Upgrade to YAML::Syck 0.85 before smoking | pugscode.org | sial.org/pbot/perl6 | ?eval [~] <m oo se> | We do Haskell, too | > reverse (show (scanl (*) 1 [1..] !! 4))
Set by agentzh on 22 April 2007.
nothingmuch veritos: i speak hebrew too 00:00
as does gaal
roiem
lightstep
a few others, too
TimToady still, there are probably a few more speakers of Chinese of one sort or another... 00:01
veritos fine, hebrew (4+) beats hanyu (2 known)
TimToady though maybe not here
nothingmuch prolly true
oh, globally def. true =)
TimToady let's see, there's audreyt and clkao and xinming and... 00:02
nothingmuch but us jews with our jew gold and our, uh... yamacas run the world, donchaknow ;-)
*yawn* 00:03
sleepy time
TimToady but in AUTHORS there are an amazing number asian names from both ends of asia... 00:06
veritos can miniperl6 be lexed by straight yacc? 00:09
TimToady you mean parsed? yacc generally needs an external lexer. but I think miniperl6 is sufficiently regular to allow it. 00:11
veritos sorry, what was that? i hit /clear too soon
TimToady you mean parsed? yacc generally needs an external lexer. but I think miniperl6 is sufficiently regular to allow it. 00:12
veritos i meant lex, sorry
i'm trying to write in straight regexes
TimToady I'm not actually a lex expert--always rolled my own lexers. but I suspect lex can do it
veritos hmm.
TimToady especially given that lex has the ability to have multiple modes. 00:13
veritos go, little lexer. go!
TimToady you might need to distinguish terms from operators though.
you could just bypass all this, translate src/perl6/Perl-6.0.0-STD.pm to Python, and you'd be done. :) 00:14
veritos i go lower level, shouldn't be a problem. i'm even lexing $foo as [T_DOLLAR_SIGN, T_WORD_STUFF]
urgh, python is evil sometimes. 00:15
TimToady well, sure, that's why you want to write Python in Perl 6, right? :)
veritos other way around
i don't go for that whole laziness thing ;) 00:16
geoffb veritos: that may have been a hint. ;-)
veritos fglock asked me to do MiniPerl6 first anyway, so that's what i'm doing now
00:16 BooK joined
TimToady don't have any problem with that, which is why my "hint" has a smiley on it. 00:17
veritos i like ruby; it just tries to be _too_ clean. my biggest problem with it though is that functions aren't first class...you need to use blocks, which have a different call syntax 00:18
jjore-w Blocks aren't just another kind of function? 00:19
TimToady In Perl 6 they're called the same but the return syntax has to differ
jjore-w er, object?
TimToady specifically "return" has to return from the "official" function, not the blocks embedded within it.
so to return from a block requires "leave" instead. 00:20
veritos oh, and ruby -current stole our new block syntax (->(){}) too! except they call it 'stabby'.
TimToady I think they require parens too 00:21
geoffb murderous lot ...
veritos TimToady: no, you can get away without.
bsb hi can someone do: svn propset svn:mime-type image/png pugs/project_planning/a_dependency_graph.png
geoffb half expecting the atheme and katana syntax to follow soon ...
veritos www.dabeaz.com/ply/ - i'm saved! 00:22
lambdabot Title: PLY
veritos can i stick something that's LGPL straight in the pugs tree (in source form, but intended as a separate library) or would that 'pollute' everything else? 00:23
TimToady as long as it's only ever dynamically linked, I don't see a problem with it 00:25
00:26 veritos joined
TimToady veritos: as long as it's only ever dynamically linked, I don't see a problem with it 00:28
veritos TimToady: it's basically a .pm lib, but for python 00:29
TimToady bsb: svn: 'pugs/project_planning' is not a working copy 00:31
It's also okay to have it as an external dependency...
veritos what would you do in this case? 00:32
TimToady would depend on how stable I thought the original was, and how stable its source webpage. 00:33
00:33 semifor joined
veritos it was intended as a teaching tool originally it says. www.dabeaz.com/ply/ if you want more 00:34
that's the author's site
lambdabot Title: PLY 00:35
TimToady bsb: okay, figured out I should delete the pugs/ :)
I will let you decide--that's why you have a commit bit. :) 00:36
but I don't think the license is a problem offhand.
veritos sounds good...i'll put my email in the readme so he can whine to me if he has a problem with it. he's a teacher though, i think he'll be fairly reasonable. (one of the SWIG guys too!) 00:37
bsb svn.pugscode.org/pugs/project_plann..._graph.png works now 00:38
lambdabot tinyurl.com/2z58bw 00:39
TimToady so it would seem here too, thanks 00:40
00:48 xdg joined 00:51 amnesiac joined 01:01 dolmans joined
TimToady A cow. A trampoline. Together they fight. A crime. 01:09
01:18 Lunchy joined 01:36 lisppaste3 joined 01:43 bonesss joined 02:10 SubStack joined 02:13 rissy joined 02:22 geoffb joined 02:40 veritos joined
veritos hello again 02:40
02:42 Limbic_Region joined 02:48 toshinori joined 02:54 mako132_ joined
veritos what happened to the svn bot? 02:56
and is anyone else working today? my tree's been up to date all 5 times i've updated it. 02:57
03:02 elmex_ joined
dduncan I'm working today 03:19
offby1 is loafing today ... as most days 03:21
03:30 agentzh joined 03:32 Alias_ joined
TimToady most of my work happens when I'm loafing 03:32
agentzh iblech, malon: the synopses on the smoke server have run out of order (see the ones under "repository snapshot / linux"). thanks.
dduncan I like bread 03:33
which comes in loaves
03:36 veritos joined
veritos hey guys and gals...just got another v6-Python commit. it now (sort of) parses (a bit of) MiniPerl6 and outputs an ugly mass that somewhat resembles Ruby. have a looksee: pugsroot/v6/v6-Python 03:37
and if you looked, try svn up - ing your tree again. 03:39
and again. 03:40
anyone actually on? 03:42
offby1 wishes he could write off loafing as work
veritos offby1: psst....svn up. 03:44
cd ~
pasteling "evalbot_r16100" at 194.145.200.126 pasted "Pugs build failure" (3808 lines, 445K) at sial.org/pbot/24451
veritos cd pugs
pasteling "evalbot_r16100" at 194.145.200.126 pasted "Pugs build failure" (3803 lines, 445K) at sial.org/pbot/24452
veritos cd v6
cd v6-Python
03:46 ruoso joined
offby1 hmm 03:46
she
offby1 svk pulls 03:47
veritos i think i'm the only one who's committed in the last ~6 hours 03:48
TimToady no, dduncan has too
veritos oh, people are here. 03:49
dduncan yes, several times
in fact, I hope to make a release today
veritos what's your little pet, pun intended, project? or did you mean for pugs in general?
