| [Coke] | timotimo: so knowing where this memory thing is happening, how can I printf the MVMstr in code? I'l recompile and see if that shows me the string. | 00:00 | |
| timotimo | oh | ||
| printf'ing the MVMString's contents is a bit roundabout | |||
| [Coke] | nothing about this bug is straightforward. :) | ||
| timotimo | it could very well be that the pointer has become garbage | ||
| if not, you should look if the storage type is "32bit graphemes" or "rope" (which is quite unlikely, but possible) | 00:01 | ||
| jnthn | [Coke]: Thanks for getting it down to being about MVMString. That helps a lot (in terms of reducing surface area). | 00:15 | |
| timotimo | jnthn: i just rewrote prefix:<!> and prefix:<not> for Bool \a to not have a jump and not load 0 and 1 as wval and decont + istrue them ... | 00:17 | |
| but can we figure something like that out in spesh? gist.github.com/timo/de0ef5d3f5e159117d2b | |||
| jnthn | timotimo: That sounds...like an improvement :) | ||
| timotimo | the only thing i could think of is duplicating the last block, pasting it onto the two conditional blocks and then being able to constant-fold istrue on the wval'd objects | ||
| jnthn | Guy helps land probe on comet. In response, gets reduced to tears because of his choice of *t-shirt*. WTF is wrong with our world... :/ | 00:19 | |
| timotimo: Yeah, it's not clear to me how to do it... | |||
| timotimo | it'll be re-writing then :) | 00:20 | |
| jnthn | timotimo: It's some kinda backwards propagation issue | ||
| timotimo | yeah | ||
| jnthn | Like, we need to know what something gets used for... | ||
| timotimo | i'd kind of like to have a pseudo-register in BBs that get called from if or unless that is set to 0 or 1 in the branch and eliminate constants and other things that way | ||
| new kernel, bbiab :) | 00:21 | ||
| jnthn | :) | ||
| timotimo | the thing i was hoping to get fixed by a newer kernel stayed broken :\ | 00:23 | |
| jnthn | aww | 00:24 | |
| japhb | jnthn: link re: t-shirt issue? | 00:25 | |
| timotimo | japhb: one of the people was wearing a hawaii shirt with ladies wearing bikinis | 00:26 | |
| jnthn | japhb: www.bbc.com/news/blogs-trending-30055278 | 00:27 | |
| timotimo | okay, that's a bit more ... risquè than just "bikini-clad ladies" | 00:29 | |
| japhb | Sigh. | ||
| japhb has feelpinions on this subject, but isn't in the mood to rant | 00:31 | ||
| jnthn | I've always found blame culture (X went wrong, QUICK LAWSUIT NOW!) ugly, and on this one I'm just sad at the pounce-on-maybe-mistakes culture that leads to a day when somebody should have been celebrating a huge scientific success instead being one they'll remember as "the day I shoulda picked another shirt" :( | 00:35 | |
|
00:38
lizmat joined
|
|||
| JimmyZ | Good morning, #moarvm | 00:40 | |
| jnthn | Oh my, it's morning already... :) | ||
| JimmyZ | 02:00 pm there ? | ||
| *am | |||
| jnthn | am :) | ||
| Yeah | |||
| Home later than expected sicne they cancelled all the trains... | |||
| JimmyZ | Oh | 00:41 | |
| timotimo | it's definitely not like he's set back women in science by a decade by wearing that shirt; it was possibly not anything he even thought about | ||
| i disagree with placing blame on him | |||
| it's more or less an expression of the general "mood"/"culture" | 00:42 | ||
| a symptom, if you like | |||
| japhb | I am personally in the camp that feels "No matter how cool your job is, work is not home -- you don't get to be your full self at work, there's a higher standard expected." I also think it's pretty clear this was a mistake, not intentional -- but if he wears that often to work, what standard is he setting for office environment? | 00:50 | |
| timotimo | right | ||
| nobody dared criticise him for his ladies-shirt | |||
| jnthn | It's sure as heck not a shirt I'd buy, let alone feel comfortable wearning anywhere, let alone work, fwiw. :) But yeah, I think it was almost certainly unintentional, and the internets have dealt their usual nuanceless harsh response. | 00:59 | |
| timotimo | oh, there's nuance in there. you just can't see it between all the rage | 01:00 | |
| jnthn | *nod* | ||
| timotimo | oh hey | ||
| can we somehow spesh "$*OUT is almost never going to change"? | 01:01 | ||
| it seems like looking that up takes a big chunk of time when we do i/o-heavy benchmarks | |||
| we already cache dynamic variables in frames somewhere, but it doesn't seem to be enough? or something? | |||
| jnthn | Yeah, we do cache those kinds of lookup | 01:03 | |
| And if it's the same string literal used int he lookup that should be cheap... | |||
| timotimo | you mean '$*OUT'? | 01:04 | |
| jnthn | Yes | ||
| timotimo | that should come from the serialization blob | ||
