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.
ayrnieu Are you sure? 00:07
brentdax Pretty sure. There's supposedly a deadline and stuff. 00:14
milna hello 00:30
Does anyone know of a good tutorial on haskell
I have tried reading Pugs, but I dont understand it
meppl i think there are many links on the haskell-website itself 00:44
www.haskell.org/learning.html 00:45
dudley milna: If you want to shell out the money, Haskell: The Craft of Functional Programming is worth it 00:49
ayrnieu milna - #Haskell also knows. 00:51
^conner_ There's been a request on #perl for Signatures::Readable::English 01:51
02:16 brentdax_ is now known as brentdax
autrijus greetings, Λcamels 04:38
Darren_Duncan howdy 04:40
autrijus :) 04:41
Khisanth that is suppose to be an uppercase lambda? 04:46
gaal morning 07:18
08:02 coral_ is now known as coral
nothingmuch hola 09:01
Aankhen`` ?eval role Foo { method bar () { "bar" } }; class Bar { does Foo; }; my $bar = Bar.new(); $bar.bar() 10:07
evalbot6 'bar'
Aankhen`` ?eval $?PUGS_VERSION 11:01
evalbot6 \'Perl6 User\'s Golfing System, version 6.2.8, July 13, 2005 (r5904)'
gaal ?eval ({a=>1,b=>2}).perl 11:18
evalbot6 '{(\'a\' => 1), (\'b\' => 2)}'
gaal ?eval {a=>1,b=>2}.perl
evalbot6 '{(\'a\' => 1), (\'b\' => 2)}'
gaal on the command line, this:
./pugs -e 'say {a=>1,b=>2}.perl'
gives this:
a 1 b 2\undef 11:19
Aankhen`` Same Pugs version?
gaal r5904
?eval $PUGS_VERSION
evalbot6 Error: Undeclared variable: "$PUGS_VERSION"
Aankhen`` ?eval $?PUGS_VERSION
evalbot6 \'Perl6 User\'s Golfing System, version 6.2.8, July 13, 2005 (r5904)'
Aankhen`` I can't confirm, unfortunately; going through `make test`. 11:20
gaal then you already have a pugs :)
Aankhen`` Oh yeah.
Silly me.
iblech ?eval say {a=>1,b=>2}.perl
evalbot6 a 1b 2\'perl6' bool::true
iblech ?eval say ({a=>1,b=>2}.perl)
evalbot6 {('a' => 1), ('b' => 2)} bool::true
gaal hey, say's back. cool..
Aankhen`` Same behaviour here, yeh. 11:21
Hiya iblech. :-)
gaal where'd that "perl6" come in from?
Aankhen`` o_O
gaal i have "\undef", which is wrong but less surprising.
iblech It seems it's parsed as (say {a => 1, b => 2}).perl -- the stringification of the hashref is "a\t1\nb\t2", and the result of the say is undef... 11:22
Hi :)
That's probably from evalhelper.p5's implementation of &say
gaal hmm. what's say's return value specced to be? 11:23
Aankhen`` wolverian: Ping.
gaal i'd assume p5 semantics?
iblech Yep, 1 if the print worked, 0 (or false) otherwise
putter: pong
gaal fail would be nice...
how useful is a 1 anyway? 11:24
the string printed might be nice. 11:25
s/nice/more useful/
iblech Hm, I like fail... Probably it is fail, as I remember reading that &open fail()s too 11:26
Aankhen`` Why does `make install` trigger re-compilation of Pugs.Run and the pre-compiled modules?
gaal weird, why did that say originally return undef? i'm looking at the code and it looks correct 11:27
return $ VBool True
Aankhen``: make smoke does, too :( i don't know how to fix this dependency business
Aankhen`` :-\ 11:28
gaal, got a minute? 11:38
gaal sure
Aankhen`` Could you see whether test 55 of t/builtins/io/io.t fails for you? 11:39
gaal perlcabal.org/gaal/smoke_xp.html will be up in a sec :-) 11:40
Aankhen`` Ah, great!
I just want to make sure I didn't break the Prelude. :-) 11:41
gaal up.
Aankhen`` Yay.
gaal++
Not Found
The requested URL /gaal/smoke_xp.html was not found on this server.
gaal missed a ~ there. 11:42
perlcabal.org/~gaal/smoke_xp.html
Aankhen`` Ah.
gaal oops :)
Aankhen`` Hehe.
gaal so, yeah, it fails.
Aankhen`` nods. 11:43
gaal has a temperature and a sore throat :(
Aankhen`` Awww. :-(
gaal does any string evaluate as false in perl 6? 11:48
?eval ?""
evalbot6 bool::false
gaal ?eval ?"0"
evalbot6 bool::false
gaal ?eval ?"1"
evalbot6 bool::true
gaal ah. :)
?eval ?"0 but weird" 11:49
evalbot6 bool::true
gaal ?eval ?"00"
evalbot6 bool::true
Aankhen`` I've created a `role Iter` in the Prelude, and modified `IO` and `Str` to `does Iter`.
`Iter` defines &prefix:<=> like so: multi sub prefix:<=> (Iterator $self: ) { $self.shift() }
Um, that's Iter $self:.
gaal cool 11:50
Aankhen`` And method shift () { ... }
gaal ?eval ?"0\n" 11:52
evalbot6 bool::true
gaal ?eval "" ~~ bool::false 11:53
evalbot6 bool::true
Aankhen`` Hmmmmmmm. 11:58
Any objections to my implementing a mutating `map` and `grep` in Prelude.pm?
gaal i know i want them... but this looks like a Q for the language list 12:00
Aankhen`` Well, it's there in S29.
gaal looks
Aankhen`` Rather, they are already specified in S29.
?eval sub foo () { }; sub bar ($foo) { }; sub baz ($foo, *@bar) { }; (&foo.arity, &bar.arity, &baz.arity) 12:01
evalbot6 (0, 1, 2)
Aankhen`` ?eval sub foo () { }; sub bar ($foo) { }; sub baz ($foo, *@bar, +$baz) { }; (&foo.arity, &bar.arity, &baz.arity)
evalbot6 (0, 1, 3)
nothingmuch is destructive map MMD?
Aankhen`` nothingmuch: Wha?
nothingmuch e.g. class array { method map (@input is rw) returns Void { ... } } 12:02
that is - it cannot be used in non void context
and it must be matched against mutable types
Aankhen`` I don't understand.
gaal A: what makes the s29 map mutating? 12:04
Aankhen`` gaal: S29 map isn't mutating.
gaal ah
Aankhen`` But there's a t/builtins/lists/mutating_listops.t which appears to require mutating `map`.
And `grep`. 12:05
gaal for a while didn't is rw work in map? i seem to remember
anyway
i need to lie down a bit
later :)
Aankhen`` See ya.
iblech re. m19s28.vlinux.de/iblech/stuff/not_p...moke.html, 43.75% ok :) 12:17
svnbot6 r5905 | Aankhen++ | * Prelude.pm: made formatting more consistent; fixed typo.
r5906 | Aankhen++ | * Prelude.pm: added Iter role, and modified IO and Str accordingly.
iblech (Spidermonkey's /bin/js)++ # Approximately as fast as normal Pugs
Firefox-- # sloow
Most tests fail because a. feature not yet implemented, b. they use eval, or c. a scalar container ($foo) holds an aggregate, not a scalar (fixing this currently) 12:19
Aankhen`` ?eval $?PUGS_VERSION 12:21
evalbot6 \'Perl6 User\'s Golfing System, version 6.2.8, July 13, 2005 (r5906)'
Aankhen`` How come my Pugs doesn't show the revision?
integral Aankhen``: checked out by svn, svk or other? 12:22
Aankhen`` SVK. 12:23
integral and you're using a local branch?
Aankhen`` Yes.
integral the script for getting the revision number (util/version_h.pl) when working with svk tries to look for the revision number of the upstream repository, not the local revision number. I think it's so there's no confusion
Aankhen`` Ah. 12:24
integral hmm, I think I wrote this, so it is to avoid confusion :) 12:25
nothingmuch seen ingy
jabbot nothingmuch: ingy was seen 1 days 15 hours 32 minutes 55 seconds ago
nothingmuch rats
QtPlatypus Is anyone else having trouble connecting to news.perl.org ? 12:26
I guess everyone isn't having any problem with it, all there all idle. 12:34
autrijus PIL2JS 43.75% OK? holy lambda. 12:57
clkao crazy
Aankhen`` QtPlatypus: news.perl.org redirects use.perl.org, which is working...
