github.com/moarvm/moarvm | IRC logs at irclog.perlgeek.de/moarvm/today
Set by moderator on 8 October 2013.
JimmyZ --output=src/stage1/QAST.moarvm src/stage1/gen/QAST.nqp 01:09
Routine declaration requires a signature at line 4103, near "(MAST::Nod"
Looks like Missing another MVMROOT
01:22 benabik joined 01:25 FROGGS_ joined 01:34 BenGoldberg joined
dalek arVM: 1a08cce | jimmy++ | src/core/interp.c:
re-order codes to avoids a MVMROOT, and some cleanups
 01:34
arVM: c7fef60 | jimmy++ | src/6model/reprs/P6opaque.c:
re-order part of code to avoid MVMROOTs
 01:57
02:06 colomon joined
dalek arVM: ce75f6d | jimmy++ | src/core/interp.c:
Add a Missing MVMROOT
 02:07
JimmyZ nvm, I was still in moarboot 02:13
switch to master still same error 02:16
.tell I got another backstrace: gist.github.com/zhuomingliang/6957379 02:22
yoleaux JimmyZ: I'll pass your message to I.
JimmyZ .tell jnth I got another backstrace: gist.github.com/zhuomingliang/6957379 02:23
yoleaux JimmyZ: I'll pass your message to jnth.
JimmyZ .tell jnthn I got another backstrace: gist.github.com/zhuomingliang/6957379
yoleaux JimmyZ: I'll pass your message to jnthn.
03:16 colomon joined
BenGoldberg nqp-m: say(nqp::index("this", "h", 1,2,3,4,5,6)); 04:20
camelia nqp-moarvm: OUTPUTĀ«Bytecode validation error at offset 22, instruction 4:ā¤register operand index 50 out of range 0..9ā¤Ā»
[Coke] why does nqp-moar have "index_s" that takes 3 args? java called it "indexfrom". 04:29
moritz hi all. What are the steps for implementing an op in moarvm? 06:16
I want to do bool_I
I looked at add_I as an example, and found that it is mapped to the MAST op add_I; but where is *that* defined? 06:23
JimmyZ moritz: src/core/oplist 06:46
moritz: add op to oplist, then run tools/update_ops.p6, then add it to src/core/interp.c 06:47
07:09 BenGoldberg joined
moritz JimmyZ: and where are the ops actually defined? 07:16
where's the implementation?
JimmyZ src/core/oplist 07:23
the implementation is in src/core/interp.c
moritz thanks 07:27
is it normal to get "use of uninitialized value of type Any in string context in any !cursor_init" from update_ops.p6? 07:35
\\o/ success 07:45
(it compiles) 07:46
JimmyZ: is it usual to separate the manual and the generated changes into two commits? 07:47
07:56 Ben_Goldberg joined
JimmyZ moritz: nope 07:56
FROGGS morning 08:15
moritz: yes, I get the warnings too since a while 08:17
moritz note to self: always pull *before* running update_ops.p6 08:27
unless you are eager to resolve conflicts in generated files
FROGGS "always pull" is a good rule, err, always :o) 08:28
moritz well, when working on rakudo, I seldomly get any conflicts when my working copy is a few days old 08:31
and when I do, they are usually trivial to resolve
FROGGS true, because there are too many different places to hack on :o)
but for moarvm there are files you touch almost every time 08:32
one is interp.c
hmmm, isbig_I wants to be implemented too, I think I pick this one
dalek arVM: 02a7a3f | moritz++ | / (9 files):
implement nqp::bool_I opcode
 08:33
moritz my first moarvm commit that actually implemented a new feature (and not just in Configure)
FROGGS humm, isbig_I seems to be there already, maybe it just needs mapping 08:35
moritz isbig_I is what I used as my model, becuse it also receives a bigint arg and returns a bool (or rather int) 08:41
FROGGS moritz: both do not seem to be mapped 08:45
this would happen in nqp/src/vm/moar/QAST/QASTOperationsMAST.nqp around line 1500 08:48
moritz: btw, we should copy this t/nqp/60-bigint.t to nqp too 08:49
08:51 odc joined
FROGGS ahh 08:59
now I know why my nqp does not build anymore 09:00
moritz: when you add an op in the middle of the list, you need to create a new stage0 for nqp using the nqp-cc
I take the chance to update sprintf 09:10
diakopter moritz: yes needs mapped innqp 09:18
moritz uhm, which one? 09:19
bool_I worked in the test I added
so I can't imagine it needs extra mapping
maybe isbig_I does 09:21
FROGGS: so how do I create a new stage0? 09:28
09:28 wsri joined
FROGGS moritz: I did it already 09:28
nwc10 is it documented how? 09:29
FROGGS dont think so
you got to MoarVM/nqp-cc
configure it, make, and the copy *.moarvm to your nqp's src/vm/moar/stage0
you always need to do it when the ops got reordered 09:30
moritz maybe update_ops.p6 should tell me that 09:32
diakopter yes don't reorder the ops 09:33
girhub cries a little at thr extra bw
moritz didn't reorder the ops
the script did 09:34
FROGGS :o)
moritz so the script needs to be more intelligent, right?
