github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
MasterDuke timotimo: when you said "in that case the last breakpoint in the trace script is too early", did you mean the list of breakpoints in the trace script is missing something? 00:50
timotimo yeah, i think so
MasterDuke ah. i was interpreting it to mean that the breakpoint i was creating manually in gdb in MVM_spesh_candidate_add was wrong 00:51
have any idea what should be added to the script?
timotimo oh, sorry about that
at the very least a breakpoint for when the whole speshing work is done 00:52
MasterDuke no worries, it's pretty clear now, i just wasn't reading clearly
timotimo but we'll want one as close to the point where it gets put in so we can sensibly "bisect" i guess
MasterDuke i am kind of surprised i didn't see anything (useful) when i added optimize_plugin 00:53
timotimo you only added "the start" of it, right?
MasterDuke ? 00:54
timotimo well, the script only uses the function names
so what you get is a dump before that function runs
MasterDuke ah, so i should also add whatever function runs next? 00:57
timotimo yeah
i'm not sure if gdb has some kind of syntax for "break on return from the function 'blah'" 00:58
that would be a suitable candidate, too
line numbers on the other hand are a very bad idea :D
MasterDuke actually bad? 00:59
timotimo yeah 01:00
when the file changes, you'd have to keep the script up to date
uuugh :)
MasterDuke afk for a bit, but i may hardcode things for this current debugging effort 01:01
timotimo that's fine. maybe the script should just offer another function that lets you add more breakpoints
MasterDuke since it's taking longer than i thought and i soon wont have quite as much time to devote
timotimo: hm, i added a breakpoint at the line immediately after optimize_plugin is called. i see a commit for optimize_plugin, but not the newly added breakpoint 01:34
MasterDuke ugh, why is this not working?!?! 02:00
Geth MoarVM: 30ab2af3b3 | (Daniel Green)++ | src/jit/graph.c
Add op name to jit "bailed completely" error
 02:01
MoarVM: 1ff55bf1cc | MasterDuke17++ (committed using GitHub Web editor) | src/jit/graph.c
Merge pull request #1031 from MasterDuke17/add_op_name_to_jit_bail_error_string

Add op name to jit "bailed completely" error
MasterDuke huh, the script hangs even if the only two breakpoints i have are optimize_plugin and the line after where it's called 02:12
even weirder, now it finished, but there aren't any commits from the breakpoint after optimize_plugin is called. i guess that means the output didn't change? 02:15
timotimo yeah 02:17
i do think i enabled "make empty commits"? or maybe i didn't
it could very well be that you'd end up with a crapton of empty commits 02:21
you can put a "--allow-empty" in the commandline for "git commit" in the trace script 02:22
MasterDuke hm, i'll give that a try 02:24
ah, now there is a commit for the next line, but yes, it's empty 02:25
timotimo OK 02:26
"no changes", then
MasterDuke heh, still no idea why the speshresolve doesn't show up. there are some in other BBs, but it never appears in the BB i care about 02:27
timotimo oh 02:28
the numbers change around a few times
MasterDuke it seems to be very consistent 02:29
timotimo no i mean
from start to end of one spesh run
the BBs get renumbered
the addresses in the parenthesis stay the same
but sometimes BBs get deleted, split, merged, etc
MasterDuke hm. afk for a bit, i'll see if i can see that happening when i get back
Geth MoarVM: pmurias++ created pull request #1033:
Trim array returned by nqp::decodelocaltime		 10:32
nine How comes that when precompiling Inline::Perl5::Parent (and only that) serialize_repossessions finds a reference to CompUnit::Repository::Staging's serialization context and even weirder through Perl6::Grammar's methods? 15:54

Powered by Cro and the Raku® Programming Language.

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

Thank you!