Redirects to, even. 13:01
Aankhen`` steps away from the computer.
nothingmuch makes large cat sounds 13:14
has anyone here got any Test::Base fu? 13:15
autrijus me. 13:20
not terribly fluent, but passable.
nothingmuch autrijus: i am trying to fudge some stuff 13:27
with skip
you know my p5orth test suite?
i just did mutual recursion this morning (3 minutes - a small change in the forth prelude... isn't that amazing?)
and now i defined a convenience word: ":deferred"
so you can say:
defer even 13:28
: odd ?dup if 1 - even else false then ;
:deferred even ?dup if 1 - odd else false then ;
but gforth doesn't have :deferredc
so in my test spec i want to say:
=== :deferred
--- skip: gforth
--- forth 13:29
...
and I just did that, it's not that bad a hack either
and I would like ingy to apply it, or judge it
and since ingy isn't here i'd like a Test::Base moose to judge it
autrijus rehi 13:32
(was rebooting)
ok. let me look at the patch? 13:33
nothingmuch sure 13:37
pasteling "nothingmuch" at 212.143.91.217 pasted "Test::Base patch" (188 lines, 5.1K) at sial.org/pbot/12084 13:39
nothingmuch oops
that's a dumb patch
pasteling "nothingmuch" at 212.143.91.217 pasted "Test::Base patch" (144 lines, 3.9K) at sial.org/pbot/12085 13:40
nothingmuch okay, that makes more sense
it does two things:
makes --- SKIP in to a test builder skip
autrijus looks pretty good 13:41
also you force a plan number
nothingmuch and --- skip into the same thing, only after eval
and yes, it forces a plan number
autrijus so three things in all
nothingmuch true
autrijus it looks good to me. send to ingy?
nothingmuch will do 13:42
i think this is a nice example of how functional programming can be more flexible than OOp
autrijus well nowadays most OOP languages admit first class functions anyway. 13:43
other than cobol, I mean java
nothingmuch true
but most people I know don't use them
except for things like callbacks 13:44
autrijus the main tension is pro-data or pro-operation
termed the "expression problem"
nothingmuch i think they are not really in a tension
autrijus I highly recommend you look into the original wadler link and Scala's solution
(both linked from my journal of yesterday)
nothingmuch you can usually slice each sub apart
line by line
and determine which line is either this or that
autrijus nothingmuch: the main problem is that it takes a lot of effort in Haskell if you add new construcor variants to a "data" declaration 13:45
while it takes a lot of effort in (say) Java to add a new method to a bunch of classes.
nothingmuch goes to read
how would you say perl 6 is in this respect?
autrijus generics + MMD solves this nicely 13:46
just as the Scala paper had it
nothingmuch hmm
what is the date on your side of the world?
autrijus (MMD let you define methods outside class; generics lets you pass type parameters around)
2005年 7月30日 周六 21時46分49秒 CST
jabbot autrijus: 2005年 7月30日是星期六
autrijus er
sorry, 2005-07-30 13:47
nothingmuch and did you wake up today, or yesterday?
autrijus I woke up 6 hours ago.
nothingmuch is trying to determine if "yesterday" really means yesterday, and then sees that there is only one journal entry that can be either
autrijus right.
nothingmuch anyway, should i just follow the Scalar link? 13:48
and will you please join the cabal temporarily so that we can have proper type support?
autrijus the "solution" link
tho the "expression problem" is fun too
eh, what does that has to do with this?
nothingmuch i'm just not getting the impression that anybody that in @Larry but luke really cares what will happen to the type system 13:49
autrijus oh Damian cares too
he's the one that brought intersection types and generics up originally
nothingmuch ah
but we still don't have a definition of parametrized and relational types, right? 13:50
autrijus actually we do, S06 has them, it's just the semantics are very unclear
I already identified the problem spots as I can see them and posted to p6l 13:51
nothingmuch remembered it as a very thin example
so can we say "this method takes an object that does Moose and returns a subroutine that returns (an object that contains an object of type Moose)"?
sub moose_container (Moose $x) { [ $x ] }; 13:52
sub moose_container (Moose $x) { { the_key => $x } };
s/sub/multi/;
autrijus multi foo (Moose $x) returns Any of Moose { ... } 13:53
multi foo (Moose $x) returns Source of Moose { ... }
(hypothetically adopting the Source/Sink behaviour)
but really, you want
nothingmuch err, i meant
autrijus foo (Moose $x) returns (Array|Hash) of Moose { ... } 13:54
nothingmuch sub moose_container (Moose $x) { -> { [ $x ] } }
autrijus oh
nothingmuch i think there is too much orthogonality in the perl 6 type system to allow 'Moose of Elk' to capture it well
autrijus returns (Code returns Moose)
nothingmuch right
autrijus returns (Code of () returns Moose)
nothingmuch can we manipulate types as first class objects? 13:55
autrijus sure 13:56
at runtime
nothingmuch what is the type of a type?
autrijus Type, supposedly
nothingmuch an object of type Type?
that is both liberating and limiting
because i would like a Type to be parametrized on it's value
autrijus that's called dependent types. 13:57
nothingmuch well, yes, and no
autrijus speak in code please :)
nothingmuch yes, i'm trying
my Moose $x;
$x.container.meta.type is Type of Moose;
it's also just a Type 13:58
damnit! someone resubscribed me to the most bullshitting list in the world
what is the type signature of 'ref'? 13:59
autrijus &ref?
nothingmuch i would say 'sub ref ((:a) $x) { Type of :a };
yes
autrijus sub ref (Any $x) returns Type; 14:00
nothingmuch see, that sort of sucks
i really meant:
autrijus my p6l proposal has this:
nothingmuch sub ref ((:a) $x) returns Type of :a { Type of :a };
autrijus sub ref (Any $x) returns Type of ref($x);
essentially having ref() as a type form 14:01
nothingmuch hmm
fair enough
autrijus sub one_of ($x, $y of ref($x)) returns ref($x) { rand(2) ?? $x :: $y }
this has the good property in that we avoid type variables.
which never felt perlish to me anyway
nothingmuch i think i want both this and type variables 14:02
side tracking:
autrijus but the only place the spec has tyvar is on generic roles
nothingmuch can you define a variable in the parameter list that isn't a parameter?
autrijus and I'm not even sure we what it there.
nothingmuch: huh?
nothingmuch sub moose ($x, ... # somehow define $y here #) { $y available } 14:03
$y is based on $x
but is not an optional parameter with a default expression that is based on $x
autrijus sub moose ($x) { my $y := $x; }
what's wrong with this?
nothingmuch because i'd like to reuse $y in the prototype 14:04
e.g. $y = ref($x); manipilate $y a bit; ... returns $y;
autrijus sub moose ($x) returns ref($x).of.blah.bleh {...}
if you want to much with types, use type operators 14:05
nothingmuch type operators?
autrijus for example, | & ^ of
selectors even
nothingmuch that's not turing complete
autrijus ah, you want the compiler to go into infinite loops
how very unkind of you.
nothingmuch =)/. 14:06
seriously though 14:07
in a language that is not completely compiled in one go
but is optionally compilable at runtime
we need two things:
to be able to give the compile time enough info to be fast
and to give the runtime enough access so it loses no features
and i think there's a bunch of deltas in the type system 14:08
autrijus and I completely agree ;)
nothingmuch that is - they cannot express everything the runtime can make up
or using them is a compile time only feature
autrijus the important thing to bear in mind is that perl6 observes the separate compilation doctrine 14:09
a module provides, to the external world, "link sets"
nothingmuch right
autrijus and as such _cannot_ trigger recompilation every time a consumer uses it
that is in direct contrast to perl5
nothingmuch but on the other hand if the module defines an import sub 14:10
which eval's some code
you get something like perl 5
autrijus nope.
nothingmuch nope?
autrijus let me try to explain
nothingmuch nono! i will never let you!
autrijus alright.
I won't explain then
nothingmuch you have to pay a fee
nothingmuch charges autrijus for one cookie 14:11
will you accept the charges?
autrijus ;)
cookie 14:12
nothingmuch i'll take that as a yes
you may continue
no, i mean an IRL one, the next time we meet
strings are just a representation
autrijus whatever.
the important thing is
nothingmuch jumps up triumphantly: I GOT A FREE COOKIE! 14:13
(rocky soundtrack in the background)
autrijus in p5, the consumer is allowed to set compilation hints that affects the compilation of the use'd module.
nothingmuch like?
autrijus use strict::with::pride;
(bad example)
a better one is 14:14
foo.pm: 14:15
say 123;
main.pl:
sub foo::say { die }; use foo;
or, more explicitly 14:16
foo.pm:
our &boo ::= \&say; 14:17
and main.pl being the same.
in p5, &boo always binds to "die"
in p6, it cannot be so.
(actually, ::= &say), but you got the idea.
nothingmuch yes 14:19
so how do you change the module's say in perl 6? 14:20
no way what so ever?
autrijus oh, easy
you load its textual content and eval it.
