Branch: refs/heads/blead
Home: https://github.com/Perl/perl5
Commit: 098735fed6bf5f779f5900473a32a8fce21435cd
https://github.com/Perl/perl5/commit/098735fed6bf5f779f5900473a32a8fce21435cd
Author: Karl Williamson <[email protected]>
Date: 2026-09-21 (Mon, 21 Sep 2026)
Changed paths:
M embed.h
M perl.h
M pod/perldelta.pod
M pod/perllocale.pod
Log Message:
-----------
Revert "Temporarily enable thread-safe locale emulation"
This reverts commit 5c89d29f2741502a089e1627d9f70450acb84a9c.
Fixes #24844
lib/locale_threads.t will not hang forever, as it sets an alarm to
prevent that. The alarm is currently at 10 minutes; but could be
lessened.
Except for rebasing, this branch has been smoke-me'd without a problem
showing up. Of course no one would notice the slowdown on those. We do
not have a *BSD CI, but I had checked the Darwin one to see if the
testing there took appreciably longer, and I didn't think it did.
The branch has been smoke-me'd multiple times before with minor changes
between them.
I think the eventual solution is to cut down the number of threads
and/or iterations in the test to an acceptable wall clock time when
thread-safe locales are emulated.
To get a feel for those numbers, I logged on to the FreeBSD machine I
have access to. It is running 14.4-RELEASE-p3. And I now consistently
get an assertion failure in running miniperl to finish the build of
blead. This is a surprise, since the branch has had no problems in
recent smoke-me builds on FreeBSD. I will investigate.
I think that here might be a good place to take a swipe at those OS
vendors that have been lying to us for years that they support POSIX
2008 thread-safe locales. Our test suite passes on those only because
we skip tests we know would fail. locale_threads.t doesn't execute on
those boxes, except long enough to realize it would fail. So of course
it would suddenly take up time when the box gets a thread-safe mode.
And the test is designed to severely test claimed safety. Its slowness
doesn't indicate that a typical work load would be significantly
affected by the emulation. And if these OS'es performed as advertised,
the emulation would not kick in, and locale_threads.t would finish in an
acceptable time. If you are reading this, and have an "in" with the
developer of any of these OS'es, please apprise them that this is more
serious than they appear to think.
Solaris appears to have full thread-safe locale handling; and cygwin
shortly will. Perl works around a bug in glibc on Linux, so they can be
thread-safe, but the workaround slows things down. Windows has had
thread-safe locales for 20 years, disregarding some bugs that have
gotten fixed, but the difference in how they store text internally has
led to a slow interface with us, as noticed in #21611.
To unsubscribe from these emails, change your notification settings at
https://github.com/Perl/perl5/settings/notifications