6.2.7 released! | pugscode.org/ <Overview Journal Logs> | pugs.kwiki.org | paste: sial.org/pbot/perl6
Set by autrijus on 13 June 2005.
stevan_ I think making roles work in an easily predictable way is extrememly important to their success 00:00
since they are just another level of class decompostion I dont expect anyone but the OO hackers to use them
but they should still be accessible
so I really want to iron out all the edge cases if we can 00:01
ok back to converstation
breadth first is basically what you get when you flatten them
except there is no overlap 00:02
it is important in role composition that it is not order dependent
which is trivial is you flatten 00:03
because instead of shadowing (like in depth first traversal) conflict are just illegal
so order doesnt matter
mugwump but it did matter at some point, when the roles themselves were combined 00:05
stevan_ nope
it doesnt matter
if they conflict, ... BOOM, .... error
mugwump ok
stevan_ so order is irrelecvant
mugwump so the resolution process basically consists of disallowing the cases where it does matter. fine by me, so long as I can still MMD in methods etc :) 00:06
stevan_ yes
although MMD is a whole other topic :) 00:07
but thats not my dept.
7 floors up and to the right, ask for autrijus, he is how you need to talk too :)
mugwump sure. of course you can simplify MMD in concept by viewing the entire signature as a part of the method "name" 00:08
but that's not important right now
stevan_ yes, but thats just C++ style, and its not as cool 00:09
we need type dispatched
Yuval (nothingmuch) has this whole idea of first class signatures which is really cool
but yes, not important right now
mugwump ok, so we've got these Roles which are simply modelled as a collection of functions, including accessors for "private" attributes which are specially named so they can't conflict or even be accessed between roles etc 00:11
stevan_ uhm
thats adding restrictions to the roles though 00:12
according to @Larry, there is no such restriction
(for the record I agree with you, but its not my language)
mugwump :)
but what restriction does Larry speak of? 00:13
stevan_ LOL
none
thats the problem :)
mugwump Unless the difference is, that role methods are actually considered a part of the class they are composed into, and so can see each other's private attributes and methods 00:15
stevan_ yu[
yup
thats the idea
Roles are flattened in , and tossed out 00:16
think of the Role as a cellophane wrapper
you open it, put the contents in your class
and toss the wrapper out
maybe hold onto the instructions (meta-data) to handle things like does(), but thats all 00:17
this is how the current perl5 prototype works (without the state part
clkao mugwump: oh, what's up with the store-password thing in svn::mirror?
mugwump clkao: well, I think the thing missing is the connection between the svn config object and the auth baton 00:19
but these objects are pretty much opaque and I couldn't find the docs when I was looking
mugwump ==> #svk 00:20
stevan_ wanders off for some foo 00:25
mugwump feeds stevan_ some foo at the bar 00:31
Khisanth what exactly is a parrot language? 00:36
mugwump PIR? 00:37
PASM?
Khisanth ah those
stevan_ ahhh the smell of schwarma 01:00
poor Luke, this room is going to stink for a while ;) 01:01
mugwump ok stevan_, I see now why there is a state problem. 01:02
stevan_ mugwump: its all about the flattening 01:03
without it, you are correct, its not an issue
but with it, its bad
mmm, the hummus is good too
Toronto has really great food I have to say 01:04
mugwump So, there are two differences between Roles and Classes. Attribute accessing and method invocation of roles is considered in the scope of the class which it is attached to
(and roles are not instantiated)
stevan_ yes
think of the {} braces on the class as a closure 01:05
which is evaluated at compile time and results in the metamodel being constructed
which can potentially have some really interesting effects if you can access the metaclass in some way 01:07
class MyClass { for ('method1', 'method2') -> $label { $?CLASS.meta.add_method($label => sub { print $label } } } 01:08
stuff like that could (theoretically) be possible
mugwump it's like Roles are non-composable Classes that have a parametric *scope*
stevan_ yes
and Roles can be parameterized too
role Something [ .... T ] { has T $.foo } or something like that, I forget the syntax right now 01:09
mugwump The Set or Tree classes are the ones that would need this and exist as working examples I guess 01:10
stevan_ yes actually it would allow for type checking of the nodes of the tree 01:11
mugwump So, the scope is basically the same thing as the "local" dispatch table (local being the composed dispatch table for a class)
stevan_ roughly yes 01:12
mugwump for now I'm putting the ::Type object, and the "local" dispatch table into the same little black box
stevan_ I am not sure about inner classes and such
mugwump so being typable means that you get a scope
stevan_ mugwump: yes thats pretty safe I think
mugwump So, still it is the ::Type object that this "magic" is happening, all that's changed is that we're now using it for the Class' methods' scope 01:13
and as the Role is not typable, their methods' scope must be attached to a ::Type 01:14
but you don't need to throw them away in the MetaModel, especially given you still need to answer .does
stevan_ currently I gather all the info I need to implement does before I toss it 01:15
QtPlatypus ?eval class Foo { has &.subat; method retsub () { .subat()}};my $foo= Foo.new(subat => {1 + 1});$foo.retsub; 01:16
evalbot6 \sub {...}
QtPlatypus That wasn't what I expected.