;)
nothingmuch okay: a distinction is in order 14:21
in perl 6 compilation is a caching strategy
err 5, not 6
that is - it's only purpose is to gain speed
it doesn't have any functional effect whatsoever
if it does, that's a bug
in perl 6, on the other hand, a compiled module and the source to compile it are not necessarily parts of the same equasion 14:22
autrijus er, no
p5 has two phases
two runtimes
or rather three
BEGIN, main, eval
eval may contain its own BEGIN main and eval
and so on
to use a module is to eval its textual content in BEGIN time.
nothingmuch yes
autrijus in p6, each module is separately compiled from scratch 14:23
nothingmuch and eventually you wind up with a tree of compiled stuff in memory
autrijus as a fresh environment
nothingmuch okay, rephrasing question:
is the object CompiledModule
a complete blackbox?
does it only have the "exported_symbols" method?
or is it a nested structure 14:24
so that $some_module->linked_modules contains Prelude and Guts
and their instances are provided by the runtime
and are the same as the consuming modules 14:25
and then the symbols can also be overridden:
Prelude can have it's 'say' replaced
or rather, say, that comes from prelude can be replaced
thus not affecting externally compiled code, just what externally compiled code links to 14:26
autrijus if Prelude has its "say" replaced
it needs to be swapped in with something of equivalent type
otherwise all typecheck was in vain.
nothingmuch right 14:27
when rebinding say if you replace it with something typed otherwise it's an error
if you use ::= it's a compile time error
if the types are coerciable 14:28
for example sub { die }
the return type of 'die' can be coerced into a string but error or string
and the parameters can be coerced into a the same thing say uses
except the parameters are thrown away 14:29
(if there are things like 'where' or defaults which are more than just bindings then this cannot be done automatically)
in a sense, this can be type sound
unless it isn't
in the many cases it is, there shouldn't be a reason for it not to
autrijus yeah. I'm fine with you hacking in the environment 14:30
as long as the linkset interface is immutable
sacred, holy, etc
nothingmuch right
and the only situation that really happens is if everything is unboxed
because you can just pass mock objects around instead
uh, s/un// 14:31
nothingmuch ponders the 'faux' operatore
$thing faux Type of ....;
let's say something can't numify 14:32
autrijus Faux pax Americana;
nothingmuch "Moose" faux Num;
+$faux; # 0
or can't stringify:
~(Something.new faux String); # "DUMMY" 14:33
nothingmuch sidetracks again: 14:36
can unboxing be generalized?
autrijus I'm of the opinion that value types is a myth, so don't ask me 14:37
nothingmuch especially WRT accessors for objects of closed classes
autrijus I think value types is valid 14:38
and value types should handle (un)boxing automagically
nothingmuch hmm
autrijus and that's all.
nothingmuch please accurately define value types
Int et al?
autrijus right, base types that has no notion of identity 14:39
nothingmuch why?
autrijus 1 =:= 1 14:40
1 =:= 2
also, it makes little sense to derive from value types.
nothingmuch right
but why stop at that?
class Foo { has $moose };
if Foo is closed and not basal, 14:41
is there any reason to make 'my Foo $x; $x.moose' a method call?
it should be an O(1) lookup
since we have all the required knowlege at compile time:
what is the definition of Foo
what is in a moose
what is the type of the method moose 14:42
whether or not Foo can change
autrijus really. what is the type?
nothingmuch and what concrete types can be a Foo
in this example Any or whatever the default type of is
and scalar
autrijus &moose:(Foo) returns Any?
nothingmuch yup 14:43
autrijus thing is Foo can't be guaranteed as closed :-/
nothingmuch $moose can contain anything
no it isn't
autrijus which calls for two compilations, but we covered that in hackathon
nothingmuch i'm talking just about the accessor
you don't need to do something along the lines of $x.meta.class.find_method("moose").
autrijus and I agree :)
nothingmuch since you can assume that lookup will be constant, given your knowlege now that you are compiling it 14:44
autrijus now, consider this
nothingmuch this is just static dispatch
no for the unboxing: 14:45
autrijus class Foo { has Foo $moose }; my Foo $x; $x.moose.moose;
can the second .moose call be done in O(1) as well?
nothingmuch given $x.meta.class.find_method("moose") { when Accessor { # compile unboxed } default { # compile a call } }
no
autrijus why?
nothingmuch wait
does method chaining in perl 6 behave differently?
autrijus it's same as p5. 14:46
nothingmuch ($x.moose) can be compiled to an access directly into $x's structure
autrijus $x->moose->moose;
nothingmuch that can be of type any
okay
autrijus but I declared
has Foo $.moose;
nothingmuch oh, i didn't notice
yes it can
autrijus but why?
because we have a type inferencer? ;)
nothingmuch yes =)
autrijus a local type inferencing propagation engine?
that's what I'm arguing for :) 14:47
nothingmuch but at this level it's just an optimization
which we already have
autrijus no, it can also reject programs.
nothingmuch oh, right
autrijus $x.moose.goose; # instant death under closed/final
nothingmuch okay, in that case it's what i've been arguing for all along =D
huraah!
nothingmuch does the Sound Type dance 14:48
autrijus and severe warning -- i.e. has to be turned off specifically -- by default.
nothingmuch i don't see why
autrijus why warning?
nothingmuch there is absolutely no scenario where this will not croak at runtime if Foo was closed
autrijus nope.
nothingmuch or am I missing something?
autrijus multi goose (Foo $x) {1}
nothingmuch foo is closed 14:49
it can't be extended
autrijus class Bar { method goose }; Foo is reopened does Bar;
nothingmuch reopened?!
autrijus (also MMD doesn't mean extension usually.)
nothingmuch you don't gain anything from closed if you have reopen
well, it might not "mean" extension in the logical sense 14:50
but in effect by declaring a multimethod you're modifying the types of all the involved parameters
autrijus ...which is why they are much better than visitor patterns...
nothingmuch since they have added functionality when they are combined
autrijus because they don't require anything on the behalf of original author of Foo 14:51
nothingmuch right
autrijus even inder closed + final
nothingmuch except that you have an unclosed version of the module
eep
multi sub &infix:<+> (Int, MyType);
autrijus although I'm also arguing that if you "close" current scope 14:52
nothingmuch that breaks all unboxing whatsoever
autrijus i.e. not only :classes<close> but also :packages
then the only way to introduce multi goose is by "my sub" or "my multi"
which is fine, because that'd be lexical.
nothingmuch hmm
autrijus and an "eval" won't be able to propagate out.
nothingmuch i think the following closing semantics are sane:
closedness of a class is lexically scoped to it's consumers 14:53
that is Foo is closed to our $x.moose.moose call
but unclosed bytecode is retained
autrijus (I agree totally :))
nothingmuch and a block that declares 'my goose (Foo $x) { 1 }' will have to use the open one
autrijus (yes. we've discussed this in hackathon.) :)
nothingmuch ah =)
really? i must have zoned out
autrijus you were drunk.
nothingmuch i was not drunk 14:54
autrijus okay, then something worse
nothingmuch i was sleep deprived
autrijus ah.
nothingmuch during the hackathon i slept about 4-5 hours a night
went to bed late for Canada
autrijus well the behaviour interface is isomorphic
nothingmuch woke up late for Israel
(which is 5-6 am canada)
except on the last days
stevan remembers nothingmuch being drunk quite often
autrijus nodnod
nothingmuch stevan: i drank a lot of beer 14:55
and i'm clumsy when sober
stevan burning pizzas and such
nothingmuch so i think you got the wrong impression
that was a type coercion problem
i explicitly said "Baking paper"
and then john said "you mean wax paper?"
and I made a bad optimization assuming "wax" means "baking" in canadian
autrijus: do @Larry agree with us? 14:57
because I was really traumatized when on the last or second day of YAPC when we were up in your room, and you mentioned that our type inferrencer is just optimizing, not type checking
autrijus nothingmuch: I have no idea. we'll see how @Larry replies to my mails in the previous two days. 14:58
nothingmuch okay 14:59
autrijus TSa is very supportive, but he argues from a far more static position, so I'm not surprised.
nothingmuch i have to reread them
admittedly i only skimmed
anyway, i feel like forthing a bit more 15:00
autrijus that's fine, the main proposal is www.nntp.perl.org/group/perl.perl6.compiler/1000
sure, go ahead. have fun :)
nothingmuch i want to port most of p5orth to forth
so that I don't have to code that much haskell ;-)
autrijus wonders if nothingmuch will enjoy Joy
nothingmuch that's not very googleable 15:01
autrijus nothingmuch: well if you like the lack of parens as much as schemers like parens, then forth is very deep and fun indeed :)
www.latrobe.edu.au/philosophy/phimvt/joy.html
nothingmuch parens ini lisp are just syntax
autrijus lisp interpreter in joy: www.latrobe.edu.au/philosophy/phimv...lsplib.joy 15:02
nothingmuch scheme is just a tree
forth is just an array
autrijus yes.
nothingmuch but when you create the array you get to pretend it's a tree
so hence forth has higher order macros, not functions =)
autrijus and Joy to Forth is a bit like Perl5 to Perl4, if I may :) 15:03
nothingmuch wow, that's sounds pretty 15:04
wowowowww, that's so cool
en.wikipedia.org/wiki/Joy_programming_language
the SQUARE
if i got it right 15:05
wait, is it pushing '*' on the stack or not?
autrijus yes. 15:06
nothingmuch *phew*
that explanation is not very good
autrijus I prefer www.latrobe.edu.au/philosophy/phimv...y/faq.html
nothingmuch okay, where is the distinction?
why isn't '.' applied to 'dup'? 15:07
how do i get a higher order dup? just quasi quote it?
autrijus the . is not there
the . is ending DEFINE 15:08
nothingmuch oh
autrijus yes, quasiquoting is happy
[1 2 3 4] [dup *] map
===> evaluates to [1 4 9 16]
fun eh.
nothingmuch so basically the only place where you can evaluate is top level?
autrijus right. that's what a pure functional language means. 15:09
see haskell :)
nothingmuch because if I said DEFINE moose = [1 2 3 4] [dup *] map .
and then I said moose
it's like i said [[1 2 3 ...] map ], right?
didn't say pure on the top of that page
wikipedia, that is
autrijus I think so, yes, it's referentially transparent
oh. www.latrobe.edu.au/philosophy/phimv...h-joy.html
it's the first sentence.
