Pugs 6.2.8 released! | pugscode.org | pugs.kwiki.org | paste: sial.org/pbot/perl6 | www.geeksunite.net
Set by autrijus on 13 July 2005.
Darren_Duncan hiya 01:55
right now I'm in a car on the way to OSCON, in north Washington
speaking to you live over a GPRS bluetooth modem 01:56
brentdax Man, $0 -> $1 is gonna take a real long time to get used to. 03:39
(Or rather, the other way around.)
...could Undef be considered Perl 6's bottom type? It's compatible with any type, any method can be called on it with any parameter... 03:48
svnbot6 r5967 | putter++ | Copied modules/Grammars/rx_grammar.pl to rx_grammar0.pl. So PMichaud can overhaul rx_grammar.pl with abandon, while putter continues incrementally with rx_grammar0.pl. 03:55
r5968 | autrijus++ | * scope defaulting for parameters should not apply to primitives,
r5968 | autrijus++ | as they are not supposed to have a lexical scope anyway.
putter a quick release engineering note - the current failures of t/unspecced/p5 are odd. roundtrip.t and method.t fail in make test harness, but work when run individually. array.t fails in both, but in make test says 05:04
No compatible subroutine found: "&my_elems"
and by itself says 05:05
No compatible subroutine found: "&elems"
Odd. Good night.
autrijus hi stevan 05:13
how's life?
svnbot6 r5969 | autrijus++ | * "-Inf" should numify as -Inf, not 0.0.
autrijus stevan: you up for changelogging for the oscon release? if not I can do it too, but I'm still chasing testfails
hm, anyone remember if there was a ruling on mutating map/grep? 05:18
map { $_++ } @_;
according to S29 they are not mutating
svnbot6 r5970 | autrijus++ | * The `end` method on lists can't be parsed as an unary op...
r5971 | autrijus++ | * some TODOs; mutating map/grep needs a ruling
autrijus but I vaguely remember some discussion
brentdax autrijus: did you see my musing about Undef above? 05:43
autrijus sub identity (Any $x) returns Undef {}
somehow not ringing a bell
brentdax True. 05:44
autrijus also in p5, undef can't instantiate a list.
or a hash for that matter
stevan hola autrijus 05:45
autrijus and unless you want to change the semantics of (@x = undef) so @x is now a unthrown exception
I don't think using Undef as Bot will fly.
heya stevan
stevan when do you want the changelog for? I am about to get some sleep actually 05:46
brentdax Alright. It just seemed like All had magic properties, and Undef had magic properties, and maybe the magic lined up.
autrijus brentdax: it seems to me that undef is still a scalar
and not (say) an array
stevan: I don't know, in the next 36 hours
stevan autrijus: ok, I can do that :) 05:47
autrijus stevan++
stevan autrijus: one quick questions before I sleep :)
autrijus sure 05:48
stevan actually more than one question,.. but i will start with one :)
is the eventual goal to have the metamodel in pure PIL?
autrijus go ahead :)
well, pure PIL is the same as pure Perl6
so to speak
stevan but isn't PIL a subset of Perl 6? 05:49
autrijus I think it makes sense to base the reference implementation on that, yes.
well, that depends on how you define "subset"
stevan ok
autrijus Perl 6 is the external representation
PIL is the internal language
stevan ok
autrijus multiple perl 6 programs may compile to the same PIL node 05:50
(say "Hello") and (say "Hello") for example.
stevan but ideally the metamodel will be written in a way that is pretty close to pure PIL
autrijus (actually, no, they will differ in PPos, but what the heck)
stevan :)
autrijus stevan: no, any idiomatic Perl 6 that does not define new classes will do
QtPlatypus asks "So Perl6 is to PIL as M-expressions are to S-expressions?" 05:51
stevan ok
autrijus QtPlatypus: yes
stevan autrijus: so what is my smallest feature set?
autrijus stevan: perl5 without AUTOLOAD
i.e. you are free to use namespaces and closures.
stevan and no bless
autrijus you can bless, you just can't do method calls :)
stevan ok
autrijus (so the bless is just tagging)
stevan excellent 05:52
autrijus that will enable the runtime to only implement the functional core
and automagically receive a OO core
stevan yes
autrijus and then it can optimize from there.
stevan autrijus++
this is what I was hoping you would say 05:53
autrijus :D next question?
stevan no more questions actually :) 05:54
I am about to check in some work from this weekend
it is kind of messy
autrijus messy is good :)
stevan but it implements a MetaClass with no AUTOLOAD, and minimal method calls (like 2 or 3 in total in the entire thing) 05:55
svnbot6 r5972 | autrijus++ | * It's now okay to use a module without &import defined in it.
autrijus oooh.
stevan it also bootstraps the MetaClass so that it isa Perl6::Object
autrijus and then you can turn them into inside-out objects
aka closure dispatch
for those 2 or 3 calls
stevan no need
autrijus and then we'll be there
stevan I just call local functions
autrijus oh? normal function calls and case-expression will do?
excellent
stevan++ 05:56
stevan :)
I am really enjoying the CLOS MOP book too
autrijus nice 05:57
stevan although I can only get a few pages along before I have to go work onthe metamodel
autrijus :)
stevan it gets my mind working :)
ok, so I am going to check in the mess, and get some sleep
autrijus okay. enjoy!
stevan I have plans to clean it up
:) 05:58
autrijus my plan is to prototype it in PIL runcore
and then use quickcheck to establish observable/provable invariants
when this thing is sound (so I can write down using scary academic notation), backport from _that_ to idiomatic perl6
meanwhile we can release 6.28.0 before that
and get people experimenting with the model/runcore 05:59
stevan I figure I can remove the rest of the AUTOLOAD (the part used by the regular classes) very easily
autrijus wonderful
I want to see the mess now :)
stevan yes yes
writing my commit message :)
autrijus :)) I'll bbiab
stevan autrijus: commited r5973 06:02
svnbot6 r5973 | Stevan++ | Perl6::MetaModel - (p5)
r5973 | Stevan++ | * MetaClass does not use AUTOLOAD anymore, everything goes through ::dispatch
r5973 | Stevan++ | * MetaClass isa Perl6::Object (once everything is bootstrapped that is)
r5973 | Stevan++ | * MetaClass test has been de-sugared
r5973 | Stevan++ | * misc. other bootstrapping things as well
stevan here is a list of a few things to look at: 06:03
MetaClass.pm - all the bootstrapping is done inside the meta() method
this is the only remaining method call for MetaClass right now
I am thinking this can be replaced by a global ::meta() function which will maybe lookup the metaclass instance in a global registry or something 06:04
(I am stealing this from CLOS, which has a find-class function)
next look in MetaModel at the ::dispatch function 06:05
it is ugly right now, but I think I can clean some stuff up.
