| AlexDaniel` woke up | 01:14 | ||
| Zoffix | #woke | 01:17 | |
|
01:56
ilbot3 joined
04:25
AlexDani` joined
04:34
coverable6 joined,
bloatable6 joined,
nativecallable6 joined,
quotable6 joined,
bisectable6 joined,
committable6 joined,
evalable6 joined,
releasable6 joined,
greppable6 joined,
unicodable6 joined,
statisfiable6 joined,
benchable6 joined,
squashable6 joined
04:38
evalable6 joined,
bisectable6 joined,
unicodable6 joined,
committable6 joined,
benchable6 joined,
squashable6 joined
04:41
releasable6 joined
04:54
bloatable6 joined,
quotable6 joined,
nativecallable6 joined,
coverable6 joined,
statisfiable6 joined,
greppable6 joined
05:14
bloatable6 joined,
committable6 joined,
coverable6 joined,
evalable6 joined,
greppable6 joined,
quotable6 joined,
nativecallable6 joined,
benchable6 joined,
bisectable6 joined,
unicodable6 joined,
releasable6 joined,
squashable6 joined,
statisfiable6 joined
07:08
domidumont joined
07:17
domidumont joined
08:26
robertle joined
08:57
patrickz joined
09:20
ggoebel joined
09:24
dogbert17 joined
09:35
rba_ joined
09:46
rba joined
10:02
rba_ joined
10:17
rba joined
|
|||
| Geth | MoarVM/jit_indexicim_ops: 00641aecdc | (Timo Paulssen)++ | src/jit/graph.c jit indexic, indexicim, indexim ops |
10:27 | |
|
10:31
rba_ joined
|
|||
| AlexDaniel` | does anybody know what's the point of including address family in the error message? | 10:58 | |
| github.com/MoarVM/MoarVM/commit/d4...cd6492L298 | |||
| c: 9dba498f7f^,9dba498f7f gist.githubusercontent.com/AlexDan...thingie.p6 | |||
| committable6 | AlexDaniel`, Successfully fetched the code from the provided URL. | ||
| AlexDaniel`, gist.github.com/bd63f6676168a39b60...11c377ec25 | |||
| AlexDaniel` | “Failed to resolve host name 'perl6.or' with family 0. Error: 'Name or service not known'” | ||
| 0 is AF_UNSPEC I think | 10:59 | ||
| .seen eater | |||
| yoleaux | I saw eater 12 Oct 2017 08:22Z in #perl6: <eater> :D | ||
| MasterDuke | so you know what family was used if you didn't specify it? | ||
| AlexDaniel` | so if I didn't specify it, it used unspecified family. Helpful | 11:00 | |
| MasterDuke | timotimo: i tried to perf record parsing the setting, which used to generate a several Gb perf.data. now it got to 231Gb and then died because it filled the disk | 11:02 | |
| any idea what has changed that would cause that behavior? | |||
| AlexDaniel` | m: perl6.or | 11:03 | |
| camelia | ===SORRY!=== Error while compiling <tmp> Confused at <tmp>:1 ------> https:⏏//perl6.or expecting any of: colon pair |
||
| AlexDaniel` | e: perl6.or | ||
| evalable6 | AlexDaniel`, It looks like a URL, but for some reason I cannot download it (Failed to resolve host name 'perl6.or' with family 0. Error: 'Name or service not known') | ||
|
11:03
eater joined
|
|||
| AlexDaniel` | e: perl6.or | 11:04 | |
| evalable6 | AlexDaniel`, It looks like a URL, but for some reason I cannot download it (Failed to resolve host name 'perl6.or' with family 0. Error: 'Name or service not known') | ||
| AlexDaniel` | eater: was it intended to be like this? | ||
| i.e. print a numeric value of a family (which is not directly useful) | |||
| especially when it's AF_UNSPEC? | |||
| eater | there is currently also support for AF_UNIX in there | ||
| so it might be useful if it showed also which family was used | 11:05 | ||
| and ofcourse you can explicitly say IPv6 or IPv4 | |||
| it may fail finding an AAAA | |||
| but there might be an A record | |||
| AF_UNIX doesn't matter here I just realize | |||
| AlexDaniel` | hm | 11:06 | |
| eater | I mostly try to make my error messages as verbose as possible so it's easy to find the problem | 11:09 | |
| AlexDaniel` | eater: any chance of having some string there instead of a number? | 11:11 | |
| or are these numbers so well known that this is not needed? I have no idea actually | 11:12 | ||
