Perl 6 - the future is here, just unevenly distributed

IRC log for #moarvm, 2017-03-13

| Channels | #moarvm index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
00:49 MasterDuke and now i've gone even deeper and it looks like i need to create a src/6model/reprs/P6biguint.c
01:15 MasterDuke .ask jnthn if you look at my last couple comments/questions, could you tell me how far off the rails i've gone? thanks
01:15 yoleaux2 MasterDuke: I'll pass your message to jnthn.
06:54 domidumont joined #moarvm
07:14 samcv MasterDuke, what are you referring to?
08:41 nwc10 jnthn: good *. The most exciting thing that happened all weekend (because it was the *only* think that happened all weekend): http://paste.scsys.co.uk/557459
08:42 nwc10 why do my fingers typo it as "think" ?
08:48 brrt joined #moarvm
08:48 brrt ohai #moarvm
09:58 jnthn I do that typo too...
09:58 yoleaux2 01:15Z <MasterDuke> jnthn: if you look at my last couple comments/questions, could you tell me how far off the rails i've gone? thanks
09:58 jnthn nwc10: How often does that happen?
09:58 jnthn I hope not "once in the whole weekend" :P
10:09 * dogbert17_ is sad, the number of SEGV's are at an all time low :)
10:16 dogbert17_ o/ btw
10:23 nwc10 once. in 3 days
10:24 jnthn Ouch.
10:34 dogbert17_ could it be the spesh bug?
10:38 jnthn I suspect so but...this is the only thing we've got that produces it, and it was already very rare
10:38 dogbert17_ how would one go about finding a bug like that?
10:39 dogbert17_ given that a golf exists I mean
10:50 dogbert17_ next qestion, are there any ENV vars or the like available with which you can perhaps make the probability of a bug showing itself increase?
11:21 jnthn MVM_SPESH_NODELAY=1 may make it more likely
11:22 jnthn Though under certain circumstances also risks hiding it
12:02 dogbert17_ thx, will try it out
12:13 * dogbert17_ notes that 10 spectest files have failures when run with MVM_SPESH_NODELAY=1
12:15 brrt dogbert17_: interesting. what happens if you don't run MVM_SPESH_NODELAY, but you 'hotten' them?
12:16 dogbert17_ what is 'hotten' in this context, run a bunch of times?
12:26 brrt yes
12:27 brrt well, within the same process
12:27 brrt so that the thing will be speshed+jitted
12:27 dogbert17_ running e.g. './perl6 t/spec/S06-currying/misc.t' when MVM_SPESH_NODELAY=1 leads to the following
12:27 dogbert17_ ok 9 - Priming :(::T  $a, Array[Positional[T]] $b, Array[Positional[Int]] :$c) with \(1) gave :($b, Array[Positional[Int]] :$c)
12:27 dogbert17_ Type check failed in binding to '$b'; expected Any but got T (?)
12:27 dogbert17_ in any  at ./blib/Perl6/BOOTSTRAP.moarvm line 1
12:28 brrt MVM_SPESH_NODELAY can expose bugs in spesh, but not necessarily the bug you're interested in or even one that could really happen
12:29 dogbert17_ ignore what I wrote above, had forgot to fudge :(
12:30 dogbert17_ brrt, i'll try your suggestion if I'm unable to find anything directly :)
12:31 dogbert17_ I liked jnthn's comment 'Though under certain circumstances also risks hiding it'
12:38 dogbert17_ brrt, jnthn: here are the failing tests, perhaps a spesh bug might be hiding in there. https://gist.github.com/dogbert17/bbbeecfeb348f5f77c2950065758b16c
13:14 dogbert17_ oh, 'ok 19 - &anonslurp.assuming((1, 2, "c", Bool::True).Seq)((Bool::True,).Seq) returned expected value ["a1 b2"] \n Segmentation fault'
13:47 brrt interesting
14:10 domidumont joined #moarvm
14:11 dogbert17_ brrt: what do you make of this gdb outpout? https://gist.github.com/dogbert17/e0cd2b873daf267a747c9301b4982707
14:13 brrt flag is unitialized, or flag_map is
14:13 brrt :-)
14:13 brrt interesting
14:15 brrt dogbert17_: can you print out the value of flag_map and flag
14:17 brrt also
14:17 brrt that looks like it's sensitive to env
14:18 brrt i wonder how difficult it'd be to have 'shareable' debug sessions
14:18 brrt either frozen (core dumps) or over the network
14:21 brrt https://developers.googleblog.com/2017/03/xla-tensorflow-compiled.html this is some impressive stuff, btw
14:26 domidumont joined #moarvm
15:33 dogbert17_ brrt, sry was forced to do $work thingies :)
15:34 brrt well, that's important, innit
15:34 dogbert17_ (gdb) info locals   # note this is from another run, had closed the previous session
15:34 dogbert17_ flag = 0
15:34 dogbert17_ flag_map = 0x3 <error: Cannot access memory at address 0x3>
