very instructive (and very clear), thanks 19:19
lambdabot Unknown command, try @list
i-blis in the same vein: www.cs.nott.ac.uk/~gmh/bib.html#fold 19:20
TimToady it makes me glad that we have list-associativity in Perl 6, and also that [op] doesn't specify l or r
rakudo: say [+] ^10 19:21
p6eval rakudo 30e2cf: OUTPUT«45␤»
TimToady it's free to do that fold in any order
i-blis you may need it with (possibly infinite) streams though
TimToady sure, though [+] is by definition eager
maybe [\\+] to get the running sum
i-blis on the other hand there is no right associative ops in perl 6 (am i right?)
TimToady rakudo: say [**] 2, 3, 4 19:22
p6eval rakudo 30e2cf: OUTPUT«4096␤»
TimToady where associativity matters, the algorithm is free to reorder only to the extent that a correct result is assured :) 19:23
however, I think [**] is quite unlikely to be seen in Real Code® 19:24
on the other hand, assignop precedence is also r2l 19:25
which matters mostly for =>
hmm
rakudo: say [=>](1..5).perl 19:26
p6eval rakudo 30e2cf: OUTPUT«say requires an argument at line 2, near " [=>](1..5"␤in Main (src/gen_setting.pm:2536)␤»
TimToady rakudo: [=>](1..5).perl.say
p6eval rakudo 30e2cf: OUTPUT«Syntax error at line 2, near "[=>](1..5)"␤in Main (src/gen_setting.pm:146)␤»
TimToady std: [=>](1..5).perl.say
p6eval std 28663: OUTPUT«ok 00:02 102m␤»
TimToady missing op in rakudo
i-blis hmm 19:27
TimToady => is more or less equiv to . in Lisp
i-blis p6eval is a parrot build?
i see 19:28
TimToady no, it runs several implemenations
diakopter nations indeed
TimToady perl6: say ~ ^10
p6eval pugs, rakudo 30e2cf: OUTPUT«0 1 2 3 4 5 6 7 8 9␤»
..elf 28663: OUTPUT«Undefined subroutine &GLOBAL::prefix__94 called at (eval 125) line 3.␤ at ./elf_h line 5881␤»
..sprixel 28663: OUTPUT«Sprixel Error: Symbolic_unary prefix__S_Caret not yet implemented; srsly!!?!??␤»
TimToady otoh 19:29
perl6: say ~^10
p6eval rakudo 30e2cf: OUTPUT«0 1 2 3 4 5 6 7 8 9␤»
..sprixel 28663: OUTPUT«Sprixel Error: Symbolic_unary prefix__S_TildeCaret not yet implemented; srsly!!?!??␤»
..elf 28663: OUTPUT«Undefined subroutine &GLOBAL::prefix__126_94 called at (eval 123) line 3.␤ at ./elf_h line 5881␤»
..pugs: OUTPUT«ÎÏ␤»
TimToady pugs is the correctest there 19:30
pmichaud need spectest?
TimToady assuming one can convert 10 to a buffer and flip its bits 19:31
probably has one that is fudged or skipped
but feel free to check
i-blis ok
TimToady i-blis: the tests and specs are also in the pugs repo, if you want to hack on them 19:32
do you have a pugs repo commit bit?
i-blis Steeve's presentation is quite hard to follow toward the end :)
TimToady if not, /msg me your email and preferred svn nick 19:33
i-blis no i don't
i don't need commit i think
TimToady yes, well, he's talking to some *very* *bright* people at ICFP
that doesn't matter, you oughta have a commit bit anyway
just to keep on your countertop in case you decide you need one 19:34
if you start reading tests/specs it's almost certain you'll run across something you'll want to fix 19:35
commitbit sent 19:36
i-blis ok, i'll see 19:37
TimToady it's customary to add yourself to AUTHORS to test the bit
around here we do forgiveness rather than permission
so if you screw up, we'll just fix it
i-blis good
i msg'ed you
19:39 jamtech joined
jnthn My word, while I was away in Asia I think they installed the brightest lighting rig ever around the city castle. 19:48
Between it and the new white-ish restored stonework, you almost need sunglasses to look at it by night now.
TimToady: I'm planning to pop something in the spec about signature introspection. 19:51
TimToady: I'm thinking .params gets you an array of SignatureItem objects (plz improve name ;-)) with properties like "name", "type", "slurpy", etc. 19:52
TimToady: Does that sounds like a reasonable direction/starting point?