FROGGS well, at least dropping a note wouldn't hurt 09:38
moritz: this is what I get in MoarVM/nqp-cc when running `make selftest`: t/nqp/60-bigint.t ..................... No registered operation handler for 'bool_I' 09:50
did you run that test using the nqp-parrot binary by any chance?
moritz FROGGS: I did a 'make test' in nqp-cc. I simply assumed it does the right thing 09:54
and, why wouldn't it?
FROGGS I'm not sure
moritz: make test does indeed use nqp-parrot 09:55
or better: it uses an nqp from PATH
moritz *facepalm* 09:56
FROGGS well, there are tests (t/qast/) that where meant tu run this way
moritz: you can't come late and expect to get everything in the first run :o) 09:59
moritz FROGGS: but I can come in late and expect the test target to do something useful :-) 10:00
FROGGS moritz: I guess jnthn can tell you why it should do exactly what it does ;o) 10:02
moritz adds a mapping 10:03
ok, does the mapping go into the nqp repo? or into moarvm/nqp-cc/ ? or both? 10:04
FROGGS nqp is enough, though, I did that for bool_I and isbig_I already 10:05
moritz ok, great
tahnks
FROGGS not ok 46 - bool_I(42)
not ok 47 - bool_I(0 10:06
(in nqp, using the runner built using MoarVM)
nqp -e 'say(nqp::bool_I(42))' 10:07
This representation (P6int) cannot unbox to other types
ahh, it is not a bigint
so, the test itself is wrong, roght?
right?
diakopter yeah 10:08
well 10:09
I'm not certain
need jnthn's assessment 10:10
bigint is kindof a gray area
as to where it's supposed to be in the type system
FROGGS hold on, the test creates a bigint using sub box() declared in that file
nqp: my $knowhow := nqp::knowhow(); my $bi_type := $knowhow.new_type(:name('TestBigInt'), :repr('P6bigint')); $bi_type.HOW.compose($bi_type); sub box($x) { nqp::box_i($x, $bi_type) }; say(nqp::bool_I(box(42)) 10:11
camelia nqp: OUTPUTĀ«Confused at line 2, near "say(nqp::b"ā¤current instr.: 'panic' pc 14748 (src/stage2/gen/NQPHLL.pir:5229) (src/stage2/gen/NQPHLL.nqp:279)ā¤Ā»
FROGGS nqp: my $knowhow := nqp::knowhow(); my $bi_type := $knowhow.new_type(:name('TestBigInt'), :repr('P6bigint')); $bi_type.HOW.compose($bi_type); sub box($x) { nqp::box_i($x, $bi_type) }; say(nqp::bool_I(box(42))) 10:12
camelia nqp: OUTPUTĀ«1ā¤Ā»
FROGGS nqp: my $knowhow := nqp::knowhow(); my $bi_type := $knowhow.new_type(:name('TestBigInt'), :repr('P6bigint')); $bi_type.HOW.compose($bi_type); sub box($x) { nqp::box_i($x, $bi_type) }; say(nqp::bool_I(box(0)))
camelia nqp: OUTPUTĀ«0ā¤Ā»
FROGGS nqp-m: my $knowhow := nqp::knowhow(); my $bi_type := $knowhow.new_type(:name('TestBigInt'), :repr('P6bigint')); $bi_type.HOW.compose($bi_type); sub box($x) { nqp::box_i($x, $bi_type) }; say(nqp::bool_I(box(0)))