?eval class Foo { has &.subat; method retsub () { .subat()}};my $foo= Foo.new(subat => {1 + 1});$foo.retsub.(); 01:17
evalbot6 2
QtPlatypus ?eval class Foo { has &.subat; method retsub () { .subat()()}};my $foo= Foo.new(subat => {1 + 1});$foo.retsub;
evalbot6 2
QtPlatypus ?eval class Foo { has &.subat; method retsub () { .subat}};my $foo= Foo.new(subat => {1 + 1});$foo.retsub; 01:18
evalbot6 \sub {...}
QtPlatypus ?eval class Foo { has &.subat; method retsub () { .subat}};my $foo= Foo.new(subat => {1 + 1});$foo.retsub.perl
evalbot6 '\\sub {...}'
QtPlatypus Anyone care to explain why it works that way? 01:19
stevan_ QtPlatypus: no :) 01:20
I think the first one "method retsub () { .subat()}" is probably using the accessor
and so it this one (maybe) "method retsub () { .subat()}" 01:21
this is ugly, but it works "method retsub () { .subat()()"
?eval class Foo { has &.subat; method retsub () { &.subat()}};my $foo= Foo.new(subat => {1 + 1});$foo.retsub;
evalbot6 2
QtPlatypus That makes sence. How would I use the attribute directly?
stevan_ like that
QtPlatypus Arh
Ok things make sence. 01:22
Odin-LAP sense, actually. :p 01:24
svnbot6 r5083, mugwump++ | Various updates to the pure Perl 6 MetaModel (non-PIL version) to reflect updated interpretation of Roles and Classes 03:07
r5084, Stevan++ | Perl6-MetaModel : there is nothing like refactoring with +93% coverage :) ... mostly just cleanup of the Perl6::Object module, it is getting pretty slimmed down now
mugwump stevan: ping 03:10
stevan mugwump: pong 03:11
I see your update
mugwump excellent. did it make any sense?
stevan havent looked yet
mugwump ok
stevan I am doing so right now 03:12
mugwump in particular is the Method.pm
stevan ok
mugwump At runtime, a method has *two* types
one type is the $object.ref
The other is the type that corresponds to the type the role was included into 03:13
stevan ok 03:14
mugwump the first is used for public accessors and methods, and is subject to dispatch through the Class' superclasses
the second is used for private accessors and methods
so, the first is bound "late"
the second is bound "early"
and a Class is merely a Role that very quickly has all the private parts of its methods bound 03:16
stevan interesting 03:18
how would you deal with conflicts? 03:19
mugwump when would they be detected?
stevan class composition time
or role inclusion time 03:20
mugwump same thing :)
stevan any time you compose the ::Type
yes, basically
mugwump yes
the actual rules for the conflict resolution are irrelevant to this, of course. 03:21
stevan yes
mugwump so long as they don't leave the ::Type objects with ambiguous stuff in them
stevan but thats the hard part :)
mugwump yes, but minimal rules which exclude many multi methods which could co-exist Shouldn't Be Hard™ to implement 03:22
svnbot6 r5085, mugwump++ | Ensure Role methods can be re-used between Classes 04:11
r5086, luqui++ | r147@Queen-of-Hearts: luqui | 2005-06-28 21:56:55 -0400 04:19
r5086, luqui++ | Added some type inferrence musings.
r5087, luqui++ | r148@Queen-of-Hearts: luqui | 2005-06-28 22:14:01 -0400
r5087, luqui++ | On disjoint sets.
r5088, luqui++ | r149@Queen-of-Hearts: luqui | 2005-06-28 22:29:45 -0400
r5088, luqui++ | Perhaps we need a new syntax for type variables, or we need to reassign ::.
