Parrot 3.9.0 "Archaeopteryx" | parrot.org | Log: irclog.perlgeek.de/parrot | #parrotsketch meeting Tuesday 19:30 UTC
Set by moderator on 19 October 2011.
dalek nxed: 2ede966 | NotFound++ | winxedst1.winxed:
start unifying handling of functions in class and namespace statements
 00:00
Coke nom: my Array $array; 00:03
p6eval nom ef4702: ( no output )
Coke nom: my Array $array; $array[0] = "WHEE"; say $array.perl
p6eval nom ef4702: OUTPUT«Cannot look up attributes in a type object␤ in method gimme at src/gen/CORE.setting:4267␤ in method exists at src/gen/CORE.setting:4261␤ in method at_pos at src/gen/CORE.setting:4648␤ in method postcircumfix:<[ ]> at src/gen/CORE.setting:1006␤ in block <anon> at …
Coke rakudo: role A[::T?] {}; class B does A[] {}; say "alive" 00:05
p6eval rakudo ef4702: OUTPUT«===SORRY!===␤Unable to parse package_def, couldn't find final ']' at line 1␤»
Coke nom: say 'X' if 'bar' ~~ /:i FOO || BAR/; 00:06
p6eval nom ef4702: ( no output )
Coke nom: say 'X' if 'BAR' ~~ /:i FOO || BAR/; 00:07
p6eval nom ef4702: OUTPUT«X␤»
Coke nom: say 'X' if 'fo' ~~ /:i FOO || BAR/;
p6eval nom ef4702: ( no output )
Coke nom: say 'X' if 'foo' ~~ /:i FOO || BAR/;
p6eval nom ef4702: ( no output )
Coke say 'Z' if 'Ab' eq 'Ab' ~~ / A :i B || C / 00:08
nom: say 'Z' if 'Ab' eq 'Ab' ~~ / A :i B || C /
p6eval nom ef4702: ( no output )
Coke nom: say "" ~~ /m ** 1..-1 / 00:09
p6eval nom ef4702: OUTPUT«===SORRY!===␤Only integers or '*' allowed as range quantifier endpoint at line 1, near "-1 /"␤»
Coke rakudo: END { say "OH HAI" }; die "oh noes" 00:10
p6eval rakudo ef4702: OUTPUT«oh noes␤ in block <anon> at /tmp/YjTm6zAnAo:1␤ in <anon> at /tmp/YjTm6zAnAo:1␤»
Coke rakudo: sub wolf { say "om nom nom" }; sub sheep { say "baah!"; callsame }; &wolf.wrap(&sheep); wolf() 00:11
p6eval rakudo ef4702: OUTPUT«Method 'wrap' not found for invocant of class 'Sub'␤ in block <anon> at /tmp/LoR8npY1O1:1␤ in <anon> at /tmp/LoR8npY1O1:1␤»
Coke rakudo: sub foo(:$a, :$b, :$c) {}; foo(:a :b :c) 00:13
p6eval rakudo ef4702: OUTPUT«===SORRY!===␤Unable to parse postcircumfix:sym<( )>, couldn't find final ')' at line 1␤»
Coke oh, crap, I've been spamming the wrong channel. Whoops. 00:14
tadzik (:
00:15 whiteknight joined
sorear RFC: disabling rakudo:, nom:, b:, perl6:, pugs:, and niecza: on #parrot 00:17
Coke rakudo/nom should probably stay. 00:22
(maybe just rakudo, have it point to the right thign.)
NotFound Did we have winxed: ? 00:23
winxed: say("hi");
dalek nxed: 133fc87 | NotFound++ | winxedst1.winxed:
refactor a bit namespace parsing
 00:48
nxed: 795a6d3 | NotFound++ | winxedst1.winxed:
group tokenizer errors
 01:16
cotto ~~ 01:49
whiteknight sorear: I like seeing them, and the traffic isn't too high 01:54
unless it's a maintenance burden 01:55
dalek nxed: abff577 | NotFound++ | winxedst1.winxed:
use a formatter function in a bunch places to simplify the emit code
 03:03
07:24 fperrad joined 07:57 fperrad joined 10:09 bacek joined 10:46 mj41 joined 11:14 alvis joined 12:11 Psyche^ joined 12:50 ambs joined 13:41 whiteknight joined
dalek p/bigint: 506177f | moritz++ | src/ops/nqp_bigint.ops:
make nqp::isbig_I consistent with set_int and get_int

