dalek arVM: 8fe2fdf | (Timo Paulssen)++ | tools/update_ops.p6:
p6 space may know about our flag constants
 01:41
arVM: 61cf617 | (Timo Paulssen)++ | tools/graph_spesh.p6:
version 1 of a speshlog-to-graphviz converter
japhb timotimo: Do you have an example of the graphviz output? 02:13
Oh, I see it in #perl6, nevermind. :-) 02:15
dalek arVM: 3510467 | (Timo Paulssen)++ | tools/graph_spesh.p6:
graph data flow (partially, ignores extops)
 02:37
timotimo t.h8.lv/add_core_op_with_dataflow.svg
japhb: feel free to praise appropriately 02:38
japhb Heh 02:45
It's actually quite cool, very useful for microanalysis
(good visualizations)++
timotimo thanks :)
dalek arVM: eaa3d23 | (Timo Paulssen)++ | tools/graph_spesh.p6:
fix handling of extops in the most naive way
 03:06
arVM: d435796 | (Timo Paulssen)++ | tools/graph_spesh.p6:
handle partial pasted speshlog snippets, too
 03:12
arVM: 5831fc7 | (Timo Paulssen)++ | tools/graph_spesh.p6:
sanitize argument names (for angle brackets)
 03:19
timotimo t.h8.lv/graph_of_gimme.svg - control flow jumps are dotted, set and decont have rounded corners now 03:48
dalek arVM: 84fa2b9 | (Timo Paulssen)++ | tools/graph_spesh.p6:
control flow jumps are dotted, set/decont rounded
JimmyZ timotimo++ # it's really awesome! 14:52
timotimo thank you ♥ 15:32
JimmyZ looked at the 'constant folding' commit and doesn't know what' wrong it is too, it's hard 15:37
[Coke] moar non jit failed a few more tests today: 32089 total.
(that's all of them)
brrt is jnthn still on holiday? 15:38
timotimo i think so 15:42
he'll be pleasantly surprised at how we drove everything against the wall while he was away %)
brrt :-)
[Coke] from my standpoint, more like "into the wall" 15:45
:P
JimmyZ Good night... 15:46
brrt will have to check that when he can access a mac again 15:47
[Coke] Happy to provide diagnostics. 15:48
timotimo i cannot reproduce the failures ;( 16:01
brrt you have a mac timotimo? 16:02
i'm afraid to say it, but i fear something may be wrong with your build system [Coke]
all tests fail at once? that's just odd 16:03
timotimo i don't
brrt hmm 16:06
i cannot explain it currently. i'll be able to run a test monday at earliest 16:07
brrt afk
[Coke] brrt: all tests fail today; over the past few weeks, I've been getting sporadic failures. 16:15
timotimo S02-lexical-conventions/begin_end_pod.t............While looking for 'CORE.setting.moarvm': no such file or directory 16:17
obviously, the build itself failed
[Coke] sure. that has happened a few times in teh past week as well. smells like a GC bug. 16:18
trying a rebuild with latest-rakudo now.
timotimo very strange.
[Coke] the re-run is happier. 16:22
but still failed 521 tests. 16:48
timotimo hum 16:56
[Coke] S03-operators/subscript-adverbs.t S04-statement-modifiers/for.t S05-mass/named-chars.t S32-str/numeric.rakudo.moar 16:57
timotimo i see the for.t failure locally, too
TimToady that one is my "fault" :) 17:06
shoudl be labeled with an RT, but I was too lazy to look
japhb
.oO( laziness with a lowercase 'l' )
 17:07
[Coke] has a few minutes to try to debug some of these aborting tests. 18:24
rove -v -e t/fudgeandrun t/spec/S03-operators/subscript-adverbs.t # malloc, pointer being freed was not allocated 18:27
but: ./perl6 t/spec/S03-operators/subscript-adverbs.t # no errors.
TimToady was trying for far too long to figure out what the "rove" command does :) 18:29
TimToady goes to get more coffee... 18:30
[Coke] :)
also, my ./perl6 is: 18:32
#!/usr/bin/env perl 18:33
exec "ulimit -t 120; ulimit -v 10; ulimit -c 0; nice -20 valgrind ./perl6-m @ARGV"
TimToady that's OSX malloc? 18:35
FROGGS btw, this script that uses DBIish fails on ubuntu when run under cron: github.com/perl6/cpandatesters.per...c-stats.pl 18:37
something about invalid target thread id in work pass or so
TimToady cron tends to be impoverished in env vars 18:38
FROGGS that what I am going to debug today
TimToady not to mention terminals :)
FROGGS but it should fail at a higher level at least
TimToady so I'd guess it's one or the other of those triggering it