camelia nqp-moarvm: OUTPUTĀ«Bytecode validation error at offset 22, instruction 4:ā¤register operand index 50 out of range 0..9ā¤Ā»
diakopter moritz: yes it needs mapped in nqp-cc also, in case it's ever included int he bootstrap 10:13
hm, apparently prove -j80 wasn't a good idea 10:15
FROGGS in case we still use the nqp-cc by then
*g*
diakopter wut. 10:16
how does t\\qregex/01-qregex.t take 5 fewer seconds today than yesterday
er, less. :D
FROGGS diakopter: maybe jnthns patch? he said that the testsuite on his machine took about 4s less or so 10:17
moritz no, fewer
seconds are countable
diakopter less
it's not countability
it's discreteness
FROGGS looks like Perl6::Ops is the next bigger thing 10:18
lunch &
diakopter next bigger thing?
FROGGS yes 10:19
diakopter bigger how 10:21
upon further thought, maybe it is countability o_O 10:23
diakopter decides to profile man-or-boy again 10:24
erm, if the cross-compiler worked o_O
moritz blog.oxforddictionaries.com/2012/08...-or-fewer/ makes me believe that "5 less seconds" is acutally correct
diakopter (where "correct" is just "what elitists thinks sounds more erudite" 10:25
)
:)
moritz but OTOH I don't understand their "5 items or fewer" example
diakopter [elitists like me]
10:25 grondilu joined
diakopter well it made me think of collective noun like sugar 10:26
"less sugar" is right. "less cups of sugar" sounds wrong, even though cups aren't discrete 10:27
er. sugar isn't a collective noun. I need to just stfu right now. 10:28
moritz diakopter: or elitists like me, who learned the language through hard work, and are annoyed when what they learned turned out to be not "correct" :-) 10:29
cups sure are discreet 10:30
:-)
diakopter *facepalm* ""what elitists thinks"
apparently it's 3:30 a.m. in my home timezone
10:31 woolfy joined 10:32 cognominal joined
dalek arVM: c76c4db | (Tobias Leich)++ | src/math/bigintops.c:
fix thinko, thruthness means non-zero
 11:18
jnthn diakopter: I think I know how to make things a bit better wrt MVMCode/frame lifetimes 11:36
Will take a look later on
FROGGS jnthn: will hllize and hllizefor be noops? 11:40
jnthn Why did we end up with a another bootstrap update?
FROGGS moritz: added bool_I to the middle of the list
jnthn (stage0)
FROGGS err, s/:// 11:41
jnthn But why not add it at the end?
And let it wait for the re-ordering before we commit to the ordering of what we have forever...
FROGGS he didn't know, and didn't thought about moving it to the end
*and I*
jnthn ah, ok
JimmyZ +1, I don't like always update stage0 because it inc nqp repo size magically 11:42
jnthn It's not a big deal, it's just that each time we do this, we make the nqp repo a little bigger...
