01:19 kjp left 01:23 kjp joined
[Coke] WIP on blin 02:07
:'
BTree – Fail, Bisected: a107b0ce342aaea278202f6b7110d3a25e9ed4d6
Control::Bail – Fail, Bisected: a107b0ce342aaea278202f6b7110d3a25e9ed4d6
a107b0ce342aaea2
linkable6 (2026-09-20) github.com/rakudo/rakudo/commit/a107b0ce34 Make RakuAST default for compilation of programs (#6704)
03:09 vrurg joined 03:13 vrurg_ left
[Coke] up to about 20 failures at 47% of the ecosystem 03:25
most on the switch to ast as default
Geth rakudo: patrickbkr++ created pull request #6713:
fix empty args in Windows script runner		 06:57
rakudo/main: dd116c27af | (Patrick Böker)++ (committed using GitHub Web editor) | src/runner/runner.c
fix empty args in Windows script runner

Previously, an empty argument (i.e. `""`) passed to a script wrapper just disappeared.		 08:18
lizmat github.com/rakudo/rakudo/issues/6714 08:21
(blocker)
looking at BTree now
Cannot look up attributes in a VMNull type object. Did you forget a '.new'? 08:23
in submethod BUILD
that feels... odd :-) 08:28
09:06 finanalyst joined
lizmat added a workaround PR for BTree now: github.com/Scimon/raku-Tree-Binary.../1/changes 09:21
Control::Bail is a Slang issue 09:24
since that module is still in p6c, and the author hasn't responded at least since 2024, and the last commit is 9 years old 09:26
and the doc states:
"Also, this module uses a lot of metamodel/internal stuff that is not necessarily nailed down by specification, so it cannot promise the same stability as one that uses only the tested 6.c features."
it looks like it's adding UNDO/LEAVE/KEEP phasers at runtime... 09:30
[Coke]: I'd say, add this to the "dead" list 09:31
Geth rakudo/remove-temp-windows-fix: 9803a81466 | (Elizabeth Mattijsen)++ | t/02-rakudo/m-flag-module-spec.t
Revert "Fix failing test in Windows CI, at least for now"

This reverts commit 75efc97ca1d58596840c6b47c1b9cf0d84b02a27.
With dd116c27af this test should just work on Windows		 09:36
rakudo: lizmat++ created pull request #6715:
Revert "Fix failing test in Windows CI, at least for now"
lizmat let's see what the CI thinks 09:37
[Coke]: any other modules I need to look at? 09:38
10:13 finanalyst left
releasable6 Next release in ≈3 days and ≈7 hours. 1 blocker. Please log your changes in the ChangeLog: github.com/rakudo/rakudo/wiki/ChangeLog-Draft 11:00
lizmat meh, patrickb 's fix did not fix the windows error 11:54
[Coke] lizmat: oh, so many. 12:58
looks like 145 failures. 12:59
Will post results as I can to the coke/* repo. 13:00
I've not done a "redo" - haven't checked the individual failures yet. 13:07
lizmat ack 13:44
just found a number of Butterfly Plan modules failing 13:46
so I have plenty to look at
[Coke] ok, have 15m, grabbing a snapshot... 13:49
raw.githubusercontent.com/coke/rak...ailures.md 13:52
... not sure why my initial count read high. looks like 82 failures. 13:53
I do see Selkie in the list of fails 13:54
(and the half dozen from the previous run)
P5__DATA is one of them. 13:55
lizmat that one should be dead... hmmm
[Coke] And I'm sure a bunch of these were classified as wontfix or similar from ugexe on a previous run
ugexe since the failure count jumped by like 100 its likely a few different rakudo regressions from various optimizations 13:56
[Coke] If you have ones that should definitely be skipped in blin going forward, we can do that, just leave me a note with the module name and the reason (or give a PR/commit to raku/blin/resources/skips.json
ugexe: it's closer to 80 something, my bad. 13:57
(total)
ugexe ah, a big source of failures is the :P5 regex removal, particularly HTTP::Tiny and things that depend on it
[Coke] ah. Well, those are a wontfix for us.
ugexe i think we'd just patch http tiny
lizmat yeah
[Coke] Cool. let me know when I can run again. back to $dayjob 13:58
ugexe++ for fixing the race condition so the blin runs are more useful first pass now
ugexe http::tinyish rather 13:59
github.com/skaji/raku-HTTP-Tinyish...akumod#L45 (in case someone wants to fix it before i get out of work in 10ish hours 14:00
lizmat I'll take that on 14:01
ugexe CI::Gen, Router::Right, RegexUtils, LIVR are the other p5 regex failures that aren't tied to http::tinyish 14:08
although some of those we may have already decided to not update (livr and regexutils maybe?) 14:09
lizmat hmmm.. I see something slightly disturbing wrt to coverage testing 14:10
looks like any method that *is* being called, does not get a "run" marking on the line with "method" and the signature 14:11
so unless a method is a one-liner, there will always be lines "not seen" 14:12
this only bothers me as a developer, as the coverage testing that my modules do, are developer only
ugexe: is that something that you were aware of? something done intentionally as an optimization? 14:13
if the latter, I can live with that :-)
ugexe i dont really follow 14:14
lizmat x method new($path) {
* opendir(my $dirhandle, $path) ?? self.bless(:$dirhandle) !! Nil
the * means it was run, the x means it wasn't 14:15
the code that was run, can only be run if the method was called
yet the method itself is marked as "not run"
with the "x"
ugexe yea i dunno, so doubt its optimization related 14:19
lizmat well, I'll just mark the cases as UNCOVERABLE for now
ugexe gist.github.com/ugexe/ed5d4ea9b285...9acf8f9010 here is the errors grouped together 14:28
indeed a lot of them are ones we are ignoring. the pdf grouping is probably from an optimization so i'll look at those later 14:29
Polyglot::Regexen is another one that needs a module level fix as its constructing stuff using RakuAST::Regex directly 14:31
probably just needs to wrap one of its arguments in [ ] 14:32
lizmat yeah, so victim of API change in RakuAST, right?
ugexe yeah 14:33
[Coke] releasable6: next 15:56
releasable6 [Coke], Next release in ≈3 days and ≈3 hours. 1 blocker. 479 out of 494 commits logged
[Coke], Details: gist.github.com/f0a0d65ee6922526ab...88bc138411
Geth rakudo: ugexe++ created pull request #6716:
RakuAST: annotate signature binding with the routine's position		 17:08
ugexe that^ and github.com/MoarVM/MoarVM/pull/2047 make Code::Coverage work correctly
lizmat cool! 17:10
Geth rakudo: ugexe++ created pull request #6717:
Use NQP's own home for the build dir runner		 17:11
rakudo: ugexe++ created pull request #6718:
RakuAST: fix the PDF modules failing in latest Blin		 17:13
lizmat made an issue with HTTP::Tinyish github.com/skaji/raku-HTTP-Tinyish/issues/16 been too long for me doing P5 regexes, felt unsure I could do a good job there 17:41
Geth rakudo: ugexe++ created pull request #6719:
Stop rejecting valid attribute and is default defaults at compile time		 18:05
ugexe rx/ <-[ \x[00]..\x[1f] \x[7f] \[ \] ( ) < > @ , ; : \\ " / ? = { } \x[20] ]>+ / 18:08
lizmat ok, I'll do a HTTP::Tinyish PR :-) 18:09
Geth nqp/main: ee17498e1d | (Elizabeth Mattijsen)++ | tools/templates/MOAR_REVISION
Bump MoarVM for coverage slot fix
 18:17
rakudo/main: 7f1c9ad625 | (Elizabeth Mattijsen)++ | tools/templates/NQP_REVISION
Bump NQP for coverage slot fix
 18:26
rakudo/main: 0768b3f3cf | (Nick Logan)++ (committed using GitHub Web editor) | 2 files
RakuAST: annotate signature binding with the routine's position

