| timotimo | jnthn: stdout and stderr will fail when you call "flush" on them because it tries to use fsync which isn't sensible on non-filesystem-backed streams | 00:42 | |
| this currently makes --target=foo end in an uncaught exception | 00:43 | ||
|
01:48
ilbot3 joined
05:47
domidumont joined
06:27
domidumont joined
07:17
domidumont joined
07:23
lizmat joined
08:20
robertle joined
|
|||
| jnthn | timotimo: Yeah, I know about that one. | 08:25 | |
| timotimo | OK | 08:27 | |
| nwc10 | good *, * | 08:34 | |
| jnthn | o/ | ||
|
08:51
zakharyas joined
09:12
brrt joined
|
|||
| brrt | ohai #moarvm | 09:18 | |
| jnthn | o/ brrt | 09:23 | |
| brrt | \o jnthn, timotimo, nwc10, * | ||
| Geth | MoarVM: 6977bc95b8 | (Jonathan Worthington)++ | src/io/syncfile.c Don't complain about fsync failure on TTY/pipe. |
09:44 | |
| MoarVM: acc73c7250 | (Jonathan Worthington)++ | src/6model/serialization.c Serialize a Decoder REPR object into null. Just as we do for file handles. This re-instates the Perl 6 level behavior that a filehandle serialized "accidentally" during pre-comp won't blow up, unless anything tries to actually use it after the deserialization. |
09:54 | ||
|
09:58
JimmyZ joined
|
|||
| JimmyZ wonders whether rakudo uses moarm's sync proc ops or not now. | 09:59 | ||
| jnthn | JimmyZ: Nope | ||
| We'll kill them off | |||
| And probably some NQP tests too | |||
| Then we only have to maintain one set of proc ops :) | |||
| JimmyZ | nice killing again :) | 10:00 | |
| jnthn likes having less code to maintain :) | |||
| JimmyZ too :) | |||
| JimmyZ wonders what's jnthn++'s plan for ea :) | 10:02 | ||
| jnthn | JimmyZ: Think it'll come after more general spesh changes, so it can be more effective | ||
| JimmyZ | jnthn: agrees | 10:03 | |
| the current spesh just like the jit in the master, which meeds some refactor | 10:05 | ||
| jnthn | Well, it's not so much refactoring it needs as re-visiting a bunch of the ways it does things | 10:13 | |
|
11:00
zakharyas joined
|
|||
| brrt | hehe, spesh can be much more clever about a bunch of things | 11:05 | |
| but that is of course true of everything | 11:06 | ||
| jnthn | lunch; bbiab | 11:20 | |
| nwc10 | ilmari: ^ | 12:07 | |
|
12:30
committable6 joined
12:31
committable6 joined
13:17
zakharyas joined
|
|||
| Geth | MoarVM: c4dbebe119 | (Jonathan Worthington)++ | src/6model/serialization.c Only deserialize decoder instances as NULL. Leave type objects as they are. |
13:38 | |
| jnthn | nwc10: That hopefully fixes the issue you saw :) | ||
|
14:04
SourceBaby joined
|
|||
| nwc10 | jnthn: it's into tests, so it's got through the build, which it didn't before | 14:27 | |
| must go, to catch a train | |||
|
14:34
travis-ci joined
|
|||
| travis-ci | MoarVM build errored. Jonathan Worthington 'Only deserialize decoder instances as NULL. | 14:34 | |
| travis-ci.org/MoarVM/MoarVM/builds/242425178 github.com/MoarVM/MoarVM/compare/a...dbebe119d6 | |||
|
14:34
travis-ci left
|
|||
| eveo | Just github issue | 14:36 | |
| error: RPC failed; curl 56 SSLRead() return error -36 | |||
| fatal: The remote end hung up unexpectedly | |||
|
14:40
robertle joined
15:29
zakharyas joined
|
|||
| Geth | MoarVM: 8fe17d3b3f | (Jonathan Worthington)++ | 3 files Improve named argument optimization. We could toss param usage tracking and turn param_sn into a hash construction in a bunch of cases, but failed to do so. |
15:38 | |
| MoarVM: 78ac0ad54c | (Jonathan Worthington)++ | src/jit/graph.c JIT decodertakeline. |
|||
| MoarVM: 2e9aff7feb | (Jonathan Worthington)++ | src/spesh/args.c Keep named-used marking if binding can de-opt. Otherwise, we end up running the instructions to test if named args are missing and may wrongly report they are. |
|||
| jnthn | Yowser, I just discovered some performance we've been missing out on... | 15:50 | |
| eveo | \o/ | 15:51 | |
| Geth | MoarVM: 21ee1a5048 | (Jonathan Worthington)++ | src/core/frame.c Fix missed opportunity to run JITted code. Often we can pick the candidate at the time of a call, to save a search to find the correct specialization. However, when we did that, we always picked the specialized-but-interpreted version of the code even if there was a JITted one available. |
15:53 | |
| jnthn | m: say 2.767 / 2.873 | 15:57 | |
| camelia | 0.963105 | ||
| jnthn | 4% off "read a million lines and count up their chars" benchmark from these various fixes | ||
| Though they should have impact across the board too | 15:58 | ||
| It's the first time I've looked at spesh output and code-gen in a while | 15:59 | ||
| eveo | Awesome. | 16:00 | |
| [Coke] | jnthn++ | 16:08 | |
|
16:30
domidumont joined
16:48
Util joined
17:11
travis-ci joined
|
|||
| travis-ci | MoarVM build passed. Jonathan Worthington 'Keep named-used marking if binding can de-opt. | 17:11 | |
| travis-ci.org/MoarVM/MoarVM/builds/242473316 github.com/MoarVM/MoarVM/compare/c...9aff7feba1 | |||
|
17:11
travis-ci left
18:12
domidumont joined
18:24
AlexDaniel joined
20:02
vendethiel joined
21:18
geekosaur joined
21:35
AlexDaniel joined
22:07
colomon joined
|
|||