Patterner yay for releases
dduncan QDRDBMS is a module in synchronized Perl 5 and 6 versions ... the Perl 5 version will probably be released on CPAN today, but the announcement will mention both versions, so people who want the Perl 6 one can go to the Pugs repository, and the Perl 5 one is on CPAn as usual 03:50
veritos oh
fun... 03:51
oh, alright...tree needed updating a couple sec ago
dduncan I also see it as appropriate to post the announcement on Perl 5 and Perl 6 lists ... in the former case, it would be database related lists, in the latter case, I figured perl6-announce plus perl6-users would be the most appropriate ... I assume that perl6-language won't be an appropriate place to announce ... not for this release anyway 03:52
a few releases later, p6l would be appropriate, because at that time QDRDBMS would include implementations of data types that I've been suggesting for mainline language support 03:53
that is, the relation data type
but the initial release won't have that, it is more of a skeleton
veritos how do you type QDRDBMS so fast? kind of a mouthful. 03:54
dduncan I'm used to it
in any event, that name is temporary, to be replaced with somethng easier to type later 03:55
wolverian what's it "short" for? :) 03:57
dduncan oh, and I'm not the packager of Pugs ... basically just audreyt releases that
veritos i see.
dduncan I'll give you a hint ... the first 2 letters mean the same as in QDOS, the latter 5 are readily understood if you know the module is about databases 03:58
and that may tell you why the name is temporary
wolverian yeah, it's the first two I'm unfamiliar with
dduncan Quick and Dirty Operating System ... which was later renamed to MS DOS 03:59
wolverian ah :)
dduncan the name denotes that this work is currently pre-alpha state
or particularly that I've cut certain corners on purpose in order to get something working out sooner
when those omissions are corrected and the project is more mature, it will be renamed 04:00
and so people aren't surprised, I state this intention in the project's README file
the other thing is that I haven't 100% decided what the longer term name will be, so I picked a bad short term name on purpose so people won't want to stick with it longer term 04:02
that is all
gaal veritos: actually (+) is not Num -> Num -> Num because Num is a typeclass, not a type. 05:07
@type (+)
lambdabot forall a. (Num a) => a -> a -> a
gaal you can ignore the forall a here; what this signature means is that for any type a that you pick, the signature is a -> a -> a so long as a happens to belong to Num. 05:08
where "belong to Num" means that you have "instance Num ThatDatatype where .. " and definitions of at least the minimum nexessary functions 05:09
05:18 laye joined 05:21 SubStack joined 05:51 buubot joined 05:54 Aankhen`` joined 06:11 rfordinal_ joined 06:13 BooK_ joined 06:30 franck__ joined 06:35 franck___ joined 06:40 BooK__ joined 06:44 benny joined 06:46 devogon joined 06:51 drupek12 joined 07:05 drupek12 joined 07:06 rfordinal_ is now known as rfordinal, sapir joined 07:09 rissy joined 07:10 marmic joined 07:19 buetow joined 07:20 justatheory joined 07:21 jisom joined 07:35 iblechbot joined 08:10 Daveman joined 08:19 buetow joined 08:23 awwaiid_ joined 08:36 rfordinal_ joined 08:53 zamolxes joined 09:02 Salzh joined 09:23 toshinori joined 09:25 dduncan left 09:56 demq joined 10:13 ludan joined
ludan hola 10:14
moritz moin ;-)
10:15 midellaq joined 10:17 devogon joined 10:23 mj41 joined 10:27 polettix joined 10:33 jisom_ joined 10:48 ruoso joined 10:50 BooK joined 10:55 BooK_ joined 11:05 ayrnieu joined 11:15 chris2 joined 11:27 iblechbot joined 11:28 rfordinal_ is now known as rfordinal
pasteling "evalbot_r16106" at 194.145.200.126 pasted "Pugs build failure" (278 lines, 15.8K) at sial.org/pbot/24459 11:32
11:32 ruoso joined
moritz I think we need a file in the pugs repository where different tasks are listed, and who usually takes care of them 11:32
for example evalbo, irclogs, smartlinks, autosmokes etc. 11:33
avar /SLAVES 11:34
moritz good name ;-)
bbiab&
OwlEye autosmokes sounds interesting :) 11:36
11:36 fglock joined 11:43 mako132_ joined 11:46 the_dormant joined
fglock veritos: ping 11:48
veritos: in case you backlog: 11:49
mp6 can be bootstrapped by writing an emitter for the target language (Python)
and then recompile the mp6 source code to python 11:50
that is, you don't need to rewrite the whole compiler
12:08 buetow joined 12:11 fglock__ joined, fglock__ left, fglock__ joined 12:13 fglock___ joined 12:17 fglock__ left, fglock joined 12:27 vel joined
avar (multiple emitters)++ fglock++ 12:28
12:59 semifor joined 13:04 BooK joined 13:16 jerrylee joined 13:19 HiTech69 joined
fglock the current version of kp6 can convert perl6 OO into pure procedural code - makes it easier to support 'simple' backends 13:36
avar how does kp6 relate to mp6? Isn't the aim of kp6 to not require mp6? 13:37
moritz I think kp6 is written in miniperl6, right?