| eater | not really I guess | ||
| but that would mean an conversion table? | |||
| AlexDaniel` | eater: thing is, my first impression was “huh, somebody left some debug info in there…” | 11:13 | |
| now I'm thinking that it is simply LTA | |||
| eater | LTA? | ||
| MasterDuke | is there any pre-existing function like uv_strerror to turn those numbers into names? | 11:14 | |
| AlexDaniel` | less than awesome | ||
| eater | :') | ||
| that can mean alot of things | 11:15 | ||
| jnthn | Probably worth trying to map it into a string, I guess just a switch/case and fall back to "family N" if it's not one we have a mapping for | 11:17 | |
| eater | I guess right now that would be the best option ye | ||
| jnthn | Agree it's probably useful info to have | ||
| samcv | hey eater | 11:18 | |
| eater | currently busy with a catastrophic ZF | ||
| S | |||
| recovery | |||
| will pick it up after that | |||
| samcv: o/ | |||
| jnthn | Sure, it's hardly urgent :) | ||
| Good luck with the ZFS recovery | 11:19 | ||
| MasterDuke | jnthn: did you see my question to timotimo about perf? would you have any insight? | ||
| eater | thanks :) | ||
| jnthn | MasterDuke: No, didn't spot that | 11:21 | |
| timotimo | MasterDuke: you can just reduce the frequency of recording | ||
| jnthn | And don't know perf super well :) | ||
| timotimo | it's either -f or -F | ||
| MasterDuke | yeah, but i never had to explicitly set it before | 11:22 | |
| trying now with -F 100 (default is 1000) | 11:24 | ||
| AlexDaniel` | eater: ok, thanks. Here's a ticket: RT #132335 | 11:25 | |
| synopsebot | RT#132335 [new]: rt.perl.org/Ticket/Display.html?id=132335 [LTA] pure numeric values of address families are not useful enough ( IO::Socket::INET.new(:host<perl6.o>, :port(42)) ) | ||
| timotimo | MasterDuke: amount of data recorded also depends on stack size. i do believe it has a default upper limit for how much of the stack it'll grab, but maybe that's a big part of what makes the file big. we don't have very deep stacks, though? | 11:26 | |
| AlexDaniel` | by the way, is there any way to specify the family from perl6? | ||
| MasterDuke | 8k is the default it'll capture | 11:27 | |
| AlexDaniel` | I'm thinking, maybe we should have family enum available to users? | ||
| AlexDaniel` shrugs | |||
| timotimo | OK, so 8k a thousand times per second (what about multiple threads though?) for how long? | 11:28 | |
| that's almost 8 megs per second, but 231 gigs is what 100000x that? | |||
| MasterDuke | could be that it's generating data at the same rate as before, it just takes much longer | 11:32 | |
| timotimo | if it were sampling too fast to write to disk it'd truncate some events i believe | 11:33 | |
|
11:36
evalable6 joined
|
|||
| MasterDuke | sure, all sorts of stuff could be going on. it's just that before all i'd do it put `perf record -g --call-graph dwarf` before the command to build the setting, wait a little bit longer than a regular build took, and i had a report i could use | 11:38 | |
| timotimo | yeah | 12:00 | |
| i'd expect that, too | |||
|
12:03
Voldenet joined
|
|||
| MasterDuke | `MoarVM panic: Memory allocation failed; could not allocate 47598128 bytes` | 12:10 | |
| on my desktop with 8g ram. running with -F 100. never had that happen before either | |||
|
12:15
Voldenet joined
|
|||
| timotimo | well, normally that's a hll-level stack overflow | 12:17 | |
| MasterDuke | trying now with MVM_SPESH_DISABLE=1 | ||
|
12:20
lizmat joined
12:25
quotable6 joined
12:40
Voldenet joined
12:49
quotable6 joined
|
|||
| MasterDuke | [07:24] <MasterDuke> trying now with -F 100 (default is 1000) # have a 6G file 2h later and still going | 13:31 | |
|
13:34
rba joined
|
|||
| timotimo | why is it doing that :\ | 13:36 | |
| does it properly run to completion without perf record? | |||