turns out that set_int and get_int only support 32bit, even on 64bit platforms. While the better fix would be to make them work on 64 bit width, this approach change nqp::isbig_i to return 1 if the stored int takes more than 32bit.		 13:43
whiteknight good morning, #parrot 13:53
nine good morning, whiteknight 13:54
whiteknight hello nine
dalek rrot/threads: eb01253 | nine++ | / (4 files):
Schedule updates of shared PMCs on the data owning thread.
 13:55
nine So the last piece missing is the read only proxy. 14:02
whiteknight wow, you're moving quick 14:03
For that Proxy PMC, I think we are going to want to use generated code, like the Object PMC uses
nine Mostly because it's been not much work at all :) Just reusing what was already there 14:04
whiteknight all the vtable definitions are available in src/vtable.tbl, and the ones that write data are marked with :write. We have several p5 utilities which read that file and generate code already
for the proxy, :write vtables should check if we are on the right thread and dispatch a message if not. All other vtables can transparently pass through 14:05
We can probably copy most of the code to do this from the same utilities that populate the Object PMC 14:06
or, if you want to avoid that mess, you can just copy the default.pmc and fill all the vtables in manually
14:18 mj41 joined
nine Re-reading your blog posts about this topic, since we don't want to guarantee any internal consistency for reading, what's the point of these proxies anyway? Just to prevent the user from doing the wrong thing (tm)? 14:21
14:43 jsut joined
whiteknight nine: the point of them is to prevent cross-thread writes 14:53
we never need to lock data if other threads can never write to it
nine So really just to prevent the user from doing what he shouldn't do. Ok. Just asked to see if I didn't get something. Anyway the proxy will be the perfect place to store a reference to the data owning interp. 14:55
whiteknight yes 14:56
basically, my thinking goes like this: Either we provide a whole library of locks and lock mechanisms, and tell the user that they are responsible for keeping things sane
...or, we use proxies to prevent data contention at the lowest level, and the user is always safe 14:57
in other words, parrot becomes a platform that handles the concurrency messy details, and as far as the user is concerned everything just works
nine But proxies still cannot guarantee read consistency. What will happen when a thread iterates over a shared array while the data owning thread updates it? 14:58
whiteknight yeah, that is a good question. We've thought of a few ways to deal with that, but haven't decided on anything yet 15:00
nine I guess it's about time to decide ;) 15:01
whiteknight yes it is :)
maybe we simply declare that read consistency isn't guaranteed 15:02
nine Then we would at least have to offer some sort of locking
Though accessing an Integer should be compeletely safe AFAICS and could be used for locking by the user. 15:04
In any case I'll have to expose disable_preemption and enable_preemption to allow atomic updates. With these operations allowed, the write subs become a quite powerful mechanism. 15:05
15:17 bluescreen joined
dalek rrot/threads: 53d28d8 | nine++ | src/pmc/task.pmc:
Mark the array of shared PMCs alive
 15:31
15:44 jlaire joined
whiteknight yeah, opcodes to enable/disable preemption will allow us to create the equivalent of critical sections 15:52
at least, at the task level
or, if not opcodes, methods on the scheduler object 15:54
scheduler.toggle_preemption(1), or scheduler.toggle_preemption(1, target_thread)
nine what's the 1? 15:55
whiteknight like a true/false 15:59
scheduler.toggle_preemption(0) turns it off
or better yet, $I0 = scheduler.preemption() 16:00
$I0 = scheduler.preemption(target_thread)
scheduler.preemption(1/0), scheduler.preemption(1/0, target_thread)
nine I find scheduler.enable_preemption and scheduler.disable_preemption more clear. What's the use case for the variant with target_thread? 16:01
whiteknight nine: internally, we would only really have scheduler.enable_preemption(target_thread), and we would have a short alias for scheduler.enable_preemption(scheduler.current_thread()) 16:02
basically, it is a tool to let you detect if a task is not being friendly, or if a thread is stuck in deadlock or something 16:03
or, a master thread controlling task priorities on worker threads, etc 16:07
nine I'm not sure if I like the idea of another thread being able to control a thread's preemption since the controller by definition cannot know what the controlled thread is currently doing. Also since a scheduler is turning preemption on/off incertain situations, this may leed to ugly interaction problems. 16:10
whiteknight okay, that's fine too 16:13
trading less control for more sanity is always fine
or, more accurately, we want as much control as we can possibly give without sacrificing sanity 16:14
nine And in any case we can always add it later 16:15
whiteknight because parrot is supposed to be a language-neutral VM, after all
so we want to expose enough functionality for languages to use
nine If I implement those as methods it would be pretty easy to call them from other threads anyway
whiteknight right 16:20
nine Maybe...I really should have created proxy.pmc manually 17:13
OTOH I still wouldn't know how to properly add it to the build system 17:14
whiteknight I *think* all you need to do is add it to MANIFEST and reconfigure 17:15
nine but the MANIFEST is auto generated
whiteknight no, MANIFEST is hand-generated 17:19
nine But it's read only and the comment says: # generated by tools/dev/mk_manifest_and_skip.pl 17:20
whiteknight it's read-only? that doesn't make any sense
un-read-only it, and just edit it. I don't think tools/dev/mk_manifest_and_skip.pl does anything. That comment is probably wrong 17:21
17:44 bluescreen_ joined
nine I have a script that generates code which is used by a script to generate code. This is so meta. 17:51
moritz
.oO( I never meta model I didn't like )
 17:55
that one is from chromatic, iirc 17:56
18:04 PacoLinux_ joined 18:11 ambs joined
dalek p/bigint: bee86c9 | moritz++ | / (6 files):
add platform dependent implementation of set_int and get_int; adjust nqp::isbig_I accordingly
 18:17
nopaste "japhb" at 192.168.1.3 pasted "'nqp bigint.pl' output on 32-bit system" (136 lines) at nopaste.snit.ch/94824 19:18
nine Why do I get a Null PMC access in invoke() when trying "ender_target = ender.'get_target'()". ender is certainly not Null 19:21
dalek nxed: 81a5796 | NotFound++ | winxedst1.winxed:
change handling of void context in expressions,