FROGGS yeah :/
jnthn JimmyZ: "marginally"? :D
more history = bigger repo is pretty lame magic :P 11:43
JimmyZ well, the stage0 files will be inc-ed x.x MB 11:45
jnthn JimmyZ: Maybe, but packfiles compress stuff too 11:46
JimmyZ i.e: it always a pain to me to do full git clone parrot. because it always break at 53% or some point. And then I need to re-git-clone. well I don't have this problem with rakudo because it's repo is much small 11:48
jnthn JimmyZ: If you don't need the full history you could maybe consider a shallow close 11:49
*clone
JimmyZ hmm, Can I git push a shallow clone with adding my commits to orgin? 11:50
jnthn JimmyZ: I guess if it's not to shallow... 11:52
JimmyZ I can't follow you :(
jnthn JimmyZ: I mean, I guess it may be a problem in some cases if you're pushing a branch and don't have the base or something...but guess you'd not hit this 11:55
JimmyZ Oh. I was fearing about something like this 11:56
nwc10 real 0m1.995s 12:31
user 0m9.288s
sys 0m2.617s
that's configure and 24-way build from clean (admittendly with ccache) 12:32
jnthn nwc10: build of....? 12:38
JimmyZ jnthn: btw, timotimo and I has the same issue, I'm on linux x64
nwc10 MoarVM
*not* wrong channel, at least, not this time :-)
jnthn nwc10: wtf, you can build MoarVM on 2s? :P
*in
nwc10 seems so
jnthn Gee, I want a 24-way...
nwc10 Seems so. It was the output of this: 12:39
for file in `find . -name \\*.git`; do (cd $file/.. && git clean -dxf); done
no, not hat
*after* that, I did this:
time sh -c 'perl Configure.pl --optimize --prefix=/home/nicholas/Sandpit/moar && make -j24 all'
JimmyZ real 0m27.326s 12:41
user 1m48.473s
sys 0m39.265s
with your cmd :(
FROGGS real\t0m9.256s 12:43
user\t0m30.956s
sys\t0m1.432s
nwc10 If I zap the cache for ccache, the cold run is: 12:45
real 0m4.033s
user 0m40.966s
sys 0m7.575s
this is not *my* machine. (It's the one I referred to as "my" machine)
anyway, this wasn't meant to be a brag about hardware.
it was meant to be that jnthn has met the goal of "short build time" rather more effectively than anyone would have guessed 12:46
jnthn ;)
FROGGS sad that nqp does not build in parallel :/ 12:47
jnthn nwc10: I'm curious, can you build NQP for MoarVM (that build won't parallelize well, because bootstrap)
FROGGS: It basically can't.
FROGGS: Because most steps depend on the one that came before
FROGGS I know
simple solution is simple: speed up moarvm :o) 12:48
jnthn Well, we did yesterday
nwc10: And then when it's built, run make test but add -j24 or so to that line ;)
nwc10 actually, I think that the other thing that the short build time indrectly demonstrates is how much we tolerate inefficient slow crap
jnthn nwc10: I can make test NQP here with -j 6 in 9 seconds :) 12:49
FROGGS real\t0m10.823s 12:51
user\t0m37.444s
-j4 that is
time prove -j4 -r --exec "./nqp" t/qregex t/nqp t/p5regex t/qast t/serialization
reordered the steps though
this might even be better: time prove -j4 -r --exec "./nqp" t/qregex t/p5regex t/qast t/serialization t/nqp 12:52
real 10.1s
bbl
jnthn hmm, setmethcache is still vulnerable even after the re-ordering 12:53
nwc10 make -j24 for nqp for that MoarVM:
real 1m58.803s
user 2m10.150s
sys 0m1.270s
jnthn oh no, setmethcache got fixed later; JimmyZ++ 12:56
nwc10: Yeah, it's the prove invocation that could be fun :)
12:57 colomon joined
nwc10 I was trying what FROGGS showed 12:57
jnthn The "time prove ..." bit? 12:58
nwc10 -j4 is fast enough, because t/qregex/01-qregex.t is slow
yes
about 8 seconds
jnthn Not Bad :) 12:59
nwc10 (total time, but really it's "how long does t/qregex/01-qregex.t take")
no. I didn't pay for this hardware. :-)
jnthn yes, well, 01-qregex.t basically does 700+ evals...
So compared to the other tests (which are compile then run) it's a heck of a lot of work 13:00
13:22 colomon joined
dalek arVM: f04eb35 | jnthn++ | / (8 files):
Stub hllize/hllizefor.
 13:25
arVM: 9ec5767 | jnthn++ | src/ (2 files):
Correct possible out-of-date pointer bugs.
