pugscode.org/ | nopaste: sial.org/pbot/perl6 | pugs: [~] <m oo se> (or rakudo:, kp6:, smop: etc.) || We do Haskell, too | > reverse . show $ foldl1 (*) [1..4] | irclog: irc.pugscode.org/
Set by TimToady on 25 January 2008.
00:04 chris2 left 00:06 icwiener left 00:12 cmarcelo left 00:30 hercynium left 00:33 Alias_ joined, Limbic_Region joined 00:40 bacek joined
pugs_svn r21303 | ruoso++ | [smop] fixing memory leak in p6opaque. test/12 is valgrind clean again... 00:41
00:50 nipotan is now known as nipotaway 01:09 alester_ joined 01:15 alester left
pugs_svn r21304 | ruoso++ | [smop] test/14_p6opaque_methods compiles and runs, but it is obviously not passing yet. 01:17
01:44 alanhaggai left 02:02 redicaps left 02:05 Southen_ joined
speckbot r14561 | larry++ | [S12] s/bool/Bool/ 02:07
02:08 rhr_ is now known as rhr 02:20 alanhaggai joined, simcop2387 joined 02:25 kanru left, Southen left 02:28 Jamtech joined 02:32 justatheory left, [particle]1 joined 02:39 meteorjay left 02:48 [particle] left
ingy hola 02:55
ingy fires up his Perl 6 machine 02:56
02:57 kanru joined
pmichaud TimToady: Are there more changes in the queue for the synopses wrt .Bool, .true, etc? Or should I go ahead and start asking questions? 03:04
TimToady feel free 03:05
pmichaud what's the relationship between .true and .Bool ?
TimToady true is an Any method
Bool is a coercion via type name
pmichaud so, given an object in a boolean context, how do we decide what to return? 03:06
e.g., for prefix:<?>
TimToady true() and prefix:<?> coerce to 0 or 1, I think 03:08
pmichaud what logic do they use to decide the boolean value of an object? Do they check .True first, and fall back to .true?
TimToady I think all types should define .true, probably 03:09
pmichaud yes, .true is defined in Any
er, Object
TimToady defaults to .defined, I think...
pmichaud (S02 says .true is defined in Object)
(and yes, it defaults to .defined)
but if I have 03:10
my $x = $a but True; say ?$a;
how does that "but True" get handled in prefix:<?>
sorry, I meant say ?$x
(I really can't type these days. I should type slower.) 03:11
TimToady prefix:<?> calls .true on the object, and the anon type overrides the .true method
pmichaud it overrides the .true method because...?
TimToady it mixes in a role that defines .true, I guess 03:12
pmichaud but that role isn't the Bool role?
TimToady why not?
pmichaud oh. Bool isn't strictly an enum, then
i.e., it's not just our bit enum *Bool <True False> because there's also a .true method in there somewhere 03:13
TimToady Bool is a valid type for an object or a property. True and False are subset types
well, .true may be built-in to bit as a 1 bit int already 03:14
pmichaud but then bit would autobox to Bit in that case? 03:15
TimToady bit :: uint1
if you had to pass it somewhere as an object, it would box to some variety of Int
pmichaud (since native types autobox to their uppercase counterparts when treated as objects)
TimToady not sure if that always has to be the case 03:16
pmichaud ah. I was reading from S02.
TimToady native types are really a funny kind of subset type
except the hardware enforces the limit :) 03:17
pmichaud yes, but adding methods to native types sounds.... tricky
TimToady well, you have to know the container type; you can't ask it what it is...
pmichaud okay, that lost me. 03:18
TimToady an "int" is a container, but not an object, so you have to implicitly know the .HOW of it 03:19
pmichaud okay
so, .true is a method on bit (or int) 03:20
and role composition causes it to override 03:21
TimToady mixin, not composition
pmichaud mixin, thanks.
(still learning all the terms here)
okay, I can follow that, I think. 03:22
TimToady hides the hands he was waving...
pmichaud one more question: what's the output of say (?0).perl ?
TimToady perl6: say (?0).perl 03:23
p6eval elf 21304: OUTPUT[0ā¤]
..pugs: OUTPUT[Bool::Falseā¤]
..rakudo 29275: OUTPUT[Bool::Falseā¤]
03:24 redicaps joined
pmichaud (rakudo does Bool::False simply because that's what Pugs and the test suite do.) 03:24
TimToady I am still slightly undecided on whether ? should be returning a bit or a Bool. can argue it either way, typologically
pmichaud oh, I was able to see lots of ways to argue it, which is why I asked :-) 03:25
TimToady there's probably a lot to be said for trying to preserve enumness 03:26
and the Bool-is-really-bit-at-runtime is kind of a compiler-centric view 03:27
where compiler is defined as that thing that throws away all the useful type info :)
certain +?$x should return 0 or 1 03:28
pmichaud I just think of it as lossy compression :-)
TimToady but darn it, computers treat all bits as booleans, in some sense...
and certainly P5 programmers are used to thinking of booleans numerically 03:29
(if you consider "" to be a variant of 0...)