I fall into an infinite loop if MetaClass does not have a specific dispatch() function. I know why, but I am not sure how to break that circularity, so right now it is specialized, and just avoids method calls 06:07
The bootstrapping takes place at the bottom of this file (line 291)
all it does it load the Perl6::Object module and then makes Perl6::MetaClass->isa(Perl6::Object) 06:08
oh, and the 3rd arg to dispatch is a boolean for SUPER calls
I am going to remove that, I think this is a bad way to go about it, and I also have some thoughts as the relevancy of SUPER 06:09
(but that is for p6l)
Also look at line 82 of Perl6/Class.pm
it is a special case for building the Perl6::Object metaclass 06:10
all other classes build their metaclasses normally, the new() method defined in Perl6::Object
I am going to completely overhaul Perl6::Class 06:11
it will look more like the JS version when I am done I think
autrijus ok. 06:12
stevan I also have some plans for how to treat methods
I can use closures to handle the details of submethods and private methods
autrijus I believe that's as lwall intended.
stevan by just wrapping them
yes, I think it is 06:13
autrijus i.e. they should not be distinguishable from closure factories
if it is that's a bug etc
ok. I'll consume the Diet MetaModel on my way to $client
stevan ok, enjoy
autrijus I need to run. see you tomorrow when you're up :)
&
stevan yes 06:14
:)
nite :)
brentdax Is there going to be a Pugs release RSN? 06:22
QtPlatypus thinks that the next release may co-incide with the new PIL. "But that is pure speculation on my part" 06:23
autrijus brentdax: yes, in time to ingy's talk 06:29
36hrs is the current plan
from now
brentdax Alright. I'll probably make sure that WWW::Kontent's first release runs with that version, even if I don't release at the same time.
autrijus cool! 06:30
brentdax (It's just a preview--no user system, parsers, or complete documentation--but I think it'll be a good demo...) 06:31
autrijus indeed
be sure to post to p6a when you're there :)
I'll bbiab
Aankhen`` ?eval $?PUGS_VERSION 06:51
evalbot6 \'Perl6 User\'s Golfing System, version 6.2.8, July 13, 2005 (r5973)'
brentdax Okay, this bug is driving me *nuts*. 07:09
It's that "call a method with the same name on an attribute" thing again, but in a different place in the code. 07:10
And with a different method name.
Although this time there are no subclasses involved...hmm.
Yes, got a test case! 07:12
pugs -e 'class A { method foo() { say "A::foo" } }; class B { has $.a; submethod CREATE() { $.a=A.new } method foo() { say "B::foo"; $.a.foo } } B.new.foo' 07:13
Should print "B::foo\nA::foo\n", but instead it repeatedly prints "B::foo\n".
...wait, argh, no. 07:15
nothingmuch stevan: ping 09:26
svnbot6 r5974 | iblech++ | t/builtins/lists/mutating_listops: Added link to p6l post. 10:29
autrijus rehi 13:07
autrijus ponders between doing some PIL2 charting and chatting and relenging 13:08
scook0 autrijus: there's some fancy stuff in src/PIL.hs that's breaking Haddock at the moment 13:17
all that %i stuff -- just so you know
autrijus hm, maybe linear implicit parameters ain't that nice after all 13:20
wolverian heh, the pizza guy looked just like Larry 13:21
autrijus oh?
3 =:= 3 # patent nonsense, right? 13:22
There is also an identity test, "=:=", which tests whether two names
are bound to the same underlying variable. "$x =:= $y" would return
true in the above example.
in that case, there need never be a tied rvalue 13:23
stevan nothingmuch: pong 13:42
good morning all 13:43
autrijus yo stevan 13:47
I'd like to discuss a weird issue with you if you have a bit time.
stevan autrijus: sure
autrijus ok. consider
$x = 1;
vs
$x = $x;
the latter actually means
$x = FETCH($x);
(or rather fetch(tied($x)) , but I'll just write FETCH) 13:48
stevan ok
autrijus and the expression can be written as
$x.STORE($x.FETCH)
vs
$x.STORE(1)
okay so far?
stevan (tied() being the means of accessing the underlying object, correct?)
autrijus yes
ok. so we are dealing with two types 13:49
one is (Scalar of Int)
one is (Int)
stevan yes
autrijus you can assign into the first but not the second
you can add with the second but not (directly) the first
stevan correct
autrijus so, by writing
sub f (Int $x) { ... }
we are actually saying 13:50
sub f (Scalar of Int named '$x') { ... }
stevan hmmm
yes
autrijus with the coercion rule that promotes Int into a fresh container Int
I mean, container of Int
but!
stevan :)
autrijus we are told that the default argument is (is constant)
which seems to me that is actually _removing_ the (Scalar of) part 13:51
instead of as we assumed in hackathon, some weird readonly shim layer on it
you agree with this intuition?
stevan yes, what you say make sense
it is not good
but it makes sense
the logic is clean
autrijus cool. 13:52
so, it seems that
stevan what does that mean though?
autrijus it means that
sub f ($x is constant) { \$x }; ${f($_)} = 3
is instadeath
stevan yes, that makes sense
autrijus as we discussed in hackathon, but this time it can be detected statitcally 13:53
instead of till runtime
since it amounts to be saying
stevan cool :)
autrijus ${\3} = 3
now, this worldview is all well if we only have objects.
but!
we also have collection :-/
if $x is (Scalar of Int) and 3 is (Int) 13:54
and if @x is (Array of Int)
what is (1,2,3,4) ?
stevan ouch
autrijus that is the core ambiguity
stevan an array of constants
autrijus an array of constants?
no, it's a constant array of constants.
stevan ok, even worse :)
autrijus but the strange thing is, if you start using it destructively 13:55
($x, $y) = (1,2);
then it magically behaves as a tuple
so, what does a constant array mean again?