 15:26
lee_ diakopter: btw nqp-moar builds for me on OS X now, but capture.t fails with gist.github.com/leedo/4d22a0ea76e6bb4dd24e 15:44
everything else passes :)
dalek arVM: 3b4b362 | jnthn++ | src/core/coerce.c:
Hardening; avoid re-count.
 16:03
diakopter lee_: o_O
16:26 benabik joined
dalek arVM: f82639a | jnthn++ | src/6model/reprs/P6opaque.c:
A couple more pointer-freshness fixes in P6opaque.

Probably not hitting these just yet, but will at some point.		 16:27
16:55 cognominal joined 16:57 ssutch joined
FROGGS current rakudo bt: gist.github.com/FROGGS/561bca9ac7cf945ebc76 17:13
dunno what it means though... 17:16
do we hit some memory limit? or the nursery-size?
jnthn oh, you get that too 17:17
I jsut get some corruption in the output...
I wonder if those two are connected... 17:19
oh, good, I get it on Windows with the debug malloc oto 17:20
I already went through the file and didn't find any places we ensure_space the wrong amount, though... 17:23
r: say 0x8C40 - 0x6C30 17:40
camelia rakudo e22ac2: OUTPUTĀ«8208ā¤Ā»
FROGGS ahh, so we check how many is available before allocating it? 17:45
jnthn yeah 17:46
we are overflowing frame_seg somehow 17:47
FROGGS ahh, hmmm
jnthn The buffer is 8192
And it claims we wrote way over the end of it
But it's kinda hard to see how/where 17:49
FROGGS and what file are you inspecting?
jnthn src/mast/compiler.c 17:50
While compiling Rakudo's Actions.nqp 17:52
maybe ASAN will find it easier...
HEAP[moar.exe]: Heap block at 000000001246F560 modified at 0000000012471570 past requested size of 2000 17:53
Is what it tells me
FROGGS not very helpful
17:54 BenGoldberg joined
FROGGS jnthn: you checked the FRAME_HEADER_SIZE? 18:07
jnthn yeah
FROGGS yeah, seems right 18:08
jnthn I dunno if these are orthogonal issues or not, but even when I bump the initial memory allocation for the frame_seg to avoid the realloc, I still get corrupt bytecdoe out 18:10
nwc10 For me, for NQP, All tests successful. building MoarVM with gcc's ASAN 18:30
which is both nice and not-so-nice
FROGGS I am trying valgrind now 18:31
nwc10 as in, "we have normality, I repeat we have normality. Anything you still can't cope with is therefore your own problem."
FROGGS I hope it finishes today :o)
nwc10 ie it's not pointing out anything obvious that needs fixing which would improve stuff
18:34 cognominal joined
jnthn nwc10: Any chance you could now try the moar-support branch of Rakudo under ASAN? 18:36
nwc10 quite a reasonable chance
jnthn Something silly is going on when compiling Actions.nqp, and while VS's debug malloc catches it, unlike ASAN it doesn't tell you where things got messed up. And it's a lot less obvious (to me) than would be desirable...
FROGGS valgrind is still running :/ 18:39
nwc10 FROGGS: it does this :-/ 18:40
jnthn: started. No idea how far it will get, or how long it will take to get there
this isn't on "my" machine, this is on (almost) my machine
which is >5 years old and only has 2 cores
FROGGS *g*
nwc10 but I am root and I can build 18:41
wait a mo - I think I can build gcc on "my" machine
no, not enough space
would have to ask root nicely
2 cores and less RAM than my laptop (but as it's a desktop likely everything on it is faster) 18:42
FROGGS ohh, jnthn blogged and I missed it
nwc10 paste.scsys.co.uk/271683 18:43
doth go boom
jnthn ha, that's th eone 18:44
Any chance you can symbolize it?