"The language Joy is a purely functional programming language." 15:10
nothingmuch "The Joy programming language is a simple functional programming language"
damn wikipedia
since when are "purely" and "simple" intgerchangiable?!
autrijus changed wikipedia.
it now says "purely".
nothingmuch =) 15:11
autrijus praises the writable web
brentdax My little project is nearly at the same level of external sophistication as a basic wiki.
nothingmuch brentdax: @{ } 15:12
autrijus brentdax: ooh
nothingmuch: so, Joy's style is utterly pointless
I mean point-free
nothingmuch =)
autrijus and because of this, it agrees with the Forth structure perfectly 15:13
I think it's a wonderful language to play with :)
nothingmuch wonders what you can do with it
brentdax You can edit now with a preview mode. I need to refactor to allow page creation, though. After that, all it needs is a markup language for formatting and a little tweaking to the link-resolving code.
autrijus larry is still thinking about my posts, I hope... he just fielded the simple ones
brentdax: demo demo 15:14
nothingmuch brentdax: by @{ } i meant: please dereference the reference to your little project, i don
't know what it is =)
brentdax Ah--WWW::Kontent, my Summer of Code project.
autrijus oh, it's Kombat, his Winter of Microsoft project
brentdax autrijus: takes about ten seconds to run a request on my laptop--I'd be insane to put it on a web server. 15:15
nothingmuch ah
autrijus brentdax: url to code?
nothingmuch brentdax: perl6.nl
feather.perl6.nl
brentdax svn.brentdax.com/kontent/trunk 15:17
nothingmuch wonders how implementing Joy in Forth might be 15:18
oh my, refactoring is indeed in order 15:22
i reimplemented ' as a word
when I had it implemented as a primitive ;-) 15:23
nothingmuch is very very evil 15:38
svnbot6 r5907 | Stevan++ | Perl6::MetaModel - (p5)
r5907 | Stevan++ | * moved all the magic in AUTOLOAD into ::dispatch
r5907 | Stevan++ | - first step in refactoring out Perl 5 sugar
autrijus oooh.
stevan++
nothingmuch thinks of adding forth code which will take listed data stack ops
copy them
relocate
and then replace all the stack primitives in use by traversing the tree
to generate RDUP from DUP, etc
and to do this in the forth prelude 15:39
stevan++
will i be bitchslapped for this?
autrijus iblech: you there? 15:42
iblech autrijus: Yep 15:43
autrijus care to lay out your preferred container semantics? 15:44
I'm working on formalism now
so I think I should start with the PIL2JS treatments 15:45
iblech Sure :)
autrijus so the bulk of code is in lib6/Prelude/JS/ContainerName.pm?
iblech No, that file doesn't exist :) Most of it is in libjs/PIL2JS.js, the .GET, .STORE, and .BINDTO methods of PIL2JS.Box 15:46
autrijus ok. 15:47
iblech The general idea is that a variable name points to a container, and containers have the .GET method (returns the unboxed value), .STORE, and .BINDTO (basically this.GET = other.GET, this.STORE = other.STORE, etc.)
Every container has a .uid attribute which is used for =:=
autrijus the STORE case is just for lvalue positions? 15:48
iblech Right. 15:49
autrijus and if BINDTO preserves uid, don't we need a clone?
i.e. just like BINDTO but with different uid
iblech PIL2JS.Box is the only real container class, PIL2JS.Box.Constant's .STORE only die()s.
autrijus or, alternatively, a "copy" primitve operation on containers
oh nvm, you already have it. 15:50
just not spelled as CLONE
iblech Yep, .STORE is used for that: my $a = $b is container_of_a.STORE(b), and .STORE sets this.value = b.GET()
autrijus // Needed for "is copy".
clone: function () {
return new PIL2JS.Box(this.GET());
},
I'm referring to this 15:51
that implementation looks suspicious 15:52
iblech .clone is only a convenience function. For is copy support, I could emit $a = $a (container_of_a.STORE(container_of_a)), too 15:53
How so? It returns a new variable container, initialized with the current contents
autrijus nvm, I was confused by the relation to perl6's .clone. 15:54
(which is deep)
never mind me :)
iblech Right. This is why I've renamed it to .copy here :) 15:55
autrijus good.
so, why is BINDTO a prim? 15:56
why is it not modifying the symbol table?
i.e. why should a box be able to bind to another box of different shape? 15:57
I can see NEW, FETCH, STORE, but I still don't quite grok BINDTO.
iblech "We can't just use C<dest = source>, because C<dest> might be the result of some
function, e.g. C<@array[$idx] := $foo>."
(from README)
autrijus I see.
so this calls to yet another underlying level in the host environment 15:59
that can maintain reallocation
in Hs we used writeTVar for BINDTO, essentially, without naming it
so, is this the shape for all containers, or just scalar containers?
iblech For all. 16:00
autrijus so if I implement "tiescalar", I'm implementing something on a higher level.
or "tiehash" etc.
iblech "Why isn't there a array_fetchElem?" -- Simply because I didn't need it -- i.e. some_array_container.GET().push(...) works fine
(or some_array_container.GET()[...], for that matter) 16:01
autrijus still thinking 16:02
iblech I think so. There's PIL2JS.Box.Proxy, too, but that's one the same level as PIL2JS.Box. It's used in &postcircumfix:<[ ]> and &postcircumfix:<{ }>
autrijus ok. 16:03
so when I run BINDTO, I basically overwrite both my uid and my content 16:04
iblech Yep.
Which is effectively the same as container_a = container_b, but works with function return values, too
autrijus in haskell: 16:05
data Container = TVar (Box a)
data Box a = MkBox { uid :: Unique , content :: a }
does that look correct?
"a container is a mutable reference to a Box that carries its type"
"a Box is a uid, with a content of some type" 16:06
iblech Yep.
autrijus now the hard question!
in general you can't replace one content with another unless they are of the same type "a".
is the type "a" essentially "Dynamic"?
i.e. you can bind anything to anything.
(I've read your posts, so I'm intentionally applying the socrates method) :) 16:07
my Int $x; my Num $y; $x := $y; 16:08
is essentially what I'm asking.
assuming that the "Int" is 'a' here, i.e. we actually allocate different storage for different types.
iblech Hm... I'd say that works, as the type is associated with the .value. But you can't %a := @b 16:09
Err
It should not typecheck, of course. 16:10
But my $x = 3; my $y = [...]; $x := $y should work
autrijus how about
my Int $x = 3; my $y = [...]; $x := $y;
? 16:11
this is really a hard question, because as you observed
%a := @b; # should be runtime error even passed typechecker
but currently nothing in p6 explicitly forbids this. 16:12
iblech Should not work, as [...] is not an Int. The type declarations are attached to the varname, I think -- if you see a "my Int $x" you should be able to be sure that $x is always a Int, even if some code rebinds $a somewhere
autrijus ?eval my %a := [1..10]; %a[0]
evalbot6 [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
autrijus ?eval my %a := (1..10); %a[0]
evalbot6 \1
autrijus okay. so 16:13
my %a
means
my Hash %a;
essentially.
(disregarding the fact that it means 'hash of hash' in external language)
(I mean in internal language)
iblech: so. try this
my $x; $x := 3; $x := "Hello"; 16:14
iblech I think the reason %h := @a should not work is not of the types of the values in the containers, but because of incompatible container types (HASH vs. ARRAY)
autrijus valid?
