|
00:20
MasterDuke joined
|
|||
| Geth | MoarVM: b03c8fdf93 | (Samantha McVey)++ | tools/UCD-download.p6 Have UCD-Download also download Shift-JIS code index |
02:04 | |
| MoarVM: 9ce054d099 | (Samantha McVey)++ | tools/generate_encoding_codetables.p6 Update generate_encoding_codetables.p6 to gen shiftjis switch For now we only generate the one used for encoding, but will later add decoding support as well which will create an array structure, since for decoding the indexes are sequential. |
|||
| samcv | yay | ||
| MoarVM: a14ecc1245 | (Samantha McVey)++ | 7 files Implement ShiftJIS encode support This implements ShiftJIS as outlined here: encoding.spec.whatwg.org/#shift_jis-encoder Encode replacements and a decoder is not implemented yet. |
|||
|
02:57
ilbot3 joined
04:01
Kaiepi joined
|
|||
| Geth | MoarVM: 5a407295f8 | (Samantha McVey)++ | src/strings/shiftjis.c Fix buffer overflow in shiftjis encode If we write two bytes to the buffer at the same time, this can overflow the buffer. Make sure to reallocate before this happens. |
05:57 | |
| MoarVM: 41711ff260 | (Samantha McVey)++ | build/Makefile.in Make sure shiftjis_codeindex.h is added to Makefile.in Otherwise if changes are made to the file, make won't recompile. |
06:00 | ||
|
06:33
brrt joined
06:50
domidumont joined
|
|||
| Geth | MoarVM: dd2072da14 | (Samantha McVey)++ | docs/ChangeLog Fix missing `:` in ChangeLog, causing website not to generate right The Perl script which generates the website's html does not detect it as a heading and causes a minor formatting error. |
06:55 | |
|
06:58
domidumont joined
06:59
AlexDaniel joined
07:40
robertle joined
08:19
domidumont joined
|
|||
| lizmat | what does it mean if I get a "Unhandled exception: const_iX NYI" when I run code with --profile ? | 08:59 | |
|
09:01
AlexDaniel joined
|
|||
| jnthn | Corruption of some sort | 10:18 | |
|
11:17
bisectable6 joined
11:43
MasterDuke joined
12:35
Util joined
13:55
SourceBaby joined
14:09
wictory[m] joined
|
|||
| timotimo | last time i looked into this it was rather confusing | 14:47 | |
| but it is on my list for sure | |||
|
15:14
domidumont joined
|
|||
| MasterDuke | jnthn: i rebased github.com/MoarVM/MoarVM/pull/689 against master, rebuilt nqp and rakudo, and all tests passed | 16:28 | |
| oddly enough there was even a passing todo in t/spec/S09-typed-arrays/native-int.t | |||
| heh, nevermind about the passing todo, that was just because of github.com/perl6/roast/commit/cabc5b3f33 | 16:38 | ||
|
16:43
AlexDaniel joined
17:36
d4l3k_ joined
17:37
synopsebot joined
17:50
domidumont joined
|
|||
| dogbert11 | .seen timotimo | 18:21 | |
| yoleaux | I saw timotimo 14:47Z in #moarvm: <timotimo> but it is on my list for sure | ||
| timotimo | ohai | 18:22 | |
| dogbert11 | hi timotimo | ||
| time for a profiler question? | |||
| this short script gist.github.com/dogbert17/f639322d...0d57c7cb15 generates an insanely large profile. Why? | 18:23 | ||
| timotimo | # This is Rakudo version 2016.03-39-gf220a5e built on MoarVM version 2016.03-46-g50c7f6a | 18:25 | |
| this could be your problem | |||
| oh, that's just old, sorry | |||
| dogbert11 | I use this script, and others, for benchmarking :) the numbers are from an old version | 18:26 | |
| it's much faster now :) | 18:27 | ||
| on my system the generated html file is 44 megs ! | 18:28 | ||
| timotimo | 41 here | ||
| somehow there's >300k call graph nodes in there | 18:29 | ||
| dogbert11 | sounds a bit much no | 18:31 | |
| I'm not saying it's a bug, just that it's a bit strange :) | 18:32 | ||
| MasterDuke | wow, i created a sql profile, but am getting an integer overflow exception when trying to run some queries | ||
| dogbert11 | uh oh | 18:34 | |
| timotimo | shouldn't be this much, no | ||
| MasterDuke | it's the exclusive_time column, i can't sum it | ||
| timotimo | did i break something, i wonder? | ||
| dogbert11: can you use some older commits and see if that changes things? | |||
| dogbert11 | how old? | ||
| timotimo | maybe a week | 18:36 | |
| dogbert11 | I'll see what I can do, is building 2017.12 atm | 18:37 | |
| timotimo | thank you | 18:39 | |
| because i just fixed something right after the last release | 18:40 | ||
| dogbert11 | timotimo: you're off the hook, 2017.12 is as bad as HEAD | 18:42 | |
| timotimo | strange, still. | 18:43 | |
| i'll have an svg soon | 18:44 | ||
| it'll be humongous, probably, but myabe i can see something | 18:45 | ||
| graphviz is still working, and i'll go afk for a significant bit | 19:06 | ||
| it'll surely finish until then. i'll also make a "minimal" version of the graph for good measure | |||