nwc10 yes, already working on that
just checking that I can run it stand alone
then will stick it under gdb
jnthn Thanks
That looks like exactly the issue
It's probably gonna be something really silly :( 18:45
nwc10 s l o w
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 18:46
9023 nick 20 0 17.0t 921m 7568 R 100.3 46.1 2:03.47 moar
nwc10 has not seen a 't' before
jnthn: paste.scsys.co.uk/271684 18:49
I am not familiar *at all* with the source, but combining that with the previous, looks like a write by write_int32() 2 bytes over then end of an allocated 8192 byte buffer 18:50
is there a way to get tmux to capture all the technicolor glory and paste it somewhere in colour? 18:51
FROGGS jnthn: valgrind: gist.github.com/FROGGS/ce17ba8f5f06ed009c7f
jnthn hmmm...that's while writing handlers
nwc10 snap! 18:52
jnthn #define FRAME_HANDLER_SIZE 4 * 4 + 2 * 2
That looks right...
grondilu shouldn't you ut parens around this? 18:54
#define FRAME_HANDLER_SIZE (4 * 4 + 2 * 2)
nwc10 src/mast/compiler.c: FRAME_HANDLER_SIZE * fs->num_handlers);
jnthn oh f**k
nwc10 I'm going for "yes"
but I would not have thought of that
FROGGS haha 18:55
that is a nice one
grondilu remember this as a common thing to know about macros in C
nwc10 um, jnthn, that macro is defined in 2 places
grondilu always put parens
jnthn It is?
nwc10 $ git grep FRAME_HANDLER_SIZE
src/core/bytecode.c:#define FRAME_HANDLER_SIZE 4 * 4 + 2 * 2
...
src/mast/compiler.c:#define FRAME_HANDLER_SIZE 4 * 4 + 2 * 2 18:56
jnthn Thing that reads vs thing that writes
nwc10 oh yes
FROGGS rebuilds
jnthn On the one hand, could try to share them. On the other, write only cares for current version whereas read may need to care for historical versions. 18:57
FROGGS src/6model/serialization.c:14:#define HEADER_SIZE 4 * 16
nwc10 so, maybe, define it somewhere as a FRAME_HANDLER_SIZE_V1 18:58
and FRAME_HANDLER_SIZE is defined as FRAME_HANDLER_SIZE_V1 (this week)
I was wondering how to handle this
slightly hacky way might end up with the reader being a C file that can be #included more than once to compile in V1, V2
until V2 is bootstrapped
repeat for V3 18:59
or maybe that sucks too much
FROGGS src/strings/utf8.c:186:#define UTF8_MAXINC 32 * 1024 * 1024
jnthn: Perl6::Actionbuilt! 19:01
err, Perl6::Action built!
src/Perl6/Grammar.nqp 19:02
Bytecode validation error at offset 5562, instruction 1011:
operand type 64 does not match register type 56
that is my next stolperstein
nwc10 That builds for me with
FRAME_HEADER_SIZE and FRAME_HANDLER_SIZE wrapped in ()
grondilu src/mast/compiler: #define FRAME_HEADER_SIZE 7 * 4 + 3 * 2 19:03
^ should put parens here as well
nwc10 paste.scsys.co.uk/271685
19:03 cognominal joined
FROGGS grondilu: yes, four files all in all 19:03
nwc10 tries `make -k` while going for food 19:04
grondilu r: say 7 * 4 + 3 * 2 19:05
camelia rakudo e22ac2: OUTPUTĀ«34ā¤Ā»
nwc10 oh, it failed before I stood up
grondilu wonders if one can write #define FRAME_HEADER_SIZE 34 // 7 * 4 + 3 * 2 19:06
.oO( I don't remember if one can put comments in a preprocessor line)
FROGGS grondilu: there are... /* ... */ comments 19:07
dalek arVM: 4b95b95 | jnthn++ | src/ (2 files):
Add needed parens around #define sizes.

  nwc10++, grondilu++ for discovering.		 19:08
grondilu yeah but I was wondering if there is not a slight waste of resource when putting the whole arithmetic expression instead of its value
jnthn The constant folder will get it :)
That fixes the segfault, but I still get wrong bytecode out 19:09
When bulding Grammar.nqp and it loads the compiled actions: 19:10
Bytecode validation error at offset 5562, instruction 1011:
operand type 64 does not match register type 56
dalek arVM: 7f49f3f | (Tobias Leich)++ | src/ (2 files):
Add needed parens around #define sizes. (#2)
nwc10 at HEAD, ASAN now expodes like this: paste.scsys.co.uk/271686 19:37
FROGGS so, we read from nowhere? 19:39
nwc10 not quite sure
FROGGS I am running valgrind now... 19:41
valgrind++ gist.github.com/FROGGS/78ff60b293aa4e996c9b 19:44
it seems to be the exit op 19:45
jnthn nwc10: What process explodes like that? 19:52
nwc10: erm, what invocation,s orry?
