github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
Geth MoarVM/derived-specializations: 4fd8740463 | (Jonathan Worthington)++ | src/6model/reprs/MVMStaticFrameSpesh.c
Mark the type tuple stored in a spesh candidate
16:19
MoarVM/derived-specializations: 61176ca3e2 | (Jonathan Worthington)++ | src/spesh/plan.c
Use candidates for existing specialization check

This is much simpler than walking the guard tree, and also will be robust in the face of derived specializations.
17:06
jnthn Hurrah, the new specialization guard tree builder that I now have locally is good enough to survive NQP make test 17:08
Two missing bits to add to make it, in theory, good for Raku also 17:10
nine Nice progress :) 17:11
jnthn This is all a yak shave for derived specializations, although it will also simplify the arg guard tree walking too, by needing a piece of state less
oh wow, turns out I also never implemented a possible optimization in the guard tree walker too... 17:13
nine I know how it is. I'm currently about 4 layers of yak shaving deep myself :) 17:20
Geth MoarVM: Kaiepi++ created pull request #1240:
Include system headers after headers local to MoarVM in the Makefile
17:28
Geth MoarVM/derived-specializations: 9a59e28fe5 | (Jonathan Worthington)++ | 2 files
Re-implement arg guard rebuilding

Now we rebuild the argument guard tree as a whole, rather than trying to copy and tweak the existing tree. This makes things far easier to reason about, especially given we're about to add support for specializations on only some argument types, which are far more challening to update an existing tree to incorporate. This approach also means we can convey the certain specialization results to all the needed places, so we should be able to remove the special-case result node for that also, just making the arg guard walker a little simpler.
17:48
jnthn Now it handles the Raku cases too, so switched over to it :) 17:50
lizmat and yet another Rakudo Weekly News hits the Net: rakudoweekly.blog/2020/02/17/2020-...eblessing/ 17:55
jnthn And reapplying the derived specialization patch atop of all that yak shave means it goes from a build failure very early in the NQP build process to a single SEGV in the NQP test suite 18:06
The rakudo build makes it to CORE.setting compilation before SEGV, and probably it'll be for the very same reason.
nwc10 but first, food?
nine So it's nice of that test case to fail, so you won't have to debug CORE.setting compilation :) 18:07
Geth MoarVM/derived-specializations: 5e7e701615 | (Jonathan Worthington)++ | src/spesh/plan.c
First attempt at derived specializations

This tries to find patterns where some types are very stable, and others vary, and then makes specializations that only specialization the stable types. The hope is that this will allow for far better scalability of our optimizations in real world programs, where some things - such as array and hash operations - encounter a broad range of types in the value, but are pretty much always called on the same kind of thing
  (Array, Hash, etc.)
18:09
jnthn nine: Indeed :)
nwc10: Yeah, though first I need to walk home, then when I get there I need to cook it... :)
Will probably get back to this tomorrow morning.
Happy ASAN barf, I guess... ;)
o/ 18:13
nwc10 jnthn: was testing the previous commit. ASAN finds the previous commit very boring 18:37
will update you when I get to test the new one
\o 18:38
Oooh, intersting, ASAN *already* excitied about part of t/02-rakudo/99-misc.t 18:40
nwc10 paste.scsys.co.uk/587848 18:59
timotimo deck the halls with barf of asan, fa fa fa fa fa, fa fa fa fa 19:20
nwc10 :-) 19:23
hmmm, I assume that t/nqp/057-construction.t is supposed to finish 19:24
(ie, it hasn't)
timotimo sha-dow bytes around the ad-dress 19:30
fa fa fa fa fa, fa fa fa fa
timotimo tweeted that joke 19:33
nwc10 2 Thread 0x7f7121237700 (LWP 1621) 0x00007f71257f011c in MVM_spesh_arg_guard_run_types (tc=0x61800000f480, ag=0x62700041fbe0, 19:34
very busy. Other thread is in pthread_cond_wait@@GLIBC_2.3.2
jnthn Hmm, that's an interesting ASAN output... 20:23
MasterDuke timotimo: seems like something you might be interested in github.com/rohanrhu/gdb-frontend 20:43
timotimo interesting 20:46
gdbgui is a bit wonky, maybe this is better 20:52
Geth MoarVM: patrickbkr++ created pull request #1241:
Revert "Dynamic library loading works with unicody paths on Windows"
21:35
patrickb ^ That PR removes a workaround that has been made obsolete by the recent dyncall bump. 21:41