|
00:46
tokuhirom joined
00:51
colomon joined
01:51
ShimmerFairy joined
02:33
tokuhirom joined
|
|||
| timotimo | i've got the Homebrew Launcher on my new nintendo 3ds now, so i can try to port moarvm to it! :P | 02:38 | |
| timotimo tries to get sleep now | 02:39 | ||
|
07:18
FROGGS joined
07:27
leont joined
07:39
domidumont joined
07:53
domidumont joined
09:41
kjs_ joined
10:04
FROGGS joined
10:22
FROGGS joined
10:31
brrt joined
|
|||
| brrt | \o | 10:44 | |
| jnthn | o/ brrt | 10:49 | |
| brrt | \o jnthn | 10:51 | |
| i've figured out a new bug in the JIT :-) | |||
| timotimo | woo! | ||
|
10:51
tokuhirom joined
|
|||
| brrt | yes. apparantly the size of AND is not propagated to NZ | 10:52 | |
| jnthn | In even-moar-jit, I assume? :) | 10:55 | |
|
10:56
zakharyas joined
|
|||
| brrt | aye | 10:58 | |
| nwc10 | sadly AND doesn't appear to be a 3 letter code for Andorra else there might be punning potential with the NZ | ||
| brrt | sorry, no puns for you today | ||
| nwc10 | oh, excellent it is | 10:59 | |
| brrt | actually, AND not propagating a size is definitely a bug, since AND is just binary arithmetic, not condition-flagging | ||
| nwc10 | but it would need to be NZL. | ||
| brrt | in AT&T notation that's just NZ-long | ||
| :-P | |||
| jnthn: are you planning to go to FOSDEM? | 11:06 | ||
| jnthn | brrt: Not sure at the moment. Having to keep my travel down for the moment. | 11:07 | |
| brrt | i suppose that is a good idea, yes | ||
| jnthn | (For health reasons :/) | 11:08 | |
| brrt | it is understood :-) | ||
| much preferable to have a healthy jnthn in prague than a sick jnthn spread over the world | |||
| jnthn | Indeed. | 11:10 | |
| brrt | hmm.. maybe you can advice me on something though | 11:13 | |
| i kind of have two topics that I may want to present | |||
| one is about the JIT | |||
| from the perspective of how we aimed to keep it simple so that it could be implemented in a reasonable amount of time | 11:14 | ||
| I mean, JIT compilers are basically known as monsterprojects | |||
| that no sane team should embark on | |||
| but i don't think we've had that experience, and I think it'd be interesting to show why | 11:15 | ||
| on the other hand, maybe people are getting bored by the JIT | |||
| the second topic is about how tech people in general and software developers in particular rarely, if ever, thing of their work as systems, and how that's hurting us | 11:18 | ||
| so that's not about perl at all | |||
| or JITs | |||
| people pick technology before they design the system. and with technology they then force themselves into a particular system architecture, and very often that technology is, shall we say, suboptimal | 11:19 | ||
| or 'wildy and embarrasingly wrong' | |||
| do you think any of these topics is worth sharing? | 11:23 | ||
| jnthn talks plenty on a couple of the architecture courses he teaches about how we love to pick tools/languages/frameworks right at the start of a project, but that this means we're doing at the point when we're at maximum ignorance about what we're actually building | |||
| And so are in a bad place to make such choices. | 11:24 | ||
| brrt | right, that's basically what i mean :-) | 11:28 | |
| and the thing is.. in env sci. we're also very interested in technology, but we're more interested in 'what does this technology *do* with it's environment' | |||
| and so we use a pretty systematic way of evaluating the functions and features and requirements, and only then start looking at specific implementations | 11:29 | ||
| and what i've found is when applied to IT that works really very well | |||
| brb | 11:32 | ||
| jnthn | *nod* | 11:33 | |
| It's interesting, dunno how to make it fit so easily into the Perl track :) | |||
| "" | 11:35 | ||
| oops | |||
|
11:48
kjs_ joined
11:55
kjs_ joined
12:24
lucasb_ joined,
brrt joined
|
|||