03:31 Limbic_Region left, alanhaggai left 03:33 nothingmuch left
TimToady I also get a bit tired of seeing Bool::True and Bool::False everywhere... 03:33
pmichaud also Order::Decrease, Order::Same, Order::Increase (from cmp) 03:34
TimToady for most purposes, if there is no conflict, True/False or Decrease/Same/Increase is sufficient for clarity 03:35
but maybe it's a false economy 03:36
maybe a pragma is the appropriate solution here 03:37
use perl :enum<explicit> 03:38
hmm
pugs: say perl(1+2); 03:39
p6eval pugs: OUTPUT[3ā¤]
TimToady perl6: say perl(1+2);
p6eval elf 21304: OUTPUT[Undefined subroutine &GLOBAL::perl called at (eval 119) line 3.ā¤ at ./elf_f line 3861ā¤]
..pugs: OUTPUT[3ā¤]
..rakudo 29276: OUTPUT[Could not find non-existent sub perlā¤current instr.: '_block11' pc 36 (EVAL_11:16)ā¤]
TimToady 'nother one of them little spec holes... 03:40
03:40 pbuetow_ joined
TimToady I wonder if we should make .repr spit out Python code. :) 03:41
pmichaud :-)
I don't need an immediate answer to any of these -- just want to be able to answer questions for Rakudo and the test suite
TimToady in some of these cases there really isn't a best answer, but it'd be nice if the implementations all agree at least 03:43
(but I do hate arbitrary decisions...)
pmichaud I have another question from the earlier questions, but I'm not quite sure how to phrase it 03:44
so bear with me
(keeping in mind I still don't _completely_ understand roles/properties)
if we have our bit enum *Bool <False True> 03:45
and then my $x = 0 but True;
then $x has a .Bool method, a .True method, and a .False method mixin'ed, yes?
and it also gets all of the methods defined on the 'bit' type as well? 03:48
TimToady if so, it would indicate the object in $x is of type Bool rather than just type bit 03:49
and the "bit" type is really just a storage hint for compaction purposes 03:50
presumably 1.False returns a Failure 03:51
03:52 cain_22 joined
TimToady you automatically get all typenames as methods, I think 03:52
03:52 pbuetow left
pmichaud as methods on.... ? 03:53
TimToady and maybe those are special cased as coercions, so can be defined either incoming or outgoing to a type 03:54
pmichaud well, that kinda brings up another question: my enum Foo @foo; # valid? 03:55
TimToady would have to value of @foo at compile time 03:56
pmichaud okay.
TimToady like in use Foo @foo;
family calling; will backlog &
pmichaud no problem. I still don't quite understand the "indicate object in $x is of type Bool" and how .true ends up in the mixin'd value 03:57
(I can understand that it's in 'bit', but I don't see how we get the methods from 'bit' into the mixin'd value -- that seems like more than a simple role mixin) 03:58
afk for a while 04:11
04:20 wknight8111 left, c1sung left 04:24 Jamtech left 04:45 casyn left 04:51 simcop2387 left 04:54 simcop2387 joined 05:07 kanru left, kanru joined 05:14 Ara5n joined 05:17 [particle]1 left, kattana_ left, cognominal_ left, Guest16015 left, p6eval left, orafu left, yves_ left, ilbot2 left 05:18 Patterner left, cj left, Psyche^ joined, Psyche^ is now known as Patterner 05:19 Grrrr left, Grrrr joined 05:22 pugs_svn left, pugs_svn joined, [particle]1 joined, kattana_ joined, cognominal_ joined, p6eval joined, yves_ joined, Guest16015 joined, orafu joined, ilbot2 joined, bacek left 05:23 baest left, bacek joined 05:24 baest joined 05:25 Psyche^ joined 05:28 alanhaggai joined 05:29 c1sung joined, Ara4n left 05:30 redicaps left, s1n left 05:31 mtve left 05:34 s1n joined 05:38 stevan_ left 05:42 Patterner left, Psyche^ is now known as Patterner, cj joined 05:43 GeJ_ is now known as GeJ 05:47 sri_work left 06:29 alanhaggai left 06:32 alester_ left 06:34 alanhaggai joined 06:50 barney joined 06:57 pmurias joined
pmurias TimToady: how does an object specifiy it's hash function? 06:58
lambdabot pmurias: You have 1 new message. '/msg lambdabot @messages' to read it.
