11 Sep 2026
[Coke] ⏳ 2177 out of 2526 modules processed (86.18%) - that seems further than it had been getting 21:54
... do I just need to make "clone blin raku" step 0 each time? :)
s/raku/repo/
Geth rakudo: ugexe++ created pull request #6665:
Rebuild a flattened method table that came from deserialization
22:07
rakudo/main: 17 commits pushed by (Nick Logan)++
review: github.com/rakudo/rakudo/compare/0...dc06e5bff3
22:51
rakudo/main: f1264ab3fa | (Nick Logan)++ (committed using GitHub Web editor) | 4 files
Make quote parsing linear in the number of escapes (#6664)

Previously a quoted string with n backslash escapes or n nested delimiter pairs took O(n**2) to parse. The nibbler loop matched <escape> and
  <starter> <nibbler> <stopper> as capturing subrules and ran a code block
after each one. A code block in a regex binds $/, and doing so rebuilds the match by walking every capture on the cursor's stack. Each escape added a ... (17 more lines)
22:55
rakudo: ugexe++ created pull request #6666:
RakuAST: fix a fourth batch of deparse holes found by round tripping raku code
rakudo/main: 7e9a743c5c | (Nick Logan)++ (committed using GitHub Web editor) | 4 files
Rebuild a flattened method table that came from deserialization (#6665)

Previously a call site that had seen many types looked its method up in the flattened method table each class caches on its meta-object. Adding a method to a class clears the table of that class only, so a subclass whose table was built before the addition kept resolving the name to the method higher in the hierarchy. The setting builds such tables for the RakuAST ... (12 more lines)
22:59
rakudo/main: ef640c9d2b | (Nick Logan)++ (committed using GitHub Web editor) | 2 files
RakuAST: reject a character class in a transliteration (#6663)

A transliteration takes literal characters and ranges. The quote language it is parsed in declared tokens for \d, \h, \s, \v and \w, but nothing ever made a value for them, so each one reached the segment list as an NQPMu and the compiler died on it with "Cannot find method 'return-type'". The documented semantics do not want them either. tr/// works as it does in Perl, over ... (14 more lines)
rakudo/main: 00feb606ab | (Nick Logan)++ (committed using GitHub Web editor) | 2 files
Measure a case-folded interpolation match against the target (#6652)

Previously a string interpolated into a regex under :i was matched with nqp::eqatic and then assumed to have taken as many graphemes from the target as the pattern has. Case folding can change a grapheme count, so
  `'ß' ~~ /:i $('SS')/` computed an end past the target and was dropped,
while `'SS' ~~ /:i $('ß')/` stopped after one grapheme.
This measures the match by folding runs of the target until the fold is as long as the pattern's, and uses that as the length the match took.
[Coke] [2026-09-11T18:28:08-04:00] ⏳ 2292 out of 2526 modules processed (90.73%) 23:21
[2026-09-11T18:28:09-04:00] 🥞🥞🥞 Testing MoarVM::Profile (old) Finished with result: signal
Main processes terminated with: code=killed, status=9/KILL
Close but not quite 23:22
Sep 11 18:28:12 maxi systemd-oomd[697]: Killed /user.slice/user-1000.slice/user@1000.service/user.slice/blin.service due to memory pressure for /user. 23:29
slice/user-1000.slice/user@1000.service being 73.32% > 50.00% for > 20s with reclaim activity
Sep 11 18:28:12 maxi systemd[1152591]: blin.service: systemd-oomd killed 33 process(es) in this unit.
that from journalctl 23:35
ugexe what are the memory limits? is there any common module being processed between all the failed runs? 23:36
[Coke] there is no easy way to know what tests were in progress at the time. 23:48
and I didn't save the overview file from each run, regardless. 23:49
What I am trying now is upping to OOMScoreAdjust=-1000
was at -750, but I think -1000 is "just leave me alone"
my recent searches indicate this might have been "you were swapping too hard" 23:50
htop reporting 30.6G memory, 40G swap
ah, now my problem is figuring out what the latest build commit is. :) 23:51
trying again... 23:53
not sure why I didn't put it at -1K to begin with. :| 23:57
(what wa being tested) 32 simultaneously tests, and it could be we had 31 fast ones running on one thread. so if I grab the last 64 (ok,63) I could look for similarities. 23:59
12 Sep 2026
Geth rakudo/main: 20 commits pushed by (Nick Logan)++
review: github.com/rakudo/rakudo/compare/0...3badfb0df4
00:16
rakudo: ugexe++ created pull request #6667:
RakuAST: fix a fifth batch of deparse holes found by round tripping raku code
00:18
[Coke] setting it to -1000 didn't help 02:23
github.com/coke/raku-blin-release-...9/suspects 02:46
grabbed several screens worth up to the end
ugexe eventually blin should probably pipe the test output directly to a file instead of storing it all in memory, which is going to add up with that many modules 03:02
it'd also presumably let someone look at the tests full output for anything that got run at all even when getting oomd 03:03
Geth rakudo/main: 16 commits pushed by (Nick Logan)++
review: github.com/rakudo/rakudo/compare/b...b6ce74e13d
04:31
rakudo: ugexe++ created pull request #6668:
RakuAST: fix a sixth batch of deparse holes found by round tripping raku code
04:32