integral the LHS isn't a List of Scalar of Int ?
autrijus integral: it is. 13:56
nothingmuch jp;as
hola stevan
actually it's no longer relevant
stevan hey nothingmuch
autrijus: maybe ($x, $y) = (1,2); is just sugar 13:57
autrijus integral: so can we spell "Array of" as "Scalar of List of" ?
think carefully before answering, as I was stuck right here :)
(and still am)
stevan for something like: (let (multi-value-bind ($x $y)) (1 2))
mjl69 hi! :-)
autrijus stevan: that's ($x, $y) := (1,2)
stevan: the assignment is really destructive 13:58
integral hrm, I'm thinking no, since p5 has taught me that List != Array
stevan (forgive my LISPishness)
autrijus: why is it destructive?
autrijus stevan: ($x, $y) = ();
integral hi mjl69
mjl69 could someone take a look at this when they have a second... nopaste.snit.ch:8001/3538 maybe it's a common installation problem...
integral: hi!
stevan why is ($x, $y) = (); not the same as ($x, $y) = (undef, undef);
integral autrijus: isn't that where a List is different from an Array?
autrijus mjl69: macosx? 13:59
mjl69 yeah.
autrijus mjl69: gcc_select to 3.3 already and make clean'ed?
mjl69 I definitely did gcc_select to 3.3. I did make clean one time but I could try it again.
autrijus mjl69: try doing that again, maybe you had 4.0 .o files 14:00
mjl69 autrijus: thanks, back in a few minutes...
autrijus stevan: it is, effectively. let me think about it
Yaakov What an eager guy, that mjl69.
autrijus my @a := ($x, $y); push @a, 1;
that should be error right?
mjl69 :) 14:01
autrijus my @a := [$x, $y]; push @a, 1; # and this should work?
integral if the List is not promoted to an Array, yes, imho
nothingmuch What a cynical guy, that Yaakov.
autrijus What a guy, that nothingmuch.
Yaakov I am the inverse of a cynic, it turns out.
Heh.
autrijus mm very cinycal
stevan autrijus: neither of those examples make sense to me 14:02
pushing on an array ref should be no different then pushing on an array
autrijus stevan: aw. let's try another form.
stevan oh wait, sorry, didnt see the :=
autrijus ($x, $y).push(1);
better? :)
stevan ok
hmm, thats a problem
autrijus what I'm saying is whether we can describe an array as a container of a list 14:03
stevan what about: my @b = (1, 2, 3).push(4, 5, 6);
autrijus in exactly the same way a scalar as a container of a value.
you mean
(my @b = (1, 2, 3)).push(4, 5, 6);
stevan no
I mean what I typed.
autrijus if the rhs is an error then the expression as a whole is an error
stevan what I am thinking is that maybe (1, 2, 3) can be autopromoted 14:04
integral you shouldn't be able to mutate a "value", right?
autrijus integral: right.
I'm using "value" in the functional programming sense
stevan yes
autrijus so
(1,2,3).push(4)
is just like saying
stevan but is perl really that strict?
autrijus 5++
integral yes, definately, imo 14:05
autrijus both are a mutable action levelled to a nonmutable value
stevan what about (1, 2, 3).shift
same thing?
autrijus $ perl
shift(1,2,3)
Type of arg 1 to shift must be array (not list)
I think perl5 agrees with me :)
stevan ok
integral push/shift should just be taking the list from the box, making a new list, and putting it back
autrijus integral: good. 14:06
then there's two questions remaining.
$a = (1,2,3); $a.push(4);
this always works, because Scalar cannot ever contain a List object. correct?
integral "yes"
autrijus and 14:07
$a := (1,2,3);
is promoted to
$a := \(1,2,3);
exactly as lwall described.
and (@a := 1) is an error.
integral what's \()? And why not $a := [1,2,3] ? 14:08
autrijus $a := [1,2,3]; # $a is Scalar of Array of Int
$a := \(1,2,3); # $a is Scalar of List of Int
in the latter form you can't push into $a. 14:09
but the first form is just fine.
the thing is [] allocates mutable storage
integral I don't understand $a = (1,2,3) then :-/
autrijus integral: oh, it's easy. infix:<,> in scalar context allocates mutable storage.
but I'm thinking about let it not do that for binding. 14:10
otoh, maybe it should always allocate mutable storage just for consistency's sake. not sure
integral owww, that's my head hurting level reached ;-)
autrijus let's elide it for now; I'll QuickCheck it later 14:11
another question is, what is Hash containing.
I'd say it contains a Map.
integral List of Pair?
autrijus but another thought is List of Pair, yeah.
so in that view, Hash is a constrained type of Array 14:12
integral otherwise you've can only contruct Maps by magic from Lists of Pairs
autrijus where elements are always pairs
and further constained to guarantee fst uniqueness
that is, key uniqueness
integral but not with an ordering
autrijus oh? (%x)[3] 14:13
hm.
integral ordered hashes by default? isn't that a bad performance thing?
autrijus no, I mean unordered
but have that return a pair
ah. I'm confusing levels
stevan which pair?
autrijus any pair
stevan the 3rd one added? 14:14
autrijus but no, .[] is a list op, surely it doesn't work on hash
integral but the same pair under the constraints of keys/values in p5?
stevan no it shouldnt (IMO that is)
Adam_ autrijus, talk to me about this thing of larry 14:15
14:15 Adam_ is now known as Alias_
autrijus Alias_: ? 14:15
integral: the value part of hash is still a mystery to me.
Alias_ the perl 5 to perl 6 thing he was working on at the hackathon 14:16
madskills?
autrijus Alias_: yeah, perl-parse-dependent
aka ppd
Alias_ is there any information on it anywhere?
autrijus it re-adds everything folded away at compile time, serialize the op tree into xml and prints it.
Alias_ or just the source 14:17
Need to add some information to my talk
(in addition to this loverly photo www.spidereyeballs.com/os5/set1/sma...1625.html)
autrijus ok. you can safely say 1)it works less well than perlcc currently and 2)it's larry's repentance for toke.c. 14:18
(i.e. developing ppd is a lot of suffering and pain and I'm not sure he'll carry it through.)
wolverian is it a christian thing? :) 14:19
Alias_ But it's going to be good enough for a first cut at Perl 6 ports of things?
autrijus Alias_: not currently, I doubt ever.