07:01 jiing left 07:20 nothingmuch joined 07:22 mtve joined 07:24 alanhaggai left 07:25 redicaps joined, nothingmuch left
pmurias pugs: module Foo {our $foo=7};say $foo; 07:40
p6eval pugs: OUTPUT[*** ā¤ Unexpected ";"ā¤ expecting "::"ā¤ Variable "$foo" requires predeclaration or explicit package nameā¤ at /tmp/y87Qe9eh8O line 1, column 33ā¤]
pmurias pugs: module Foo {our $foo=7;};say $foo;
p6eval pugs: OUTPUT[*** ā¤ Unexpected ";"ā¤ expecting "::"ā¤ Variable "$foo" requires predeclaration or explicit package nameā¤ at /tmp/ikMRRsLrnP line 1, column 34ā¤]
pmurias pugs: module Foo {our $foo=7;};say $Foo::foo;
p6eval pugs: OUTPUT[ā¤]
pmurias pugs: module Foo {our $foo=7;};say $*Foo::foo;
p6eval pugs: OUTPUT[ā¤]
pmurias pugs: $Foo::foo=7;say $Foo::foo;
p6eval pugs: OUTPUT[7ā¤]
pmurias pugs: $Foo::foo=7;say $*Foo::foo;
p6eval pugs: OUTPUT[ā¤]
pmurias pugs: $*Foo::foo=7;say $*Foo::foo;
p6eval pugs: OUTPUT[7ā¤] 07:41
07:43 tuxdna joined
pmurias ruoso: hi 08:05
08:08 iblechbot joined
pmurias ruoso: is there any sort of reference counting debugging mode? 08:11
i have found SMOP_LOWLEVEL_MEM_DEBUG, but it floods me with not very usefull data 08:31
08:44 kattana_ left 08:50 jiing joined
pmurias moritz_: the current evalbot backends don't offer any security right? 09:11
moritz_ pmurias: pugs does 09:12
pmurias: and the others offer security by inability ;)
at least a bit ;)
actually kp6 also has a safe mode
let's ask the other way round - how much would you benefit from evaling sm0p code on evalbot? 09:13
ruoso pmurias, SMOP_LOWLEVEL_MEM_DEBUG is one way... other way is valgridn... it will show you where the unfreed memory was allocated... 09:21
pmurias moritz_: kp6 does not have a *working* safe mode 09:24
09:26 meppl joined
pmurias ruoso: i'm debuggin memory being freed twice 09:27
09:28 chris2 joined
pmurias * debugging 09:28
is there an equivalent of caller in c? 09:30
09:34 elmex joined
pmurias i could add a log of reference counter increments and decrements to SMOP__Objects if it did 09:36
rakudo_svn r29285 | bernhard++ | [Rekudo] Some tidbits in docs/compiler_overview.pod and README 09:38
pmurias maybe scripting the gdb would work... 09:39
ruoso: would a tool showing where was the reference counter of a prematurly free'ed of leaking smop counter changed be helpfull to you? (i don't know if it's worth doing something like that) 09:42
buying a gsm modem& 09:46
09:46 pmurias left 10:06 ruoso left 10:09 iblechbot left 10:19 wknight8111 joined 10:21 bacek left 10:22 meppl left 10:24 donaldh joined 10:26 Kattana joined 10:29 meppl joined 10:38 redicaps left 10:58 b_jonas joined 11:04 wknight8111 left 11:35 pbuetow_ left, pbuetow joined 11:43 ruoso joined 11:47 wknight8111 joined 11:57 broquaint left 12:04 broquaint joined 12:09 tuxdna left 12:11 mjk joined, alanhaggai joined 12:38 wknight8111 left 12:40 rindolf joined 12:44 cmarcelo joined 12:47 iblechbot joined
moritz_ is offline for the weekend, see you call 12:58
13:01 alc joined 13:04 rurban joined 13:08 awwaiid_ is now known as awwaiid
rakudo_svn r29290 | pmichaud++ | [rakudo]: spectest-progress.csv update: 95 files, 1691 passing tests 13:09
13:09 alanhaggai left 13:19 rindolf left 13:21 ruoso left 13:36 rindolf joined, mjk left 13:40 alc left 13:41 Schwern joined 13:42 pmurias joined 13:50 rdice joined 14:00 wknight8111 joined 14:02 rindolf left 14:05 rindolf joined, barney left
rurban Can somebody please update the pugs webpage with the recommended installation method? (Install ghc, Cabal, cabal-install, HTTP, zlib, Pugs) 14:05
14:06 thepler_ joined
rurban BTW: How to submit an pugs installation guide for cygwin? 14:09
14:18 rindolf left
pmurias rurban: isn't the general windowsy best practice to make a installer? 14:22
rurban win32 yes, cygwin not.
