00:43 zakharyas joined 01:48 ilbot3 joined 06:16 domidumont joined 06:21 domidumont joined 07:19 domidumont joined 07:51 FROGGS joined
psch geez moar is confusing :P 07:56
also i'm clearly not awake enough yet to hack on it either
'cause now i've pushed (to my fork behind the PR, but still) a change that fails with (something like) "invalid identifier" or so 07:57
alright, i think i've got it now :l 08:09
i've pushed the adjusted error message on nqp-j already, so i'd say after the merge we bump moar & nqp rev and then i push the adjusted test 08:10
08:12 domidumont joined 08:54 domidumont joined 09:21 Dunearhp joined
jnthn wonders what's confusing about moar... :) 09:38
dalek arVM: cbb1178 | peschwa++ | src/6model/reprs/CStruct.c:
Die on CStruct without any fields.
arVM: 98975c8 | peschwa++ | src/6model/reprs/CStruct.c:
Adjust error message for zero size CStruct, jnthn++.
arVM: 174fc71 | peschwa++ | src/6model/reprs/CStruct.c:
Pass the STable into compute_allocation_strategy.

Otherwise, well, it's not there and we cannot ask it for the debug_name.
arVM: 6b8fa2e | jnthn++ | src/6model/reprs/CStruct.c:
Merge pull request #416 from peschwa/master

0672713 | cygx++ | src/io/syncfile.c: make MVM_file_open_fh() throw if the file we opened was a directory
arVM: 727114d | cygx++ | src/io/syncfile.c:
Document why it's ok to ignore fstat errors in MVM_file_open_fh()
arVM: fb40d63 | cygx++ | src/io/syncfile.c:
Fix bug and memory leaks in MVM_file_open_fh()
arVM: 0dce897 | jnthn++ | src/io/syncfile.c:
Merge pull request #406 from cygx/fail-open-if-dir

8726735 | labster++ | build/probe.pm: Add error message for likely macOS build failure (that I personally experienced)
arVM: 34c375a | jnthn++ | build/probe.pm:
Merge pull request #408 from labster/master

Add error message for likely macOS build failure
10:05 travis-ci joined
travis-ci MoarVM build failed. Jonathan Worthington 'Merge pull request #416 from peschwa/master 10:05
travis-ci.org/MoarVM/MoarVM/builds/164237184 github.com/MoarVM/MoarVM/compare/9...8fa2ed3a27
10:05 travis-ci left
psch --no-jit and no --has-libffi (which i assume is also --no--libffi..?) 10:06
okay i don't think i understand what travis tests there..? 10:09
'cause, well, it creates install-jvm-runner.pl and then dies because of an too old moar version..? 10:10
10:10 domidumont joined
psch ohh, it builds nqp 10:10
right, moar doesn't have tests
10:21 travis-ci joined
travis-ci MoarVM build failed. Jonathan Worthington 'Merge pull request #406 from cygx/fail-open-if-dir 10:21
travis-ci.org/MoarVM/MoarVM/builds/164237500 github.com/MoarVM/MoarVM/compare/6...ce8974453b
10:21 travis-ci left 10:55 zakharyas joined 11:43 btyler joined 11:52 domidumont joined 12:23 Ven_ joined
dalek MoarVM: 8e92125 | (Dominique Dumont)++ | Configure.pl: 13:58
MoarVM: Fix build without libtommath source
MoarVM: Debian policy requires (as far as possible) to remove convenience
MoarVM: copy of source code before build.
13:59 dalek joined 15:31 Ven_ joined 15:45 Ven_ joined 16:05 domidumont joined 16:56 btyler joined 17:15 brrt joined 17:45 btyler joined
brrt good * #moarvm 17:45
\o btyler
btyler hiya brrt \o 17:46
brrt is finally progressing on the JIT by deciding not to care about some design things 17:47
hmm, just had an idea… let me put it here before it slips my mind 17:49
optimistic store removal can be a completely local thing 17:50
(in which case, it doesn't help much, but okay)
in which case, the only 'natural storage position' for a value would be the last
e.g.: suppose we have a sequence of LOAD, COPY, (equalling SET), another COPY, and then a STORE 17:51
if the definition of the first COPY would be the last one, the STORE would be inserted anyway; but it'd be equally safe just to store it in the final place 17:52
(in which case… *drumroll* we can detect-and-eliminate that double store
not sure if this is anything but hypothetical, but still a nice simplication 17:53
trainswitch &
17:54 btyler joined 18:03 FROGGS joined 18:09 brrt joined
brrt anyway… any plans for LPW? 18:10
(just yesterday i finally installed Fedora 24. And I finally got rid of selinux. I wonder which one contributes most to the thing feeling much faster 18:14
dalek arVM/even-moar-jit: c3fd80c | brrt++ | src/jit/linear_scan.c:
Implement naive spilling and splitting

The actual logic of the 'spill' decision needs some work because it needs to decide whether or not to split a range in the first place. Also, splits should probably be rangeable (which means that the split logic may be more complicated).
And it needs the mysterious function 'create_new_live_range' and 'get_storage_location', which probably aren't really very complicated but I don't want to think about them right now :-)
brrt i wonder how hard it'd be to have a minor program that links moar, builds an expr tree (from a file?), compiles and runs it 18:47
possibly the parsing bit is hardest, but, sexpr
timotimo why not write it in nqp? 18:48
brrt because NQP is itself jitted
timotimo ah, you don't want that 18:49
brrt or, you mean, the parser
i want some more fine-grained control for it, anyway
timotimo well, moar will only really run stuff it has bytecode for ...
brrt nqp::jitdisable(), nqp::jitenable(), slt
timotimo sounds sensible
and easy to do, i guess?
brrt not sure
timotimo though i'm not sure if we rely on that flag not changing :D
brrt yeah, i wonder how that interoperates with speshing 18:50
which is kind of asynchronous from the viewpoint of the user
timotimo right
timotimo BBL 18:51
brrt (furthermore, it has clear action-at-a-distance for mutlithreaded applications, unless we bind the flag to threadcontext)
see you :-)
timotimo right, the flag is in the instance right now 18:52
brrt which… is really easy to change, but also, probably not worth it 18:53
19:41 Ven_ joined 21:08 Ven_ joined
timotimo i think we could also trigger a major collection when the number of gen2 roots has increased a noticable amount 21:11
21:28 Ven_ joined 21:38 Ven_ joined 22:18 Ven_ joined 23:05 dalek joined