Building Actions.nqp still? 19:53
nwc10 /home/nick/Sandpit/moar/bin/moar --libpath=/home/nick/Sandpit/moar/languages/nqp/lib /home/nick/Sandpit/moar/languages/nqp/lib/nqp.moarvm --target=mbc --output=blib/Perl6/Grammar.moarvm --encoding=utf8 src/Perl6/Grammar.nqp
jnthn nwc10: Does it give an error? 19:54
nwc10 Bytecode validation error at offset 5562, instruction 1011:operand type 64 does not match register type 56frame_name_36
jnthn ok, same here then
nwc10 and then ASAN aborts
well, but your terminal probably didn't mangle the newlines on pasting 19:55
I tried to run it (a second build without ASAN) under valgrind, but valgrind itself seems to SEGV
jnthn Grrr...--dump shows that there is indeed a bizare code-gen fail... 19:58
FROGGS nwc10: my valgrind out looks like output of asan fwiw 19:59
nwc10 jnthn: OK, so you can likely solve that without external help? 20:04
[Am I allowed to go to bed? :-)]
but also, it implies that MoarVM is not yet hardened against "garbage in" 20:05
jnthn nwc10: Well, thing is, the validator correctly points out it's wrong 20:06
nwc10: It then crashes *after* the validator has thrown an exception refusing to execute the malformed code.
nwc10 oh.
if I understand it correctly, "that would be bad"
jnthn Well, it'd be worse if it tried to actually run the invalid code. :) 20:07
FROGGS: It fails at exit, iiuc?
FROGGS jnthn: looks like
jnthn nwc10: Anyways, sleep is allowed ;-) 20:08
FROGGS both my valgrind and nwc's asan point to an invalid read of size 8 at interp.c line 3180
which is exit(GET_REG(cur_op, 2).i64);
so, the .i64 would fit
jnthn Yeah...But that's really odd
FROGGS yeah, I don't see anythong obvious in the declaration of the op, nor do I see a related call to exit 20:09
jnthn Well, the exit makes some sense 20:11
I think it's the one in HLL::Compiler
diakopter jnthn: you have asan 20:13
FROGGS which one of the six?
jnthn diakopter: You gave me access to a box with it 20:14
diakopter: I was gonna try that after dinner, but before I could get there others had already done it :)
nwc10 "the box", or "dinner"?
diakopter yeah I was reminding you :D
"there" 20:15
nwc10 jnthn: I susggest that you get there, as that way you can (re) validate things after I've turned in
diakopter 2>&1 |asan_symbolize 20:17
nwc10 monochrome :-(
I want symbols and colour.
pony is optional 20:18
FROGGS jnthn: you are right, if I comment out the first three exits from HLL::Compiler, it dies differently 20:24
I'd think printing to stderr followed by nqp::exit(1) would look identical 20:26
nqp: my $err := nqp::getstderr(); nqp::printfh($err, 'a'); nqp::exit(1); 20:28
camelia nqp: OUTPUTĀ«aĀ»
FROGGS nqp-m: my $err := nqp::getstderr(); nqp::printfh($err, 'a'); nqp::exit(1); # it needs a rebuild Ć³.Ć²
camelia nqp-moarvm: OUTPUTĀ«Bytecode validation error at offset 22, instruction 4:ā¤register operand index 50 out of range 0..9ā¤Ā»
FROGGS weird, it does not 20:31
nqp -e 'my $err := nqp::getstderr(); nqp::printfh($err, "a\\n"); nqp::exit(1);'
a
no error
while uv_loop_delete explodes when I strip that nqp::exit(1) 20:32
gnight 20:43
diakopter o/ 20:46
what are we diagnosing now
FROGGS diakopter: gist.github.com/FROGGS/78ff60b293aa4e996c9b 20:49
jnthn diakopter: As for me, why Actions.nqp compiles to something wrong 20:51
diakopter Perl6? alrfeady? 20:53
jnthn yeah
FROGGS there were no more bugs to explore in nqp 20:54
jnthn It ends up trying to getcode into a str reg
Validator chokes on it, naturally
--dump confirms it
--target=mast also
diakopter hunh.