I'm just writing it to my use.perl journal....
14:26 rindolf joined
pugs_svn r21305 | pmurias++ | [smop] 14:34
r21305 | pmurias++ | hashes are DESTROYEDALL'ed corretly
r21305 | pmurias++ | fixed sm0p.pl bug
r21305 | pmurias++ | added untested infix:eq
14:35 pmurias left 14:38 hercynium joined 14:43 scrottie left, jhorwitz joined
rurban cygwin pugs installations at use.perl.org/~rurban/journal/36897 14:48
lambdabot Title: Journal of rurban (7989)
14:54 alanhaggai joined 14:56 [particle]1 is now known as [particle] 15:01 alester joined 15:04 rurban left 15:12 rdice_ joined 15:19 barney joined 15:28 rdice left, donaldh left 15:31 lumi left 15:33 nothingmuch joined, jjore left 15:38 lambdabot left 15:40 lambdabot joined 15:42 lumi joined 15:43 b_jonas left 15:44 Lorn_ joined 15:50 jan_ left 15:51 justatheory joined 15:57 Lorn left 16:11 zamolxes joined 16:16 b_jonas joined 16:35 zamolxes left 16:56 Schwern left 17:02 ruoso joined
ruoso @tell pmurias, please remember of adding your changes to the Changelog in the smop wiki 17:08
lambdabot Consider it noted.
17:10 cain_22 left 17:13 nipotaway is now known as nipotan 17:48 kanru left 17:49 pmurias joined
pmurias ruoso: rehi ;) 17:50
ruoso pmurias, hi
pmurias the changelog should contain only full changes ? 17:51
pugs_svn r21306 | ruoso++ | [smop] More advances in p6opaque_methods. skeleton for the p6opaque methods array proxy object
ruoso pmurias, not really... I tend to log as something advances, even if not finished yet...
my idea is that someone would have how to help in this something I still am working on 17:52
pmurias why do we need a changelog, isn't svn log enough?
ruoso hmmm... well...
I don't know... to have it side-by-side with the roadmap?
for instance... you're working in a lowlevel hash 17:53
that is in the roadmap...
pmurias ruoso: any other reasons? 17:57
ruoso it will also serve as a report on the progress for the grant 17:58
pmurias having the svn log side by side could be arranged
ruoso pmurias, I know... but I usually don't add in the changelog as often as I commit...
Usually, for an unfinished activity, I log once a day
18:01 rindolf left
pmurias food& 18:03
ruoso: once i finish the Hash i'll add my changes to the changelog 18:09
ruoso that's cool 18:10
pmurias ruoso: i haven't finish it yet ;(
* finished
ruoso no problem... 18:11
pmurias it's strange that $a eq $b compiles to &infix:<eq>($a,$b) not to $a.infix:<
18:11 rindolf joined
pmurias sorry, it's the reverse actually 18:11
rindolf: hi
how is your lisp dialect
?
rindolf Hi pmurias 18:12
pmurias: what's up?
pmurias: it's in deep design stages.
pmurias: and it would be hard for me to work directly on Parrot.
Because I'm banned from both IRC and email.
pmurias rindolf: you can write a smop backend (once smop matures a bit) 18:13
rindolf pmurias: ah.
pmurias: what is smop?
ruoso rindolf, www.perlfoundation.org/perl6/index.cgi?smop 18:15
lambdabot Title: SMOP / Perl 6
rindolf ruoso: checking. 18:16
My Internet is stupid now.
pmurias my internet is stupid two after the so called upgrade
rindolf pmurias: not sure I understand what it is exactly. 18:17
pmurias * stupid too
rindolf: it's a vm/interpreter 18:18
rindolf pmurias: ah.
ruoso rindolf, it's actually a runtime library for Perl 6 very much in the same sense perl 5 is the runtime library for Perl 5 18:19
rindolf ruoso: ah. 18:20
ruoso several ideas were stolen from the p5 runtime...