fglock kp6 is currently the language in which kp6 is written 13:38
mp6
moritz: yes :)
avar don't trip over each other:)
fglock kp6 has better semantics, but requires less backend support; OTOH, it does more emulation and can be slower 13:41
but the level of emulation is configurable 13:42
avar how were you doing with the parrot backend? Past hello world?:)
fglock avar: with the latest kp6 changes, it could now support full OO 13:43
moritz PIR hello world? ;-)
fglock the latest tested code is:
svn.pugscode.org/pugs/v6/v6-MiniPer...6parrot.pl 13:44
the current plan is: kp6 -> non-OO mp6 -> p6parrot 13:50
avar fglock: does it emit pir that you can then run without [mp]p6? 13:55
fglock I could not make p6parrot emit pir yet; not sure if it's my fault... 13:56
but mp6 emits simple-p6 that p6parrot can run 13:57
without mp6
14:03 rlb3_work joined 14:20 rlb3_work joined 14:21 [particle] joined 14:24 perlmonkey2 joined 14:28 fglock_ joined 14:29 DarkWolf84 joined 14:30 rindolf joined
rindolf Hi all. 14:30
14:30 hcarty joined
moritz hi rindolf 14:30
rindolf Hi moritz
moritz: sup?
moritz rindolf: nothingmuch ;-) 14:31
rindolf moritz: OK.
14:32 fglock__ joined
rindolf I've started working on perl.net.au/wiki/Tactfulness_for_Geeks 14:32
lambdabot Title: Tactfulness for Geeks - PerlNet
moritz rindolf: that's like trying to preach to the devel ;-) 14:34
geoffb moritz: oh, man, that one *hurt*
moritz geoffb: no harm intended ;-) 14:35
nothingmuch rindolf: tactfulness doesnt involve protocols and latency, it involves applying a filter to what you say 14:38
90% of the time you can avoid needless confrontation by just stfuing for another 10 seconds and thinking if that's really what you want to or should say, and if that's how you want to say it 14:39
and that applies to real life also
i don't think the concept of tact varies from one medium to another
and also, i believe it's media, not mediums (i think mediums is plural for the synonym of the fortune teller) 14:40
at least that's the only way i've seen it used that way withoutu being corrected by others quickly
rindolf nothingmuch: often if you don't say anything it will be worse than saying something. 14:41
nothingmuch: tact is not only about shutting up.
nothingmuch: which you should know if you've read PG. 14:42
nothingmuch PG?
moritz rindolf: but in 90% of the cases shutting up helps ;-)
nothingmuch for what it's worth I don't agree, it's very uncommon that saying something is better than saying nothing if there's a dillema between the two 14:43
moritz ?eval my @a = 1..3; @aĀ»++
evalbot_r16111 (*** Can't modify constant item: VInt 1ā¤ at <eval> line 1, column 15 - line 2, column 1, *** Can't modify constant item: VInt 2ā¤ at <eval> line 1, column 15 - line 2, column 1, *** Can't modify constant item: VInt 3ā¤ at <eval> line 1, column 15 - line 2, column 1)
nothingmuch and I didn't say it was only, i said it was mostly
moritz should that work?
nothingmuch moritz: yes, it should
moritz ?eval my @a = 1..3; @a=@aĀ»+1
nothingmuch oh wait
evalbot_r16111 Error: ā¤Unexpected "\187+"ā¤expecting "::", term postfix, operator or ","
nothingmuch ?eval my @a = (1..3); @aĀ»++ 14:44
evalbot_r16111 (*** Can't modify constant item: VInt 1ā¤ at <eval> line 1, column 17 - line 2, column 1, *** Can't modify constant item: VInt 2ā¤ at <eval> line 1, column 17 - line 2, column 1, *** Can't modify constant item: VInt 3ā¤ at <eval> line 1, column 17 - line 2, column 1)
nothingmuch hmm, no
i remember something vague about binding semantics in array assignment
rindolf perlbot: ph
perlbot: pg
nothingmuch but i'm out of touch
rindolf nothingmuch: PG == Paul Graham
14:44 hcarty left
moritz but the version with @aĀ»+1 should work, right? 14:44
nothingmuch rindolf: i doubt paul graham could teach me much about tact or life in general, my parents have first dibs 14:45
rindolf nothingmuch: "A wise man can learn from a fool, much more than a fool can ever learn from a wise man".
nothingmuch: and Paul Graham is very wise.
nothingmuch also, most of what I read by him (and it wasn't much) may have been interresting, but I definitely didn't consider it to be The Word
so, if i read his stuff he will learn from me? ;-) 14:46
rindolf nothingmuch: www.paulgraham.com/nerds.html
lambdabot Title: Why Nerds are Unpopular
rindolf nothingmuch: I didn't say he was the absolute truth.
nothingmuch: but I got many important insights from him.
nothingmuch: no. Not unless you blog about it or whatever. 14:47
nothingmuch: and he reads it.
nothingmuch: I don't consider you a fool.
What my quote means is that a wise man always learns more than a fool.
A fool is stubborn, he's not open.
nothingmuch i know what it means, i was kidding 14:48
rindolf He may be very intelligent.
nothingmuch: oh. Sorry.
:-)
/me can sometimes not detect when people are saying stuff tongue-in-cheek
nothingmuch that's why i put the smiley in there
internet lacks tone 14:49
PG's problem is that he writes a lot of text to convey a simple idea
ah, there we go 14:52
i agree with that paragraph in spirit
"In fact, it's derived from the same root as "tactile," and what it means is to have a deft touch. Tactful is the opposite of clumsy. I don't think I learned this until college."
to me that means knowing when to shut up and when to speak up
not to shut up most of the time
14:52 toshinori joined
pasteling "evalbot_r16111" at 194.145.200.126 pasted "Pugs build failure" (6 lines, 228B) at sial.org/pbot/24466 14:53
fglock_ nothingmuch: I wrote a very simple OO model for kp6, it would be nice if you could review it 14:57
nothingmuch sure
fglock_ run tests in v6/v6-KindaPerl6, with 'sh run_tests.sh' 14:58
the lib is: svn.pugscode.org/pugs/v6/v6-KindaPe...rl5/MOP.pm 14:59
lambdabot tinyurl.com/2apyw6
rindolf nothingmuch: it also means knowing what is the right thing to say. 15:00
nothingmuch: and how to say it.
nothingmuch: what's up, BTW?
nothingmuch knowing what the right thing to say and when to say it is omiscience 15:01
tacfulness, for me, is minimising damage when you're not 100% sure
rindolf nothingmuch: pardon?
nothingmuch when you know the right thing to say there's no need for tact
rindolf nothingmuch: you can often say the same thing in two different ways, and one of them will be much better.
nothingmuch: no, but you can often have a good judgement of what is better to say. 15:02
How will people interpret it, etc.
nothingmuch in practice I find that is often not such a big problem as the other half
15:02 toshinori left
nothingmuch fglock_: why go back to "traditional" OO 15:03
rindolf nothingmuch: I agree with you that not saying anything is something you have to do.
nothingmuch ?
rindolf nothingmuch: as opposed to prototype-based OO?
