niner

Those reverts improved the situation a lot, but did not actually fix it. I still got 2 module that show these symptoms: [ 18s] 1 RMD: Precompiling /home/abuild/rpmbuild/BUILDROOT/perl6-JSON-Name-0.0.6-1.89.x86_64/usr/share/perl6/vendor/sources/94361D38EBEDEAA5B55B6FEDD31CF9FC44EEF7E7 into /home/abuild/.raku/precomp/D763EACC8498FC80A334E6E77D38EADF9C9A3F81/94/94361D38EBEDEAA5B55B6FEDD31CF9FC44EEF7E7.bc...

I finally got it, for real this time. The culprit is: https://github.com/rakudo/rakudo/commit/6b50a940af77211709ef872560b2ced35f137c11#diff-1d976180c81a8bba84a9e361d63676c61d40398d80ca17adf58d095eeba3feabR335

Fixed for real in https://github.com/rakudo/rakudo/commit/cc7193d0a6d95bd580a4b2783795d082a56d8e6d

What the posted backtraces don't show is that the issue is an infinite recursion in the spesh optimizer. Which explains the different places the segfault happens and the mysterious changing...

On Samstag, 2. April 2022 00:21:05 CEST Elizabeth Mattijsen wrote: > But doesn't the CURS already **replace** the CURI in the chain with this: No, that just replaces the CURI...

> So maybe it should also replace it in the `$*REPO` chain as well? That > could be done in `TWEAK`, and then have `self-destruct` restore the > original situation?...

I think inadvertent sharing of the $/ variable between threads is often an issue with regexes.

We really use CPAN just as storage space. The CPAN indexer ignores Perl6 directories, so CPAN doesn't care if the same distribution or module name appears in multiple author directories....

Yes, my script does: `RAKUDOVERSION=$(echo -n $RAKUDOGITVERSION | sed 's/-/+git/' | sed 's/-/./g')` This is just the convention that openSUSE seems to use for git versions. Both `.` and `-`...

dod38fr: I have spent loads of time thinking about how rakudo's precompilation can work with distribution packaging and have implemented it in a way so all your issues should actually...