18:20 Alias_ left
pmurias ruoso: how should i pass a value from a q:sm0p {...} to the enclosing C? 18:24
ruoso you don't
because the code inside q:sm0p won't be executed at that time 18:25
so you can't get the value from that result
you can have a callback method in your object
that will receive the value you want
and do something about it...
take a look at the p6opaque DESTROYALL code...
pmurias a SM0P_RUN(q:sm0p {...}) would be handy 18:26
ruoso pmurias, you don't want to do that
because you will be recursing in the C stack
and that's a bad thing when you want to interact with interpreted code
pmurias ruoso: interpreted code = sm0p? 18:27
ruoso yes
sm0p is interpreted in the interpreter runloop 18:28
that's why I always return immediatly with false from the functions where I have a sm0p block
to let the interpreter call the sm0p instructions later
pmurias callbacks and c don't play very well 18:30
ruoso why not?
but you can always think of your callback as a method call 18:31
as I do in p6opaque.DESTROYALL
which in the sm0p block calls .^!DESTROY
that call is received as a standard method call... 18:32
pmurias i'll have to implement sm0p labels ASAP 18:36
ruoso pmurias, yes... it's starting to be a real pain 18:37
pmurias will need something like SM0P_RUN if we want sm0p to be embeddable 18:38
ruoso but you can call loop on the interpreter 18:39
as the tests do
but you shouldn't do that inside the code...
18:39 tuxdna joined
ruoso because you would be recursing in the C stack 18:39
18:40 Alias_ joined
pmurias re callbacks c doesn't even have closures and complex control flow gets turn into state machines 18:40
* turned
ruoso: the Hash is temorary 18:41
ruoso pmurias, but you have a capture that you can build in the sm0p block and call your method
instead of returning the value, use it as a parameter to a method call
you can build the capture in the sm0p block
and receive the value in the other dispatch call 18:42
pmurias the sm0p call is in a loop
18:44 gaal joined
pmurias the s1p Hash will be replaced by a Hash in perl6 shortly after in the initialisation sequence 18:44
18:44 Exodist joined
ruoso pmurias, but still, recursing in the C stack is a bad idea... 18:44
pmurias i don't think any body will be calling continuations at this stage 18:46
ruoso pmurias, smop is doing it already 18:47
smop_lowlevel counts on continuations...
it already uses Continuation-Passing-Style
that's how it intercepts the execution to call DESTROYALL
pmurias i'm using only constat identifiers keys than 18:49
* constant
ruoso that's an acceptable restriction for the moment
no object is accepting non-constant idenfier for method dispatch yet 18:50
pmurias shower& 18:51
18:59 Alias__ joined, Alias_ left 19:02 chris2 left
pmurias ruoso: how do you plan to achive half-decent performance with smop? 19:03
ruoso pmurias, by someday replacing SLIME with something less lame 19:04
but do you think the performance is too bad already? 19:05
I still couldn't find anything worth benchmarking
pmurias ruoso: we don't execute anything yet, and most of the objects are still in C 19:11
ruoso I think test/14 will already give us some idea on how heavyweight it is, since the method dispatch will already be delegated by non-c code... 19:13
pmurias was just comparing the way we do things with parrot
ruoso not used to handle parrot... 19:14
how do we do things with parrot? 19:15
pmurias i meant i compared the way we do things with the way parrot does them
ruoso and I meant, how do parrot do them? 19:16
;)
pmurias faster
;
)
they have vtables, int and string registers 19:17
ruoso hm... but they still don't know how they will handle representation polymorphism
19:17 tuxdna left
ruoso I was talking with jonathan some time ago, and they're assuming they'll find a way to handle it 19:18
pmurias handle is as a special case most likely
s/is/it/ 19:19
19:19 Chillance joined
ruoso but if the vtable is visible to the interpreter... there isn't really a way to special case it 19:19
(of course they are probably smarter then me and they'll find a way... but I just can't think on a way to do it) 19:20
pmurias you could use SMOP_OBJECT's as pmc if you want
19:20 barney left
pmurias ruoso: the vtable contains only the most common methods 19:20
ruoso the most common being?
smop abstracts one more level 19:21
spinclad i expect vtables will not map one-to-one with .HOWs
19:22 Alias__ left
ruoso spinclad, I'm not sure about it... but afaics... it's something like that... 19:22
pmichaud and [particle] could probably shed a brighter light on it
pmurias ruoso: the most common ones are the ones parrot developers feel will be called most 19:23
ruoso pmurias, but at which level?