Or what?
nothingmuch as opposed to the generalized metamodel thing that was the result of roughly 2 years of continual redesign by stevan and then by me 15:04
which is capable of both classy OO and prototype OO
and handling things like AUTOLOAD easily 15:05
fglock_ nothingmuch: I was after something that I could implement fast; I mean to upgrade the model later 15:06
nothingmuch why not reuse MO? it's already implemented 15:07
fglock_ nothingmuch: I need to translate this to MiniPerl6; it's easier to translate my own code (not a good excuse) 15:08
nothingmuch fairynuff
it 15:09
[particle] fglock: i think that's one reason other folks aren't helping you much (you keep translating your own stuff)
nothingmuch s just a bit of a shame to see a regression in terms of the limitations we tried to take out
fglock_ [particle]: mp6 is the stable implementation 15:10
[particle] nm: is MO the latest and greatest metamodel? 15:11
or class::mop?
fglock_ nothingmuch: do you think I can keep the API and change the implementation later?
clkao Win 47
rindolf www.m-w.com/cgi-bin/dictionary?book...;va=medium - Inflected Form(s): plural mediums or meĀ·dia /-dE-&/
lambdabot Title: Definition of medium - Merriam-Webster Online Dictionary
nothingmuch [particle]: MO is later 15:12
Class::MOP is perl 5 specific
MO is inspired by it (and as such by CLOS and smalltalk's metamodel)
but attempts to make compile time and runtime separatable
[particle] great. maybe i'll port MO to parrot 15:13
nothingmuch the part that should be ported first is very very simple -- reusable responder interfaces
[particle] now that we have a fancy new object model
nothingmuch (not necessarily the ones I sketched in p5)
responder interfaces are dumbed down versions of classes meant to be easy to dispatch on
fglock_: MO's API requires a new layer
what you sketched seems to be like the responder interface layer 15:14
[particle] where is MO? 15:15
nothingmuch nothingmuch.woobling.org/MO/
lambdabot Title: Index of /MO
[particle] cpan search is less than helpful
great!
nothingmuch it's not CPAN worthy yet
[particle] ah, i see
nothingmuch $work ;-)
the problem is that the "stock" objects for Class, etc are more experiments
to stretch the boundries isntead of provide a good base
it's noncomittal
[particle] reads the ten thousand inch view 15:16
nothingmuch nothingmuch.woobling.org/MO/lib/MO/Overview.pod
?
[particle] yep
nothingmuch lk
oi, i see
[particle] s/"/'/
nothingmuch ;-)
patches welcome
;-)
fglock_ nothingmuch: I'm mostly interested in working 'Code' objects right now; the model can be improved later
nothingmuch fglock_: fairynuff, i agree that the perfect is the enemy of the good 15:17
but I would make sure you are not using the meta model in a way that is too restricting for an MO style one to come by later
theoretically in MO you can swap out your entire object system (new concept of method calls, new concept of objects, new concept of "concepts" (defining meta objects such as prototype objects or classes)) 15:18
without stepping over the language's native ones
(part of the inspiration was parrot's intended multiple language thing, and the problems that might arise from that)
fglock_ is there something I should change right now?
nothingmuch it's tricky to say
i don't know kp6 or your mop very well 15:19
[particle] nm: in parrot, Class and Object are coupled... if you override one, you need to override the other. is this common?
nothingmuch [particle]: Class is the class of the meta class and Object is the class that is the base class of all instances?
fglock_: the problem with MO is that it works in cycles 15:20
the compiler or the runtime creates meta objects whose job is to retain the declarative data closely 15:21
kinda like a high level AST
[particle] nm: yes, iirc
nothingmuch [particle]: in the current example metaclass (Class::MI and Class::SI) there is no concept of a base class of everything
instead "core" methods are generated
but this can be changed
as for whether or not it's common - oi 15:22
i've never seen that restriction before
i don't think it matters much
if you subclass Class into e.g. My::ORM::Metaclass
then maybe $db_bound_object has a separate default meta class
but maybe it doesn't and it's just the meta protocol (e.g. accessor generation) that's being changed
i think they are orthogonal but that it's not a horrible restriction to enforce 15:23
... fglock_:
these objects may refer to eachother (by value or by name, it doesn't matter)
for things like subclassing, etc
then you compute standalone structures from these
either at compile time in advance or as necessary during boxing 15:24
and you reference this structure from the object's type box
instead of the original metaclass
if you keep this runtime/compile time separation possible (even if it's all runtime for now) then you should be able to plug MO later by just generalizing the method call opcioode
but there needs to be room for the multiple pass arch 15:25
you should probably ask audreyt for advice, if she knows your stuff better
because she also knows MO well
btw, how is ingy? does anybody have an update?
fglock_ nothingmuch: thanks 15:26
nothingmuch btw, don't take this the wrong way but WRT the perfect being the enemy of the good
this is the Nth mini language, right?
at some point i think you need to try to get at least the most difficult part closer to perfect 15:27
even if it slows you down
because in the long run it's going to have to happen sometime
fglock_ nothingmuch: most difficult part == MO ? 15:28
nothingmuch the most difficult part is interlocking parts that need to get bootstrapped somehow
the compiler is written in it's own language, right?