Alias_ oh yuck
I thought that was the point? 14:20
autrijus o/~ Larry walked this onesome valley / He had to walk it by himself o/~
s/one/lone/
Alias_: that was the point, but I think @Larry seriously underestimated the efforts involved.
Alias_ chuckles
wolverian do you guys have recommendations for a Linux laptop?
Alias_ it is indeed a long and ugly road
wolverian ack, maybe I should ask in a better channel, sorry. 14:21
autrijus Alias_: I know you don't want PPI to be mistaken as a source parser, but really, a "regular form" of perl5 that can make both PPI and PerlTidy happy, may fare batter.
I mean, better.
Alias_ sure 14:22
And if anyone else can get around the whole "impossible" problem, I'll be happy
s/else//
(since I cheated)
autrijus also we can take two stages
a better, more-loseless more-regular B::Deparse for example. but it's still very tricky 14:23
Alias_ B::Deparse is a CODE serializer though
(in the CODE reference) sense 14:24
autrijus I'm sure we both understand the huge gap involved :)
Alias_ right
autrijus in comparison, ponie is relatively easy, and pugs-compiled-to-perl5 is downright trivial.
Alias_ But I think it does have a lot of value as a CODEref serializer
autrijus (but just in comparison; individually they are large tasks.)
Alias_ Not so sure trying to push it to a full blown thing is worthwhile 14:25
But then I'm willing to cut @Larry a lot of slack on the subject, since I don't know any internals
autrijus really the only one who knows that much about toke.c is $Larry.
$*Larry, that is.
you can ping him again on p6c if you like 14:26
or mail him privately to get a more uptodate status
Alias_ That's ok 14:27
Might hunt him down today...
I've got a few days yet to update (er... finish) the talk
autrijus integral: you still there? I'm thinking about @x as Data.Array.IO, %x as Data.HashTable and $x as Data.IORef. 14:28
Alias_: nod, please mail him to get the recent status; I'm only updated up to the hackathon
mjl69 autrijus, I made sure I was using gcc 3.3 and did make clean and got the same errors. 14:29
autrijus integral: modulo the fact that Data.Array.IO is not resizable, suppose it does
integral autrijus: I'm here, but let me check my haskell docs
autrijus integral: and the two fundamental value types are Obj and [Obj].
mjl69 busy working at his pesky day job, will peek back in in a few...
autrijus integral: and very strangely, you can't store [Obj] into $x directly. 14:30
mjl69: you using any embedded flags?
integral: the evaluation rule has that STORE on $x will promote [a] into (Array of a), recursively
so
$x = (1, (2, 3), 4);
er
integral bother, I'd forgotten nested lists were darn weird 14:31
autrijus bummer, I can't even write nested lists.
that is a good sign actually 14:32
[Obj] never gets promoted to Obj; it's always (Ref of Array of Obj)
so if I write
$x = (1, (scalar(2, 3)), 4);
then it first turns the inner (2,3) into mutable [2,3] 14:33
then store [1,[2,3],4] (all mutable) into $x.
integral yeah, that makes sense 14:34
autrijus so the moral of story is, to put [Obj] into Scalar, it's promoted to (Ref of Array of Obj); to put Obj into Array or Hash, it's promoted into [Obj] with one element
but then, what does
sub f (%x is constant) { }
means, really?
does it mean %x is a List of Pair that happens to be viewed as a hash? 14:35
integral autrijus: I think it should be a ConstantHash does Hash, and it's just an implementation detail if there's really no container
autrijus List of (Pair of (Str, Scalar))
integral: then why not ConstantArray does Array, why list? 14:36
and why not ConstantScalar does Scalar, why value?
or is hash really second-class here?
integral oh, I was thinking it should be same for the other two
autrijus oy, but there is an important difference. a ref to ConstScalar is not the same as a ref to rvalue 14:37
${\3}++ # compile time error
${\(my $x is constant)}++ # not really detectable
(if we adopt the ConstScalar view) 14:38
let's try this simple experiment
my $x is constant = 3; my $y = 4; $x := $y; $x = 5;
legal?
integral umm, no? 14:39
autrijus why not? "is constant" is just a normal tie-like layer
surely you can destroy the association that ties you to ConstScalar via rebind?
(I say 'no' too, because I think $x is as good as a rvalue, and binding should be disallowed) 14:40
it's like saying (3 := $y) to me.
integral yeah, it should. But then from a maintenance standpoint "is constant" is upfront, and shouldn't change
but "is constant" isn't a normal tie if it means the container is a ConstScalar not a Scalar
ugh, that didn't help 14:41
autrijus didn't help.
?eval my $a := 3; my $b; $a := $b; $a = 5 14:42
evalbot6 \5
autrijus I'd like that behaviour to die
?eval my $a is constant := 3; my $b; $a := $b; $a = 5
evalbot6 \5
autrijus er, sorry.
integral hmm, I was about to say that is constant restricts the type allowed in the binding for the container 14:43
autrijus hm. 14:45
that is a valid view. that says the pad always maps symbols to containers, never to values.
({return 3}.()) := 4; # but this is patently absurd? 14:46
integral yes, must be 14:47
autrijus note that (my $pi is constant = 3) is a special form already
unless we take "constant" to mean single-assignment. 14:48
mmm that's an even more interesting view.
integral hmm, what exactly is the LHS to := ? Is it a container, or is it a binding?
autrijus the LHS to := is a container.
or rather, a signature that contains containers.
integral so, if $a := $b; And then I do $b := $c, then $a =:= $c too? 14:49
autrijus for sure.
integral oh, so it's more like it mutates the LHS container into a proxy for the RHS container... 14:50
autrijus no... not so
see PIL.hs :) 14:51
it overwrites the LHS container with the Box of the RHS.