pmurias, is it just an optimization? 19:24
pmurias MESSAGE in smop terms
sort of
it's designed in
ruoso except that in smop, the vtable is abstracted inside the message call
pmurias which is slower
ruoso but might be required to support different representations 19:25
which in turns can result in faster again
because I can use optimized low-level values as if they were standard objects
pmurias the parrot folks can to
ruoso well... 19:26
as I said... I'm not really aware on the internals of parrot...
pmurias they sort of assume standard objects are optimized low-level values
ruoso but from what I've talked with jonathan, they still didn't sort out how to deal with different representations...
pmurias * as special case of optimized low-level values
19:27 Alias_ joined
spinclad vtable perhaps corresponds with responder interface 19:27
pmurias spinclad: no
ruoso spinclad, that's what smop does
spinclad no?
ruoso parrot is one level more concrete 19:28
pmurias spinclad: sort of yes
;)
ruoso smop is one level more abstract
spinclad, the vtable in parrot is exposed, in object is hidden inside a responder interface
pmurias spinclad: they are responder interfaces with hard coded common operations
ruoso s/object/smop/
spinclad the vtable is exposed to parrot opcodes, unlikely (as i see it) exposed to Perl 6 (or other HLL) level 19:29
ruoso pmurias, it would be like if we take the lookup inside the message calls of smop and turn it into some static vtable lookup
the difference is one level of abstraction 19:30
spinclad sorry, is exposed to parrot opcode dispatch, as RI is to operation invocation 19:31
ruoso but again.. I'm not familiar with parrot internals... pmichaud and [particle] would probably give a correct explanation on that 19:32
pmurias ruoso: but if an operation is not in the hard-coded set it's handled by a special method code operation, just as in sm0p 19:33
ruoso pmurias, which means that in the end parrot will need as much indirection as smop does? 19:34
pmurias for the less common operations yes ;)
19:35 Jedai joined
ruoso is the vtable subject to cache invalidation or something like that? 19:35
pmurias it's propably pmc specific 19:36
i haven't seen one with cache invalidation but i haven't looked much
spinclad if you replace a sub it points to? you may generate a new vtable and start using it; don't think that's a core operation now.
ruoso well...
btw... it's intersting to make clear that I really think parrot in the end will be a full fletched vm that supports Perl 6 entirely... but one core feature (representation polymorphism) is not addressed yet... 19:39
[particle] parrot has four core data types, with register types to support each: int, num, string, pmc
pmc is an abstract data type
each core data type can be converted to the other, for example $I0 = $S0
parrot uses ops for this 19:40
set_i_s $I0, $S0
pmurias get back to coding seeing much more competent people took up the discussion ;)
[particle] these are hardcoded for int num string types 19:41
spinclad i think vtables apply only to pmcs?
[particle] since pmcs are abstract, the vtable is there to allow pmc writers the ability to write their own functions
for example to convert pmc to int
set_i_p $I0, $P0 19:42
this calls the pmc's get_integer vtable function
$I0 = $P0.'get_integer'()
all pmcs inherit from default.pmc to get some common methods
and all vtable methods can be overridden 19:43
freeze, thaw, init, invoke, get_string, etc
additionally, pmcs can have attributes and methods
19:43 rindolf left
[particle] for example, the p6meta pmc has a 'register' method to register a class 19:44
so, you're not limited to use vtable functions in your pmcs
spinclad for get_attribute operation, does this go through the vtable?
[particle] in fact, the Exporter pmc (for namespace export) only uses methods, no vtable functions
(other than freeze/thaw)
get_attribute is a vtable function. 19:45
ruoso get_attribute meaning high-level object attributes? 19:46
[particle] meaning pmc attribute
pmc is an abstract data type. it defines an structure and an api
spinclad mid-level anyway; not necessarily high-level
ruoso let me rephrase 19:47
is get_attribute vtable function going to be used to fetch the attribute of some Perl 6 Object?
or that is never part of the HLL runtime... 19:48
spinclad yes, 'has $.x' translates to a pmc-level attribute
and accessing it to get_ and set_attribute
[particle] actually, get_attribute is a macro 19:49
something like SET_ATTR_ns_src(interp, SELF, src);
19:49 lde joined
[particle] sorry, GET_ATTR_ns_src(interp, SELF, tmp_ns_src); 19:49
spinclad (at C level?)