| lucasb_ | I don't know how much the 3 dependencies libuv, dynasm, and dyncall are used in moarvm code. I'm just reminding that the 3 are outdated in moar's repo. I don't know if there is time to update them before xmas, or if it's even relevant to do it. | 12:24 | |
| timotimo | there is surely time before christmas to update | ||
| how far are we behind? | |||
| lucasb_ | I think libuv is the most important. I know nothing about dynasm,dyncall | 12:25 | |
| libuv is at v1.7.5; moar's libuv is 1.0 I think | |||
| brrt | hi lucasb_. libuv could well be outdated. we don't lightly undertake an update though | 12:29 | |
| timotimo | we're that far behind, huh? | 12:30 | |
| brrt | although iirc i did create a 1.6 branch some time go | ||
| ago | |||
| timotimo | didn't we recently get an update? | ||
| yeah, i remember that | |||
| i thought it had been merged | |||
| brrt | i'm not sure if that ever got merged | ||
| anyway. DynASM cannot be outdated for the simple reason that we maintain our own fork | |||
| (it's not normally a separate project) | 12:31 | ||
| dyncall, i don't know about that | |||
| iirc FROGGS did some work to 'unbundle' that for maintainers | |||
| lucasb_ | I mean MoarVM/dynasm has a newer revision MoarVM's submodule | ||
| brrt | well, don't use that newer revision | 12:32 | |
| lucasb_ | but I think is few code | ||
| oh, ok | |||
| brrt | it is the way it is for a reason :-) | ||
| lucasb_ | thanks for the explanation, brrt++ | ||
| brrt | specifically, i hacked dynasm to support dynamic addressing for the x64 extended registers. i've had many problems with that. | 12:33 | |
| lucasb_ | is it extracted from the luajit project? I see there is a LuaJIT 2.1.0-beta1 from august | ||
| brrt | yes. i don't think they have had many dynasm updates since 2.0 | ||
| i try to keep up with that, but not entirely succesfully | |||
| i think you can create a new extraction by git tree-filter or something | 12:34 | ||
| jnthn: yeah, i think you're right about that. doesn't really fit within perl | 12:39 | ||
| flussence | fwiw I've been building moar against system libuv(1.7) and libtommath for months now and it's worked fine for me | 12:40 | |
| lucasb_ | flussence: That's interesting to hear. I'll try that myself. But if I hit some problem I'll never know if it was my fault. I see that the last libtommath | 12:44 | |
| ... the last libtommath release is 0.42, but there is newer commits in its github repo | 12:45 | ||
| brrt | hmmm | 12:46 | |
| lucasb_: i would personally recommend against just checking out the latest and greatest github commits | |||
| unless there are specific bugs or issues that have been fixed | 12:47 | ||
| i mean, many of the commits in moar end up broken at the end | |||
| FROGGS | the libtommath repo contains patches made by moritz, me and others... | ||
| brrt | oh really? | 12:48 | |
| well, i didn't know that | |||
| FROGGS | yes, in their development branch | ||
| which is currently their default | |||
| lucasb_ | brrt: I think I'm a little OCD about having the lastest thing. Thanks for the advice :) | ||
| *latest | 12:49 | ||
| brrt | that is understandable :-) | 12:57 | |
| jnthn | Mmmm...chili win :) | 13:03 | |
| brrt: Yeah; the JIT would more | |||
| brrt | ok then :-) | 13:06 | |
| it's always possible to reject my proposals | |||
| or not show up | |||
|
14:08
kjs_ joined
|
|||
| brrt | it appears my nz and fix doesn't... fix it :-( | 14:45 | |
| well, it fixes the missing check, not the segv | |||
| dalek | arVM: a4c5058 | jnthn++ | src/strings/normalize.h: All Latin-1 controls are normalization terminators With the exception of \r if we're doing NFG, since we'll need that to become a grapheme if followed by \n. |
14:52 | |
| arVM: 808b772 | jnthn++ | src/strings/normalize.h: Stop fast-pathing \r when computing NFG. Since we'll need \r\n to become a single grapheme. |
|||
