00:32 zakharyas joined 01:25 zakharyas joined 03:44 vendethiel joined 04:43 vendethiel- joined 06:08 agentzh joined 06:33 zakharyas joined 06:58 zakharyas joined 07:01 domidumont joined 07:51 nebuchadnezzar joined 08:14 brrt joined
brrt good * #moarvm 08:20
i survived attempts by the national railways to suffocate its travellers once more
and on the way i realized
there's only one teeny, tiny bit to fix before the expr JIT can compile c function calls 08:21
samcv hey brrt do you know about MVMROOT 08:26
i would like to learn its magic and how it works :)
not sure when it is needed vs when it isn't needed, which variables need it etc 08:27
brrt i know something about it 08:29
what it does, iirc, is add its argument as a 'temporary GC root' 08:37
so
we have a precise garbage collector, right 08:38
samcv ok so it helps out the GC. so why does adding MVMROOT for a variable make it not crash. does it GC those values sometimes? 08:44
and then ruin everything?
and why doesn't that need to be done for _everything_ 08:45
09:21 domidumont joined
brrt okay, sorry, i was off for a bit 09:26
so
precise 2 gen garbage collector
young objects, nursery, bump allocation, two-space
09:28 domidumont joined
brrt during collection, collector walks all 'roots', i.e. all active frames and their work space / lexicals 09:28
however, what if you create a new object, and then directly allocate another (during initialization) 09:29
then, prior to assignment to the work space registers, the collector can't find that object, and so would consider it's memory dead
the garbage collector could move the object out underneath it's pointer 09:30
that's what the MVMROOT thing does
it pushes the pointer (or rather, i think, a pointer-to-the-pointer), on a temporary stack, and if any GC would happen during the block it covers, it will a): scan that pointer, b): update that pointer if it were moved 09:31
samcv: does the above make sense? 09:53
09:58 domidumont joined
samcv yeah i think so 10:06
so if you create a value from another one. and the GC applies to only MVM type objects or how does it work 10:07
like MVMString * i know it will do at least. what about just plain integers? that's just on the stack right 10:08
jnthn Yeah, ints and stuff just live on the stack 10:10
Basically it's anything that's an MVMCollectable
So, MVMObject, MVMSTable, etc.
And MVMString is just a type of MVMObject 10:11
samcv cool
well gonna run spectest on this and see how it fares. hopefully doesn't fail anything
jnthn Lemme know if you want me to take a peek at some patch
samcv yeah check my PR and you can see the last 2 commits i did 10:12
cause before there was a pretty big flaw as i commented. this way is much nicer anyway. plus less work changing case theoretically 10:13
jnthn Ah...so my tired "what if the chars that changed aren't before the offset" was sorta an on something 10:14
jnthn finally got decent sleep and might be more useful today...
samcv and this doesn't have to deal with any of the possible expansions beforehand, so the start point doesn't ever have to have an offset 10:15
m: say staa' ~~ m:i/st/ 10:20
camelia ===SORRY!=== Error while compiling <tmp>
Two terms in a row
at <tmp>:1
------> say staa⏏' ~~ m:i/st/
expecting any of:
infix
infix stopper
postfix
statement end
statement modif…
samcv m: say 'staa' ~~ m:i/st/
camelia ｢sta｣
samcv that is not fixed but that's on the nqp or perl6 side
jnthn Looking at this patch...we fc the whole bunch just to see if we need to do the smarter thing below 10:21
samcv yeah that may get removed and just use the same for both 10:22
jnthn But the smarter thing below would be cheaper
Right
I'd say it probably should; will be a lot simpler/clearer then
Not to mention cheaper
And probably allocation-free also
Meaning regexes with :i would create less chunk 10:23