fglock_ yes
nothingmuch to write a compiler you want good uspport for $nifty features
but to support these features you need a good compiler
(parser, meta model, interpreter, emitter, yadda yadda yadda)
all these parts depend heavily on one another and if you kludge one of them you will at some point suffer in another part due to that 15:29
15:29 toshinor_ joined
nothingmuch i don't think you can really write everything up front 15:29
15:29 |OUF|HiTech69 joined
[particle] can't you build a better metamodel from a simpler one? 15:29
nothingmuch the point is that there have already been many simpler ones 15:30
we've learned our lessons =)
i'm not preaching for big design up front
but it looks a bit like we're avoiding the next big step by looking for slightly more immediate results
s/we're/you're
[particle] i'm wondering if MO can sit on top of what fglock has
nothingmuch i don't deserve to claim i'm doing anything ;-)
[particle]: yes and no
MO sits on Moose in perl 5 land 15:31
i've experimented with MO in perl 5 bootstrapping itself
and it works OK too
but the thing is that the current stuff is in perl 5, with some integration it could be reused quite easily
eventually MO will need to get ported to perl 6
and then get compiled by it's perl 5 counterpart
and then it can compile itself 15:32
[particle] and if it's ported to kp6, then that can happen *soon*
nothingmuch whereas in the current scenario it looks like the simpler model will get converted to perl 6, and then MO will be written in that, and then it will be integrated
fglock_ nothingmuch: I think the simpler model can be converted to MO by hand, with some help of the compiler 15:33
nothingmuch i don't see how the compiler helps 15:34
but suure, it can be converted later
the question is at what cost
and if it's really cheaper (time wise) to do that later instead of now
fglock_ the compiler just takes care of the syntax sugar 15:35
nothingmuch superficially
fglock_ I'm ok to do it now,
nothingmuch in practice it's also about the compilation structures' lifecycles
fglock_ kp6 milestones are all met,
nothingmuch which are crucial for MO to be used correctly
fglock_ so there is no reason not to do it now
nothingmuch i can't help right now ($work schedule is hard)
but maybe come may 15th i'll have more time to hack again 15:36
things to keep in mind -- a port of MO does not necessitate direct porting of the classes therein
it's mostly about the most basic design decisions
and about responder interfaces (which are the easiest part to port)
so you can simplify MO to very few LOCs, i think
and it remains extensible when the language requires richer meteaclasses 15:37
[particle] nm: based on this conversation, and your work on MO, i'd appreciate your comments on PDD15. think you can give it a look when you have time?
nothingmuch [particle]: i'll give it a shot
link?
fglock_ what if I ported some MO tests to kp6, and worked on it
[particle] www.parrotcode.org/docs/pdd/pdd15_objects.html 15:38
lambdabot Title: Object and Class semantics for Parrot - parrotcode:
nothingmuch fglock_: i think the smartest thing to do is to take your model
and refactor it to use MO style cycles
that way you don't need to make big changes later
and you don't need to adopt a foreign code base alltogether
fglock_ are there docs on MO style cycles? 15:39
nothingmuch and you can even fake it
because the responder interface can just be the metaclass
it's for when it's inconvenient for the metaclass to be a responder interface that MO brings benefits
that overview that [particle] was reading
fglock_ ok
nothingmuch i've gone further and implemented all the objects involved in MO in a purely functional style 15:40
so that they're portable to haskell, too
but that's not mandatory
(though ideally it'll be refactored back into that once it's self hosting in perl 6)
[particle] we're currently redesigning the PMC object 15:41
nothingmuch [particle]: skimming it it looks like Class tries to be the one true responder interface in MO speak
[particle] i've just started reading about responders now in MO
nothingmuch (not necessarily good or bad)
i'll read into it more 15:42
fglock_ brb
[particle] for PMCs, the responder interface is a vtable
hrmm, actually that's incomplete
it's the vtable plus non-vtable methods
nothingmuch from what I remember the pmc is practically it's own responder interface 15:43
[particle] we've been talking about role-based pmc composition
nothingmuch becaus eit's data coupled with a type
ingy nothingmuch: kiss
nothingmuch ingy: yay, you're concious!
[particle] hey there ingy! 15:44
ingy hi
cls_bsd ingy: hi !! 15:45
and night!!!
nothingmuch [particle]: the original vision is that by extending the VM or implementing responder interfaces with the VMs language (both should be possible) the metaclass author can support an idealized runtime version of the object's behaviro
== performance 15:46
for example compiling a flattenned vtable
indexing method names
etc
[particle] okay
nothingmuch all lead to the construction of a simpler data type
90% of the itme this will be a dispatch table
[particle] we definitely want vtable flattening, rather than storing pointers to the vtables of the composers 15:47
ingy hi cls_bsd
[particle] hopes ingy has found a friend in tab-completion 15:48
nothingmuch when the process is separated then the metaclass's advanced features (e.g. roles, attribute grammars, complex attributes, various construction behaviors etc) can be implemented in the host source language
ingy heh
alwayz
nothingmuch as long as the output is a simplified structure that can then be transliterated into a natively supported responder interface 15:49
example: for every possible target class, precompute the final table of all applicable methods and save them into a tabular structure
if optimising for size, fragment the tables into shared ones with pointers
if optimising for speed make only one vtable for every class
the two optimisations can coexist as long as there is either a polymorphic dispatch
(the responder interface has a type) 15:50
or a different opcode is generated for the method call
ingy i'm not much slower typing one handed then two
nothingmuch inlining the lookup logic at the callsite
rindolf Hi ingy
ingy: are you feeling any better?
ingy hi rindolf
still hate me?
rindolf ingy: I never hated you.
ingy: oh, you mean about speeding?
ingy ;) 15:51
nothingmuch speeding is fun
[particle] everyone hates ingy!
rindolf nothingmuch: not on the road.
nothingmuch rindolf: speeding is fun on the road
ingy Ricky Bobby is my hero
nothingmuch skydiving is also fun
rindolf "Everyone hates Ingy dƶt Net"
[particle] skydiving *is* fun
nothingmuch risk and consequences don't necessarily deduct from the fun of the activity before they are realized 15:52
hence speeding is fun even if it's reckless
ingy nothingmuch: well said
nothingmuch ingy: i've been studying for this 15:53
played wiuth snowboards and ice axes
i can rationalize stupid hobbies very well now =)
[particle]: are you seeing what I'm getting at? rich semantics in the language's compiler, and low level reusable components in the runtime 15:54
ingy www.cafepress.com/buy/movie/-/pv_de...pt_/c_360/
lambdabot Title: CafePress.com : Error, tinyurl.com/29lbs5
nothingmuch if required the runtime can use rich semantics too, but it doesn't have to all the time
for example in perl 5 it's rich semantics all the way (the interperter walks the stash)
rindolf I've uploaded to Flickr a photo of a Hyrax I took in the Carmel : www.flickr.com/photos/shlomif/471195681/ 15:55
lambdabot Title: Hyrax in Little Switzerland on Flickr - Photo Sharing!