(BTW, anyone else welcome to chime in too ;-))
TimToady why not just Parameter objects? 19:53
jnthn TimToady: See, this is why I ask you to name stuff. :-)
Parameter it is.
jnthn feels silly for not thinking of that now. 19:54
TimToady it would be nice if there was a way to bind to the front of an argument lists and return the position where it stopped 19:55
then you wouldn't need any introspection to do C<for>
jnthn Use case?
We need that now? 19:56
TimToady optional params
for @list -> $x, $y? { ... }
jnthn Oh, you mean work through an iterator up to the point where the signature has taken all it can?
TimToady yes 19:57
I think that should be some kind of primitive
19:57 Lorn joined
TimToady and not dependend on polling the sig for .arity or .count 19:57
*ent
just something to bear in mind
jnthn Well, the only real difference is that you don't panic when you find you have some stuff still in the list.
TimToady right, and find a way to give the pos back 19:58
or iterate the iterator
jnthn If it's an iterator, it manages the pos though?
TimToady presumably, if it's destructive readout
some forms of binding want to give the whole thing back if they dont match completely 19:59
jnthn I guess then it's just making sure the binder can handle an iterator as input rather than a capture.
erm
as well as
I don't see that being too hard to achieve.
Hmm. Give the whole thing back is a little trickier though. 20:00
TimToady I just see that as a more natural approach than introspecting
jnthn Yeah, it likely is.
TimToady well, but you have to do that when trying against multiple candidates
jnthn oh hmm
yes.
That wouldn't play nice with multi dispatch otherwise. 20:01
TimToady I almost envision it as an extra * parameter that just binds the rest of the list
jnthn That could work too. 20:02
TimToady so for @list -> $x, $y turns into -> $x, $y, *@rest, and then we replace @list with @rest and continue
assuming binding to slurpy is also lazy
jnthn Not sure how the last step would look
(replace...)
Well, that is also an assumption... :-)
20:07 SmokeMachine joined 20:12 hercynium joined
jnthn (though not an assumption that I find unreasonable) 20:13
jnthn was distracted sorting out laundry
TimToady: Where does the sig introspection spec belong? 20:22
TimToady: S06?
moritz_ you could also put it in S32::$somewhere, but then nobody would take it serious ;-) 20:24
jnthn :-P
sbp tell them the S stands for Serious 20:26
moritz_ it stands for Speculation
sjohnson heh 20:29
S == Srsfknbzns
TimToady rakudo: sub myfor (@list,&block) { 1 while block(|@list) }; myfor(0..10, -> $x, $y?, *@x where { @CALLER::list = @x } { say $x, $y//'undef' }) 20:30
that almost works
p6eval rakudo 30e2cf: TIMED_OUT
TimToady says 01 over and over
moritz_ no CALLER yet
TimToady ah 20:31
jnthn Aye, it'll be putting it - quite literally - into the package CALLER. :-)
20:31 snearch joined
TimToady but you get the idea 20:31
jnthn Yes.
TimToady course, it shouldn't work anyway for at least two other reasons
jnthn One being @list isn't a context var? 20:32
TimToady and the other is that @list is not rw
well, in any case, we can't rely on the user to tack on the *@x parameter 20:33
so we need some way for 'for' to request the transformation on the block's sig 20:34
20:34 payload joined
jnthn Feels like we're a generalization away from a neat solution. 20:37
Or something.
TimToady it's either some special invoke that assumes the extra slurpy, or some way to derive a block/sig with the return path via a wrap or some such indirection 20:39
or a way to pull out the block's sig and use it as part of a larger sig 20:40
anyway, my point is that we should treat the need for .arity and such to be a design smell 20:41
nap & 20:42
pugs_svn r28664 | jnthn++ | First cut of specification for signature introspection. 20:49
20:52 payload1 joined
pugs_svn r28665 | jnthn++ | [spec] Add missing invocant and multi_invocant traits from signature introspection interface. 21:04
moritz_ jnthn++ 21:10
21:13 pmurias joined 21:14 quietfanatic joined
jnthn moritz_: In theory, this makes the signature operator fix deleting two characters. ;-) 21:17
pugs_svn r28666 | kyle++ | typo fix 21:19
21:20 Whiteknight joined
moritz_ jnthn: ;-) 21:20
PerlJam (progress by removing things)++ 21:21
jnthn ...while I like typo fixes, I can't see the typo in that diff? 21:22
oh
way
:-)
meh
thanks kyle :-) 21:23
KyleHa *bow*
dalek kudo: 10069b6 | chromatic++ | src/ops/perl6.ops:
[ops] Fixed some compiler warnings by casting malloc() results.
