| IRC logs at
Set by AlexDaniel on 12 June 2018.
00:02 sena_kun joined 00:04 Altai-man_ left 00:41 vrurg left 00:48 vrurg joined 00:52 vrurg left 01:11 vrurg joined 01:29 MasterDuke left 01:55 lucasb left 02:01 Altai-man_ joined 02:02 vrurg left 02:03 sena_kun left 02:18 vrurg joined 02:38 AlexDaniel left 02:40 AlexDaniel joined, AlexDaniel left, AlexDaniel joined 04:02 sena_kun joined 04:03 Altai-man_ left 06:01 Altai-man_ joined 06:03 sena_kun left
nwc10 good *, #moarvm 06:19
07:45 zakharyas joined 08:02 sena_kun joined 08:04 Altai-man_ left 08:13 MasterDuke joined 08:25 MasterDuke left 09:01 MasterDuke joined
jnthn morning o/ 09:29
tellable6 2020-06-28T10:45:40Z #raku-dev <Kaiepi> jnthn, is ok on the condition that nqp::listen always be called directly after nqp::bind_sk when binding streaming sockets?
nwc10 \o
lizmat /o 09:30
sena_kun o/ 09:31
MasterDuke morning. i would ask if now is a good time to merge, but it seems github is down 09:32
10:01 Altai-man_ joined
jnthn Yes, I'm now onto my third task of the day that I'm attemting to do, but finding I can't quite do 'cus github is down :/ 10:02
10:04 sena_kun left
nwc10 So ahead of its time :-) -- 10:07
(not a Rickroll.)
MasterDuke ah, back now 10:08
10:44 gmoshkin joined 10:47 gmoshkin left 11:46 zakharyas left 12:02 sena_kun joined 12:03 Altai-man_ left
jnthn Hm, this is odd. Somehow we end up in a situation where there's something on the stack top (a frame) that doesn't match up with tc->cur_frame (it's one frame deeper than that), and I can't figure out how it gets left behind there 12:40
It's got bogus content, and we end up tripping over it during GC
moritz sounds like something that deopt could do? (just spitballing, have no real knowledge about it) 12:41
does it happen with spesh disabled? 12:42
jnthn Not that I can see; I did already consider deopt as a possible cause
Oh...the thread that is being marked is not the one that triggers GC
timotimo whew 12:43
jnthn ahh...I see it 12:44
12:46 zakharyas joined
timotimo a bad mark_thread_blocked? 12:51
Geth MoarVM/new-disp: 008d4afae0 | (Jonathan Worthington)++ | src/core/frame.c
Reorder frame invoke logic for safety