Box being the content of the container
which includes Id, Value, and maybe tiedness.
integral oh, a Box has a *pointer* to the storage, not the storage itself
autrijus aye.
three layers.
four if you count symbol table.
value -> box -> container -> name 14:52
integral hmm, well isn't the constness thing ok then?
autrijus how so? I'm saying that name may map to either value or container
integral my $a is constant; $a := $b; # oh, but then you can't update via $a, so you shouldn't be able to update via $b
autrijus right but that allows
oh wait 14:53
who decides that you can't update via $a?
what you said describes ($b := $a)
not the other way around.
integral but $a has "is constant" up front, so $a = 5 should be always disallowed, no?
autrijus so you mean "is constant" is annotating the symbol table? 14:54
integral yes
autrijus hm.
integral but then that's why I showed that action-at-a-distance := use ($a := $b; $b := $c; $a =:= $c) 14:55
autrijus ok, I think we are converging
what you are saying is that
instead of
value -> box -> container -> name
we have
1. (value -> box -> container) -> name
2. (value) -> name
and "is constant" is not a "tie" in box anymore 14:56
i.e. no ConstScalar
it's directly changing the shape of the thing the name points to.
integral hmm 14:57
yes
autrijus hm 14:58
my $x is constant is Tieable = 3; tie($x...)
allowed?
if allowed, that means (box) -> name
instead of (value) -> name
integral hmm, if it is allowed STORE will never be called, and if FETCH is called, you can't get any constant optimizations
autrijus that's fine, Tieable defeats optimization anyway 14:59
I'm just talking about the usefulness of FETCH.
I think we should keep that.
integral with the semantic that FETCH is called for every mention of $x ? 15:00
autrijus whenever $x occurs that demands a rvalue. 15:01
note that perl, like C and Java etc, adds implicit FETCH and STORE depends on context
and unlike ML etc.
integral ok
mjl69 autrijus: sorry to ask a question and disappear. Were should I look for embedded flags? 15:02
autrijus mjl69: if you didn't modify your environment, then you didn't add embedded flags
gaal @papers.push("Threads cannot be implemented as a library" => portal.acm.org/citation.cfm?id=1064978.1065042)
mjl69 oh yeah. I did set various environment variables as directed in the directions. Let me check which ones.. 15:03
autrijus mjl69: try disabling all embed first and slowly add more... 15:13
dudley mjl69: Those errors are usually caused by gcc 4.0 pollution somewhere. 15:27
svnbot6 r5975 | autrijus++ | * New formulation of containers, taking into account the
r5975 | autrijus++ | four combination of "tieable" and "constant" traits.
r5976 | autrijus++ | * $CALLER::_ is now again usable in defaults.
autrijus I'm running out of names. ;) 15:28
dudley mjl69: For example, if you've build parrot with gcc 4.0, you'll get those errors when you try to embed it into pugs
autrijus a Pad maps Symbol to Container; Container is a Cell typed with Scalar/Array/Hash; a Cell is either mutable or a constant Box; a Box is either a tieable box or a nontieable box. 15:29
the only mutable point here is the Cell; everything else is statically bound.
QtPlatypus Any problem in computer scince can be solved by an extra leval of indirection. 15:30
autrijus (and if it's constant, there's no way of changing the value except by introducing a new symbol with the same name in an inner scope)
QtPlatypus: how very sadly true
better names welcome though.
mjl69 dudley: maybe that's it. It could be the parrot compilation. 15:36
svnbot6 r5977 | autrijus++ | * reformat and todoify splat_rw.t
dudley I just confirmed that definitely causes it. Maybe that should be documented somewhere... :-) 15:37
rafl Hm, my svk eats up 80% of 512MB RAM. :-/ 15:39
mjl69 speaking of parrot... should PARROT_PATH be set to the path of where the source files are (the ones I downloaded) or to where the installation of parrot is? I installed parrot to the default location, but set PARROT_PATH to the directory of source files.
dudley mjl69: The source dir 15:40
clkao rafl: what are you doing?
mjl69 dudley: that's what I thought. Just wanted to be sure. Thanks! I am going to start over with everything when I get a few more free minutes. 15:41
dudley np
rafl clkao: A simple pull.
clkao rafl: eek. how simple? large repository? many changes? 15:44
rafl: or it's the first time you run pull. it's probably building some cache
and you are using 1.00 rather than 1.01, right?
rafl clkao: the pugs repository, small changes to 3 files. Not the first pull. v1.00 (debian) 15:45
clkao rafl: hmm. do you see 'auto-merge... ' message ? where does it hang? 15:48
rafl clkao: I don't see such a message. It doesn't really hang. It just uses a huge amount of memory which makes the system swaping. This occurs before any output is made. 15:51
svnbot6 r5978 | dudley++ | *Document link failures on MacOS due to inadvertent linking of gcc 4.0 objects. *Add me to AUTHORS.
clkao rafl: ya, that's a known problem for 1.00 when building copy cache. 15:52
rafl Hm. The Debian wishlist bug for 1.01 is already filed 11 days with no reaction from the maintainer.. 15:56
clkao eek 15:57
Aankhen`` Yay! I managed to build SVK 1.01! 17:00
autrijus woot! 17:01
Aankhen`` Though only 84% of the tests pass...
Then again, that's not as bad as I expected. :-D
clkao Aankhen``: huh? 84% of svk tests? how come? 17:04
GeJ Aankhen``: on which platform if i may ask? 17:06
Aankhen`` I have no idea.
I just followed the instructions on the Wiki.
(clkao)
GeJ: Win32.
GeJ hum, ok so it's not related on the problem I'm thinking of on FreeBSD.
17:11 brentdax_ is now known as brentdax
svnbot6 r5979 | autrijus++ | * change PIL from ST monad to STM monad. 17:22
mjl69 dudley++ # you were right, had to make clean parrot and do all over. no link errors this time. 17:24
dudley mjl69: cool :-) 17:30
gaal autrijus, ping 17:46
autrijus hm, treacherous perl5 17:57
gaal: pong
$ perl -e '$_++ for 1,2'
Modification of a read-only value attempted at -e line 1.