only thing I can think of that could cause that 20:55
is releasing a register in the qast compiler into the wrong list
jnthn But it's meant to be an exclusive register
diakopter line no? 20:56
well yeah
it's an exclusive one sure
but it wasn't before it was returned to the list
previously
where's the line no of the getcode emission 20:57
probably typo by me in some prior line
jnthn It's one of the ones generated for closure handling in QAST comp
See clone_inner
But it's stuff I added :)
diakopter looking 20:59
line 937 21:04
could return a string register when it wants an obj
jnthn If it did, the stuff in the proto would take care of it 21:05
diakopter: It's very odd. I commented out all of actions.nqp and have been gradually uncommenting it. Reached a point where any one more method uncommented triggers the error. 21:55
And once again it looks like: 21:57
00964 getcode loc_7_str, Frame_447
00965 takeclosure loc_7_str, loc_7_str
oh my... 22:00
00948 getcode loc_255_obj, Frame_439
00949 takeclosure loc_255_obj, loc_255_obj
00950 getcode loc_0_obj, Frame_440
00951 takeclosure loc_0_obj, loc_0_obj
...
eek! 22:03
write_int16(ws->bytecode_seg, ws->bytecode_pos, (unsigned char)l->index);
char?!
dalek arVM: 5c293a8 | jnthn++ | src/mast/compiler.c:
Don't cast local/lexical indexes to 8-bit!

They're meant to be 16. Also detect the out-of-range local case.		 22:08
timotimo i'm still getting the "Routine declaration requires a signature" error even though i set an absolute path as the prefix like i did in /tmp the one time it worked 22:16
though ... the nqp checkout is *under* the moarvm folder .. can that be it? 22:17
jnthn Feels unlikely to be; feels more like a weird heisenbug...
We now compile the Perl 6 actions and grammar 22:19
timotimo dependent on the strlen of the path? >_>
jnthn timotimo: Well, something memory-layout-ish / GCish 22:20
To progress further with Rakudo on MoarVM we need to finish the extops wiring :)
grondilu hang on: nqp.moarvm can compile the Perl6 grammar and actions, now? 22:27
timotimo getting it to compile isn't terribly much more than compiling all of nqp ...
grondilu the difficult part will be the CORE.settings? 22:28
timotimo i imagine the metamodel is a bit complicated, too
grondilu ok 22:29
diakopter jnthn: how muxch mem 22:38
jnthn diakopter: didn't measure
jnthn does so 22:39
diakopter time?
jnthn Grammar.nqp tops out at 260MB 22:40
Which is the highest anything reaches
Actions is less than that.
The stuff that builds so far takes around 47s here
diakopter my goal: 5s 22:41
jnthn :)
Yeah, we've lots of room to improve yet :) 22:42
diakopter .. and then 1s
jnthn Let's focus on making it work first, though ;)
diakopter *cry*
gerd said he'd work on extops today 22:43
jnthn Well, I'll be tied up tomorrow but can help with 'em a bit also on Tuesday maybe... 22:44
I reckon by this time in a week we can have everything that comes before CORE.setting compiling.
diakopter .ask not_gerd , 22:48
yoleaux diakopter: I'll pass your message to not_gerd.
jnthn Anyway, this is fairly nice progress... 22:50
'night o/
lue I get a segfault trying to compile NQP (using Clang for this): gist.github.com/lue/6968277 22:51
diakopter oop
22:56 benabik joined
diakopter okay let's see about getting my phone to build this thing 22:58

Powered by Cro and the RakuĀ® Programming Language.

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

Thank you!