[particle] yes
this is converted to PMC *tmp_ns_src = exp->ns_src; 19:50
where exp is Parrot_Exporter *exp = PARROT_EXPORTER(pmc);
Parrot_Exporter is a c struct
pointer
obviously :)
ruoso so every PMC must have the same Parrot_Exporter struct format... 19:51
[particle] i mean, exp is a pointer. uggh
every Parrot_Exporter has the same format
this is for exporter.pmc
float.pmc has a Parrot_Float struct
ruoso so the attribute lookup is static...
[particle] the pmc header is the same struct everywhere
spinclad (i would say 'details, details', but some of the details at least matter for judging how things map between smop and parrot) 19:52
[particle] the pmc data segment is different, based on the attributes it contains
ruoso but the attribute-lookup-process is static, I mean
[particle] what do you mean by static?
ruoso meaning that if you have a Glib object, you need to make a proxy object that map the glib object attributes to pmc attributes 19:53
[particle] yes, then you do your access through the proxy
spinclad the glib object might be a single attribute
ruoso that's the major difference between parrot and smop
[particle] but it's not an incompatible difference
ruoso sure... 19:54
but it's the same difference smop has from perl 5...
one more level of abstraction... 19:55
in smop, the attribute lookup will be entirely delegated to glib 19:56
spinclad one difference i see: the RI for an object need not be found at a fixed slot in it; as long as at invocation time you have the object and its RI, you can invoke an operation of it
ruoso yes... and that's everything smop defines... not a milimeter more... 19:57
everything is delegated to the object itself, not a single static lookup.... 19:58
(except for the RI, of course)...
spinclad as in haskell the type of an object need not be stored in it, and may be compiled away entirely 19:59
merely implicitly known throughout the code
ruoso that's where the one more level of indirection might buy extra room for optimization...
pmurias ruoso: i'm giving up on writing hash tests and hacking in labels to get rid of those stupid integers 20:00
ruoso pmurias++ heh
that's why it's called SLIME ;)
pmurias the target integer in capturize is relative forward :( 20:01
ruoso pmurias, yes.
20:02 icwiener joined
ruoso spinclad, for instance... in p6opaque, the REPR api returns other objects... but this objects will belong to the p6opaque "community", which means that they will know the internal format of p6opaque and navigate directly in it... but the interpreter doesn't know about it, so it is still completely dynamic... 20:02
in fact... the only way to get instances of the other types of the p6opaque "community" is by calling p6opaque REPR methods 20:03
pmurias ruoso: it's possible to that in parrot too
sometimes you will just end up having an extra layer of indirection to esacape the optimalisation 20:04
* escape
ruoso anyway... I think it doesn't matter if smop is superceded by parrot in the very near future... the experience of stressing the representation polymorphism is already fruitfull enough imho... 20:07
spinclad then in the interpreter either you pass around two things for the object (object and RI) or you put together one (p6opaque?) thing that refers to both 20:08
ruoso spinclad, actually every smop object either has a RI or is a RI
p6opaque implements the REPR api 20:09
spinclad 'has an RI': perhaps implicitly? does it have to be represented in its storage footprint?
ruoso spinclad, in smop, yes... 20:10
20:10 nothingmuch left
ruoso spinclad, I couldn't think on a way of doing it other way... 20:10
spinclad, www.perlfoundation.org/perl6/index...._bootstrap 20:11
lambdabot Title: SMOP OO Bootstrap / Perl 6
spinclad an int object, for instance: when used in a community where its RI is known to be Int_RI (say), that could be supplied whenever an operation on it happens
and the int itself takes up just its 2^n bits 20:12
ruoso but how do you introspect?
pmurias ruoso: re superceded, do you plan on throwing away smop?
spinclad i.e. in the scope of my int $n
ruoso pmurias, not really... in fact I think smop will merge with p5 loop somewhere in a not-so-far future
and I think smop will be specially succesfull in the p5-p6 bridge
spinclad, hm... the thing is that in smop even the interpreter is a SMOP__Object* 20:13
spinclad introspection is just such operations, you supply the Int_RI (at least implicitly) and do introspection ops through it
ruoso spinclad, probably some optimizations could be done like that... 20:14
but the standard is that the object must be self-contained
pmurias spinclad: as i have it planned such optimalisations belong to the JIT 20:15
s/to/in
spinclad if by 'the standard' you mean the same as 'outside its community, e.g. the interpreter', i agree, it will need some wrapping then
ruoso yes... but inside p6opaque, for instance, the additional data that builds p6opaque is not SMOP__Object* 20:16
the objects in the same "community" can use the data as plain C structs... 20:17
spinclad agreed
ruoso spinclad, but a simple test like if (SMOP_RI(foo) == SMOP__NATIVE__int)
allows me to call C-level functions for native types
instead of high-level method calls 20:18
and that is how SMOP is bootstrapping its type system
by special casing the known responder interfaces
and using the high-level calling convention otherwise 20:19
for instance...
