Parrot 6.3.0 "Black-cheeked Lovebird" | parrot.org/ | Log: irclog.perlgeek.de/parrot | #parrotsketch meeting Tuesday 19:30 UTC
Set by moderator on 15 April 2014.
00:13 rurban1 joined 01:14 rurban1 joined 01:39 rurban1 joined 03:43 cxreg joined 04:06 plobsing joined 04:31 Psyche^ joined 04:54 woosley joined 05:11 woosley joined 05:21 woosley joined 05:25 woosley joined 05:28 woosley joined 05:33 woosley joined 05:37 woosley joined 05:42 woosley joined 05:52 woosley joined 06:04 woosley joined 06:46 cooper joined 06:59 FROGGS joined 07:28 basiliscos joined 12:38 cogno joined 13:22 kshannon joined 13:42 bluescreen joined 14:25 benabik joined 14:26 rurban1 joined 15:07 FROGGS joined 16:42 Chirag joined 17:17 basiliscos joined 18:01 user_2906 joined 18:03 benabik joined
Chirag rurban: Hey! Thanks a lot for your help with my proposal :) .. My (final) exams are from 4th May to 15th May (and that would mark the end of the semester :D) 18:14
I prepared my github account today, forked parrot and created a remote to parrot (the original repo) .. 18:18
Since I still have a few days before my exams, I was wondering if I should start working on the first task - adding GC write barriers? 18:21
18:29 rurban1 joined
rurban If you feel well enough, why not. 18:35
you can add them manually to the ops or improve the compiler to add it. Best would be to add {} to each branch which does an early return. Only early returns need to add these barriers 18:36
git checkout rurban/pmc2c_orig 18:38
and then check the TODO GC write barrier warnings
Like TODO GC write barrier to default.pop_float
TODO GC write barrier to default.pop_integer
TODO GC write barrier to default.pop_pmc
TODO GC write barrier to default.pop_string
TODO GC write barrier to default.shift_float
TODO GC write barrier to default.shift_integer
TODO GC write barrier to default.shift_pmc
TODO GC write barrier to default.shift_string 18:39
e.g. look at capture.c then: 18:42
CAPTURE_array_CREATE(interp, _self, array);
return VTABLE_shift_float(interp, array);
/* need PARROT_GC_WRITE_BARRIER(interp, _self); before return */
so it needs to transform return VTABLE_shift_float(interp, array); to
PARROT_GC_WRITE_BARRIER(interp, _self);
return VTABLE_shift_float(interp, array); 18:43
nope. better use intermediates for the result.
so FLOATVAL tmp; 18:44
CAPTURE_array_CREATE(interp, _self, array);
tmp = VTABLE_shift_float(interp, array);
PARROT_GC_WRITE_BARRIER(interp, _self);
return tmp;
Chirag so all other barriers have to be written in the same way? 18:46
rurban so basically every PMC which returns a value and calling return VATBLE_method needs to add the write barriers for self
yes
Chirag after writing them, how do i test? 18:47
rurban rurban/pmc2c_orig already detects and marks all the cases correctly. it just doesn't do the transforms yet
make should print no TODO GC write barrier warnings
benabik is right person to ask how the write barriers internally work 18:48
I need to look it up also later
benabik What?
rurban hi. do we need write barriers in pmcs only for self, or also for intermediate return values?
like in Capture_shift_float 18:49
benabik The specific rule is any time you save a value to a new PMC in an old PMC.
The old PMC is the one that gets barriered.
rurban sounds good 18:50
benabik Lemme check Capture_shift_float. Been a while since I’ve used Parrot’s WB...
rurban so it's mostly self, but sometimes we have some other pmc args (pmc *old) 18:51
Capture_shift_float is just some simple example
CAPTURE_array_CREATE(INTERP, SELF, array);
return VTABLE_shift_float(INTERP, array);
Here we can add the write barrier between the CREATE and the VTABLE call 18:52
so we don't need an intermediate value for the result value
because the VTABLE call doesnt touch SELF 18:53
benabik In that case, I’ll point out that CAPTURE_array_CREATE already calls PARROT_GC_WRITE_BARRIER 18:54
rurban so these rules need to be codified in the pmc2c transformer
ah yes. hmm. maybe we should add all the missing barriers manually 18:55
benabik Yeah, WB are tricky to add automatically.
rurban and just mark it in pmc2c as already done
benabik The best you can really do is WB any PMC you use but don’t create yourself. And that’ll pessimize GC a lot.
Chirag a new PMC is when SELF is passed, right? 18:57
rurban no. SELF is the old pmc. a new is either the return value of PMC, or another PMC arg which is modified upon some values from SELF
benabik The simplest rule is “if you use a PMC you didn’t create, write barrier it” 18:58
rurban good
Chirag will remember that
:)
benabik You’ll probably add some you didn’t need, but that’s better than missing some you did.
rurban So e.g for capture.pmc add a comment like /* no PARROT_GC_WRITE_BARRIER, already in CAPTURE_array_CREATE */
and then check the compiler for this comment 18:59
I'll create a ticket for it
Chirag alright!
rurban github.com/parrot/parrot/issues/1069 19:15
can some gh admin add user chirag. I still have no admin rights there. 19:17
Chirag: what's your github username?
you need to fork parrot 19:18
your fork will then appear at github.com/parrot/parrot/network/members 19:19
and then you can add simple pull requests
note that we write SELF in the .pmc file, which gets transformed to _self in the .c file 19:21
same for INTERP
Chirag i use ZYROz as a nick 19:23
rurban updated
Chirag forked already 19:24
rurban name your branch ZYROz/pmc2_orig then
Chirag ok..
rurban or better ZYROz/pmc2_orig2-gh1069, it will be an improvement over rurban/pmc2_orig 19:25
Chirag sure
rurban this should be good enough work for a few days
Chirag yes :) 19:26
I ll keep looking up the docs as I go along..
Util #ps time 19:30
19:30 rurban1 joined 19:52 rurban1 joined
Chirag I cant get past this - make: *** [runtime/parrot/library/nqp-setting.pbc] Aborted (core dumped) 20:13
this happens during make install 20:14
rurban: make install fails for rurban/pmc2_orig .. make: *** [runtime/parrot/library/nqp-setting.pbc] Aborted (core dumped) 20:34
rurban don't do make install. work just from the git dir 21:04
because rurban/pmc2_orig is not ready yet. lots of wrong WB 21:05
see commit d2a039e48567a0d49945bf89b36cae6974273484 where I added the GC WB checks 21:07
basiliscos Good to know that Chirag is going to devep parrot with new forces. 22:06
I'm still stucked at documentation,
23:24 rurban1 joined 23:26 rurban2 joined 23:51 benabik 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!