optimize a bit shortcut boolean ops by the way		 19:22
sorear nine: that line hides several instructions 19:23
nine: perhaps $Panon = find_method ender, "get_target" returned Null
nine: or, the invokation of get_target returned a null CallContext somehow(?)(not sure of the details of the unpack-results step, it may be impossible for this to NPMCA) 19:24
nine sorear: ah, ok. Since I just added the Proxy PMC class, I guess I missed something, so it doesn't find the method. 19:26
whiteknight yeah, that null pmc invoke error message is not very informative 19:28
it could mean the method is not found, or that ender is null
nopaste "japhb" at 192.168.1.3 pasted "'nqp bigint.pl' output on 32-bit system, now from i=28" (144 lines) at nopaste.snit.ch/94825 19:29
sorear nine: oh, someone is finally dealing with the PMCProxy cesspool?
nine '$P0 = find_method ender, "get_target"' gives me a Null PMC access in defined()
whiteknight sorear: no, a new proxy type
PMCProxy is still a cesspool 19:30
nine sorear: no, I've no idea what PMCProxy even is. I'm adding a new Proxy PMC class for threading
Ok what would I have to do to add some method to a PMC class that previously had none? 19:31
NotFound nine: a PMCProxy if sort of the class of a non-Object PMC 19:32
nine: add? You mean at runtime? 19:35
nine NotFound: no, in the source 19:36
NotFound nine: In a .pmc file?
nine NotFound: yes
NotFound: added this: paste.scsys.co.uk/159745 19:37
NotFound nine: look for METHOD in src/pmc/*.pmc
nine: looks good. Make sure it is inside the pmclass block. 19:38
pmc2c is not good at diagnosing errors, usually just skip whatever he doesn't understand. 19:39
nine It definitely is inside the block. paste.scsys.co.uk/159746 is the whole source 19:40
and the method gets translated to Parrot_Proxy_nci_get_interp by pmc2c 19:41
NotFound Delete the generated C to be sure it gets regenerated. The dependencies can fail.
And its .o
nine no change :8 19:42
:(
NotFound Strange... 19:43
Do you have that on a branch?
Now that I think about it... If that PMC is a proxy for other PMC... Isn't find_method looking at the proxy target? 19:48
nine oh, that could very well be
Should have thought of that. The reason for creating accessor methods was get_attribute getting proxied in the first place 19:49
19:59 snearch joined
nine Now I just have to fight the GC again... 20:12
whiteknight what's the fight with the GC? 20:13
nine Ah no it's a logic error on my part. I create the proxies on task.pop() and intitialize proxy->interp with INTERP which is the running thread and not the owning thread 20:14
whiteknight ah, okay 20:18
20:34 mj41 joined
nine I think, I'll give up for today... 20:40
Good night, #parrot 20:49
tadzik g'night 20:50
21:17 contingencyplan joined
dalek nxed: d7389d2 | NotFound++ | winxedst1.winxed:
store class constants in items instead of a separated container
 21:47
21:54 rfw joined 22:29 rfw joined 23:26 aloha joined 23:36 bacek_at_work joined
Coke whiteknight, nine; tools/dev/mk_manifest_and_skip.pl does actually do something. 23:37
23:48 autark_ joined

Powered by Cro and the Raku® Programming Language.

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

Thank you!