r5089, mugwump++ | Method dispatch only happens on the compiled dispatch tables, not bothering with the Meta objects 04:26
obra seen autrijus 04:43
jabbot obra: autrijus was seen 7 hours 20 minutes 34 seconds ago
nothingmuch evening 05:47
seen autrijus 05:49
jabbot nothingmuch: autrijus was seen 8 hours 26 minutes 22 seconds ago
nothingmuch he was supposed to wake up by now ;-)
Khisanth go knock on his door? 05:51
nothingmuch i'm not staying at the hotel
and it's not nice to wake someone up in the middle of their beau^H^H^H^H^Hhacking sleep 05:52
. o O ( why are there so many conflicts ? ) 05:58
svnbot6 r5090, nothingmuch++ | array container implementation
r5091, nothingmuch++ | array is compilable, and tested.. still broken
r5092, nothingmuch++ | Sorry stevan++, here's Scalar 05:59
Khisanth wonders if you can do ../../../../method 06:04
nothingmuch Khisanth: .. being superclass or caller? 06:05
;-)
jql Didn't &prefix:<../> get shot down? 06:06
nothingmuch uh
i thought it was a joke 06:07
mugwump That's a really good idea!
jql well, considering how ./ isn't a joke. :)
revdiablo I think Juerd was serious about ../ for about 12 hours
jql heh
mugwump and then? :)
QtPlatypus I thought that it got shot down.
jql and then we wondered "how could anyone use that?" 06:08
well, I wondered it
revdiablo he woke up the next day and realized
jql I dunno what @Larry thought
Khisanth well if there is going to be the ugly ./method why not go along with it and have ../../method too? :)
mugwump ok that's fine for SUPER, but what about NEXT? 06:09
jql ./+method # bring it
revdiablo .->method # ;)
nothingmuch that's why file systems need continuations 06:10
coral i like talk like that 06:11
nothingmuch++
coral gets out the pom-poms
nothingmuch bah, it's 2 am
coral could perlfs.sourceforge.net/ be ported to pugs? 06:12
nothingmuch and I'm screwing around with PIL underlying runtime stuff
instead of working on my talk
Khisanth nothingmuch: following in tradition? 06:16
svnbot6 r5093, nothingmuch++ | Testing is good for you
gaal hello. what should the prelude be compiled to with the fast-prelude scenario? pugs ast? something else? what target do we always have? what would be easy to dump partially and later inject to the environment? 06:17
PIL? 06:18
revdiablo previously known as PAST
nothingmuch hola PIL 06:19
err, gaal
gaal that's why it was called that.
it was called past in the past.
revdiablo heh
in the past PIL was called PAST, duh.
Khisanth DR. PHIL 06:20
gaal aloha nm
nothingmuch gaal: i have a new laptop 06:21
gaal "become new"! :)
nothingmuch i actually had it before, but i forgot to tell you
thanks =)
gaal (not sure they speak inglezit on the chanell)
nothingmuch uh, how did an ingy module work without YAML installed?! 06:23
holy crap, this computer is so fast 06:24
nothingmuch is going to run the Regexp::Common test suite, just to see it go fast
gaal why does script/pugscc still mention 'dump.ast'? i thought that was obsoleted?
....why is pugscc still written in p5? :)
nothingmuch uh 06:25
gaal i 80% ported it once when i wanted to fix the cygwin path ickiness, but was missing some features. i imagine those are already available 06:26
but well, i guess it works now and is probably faster than a p6 version... 06:27
gaal looks for food 06:29
Khisanth hmm
where can I find more info on .as()? or can that bible thing tell me? :) 06:46
nothingmuch woah, i'm 20 07:06
gaal happy birthday!
obra happy birthday, nothingmuch 07:08
gaal ?eval sub f {say "not said"}; f(); say "said" 07:39
evalbot6 said bool::true
gaal why is "not said" not said?
QtPlatypus could be the say saftly thing 07:40
gaal ?eval sub fatal { die; die; die } fatal(); say "immortal"
evalbot6 immortal bool::true
gaal nope, same thing in one-liner. 07:41
Khisanth hmm
gaal dispatching is very b0rked!
?eval no_such(); say "alive"
evalbot6 Error: No compatible subroutine found: "&no_such"
gaal this is very bad :) 07:42
Khisanth hmm do that enough times and I get undef 07:43
instead of bool::true
gaal the bool::true is simply the valuation of the last say
QtPlatypus ?eval my $a=1;sub foo () {$a++}; foo(); $a 07:44
evalbot6 \1
Khisanth but why should say start returning undef all of a sudden?
gaal why should functions not be called all of a sudden? the current r is not sane.
Khisanth well that is for sure :) 07:45
gaal svn log -v's for possible recent culprits 07:46
Khisanth ok returning doesn't seem to work at all 07:47
gaal backtracing, 5039 now 07:51
make unoptimized++
b0rkage == r5040 07:53
hmm, i don't understand why this breaks! evalExp circa line 309 *does* get called. 08:18

Powered by Cro and the Raku® Programming Language.

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

Thank you!