$ perl -e '$_++ for 1..2'
# ok 17:58
gaal autrijus, i'm assuming more or less (trecherous) perl5 semantics in use/import
wow
one's a list and one's a... list? :)
autrijus .. apparently promotes values into containers
gaal is it because of ..'s special iterator magic added sometime in 5.00x?
autrijus I have no idea. 17:59
I'm so happy I'm not reimplementing perl5.
gaal so i'm adding to EvalStyle a Bool evalInit member: if true, opEval attempts to run the function &import in the evalled code
problems:
1. how do i know the namespace to look in? 18:00
2. when we introduce finer export controls, we'll have to pass information to the parser, since that's where importation takes place now
autrijus the parser currently calls &import 18:01
gaal oh!
really?
autrijus yeah.
l586
gaal i'm talking about init actually, not symbol importation
autrijus Parser.hs
gaal looking
autrijus init too.
gaal cool!
autrijus :)
gaal does this get called in "require" too? or does dumpEnv control that? 18:02
ah, it's directly on &use, i see
does use Foo; use Foo call Foo::import twice? (it should) 18:03
anyway: i'm adding &no. 18:05
autrijus cool 18:06
gaal i'm not sure why Prim has op1 "use" at all though if the parser reduces this?
l255
autrijus to give it a different form than "require"?
I have no idea either
gaal ah, i think they share the responsibility. if i'm reading this correctly, l583 in Parser does 'App (Var "&use")...' which delegates some of the work to the prim (and eventually to Prim.Eval) 18:08
autrijus nod. 18:10
svnbot6 r5980 | autrijus++ | * lift up the "Id" from Container so that each container's Id 18:16
r5980 | autrijus++ | remains mutable even if the content is immutable.
gaal i don't understand the hack in Parser:588 18:23
it's supposed to tell whether the module defines &import 18:24
autrijus gaal: the hack is to check for &Foo::import.name
because there's no .defined yet
and autovivified Code has empty names.
gaal but what does .name mean?
aaahah!
i didn't know that existed. :)
autrijus :) 18:25
?eval &say.name
evalbot6 '&*say'
gaal well, &no is almost exactly the same as &use
autrijus ?eval &say::goodbye.name
evalbot6 ''
gaal ~/o say my name, say my name ~/o
ahem.
except that the hook it calls is unimport instead of import, and that it doesn't eval.
i'll see if i figure out how to factor this out. 18:26
mjl69 I hope that 'make test' finishes before I have to pack up and go home... can't wait to finish install and start playing. 18:27
gaal oh: what's val doing in line 589? why does App sub need it?
mjl69: you can start playing already, you have a pugs :)
stevan autrijus: any thoughts on the metamodel? should I continue along this direction?
autrijus stevan: yes, I think it's sanity
mjl69 gaal: that's what I mean, play pugs :-) 18:28
autrijus and I already has some good ideas of how to integrate it to PIL.hs runcore
stevan autrijus: I will take it that sanity is a good thing in this situation :)
gaal mjl69, i know. but if you're already testing then you already have an executable and you don't need for make test to complete to start playing :) 18:29
svnbot6 r5981 | iblech++ | * INSTALL: Wrap dudley++'s new paragraph.
r5981 | iblech++ | * ext/Algorithm-TokenBucket/t/test.t: Skip test which depends on low system load.
r5981 | iblech++ | * PIL2JS: s:g/GET/FETCH/ to align to PIL.Internals and Perl 5.
autrijus gaal: I think it's a typo. supposedly import's invocant is the caller package
stevan: yes. :)
gaal good. :)
mjl69 gaal: oh. good point. didn't think of that. Time to cmd-n a new terminal window... 18:30
dudley doh. Damn soft wrap.
mjl69 you know how we Mac users are. They want to take are terminals away because we are like a bunch of dangerous babies. 18:31
s/are/our
autrijus parsed "degenerate barbies" 18:32
obviously I need to sleep
mjl69 same thing :-) 18:33
EEEh! it's true, pugs is already there!
gaal unregenerate babbies?
mjl69 hello, worldbool::true 18:35
the output of my first pugs program
autrijus ?eval "Hello, world"
evalbot6 'Hello, world'
autrijus ?eval say "Hello, world"
evalbot6 Hello, world bool::true
gaal (i have the feeling the vast majority of people living in babbie, alabama are christians, though.)
autrijus waves. sleep - bbl
mjl69 autrijus++ 18:36
svnbot6 r5982 | iblech++ | PIL2JS: Step 1 of renaming PIL::Nodes to PIL and PIL::Papp, PIL::PLit, etc. 18:58
r5983 | iblech++ | PIL2JS: Step 2/3 of renaming.
r5984 | iblech++ | PIL2JS: Step 3/3 of renaming, PIL2JS works again now.
Aankhen`` G'night. 19:23
svnbot6 r5985 | gaal++ | have &use call &import with the caller package as first arg 19:36
integral hmm, &import is a sub rather than some kind of "package method" 19:40
iblech autrijus: Will the PIL2 for { my $a; { say $a } } look like { my $a; { say $OUTER::a } }? If so, implementing lexicals in PIL2JS without having to rely on JS's "lexicals" would be a bit easier :) 19:45
autrijus: (But it's not a problem if the answer is no :)) 19:46
xinming anyone here can do a prediction of perl 6 release time? :-) 19:49
integral Christmas ;-)
dudley xinming: This is the part where you say, "Which Christmas?" ;-) 19:50
Limbic_Region integral - that has been long proven to be misrepresented truth - it is no mystery why p6 will have a built in way to count the number of days till Easter - it must be better than PHP in all ways afterall
integral heh
Limbic_Region the count does indeed start from Christmas though which is where the rumor got started 19:51
Limbic_Region ducks
xinming In fact, I ever heard that there might be a rc of perl 6 relase this year. No sure. 19:52
integral had the impressive pm was still being optimistic
Limbic_Region xinming - there was a FAQ stating that there might be something out by the 3rd quarter of this year - it has been removed 19:53
no one knows for sure when it will be out - hence the joke about "Christmas" with no year 19:54
Pugs happens to be the furthest along (and AFAIK the only actively developed) p6 compiler
you can write an impressive amount of p6 code today 19:55
integral the problem comes when you encounter an area that was made up on the spot ...
