|
01:36
ggoebel joined
01:59
ilbot3 joined
02:35
AlexDani` joined
02:52
greppable6 joined
05:35
yoleaux joined,
ggoebel joined
05:44
domidumont joined
05:47
lizmat joined
05:50
domidumont joined
05:56
brrt joined
|
|||
| brrt | good * #moarvm | 06:11 | |
| yoleaux | 16 Oct 2017 19:09Z <AlexDaniel`> brrt: Hello. Any news on RT #132269 ? | ||
| synopsebot | RT#132269 [new]: rt.perl.org/Ticket/Display.html?id=132269 [REGRESSION][BUG] JIT removing loop construct and confusing last() | ||
| brrt | .tell AlexDaniel not yet | ||
| yoleaux | brrt: I'll pass your message to AlexDaniel. | ||
| AlexDaniel | Ok | 06:32 | |
| yoleaux | 06:11Z <brrt> AlexDaniel: not yet | ||
|
06:41
brrt joined
06:52
brrt joined
|
|||
| brrt | it's a fun one that one | 06:53 | |
| requires network access, so no easy debugging | |||
| multiple threads, so bisect doesn't quite work | |||
|
07:04
domidumont joined
|
|||
| brrt | oh, and the exception isn't an adhoc one | 07:04 | |
|
07:08
patrickz joined
|
|||
| brrt | good news | 07:46 | |
| under jit-dump.pl, the bisect works with one frame offset | 07:47 | ||
|
07:48
rba joined
07:56
domidumont joined
08:04
rba_ joined
08:05
rba__ joined
08:26
robertle joined
08:28
rba joined
|
|||
| brrt | the generated code, that appears to be correct | 08:33 | |
| i'm a little surprised not to see the idx->const thing take effect yet | |||
| however, i'm also seeing a missing label | 08:36 | ||
| or what i think is a label that's missing | |||
|
08:40
zakharyas joined
|
|||
| samcv | hey brrt | 08:47 | |
| brrt | the JIT log is unfortunately pretty silent about what it is that we're missing | ||
| hey samcv | |||
| samcv | buggable, tags | 08:50 | |
| buggable | samcv, Total: 1629; 6.D: 2; 9999: 9; @LARRY: 28; ANNOYING: 8; BOOTSTRAP: 4; BUG: 592; BUILD: 12; CONC: 44; DOCS: 1; EXOTICTEST: 3; FLAP: 1; GLR: 3; IO: 20; JVM: 48; LHF: 7; LTA: 176; MATH: 5; META: 2; MOAR: 2; MOLD: 233; NATIVECALL: 21; NYI: 57; OO: 13; OPTIMIZER: 8; OSX: 2; PARSER: 5; PERF: 27; POD: 19; PRECOMP: 15; REGEX: 46; REGRESSION: | ||
| samcv, 38; REPL: 6; RFC: 61; RT: 2; SECURITY: 2; SEGV: 28; SINK: 1; SITE: 1; SPESH: 1; STAR: 7; TESTCOMMITTED: 12; TESTNEEDED: 35; TODO: 13; UNI: 27; UNTAGGED: 282; WEIRD: 2; WINDOWS: 4; See fail.rakudo.party/ for details | |||
|
09:33
AlexDani` joined
09:57
rba joined
09:58
releasable6 joined
10:14
rba_ joined
10:27
patrickz joined
10:49
timo joined
11:01
brrt joined
11:08
rba joined
|
|||
| brrt | so, i'm missing a throwish control guard | 12:10 | |
| now to figure out why i'm missing it | |||
|
12:21
zakharyas joined
12:42
rba joined
12:54
rba_ joined
15:06
brrt joined
|
|||
| brrt | grr, integration tests | 15:45 | |
| timotimo | did you mean: infuriation tests? | 15:46 | |
| brrt | oh well | 15:49 | |
| other tests didn't catch this | |||
| timotimo | you're talking about the jit frame handler thing? | ||
| brrt | i might have a few words to say about how the implementation of Net::HTTP | ||
| well, my basic question is | |||
| how does a deopt_all / deopt_ins get stuck on a set op | 15:50 | ||
| ugexe | it was written to work on JVM and pre syncsocket rework + threads, so its workarounds ontop of workarounds | ||
| brrt | i can believe that | ||
| timotimo | another op might have been deleted and the annotation would have fallen onto the op before or after it | ||
| brrt | otoh, it has a loop that calls .recv with a single byte | 15:51 | |
| not super elegant | |||
| ugexe | its the only way to handle that situation without a deadlock waiting on data without knowing the size before hand | ||
| or at least it was | |||
| timotimo | if this is about something happening in spesh, should be able to use the spesh dump-at-every-change thingie for gdb | ||
| ugexe | without Connection: close (Net::HTTP needed towork with keep alive) | ||
| brrt | nah, this is the expr JIT not handling something it should | 15:52 | |
| but i want to figure out how | |||
| timotimo | hmm, ok | ||
| brrt | ugexe, respectfully, that makes no sense to me. recv() in unix should return any bytes it has, not just block until it has N bytes | 15:53 | |
| if that's not how recv() works in perl6, it's recv()'s bug | |||
| or perl6-s recv()s bug | |||
| anything else is madness | 15:54 | ||
| on th eother hand | |||
| i'm happy for this madness because how would i have found my bug otherwise | |||
| ugexe | it only does that to read the header. the body is read in large chunks | 16:03 | |
| the 1 byte-at-a-time is so it can do .lines on binary data to read the header without dealing with \r\n encoding changes that kept happening | 16:04 | ||
| (in addition to the recv size being too large problem that was probably jvm related) | 16:05 | ||
|
16:12
rba joined
16:44
robertle joined
16:49
AlexDaniel` joined
17:15
ggoebel joined
17:34
zakharyas joined
17:55
cog_ joined
19:08
brrt joined
19:09
domidumont joined,
Util joined
19:14
zakharyas joined
19:27
patrickz joined
|
|||
| brrt | hmm, bailing on the DEOPT_ONE at the set doesn't actually do what i expected | 19:47 | |
| as in, it doesn't resolve the problem | 20:26 | ||
| that suggests that the duplicate dynamic label is | 20:30 | ||
|
20:59
rba joined
21:35
rba joined
22:31
evalable6 joined
22:58
ZofBot joined
23:43
evalable6 joined
|
|||