The call stack entry needs to be sufficiently set up before we try to GC mark it. Sending spesh log entries can cause us to enter GC. Make sure we have set code_ref and outer before that can ever happen, so that we don't mark leftover junk pointers.
jnthn hah, and get got rid of one more `make test` Rakudo failure too, it seems 12:56
I think all the current ones are known NYI of things
13:01 dogbert17 left
jnthn Guess I'd better do some kind of wiring up of multi dispatch to the new dispatch mechanism, since there's so many failures relating to that, that I can't actually see if there's any other interesting ones. :) 13:04
Geth MoarVM/new-disp: 3be0031c4b | (Jonathan Worthington)++ | 3 files
Make captureposelems work on MVMCapture
13:46 AlexDaniel` left, dogbert17 joined 13:50 Geth left, Geth joined 13:53 Kaiepi left 14:01 Altai-man_ joined 14:04 sena_kun left 14:07 AlexDaniel` joined
timotimo - here's my patch for logging the resolution of a dispatch program based on the correlation id that was there at the beginning 14:20
ooh, wasn't there a thing where i was supposed to implement the correlation id lookup based on what's on the stack, i.e. if multiple run frames are on top of it? 14:21
14:26 MasterDuke left
jnthn timotimo: Will have a look at it when I'm not trying to re-work multi dispatch... :) 15:02
timotimo aye 15:03
jnthn (Which is quite tricky, given I'm giving the algorithm a bit of an overhaul too, so we can, for example, avoid some of the repeated nominal type checking we do today when there's candidates that need bind checks)
nwc10 but surely you can mutli-task when you're doing mulit-dispatch?
jnthn Yes, but only if the other task is undemanding, like "drink tea" 15:04
nwc10 and by some coincidence, that's also what I'm doing
lizmat as am I :-) 15:10
darn, out of it again :-)
Geth MoarVM/new-disp: 08c34ce157 | (Jonathan Worthington)++ | 3 files
Implement captureposprimspec for MVMCapture
15:43 rypervenche left 15:46 rypervenche joined 16:02 sena_kun joined 16:03 Altai-man_ left
jnthn Hm, turns out a fairly incomplete multi dispatch impl gets a lot of the tests back 16:17
(To be clear, I'm not actually using it for all multi-dispatch yet, only those cases where a coercion on a return type bottoms out in doing a multi dispatch) 16:18
16:59 Kaiepi joined
jnthn Uff, probably at the point where I should rebase all the new-disp things onto Rakudo HEAD 'cus I suspect many of my remaining spectest failures are due to missing HEAD fixes 17:15
Well, onto everything HEAD really 17:17
17:23 japhb left, japhb joined
jnthn Wow, after rebasing, only 8 spectests showing issues on new-disp. (Noting that we're only using it in place of spesh plugins so far, for the most part) 17:34
And it seems 5 of them are for the same bit of NYI multi dispatch stuff 17:35
Geth MoarVM/new-disp: 141 commits pushed by (Jonathan Worthington)++, (Timo Paulssen)++
jnthn All rebased, so will need some twiddling for anyone testing it 17:38
(There were assorted conflicts on MoarVM, and I find those easier to deal with - not to mention getting cleaner history - with a rebase workflow.)
nine Are you planning on merging it as soon as it can fully replace spesh plugins or only when it replaces all dispatch? 17:39
jnthn Undecided. 17:41
There's a horrible hack or two that would probably need to be turned into something more robust if we want to merge it sooner
While if we wait until it replaces all dispatch then the hacks in question become dead code and so I just delete them :) 17:42
Home time o/ 17:46
17:48 zakharyas left 18:01 Altai-man_ joined 18:03 sena_kun left 18:07 MasterDuke joined
Geth MoarVM: 193748846b | (Daniel Green)++ | src/jit/core_templates.expr
Add JIT templates for return_(i|n|s)

Based off of the pre-existing one for return_o.
MoarVM: 5832b26289 | MasterDuke17++ (committed using GitHub Web editor) | src/jit/core_templates.expr
Merge pull request #1320 from MasterDuke17/jit_templates_for_return_ins

Add JIT templates for return_(i|n|s)
18:50 dogbert17 left 18:53 dogbert17 joined 18:57 zakharyas joined
timotimo trying to run update_ops has callstack.c's mark stumble over a MVM_CALLSTACK_RECORD_FLATTENING 19:03
so i'll implement that
nwc10 jnthn: ASAN excited by new-disp -- 19:06
there were other failures in the test summary, but when I picked a different one and re-ran it, it seemed fine
so not sure how many bugs there are 19:07
but nonzero, which makes ASAN's day
it's a use-after free. The free is in remove_one_frame
might need FSA_DEBUG to expose it 19:08
timotimo thanks to a MVM_callsite_mark, it was trivial 19:09
Geth MoarVM/new-disp: 6885813817 | (Timo Paulssen)++ | src/core/callstack.c

using MVM_callsite_mark on the &record.produced_cs.
20:02 sena_kun joined 20:03 Altai-man_ left
jnthn timotimo: That's incomplete; the MVMRegister * or args also must be marked per the flags in the callsite 20:18
*of args 20:19
nwc10: ooh, interesting...
nwc10 I had been just about to go to bed. But if you think you want any more info, I can delay for a bit
timotimo ah, so i do have to iterate through the flags, then 20:21
20:31 zakharyas left
timotimo can i just keep one index for the callsite entries and one for the register "addresses" and when i see one with MVM_CALLSITE_ARG_NAMED (perhaps also NAMED_AND_FLATTENED) i skip the entry but not the register? 20:33
jnthn nwc10: No, was doing homework. Will probably look on Wednesday 21:07
timotimo: No, the data in a flattened record is *already* flattened
timotimo: You don't have to care about nameds even, just just check if it's an object or string arg in the callsite 21:08
(Because the whole "named is two args" thing is totally gone in the new calling conventions
Now the names live only in the callsite object
(The point of a flattening record is that we can allocate the space for the result of the flattening right there on the callstack) 21:09
lizmat and yet another Rakudo Weekly News hits the Net: 21:21
timotimo whee 21:43
[Coke] lizmat++ 21:45
22:01 Altai-man_ joined 22:04 sena_kun left 22:43 Kaiepi left 22:44 Kaiepi joined 23:06 AlexDaniel left, AlexDani` joined 23:20 vrurg left