[particle] yes, i can see it, but i'm not exactly sure how to get there
nothingmuch hosting perl 5 on a MO VM will mean that the metaclass (prolly an object wrapping the stash) is probably going to be it's own responder
whereas in C++ for example, you have the one true metaclass
rindolf Licence is PD.
nothingmuch that compiles to the one true responder interface
because everything is static it's easy:
either it's virtual and you run it through a vtable or it's just a function call with OO syntax ;-)
but C++ is not rich 15:56
MO tries to make both worlds possible depending on the investment the VM and the language compiler are willing to make
[particle] ultimately, we'd like PMC and PIR oo to look the same, so you can use interchangeably
nothingmuch i'm not sure exactly what you mean 15:57
from whose POV?
[particle] it's a bit tangential
from the user's perspective 15:58
nothingmuch from the user's perspective it should be possible
(you mean the metaprogramming, compiler writing user, right? not that user's user?)
[particle] yes
nothingmuch .. provide implementations of RIs and metaclasses for PMCs
[particle] correct.
15:59 ruoso joined
nothingmuch it should be simple 15:59
then the code gen emits abstract opcodes like "box"
and the static analyzer can decorate these opcodes with more info 16:00
like a set of types that this box might be
and then this can be optimized into lowlevel opcodes
which are different
e.g. if you say $class.new(...); then it's always the abstract opcode in the runtime
but if it's Str.new
then the metaclass is known, hence the RI getting boxed in the .new is also known 16:01
and since this is really a proxy for the PMC stuff in the low level it's not "native" OO but FFI OO
(in the sense that Str is a class implemented by the VM or a library the VM uses or whatever)
this applies to multiple languages at once too, btw
Str might be implemented in python 16:02
and $str.foo will be dispatched according to pytjon rules
using python metaclasses
it's not "brought over" to the perl side
is this any clearer?
[particle] yes, thanks 16:04
nothingmuch i should probably read the PDD before advising more
i only have a vague picture so i need to keep things too generalized ;-) 16:05
[particle] sure.
and i've only just begun writing tests for the implementation
mostly i'm testing api's and other unit tests
16:05 bernhard joined
[particle] we need some solid functional and performance tests, though 16:06
fglock_ backlogging... 16:22
16:23 veritos joined
veritos morning 16:23
i already told fglock this, but apparently Python lets you compile to c code too. 16:24
it's only dependent on your system's libc
basically like B::C, but that works.
16:31 amnesiac joined 16:34 kanru joined, midellaq left
fglock_ re nothingmuch.woobling.org/MO/lib/MO/Overview.pod - the compile-time metamodel look just right; now looking at the runtime... 16:38
16:45 fglock_ left
nothingmuch /w 35 16:47
16:48 franck__ joined 16:53 buetow joined
gaal looks like a great resource: www.diku.dk/hjemmesider/ansatte/torbenm/Basics/ 16:54
lambdabot Title: Basics of Compiler Design
gaal # free book
moritz looks 16:55
16:57 fglock joined
TimToady nothingmuch: I'm wondering if P6's "undefined prototype" objects can be thought of as essentially just the pure, abstract responders without any particular attached object. 17:05
17:06 penk joined
TimToady the responder representative, if you will... 17:06
I was always thinking of the RI as "something beyond .HOW", but those undefs are pretty close to the same notion, insofar as they respond to the same methods (to the extend possible without an object) 17:08
nothingmuch TimToady: IMHo there's no need to burdon the language with RI details 17:09
that's for language implementors
granted prototype object based OO is very easy to translate to RIs
moritz gaal++ # that books looks fine ;-)
nothingmuch so the affinity is high between the metaclass and the RI of prototype OO stuff
but I'm not sure p6 should care 17:10
i mean, technically, yes, that's what they'll be (a box with just a type, i suppose)
TimToady sure, just trying to figure out if it saves me mental energy to identify them more closely in my mind.
nothingmuch but i don't think it's a fundamental change in the thought modeling 17:11
TimToady if RI is fundamental to the MO view, and MO is fundamental to P6, then at some point RI shows up as fundamental in P6...
nothingmuch RI is fundamental in the same sense that target opcodes are
compile time metamodeling is fundamental in the same sense that prelude functions are 17:12
TimToady except there's a profound difference between a singular interface and a bunch of opcodes.
nothingmuch but it's not singular 17:13
RIs are menat to duck type in the interpreter
17:13 ozo joined
TimToady well, sure, the whole OO mindset is that things that are singular on the outside are plural on the inside somehow... 17:13
nothingmuch si 17:14
17:14 justatheory joined
TimToady and I'm not saying that RI's have to be real objects, only that to the extent they are, an undefined prototype is very like it. 17:14
nothingmuch which is why i think it's invisible to the language
oh
sorry, i misunderstood
hmm
that's aesthetic 17:15
but I don't know if it actually has any technical meaning ;-)
TimToady probably not, unless they also happen to respond to ".WHATTHEHECKKINDOFRESPONDERINTERFACEAREYOU" 17:16
nothingmuch hmm
rindolf Hi TimToady
TimToady or .WTF maybe. :)
nothingmuch i always ssaw RI as the return value from blessed() in perl 5 terms
TimToady but yeah, mostly just mental construct 17:17
nothingmuch and likewise what you give to bless() to box some data with a type
rindolf Heh. Today is Tuesday, when God said "And it is good" twice.
TimToady gotta take my kid to the bus stop for ~20 min.
afk &
rindolf Or when God saw it was good twice.
