00:13 ssutch joined 00:20 ssutch joined 00:23 benabik joined 00:33 ssutch joined 00:40 ssutch joined 00:47 ssutch joined 00:49 colomon joined 00:56 ssutch joined 01:12 ssutch joined 01:20 ssutch joined 01:47 ssutch joined 01:53 ssutch joined 02:00 ssutch joined 02:12 ssutch joined 02:27 ssutch joined 02:38 ssutch joined 02:48 ssutch joined
timotimo my friend is now writing (with a tiny bit of my help) python code for gdb to isnpect stack frames for things that aren't but should be reachable from one of the roots 02:53
02:54 ssutch joined 03:00 ssutch joined
TimToady friend++ 03:01
diakopter heh 03:29
timotimo i see debug output on his screen 03:36
TimToady Speak "friend" and press ENTER. 03:39
timotimo how do i find what -> allocate calls in MoarVM's source? 03:42
found it 03:45
it's static allocate functions in the repr codes
i need to pack some stuff and go to bed, i'm getting pretty tired
my friend, however, is still wide awake :)
actually, he's goning to bed, too 03:50
05:22 cognominal joined
FROGGS morning 09:00
09:08 _ilbot joined 10:42 isBEKaml joined
isBEKaml Hi, does moarvm have support for building on cygwin? 10:55
FROGGS not_gerd did some patches for this scenario, but I never tested it 10:58
isBEKaml well, I just tried building it and it failed right at the first step over pthread posix barriers. I guess cygwin doesn't support it and moarvm assumes it does. 11:00
also, perl's $^O gives cygwin but moarvm took it to be unix. 11:01
FROGGS isBEKaml: patches welcome :o) 11:03
because it is hard to fix these things on a non windows+cygwin platform
isBEKaml ok, I got some way past that. 12:07
now, I'm not sure if I should be using WIN32_PLATFORM while building on cygwin. I'm removing that to see how far I can go. 12:08
that is, build on cygwin fails on src/platform/win32/io.o task. 12:09
12:28 isBEKaml_ joined 12:30 isBEKaml joined
nwc10 [Coke]: you're building MoarVM on OS X, right? If so, is all happy with dylib loading once it's installed? 13:21
13:55 isBEKaml joined 15:20 brother joined 15:29 hoelzro_ joined 15:42 isBEKaml joined
isBEKaml Is this just me or a really obvious miss? github.com/MoarVM/MoarVM/blob/mast...win32/io.c - is windows header set somewhere? 15:43
nwc10 nursery torture now down to 92K and still OK 15:55
build times increasing
figured I could run 3 in parallel, so did 15:56
(the machine has enough RAM and CPU)
isBEKaml okay guys, I'm signing off now. here's my work so far(bad or otherwise :-)): gist.github.com/anonymous/8171888 16:05
feel free to pick good ones or reject them altogether.
'night, folks. 16:06
timotimo nwc10: we are currently walking up the stack fro every invocation of MVM_gc_allocate_object and looking at what things are not rooted from the C stack 16:11
do you think that's helpful?
we get lots of false positives, though, tsill
do you have a method i can use to figure out if some pointer is rooted? 16:12
nwc10 I honestly don't know. I'm suspicious that it will have far too many false positives
I don't know conclusively. I know that if you push something to temporarily root it, the call is into MVM_gc_root_temp_push() 16:13
timotimo where do i find the bounds of the current tospace?
nwc10 and that puts them into the array tc->temproots
so everything in *that* array is rooted
timotimo yeah, i look through all of those
as well as the permroots
and the threads, compiler_registry, hllsyms and clargs
[Coke] nwc10: my day to day box is OS X, but I'm having failures on straight linux (diakopter's host06 box) 16:14
nwc10 does -rpath work on your version of darwin? (to get an installed moar executable that can find its dylib) ? 16:15
[Coke] I haven't touched moar on osx in a while. checking. 16:16
timotimo stuff from the tospace start with 0x7777fff or something like that?
7fffff
i think i'm not yet looking through the registers, as those are rooted, too 16:17
nwc10 to space starts at nursery_tospace in the thread context
and from space starts at nursery_fromspace;
look at MVM_ASSERT_NOT_FROMSPACE in gc/debug.h 16:18
but I have a suspicion that the way that code is written (using MVM_exception_throw_adhoc()) that the exception can sometimes get caught and ignored
[Coke] nwc10: um, do I need to add "rpath" somewhere? 16:19
timotimo thanks
nwc10 [Coke]: no, I think the problem is (At least on Snow Leopard) that rpath is ignored
Perl 5 doesn't use it on OS X
[Coke] I'm on 10.8.5 here. 16:20
timotimo before we run allocate, is the fromspace where we should look?
when does the swap between the two happen? 16:21
nwc10 I'm on 10.8.0
[Coke] 3rdparty/dyncall shows as modified and I can't seem to reset it.
[Coke] gives up, and blows away MoarVM entirely. 16:25
nwc10 aaaaargh bollocks
# setup library names
$config{moarlib} = sprintf $config{lib}, $NAME;
$config{moardll} = sprintf $config{dll}, $NAME;
That's not very general
FROGGS that was me... what is the problem? 16:28
nwc10 paste.scsys.co.uk/288153 16:29
I want to use lib%s.dylib in a third entry
FROGGS nwc10: then you need to fiddle the lib and dll entry, no? 16:30
nwc10 no, they already have that
see the paste - I hope it's obvious
Or I want to build the value ldshared using the value of dll
which I can't do at that point 16:31
or I just want to use a Makefile macro
which I can't do, because it's not done that way in the Makefil
I need to add the bit in bold to ldshared => '-dynamiclib -install_name $(PREFIX)/lib/libmoar.dylib'
but I'd like to express it as ldshared => '-dynamiclib -install_name $(PREFIX)/lib/lib%s.dylib' 16:32
because that's how dll is done
but I can't, as it's done later
FROGGS can't you set $config{ldshared} in the configure script?
hmmm
I don't see bold text but I think I know what you mean
nwc10 well, yes, I can. But I think that the build is currently broken on all Snow Leopard
if not all darwin 16:33
and I know that sounds alarmist, but it would not surprise me
Perl 5 is doing it very differently, with -install_name
not that I like dylib.
seems to have a lot more pain than dlopen
I can't see an easy simple way to both fit within the current system and avoid hard coding something that is purpusfully not hard coded elsewhere 16:34
and I can't type
FROGGS :( 16:36
timotimo well, we get a whole lot of output 16:46
17:15 wsri joined 17:16 eternaleye joined 17:21 krunen joined 17:24 eternaleye joined
nwc10 59K fine, building 58K, 57K and 56K 18:44
maybe this is famous last words, but I would have expected to hit a few more bugs than I did
19:57 benabik joined 20:10 colomon joined 20:21 ssutch joined
timotimo well, that sounds good 20:25
maybe now a different class of bugs lurk?
nwc10 possibly 21:22
38K fine, but took 22 minutes 21:23
IIRC we were about 3 minutes with a sane sized nursery
much GC-ing for no good reason (so, we also get to see if anything gets "prematurely" collected)
nwc10 would assume not, as we ought to have seen those already 21:24
35K fine, but took 30 minutes 22:05
diakopter heh 22:13
23:11 ggoebel118 joined 23:31 ggoebel118 joined
[Coke] if I have just Moar, how can I test it? 23:57
(not sure how I can test nwc10++'s earlier request. Guess I'll upsize to nqp)

Powered by Cro and the Raku® Programming Language.

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

Thank you!