00:46 Ven_ joined 01:50 Ven_ joined 02:55 Ven_ joined
JimmyZ psch: at a first quick look: can_store(tc, MVM_args_get_pos_obj(tc, apc,i ,1).arg.o))) can be can_store(tc, arg)) 03:04
psch: and can_store(tc, args[i].o) -> can_store(tc, arg) 03:05
psch: oh, really should be (contspec && !MVM_is_null(tc, arg) && contspec->can_store(tc, arg))) 03:11
psch: hope it fixs the segfault 03:12
psch: better: (!MVM_is_null(tc, arg) && contspec && contspec->can_store(tc, arg))) 03:13
03:59 Ven_ joined 04:34 tokuhiro_ joined 04:40 stmuk joined 04:51 vendethiel joined 05:02 Ven_ joined 06:50 domidumont joined 07:09 Ven_ joined 07:27 lizmat joined 08:06 zakharyas joined 08:57 brrt joined
brrt \o #moarvm 09:12
nwc10 o/ brrt
timotimo mvraom# o/ trrb
brrt \o nwc10, timotimo 09:13
i've a designish question again
currently i have (in the expression jit) a node called LABEL, which takes a single arguments, which must be a CONST node, and which either loads the address (using lea) or may be used to jump 09:14
i.e. (branch (label (const 4 4))) turns into jmp =>(4)
this =>(4) is the notation of a dynamic label in dynasm
anyway, it occured to me that the const makes no sense 09:15
there is no way to do anything but a const there
timotimo we can't calculate jump targets?
brrt e.g. the same should rather be (branch (label 4))
oh, yes we can
but then you should have the form (branch reg)
or even
(branch (load (addr reg $ofs) $size)) 09:16
timotimo right
brrt ok, so with that said 09:18
i need to add markers
i.e. not only to *refer* to labels, but also to *insert* them explicitly
my initial guess was (MARK (LABEL 4)) for the expression =>(4):
however, this ... runs against the same complaints as (LABEL (CONST 4)) did 09:19
timotimo as in: "when jumping to label 4, this is where you'll end up"?
brrt yes
it makes sense verbally, but not in the sense of 'compute label 4, then mark it'
if i say (MARK 4) then, why is it called 09:20
mark
so i'm almost going to say, why not (MARK_LABEL 4)
timotimo my bikeshed sense is tingling :P
brrt design is like that :-P
but in seriousness, i want to make the tree make sense to people 09:21
(maybe i should output graphviz for the jit tree) 09:22
that would be a fun miniproject
timotimo sure. i like outputting graphviz :)
brrt timotimo.. do you have time left? (oh, how do i dare even)
basically, in even-moar-jit, in src/jit/log.c, we have a tree-walking log.c 09:24
but the tree isn't a tree really, so it gives a quite confusing sotry
story
timotimo right, that's the thing i'm worried about 09:26
brrt which is why i think graphviz dot graphs may help :-) 09:27
timotimo i don't understand enough about the tree - especially with the way things are hooked in-place
and such
er
brrt oh, i can help with that....
