01:32 FROGGS_ joined 03:46 vendethiel joined 03:52 JimmyZ joined 09:40 JimmyZ_ joined 13:03 jnap joined 13:05 brrt joined
brrt hi #moarvm 13:07
moritz \o brrt 13:09
brrt hi moritz
long time no chat i guess? :-) 13:10
moritz chats dayly, but can't always remember with whom 13:11
brrt :-) 13:12
how's life now?
brrt is already wondering if dynasm can be used for purposes of nativecall
13:58 oetiker joined 14:31 FROGGS joined
FROGGS o/ 14:32
14:53 brrt joined 15:00 zakharyas joined 15:05 zakharyas1 joined 16:31 brrt joined 17:06 jnap joined
FROGGS I know some details about this bug: 18:46
yes 1 | head -20 | perl6-m
# like 13 lines of output
Spesh: failed to fix up handlers (-1, 114, 114)
timotimo good to know. FROGGS -v
FROGGS it is caused by optimize_can_op()
timotimo oh no!
that's my work :(
timotimo so no terribly big surprise it can break ;)
FROGGS :P 18:47
timotimo so i guess the MVM_6model_can_method_cache_only function isn't always safe? 18:49
can you tell more exactly where that error is happening? 18:56
because i see nothing in optimize_can_op or MVM_6model_can_method_cache_only that could cause that error
FROGGS I am spectesting rakudo right now 18:59
18:59 brrt joined
timotimo oke 18:59
FROGGS I can do more perhaps in a few minutes
19:05 zakharyas joined
FROGGS timotimo: it seems to be optimize_iffy in combination with optimize_can_op 19:09
because I only have turned on these two, and there error is still there 19:10
(and vanishes when I disable one of them)
timotimo ah, so it's not the method itself that's exploding, but the resulting bytecode is bogus? 19:11
FROGGS I guess so 19:14
open the repl, and just type 13 times 1 and enter
timotimo we should be able to find out more using the spesh log
masak (also see irclog.perlgeek.de/perl6/2014-05-31#i_8800771 where I check what does and does not trigger it) 19:57
lizmat m: EVAL "1\n" for ^100 # no problem, oddly enough 20:01
camelia ( no output )
lizmat so I would assume it's something around the repl 20:02
timotimo it could be that the "1\n" gets compiled to a new code object each time 20:25
and perhaps that doesn't trigger the spesh
10 runs is when spesh inserts the logging bytecodes, 3 runs later it tries to specialize based on the seen information
lizmat timotimo: do you know where the --stagestats code lives ? 20:31
I want to kill "Stage start", as it's non-info nowadays
FROGGS it is in HLL::Compiler or so 20:45
just grep for usage of nqp::sprintf in nqp 20:46