Previously the QAST that binds a block's signature had no node, so it got no line annotation. MoarVM's line coverage attributes the frame's entry basic block to the first annotation in the frame, which is the first statement of the body. A called routine never reported its declaration line, leaving a "not covered" mark on every multi-line method or sub ... (5 more lines)		 18:28
rakudo/main: 7 commits pushed by (Nick Logan)++ 18:42
rakudo/main: 8bf3e83d47 | (Nick Logan)++ (committed using GitHub Web editor) | tools/templates/moar/Makefile.in
Use NQP's own home for the build dir runner (#6717)

Previously the build dir rakudo-m runner set NQP_HOME to share/nqp under Rakudo's install prefix. When configured with --with-nqp pointing at an NQP installed elsewhere that directory does not exist, so anything that reads files from NQP home fails. For example --profile could not find profiler/template.html and fell back to sql output.
This takes the NQP home reported by the NQP being built against.
patrickb lizmat: I have one hour at hand and have booted up my Windows machine. 19:02
How can I repro the runner space issue?
Geth rakudo: ugexe++ created pull request #6720:
RakuAST: fix the two Selkie failures from the latest blin run		 19:05
lizmat patrickb: check commit 75efc97ca1d58596840c 19:09
linkable6 (2026-09-22) github.com/rakudo/rakudo/commit/75efc97ca1 Fix failing test in Windows CI, at least for now
patrickb hm. 19:11
lizmat that is the test that failed 19:12
patrickb this doesn't seem to be about the script wrappers at all.
`is-run` is running rakudo directly, isn't it? 19:13
lizmat afaik, yes?
you mean it's an issue with is-run on Windows?
patrickb is-run is calling `rakudo.exe`. Either the temporary one created during the build, or the final one. 19:15
That executable is probably to blame.
The script wrappers only come into play when installing a module that contains a bin script
19:20 finanalyst joined
lizmat so Claude's reasoning for the error was way off 19:20
good to know! but perhaps it was *also* an issue with the runner ? 19:21
afk& 19:23
patrickb it was. the find and fix are legit.
ugexe yeah, its the same issue but for installed modules 19:24
Geth rakudo: patrickbkr++ created pull request #6721:
fix empty args in Windows build dir rakudo binary		 19:29
patrickb There you go. The build dir rakudo and the script wrappers share the same argv quoting impl. So they had the same bug. 19:31
Geth rakudo: ugexe++ created pull request #6722:
RakuAST: decide a generic coercion needs the full binder at begin time		 19:42
[Coke] Do we want that to go in this release? 19:49
Guessing yes.
It's Wednesday. I know we're cramming a lot into this release, but how late are we expecting commits to come in? (I can delay the release until Sunday) 19:50
(or later)
ugexe probably the rest of the day 19:56
[Coke] ok. ping me when I can kick off the next run (last one took about 5 hours) 20:02
I can do a "short" one with just those that failed last time first, if that would be helpful
ugexe we should keep doing all of them to make sure the fixes don't regress something else 20:14
Geth rakudo: ugexe++ created pull request #6723:
RakuAST: reduce <!before .> to the end anchor		 20:15
21:05 finanalyst left
Geth rakudo/main: a6cfbf4526 | (Nick Logan)++ | 3 files
RakuAST: only reject an attribute default no value could satisfy