I expect you to say valid. :)
iblech Yep, as $x is of type Any, and both 3 and "Hello" are Anys
:)
autrijus ok. so our definition is wrong
data Container = TVar (Box a)
data Box a = MkBox
{ uid :: Unique
, value :: forall b. b < a 16:15
}
is the desired semantics
now, who decides the '<' relationship.
if I had said 16:16
iblech Probably, but as I don't know what forall b. b < a means, I can't say for sure :)
autrijus oh, it means "all values that instantiates this type needs to be a subtype of a"
< is the subtype notation
iblech Ok
autrijus so Int < Any
but not the other way around
so, now 16:17
my Int $x; $x := &say;
is obviously bogus
iblech Right.
autrijus but can I say
role Code is reopened does Int {...}
?
and expect that to work 16:18
iblech I think so.
autrijus good. we are in agreement then :)
iblech :)
autrijus ok, now let's look at the scalar type. 16:19
integral iblech++ autrijus++
autrijus formalization++ :)
what does scalar provide beyond a simple container?
iblech Nothing, I think. 16:20
autrijus I can think of one thing 16:21
namely the TIE and UNTIE primitives.
my $x; tie($x => OtherClass); untie($x);
for the duration of the program, OtherClass is supposed to provide STORE and FETCH 16:22
until untie()
iblech Oh, right.
integral does tie like that exist? (or need to?) 16:23
autrijus integral: yes and yes
iblech But OtherClass doesn't have, of course, any influence on BINDTO.
autrijus yes.
autrijus wonders when will we start to disagree
autrijus is thinking 16:24
so, a scalar augments a container's interface 16:25
making it a subtype of a container
still thinking
we can adopt an open-world view or a closed-world view. 16:26
a closed-world view says there are limited forms of container types
and &tie can deduce them base on simple casing
wolverian Aankhen``, pong 16:27
Aankhen`` wolverian: Hola.
autrijus under that view, the language provides several mutually incompatible tie interfaces, and that's all
wolverian hello. I've been away a bit. :)
Aankhen`` I was wondering: are you planning to port URI?
s/port/refactor/ as appropriate.
wolverian Aankhen``, yes, but I've been really tired now; my medication hasn't kicked in yet. 16:28
autrijus see the definition of IVar for an example of closed world
do you think it makes sense?
Aankhen`` wolverian: What happened?
wolverian I sleep 16 hours a day and I'm tired the rest.
Aankhen``, hypothyroidism.
autrijus (this is about tieability, not about container allocation per se) 16:29
Aankhen`` Ah.
iblech autrijus: I think that'll work. 16:30
autrijus an open-world view will mean that you can selectively tie anything.
i.e. only tie FETCH but not STORE
wolverian Aankhen``, I guess I'm out of it for a month still, or so. :\
iblech But one can emulate not tieing STORE by simple relaying STORE requests to the original container 16:31
Aankhen`` wolverian: S'okay. Take your time.
autrijus or tie both PRINTLINE _and_ KEYS.
wolverian thanks. I will. :)
Aankhen`` wolverian: Is it curable?
autrijus although with the ability of MMD'ing &{} %{} @{}, there's little gain in doing that. 16:32
wolverian Aankhen``, yes, simply by administering thyroxine, the hormone the thyroid gland is supposed to produce.
Aankhen`` wolverian: I see.
wolverian Aankhen``, it however takes a long time for the hormone levels to rise back to normal. it is also tricky to find the right amount to take daily. 16:33
autrijus okay, closed world then. next question. who inhabits this closed world?
Scalar, Handle, Array, Hash, obviously.
Aankhen`` wolverian: OK. Anyway, don't mind me, it was nothing important. Go rest.
iblech Right. I think many of Perl 5's SPECIALTIESUBS aren't needed in P6, as they're simple method/mmd calls
autrijus yup.
but still tie acts on container and mmd on value 16:34
so mmd can be open but it makes sense for tie to be closed.
wolverian Aankhen``, nah, I have a few hours now of being semiawake. :) or an hour. that goes pretty much to checking the mailing lists and planets. and you guys here. :)
Aankhen`` wolverian: Ahh.
iblech Hm, do we need Handle? As $fh is a simple scalar (which happens to contain an IO object)
Right.
Aankhen`` AFK for a bit. & 16:35
autrijus iblech: you mean
$*ERR := MyInterceptedThing.new
?
iblech Right.
autrijus oh, btw. 16:36
my $x = 123; my Int $y; $y := $x;
valid?
iblech Hm, unsure. What if somebody does $x = "hi" later... 16:37
autrijus (it's a very tricky question.)
my $x = 123; sub f (Int $y is rw) { $x = "hello!" }; f($x); 16:38
what really happens? :)
iblech Type error. 16:39
autrijus at which stage?
function application or assignment?
iblech $y = "hello!"
autrijus I didn't say $y.
$x = "Hello"
iblech Oh, I assumed you typoed $x 16:40
autrijus I didn't. :)
iblech Ok, then no error at all, as $x isa Int
autrijus not after the assignment tho. 16:41
my $x = 123; sub f (Int $y is rw) returns Int { $x = "hello!"; $y }; f($x);
iblech Oh!
autrijus according to my treatment of Box above 16:42
the function application should fail.
but that's weird.
iblech Ok, that's a type error then, and the appl shouldn't typecheck.
But it's correct, I think...
autrijus nod. the problem does happen even if you had used normal $y 16:43
i.e. read only
only by "is copy" you avoid this.
this just in from larry!
$x = @array;
$x := @array;
Hey, who said anything about consistency? :-)
(he thinks those two forms should be identical) 16:44
iblech He's wrong I think: If you re-assign the first $x, @array is unchanged, but if you change the second $x, @array is changed, too 16:45
my $x = [1,2,3]; my $y = $x; $x = [4,5,6]; say $y # 1 2 3 16:46
my $x = [1,2,3]; my $y := $x; $x = [4,5,6]; say $y # 4 5 6
autrijus my @array = (1,2,3); $x := @array; $x = 10; 16:47
this is a STORE call to an array container.
should it work by pun? 16:48
or should it be rejected?
I don't like punning, so I'd say rejected.
iblech Hm, I'd say it should be rejected, too. 16:49
But $-variables may never point to an array container, I think.
autrijus why so?
ah. because you are also a close worldist and think $ stands for Container::Scalar. 16:50
iblech Yep, $ is always scalar, @ always array, % always hash.
autrijus so the := fails due to container type mismatch.
but larry says we vivify @array to a new anonymous scalar container that contains a ref to it. 16:51
iblech Or, alternatively, should that reall be $x := \@array, which would work?
autrijus and then BINDTO $x to that.
and discard the anon scalar container
iblech s/reall/really/
autrijus this does have the net effect of having := to be same with =
again, all this logic needs to be hard coded in the BINDTO. 16:52
so, amazingly, we three are in agreement this time :) 16:53
iblech Ah, right, I see it now :)
autrijus ok. so we agree that we ditch TIEHANDLE. 16:54
that plays well with the removal of the * sigil.
we still have two sigils left, though.
& and ::
let's tackle & first
tie(&say => SomeOtherClass) 16:55
iblech I'd say & is really $, but with its type fixed to Code, so one can't say &foo = 3
autrijus no.
unless you want &foo = &say;
iblech I want that :)
autrijus oh, you want that. :)
the idea of a separately tieable &foo is that we can intercept calls. 16:56
or lie about parameters
and .assuming.
but all these can be done with function wrapping
iblech Right. 16:57
autrijus under &say;
er
undefine &say;
what does that mean?
iblech FWIW, TSa would probably say &foo(...) is really &postcirumfix:<( )>(&foo, ...), but this has got the problem of calling &postcirumfix:<( )>... 16:58
autrijus TSa's view is essentially a tieable &foo with APPLY.
which has its merits
(APPLY, like FETCH, would be tieable)
and &foo won't have FETCH and STORE; whenever you want FETCH, it yields a ref to it. 16:59
iblech I think that's ok, but of course calling say won't work afterwards. Consider $x = &say; undefine $x; $x(...)
autrijus ok. 17:00
hm, we don't distinguish between an undefined array and a zero-length array. 17:01
nvm.
hm. 17:02
internalls perl5 always saw CV as something other than SV
svnbot6 r5908 | chromatic++ | Make Test::Builder::Tester test pass (ignore tested module's test header).
autrijus hm. 17:03
ok. if we say &foo is merely like Foo $foo 17:04
er
Code $foo
then it follows that tie(&foo => ...) would work
with only FETCH and STORE
since it's tiescalar as usual.
iblech Yep.
autrijus i.e. it's a constrained scalar container.
I'm actually fine with that. 17:05
iblech :)
autrijus okay. finally, ::
which is the package type
that implements exciting things such as nested namespaces, type inhabitation, dispatch, etc. 17:06
allow tying on :: is akin to tying the symbol table, plus type checking and dispatch.
perl5 never let you do that.
iblech But we certainly want that in P6. 17:07
autrijus ::CGI = ::Main;
what does _that_ mean anyway.
iblech our $foo = 42; ::CGI == ::Main; say $CGI::foo # 42
autrijus aha, but there's more! 17:08
iblech err, s/==/=/
autrijus CGI.new.does(Main); # true
iblech Right.