iblech And you can even compile 62.68% to JavaScript :)
integral iblech++
Limbic_Region iblech++ 19:57
iblech Thanks, thanks :)
Limbic_Region but only 62.68% of the X% of p6 that Pugs currently supports (and has tests for I would imagine)
or rather - what exactly does that 62.68% represent? 19:58
iblech Oh, right. 62.68% of Pugs' testsuite
(36 unexpected success :) -- But many tests (t/oo, t/packages) don't even compile to PIL) 19:59
Limbic_Region so it is 62.68% of the Pugs' test suite that compiles to PIL, which represents some percentage of p6 implemented by Pugs (hopefully we have 100% code coverage but I am not sure), which represents some percentage of how much of the p6 specification that has been completed (my guess is around 85%) which represents some portion of the entire spec (as in it hasn't been written down yet), which is more likely around 40% 20:01
iblech Right. 20:02
Limbic_Region still quite impressive
autrijus iblech: the answer is "yes". 20:15
svnbot6 r5986 | autrijus++ | * thoroughly decouple Tieable and "is rw"; if a container is neither,
r5986 | autrijus++ | then it is a pure value. This means constant containers cannot
r5986 | autrijus++ | ever be rebound.
autrijus Limbic_Region: instead of 85% I'd say 25%. 20:16
but the other numbers agree with my gut feeling
iblech autrijus: Great :)
szabgab hi nothingmuch ! 20:17
autrijus iblech: the idea is to enable all lexicals always addressable by absolute number anywhere.
iblech: you still need to provide closures
but at least that invariant makes closures far easier to simulate.
(this is also so we can avoid the completely broken parrot lexical pad.) 20:18
iblech autrijus: Makes perfect sense
autrijus iblech: good. I hope my r5986 also makes perfect sense to you. 20:19
iblech Looking
autrijus the basic premise is that anything marked "is constant" is indeed constant and is always inlinable.
if it is constant and tieable, then you need to point to the mutable cell that represents the tiedness 20:20
some obvious consequences are: 20:21
sub f ($x, $y) { $x := $y } # error
sub f ($x is rw, $y) { $x := $y } # fine
iblech Makes perfetct sense. :)
autrijus yay. :)
also, suppose %ENV is constant is tied 20:22
(which it wasn't in the current example)
Limbic_Region autrijus - of the known specifications available - you think only 1/4 has been implemented?
autrijus Limbic_Region: aye
iblech: then you can freely tie/untie it
Limbic_Region from the synopses - or from decisions made on the list that aren't reflected in the docs or both? 20:23
autrijus iblech: but it's =:= stays invariant.
so if you have a constant but tied scalar, it is conceivable that you can still assign to it whilst it is tied.
iblech Right, makes sense. 20:24
autrijus and all these information are made upfront at compile time
and never changeable, so reduction rule is straightforward.
Limbic_Region: both
Limbic_Region: but the delta is insignificant 20:25
Limbic_Region: see, we don't even have roles :)
and the container types are quite bogus, as anyone using "is rw" and references can attest
etc.
but we're fixing it faster than before. :) 20:26
Limbic_Region hmmm - I think we are looking at this from two different perspectives - but that's the great thing about statistics - you can twist them to lie how ever you want 20:28
I was thinking of objects as 1 defined unit though that 1 unit likely represents a much larger portion of the language then I was previously giving it credit for 20:29
nothingmuch evening
Limbic_Region salutations
nothingmuch szabgab: ping 20:31
svnbot6 r5987 | autrijus++ | * Remove the old vtable-based tie interface. 20:37
r5987 | autrijus++ | * Make all container contents generatable via Arbitrary.
iblech sub foo (?$a = $a) # ok? i.e., does the second $a refer to the first $a? 20:41
Hm, I think so.
autrijus yes. that line is as good as saying ?$a. 20:46
my $a = $a;
hm. perl does not ever have list of lists as a value, right? 20:48
integral hmm, so this rule means you can do stuff like my &bottom = { &bottom.() }
autrijus it's always list of array references.
integral: yes.
20:48 moan is now known as moan_
autrijus ?eval my &bot = { &bot() }; bot 20:48
evalbot6 (no output) 20:49
autrijus good.
Khisanth that is not infinitely recursive?
autrijus that is.
integral pity _|_ isn't a word
iblech autrijus: re LoL, right, I think.
20:51 moan_ is now known as moan
autrijus integral: _L 20:52
?eval my &_L = { _L }; _L
evalbot6 (no output)
integral is that the time out, or some really cool bottom detection?
autrijus timeout. 20:53
integral hmm, I suppose the hard bit of really good detection would be signals 20:54
21:01 brentdax_ is now known as brentdax
moan Kind of a faq, i guess, but nevertheless: Are the synopses in sync with the language pugs is currently implementing? 21:01
mugwump that's the idea, moan, but pugs doesn't cover all the synopses specified behaviour 21:05
nor do the synopses cover certain more 'out there' features aiui, like continuations. tests for those features are in t/unspecced/ 21:06
moan thanks. Finally, I got something fresh to read, then... and to type, afterwards 21:07
autrijus iblech: still there? 21:10
iblech autrijus: Yep
autrijus ok. r5988 21:11
finally gave a reasonable (to me) semantics to evaluation results
-- | Any PIL expression can only evaluate to one of three value results.
svnbot6 r5988 | autrijus++ | * formalizing the evaluation results domain
autrijus Void | Item Item | List [Item]
integral hmm, is ./pugs -I ext/Test/lib t/unspecced/cont.t the best way to run a test file manually?
autrijus integral: or "prove"
integral prove doesn't seem to work for uninstalled pugs 21:12
autrijus iblech: the definition of "Item" is from S02
integral: right.
iblech: then there's the fact that Ref always point to a container.
the last bit I'm less sure of; conceivable we can distinguish between container refs and value refs.
or we can elect to promot values to constant containers and ref them. 21:13
the delta seems to be on Void; do we have a way to obtain a reference to Void?
(no container can contain Void directly)
in perl5, \() is the same as ()
iblech I'd take option #2, promoting values to constant containers 21:14
autrijus that means
\(Foo.new)
is actually
my $temp is constant = Foo.new; \$temp;
and \(1,2,3,4) would be 21:15
my @temp is constant = (1,2,3,4); \@temp;
(disregarding the fact that \ is a listop for now, suppose it isn't)
that sounds sane to you?
iblech Right, makes sense.
Khisanth so it is no longer \1,\2,\3,\4 ?
autrijus Khisanth: I don't know, I'm assuming it still is 21:16
?eval \(1,2,3,4)
evalbot6 [1, 2, 3, 4]
autrijus heh.