spinclad at that point, within the community, that '(SMOP_RI(foo) == SMOP__NATIVE__int)' constant-folds to 'true'
ruoso yes...
in smop_lowlevel.sm0p
spinclad and if you have to ask what its SMOP_RI is, you're not a full insider 20:20
ruoso if the object is SMOP__NATIVE__int, the destruction code recurses in the C stack, otherwise it uses CPS to call DESTROYALL on the object...
spinclad, although it's probably a good idea to always ask that, to avoid segfaults... 20:21
pmurias what is a good label sigil?
spinclad i see great potential to compile away overhead here
ruoso pmurias, ^\s*([A-Za-z]): ? 20:22
ah... you mean to use the label later...
can you use bare word? 20:23
pmurias ruoso: not really
ruoso and check for a label before assuming it's a constant identifier?
spinclad (another representation: foo lives in a pair of variables/registers, $foo and ^$foo; SMOP_RI(foo) -> ^$foo)
pmurias ruoso: i can only check for a label after assuming it's a constant identifier 20:24
[particle] can you use the colon?
ruoso spinclad, except that ^$foo means $foo.HOW which is something completely different...
spinclad (if all references to the latter in the code get constant-folded away, the register can go too)
pmurias [particle]: you mean $SMOP__SLIME__CurrentFrame.move_responder(:foo,:bar); ?
ruoso I like that 20:25
[particle] yeah, that
spinclad sorry, i do mean the low-level SMOP_RI, not the high-level .HOW here
ruoso except that it looks like named arguments
but that's ok for sm0p 20:26
[particle] you can use ` 20:27
that's reserved for the user, so not in any normal perl 6
ruoso spinclad, anyway... compiling away overhead happens all the time in smop... it happens by hand yet ;) but it does 20:28
and it's actually the way smop is bootstrapping the type system
spinclad (sure, it has to be, to break the meta-loop) 20:29
(the MOP loop)
ruoso spinclad, have you seen the SMOP slides? 20:30
pmurias ruoso: :label or `label?
spinclad i read at least part way through them, sometime in the spring
ruoso pmurias, `label... :label is named arguments (although sm0p doesn't support it)
spinclad, please, consider taking another look, after this conversation it will probably have another meaning ;) 20:31
spinclad sure. i also got partway through the wiki pages but got bogged down by a slow browser... i'll read them again in links (much easier i expect) 20:33
pmurias spring? are the SMOP slides so old? 20:35
spinclad unless i misremember, which is not at all unlikely, i thought there was some slide set early on. please remind me of a best slide set's URL? 20:38
might as well read afresh or anew, whichever it be. 20:39
(misc/sm0p/README could use a pointer to the wiki at least)
pmurias ruoso: what was the slides url? ;)
ruoso pmurias, the wiki has a link 20:40
pmurias spinclad: misc/sm0p/ is not smop
ruoso spinclad, the sources are in v6/smop
pmurias, moving misc/sm0p to v6/smop/sm0p is probably a good idea
pmurias is trying to get label to work 20:41
spinclad thanks
(i've been in there before, and keep forgetting. forget me own head next.) 20:42
pmurias i can only one label per q:sm0p to parser, argh 20:56
even though i have <node_with_label>* 20:58
pugs_svn r21307 | ruoso++ | [smop] $p6opaque.^!methods() works. "push" and "shift" are not implemented yet, but the test metaclass already tries to call the method. three values leaking atm. 21:02
pmurias i'll attempt labels in the near future again (tommorow likely), in the meantime if anybody is intrested feel free 21:04
21:07 rdice_ left
ruoso home & 21:08
21:08 ruoso left 21:13 cggoebel left 21:16 donaldh joined
pugs_svn r21308 | pmurias++ | [smop] primitive label support 21:16
pmurias got something working somehow... 21:17
sleep&
21:17 pmurias left 21:29 donaldh left 21:41 Alias_ left 21:43 jan__ joined, hercynium left 21:50 Exodist left 21:53 iblechbot left 22:33 Jedai left 22:58 armagad joined 23:06 yewenbin joined 23:20 alester left 23:23 Alias_ joined 23:26 silug left 23:40 cmarcelo left 23:43 nothingmuch joined 23:47 Chillance left 23:54 jhorwitz left

Powered by Cro and the RakuĀ® Programming Language.

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

Thank you!