TimToady: bye.
TimToady: have fun.
nothingmuch rindolf: what a coincidence!
rindolf nothingmuch: it's mystical I tell you! 17:18
nothingmuch yep... this tueday just keeps popping up
rindolf It only happens once every 7 days.
nothingmuch TimToady: bless( $instance_data, Dog ); # consistent 17:19
also querying the type of is consistent
but i think it's more of a parallelism
rindolf nothingmuch: you're using vim, right?
nothingmuch because the proto is the type of the object to some extend (loose definition of type here)
and the RI is the support object that the VM uses to figure out how the type behaves 17:20
rindolf: yes
rindolf nothingmuch: ah OK.
nothingmuch what value will that object give, though?
i mean, besides proto stuff
rindolf nothingmuch: how long is your .vimrc file? 17:21
160 641 5679 .vimrc
But I have some stuff in other files.
nothingmuch Dog.handle_method( method => Method::Named.new( :name<foo> ), responder => $dog ); # possibly useful?
especially if $dog is really something else 17:22
rindolf Hmmm... www.flickr.com/services/api/upload.api.html
nothingmuch rindolf: about the same
lambdabot Title: Flickr Services
rindolf nothingmuch: how do you upload photos to Flickr? 17:23
nothingmuch i don't
rindolf nothingmuch: ah.
nothingmuch: then who does?
nothingmuch flickr users 17:25
rindolf Hmmm... search.cpan.org/~cpb/Flickr-Upload-1.25/
lambdabot Title: Christophe Beauregard / Flickr-Upload-1.25 - search.cpan.org
rindolf nothingmuch: I meant in regards to Perl 6.
nothingmuch who uploads photos to flickr in regards to perl6? 17:26
rindolf And Mandriva has it. 17:29
nothingmuch: from conferences, etc?
nothingmuch the intersection of the set of flickr users and the set of people who attend conferences and take pictures 17:30
i really don't understand what type of information you're trying to get with these questions
17:31 ozo_ joined
nothingmuch food & 17:35
17:38 Psyche^ joined 17:39 [particle] joined 17:54 Psyche^ is now known as Patterner 18:06 jisom joined 18:10 fglock left, fglock joined 18:14 the_dormant joined
fglock nothingmuch: I think I got it now - my 'CALL' is the 'responder' implementation; I need to box it into a 'responder object', and make a 'responder interface' 18:15
hmm - simple like that?
nothingmuch CALL($obj, "foo") should have a body of something like UNBOX_RI($obj)->dispatch( method => Invocation::Method::Named->new("foo"), responder => $obj ); 18:18
rindolf is doing urpmi.update -a now. 18:19
fglock you mean, the objects knows it's dispatcher 18:20
s/objects/object/
the reason for this is so that you can have several MOs in the same program? 18:23
18:30 stevan__ joined
fglock nothingmuch: MO/t/si.t has commands like MO::Run::Aux::method_call( $base_obj_box, "elk" ) 18:31
nothingmuch those are helper routines 18:32
or opcodes, if you will
method_call( $mo_obj, $string ) is equiv to $perl_obj->$string
oops, missed the prev part 18:33
yes, the reason is that you can have several responder interfaces
from even more varied metaclasses
multiple languages' systems
multiple "concepts" in the same language 18:34
hand coded optimisations
whatever is fit
fglock how about the bridge between the implementations? does it just fits together
I mean, what if the method arguments belong to a different MO - it's probably handled by a 'coerce' routine somewhere else 18:37
ah - no, it's just the API that matters
18:38 calamare joined
nothingmuch nope 18:38
if the languages share a similar enough Invocation object 18:39
then there's no need for coercion
if they don't then that glue has to be done using a wrapper RI
the boxed type presented to e.g. perl 6 from $lang_foo will have a wrapper RI that translates invocations before delegating to the original ones
so basically what you said afterwords =)
fglock wrapper = MO::Run::Aux::method_call for example 18:40
nothingmuch nope
more like class ShimRI { method dispatch ( :$obj, :$method ) { Invocation::DifferentKind.new( .... ); # based on $method; $.wrapped_ri.dispatch( $obj, $new_invocation ) } } 18:41
and, of course, it wraps all return values as necessary, too
fglock ok - I guess I'll do some implementation and come back with more questions :) 18:42
nothingmuch cool
fglock brb 18:46
18:46 fglock left 19:17 fglock joined
fglock unbox($obj)->dispatch( $obj, $meth ) == now works in kp6 19:34
any MO people still around? 19:58
one remaining problem in kp6 OO: 19:59
containers currently have 2 dispatchers, one for plain methods, and one for VAR() methods 20:00
20:01 penk joined
fglock having 2 dispatchers in the same object doesn't seem to be against MO, so I'll just keep it 20:02
20:03 geoffb joined, marmic joined
Juerd Feather is going to change IP address soonish. 20:04
kolibrie new network? 20:07
20:11 Aankhen`` joined 20:15 veritos joined
veritos heya 20:17
pasteling "fglock" at 201.54.129.80 pasted "kp6 - working "new-MO" output of t/kp6/20-bit.t" (51 lines, 1.5K) at sial.org/pbot/24472 20:18
fglock veritos: hi 20:19
20:20 _bernhard joined
avar yay I sort of understand how mp6 works now:) 20:23
fglock avar: cool
"sort of"? :) 20:24
avar: now you can start with kp6
avar well how the p5 emitting backend differs from parrot
veritos phew. 20:25
i'm working on the python code generator.
fglock: so what did you think of that Freeze thing i showed you? certainly one thing python did right... 20:26
avar v6-Python?
veritos avar...not now
fglock the kp6 emitter is much simpler than mp6's
veritos i'm working on a backend for miniperl6
avar neat
[particle] compiling perl 6 to python 3000?
veritos [particle]: not 3000 20:27
avar hah
20:27 dduncan joined
fglock veritos: it would be nice to see v6->python->C working 20:27
[particle] vaporlang on vaporlang
veritos fglock: once i get this python emmiter working
fglock: we could have almost skipped parrot.
