01:08 TimToady joined 03:34 jepeway joined 04:07 vendethiel joined 04:49 vendethiel joined 06:01 harrow joined 06:09 vendethiel joined 07:00 FROGGS joined 07:03 vendethiel joined 07:04 Ven joined 07:20 FROGGS joined 07:54 zakharyas joined 08:07 colomon joined 08:12 Ven joined
FROGGS ha! 08:12
why does push_X in interp.c has MVM_SC_WB_OBJ while unshift does not? 08:13
jnthn: commenting out that MVM_SC_WB_OBJ for push_X makes push work...
the question is: what does that mean?
08:15 vendethiel joined 08:18 Ven joined 08:58 vendethiel joined 09:00 colomon joined 09:55 vendethiel joined 10:28 FROGGS joined 10:47 Ven_ joined 10:51 FROGGS joined 12:16 brrt joined 12:25 vendethiel joined
masak FROGGS: well, it's a write barrier, but you probably know that already. 12:27
FROGGS: at a guess, removing the write barrier "makes push work" within quotes because there is some race that could be triggered but that you're not triggering. 12:28
FROGGS: the only reason I could see unshift not needing a wb is if it does everything completely safely by atomically moving the pointer to the array itself. but more likely that's just an oversight, and a race waiting to happen. 12:29
FROGGS I just wonder why/how hitting a write barrier makes my playing cards building collapse
masak that is a really good question. 12:30
FROGGS and I dont understand enough about that topic
I mean, I dont understand how hitting a WB has an influence of (dis)owning objects on an SC 12:32
12:53 Ven_ joined
brrt FROGGS, silly question, does it still collapse on MVM_JIT_DISABLE=1 13:05
:-) 13:06
because the JIT also does direct sc access
FROGGS I disabled spesh yesterday and that did not change anything
brrt ah well 13:07
shame
FROGGS and it is nice that it is not a spesh or jit(ter)bug 13:08
brrt well, yes :-) 13:09
nwc10 FROGGS: is the testcase more than a branch in the 3 repositories?
FROGGS nwc10: no, it is exactly that 13:11
nwc10: jsoff in all three, then running: perl6 ser.pl thrice
"thrice"
hmmmm, does that mean that I do not clear the SC ownership correctly? 13:15
so that MVM_get_idx_of_sc still reports something > 0 for my (meant to be) disowned objects
okay, I've got something to look at
hmmm no, I'm doing that right 13:23
so if the thing I am pushing to was one of the objects of my SC, the WB code would be (correctly) a noop 13:24
13:33 FROGGS joined
FROGGS hmmm, how does one translate fremdschämen... 13:41
nebuchadnezzar FROGGS: en.wiktionary.org/wiki/fremdsch%C3%A4men 13:42
?
FROGGS ohh, it actually is a word...
nebuchadnezzar: that expresses my feelings at the workshop right now 13:43
nebuchadnezzar erf, so you are embarrassed because someone else embarrassed himself witout noticing it 13:44
I may you that work for our french deputies, but I'm more ashamed by the whole country :-/ 13:45
FROGGS nebuchadnezzar: the current speaker seems to be on another level than the audience it seems 13:49
it is hard to put in words
nebuchadnezzar I think I see, he does not notice that nobody follows his explanation because it's too high level, or something like that 13:50
FROGGS nebuchadnezzar: sort of, the audience now asks: why are you doing that? what is its goal? what problems do you want to solve? 13:54
and it is getting more strange every minute it goes
nwc10 FROGGS: ser.pl under valgrind produces no errors 14:22
timotimo right; that's not the failure mode, though :(
nwc10 (run 3 times, "version skew" error on third run
also, FROGGS++ is up on stage currently
so might be a bit slow in replying :-)
timotimo FROGGS: there ought to be an SC WB in all "put something into a list" instructions 14:23
the SC WB triggers reposession, AFAIK
14:43 vendethiel joined
jnthn FROGGS: See src/6model/6model.h:157; clearing the SC is a bit more involved than just setting sc_idx to 0, also idx needs to be, and see the bit in the #ifdef two which may give somehting to NULL, barrier, and free 14:48
Uh, darn, that'll still be a data race
FROGGS ohh, MVM_USE_OVERFLOW_SERIALIZATION_INDEX is actually defined 15:12
jnthn :) 15:13
FROGGS I totally missed that
ohh no, it isnt 15:14
15:30 vendethiel joined
FROGGS okay, setting sc_idx *and* idx to zero does not improve it much :o( 15:59
japhb FROGGS: What did you give a talk on? (nwc10 mentioned you were on stage earlier) 16:02
FROGGS japhb: Perl5 XS, XSpp, SWIG and NativeCall 16:03
japhb FROGGS: Ooh, nice. Slides? 16:04
timotimo i surely hope NativeCall made a good impression 16:05
even though we still don't have arrays inside chars ;(
structs*
japhb We have AoS but not SoA? 16:06
timotimo %) 16:07
16:22 tadzik joined 16:38 vendethiel joined 17:24 FROGGS joined
timotimo oh, wait, that was serious and you meant "Array of Struct" and "Struct of Arrays" rather than "Array of String" and "String of Array" 18:08
the problem is that we don't have a way to say "this in-line array has a size of $n" yet 18:09
FROGGS right, that's blocking it
timotimo i'm close to just saying "fuck it" and implement *something* that brings us closer 18:20
FROGGS :o)
timotimo "well, you have to be root and load this textfile into /dev/mem and then use the sha256sum of your kenrel config as the name of the class, but then it works!"
"also, it only works if the in-line array has a prime number of elements in it" 18:21
"on a friday"
FROGGS "but not this friday" 18:23
timotimo so basically, if i put a trait "is inline-array-size['@!foobar', 10]" on the class, that'd be a workable first step so that we can at least experiment with the rest? 18:25
18:47 Peter_R joined 18:49 danaj joined, colomon joined 18:50 harrow joined 19:19 danaj joined 19:22 lizmat joined 19:30 Peter_R joined 19:48 Peter_R joined 19:53 lizmat_ joined 19:54 Peter_R joined 20:01 lizmat_ joined 20:05 rurban joined 20:33 colomon joined 20:50 brrt joined 20:56 Peter_R joined 21:05 lizmat_ joined 21:09 lizmat__ joined 21:39 colomon joined 21:49 Peter_R joined 23:11 vendethiel joined

Powered by Cro and the Raku® Programming Language.

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

Thank you!