FROGGS yeah, we'll see
TimToady though maybe it's related to the thread problems that seem to be more prevalent in child processes 18:39
[Coke] TimToady: I'm on OSx. I don't know if it's osx or moar 18:56
TimToady I think we've only seen that error on OSx, but maybe that's just because it's checking... 18:57
timotimo [Coke]: can you build a moar with asan perhaps? 19:02
[Coke] timotimo: if someone pointed me at how, sure. :) 19:27
timotimo perl Configure.pl --asan in moarvm 19:29
may need to install the asan development package, but it should come with clang itself
[Coke] looks like moar uses it by default if you build with clang, but not with gcc? (but on the mac, those are the same thing, I think) 19:30
timotimo what? asan by default? 19:31
that sounds crazy
[Coke] still no way to pass moarvm config flags down from rakudo through nqp, right?
timotimo not 100% sure, i think there's --moar-option in nqp at the very least 19:32
[Coke] timotimo: look at MoarVM/build/setup.pm for sanitize
moarvm doesn't create a "config.status" when you config it. :(
TimToady no, it doesn't :(
timotimo ... wtf? why would -fsanitize=address be on by default? 19:33
TimToady so don't forget to set prefix
timotimo that's a crazy performance penalty
[Coke] timotimo: clang: error: unsupported argument 'address' to option 'fsanitize=' 19:34
timotimo clang too old? asan has been built-in to clang for a *long* time now
[Coke] so this will take a while to enable.
Apple LLVM version 6.0 (clang-600.0.54) (based on LLVM 3.5svn)
timotimo AddressSanitizer is fully functional on supported platforms starting from LLVM 3.1. 19:35
FROGGS lizmat had that problem too
[Coke] note that my current issues are on a nojit build. perhaps since that's not the default, people aren't seeing that on their platforms. 19:36
seems like we could do a nojit asan build elsewhere and maybe get an explosion.
TimToady wonders about the line in alloc.h that says: if (!ptr && size > 0) 19:47
and whether we're doing a realloc to size 0 somewhere that is assuming the pointer will be valid afterwards
for sure it fails if you take out the "&& size > 0", so there are things reallocing to 0 19:56
so MVM_realloc can return unchecked null pointers
[Coke] looks like mac needs a slightly different clang invocation 19:58
testing...
dalek arVM: bc1c0b9 | (Timo Paulssen)++ | tools/graph_spesh.p6:
also show annotations in the control flow
 20:20
[Coke] running one of those failing tests through an osx asanified build does nothing. 20:25
if I use the prove invocation, I get the same malloc error, but no new output 20:26
TimToady would be funny if it were in the prove program, not in moar... 20:27
timotimo o_O 20:32
TimToady there appears to be an unchecked malloc at ./3rdparty/libuv/src/unix/darwin.c line 80 20:42
there are a lot of places in libuv that return -ENOMEM; gonna be a bear to eyeball all the uses of those to see if anyone ignores the error 20:51
./3rdparty/libuv/src/unix/proctitle.c line 58 is weird, if the malloc fails, returns argv, which is maybe not from malloc 20:58
otoh, can't just be an unchecked malloc, since that would return NULL, which would not complain when freed like that 21:03
FROGGS timotimo: you probably broke it on windows: gist.github.com/FROGGS/bad02c1269672397b639 21:22
timotimo oh wow 22:30
i'm sorry about that
FROGGS that might be even easier to fix than the IO::Path problems :S 22:32
dalek arVM: eba41f9 | (Timo Paulssen)++ | src/spesh/args.c:
fix Microsoft Visual C compliance
timotimo FROGGS: that should fix it?
timotimo i don't know of any IO::Path related problems 22:33
FROGGS $*CWD and dir() return paths containing backslashes, but rakudo can't handle backslashes anymore 22:34
timotimo oh
FROGGS timotimo: aye, moarvm compiled fine again 22:44
timotimo thanks 23:00
still can't get used to that limitation
we may really want to set -Wpedantic or what it's called on gcc

Powered by Cro and the Raku® Programming Language.

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

Thank you!