fglock [particle]: mp6 new-MO is out
[particle] great
that was fast! 20:28
fglock veritos: heh
[particle]: see the last paste
avar fglock: the kp6 emitting backends end up in mp6 though right? 20:29
demq hey particle did you notice my message the other day?
fglock [particle]: kp6->parrot is probably more doable now; it requires simpler semantics
20:29 demq is now known as dmq
veritos generating python is hard, you have to keep track of indent and everything.w 20:29
[particle] dmq: no 20:30
fglock avar: mp6 is just one backend; it does kp6->perl5 directly
avar veritos: ya, that sucks:)
dmq oh, just wanted to thank you for the pointer to the \v/\h esclists, and to say that ive added such support to p5.
so, thanks. :-) 20:31
was kinda curious actually what the rules were for generating the lists, just in case unicode X.0 adds new ones... 20:32
veritos maybe i could implement it as a two-pass compiler. first would leave comments like # DEDENT or # INDENT, and the second pass would change those to actual indents.
avar fglock: and it can do kp6->python|php|lisp|whatever easier than mp6->whatever?
fglock veritos: I wonder if there is a "pythontidy" that would do it for you
dmq just use { }
[particle] dmq: fantastic news
avar if 1+1: # { 20:33
print "ook":
# }
kolibrie fglock: should I start looking at kp6 as a replacement for v6.pm and PCR?
fglock avar: yes, the ast transforms make it easier to desugar
veritos fglock: no, because there's no way to tell. indentation works like {}, you can't work otherwise.
dmq i also added support for \R as per uniocde tr18: unicode.org/unicode/reports/tr18/
lambdabot Title: UTS #18: Unicode Regular Expressions
fglock kolibrie: sorry, not enough runtime yet
avar kolibrie: yeah it looks much nicer for what PCR does 20:34
kolibrie: not that I know...:)
kolibrie fglock: no problem (I'm not doing much to help)
I am getting ready to update my 'Practical Extraction with Perl6' talk 20:35
it's experienced some major bitrot I'm sure
avar veritos: it would be awesome if you could document some of this stuff as you go along:) 20:36
fglock kolibrie: but I guess it's time to start 6-on-6-on-kp6; the runtime will be written as-needed
avar veritos: writing custom backends that is
*nudge* *nudge*
veritos avar: i'm mostly taking the perl5 one and changing the emitted code.
kolibrie fglock: sounds great
[particle] dmq: thanks for the \R reminder, >>@todo
avar veritos: mm, and writing a match.pm in pythoon?
veritos avar: www.dabeaz.com/ply/ 20:37
lambdabot Title: PLY
dmq i dont think p6 is supposed to or needs \R in the sense of that article.
\R is supposed to be [^\r] in p6.
avar p6 pretty much doesn't need any escapes since you can just add them:)
<-[\r]> you mean:) 20:38
[particle] hrmm... i guess pge is *mostly* perl 6-based
dmq avar: er, right. :-)
[particle] it will support negated \r, so maybe that'll do
avar I don't think you can define custom \ escapen in p6 though 20:39
then again I don't know p6:)
veritos you could in p5...there was an example in ze camel.
fglock [particle]: re translating my own stuff - hmm - do you think people get lost?
[particle] fglock: yes, i do 20:40
dmq so no idea how those lists were determined? 20:41
[particle] it's a combination of things--not many understand compilers, nevermind bootstrapped compilers written in a subset of perl 6 :)
dmq: the \v \h lists?
i copied them from the spec
moritz avar: you _can_ define \-sequences in p6 ;-) 20:42
avar moritz: howz?
moritz token rule_backslash:<w> { ... } # define your own \w and \W
avar neat 20:43
moritz perlcabal.org/syn/S05.html#Syntactic_categories
lambdabot Title: S05
veritos python has some quirks though. there's no bare 'do this once block' like do{}, you have to say something like if True: yada 20:44
fglock: what's class Bind supposed to be? 20:45
a list of statements?
fglock veritos: $a := $b 20:46
veritos hmm...it seems to look like ->(){} from the syntax here 20:47
fglock or: [any-structure] := [other-structure] - but you don't have to support this syntax
20:47 vel joined
fglock Bind is used in many places, like parameter-passing 20:47
veritos go to like line 160 in the perl5 backend
fglock one sec
veritos: that part of the program transforms [$a,$b] := [1,2] into $a:=1; $b:=2 20:49
veritos fglock: oh, ok.
that works
right now i'm basically translating the bits of perl5 into python.
i haven't typed a := since i learned pascal. 20:51
fglock later & 20:53
avar perl6: the new pascal 20:55
geoffb As a young kid, my parents sent me to a computer day camp. It was tiered into basic and advanced students. I was considered a "basic" student because I didn't know Pascal, despite knowing at least one assembly language at the time. My, how times have changed.
[particle] you go to advanced day camp now?
Tene geoffb: that's odd.
geoffb [particle]: heh.
On the plus side, the basic students got to play with a programmable robot as their final project. So I spent most of my time coding it to drive various patterns around the classroom. 20:56
Early 80's robotics not so precise, btw. 20:58
Some day when I'm rich and bored I ought to get back to playing with robotics again ... 20:59
geoffb hopes embedded Parrot will be a reality by then
[particle] i sure haven't programmed in VAL-II in a while 21:00
operating a robot that could position 44lbs at any angle was fun, but without force feedback, quite dangerous 21:01
21:07 Caelum joined 21:09 desiderato joined 21:19 desiderato left 21:39 sapir joined 23:17 SubStack joined 23:25 kunwon1 joined
TimToady wonders how well Perl 6 will run on a TRIPS processor... 23:31
mugwump do they come from the frig? 23:40
23:42 kunwon1 joined
jrockway TRIPS = TimToady work project? 23:44
TimToady hardware.slashdot.org/article.pl?si...24/2021243
lambdabot Title: Slashdot | Next-Gen Processor Unveiled
jrockway ah, cool 23:45
TimToady we already have the requisite references to Erlang and Haskell in the commentary...
specbot6 r14378 | larry++ | cut-n-paste typo noticed by TheDamian++ 23:51
23:58 forsaken joined

Powered by Cro and the RakuĀ® Programming Language.

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

Thank you!