| jnthn | Well, the string lit comes from the string heap | 01:05 | |
| timotimo | that one | ||
| yeah | |||
| i hope we cache it after unboxing :) | |||
| japhb | I thought TimToady++ had experimentally shown that the current caching scheme for dynamics was insufficient (average frame distance too high) | ||
| jnthn | I don't think the code-gen boxes it? | ||
| timotimo | oh? | ||
| jnthn: i shall have a look at some spesh log output | |||
| segfault :) | |||
| jnthn | timotimo: oh, though... | 01:06 | |
| Perl 6 lookups of those end up being calls to DYNAMIC | |||
| TimToady | well, it's not yet optimal, I think | ||
| jnthn | And we fail over to PROCESS for $*OUT... | ||
| So that misses all the caching entirely :( | |||
| TimToady | but we got the average frame search from 20+ down to under 5 | ||
| jnthn | It's almost as if we'd like to be able to cache "I sure as heck didn't find this symbol beneath this point" too | ||
| timotimo | ooooh | ||
| jnthn | If we can't already | ||
| timotimo | that could explain it i suppose? | 01:07 | |
| jnthn | I found that with the method cache fwiw | ||
| We used to waste a load of time doing MOP fallback stuff in nqp::can on the actions class for things like ws, which had no action method | |||
| Once I made the method cache authoritative so we knew a miss in it meant "there really ain't one", we got a decent win. | |||
| timotimo | "OUT" doesn't even show up in the spesh log | ||
| perhaps it's *that* bad | |||
| that it doesn't even get spesh'd | 01:08 | ||
| TimToady | we could do better caching if we allocated space for it in the pad | ||
| especially for known read-only values | 01:09 | ||
| otoh, single use caching isn't very useful, so you really want to cache up a level or two from the singletons | |||
| jnthn | *nod* | 01:10 | |
| TimToady | iirc the current scheme only gets desperate and starts replacing cache entries if it finds a really long lookup | ||
| and it's also a bit ad hoc; since we have to look in each lexpad anyway, it kinda makes sense to cache values there instead of using a special frame slot | 01:11 | ||
| and in that case we could be smarter about usage patterns, since we could cache multiple things at any level | 01:12 | ||
| but that also potentially makes our lexpads not quite readonly at runtime, which is the downside | |||
| so it's not all win | |||
| you probably shouldn't trust much of what I say right now, since I'm coming down with Yet Another Cold, courtesy of family get-together-ness | 01:13 | ||
| jnthn | Urgh | ||
| Meanwhile, I'm just about over jetlag, and am flying longhaul again in a couple of days... | 01:14 | ||
| TimToady has been working on the longest-literal stuff, which turned out not to need a rebootstrap | |||
| but has other issues | |||
| currently fails a bunch of tests which may be an inordinate reliance on ordered rules in the grammar, or maybe just bugs | 01:15 | ||
| getting lit len to propagate up and down the nfa derivations is...fraught... | |||
| japhb | jnthn: Where to this time? | 01:16 | |
| jnthn | japhb: South China | ||
| Just for several days. | 01:17 | ||
| (Work-related, not vacation this time. :)) | |||
| japhb | More from the same place you and masak spent a month? | 01:18 | |
| jnthn | Aye | ||
| masak comes to, so at least I've got somebody to tell the terrible puns I come up with during the plane trip to... :) | 01:19 | ||
| *too | |||
| japhb wonders how a Swedish training company gets serious business in South China | |||
| jnthn | I dunno. :) Sales is a black box that we feed the sort of stuff we like to do into, and offers of work for us to accept or run screaming from come out of. :) | 01:23 | |
| japhb | Heh | 01:25 | |
| jnthn gets some rest | 01:41 | ||
| [Coke] | good sales people are a wonder. I'm kind of happy I ended up in an industry where I don't have to rely on sales. :) | 01:42 | |
|
03:15
synopsebot joined
09:51
JimmyZ joined
11:04
ggoebel111111111 joined
14:24
camelia joined
14:30
woolfy joined
16:19
zakharyas joined
|
|||
| nebuchadnezzar | hello | 17:27 | |
| maybe the travis build matrix could be updated? travis-ci.org/MoarVM/MoarVM/jobs/41102955#L155 | |||
| timotimo | ah | 17:29 | |
| yeah, could | |||
| fortunately it's entirely safe | |||
|
17:51
FROGGS_ joined
18:48
BinGOs joined
18:54
carlin joined
18:55
woolfy joined
18:59
hoelzro_ joined
19:01
Juerd joined
19:21
nebuchad` joined
19:22
leedo_ joined
19:25
nine joined
19:26
timotimo joined
19:53
carlin joined
20:00
Juerd_ joined,
jnthn_ joined,
Util_ joined
20:11
[Coke]_ joined,
lue joined
20:23
oetiker joined
20:49
dalek joined
21:51
ggoebel111111113 joined
22:48
woolfy1 joined
22:56
lizmat joined
23:08
colomon joined
|
|||