15:36 brrt seems pretty likely that flag_map ain't initialized properly
15:38 dogbert17_ there seems to be something 'speshial' going on, unsetting the env var makes the test file pass
15:38 brrt well, that's not very special
15:38 brrt the env var causes an extra allocation
15:38 dogbert17_ aha
15:38 brrt so something is being GC'ed without being fully initialized, it would seem
15:39 brrt that's pretty weird, i think
15:41 dogbert17_ are you in a position to try it? Could perhaps be something fishy with my setup
15:44 brrt not really, not now
15:45 dogbert17_ am beginning to suspect that there's something on my end causing this ...
16:28 * dogbert17_ but then again maybe not ...
17:35 dogbert17 jnthn: here's some ASAN output of what 'might' be a spesh bug: https://gist.github.com/dogbert17/8e96d144505ae8c0434242a852f607d0
17:54 Ven joined #moarvm
18:13 Ven joined #moarvm
18:22 jnthn eww, that doesn't look to nice...
18:23 jnthn dogbert17: Normal nursery?
18:23 dogbert17 jnthn, yes
18:25 timotimo i've always had as neaking suspicion that callsite lifecycle management would end up problematic at some point
18:34 Ven joined #moarvm
18:50 jnthn Aye, though I simplified it a bunch a while back
18:51 jnthn So not quite sure how it's going awry this time
18:53 dogbert17 on the plus side it should be easy to reproduce
18:53 Ven joined #moarvm
19:00 timotimo i remember hoelzro struggled with callsites a while back. or maybe that was something else
19:07 hoelzro timotimo: I think so: https://hoelz.ro/blog/hunted-by-a-leak
19:13 Ven joined #moarvm
19:17 [Coke] hoelzro: I cannot tell when that published. :(
19:17 timotimo oh, indeed
19:17 timotimo that's a pet-peeve of mine
19:18 timotimo when you can't tell what the date on a blog post/article/whatever is
19:19 * [Coke] reads all the entries, though, nifty
19:19 [Coke] hoelzro++ fixing things and taking names.
19:24 agentzh joined #moarvm
19:33 Ven joined #moarvm
19:37 hoelzro oh - maybe I should add that, then?
19:37 hoelzro I read a post once that said you *shouldn't* do that
19:38 timotimo huh?
19:38 timotimo tell me where the author lives, i'll fight them irl!
19:38 hoelzro timotimo: https://training.kalzumeus.com/newsletters/archive/content-marketing-strategy
19:39 [Coke] hoelzro: My best counter example is anything involving rakudo on parrot. You wouldn't want someone to think they were reading the state of the art.
19:39 hoelzro that's a really good point
19:39 timotimo "content marketing strategy"
19:39 timotimo >_>
19:39 hoelzro and considering how fast Moar moves...
19:39 hoelzro timotimo: it was linked from a post about writing a great programming blog
19:39 timotimo if your goal is to make people think your stuff is still relevant by just not telling them when it was written ...
19:40 hoelzro ...which mine is not, so...
19:40 hoelzro maybe I
19:40 hoelzro maybe I'll just add those date.
19:40 hoelzro *dates
19:40 hoelzro jeez
19:40 hoelzro can't type today!
19:41 nwc10 it's all the fault of the DST switch?
19:42 nwc10 (I forget exactly where you are, but I was thinking the US, but also know that not all parts of the US do the DST dance)
19:42 [Coke] dumb dumb DST. :|
19:42 nwc10 UGT to UGT+1
19:42 nwc10 I'm not sure how/if that works
19:42 * [Coke] wonders how amenable NYS would be to a "kill DST" bill.
19:43 nwc10 it's somewhat like ∞ + 1
19:44 hoelzro I did indeed do the DST dance =(
19:47 hoelzro ok, publish dates added
19:47 hoelzro ...at the bottom - adding them in a nice place on the top will likely take a little more work
19:53 Ven joined #moarvm
19:56 * [Coke] thinks that's fine.
19:59 lizmat datapoint: I haven't been able to crash HARNESS_TYPE=6 make spectest after 30 runs
20:01 nwc10 I haven't been able to crash it for 48 hours now
20:01 nwc10 it's looking good
20:07 timotimo <3
20:13 Ven joined #moarvm
20:19 timotimo oooh, the gdb-moar.py can benefit seriously from the debug name featuer
20:19 timotimo feature*
20:33 Ven joined #moarvm
20:53 Ven joined #moarvm
21:05 Ven joined #moarvm
21:57 zakharyas joined #moarvm
22:05 samcv good *
22:05 lizmat samcv o/
22:07 samcv i'm feeling kind of like writing a case insensitive index op today
22:10 lizmat samcv: go for it!  :-)

| Channels | #moarvm index | Today | | Search | Google Search | Plain-Text | summary