Previously a RakuAST compiled attribute declaration refused to compile when its default was read from a variable whose type did not match the declared type exactly. `my @d = 1, 2; class C { has Int @.x = @d }` gave "Can never assign default value "Positional" to attribute '@!x', it expects: "Int"". So did `my Any $d = 1; class C { has Int $.x = $d }` ... (11 more lines)		 21:47
rakudo/main: 2b844e3fb5 | (Nick Logan)++ | 5 files
Accept a role type parameter as an is default value

Previously `role R[::T] { has Mu:U $.x is default(T) }` failed to compile with "Can never assign default value T (T) to attribute '$!x', it expects: Mu:U", and `my Mu:U $x is default(T)` in a method of such a role failed with "Default value 'T' will never bind to a variable of type Mu:U". Only a generic declared type skipped the type check of the ... (7 more lines)
rakudo/main: c3aef3adca | (Nick Logan)++ (committed using GitHub Web editor) | 7 files
Merge pull request #6719 from ugexe/ugexe/rakuast-attribute-default-check

Stop rejecting valid attribute and is default defaults at compile time
rakudo/main: 0dc4f983a9 | (Patrick Böker)++ (committed using GitHub Web editor) | tools/templates/moar/rakudo-m-build.c.windows
fix empty args in Windows build dir rakudo binary (#6721)

The fix in dd116c27af34f1da847e68f06f33f2173b88f8aa applies to the build dir rakudo runner as well.
rakudo/main: facfdfee64 | (Nick Logan)++ (committed using GitHub Web editor) | 3 files
RakuAST: decide a generic coercion needs the full binder at begin time (#6722)

Previously a parameter typed with a coercion to a type capture, such as T(), only forced custom-args on its routine at CHECK time. A role body compiles and lowers its lexicals at begin time, so the routine's parameters were already lowered to frame-locals. The unit then formed the routine again with the full binder, which writes the lexicals by ... (5 more lines)
rakudo/main: ef1ebb037e | (Nick Logan)++ (committed using GitHub Web editor) | 2 files
RakuAST: reduce <!before .> to the end anchor (#6723)

Previously a negated before assertion of any character reduced to the end anchor paired with a negated zerowidth `.` class. The cclass op ignores negation for `.`, so that branch held wherever a character remained and `<!before .>` matched everywhere. Under ratchet a sequential alternation like `<!before .> || \w+` then committed to the first branch and never tried the second. This reduces the assertion to the end anchor alone.
rakudo/main: b07ba79aac | (Nick Logan)++ | 2 files
RakuAST: keep a doc continuation line that starts with = and no word character

Previously a line of an abbreviated or paragraph block that started with a single = ended the block, even when no word character followed. A markup code continued onto a line starting with => then died with
  "RakuDoc markup code C missing endtag". This ends the block only at
  = followed by a word character, as the legacy grammar does.		 21:55
rakudo/main: 182b0a20fa | (Nick Logan)++ | 3 files
RakuAST: keep conditional code that was compiled ahead of the unit

A role method or a BEGIN block is compiled before the optimize walk of the unit, and each thunk and block in it gets a code object then. Previously the walk could still drop that code from a branch with a constant condition, such as `if False { try 1 }` in a role method, and precompiling the module died with "QAST::Block with cuid N has not appeared". This keeps a branch or operand holding code whose block is already formed, the way it keeps one holding a declaration.
rakudo/main: d56246d410 | (Nick Logan)++ (committed using GitHub Web editor) | 5 files
Merge pull request #6720 from ugexe/ugexe/rakuast-selkie

RakuAST: fix the two Selkie failures from the latest blin run
ugexe github.com/raku-community-modules/...eam/pull/5 22:19
github.com/raku-community-modules/...ica/pull/1
github.com/raku-community-modules/...lic/pull/1
i still have those open prs as well
[Coke] wanty me to kick off a blin run? 22:20
lizmat ugexe: looking at the community PRs now 22:21
ugexe github.com/raku-community-modules/...er/pull/13 22:25
[Coke]: no not yet
lizmat ok, I fear that some of these will have to wait until tomorrow :-) 22:26

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!