Everywhere you can say CGI you can say Main now. 17:09
autrijus BEGIN { tie(::CGI => FakeCGI) }; package CGI; sub foo {...}
gaal ?eval my IO $x; say ($x.say("should not print")).perl
evalbot6 should not print bool::true bool::true
autrijus FakeCGI gets to handle registration of new symbol! 17:10
and reopend!
and other things!
gaal ?eval undef.say("hmm")
evalbot6 hmm bool::true
autrijus do you enjoy this level of madness? :)
iblech Of course! :)
autrijus now, hard question time 17:11
::CGI = (::Main | ::DBI);
::PPI = :(Int, Int);
if your intuition is the first is okay and the second is not 17:12
then that means packages and types are really orthogonal.
gaal hey, a new planet was discovered. 17:14
iblech Hm, I'd say both work... ::Foo can either be the same as Foo (i.e. class object), or it's a "real" type like :(Int, Int)
brentdax & #Touring an unconsecrated Mormon temple 17:15
(of all the random things to be doing.)
iblech BBIAB, food 17:16
autrijus iblech: but, it makes sense to tie packages
but less sense to tie types
and if you are saying Type is the top level container and it just happens Package is Type
then the tie interface on non-package types would be bogus. 17:17
on the other hand, the tie interface fo scalar is also bogus for read-only things. 17:18
so that's not a big problem.
gaal autrijus: can you give me a hint on implementing 'use fatal'?
autrijus gaal: see Fatal.pm 17:19
gaal what, on p5?
autrijus yup
gaal there's already a fail_; should use fatal "just" set the caller's $?FATAL_SHOULD_DIE? 17:20
autrijus oh, you mean _that_ fatal
not Fatal
sorry.
you want it to be lexical, so you need to get lexical export to work 17:21
other than that it's fine
gaal %CALLER::<%MY::FATAL_SHOULD_DIE> modulo syntax etc? 17:22
autrijus that works too.
yes.
gaal we don't have 'no' yet, do we?
autrijus nope.
iblech: ok. so, provisional recap
gaal thanks.
autrijus in Perl 6, container is a box with a type, an ID and a content. 17:23
but users won't see containers directly.
there are scalar, hash, array and type containers.
those four can refer to a primitive container 17:24
or be overriden with tie.
if that sounds okay with you, I'll get a prototype going tomorrow :) 17:25
meanwhile I need to nap a bit... bbiab.
(or bbl, depends)
iblech autrijus: re. Yep, that's exactly what I had in mind, I think :) 17:32
nothingmuch shit! someone put my pants in the laundry before I took my wallet out 17:44
my thumb drive is now cleaner 17:45
but probably not functional
gaal so much money laundering takes place in this country
autrijus rehi. 18:16
nothingmuch hola 18:19
autrijus I'll start prototyping in PIL.hs
and see how far it goes.
nothingmuch the haskell PIL evaluator? 18:20
autrijus yes, starting with the container primitives
unlike certain other languages which only have "ref"
as the sole container
nothingmuch btw, do we have a perl 6 impl?
autrijus we have mutable "scalar", "array", "hash" and "type".
perl 6 impl of that?
nothingmuch yes
of a PIL evaluator 18:21
autrijus metacircular evaluator?
nothingmuch yes
autrijus what fun is that?
(sure, we need to have that, but I don't think it's neccessary or fruitful for this stage)
although you are certainly welcome to work on it :)
nothingmuch well, if it's too long then PIL<->Perl 6 is perhaps not a good mapping as we thought
svnbot6 r5909 | autrijus++ | * cleanup to remove obsolete IMC code.
autrijus er. 18:22
sub eval_pil (PIL $pil) { eval $pil.dump_as_perl6 }
done!
nothingmuch bah
i mean implementing the perl 6 types in perl 6, using perl 6 types
and really walking the structure, not compiling it 18:23
autrijus yes, I know.
and I'm saying sure, this needs to happen eventually, but it makes little sense until we can run that underlying code.
nothingmuch again, not as something of any practical use
autrijus no, eventually that will be of much use
see docs/notes/plan
nothingmuch i mean right now it doesn't have a practical use
except to meter how pewrl 6 and pil juxtapose 18:24
autrijus right.
I don't think that meters anything really
at this stage
nothingmuch fair enough 18:26
autrijus iblech: hm, I think the argument against TIEHANDLE applies to TIETYPE. 18:28
iblech: i.e. container types should concern with data storage; types are always scalars in term of data storage
hence any interception can be done by binding to a proxy 18:29
iblech Oh, yes, of course
autrijus I'm thinking about unifying Array and Hash into Keyed (or "Aggregate").
since Hash is list of pairs and lookup of strings
Array is list of values and lookup of ints 18:30
Aankhen`` A la JavaScript?
autrijus but I'm not sure about that (the list/lookup relationship differ), so I'm keeping those two. 18:31
in certain other languages, arrays are always bounded 18:33
but in perl6 arrays are more like doubly linked lists.
Aankhen`` Bounded? 18:34
autrijus having a finite set of possible keys.
usually ordered
Aankhen`` Ah. 18:36
autrijus instance Arbitrary Container where arbitrary = oneof $ map return [ Scalar, Array, Hash ] coarbitrary = assert False undefined 18:38
er.
instance Arbitrary Container where
arbitrary = oneof $ map return [ Scalar, Array, Hash ]
coarbitrary = assert False undefined
autrijus praises the infinite conciseness of QuickCheck
Aankhen`` ?eval sub foo () { sub bar () { "BAR" }; bar() }; say foo(); say bar(); 18:47
evalbot6 BAR BAR bool::true
iblech Aankhen``: re. That's correct, sub bar {...} is really BEGIN { our &bar = sub {...} }, so &bar is not installed lexically 18:50
You've to use my sub bar if you want the second bar() to fail.
Aankhen`` Ah, I see. 18:51
Is Hash::kv known to bomb in PIR?
autrijus probably. 18:54
nothingmuch www.cpushack.net/ 19:28
specifically www.cpushack.net/CPU/cpu.html#tableofcontents 19:29
interesting processors: 19:31
ARM
AMD 29000
intel IA-64
DEC Alpha
Darren_Duncan I'm happy to report that my OSCON Lightning Talk was accepted 19:35
darrenduncan.net/OSCON/OSCON2005Lig...ngTalk.txt
I appreciate any input to chopping about a quarter of the speech length off while retaining the highest impact parts and/or say everything more tersely
so who here is going to OSCON? 19:36
svnbot6 r5910 | iblech++ | PIL2JS: 19:37
r5910 | iblech++ | * my $a := @a, my $a := %a work correctly now (hopefully).
r5910 | iblech++ | * The @foo in sub foo (@foo) {...} is guaranteed to be an Array and not a
r5910 | iblech++ | Ref, even if called like foo [1,2,3].
r5910 | iblech++ | * The $foo in sub foo ($foo) {...} is guaranteed to be a Scalar (for example,
r5910 | iblech++ | a Ref of Array), even if called like my @a = <a b c d>; foo @a.
r5910 | iblech++ | * Removed Prelude::JS::OO as &ref and &isa need to operate not only on
r5910 | iblech++ | references, but on aggregates, too. Moved the implementation of &ref and
r5910 | iblech++ | &isa to libjs/PIL2JS.js. 19:38
r5910 | iblech++ | * PIL::Nodes -- my @a is not my @a = (undef) any longer, but my @a = ().
r5910 | iblech++ | Similar for %h.
r5910 | iblech++ | * Probably many tests broke, will go through them and fix.
coral wow 19:39
iblech %hash.elems and $pair.elems should not work, right?
(Only @array.elems)
nothingmuch coral: don't be alarmed, iblech simply likes karma D;-) 19:42
for example "Probably many tests broke" should not be a ++ at all ;-)
iblech But on the same line there's "will go through them and fix" ;)
nothingmuch that's +=0
coral no alarm here 19:43
i've been scaring people by talking about PIL2JS
nothingmuch okay then s/alarmed/impressed/
seriously though, iblech++ # wonderful work
coral yes
iblech++
iblech Thanks :)
nothingmuch Darren_Duncan: please expand on Rosetta::Emulator::DBI 19:44
autrijus iblech++ # my task is reduced to documenting and formalizing your work :)
nothingmuch and give the state of current "drivers"
ah, the term I was looking for is "Engines" 19:45
"An app can easily query whether an Engine provides for its needs." <-- what does this mean?
coral jini!
autrijus :r
oops.
nothingmuch does Rosetta offer support certain features on top of the engines, in a slow way?
coral nothingmuch: that phrase you asked about reminds me of an old sun idea called "jini". 19:46
Darren_Duncan The query thing can be done by the Rosetta method features()
Rosetta::Emulator::DBI has the same API as DBI but uses Rosetta as a back-end 19:47
nothingmuch coral: what does it do? 19:48
Darren_Duncan given the length of the talk, I consider this module to be a tersest-priority ... important to name, but not go into much detail
nothingmuch i just read about not very old but pretty weird sun idea called "picoJava" on that cpushack page 19:49
coral nothingmuch: imagine zeroconf crossed with api discoverability
nothingmuch Darren_Duncan: that's one sentance that makes it obvious.
coral "hm, i seem to be on a network. anyone know how to get to the internet?"
nothingmuch coral: polymorphic zeroconf?
coral "hm, i seem to be on a network. anyone want to get to the internet?"
Darren_Duncan Rosetta itself mainly just defines an API ... its added features mainly amount to doing input checking
nothingmuch that was always my fantasy about what apple's rendevouz is or will be 19:50
Darren_Duncan I wanted to give Engines the maximum flexibility to implement how they like
coral i very like the idea of advertising web services over rendezvous
nothingmuch until i realized that it's just "i'm ._local.tcp.moose"
Darren_Duncan putting too much in the API may constrain Engine developers from using best practices
nothingmuch Darren_Duncan: what engines are available?
