cognominal what p6bindattrinvres stands for? Perl6 bind attribute... ? 00:28
tadzik inverse resolution? 01:08
guessing :)
leont What is the provides meta-data used for currently? 01:17
Wrong channel! 01:22
timotimo oh, ops.markdown doesn't have it ... because it's a perl6 op ... 05:49
ah, there it is 05:51
huh. weird. p6bindattrinvres seems to turn pretty much directly into a a bindattrs_o, it just deconts the second argument 05:53
nwc10 good *, #moarvm 06:58
FROGGS o/ 07:04
timotimo good brrt, good nwc10, good froggs 07:10
brrt good timotimo 07:12
hey, timotimo, i thought of something
you were intereseted into copy removals, weren't you?
i've figured out a safe path to implementing itg 07:13
(if i'm ever going into consulting, safe implementation paths are going to be my catchphrase and my moneymaker)
very simple strategy
* you figure out which copys (sets) are redundant, and 'virtually' remove them 07:14
timotimo oh, you mean the "set" operations, yeah?
brrt i.e. at first you don't remove them, but you emit them, *but* create both the deopt mapping and translate all users to the old registers
so you get a bunch of completely redundant set-s which are never used 07:15
this keeps the memory layout exactly the same as before, just tests that the translation mapping is safe
eh use mapping
fine, next step, during deopt, you copy all values from their 'reduced' registers to the redundant registers 07:16
now logically that is still completely redundant? but it demonstrates that your deopt logic works
last step, you remove the set ops proper, and verify everything works as before 07:17
brrt you are done. q.e.d. :-) 07:18
watch as performance blazes
FROGGS :D 07:20
timotimo not so sure about that part :)
brrt no, neither am i 07:25
but you see what i mean, right
timotimo i *think* so 07:28
my moar and perl6 related productivity has been taking a nosedive recently :( 07:29
brrt yeah, mine too 07:40
although my science-related productivity has increased to compensate :-)
timotimo oi, not good, your stuff is important to me! :P
brrt well, i've been refactoring the tiler table generator into a oo format, and i hope to attack the false-cost-table problem 07:41
and aside from that, to mark arguments you're intereseted in and have them passed as well
i can do that now, but it yields to global variable explosion 07:42
that might still be worth it, but understanding suffers
but the dynvar bug is still eluding me, and that hurts 07:43
and another thing i should've done much earlier, is give each jitted code object a sequence number for identification
timotimo a sequence number, eh? 07:44
for cross-reference between what exactly?
brrt well, for instance, after adding the sequence number it became obious that quite a few frames are compiled twice (probably different spesh candidates) 07:45
means that i could be trying to debug a certain frame, and look at another frames binary code altogether
s/obious/obvious/ 07:46
timotimo oh
yeah, that's a very good idea then
brrt i've not yet a complete implementation plan for linearizing tiles 07:47
one thing that bothers me a bit is that i have to mark the condiitional ranges
jnthn morning, #moarvm 09:41
p6bindattrinvres = do a bindattr, and evaluate to the invocant as the result of the op (as opposed to the bound value); really useful in a few places
timotimo oh, i see 09:42
brrt good morning jnthn 09:47
jnthn o/ brrt 09:48
dalek arVM: aac89bd | jnthn++ | src/strings/ (3 files):
Refactor MVM_nfg_get_case_change interface.

It will now be able to return multiple graphemes (though never does so far).
10:52
jnthn m: say "\c[LATIN SMALL LIGATURE FF]".tc 11:22
camelia rakudo-moar 70a94d: OUTPUT«FF␤»
jnthn m: say nqp::tc("\c[LATIN SMALL LIGATURE FF]")
camelia rakudo-moar 70a94d: OUTPUT«Ff␤»
jnthn mm
m: say "\c[LATIN SMALL LIGATURE FF]".tclc
camelia rakudo-moar 70a94d: OUTPUT«Ff␤»
jnthn oh, method tc in Rakudo doesn't actually use nqp::tc o.O 11:23
nwc10 it forks out to Perl 5? :-) 11:24
jnthn No, it uses uc 11:24
wtf :)
tadzik ......really? :D 11:25
jnthn fixes that and crosses fingers :)
timotimo fixes fingers
jnthn yeah really
Better :) Now my tests pass :) 11:27
dalek arVM: ee9fc2b | jnthn++ | src/strings/ (3 files):
Complete length-changing casefold/NFG integration.
11:30
brrt \o #moarvm 15:31
i'd be very pleased if you could review my next blog post for me
brrt-to-the-future.blogspot.com/b/...;type=POST
now i have to be off again :-)
dinner &
Guest87784 so, I finally isolated... a bit... my "invalid GC status" 15:34
Ven_ I guess I'll make a repro repo 15:37
jnthn Ven_++ 15:38
Ven_ github.com/vendethiel/moarvm-gcerr...solated.p6 here's the file. I havn't found a way to isolate it without GTK::Simple, though. 15:39
This errors with "Invalid GC status observed; aborting" (and used to crash badly)
arnsholt .tell brrt +1 on the blog post from me 15:47
timotimo arnsholt: we don't have a bot here, but brrt does backlog dilligently 15:51
Ven timotimo: ^ that's the SPW bug, btw 15:53
timotimo yeah, i thought it'd be
arnsholt Ah, excellent 16:05
retupmoca I'm having trouble building the latest rakudo with VS2015 (both with --gen-moar and master/master/nom): moar.exe dies in CORE.setting stage parse with what appears to be STATUS_STACK_BUFFER_OVERRUN 17:14
is this known? Or (more likely) did I screw up my environment somehow?
[Coke] retupmoca: OOC does it work with */*/nom ? (that is, letting nom pick the versions?) 17:16
ISTR that we've had issues with VS-latest in the past, and I think our devs are using an earlier version... but I'm on mac, so not entirely sure. 17:17
jnthn Yeah, I'm on VS 2013 I think. But I thought somebody patched our build issue...
And it was an issue compiling MoarVM, not using it to build CORE.setting
retupmoca [Coke]: yeah, doing --gen-moar and letting rakudo pick whatever it wants still throws the error 17:24
jnthn: I had it building on VS2015 a month or two ago, but I haven't tried in a while
[Coke] retupmoca: you're the canary for vs2015, then. :) 17:40
it might be helpful if you could bisect the moarvm build and see where it goes south.
(assuming that the core issue here is the moarvm build failing because rakudo config calls it, not because moarvm builds ok and then something dies in rakudo's build. 17:41
retupmoca ok, rakudo 92be579 (--gen-moar) from september 30 builds fine, so I'll try to get a bisect running tonight or so 17:57
[Coke] retupmoca++ 18:18
itz_stmuk 32 bit at least worked OK for me some time last week on VS2015 so it might be recent change 19:04