timotimo i've been meaning to look through the moarvm utils i've built some time ago to make sure they still work properly
brrt letmesee how dot graphs really work 09:28
timotimo super easy
digraph Foo { a -> b; c -> d; a -> c; }
brrt ok, what i'm thinking is doing postorder traversal 09:29
or maybe preorder plus inorder
timotimo if you want to put attributes on a node, you'd put a line like "node a [color = red]" or something
if you want nodes to have more "complicated" names, you can quote 'em
brrt so, something like (LOAD (ADDR (LOCAL) 16) 8) would become... 09:31
timotimo oh, you can also write "a -> b -> c -> d -> e;" in a single line
brrt digraph Foo { node n_1 [label="LOAD"]; node n_2 [label="ADDR"]; node n_3 [label="LOCAL"]; node v_1 [label="16"]; node v_2 [label="8"]; n_1 -> n_2; n2 -> n3; n_2 -> v_1; n_3 -> v_2; } 09:33
timotimo yeah, that'll work
pro-tip: use dot -Tx11 and you'll get a scrollable, zoomable thingie 09:34
d'oh, syntax error 09:35
"near '['"
ah, "node" is not actually a thing 09:36
brrt oh, thanks
i can do that 09:37
10:22 donaldh joined
jnthn morning, #moarvm 10:57
11:08 pyrimidi_ joined
JimmyZ morning, jnthn :) 11:14
brrt good moarning jnthn 11:25
13:12 domidumont joined
psch JimmyZ: directly passing arg to can_store is what i had before. that always segfaults 13:34
why afaiu is because the != MVM_REPR_ID_NativeRef if already deconts..? 13:35
and adding only !MVM_is_null still gives me a SEGV for Scalar.WHICH 13:38
s/why/which/ 13:40
JimmyZ psch: then I have no idea :) 14:55
psch: jnthn++ may have 14:56
dalek arVM/even-moar-jit: 04b540b | (Pawel Murias)++ | src/strings/ops.c:
Make escape handle \r\n properly
 15:02
arVM/even-moar-jit: cdf1067 | jnthn++ | src/strings/ops.c:
Merge pull request #290 from pmurias/fix-escape-crlf

Make escape handle \r\n properly
arVM/even-moar-jit: 4a35f2c | timotimo++ | src/io/syncfile.c:
handle files where stat falseley reports 0 bytes size
arVM/even-moar-jit: 04c431f | brrt++ | src/ (2 files):
8a1ad8b | brrt++ | src/jit/label.c:
15:24 brrt joined
15:43 domidumont joined
dalek arVM/even-moar-jit: cf7f90d | brrt++ | src/jit/label.c:
Another typo fix
 15:45
15:50 tokuhiro_ joined
hoelzro good moarning #moarvm 16:04
jnthn o/ hoelzro
hoelzro o/ jnthn
brrt good * hoelzro 16:05
hoelzro greetings brrt 16:06
dalek arVM/even-moar-jit: 2a41347 | brrt++ | src/jit/log.c:
Dump dot digraphs of expression trees

The expression 'trees' are really graphs, and this way we can visualize them much more readily.
brrt timotimo++ for the idea 16:07
dinner & 16:09
16:18 zakharyas joined
nwc10 hoelzro: did you get enough anwser about ASAN? 16:45
16:46 colomon joined
hoelzro nwc10: timotimo told me about --asan, but I wasn't able to get it working last night =/ 16:46
nwc10 :-( 16:47
do you remember what the testcase that triggers it is?
I have one built
hoelzro I can build MoarVM, but once I try to perl Configure.pl --gen-nqp in rakudo, MoarVM gets mad
nwc10 oh. odd
hoelzro I don't offhand; I was hoping that pastebin link would be around
nwc10 did you have an irlog for when it was? 16:50
hoelzro yes, one second 16:51
nwc10: irclog.perlgeek.de/moarvm/2015-09-30#i_11296805
17:04 kjs_ joined 17:19 domidumont joined 17:52 tokuhiro_ joined 17:56 FROGGS joined
FROGGS o/ 17:57
18:05 zakharyas joined 18:11 ggoebel joined 18:46 Ven joined 18:54 ilbot3 joined
hoelzro o/ FROGGS 19:15
nwc10: may I ask what your OS/toolchain is like? I would like to get an ASAN MoarVM built, and perhaps fix up the issue I'm running into 19:21
nwc10 it' snot my machine. It's fairly old Centos 19:22
and root built: gcc (GCC) 4.9.0 19:23
... 2.6.32-358.el6.x86_64 #1 SMP Fri Feb 22 00:31:26 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux
hoelzro so you're using the GCC ASAN? 19:24
nwc10 yes
hoelzro hmm...maybe I can try a build in a VM, see if that works
I'm using a pretty new GCC
but it fails for me with clang too
19:30 kjs_ joined
nwc10 I've not yet figured out what version of nqp and rakduo I was using at the tme 19:31
timotimo i kind of wish brrt had put online an example of the dot output
now i'll have to rebuild moar with even-moar-jit to see it :|
hoelzro I can play around with the ASAN stuff later tonight 19:33
timotimo hoelzro: in what way does asan fail for you? 19:35
hoelzro timotimo: an ASAN configured MoarVM will fail to build NQP
I can paste the output in a bit, if you'd like
timotimo sure 19:36
i was hoping it'd be something simple like "cannot link moar because something's missing"
to which i'd say "yeah, install the dev files for asan"
hoelzro moar *did* get mad when I did --asan --compiler clang 19:38
but that moar fails to run entirely
let me go break Moar real quick
timotimo heh 19:39
hoelzro timotimo: gist.github.com/hoelzro/f92385cef4e2f7a7f40f 19:46
timotimo er, wha? 19:47
it's not supposed to quit out if there's leaks at the end 19:48
thing is, moar will just exit(0) out when the main code is finished
so we *expect* lots of leaks at the end
for things that are life-of-program scoped
i mean, we still do throw out bytecode objects when they get garbage, but mostly they don't 19:49
huh
a few of these are unexpected, though. why would MVM_string_decode_stream_sep_default leak? 19:50
or does that just mean the MVMString it belongs to didn't get cleaned up?