 21:25
KyleHa When I build Rakudo with 'make' it works, but with 'make -j 3|5', it breaks. Undeclared dependency somewhere? 21:26
moritz_ KyleHa: it works with make -j 3 (and 2 cores) here... how does it fail for you?
KyleHa perl6_ops.c:29:38: error: ../pmc/pmc_p6lowlevelsig.h: No such file or directory 21:27
jnthn Oh hmm
Maybe PMCs have to come before ops is not enforced.
21:28 ShaneC1 joined
moritz_ how do I write a multi that takes an undef as its only argument? ($x where undef)? 21:29
jnthn moritz_: Should work, if smart-matching agaisnt undef checks undefinedness.
moritz_ it does 21:30
jnthn Then yes, should work.
moritz_ rakudo: say undef ~~ undef, 0 ~~ undef;
p6eval rakudo 30e2cf: OUTPUT«10␤»
jnthn Should do what you want them. :-)
I don't think you'll get away with...
moritz_ just wandering if there was a cooler way ;-)
jnthn std: multi foo(undef) { }
p6eval std 28665: OUTPUT«[31m===[0mSORRY![31m===[0m␤Invalid typename undef at /tmp/ixjQYhJGNa line 1:␤------> [32mmulti foo(undef[33m⏏[31m) { }[0m␤FAILED 00:02 100m␤»
jnthn aye
'fraid not. :-)
moritz_ rakudo: multi sub foo(undef) { say '\\\\o/' }; foo(undef) 21:31
p6eval rakudo 30e2cf: OUTPUT«Malformed routine definition at line 2, near "foo(undef)"␤in Main (src/gen_setting.pm:146)␤»
quietfanatic std: multi foo(Any:U) {...} 21:33
p6eval std 28665: OUTPUT«[31m===[0mSORRY![31m===[0m␤Invalid typename Any at /tmp/CWGlMiJEUK line 1:␤------> [32mmulti foo(Any:U[33m⏏[31m) {...}[0m␤ expecting circumfix␤FAILED 00:01 99m␤»
moritz_ now that's a bit LTA error message ;-)
quietfanatic That's not quite the error I expected.
moritz_ rakudo: multi a(Int $x) { say 'Int' }; multi a(Object $x) { say 'Object' }; a(3|4) 21:34
p6eval rakudo 30e2cf: OUTPUT«Object␤»
jnthn rakudo: multi a(Int $x) { say 'Int' }; multi a(Any $x) { say ' }; a(3|4)Any' 21:37
p6eval rakudo 30e2cf: OUTPUT«Unable to parse block; couldn't find final '}' at line 2, near ""␤in Main (src/gen_setting.pm:146)␤»
jnthn rakudo: multi a(Int $x) { say 'Int' }; multi a(Any $x) { say 'Any' }; a(3|4)
p6eval rakudo 30e2cf: OUTPUT«Int␤Int␤»
jnthn \\o/
21:37 tak11 joined 21:38 payload joined 21:40 Intensity joined
jnthn moritz_: Yay. Series fixed. :-) 21:41
moritz_ \\o/ 21:42
jnthn Hey, this means we can write Signature.perl in Perl 6.
moritz_ \\o/ 21:43
jnthn Time to re-triage the bugs. 21:45
21:52 rgrau joined
quietfanatic rakudo: sub add ($a, $b) {$a + $b}; sub wrapper (*@_) {add(@_[])}; wrapper(3,4) 21:53
p6eval rakudo 30e2cf: OUTPUT«too few arguments passed (1) - 2 params expected␤in sub add (/tmp/tP7MizKgkh:1)␤called from sub wrapper (/tmp/tP7MizKgkh:2)␤called from Main (/tmp/tP7MizKgkh:0)␤»
quietfanatic rakudo: sub add ($a, $b) {$a + $b}; my @x = 3, 4; add(@x[]) 21:54
p6eval rakudo 30e2cf: OUTPUT«too few arguments passed (1) - 2 params expected␤in sub add (/tmp/leGcr4PgkW:1)␤called from Main (/tmp/leGcr4PgkW:0)␤»
quietfanatic rakudo: sub add ($a, $b) {$a + $b}; my @x = 3, 4; add(@x.values)
p6eval rakudo 30e2cf: OUTPUT«too few arguments passed (1) - 2 params expected␤in sub add (/tmp/oAetf5H0DR:1)␤called from Main (/tmp/oAetf5H0DR:0)␤»
quietfanatic What's wrong with what I'm doing?