I'm defeated by pugs :)
I don't know, I can argue bothways
Khisanth ah to be defeated by you own creation ... isn't that always a goal? :) 21:17
autrijus :)
iblech: do you think Void/Item/List is comprehensive enough? I'm a bit uncertain about whether Junc is also a fundamental val type. 21:18
otoh, I can argue that Junc is always an Item
but there is still the question whether it's a Obj or a intrinsic.
s/or a/or also a/ 21:19
i.e. whether it makes sense to talk about (1|2) as an unboxed intrinsic.
iblech Hm, I'd probably see Junc as an Item.
autrijus but what about Pair? 21:20
where does the madness end?
:)
iblech :)
Hm, the pair thing
I'd take this to Item, and specialcase only in the binding code
autrijus so you mean both are intrinsic?
iblech I.e. treat Pair as an otherwise normal object
autrijus nono, normal objects are Obj 21:21
I shouldn't confuse you with the Item terminology
let's call it something else.
iblech :)
autrijus data Val = Void | Single Single | Plural [Single] 21:22
data Single = Bit Bit | Int Int | Num Num | Str Str | Ref Ref | Obj Obj
a bit better?
iblech Ok.
Yep.
autrijus ok. the question is whether Junc and Pair are Single variants.
i.e. whether they can exist in an unboxed form. 21:23
iblech I'd place Pair and Junc on the same level as Bit, Int, etc., i.e. make them Singles
autrijus i.e. non-object values
anything else that come to mind?
type?
iblech I'd make Type a Single, too. 21:24
autrijus do note that each Single has its lowercase types :)
my pair $x = (1=>2);
iblech Even my obj $x? 21:25
autrijus no, that is the assumed form
you always get Obj unless you declare one of the other forms.
iblech Ok. 21:26
autrijus ok. so we have pair, junc and type as the non-object intrinsics.
dare I suggest Code?
iblech Yes. 21:27
autrijus ok, I'm out of madness bits :)
mugwump I thought function application was the only intrinsic!
iblech :)
autrijus beats mugwump with church numbers
handles and rules can survive without being intrinsics. 21:28
ok. I think I can argue that pair, junc, type and code fits in the "compact storage" mental model 21:30
while everything else can't 21:31
so, is Pair just two containers 21:32
or two singles?
two singles, I'd argue. 21:33
hm, not neccessarily. they mean different things
two containers feel less broken. 21:34
iblech Right, and that's the way pairs are handles in PIL2JS, too.
Consider $pair.key = ...
autrijus nod. 21:35
but junctions are Singles
as lvalue junction is doomed
ok, the new Val is in. I can rest in peace for a bit... the new Eval would be fun. 21:45
svnbot6 r5989 | autrijus++ | * Completed first cut at the Val domain of new PIL runcore.
r5989 | autrijus++ | iblech++ for joining my madness in adding four apocryphal
r5989 | autrijus++ | unboxed types: Pair, Junc, Type, Code.
autrijus &
QtPlatypus autrijus: Shouldn't the unboxed types be called pair, junc, type, code ? I thought the convention was caps for objects and boxed types, no caps for unboxed types. 21:58
iblech Right. But Haskell demands that constructors are ucfirst. 22:00
Of course, the unboxed types as seen from Perl 6 will be lowercase.
QtPlatypus nods. 22:12
buu perlbot: karma autrijus 22:16
jabbot buu: autrijus has neutral karma
perlbot Karma for autrijus: 11
nothingmuch perlbot: karma iblech 22:17
jabbot nothingmuch: iblech has neutral karma
perlbot Karma for iblech: 27
svnbot6 r5990 | iblech++ | * t/var/var.t: Added 25 more tests.
r5990 | iblech++ | * PIL2JS: Real lexicals. $OUTER::. Tree traversing is fun :)
r5990 | iblech++ | * Previously, PIL2JS relied on JavaScript's broken "lexical" semantics (i.e.
r5990 | iblech++ | "alert(foo); var foo" works). Now, PIL/PIL2JS handles lexicals.
r5990 | iblech++ | * Thus, "say $a; my $a" is now illegal, and the $sub in
nothingmuch that's not making any sense
svnbot6 r5990 | iblech++ | my $a = 3; my $sub = { $a++ }; { my $a = 9; say $a; say $sub() }
r5990 | iblech++ | captures the correct $a. The downside is that 05-sub.t stopped to pass, but
r5990 | iblech++ | that problem will go away as soon as PIL2 enters the world. :) (Other tests
r5990 | iblech++ | still work.)
r5990 | iblech++ | * Because PIL2JS handles part of the lexical allocation thing, we got
r5990 | iblech++ | $OUTER:: for free. :)
iblech nothingmuch: Probably the database was reset or so 22:18
perlbot: karma iblech
jabbot iblech: iblech has karma of 337
perlbot Karma for iblech: 27
iblech hrm
nothingmuch jabbot: you need to stop drinking
jabbot nothingmuch: Oh, I need to stop drinking?
svnbot6 r5991 | iblech++ | * t/magicals/dollar_underscore.t: s/eval/try/. 22:55
r5991 | iblech++ | * t/builtins/arrays_and_hashes/pick.t: Very minor comment fix.
r5991 | iblech++ | * PIL2JS:
r5991 | iblech++ | * PIL::PVal: Fix compilation of VBools (I accidently broke it with the
r5991 | iblech++ | ->fixup commit r5990).
r5991 | iblech++ | * Prelude::JS::ControlFlow: &statement_control:<loop> should respect Perl's
r5991 | iblech++ | idea of booleaness. Pro: Some more tests should pass (resmoking), con:
r5991 | iblech++ | additional slowness... but "I can get wrong results fast, too" :)
brentdax Is postfix:<++> (Str) still magical in Perl 6, and if so, is that implemented in Pugs? 23:35
iblech Both yes: 23:36
?eval my $a = "a"; $a++; $a++; $a++; $a 23:37
evalbot6 \'d'
buu ?eval my $a = "d"; $a--;
evalbot6 'd'
buu awhh
ayrnieu ?eval my $a = "d"; $a--; $a 23:40
evalbot6 \-1.0
wolverian haha.
ayrnieu similar to perl 5. perl -le '$a = "a"; $a--; print $a'
postfix ++ hath magic. 23:41

Powered by Cro and the Raku® Programming Language.

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

Thank you!