anyway, if you can get the exact invocation that it uses, you can pass that flag to moarvm that asks it to do a full cleanup when it shuts down
that'll either make the leak sanitizer happy or it'll at least show "real" leaks
hoelzro I just did; still got a bunch of warnings and a non-zero exit code 19:51
=/
timotimo update the gist please? :) 19:55
(i'm hoping that the output would have changed!)
hoelzro moment 19:56
gist.github.com/99a68f84f1d8034e7d5e 19:58
timotimo urgh, it has gotten *more*? 20:01
hoelzro it varies from run to run
timotimo perhaps --full-cleanup has rotted a tiny bit and more things need to be handled there? :\
and we can't ask asan to ignore leaks?
hoelzro oh wait, now it's a consistent 2388 lines
that's a good question 20:02
timotimo can you ensure that the two gists come from the exact same commandline?
hoelzro they are from the same session, with the same build of MoarVM
timotimo i mean
perhaps we're comparing two different invocations of moar? 20:03
hoelzro I copied the output from Configure & make for the first, and then I ran that raw command with --full-cleanup for the second
timotimo like, perhaps one of them is for -e 'nqp::getconfig().str.spurt'
(the actually working equivalent of that)
hoelzro ah ha
timotimo wait, this is about compiling nqp
it can't use that %)
hoelzro yeah
heh
timotimo or perhaps it does, with the bootstrapped nqp, but that'd be nonsense
okay, whatever :| 20:04
we must find a way to tell asan to not use the leak sanitizer
hoelzro or fix the leaks ;)
I'd rather take on the former task =)
20:05 vendethiel joined
timotimo a gigantic amount of those leaks are from the fixed size allocator; i'm not sure if that was already around the last time i used --full-cleanup 20:07
hoelzro "The leak detection is turned on by default on Linux; however, it is not yet supported on other platforms." 20:14
that explains some things 20:15
timotimo the last time i used asan (also with gcc) it might have reported leaks, but it didn't set a failing exit code 20:17
anyway, you can compile moar without asan to build things, then recompile maor with asan to test your stuff 20:18
i should really have pointed that out sooner!
hoelzro I did that, but then the Makefile helpfully noticed that $(which moar) hand changed, and subsequentially rebuilt rakudo =/
I *could* just modify the Makefile for now =)
s/hand/had/
20:18 kjs_ joined
timotimo what?! 20:19
when did that happen
hoelzro when I tried last night 20:20
timotimo moar shouldn't cause rakudo to rebuild
hoelzro it looked like it did
timotimo ... at least when you're a dev like us :|
hoelzro hehe
I'll try right now
timotimo shouldn't as in: "IMO that's not what it should do", not as in: "i know it doesn't do that"
hoelzro I hate that we don't make that distinction in English =/ 20:21
maybe it ought not to?
hoelzro frequently struggles with should vs ought vs supposed to
grrr, there's ASAN_OPTIONS 20:22
but no docs on it!
timotimo right, i have probably ought to have used "ought"
ASan's behavior can be changed by exporting the ASAN_OPTIONS env var. Some of the useful options are listed on this page, others can be obtained from running an ASanified binary with ASAN_OPTIONS=help=1. 20:23
www.chromium.org/developers/testin...ssanitizer
isn't it obvious that the docs for ASAN_OPTIONS would be found under chromium.org?