moritz_ arrays don't flatten automatically in argument lists 21:55
you have to do that with |@_
quietfanatic I thought the [] or .values would flatten...
moritz_ not in argument lists.
quietfanatic rakudo: sub add ($a, $b) {$a + $b}; my @x = 3, 4; add(|@x)
p6eval rakudo 30e2cf: ( no output )
moritz_ they are essentially slice context
quietfanatic I guess I haven't quite got my head around the flattening/nonflattening business. 21:56
moritz_ it's a hairy business 21:57
quietfanatic Hmm, still not working for what I'm doing... 21:59
Perhaps it doesn't work on Parrot NCI functions. 22:04
jnthn That wouldn't be entirely surprising. :-/ 22:06
quietfanatic In which case, I shall have to know the number of parameters ahead of time. 22:07
It can be done. It's just a bit harder.
This is odd. 22:09
rakudo: my @x = 3, 4; say |@x.perl
p6eval rakudo 10069b: OUTPUT«argument doesn't array␤in sub say (src/gen_setting.pm:1730)␤called from Main (/tmp/QYB3fZHi1v:0)␤»
quietfanatic "argument doesn't array"
I came upon this by accident, but...
moritz_ @x.perl returns a string 22:10
lambdabot Unknown command, try @list
jnthn rakudo: my @x = 3, 4; say |(@x.perl)
p6eval rakudo 10069b: OUTPUT«argument doesn't array␤in sub say (src/gen_setting.pm:1730)␤called from Main (/tmp/aBW29d9j4b:0)␤»
moritz_ |Str doesn't make any sense
jnthn rakudo: my @x = 3, 4; say (|@x).perl)
quietfanatic rakudo: say |"Str"
p6eval rakudo 10069b: OUTPUT«Syntax error at line 2, near ")"␤in Main (src/gen_setting.pm:146)␤»
rakudo 10069b: OUTPUT«argument doesn't array␤in sub say (src/gen_setting.pm:1730)␤called from Main (/tmp/jSdUGar7Fl:0)␤»
jnthn rakudo: my @x = 3, 4; say (|@x).perl
p6eval rakudo 10069b: OUTPUT«[3, 4]␤»
quietfanatic That was how it happened, yes. I removed the parens.
And forgot to remove the | 22:11
dalek kudo: 36e82b6 | chromatic++ | src/pmc/p6lowlevelsig.pmc:
[PMC] Fixed compiler warnings in P6LowLevelSig PMC.
kudo: 77088bc | chromatic++ | src/pmc/perl6multisub.pmc:
[PMC] Fixed compiler warnings in Perl6MultiSub PMC.
jnthn Precedence strikes.
dalek kudo: 4f997a0 | chromatic++ | src/pmc/p6opaque.pmc:
[PMC] Fixed a compiler error in the P6Opaque PMC.
jnthn chromatic++ fixes my crappy C code :-)
quietfanatic But that error message is weirder than I could've hoped for. :)
jnthn does a spectest run, hoping this means his branch is mergable. 22:14
22:15 justatheory joined 22:16 Whiteknight joined, TopoMorto joined
jnthn reads the regex actions.pm file and is impressed 22:17
22:17 Lorn joined
jnthn method metachar:sym<$>($/) { # ah, so that's how we define actions for 'em. :-) 22:18
quietfanatic: btw, was it you that was asking about signature introspection? 22:21
quietfanatic Yes. 22:22
jnthn quietfanatic: I put something in the spec this evening.
quietfanatic: And implemetned said bit of spec in my branch.
If the current spectest run passes, I'll merge. :-)
quietfanatic I see. I shall pull and try it.
