Geth nqp/main: 2fc2375fa7 | (Nick Logan)++ (committed using GitHub Web editor) | 2 files
Give character class methods a declarative prefix (#873)

Previously the character class methods of the cursor, such as alnum and digit, had no NFA. A branch calling one contributed nothing to longest token matching, so a shorter branch listed earlier won. Under ratchet that lost the match, since the group is not backtracked into. This gives each of those methods the one character it consumes as its NFA when merging subrules, using the same character class the method checks. A regex overriding one of these names keeps its own NFA.
Fixes rakudo/rakudo#3830
00:37
linkable6 RAKUDO#3830 [closed]: github.com/rakudo/rakudo/issues/3830 [regex] Strange token alternation errors with <alnum> and <graph> when they are in the second part of the alternation
Geth rakudo: ugexe++ created pull request #6662:
Bump NQP to give character class methods a declarative prefix
00:40
rakudo/main: 3703bd6488 | (Nick Logan)++ (committed using GitHub Web editor) | tools/templates/NQP_REVISION
Bump NQP to give character class methods a declarative prefix (#6662)
01:20
01:54 kurahaupo left
Geth rakudo/main: 4 commits pushed by (Nick Logan)++ 02:04
04:23 Pixi` joined 04:27 Pixi left, Pixi joined 04:29 Pixi` left 04:44 Pixi` joined, Pixi` left 04:46 Pixi` joined 04:47 Pixi left, Pixi` left 04:53 Pixi joined 05:14 Pixi` joined 05:18 Pixi left 05:35 Pixi joined 05:37 Pixi` left 06:08 Pixi` joined 06:10 Pixi left 06:57 lizmat joined 08:06 lizmat left 08:59 lizmat joined 09:04 kurahaupo joined
lizmat . 09:20
09:20 lizmat left 11:31 Pixi` is now known as Pixi 12:00 kurahaupo left
[Coke] builds caught up with ugexe. Re-starting blin 13:15
13:23 lizmat joined 13:30 lizmat left 15:05 shareable6 left 15:07 shareable6 joined 15:36 sivoais left 15:38 sivoais joined
[Coke] ... whoops. ran out of disk. Whoops, had a copy of blin directory and EVERYTHING from a build in it lying around. 16:11
ab5tract Sorry, more disk will cost you $1010 16:20
I recently looked up the now-price of my last non-nvme drive and... it wasn't a pleasant experience
I didn't have the heart to look at the last nvme drive vs its now-price 16:21
[Coke] oof moarvm.org git repo is 2.1G
ok, that's the second blin run in a row where the main thread was killed. 17:56
Finished with result: signal
Main processes terminated with: code=killed, status=9/KILL
Any commits in the last 24h intended to impact memory usage?
really not looking forward to doing a bisect of blin runs. 17:58
(having said that... kicking off a bisect of blin runs) 18:00
ugexe maybe one of the regex one, although i don't have a more targeted guess 18:15
[Coke] yah, I figured it was not an answerable question. :) 18:20
19:46 lizmat joined 20:09 lizmat left
[Coke] I suppose it could also be a module that is now generating a lot more output. (if the module itself had a leak, OOM killer should have preferred it) 20:18
ugexe i tested a bunch of grammar heavy distributions and their memory usage all went slightly down 20:22
[Coke] ok. I'll figure it out. (could just be my box) 21:39
ugexe the output thing you mentioned does seem like it could cause that though since it looks like blin gets all the output before calling .decode on it 21:41
m: my $t = "abc-"; say .Str with $t ~~ tr/\-abc/1234/; 21:43
camelia 123-
21:59 lizmat joined 22:00 lizmat left 22:19 kurahaupo joined