»ö« Welcome to Perl 6! | perl6.org/ | evalbot usage: 'perl6: say 3;' or rakudo:, niecza:, std:, or /msg p6eval perl6: ... | irclog: irc.perl6.org/ | UTF-8 is our friend! | Rakudo Star Released!
Set by diakopter on 6 September 2010.
sorear I hesitate to use the word "bad", it's very judgemental 00:00
tadzik the PIR (Parrot intermediate representation, quite a low-level language) is barely faster than Perl 5 now, and not in all the cases I think 00:01
sorear parrot's scope is incredibly large for <20 active contributors
javuchi i was planing to implement a language on it 00:02
but if you say it is so bad
i have my doubts
sorear javuchi: are you doing high-performance computing work?
flussence PIR as fast as p5? That sounds quite encouraging actually 00:03
javuchi i'm looking for a good comprimise
sorear Parrot never had speed as its only goal. if speed is *your* *only* goal, you want a VM that was designed for speed
javuchi can you point me (even if it is off-topic) a place for a good comparison between different open source virtual machines? 00:04
just fo the matter to choose for my project
tadzik flussence: well, I have my stupid benchmark I run from time to time
sjohnson Trashlord: that makes two of us (soon)
sorear there are, AFAIK, no good general VMs for high-performance work
you'll need to put together your own... LLVM is a good starting point
sorear javuchi: the great computer language shootout might be a good start 00:05
flussence bah, sorear beat me to it
link: shootout.alioth.debian.org/
sorear javuchi: if you have other goals besides speed, now would be a good time to mention them
javuchi i'm looking for something generalist, like parrot is 00:06
easy assembler
small
well mainteined
that kind of things
sorear Parrot's #1 goal is to make developing languages easy 00:08
javuchi that's why i like parrot 00:09
the problem is speed
when is it getting better?
tadzik every day. Slowly, yes. But the GC is being worked on, and there are plans for the new opcode set, leading to better JIT... one day 00:10
javuchi well... it is funny, because it started in 2003 if i'm not wrong
should i wait until 2020 for a reasonable speed? 00:11
dukeleto javuchi: would you like to help make parrot faster? 00:13
tadzik I'm sure #parrot folks will give you more appropriate answers. However, I see Parrot getting faster every release, sometimes twice as fast, sometimes 30% faster. And the big changes are still before'em 00:14
javuchi sure, why not dukeleto 00:15
sorear if you read about the history of the parrot project you'll see that it's studded with infighting and politics 00:20
that's basically behind us now, but it still means that you can't count 10 years as progress 00:21
dalek odel: 3629a41 | jnthn++ | dotnet/ (3 files):
[dotnet] Various tweaks to get handler setup working enough for us to set up return exception handlers.
 00:24
odel: b11ad22 | jnthn++ | t/nqp/ (2 files):
We now pass 20-return.t and 48-closure.t.
Limbic_Region actually, I am surprised I can't find a better description of the road map than trac.parrot.org/parrot/roadmap 00:25
Limbic_Region is going to wander over to #parrot and ask there
tadzik goodnight o/ 00:44
jnthn -> sleep 01:34
eternaleye rakudo: .say for (Int ~~ Nil), (1 ~~ Nil), (Mu ~~ Nil) 04:40
p6eval rakudo 479650: OUTPUT«0␤0␤0␤»
eternaleye Oh good, it matches the spec' now 04:41
Or at least, what was stated to be correct last time I tried that and got 1\n0\n1 04:42
sorear I wonder if there's any way to implement lists in Perl 6 that's as efficient as the Perl 5 way. 05:01
sorear TimToady: Who do I get to credit with the devilish genius of the MARK stack? 05:06
sorear current niecza, before lists refactor: 56 us/cycle on perf/itmark.pl 05:28
600us rakudo; 158ns perl5 :/ 05:30
still 400 times slower 05:31
117ns for Parrot. I'm amazed. 05:35
dalek ecza/master: eb2e899 | sorear++ | perf/itmark.p (2 files):
Add Perl 5 and PIR versions of itmark
 05:40
tadzik hello 07:59
x3nU omg difference in speed beetwen rakudo 2010.10 and older is amazing 10:20
it's because that .flip change probably
my euler problem 36 script
now completes in much less than half hour
on older versions it took half day 10:21
frettled nice! 10:32
jnthn oh hai, #perl6 10:56
colomon \o 11:20
tadzik o/ 11:21
jnthn o/ colomon, tadzik 11:22
dnl moinsen 12:20
dalek odel: e455223 | jnthn++ | dotnet/runtime/Metamodel/Representations/RakudoCodeRef.cs:
[dotnet] Comment improvements, and give our code object representation a slot for any dispatch candidates it will operate over, in the case that it's a dispatch routine.
 14:10
dalek odel: 2ad10c2 | jnthn++ | / (6 files):
[dotnet] This does the core of the multi-dispatch switch. Rip out the old separate candidate list storage. Store multi candidates in the dispatch routine's dispatchees slot (instantiations for nested scopes NYI, though). Use nearest outer dispatch list when entering the multi dispatcher, tossing all the slow and failure-prone named lookup stuff. Update NQPSetting to not pass the name along to the dispatcher entry call. This passes all tests we did bef
 14:58
dalek kudo: 5f5bae4 | Util++ | src/core/Complex.pm:
[core/Complex.pm] Fixed indenting
 15:43
dalek odel: 0d78282 | jnthn++ | dotnet/ (2 files):
[dotnet] Add experimental support for :scope('outer') to PAST::Var (curious to see how pmichaud++ reacts), which does a lexical lookup but skipping the current scope. Currently, bind not supported, and of course :isdecl is invalid.
