github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
nwc10 good *, #moarvm 10:39
jnthn: would you like the good news, and brrt the not so good news? :-)
jnthn Well, this seems like a better deal for me than brrt++... :) 10:54
nwc10 jnthn: I can get to the spectest again with +#define MVM_SPESH_CHECK_DU 1 10:55
no idea what changed when, but something you (singular or plural) did fixed the underlying bug that that had hit
and for brrt (and everyone else I guess), the --ubsan option to Configure is somewhat Less Awesome than it ideally should be, because (I forget which bit of 3rd party code) the JIT headers do undefined behaviour, so it always spits out pretty warnings 10:57
which messes up spectests that check output
(can't trivially check right now, as I'm spectesting again)
but it's some header with x86 in the filename, so it knows what it expects
but strictly it's undefined behaviour
dogbert2_ hello brrt 13:22
brrt ohai dogbert2_ 13:27
nwc10: infopls
dogbert2_ brrt: trying to build MoarVM with your branch expr-jit-invoke failed for me. Got the message 'make: *** No rule to make target 'src/jit/nativecall.h', needed by 'src/main.o'. Stop.' 13:28
brrt did you make reconfig
I might have not committed that file
dogbert2_ timotimo suspects something might be missing in makefile.in
Geth MoarVM/jit-expr-optimizer: 4 commits pushed by (Bart Wiegmans)++ 13:29
brrt maybe, let me have a check 13:29
Geth MoarVM/expr-jit-invoke: a9a46d975d | (Bart Wiegmans)++ | 4 files
[JIT] Implement invoke sequence in expr JIT
 13:31
MoarVM/expr-jit-invoke: e546eb1a23 | (Bart Wiegmans)++ | src/jit/graph.c
[JIT] Eliminate special consume_invoke opcode

Except for nativecall_invoke, we no longer need this part of the legacy JIT.
MoarVM/expr-jit-invoke: f562106d02 | (Bart Wiegmans)++ | 7 files
[JIT] WIP on nativeinvoke in the expression JIT
brrt dogbert2_: please try again? 13:34
nwc10 brrt: with --ubsan 13:41
3rdparty/dynasm/minilua.c:1609:4: runtime error: signed integer overflow: -2147483648 + -1 cannot be represented in type 'int'
oops
3rdparty/dynasm/minilua.c:1609:4: runtime error: signed integer overflow: -2147483648 + -1 cannot be represented in type 'int'
there. that's the (minimalist) pavement pizza line 13:42
and that appears in the spectests
so it confuses things
and I'm misremembered where it came from
brrt hmmm. I see 13:44
but.... we don't link minilua?
we do link dasm_x86.h
and I'm ready to believe there's UB there
dogbert2_ brrt++ it builds now. have to wait until I get home before I can test it though 13:56
nwc10 oh, not totally misrembmbering 14:04
Stage parse : 3rdparty/dynasm/dasm_x86.h:224:16: runtime error: signed integer overflow: 2147483647 + 128 cannot be represented in type 'int'
3rdparty/dynasm/dasm_x86.h:590:13: runtime error: signed integer overflow: 2147483647 + 128 cannot be represented in type 'int'
(I spared you the bold and the red)
so that's a second place
that one is quite common 14:13
I'm seing it slowly through the build
Geth MoarVM: 304f568f19 | (Jonathan Worthington)++ | 3 files
Correct deserialization handling of parametrics

It's fine to poke the interned value into the target SC, however we must not set the SC index of the current SC onto it. Otherwise, we get a bogus index, and can end up looking up a completely bogus object.
Addresses github.com/rakudo/rakudo/issues/2345.		 14:33
timotimo whoa. 14:37
jnthn That took a bit of finding 14:38
timotimo now i don't have to have two versions of perl6 installed any more to both have my newest profiler output stuff and a working cro :) 14:40
jnthn :) 14:43
Yeah, I've had to run with the september release this week on $dayjob app too :)
(Verified it runs nicely on HEAD after that fix)
brrt jnthn++ 14:52
dogbert2_ with some luck there are no blockers left 14:53
brrt nwc10 - noted
hmm, ubsan is probably right, but I'm not sure I see the problem in practice 14:56
I'll have to think up a formulation in which this works

Powered by Cro and the Raku® Programming Language.

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

Thank you!