Darren_Duncan there's no reason one Engine can't subclass another if they want to share 19:51
only one Engine exists currently, Rosetta::Engine::Generic, and it is technically a reference implementation, not the be all and end all
but until more are made, it should service many needs
nothingmuch Darren_Duncan: so basically in my Rosetta::Mapper::OO i will give a spec of "These features of the rosetta api" and anything I use under that is interchangiable? 19:53
Darren_Duncan all Engines are interchangeable that declare support for the features() you want to use ... you can use that function to see programmatically that a user-chosen Engine will work for your Mapper::OO 19:54
coral ooo, i like that
Darren_Duncan part of my debate is how much "how to" info I put in the talk, and how much is left for people reading the POD 19:55
I would love to say a bunch of "how to", but time is short, the talk is mainly to get people interested to find out more 19:56
this said, I will tweak it now to account for your questions, to help aid in *understanding* 19:57
autrijus iblech: formalization #1 is in 19:58
svnbot6 r5911 | autrijus++ | * container type formalism, draft #1
iblech oooh :) /me goes to look
autrijus I need to write it down in English too, so I'll do it here
if you can help integrating it into comments, all the better. I'll also post to p6c :)
(or is this p6l? not sure) 19:59
I suspect p6c
the gist is that there are only three kinds of Box
a symbol is only bound to one of them, and never changes the type of its binding
so we essentially have three zones, so to speak 20:00
there is no way to cross zones
this is neatly signified by sigils, but in the sigiless variant one would need to use explicit annotation.
now a box is a mutable cell containing two things. one is its id 20:01
and one is a possibly-tied content.
toggling tiedness does not change their IDs
now, there is an open question 20:02
what happens when you do
untie(%ENV); ?
is that even allowed? 20:03
iblech Then +%ENV is 0?
I.e. %ENV is then a plain normal hash
autrijus but that means %ENV need to keep a shadow normal hash around
because when you tie something, untie, manip the content, tie again, untie again 20:04
we expect the content to be still there
right?
iblech Right.
autrijus okay, then every tied variant needs to keep an untied counterpart around. 20:05
(also this is in marked contrast with perl5)
where you can't untie %ENV
20:06 brentdax_ is now known as brentdax
iblech Yep, as in P5 %ENV is simply a special variable supplied by perl, and in P6 it's an ordinary tied variable which happens to call Perl6::Internals::get_env($key) 20:06
Darren_Duncan okay, I just updated darrenduncan.net/OSCON/OSCON2005Lig...ngTalk.txt ... 20:08
mainly at the end
one sentence was edited to "An app can easily query the features() method to see whether an Engine provides 20:09
for its needs."
onother became "Use the 'Rosetta::Emulator::DBI' module, a DBI clone, and your legacy apps will
gain freedom with nary a change; so no porting is necessary."
the ", a DBI clone" was the added part
that's how I shortened "provides the same API as DBI does" 20:10
autrijus iblech: ok, then I changed the model to reflect this
r5912
svnbot6 r5912 | autrijus++ | * tied stuff keeps the untied version around. 20:11
autrijus each box now has three things: an id, a boxed value, and maybe a tied version of the value.
iblech Ok. 20:12
autrijus I wonder if there can ever be that we have a box with scalar value 20:13
but somehow tied as array.
does this make any sense?
nothingmuch tied version of the value == itype interface? 20:14
and value = representation of box?
Darren_Duncan: that sounds better 20:15
geoffb (coming in very late to this discussion) -- autrijus, do you mean like tying a string as an array of chars?
Or a blob as an array of lines?
iblech autrijus: Hm, I'd say no. $s are always scalars, no matter what. You can make $foo an arrayref, or you could $foo = Foo.new; method Foo::postcircumfix:<[ ]>($idx) {...} 20:17
svnbot6 r5913 | autrijus++ | * not always tied.
Darren_Duncan also prepended "Engines are interchangeable, and " to one sentence
nothingmuch, thanks for your input so far
nothingmuch when is OSCON? 20:18
Darren_Duncan OSCON starts in 2 days, on Monday August 1st
the Lightning Talks are on thursday the 4th
I have until then to revise the speech to make it better and fit it in 5 minutes
the current one takes about 6-7 minutes to read aloud at a normal pace 20:19
nothingmuch do you think you can make an engine for a specific database in that time? 20:20
autrijus iblech: ok, I'll reflect that then.
nothingmuch i think Generic is not enough for most people to say "i'm trying that out as soon as I get home", but is enough to get most people to say "i'll try that out in a few months"
autrijus that makes it extremely easy to model.
iblech :) 20:21
autrijus r5914.
we're down to 6 lines!
so, a Box of something is either untied, where it contains an Id and a something 20:22
svnbot6 r5914 | autrijus++ | * Tied box variant is not allowed to disagree with untied.
autrijus or tied, where it contains Id, something, and a tied something.
during tie/untie, that something stays invariant. 20:23
Darren_Duncan nothingmuch; no, it would probably take longer
I'm not exactly a fast worker
autrijus thinks that is miles better than SvMG
coral aiee, SvMG
coral chants the ritual cleansing curse
and when the magic has passed through me, only perl remains
autrijus heh. 20:24
now, this treatment is different from Mg
in MG, you can layer multiple tiedness on a container
Darren_Duncan its already taken me 3 years of development to get where I am now, and basic selects etc aren't fully implemented yet
autrijus and selectively pull away ones
do we want that?
geoffb autrijus, can you explain what's wrong with my question earlier (Why is making a string look like an array of chars not a tie of different container types?)
autrijus geoffb: oh, sorry, missed it
geoffb np 20:25
Darren_Duncan what does work in that regard is that you can define selects, and generate the SQL, and connect to databases ... I have to connect the dots where the SQL is actually issued
autrijus geoffb: $str ~= "hello"
you want this to trigger PUSH?
Darren_Duncan brb
geoffb yes, or alternately push @str, ['h','e','l','l','o'] doing a concat
autrijus so 20:26
$str = DBI.connect(...)
is typefail
since it's not a string 20:27
geoffb nod
autrijus then this is indeed tying a scalar to an array. 20:28
svnbot6 r5915 | iblech++ | * t/builtins/arrays_and_hashes/kv.t -- s/%hash.elems/+%hash/, as, according to
r5915 | iblech++ | S29, .elems works only on arrays. s/$pair.elems/1/ for the same reason.
r5915 | iblech++ | * New t/builtins/arrays/elems.t -- Tests for .elems.
autrijus I mean, as an array.
geoffb My view of tie was always "This allows me to make anything look like anything else -- I can really pretend this object is an array, or in particular this string is an array, this DB is a hash . . . "
nothingmuch the way i saw ties is polymorphism extended into N syntaxes
you are no longer bound to $array.index($index) for polymorphism
autrijus geoffb: that is not perl5's tie anymore 20:29
nothingmuch you can use a more intuitive interface: @array[$index]
autrijus that is something far more radical :)
nothingmuch autrijus: example please =)
autrijus nothingmuch: geoffb's.
nothingmuch the str push? 20:30
autrijus this has the net effect of implementing
$str := @chars;
geoffb Sure, perl 5's tie required an object (ref) underneath, but I always saw that as merely working around limitations of Perl 5's object model, not anything essential to the concept
nothingmuch tie is just 'does on containers'?
geoffb nothingmuch, if I understand you right, yes
autrijus and then all manip on @chars gets reflected to $str, and all manip to $str reflects on @chars.
nothingmuch doesn't that introduce N-ary propagation issues 20:31
autrijus nothingmuch: yes.
geoffb: you can already establish such relationship in perl5 using mutual refs
nothingmuch like if the middle of a string was slices, what happenes to the object
autrijus or one way ref and overload
geoffb: but that is going to hurt performance a lot
so are you asking for a more native support for this?
nothingmuch if a char is inserted between two arrays, which does it belong to?
geoffb I'm not saying perl 5 can't do that -- I was asking why you were insisting that a container has to have the same type in both tied and untied case
autrijus geoffb: oh, it's just agreeing with perl5. 20:32
geoffb OIC
autrijus I wasn't insisting :)
I was asking.
nothingmuch does uising a string as an array create a new container containing an array value tied to the string?
autrijus nothingmuch: if so, there's nothing new
geoffb I would like the bigger concept -- but I grant that it may be difficult/non-performant
autrijus nothingmuch: geoffb I believe is asking for the same container to act polymorphically.
nothingmuch okay, so how does the value behave? 20:33
autrijus i.e. you have a Box Scalar constrained to contain Str, and then tie it to Tie::Array::Something
nothingmuch: it's not about the value
it's more like ~= now calls PUSH
instead of FETCH+STORE
geoffb shrugs. I like having expressive power, and a cheat sheet that says "this will kill performance, but it's there if you ever need it" :-)
autrijus geoffb: but you can simulate that pretty easily :)
so there is expressive power
geoffb Well, fair enough.
nothingmuch i don't understand why it's so cool, frankly 20:34
autrijus like, if you mmd over all Str methods and translate it to PUSH calls etc.
that will instantly work.
just not very fast.
geoffb Hmmm.