Hmm
Your branch...how do I get that branch?
jnthn quietfanatic: Wait 30 mins or so, with luck it'll be master soon. :-) 22:23
quietfanatic Or maybe I'll wait for the merge then.
jnthn If you git pull and then git checkout llsig though
Then you'll probably have it.
moritz_ ah, now I also get the parallel build failures 22:25
jnthn grabs a beer and relaxes a bit
quietfanatic I am almost done with a preliminary-version Link::C module. 22:26
jnthn quietfanatic: ooh, nice 22:27
moritz_ jnthn: how are you supposed to get the sigil of a parameter through introspection?
quietfanatic It only works with functions on native data types right now.
jnthn .name.substr(0, 1)
quietfanatic (such as int, char *, but no structs)
moritz_ ok, so it's in name
jnthn moritz_: yes 22:28
moritz_: name The name of the lexical variable to bind to, if any
moritz_ ok
jnthn The name of the lexical includes its sigil.
Though
heh
moritz_ that implies sigil, right
jnthn This definition is kinda sucky once we hit attributive parameters.
So it'll likely want a tweak of some kind. 22:29
Anyway, now there's a spec, if anybody fancies writing some tests for sig introspection, that'd be great.
A "git pull" updates all branches, yes? 22:30
moritz_ all tracked branches 22:31
22:31 i-blis joined 22:40 NorwayGeek joined 22:42 Lorn_ joined
pugs_svn r28667 | lwall++ | [S12] revise type subset pair notation a bit 22:55
jnthn OK, going for merge! 22:58
Of course, it's the next branch where The Real Fun happens. :)
pmichaud branch*es* :)
jnthn ;-) 22:59
Well, the next one is The Really Big One for the binding changes.
But yes, for other stuff I see many more. :-) 23:00
Juerd really needs to learn how to use revision control thingies, so he can decipher the other half of irc.
jnthn I expect to do some more things in master first.
Cleanups, preparation, etc. 23:01
Merged. Please note re-configure needed. 23:02
23:02 dalek joined
jnthn And this includes the parameter introspection stuff. 23:02
quietfanatic: ^^ 23:03
quietfanatic Yay!
jnthn No doubt we'll extend and tweak it a bit. 23:04
obra_ 2C 23:05
23:06 i-blis left
quietfanatic I can't clone...no doubt the fault of my college's network... 23:12
Oh there we go, never mind. 23:13
23:17 payload joined 23:19 NorwayGeek joined
quietfanatic rakudo: .type.say for (sub (Int $a, Str $b) {...}).signature.params 23:25
p6eval rakudo 4f997a: OUTPUT«Attributes of type 'struct llsig_element **' cannot be subclassed from a high-level PMC.␤current instr.: 'perl6;Perl6Role;!select' pc 5073 (src/classes/Role.pir:55)␤»
23:25 lumi joined
quietfanatic Worked over here. 23:25
anyway, jnthn++
jnthn quietfanatic: If that works for you, I suspect it's just that rakudo bot here is not updated yet.
quietfanatic makes sense 23:26
23:28 Confield joined 23:35 dalek joined 23:50 NorwayGeek joined
pugs_svn r28668 | lwall++ | [S06,S12] clarify when yada is taken to be a stub 23:51
r28668 | [S06] clarify semantics of package subs
r28668 | [S06] document that subs default to 'my' scope now
jnthn TimToady: Subs are now lexical by default? 23:53
TimToady: How much else is going to become lexical by default?
23:54 mepplock joined
jnthn Well, better phrased, are the other things you expect to become... 23:55
pugs_svn r28669 | lwall++ | [STD] improve message on unrecognized type names
r28669 | [STD] recognize :D, :U, and :_ type modifiers
r28669 | [STD] allow sub &*foo form of contextual sub declaration
TimToady I've been parsing subs as 'my' by default for a month or two now, and it doesn't seem to cause too much indigestion 23:56
I don't expect classes to become lexical only, since people want to mention those globally often
jnthn TimToady: OK, makes sense.
TimToady: One other thing. You mention :D and :U are syntactic sugar for a where constraint. 23:57
TimToady: Does this definition allow for a compile to do something smarter, though?
23:57 aleK69 joined
TimToady sure 23:57
jnthn OK
TimToady .defined is deep magic in any case :) 23:58
jnthn I note that this definition means that :(Int:_) and :(Int:D) sort to the same level in the multi-dispatch ordering.
Oh, wait
They don't
Since :_ is a no-op 23:59
TimToady otoh, it means nearly all sigs are constrained, which is a problem
jnthn In fact, I'm curious why we even have :_
Since it's the default.
Well, not really
TimToady reread the section, it's not always
jnthn You normally don't care, no?
TimToady we currently default non-invocant args to :D
but that's negotiable, I suppose

Powered by Cro and the Raku® Programming Language.

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

Thank you!