 15:58
odel: aa4bd85 | jnthn++ | dotnet/ (2 files):
[dotnet] Instantiate protos in inner routines to create dispatch routines with the candidate list for that block. We actually do this by copying the most immediately outer instantiation's candidate list and then adding the new ones to the list, which seems likely-ish to get closure semantics right while giving the semantics as per spec. It's very much an instantiation in that it shares body and other bits with the original proto, though.
jnthn TimToady++ TimToady++ # updated multi spec is SO much saner to implement
dalek odel: b3e555f | jnthn++ | dotnet/runtime/Runtime/ (2 files):
[dotnet] Tiny refactor to enable a start to be made integrating the multi-dispatch cache.
 16:30
odel: 92997a3 | jnthn++ | dotnet/runtime/ (2 files):
[dotnet] Integrate that multi-dispatch cache I designed on the train a while back. Seems to help, or so says the (RedGate profiler)++.
araujo making some changes to get 'infix' notation 16:39
dukeleto If anyone would like to be involved with Google Code-In, you are cordially invited to #gci on irc.perl.org 18:54
mberends joins 19:00
rindolf Hello everyone. 19:12
dalek odel: cac2fa6 | jnthn++ | dotnet/compiler/ (2 files):
[dotnet] Update JnthnNQP to parse a * as the sole body of a proto and turn it into a dispatcher call. Do it a bit differently from STD, since we don't parse * at all otherwise.
 20:11
odel: 413bae8 | jnthn++ | common/NQP/NQPSetting.pm:
[common] Replace explicit multi-dispatcher ops with a proto with a plain body of *.
jnthn TimToady: You ever considered what happens in a proto if instead of actually doing the multi invocation, somebody returns a closrue that will go and do it later on? Seems that it just works out naturally though... gist.github.com/655709 20:26
dalek odel: aa95937 | jnthn++ | dotnet/compiler/ (2 files):
[dotnet] Implement {*} parsing in JnthnNQP so that we can enter the multi-dispatcher at any chose point in the proto.
 20:27
odel: df06e98 | jnthn++ | dotnet/runtime/Runtime/Ops.cs:
[dotnet] Fix a silly thinko - we want the capture passed to the proto, always.
TimToady jnthn: that's fine, as long as we still end up with a new &foo in each new multi scope 21:05
jnthn TimToady: Oh, we certainly do 21:06
TimToady: It's a seperate object instance. :-)
jnthn With a separate candidate list stored per instance. 21:09
jnthn rakudo: my @x = 1,2,3; my $i = @x.iterator; say $i.WHAT 21:56
p6eval rakudo 5f5bae: ( no output )
jnthn rakudo: say "alive"
p6eval rakudo 5f5bae: OUTPUT«alive␤» 21:57
jnthn rakudo: my @x = 1,2,3; my $i = @x.iterator; say "alive"
p6eval rakudo 5f5bae: OUTPUT«alive␤»
jnthn rakudo: my @x = 1,2,3; my $i = @x.iterator; $i.^methods(:local)>>.name>>.say
p6eval rakudo 5f5bae: OUTPUT«new␤reify␤»
jnthn rakudo: my @x = 1,2,3; my $i = @x.iterator; say $i.get; say $i.get;
p6eval rakudo 5f5bae: OUTPUT«Method 'get' not found for invocant of class 'ListIter'␤ in main program body at line 22:/tmp/x1tyf9USTf␤»
sorear jnthn: Would you be comfortable with declaring Iterator to be an implementation detail and requiring portable code to use gather/take, series ops, for loops, etc?
jnthn sorear: I'd guess "no" 21:58
sorear: But pmichaud is the guy who understands this stuff. :-)
jnthn tries to work out what to call on an iterator to get a value from it.
rakudo: my @x = 1,2,3; my $i = @x.iterator; say $i.munch; 21:59
sorear You can't
p6eval rakudo 5f5bae: OUTPUT«Method 'munch' not found for invocant of class 'ListIter'␤ in main program body at line 22:/tmp/Grqu8UBviF␤»
sorear That's not how Rakudo iterators work
I do -not- want to be constrained to the Rakudo iterator model
dalek odel: d6d3205 | jnthn++ | dotnet/LHF.txt:
[dotnet] Add a Low Hanging Fruit file to suggest jumping in points for anyone interested in doing so.
sorear if $i is a RakudoIterator, $i.reify returns a Parcel 22:00
each element of $i.reify is either a value or a fresh RakudoIterator
look at !fill in src/builtins/List.pir for how to extract a value from a RakudoIterator - it requires a loop and the splice op 22:01
jnthn Ah, yes, it's an immutable iterator model. 22:02
sorear: I'm mostly looking 'cus I'm wanting to do for loops in NQP.Net. 22:05
NQP on Parrot just uses the Parrot idea of iterators.
sorear I spent much time last week redesigning for loops for Niecza 22:06
lue ohai o/ 22:08
jnthn sorear: How did that end up looking? 22:36
sorear: Does it call map?
I guess the devil is in how your .map looks. :-) 22:37
*devil in the detail 22:38
sorear Well, I haven't had tuits to finish coding the new design
jnthn Rakudo went through a LOT of iterations of list design before getting to what we have today, fwiw. This has been the least problematic design so far. 22:39
sorear I'm using a lot of elements of the design 22:41
I could probably even make it compatible, but only by using horribly wrong names for stuff
jnthn Eh well, I'll try and figure out for in NQP.NET tomorrow or so. :-) 22:45
Quite happy to have got the multi dispatch stuff in today. :-)
Util .new() vs .^new() 23:25
When creating a new object of class Foo from within some other method of that class, 23:26
which is the more correct way to call new() without hardcoding the classname?

Powered by Cro and the Raku® Programming Language.

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

Thank you!