| dogbert11 | timotimo: removing a bunch lines making up $num will make the profile smaller | 19:07 | |
| timotimo: until later :) | |||
| timotimo | how much smaller? | ||
| dogbert11 | 3 megs | 19:12 | |
| MasterDuke | dogbert11: does it change anything if you re-write it to use .rotor on a .comb of $num instead of a .comb on a .substr of $num? | ||
| dogbert11 | timotimo: you can cut lines 8-25 in the gist | 19:13 | |
| MasterDuke: will check | |||
| MasterDuke: that change didn't help much gist.github.com/dogbert17/1b45cc1d...6bcf630f96 | 19:34 | ||
|
20:03
domidumont joined
|
|||
| MasterDuke | huh. does it run any faster? | 20:23 | |
| dogbert11 | roughly the same speed | 20:26 | |
| timotimo | haha, remember when i thought graphviz would be finished when i come back? | 21:46 | |
| dogbert11 | let me guess, it wasn't | 21:50 | |
| timotimo: change to this instead: my $num = '731671765313306249192251196744'; | 21:51 | ||
| timotimo | i suppose it's not faster because we don't call &infix:<~> as often? ;) | 21:52 | |
| dogbert11 | the profile will be 250k but I suspect the 'problem' will be visible anyway | ||
| perhaps that's easier for graphviz to handle :) | |||
| samcv | hmm i have this utf8 file and it seems to be changing the line endings from \r\n to \n i know that's an option but i didn't think it was the default | 21:53 | |
| dogbert11 | I could see the profile in the browser but something was seriously wrong with the call graph | ||
| samcv | or maybe i altered something | 21:54 | |
| jnthn | samcv: It's the default (and should be), and is controlled by an option passed to the decoder | 21:56 | |
| samcv: Never quite got around to wiring up exposing it on IO::Handle yet | |||
| Or maybe did...I forget :) | |||
| samcv | hah | 21:57 | |
| jnthn | (Also, should be the default for IO::Handle, but not for sockets.) | ||
| (So the latter configure the decoder differently) | |||
| samcv | i don't see translate-nl anywhere in roast | ||
| jnthn: i don't see :$translate-nl in IO::Handle.encoding | 21:59 | ||
| well there is $!decoder := $encoding.decoder(:translate-nl, :$replacement, :$strict); inside the method, but there's no translate-nl argument to control the setting | 22:00 | ||
| well that is good news. that means my shiftjis decoder works :) only issue was getting different line endings and that is just the utf8 decoder altering the text | 22:01 | ||
| though maybe that means i need to implement line ending changing in shiftjis! | |||
| hmm though it should work since i use graph = MVM_unicode_normalizer_translated_crlf(tc, &(ds->norm)); | 22:02 | ||
| MasterDuke | timotimo: not just the size of the profile is weird. it also puts all the info on the front/overview page (i.e., routines, allocations, call graph, etc.) | 22:05 | |
| samcv | ah it seems to work. great. made some changes to rakudo and i think that was causing it. forgot to pass that through | ||
| jnthn | samcv++ # shiftjis \o/ | 22:07 | |
| samcv | wonder if a compiler will optimize a switch that goes from 0 to 11103 almost seqentially into an array | 22:10 | |
| jnthn | Hopefully...I mean, in theory it does it for our interp :) | 22:11 | |
| jnthn afk a little | |||
| samcv | there's 10 gaps between those two numbers in the shiftjis switch I have now. I was going to leave the codepoint -> shiftjis index switch. since the codepoints are not at all in order and many many gaps | ||
| though then i also have to make sure that it catches an error if it hits a point where it dosen't have an index. so i may leave it how it is for now. I don't want to introduce any bugs | 22:12 | ||
| by complicating it | |||
| dogbert11 | timotimo, MasterDuke: a bizarre observation, if I run the profile under valgrind the output suddenly gets a lot smaller !?!? | 22:24 | |
| perhaps something to do with threading | 22:25 | ||
| timotimo: setting MVM_SPESH_DISABLE=1 on the original gist makes the profile go from 44 Meg to 93k | 22:28 | ||
| MasterDuke | dogbert11: what about disabling just parts of spesh? | 22:34 | |
| dogbert11 | MasterDuke: yeah, it seems to be something with inlining, setting MVM_SPESH_INLINE_DISABLE=1 also makes the profile <100k | 22:35 | |
| MasterDuke | timotimo: ^^^ seems relevant, didn't you fix a related bug recently? | 22:36 | |
| dogbert11 | 2017.12 behaves the same way, i.e. normal profile is 40+ megs while a run with inlining disabled goes down to ~90k | 22:38 | |
| Geth | MoarVM: samcv++ created pull request #820: Implement ShiftJIS decode and decodestream |
22:41 | |
|
22:46
AlexDaniel joined
|
|||