Please correct my assumption here: My assumption is that containers come in two flavors, tied and untied. Untied containers allow the compiler/prelude to be as dirty as desired to be fast. Tied containers force the compiler to emit completely primitive operations. 20:35
(er, and part of that last sentence was, "but everything still works, just slowly") 20:36
geoffb is clearly missing something, sigh 20:38
nothingmuch is confused 20:56
svnbot6 r5916 | autrijus++ | * reflect the design decision that "is Tieable" is a compile-time
r5916 | autrijus++ | decision that cannot be reverted at runtime.
nothingmuch i stretched my arms up
and the light flickered
so i drew them back, and did it again, and it flickered again
turns out it was the auto-brightness adjustment thing on the laptop 20:57
for some reason i *still* think it's cool
autrijus geoffb: no, your understanding is sound 20:58
see PIL.hs 20:59
r5916
nothingmuch oh shit, i have 50 browser tabs to close again 21:00
uptime is a curse in the days of tabbed browsing
autrijus geoffb: there are some comments finally. please check if they disagree with your intuition
iblech: 21:02
Aankhen`` G'night.
autrijus my %x; %x := %*ENV;
what happens?
or, try vice versa.
nothingmuch bzzt
autrijus according to my current model, it's either error bothways, _or_ %x only binds to the normal storage of %*ENV and ignore the magic 21:03
while %*ENV := %x would drop the existing magic. I think.
iblech Hm, I'd say %x := %*ENV is error, as %x is not tieable, and %*ENV := %x causes %*ENV to loose its tiedness and becomes a normal a hash then 21:04
autrijus but.
%*ENV := %x; tie(%*ENV)
is sound
so what really happened is just an implicit untie() 21:05
to match the tiedness of %x.
nothingmuch holy crap!
it seems my disk on key is alive!
it's data got zapped
but it's still functional!
iblech Yes, but where's the contradiction with my assuption? :) %*ENV := %x causes %*ENV to still be marked as potentially tieable, but, as %x is not tied, %*ENV isn't either after the binding 21:07
But you can re-tie %*ENV, while you couldn't tie %x, as %x isn't marked as tieable
autrijus in that case %x := %*ENV is a compile time error, even. 21:11
iblech Yep. 21:12
autrijus cool. 21:13
iblech :)
autrijus so 21:15
untie(%*ENV); %x := %*ENV;
sound?
iblech Yep. 21:16
But
autrijus but? :)
iblech then we can't generally detect %x := %*ENV as a compile-time error, consider: (&untie, &id).pick()(%*ENV); %x := %ENV
nothingmuch so how do you alias %*ENV to %x?
::= ? 21:17
iblech (As we can't know at compile-time if %*ENV is tied or not)
s/if/whether/
autrijus why?
oh, you mean at runtime.
right.
that's why I asked :) that means no compile time error checking in general. 21:18
you meant "we can't know, at compile time, whether %*ENV will be tied at runtime)
iblech Right, exactly.
svnbot6 r5917 | iblech++ | * Usual svn props.
r5917 | iblech++ | * t/builtins/lists/join.t -- Tests for join on one-element lists (should work),
r5917 | iblech++ | and "hi".join(...), which should not work (but ("hi",).join(...) should work
r5917 | iblech++ | -- that's the same as (42).kv (should not work) and (42,).kv (should work)).
r5917 | iblech++ | * PIL2JS: Unbroke many tests:
r5917 | iblech++ | * Prelude::JS -- Stringification of one-element arrays and hashes works now.
r5917 | iblech++ | * "Fixed" the stringification of pairs with keys eq "a" -- it's the Pair
r5917 | iblech++ | being used in named parameter binding as well as normal object thing. 21:19
r5917 | iblech++ | * Prelude::JS::Array -- join on one-element lists should work.
r5917 | iblech++ | * Prelude::JS::Keyed -- Unbroke delete on arrays.
r5917 | iblech++ | * Prelude::JS::Hash -- Unbroke { $only_one_pair }.
nothingmuch damn, it won't even partition 21:24
nothingmuch wonders if he needs 1gb or 512mb
svnbot6 r5918 | kolibrie++ | added make_swatch.p6 for making a paint colors swatch 21:45
r5918 | kolibrie++ | autogenerates an SVG image
r5919 | autrijus++ | * bind and untie
r5920 | iblech++ | * Renamed t/builtins/lists/undef_in_map.t to empty_list_in_map.t and added two
r5920 | iblech++ | more tests to it.
r5920 | iblech++ | * PIL2JS: PIL::Nodes -- Unbroke &try -- many tests should pass again now, resmoking...
Darren_Duncan are there any plans to do another Pugs release before OSCON? 21:49
svnbot6 r5921 | autrijus++ | * cosmetic polishing and avoid warnings
autrijus hm. 22:04
Darren_Duncan: when is that?
iblech: consider
$x := \@y;
$x = 3;
# error: 3 is not Ref ?
iblech Yep, think so.
Darren_Duncan OSCON starts in 2 days
Aug 1-5
autrijus will ($x = \@z) work in place of ($x = 3)?
Darren_Duncan the conference proper is Aug 3-5, tutorials on 1-2 22:05
autrijus the question is whether \@y produce an rvalue container
which I think it does
iblech Yep. And @y ~~ @z will be true then, I think
autrijus otherwise you have
\@y = \@z;
which is highly dubious. 22:06
Darren_Duncan see conferences.oreillynet.com/os2005/
I'm heading off there at this time tomorrow 22:07
autrijus ok , then I'll try a release. 22:08
svnbot6 r5922 | iblech++ | PIL.hs -- Hadockified comments and added a #ifndef HADDOCK around the GADT def. 22:15
autrijus iblech: 22:31
%a =:= $b;
always false or always error?
I'm inclining toward false.
iblech Hm, what about: my $b := @a; @a =:= $b?
Probably false, as $b is a Ref of Array and @a is a Array 22:32
Both false then
autrijus nod. 22:33
I can argue the other way too
but this way feels a bit cleaner.
(I could argue that \@x always retain the Id). 22:34
\@x =:= @x;
iblech Hm.
autrijus I don't think it's wise though :) 22:35
iblech Consider: "How can things of different types (Ref of Array, Array) ever be identical" -- so, yes, \@x =:= @x should be false. 22:36
BTW, is there a negated =:= op (like ne or !~)?
autrijus !:! ? I don't think so 22:37
iblech You're right, according to S03. Probably because a. it wouldn't be needed that often anyway and b. !:! is *ugly* 22:38
Darren_Duncan I was thinking about that the other day, because I want a negative =:= 22:40
I thought perhaps calling it !=:= or !:=, to be parallel with != for == 22:41
spinclad =!= ? 22:42
svnbot6 r5923 | iblech++ | PIL2JS:
r5923 | iblech++ | * PIL2JS.js -- Hack, all things which .isa("Array") are .isa("List") now, too.
r5923 | iblech++ | * Prelude::JS::Keyed -- Fixed &pairs, &kv, and &values on one-elem
r5923 | iblech++ | lists/arrays/hashes.
spinclad like a symmetrized != 22:44
it doesn't keep the : (from :=), but preserves some of =:='s flavor. 22:49
autrijus sure, but I highly doubt the practical need for this construct :) 22:50
spinclad it forwards the crusade to program solely in punctuation! 22:53
autrijus riight. 22:54
svnbot6 r5924 | autrijus++ | * nomenclature fixups; equivalence test. 22:55
iblech Is stringification/numification/booleanification of real refs, i.e. refs which do not autoderef, specced somewhere? 22:56
autrijus 3 + \$obj; ?
iblech Right. 22:57
autrijus my impression was that reference is always true, and does not have str/numify forms.
iblech But err, \$obj will autoderef if $obj contains a "real object": www.nntp.perl.org/group/perl.perl6....uage/22532
autrijus right 22:58
...so?
iblech so 3 + \$obj will evaluate to 3 + $obj, which in turn will eval to 3 + coerce:<as>($obj, Int) or somesuch 22:59
How about 3 + \3?
autrijus 3+\3 is instant death, as no auto deref is done
I was talking about that.
iblech oh, ok. 23:00
autrijus 3+$obj is also death if $obj has no coercion forms.
iblech Right.
autrijus good, we agree again
iblech :)
svnbot6 r5925 | iblech++ | New test: t/var/autoderef.t -- Tests for autodereferencing/non-autodereferencing. 23:12
autrijus I need to sleep. ciao! :) 23:13
iblech Night :)
autrijus++ # "Definition of containers."
wolverian hmm. debian is stopping mozilla support in stable, apparently. 23:16
iblech Need to sleep, too. Night all! :) 23:55

Powered by Cro and the Raku® Programming Language.

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

Thank you!