| jnthn | Curiously enough, a4c5058 actually busts a spectest | ||
| Looks like 'cus of a badass umption in Rakudo's Str.perl | 14:53 | ||
| Gets me down to 70 fails in grapheme-break.t from 83, though :) | 14:54 | ||
| brrt | \o/ | ||
|
14:57
tokuhirom joined
|
|||
| jnthn | Now got a patch getting me down to 57 | 15:37 | |
| dalek | arVM: cc60af3 | jnthn++ | src/strings/normalize.h: Slight clarification in a comment. |
15:44 | |
| arVM: 1ce151f | jnthn++ | src/strings/normalize.c: Full handling of Control in grapheme breaks. Covers a bunch more cases where we should break into graphemes, but wrongly did not in the past. |
|||
|
16:19
FROGGS joined
16:59
tokuhirom joined
|
|||
| dalek | arVM: 816c8b6 | jnthn++ | / (5 files): Introduce an NFG quickcheck property and use it. This plays the same role as the other normalization form quickcheck properties, but also incorporates characters that may serve as grapheme extenders despite having a CCC of zero. |
17:27 | |
| arVM: deedb4b | jnthn++ | src/strings/normalize.c: Update a comment with a note on NFG. |
|||
| nwc10 | jnthn: ASAN still likes you (or maybe it just tolerates you - I guess it's more cat-like than dog-like) | 17:33 | |
| but have not yet got to that laest pair of commits | |||
| the problem I mentioned some days ago on channel, I can't now recreate | |||
| I could reliably recreate it | |||
| no idea what caused it, or what fixed it | |||
| FROGGS | jnthn++ | 17:42 | |
| nwc10: I like to think that the relationship between ASAN and the jnthn is more like the one of Yoda and The Force™ | 17:44 | ||
| in short, ASAN believes in jnthn :o) | |||
|
18:04
kjs_ joined
18:15
leont joined
|
|||
| jnthn | Didn't actually do the most fiddly bit yet though :) | 18:27 | |
| And it remains to be seen what "\r\n".chars returning 1 (and similar for .substr) will do | 18:29 | ||
| (To the tests/ecosystem) | |||
| Anyway, food's ready :) | 18:31 | ||
| leont | Food is good, I like food! | 18:34 | |
|
18:51
tokuhirom joined
19:33
kjs_ joined
19:55
vendethiel joined
20:18
ghostin joined
20:20
kjs_ joined
20:22
ghostin left
20:34
colomon joined
20:42
rarara__ joined
20:43
lizmat joined
|
|||
| rarara__ | I would like to ask a question about MVM_string_substring: | 20:43 | |
| what is a single stranded string? | |||
| in which case we memcpy? | |||
| I'm trying to understand why Str.split is so slow; it can be either because of the simple cycle inside it, or because of substring | 20:44 | ||
| well, prob because of the cycle. | 20:48 | ||
|
20:53
tokuhirom joined
|
|||
| dalek | arVM/even-moar-jit: 47f8f08 | brrt++ | src/jit/ (4 files): Distinguish simple and end-of-block branches End-of-block branches can have a GC syncpoint, while regular branches had best not done this, given that it trips up the register allocation. |
21:14 | |
| MoarVM/even-moar-jit: 9e5d7ed | brrt++ | src/jit/x64/tiles.dasc: | |||
| MoarVM/even-moar-jit: Remove useless casting of small values | |||
| MoarVM/even-moar-jit: | |||
| MoarVM/even-moar-jit: If a cast is necessary it will be inserted by the expression | |||
| MoarVM/even-moar-jit: tree compiler; it should not be added by the code generator. | |||
|
21:32
colomon joined
21:45
zakharyas joined
21:49
kjs_ joined
21:54
Ven joined
22:00
Ven joined
22:24
kjs_ joined
22:41
Hotkeys joined
|
|||
| Hotkeys | hey | 22:41 | |
| jnthn, I'm having an issue with \r\n and such in a test and was wondering if you think it's related to the stuff you've been doing with that stuff pastebin.com/UqQWwPX4 | 22:42 | ||
| it was suggested I ask you here | 22:43 | ||
| TimToady | jnthn is hopefully in bed right now, given he lives in Prague | 22:44 | |
| Hotkeys | oh | 22:53 | |
| whoops | |||
| psch | Hotkeys: afair, having byte-less \r\n seamlessly translate to grapheme-level \n is something jnthn++ already has on his list | 22:54 | |
| err, s/less/level/ | |||
| Hotkeys | alright | 22:55 | |