*junk
samcv yeah
well we will need needle_fc though 10:24
we will still apply fc to needle always
jnthn Ah, true
Though we could iterate through the graphemes of that fc-ing them also
samcv so if we don't use needle afterward, then it doesn't have to be in MVMROOT right? 10:25
jnthn Right
Which would mean we don't expend effort fc-ing all of needle if smething early on didn't match
samcv very true
jnthn Though the needle is usually gonna be much smaller than the haystack, so you're already getting the big saving :-) 10:26
samcv it became much easier when I renamed the variables h and n for haystack and needle rather than a and b 10:29
jnthn :D 10:30
Yeah, I noticed that in the PR 10:31
dogbert17_ o/ 11:04
samcv o/ dogbert17_
dogbert17_ any new SEGV's today? 11:05
I have run a bunch of harness6 test runs without any problems, looks very good. jnthn++ 11:06
the question though, is lizmat happy :)
jnthn Nice :) 11:08
dogbert17_ we're running out of bugs :) 11:14
brrt well, i'll be soon introducing fresh new ones 11:28
don't worry
samcv jnthn, added comment to PR curious what you think on this 11:44
jnthn samcv: Lunch now; will look after :) 11:52
timotimo looking forward to all those new bugs 11:53
nwc10 don't worry, I'm sure Zefram will find lots
timotimo zefram is really good at that 11:56
jnthn Bah, forgot I needed to make some rice too 12:03
jnthn looks while that cooks
samcv i'm prolly going to bed now. see you all tomorrow 12:05
jnthn Left a comment 12:09
Rest well o/
timotimo gnite samcv
dogbert17_ runs a real stresstest to be sure (nursery = 64k, MVM_GC_DEBUG=2, FSA_SIZE_DEBUG=1 etc) 12:15
dogbert17_ no SEGV or panic at least but a couple of extra tests complained 12:44
timotimo nice 12:45
tests aren't clean on my end either
don't know if they should be 12:46
dogbert17_ I think a few are expected to fail atm, e.g. t/spec/S11-modules/require.t 12:48
jnthn I thought a fix went in for that one this morning 12:49
(Only a couple of hours ago) 12:50
lizmat t/spec/S11-modules/require.t is clean for me now 12:57
m: my $s = 0; ($s++,) Zxx 9; dd $s
camelia Potential difficulties:
Useless use of Zxx in sink context
at <tmp>:1
------> my $s = 0; ($s++,) ⏏Zxx 9; dd $s
Int $s = 9
lizmat oops
jnthn m: my $s = 0; $s++ Zxx 9; dd $s 13:14
camelia Potential difficulties:
Useless use of Zxx in sink context
at <tmp>:1
------> my $s = 0; $s++ ⏏Zxx 9; dd $s
Int $s = 1
jnthn Why are the parens needed to make it thunk properly? 13:15
Oh, wait, maybe the list matters? 13:16
m: my $s = 0; $s++ xx 9; dd $s 13:18
camelia Int $s = 9
jnthn Curious that doesn't warn but the Zxx form does
14:34 brrt joined 14:37 lucasb joined
lucasb I just heard about a workshop called MoreVMs 2017 Workshop on Modern Language Runtimes, Ecosystems, and VMs 14:39
from this: 2017.programming-conference.org/ 14:40
I'm just sharing because I found the name funny :)
jnthn haha :) 14:41
timotimo does nqp::ordat ever do anything with ASCII? :\ 14:59
Error encoding ASCII string: could not encode codepoint 773
points at
if nqp::ordat($text, '"', $pos) == 34 { # "
oh wait
m)
yeah, that wasn't so clever
jnthn It's a string operation, and strings aren't encoded, so...
timotimo the problem was it used to be eqat, but i forgot to remove the '"' when i turned it into ordat 15:00
perl6 --ll-exception -Ilib -MJSON::Fast -e 'say from-json(q{ { "̅haha": "lol" } }).perl'
${"\x[305]haha" => "lol"}
moritz \o/ 15:16
timotimo anyway, i pushed that along with a version bump 15:17
but now alexdaniel isn't here to test it 15:18
and the escaping thing with control characters is still missing 15:23
15:36 brrt joined 16:05 brrt joined 16:27 lizmat joined 16:32 lizmat joined 17:42 domidumont joined 18:57 agentzh joined 22:04 agentzh joined 22:58 zakharyas joined 23:00 agentzh 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!