| how exatcly are you invoking it? just what you wrote before? | |||
| -g --call-graph dwarf /path/to/moar --libdir=blah /path/to/perl6.moarvm --target=... ? | 13:37 | ||
|
13:46
rba_ joined
|
|||
| MasterDuke | yes to both | 13:47 | |
| gist.github.com/MasterDuke17/acfe6...b018307277 finally just killed it and ran perf report on what had been recorded | 13:49 | ||
| timotimo | that's weird | ||
| you don't have a mini-nursery or something per chance? | |||
| MasterDuke | `perl Configure.pl --gen-moar && make`, then ctrl-c when it started building the setting and re-ran just that command with perf | 13:51 | |
| timotimo | weird! | ||
| in theory you could grab a telemetry log to see what the difference between "start of gc to end of gc" and "end of gc to start of gc" is | 13:52 | ||
| and if it's about a 90-10 split | |||
| MasterDuke | oh, hm...just tried re-running without perf and now *that* still hasn't finished... | ||
| timotimo | well, that's certainly not right | 13:55 | |
| MasterDuke | seems like it might be setting --target=parse | 14:01 | |
|
14:01
rba joined
|
|||
| timotimo | oh!! | 14:02 | |
| it's probably trying to dump the thing! | |||
| MasterDuke | ? that used to work | 14:03 | |
| timotimo | hmm | 14:04 | |
| MasterDuke | yup, finished in just ~20s longer than usual with --target=mbc, 3.7G file | 14:06 | |
| updated gist with its report | 14:07 | ||
| timotimo | i've gotta go now, though | 14:08 | |
| i'm glad we figured this out | |||
| MasterDuke | and that was without setting -F 100 | ||
| later... | |||
| timotimo | once TT's optimization to push info from nfa runs down lands we'll get less time spent in nqp_nfa_run | 14:09 | |
| i wonder if the dynamic var cache needs tuning again (that's what influences find_contextual_by_name i expect) | 14:10 | ||
| spesh_arg_guard_run could decrease once we jit those | |||
|
14:16
rba_ joined
14:31
rba joined
14:39
eater joined
14:47
rba_ joined
|
|||
| MasterDuke | how do i read the dynvar log? setting MVM_DYNVAR_LOG=dynvar.log while building the setting resulted in a 7.8M line file | 14:49 | |
|
14:49
eater joined
|
|||
| MasterDuke | N = not found, F = found, C = cached ? | 14:49 | |
| 26119 I, 110328 N, 3791437 C, 3914666 F | 14:53 | ||
|
14:57
eater joined
|
|||
| MasterDuke | some quick stats from the log gist.github.com/MasterDuke17/b9654...f9b30d458b | 14:58 | |
| timotimo | it's supposed to also have info about how deep it had to search to find something | 16:05 | |
| i'm sure TT had a tool to evaluate the raw data | |||
| MasterDuke | yeah, there are a whole bunch of numbers also | 16:15 | |
|
16:48
domidumont joined
|
|||
| MasterDuke | some quick stats from a jit log of building the setting gist.github.com/MasterDuke17/75dd8...71581ebefc | 17:06 | |
|
17:17
zakharyas joined
17:41
committable6 joined
17:45
evalable6 joined,
bisectable6 joined
17:48
committable6 joined
|
|||
| Geth | MoarVM: MasterDuke17++ created pull request #729: JIT the getexcategory op |
17:58 | |
| MoarVM: MasterDuke17++ created pull request #730: Jit param_rp_s op |
19:19 | ||
|
19:41
cognominal joined
|
|||
| cognominal | AlexDaniel, about irclog.perlgeek.de/moarvm/2017-10-21#i_15334051 it seems that hiding everything about the resolver API in high level API like connect is less than ideal. Verbose messages may be a good idea github.com/MoarVM/MoarVM/issues/683 | 19:48 | |
|
20:35
zakharyas joined
20:39
zakharyas joined
|
|||
| samcv | going to start making the changelog now | 23:00 | |
| jnthn | samcv++ :) | 23:01 | |
| samcv: Anything more you need from me on release stuff? | |||
| I'll be around for the next ~45 mins. I may get chance to pop by tomorrow before my train, or maybe not. | |||
| samcv | the release guide is pretty clear i think | ||
| jnthn | OK, cool :) | 23:02 | |
|
23:32
patrickz_ joined
|
|||