hoelzro of course!
twitter.com/hoelzro/status/664901403509395456 20:24
export ASAN_OPTIONS=alloc_dealloc_mismatch=0:detect_leaks=0
from the sanitizers repo
nwc10 hoelzro: paste.scsys.co.uk/501378 20:25
hoelzro yaaay nwc10++
now I don't have to fix my ASAN MoarVM =P
(just kidding)
nwc10 it would be useful if you did, as I'm going to go to bed at some point 20:26
hoelzro don't worry, I will =)
thank you for reproducing that for me
nwc10 it was mostly a matter of time, waiting for things to build
(and having to guess suitable revisions) 20:27
20:33 Peter_R joined 20:35 ilbot3 joined
hoelzro timotimo: just tried the rebuild thing; Rakudo did *not* rebuild after I built a new Moar 20:39
nwc10: thanks for putting in the work! hopefully I'll finish before the paste expires this time =) 20:40
ASAN tip: ASAN_OPTIONS=detect_leaks=0 20:47
timotimo phew 20:54
hoelzro: do you think this warrants an entry to the perl6 faq?
"i tried running rakudo under asan/valgrind/... and it reported oodles of memory leaks"? 20:55
hoelzro hmm
seems rather low level
we might want to include it under the --asan option help
timotimo fair enough
moritz timotimo: #perl6 hasn't been swamped by that question :-)
timotimo that's true
hoelzro timotimo: re MadcapJake's issues, do you think we can detect if we're running under valgrind and imply --full-cleanup? 21:04
same for ASAN? 21:05
timotimo oof
what's the thing called that you can load up symbols with?
we could use that to figure out if a random asan symbol is in our executable already 21:06
hoelzro dlopen/dlsym?
we could also have --asan set -DALWAYS_BE_CLEANING or something
timotimo yeah, that 21:07
hoelzro valgrind.org/docs/manual/manual-cor....clientreq
valgrind.h provides a RUNNING_ON_VALGRIND macro
21:08 tokuhiro_ joined
hoelzro but that means you need to configure MoarVM after having installed valgrind 21:08
and a lot of people install valgrind once they need it
timotimo yeah, i wanted to add valgrind clientreq at some point
but it was a bit harder than i thought
it did improve things a bit 21:09
like, you'd usually get things like "object is 1000000000 bytes into a block of 1000000000000 allocated at MVM_instance_create" 21:11
when things in the nursery go wrong
21:30 lizmat joined
psch github.com/MoarVM/MoarVM/pull/291 21:58
i'll push the corresponding changes to rakudo to a branch there for now
hoelzro psch++ 21:59
psch that'd be 3 days for an xmas ticket... feels a little underperforming considering jnthn++ pulls one or two a day... :P 22:00
hoelzro psch: jnthn is magic, though =)
and it's still work that goes towards xmas! 22:01
psch yeah, i'm not really comparing myself, nor taking it as a benchmark
hoelzro your 3 days is better than I've done for the last few months =/ 22:02
22:02 kjs_ joined
psch i was really surprised having to get into mvm code with that though 22:04
RT #74414, rt.perl.org/Ticket/Display.html?id=74414 for reference
but yeah, it happens
the rest of the dependencies to #123766 seems really involved though :/ 22:05
not sure i can find something i can do anything with...
hoelzro that's a complicated issue
psch hoelzro: #74414? nah, in itself not so much
hoelzro: the trouble was caching, in the end :) 22:06
skipping multicaches it was solveable completely in BOOTSTRAP.nqp
but that's obviously regressive wrt performance at least, so...
well, at least that's how i felt about it. maybe it's actually complicated and i just have too much spare time... :S 22:10
hoelzro =)
22:12 FROGGS joined
hoelzro wow, ASAN spectest sure takes a long time 22:17
22:30 Ven joined 22:46 Ven joined 22:49 Ven joined 22:53 FROGGS joined 23:16 tokuhiro_ joined 23:34 Ven joined 23:46 Ven 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!