Hey everyone,

Following is the p5p (Perl 5 Porters) mailing list summary for the past week (including Monday the 26th). Enjoy!

October 20th-26th

Updates

Additional grant reports from Dave Mitchell. The majority of the work was spent on overhauling the context stack. End result? Fixes for the original issue ("death during unwinding causes crash") and entering and existing subroutines and loop will be a lot faster!

Additional grant report by Tony Cook. Tony has spent around 16 hours, reviews or worked on approximately 12 tickets and applied 2 patches.

Ricardo Signes updates that November will have an Onion Sketch. The Onion Sketch was mentioned in the last update.

Ricardo also applied a patch by Peter Rabbitson (Ribasushi) in order to get Carp passing on older toolchains. Ricardo asked for additional eyes on the commit before a release - Craig A. Berry gives an A-OK on it.

Bulk88 provided a now-merged patch in Perl #126412 to address the GCC builds failing on Windows, mentioned on this discussion thread.

Tony Cook adds more information on the __DIE__ + goto leading to infinite recursion and segfault.

Tony also cleaned up tickets: * Perl #125891 was a ticket about a behavior with tied filehandles and I/O which is deemed not a bug. * Perl #126062, which raises a possible issue with Digest::SHA, but is not maintained by p5p and the discussion seems to have dropped.

Ævar Arnfjörð Bjarmason had issues reintroducing the new warnings pragma category due to technical reasons, and - as he put it:

[...] about to spawn a subprocess.

All the best and have a merry PID. :)

The Thread::Queue in core is updated to version 3.07.

Dennis Kaarsemaker provides some improvements for the simple .travis.yml file mentioned in Perl #123981.

Karl Williamson updates that perl now passes all tests on z/OS. Karl gave a great summary of his work.

Bulk88 provides a patch in Perl #126452 to optimize more the build and compilation of perl of Windows. Interesting insight there.

Bulk88 also provides a patch in Perl #126453, related to a discussion mentioned below in this update regarding version.

Bulk88 provided a patch (merged by Tony Cook) in Perl #126431 to clean up the API by making some functions static.

Bugs

Reported bugs

Perl #126414, opened by Bulk88, discussed perl storing inodes in NVs/doubles, and mentions possible rounding for inode integers, possibly causing a mismatch when comparing them.

Perl #126410, reported by Todd Rinaldo, tries to raise (and address, with a provided patch) a situation in which perl creates a destructor cache when there isn't an AUTOLOAD or DESTROY method. The patch prevents that from happening.

Perl #126407, reported by Vadim Pushtaev, asks about AUTOLOAD being called for the DESTROY method. This seems related to the ticket above, opened by Todd Rinaldo. Vadim also found this already exists as Perl #124387.

Perl #126437, reported by Ed Avis, asks about hex's behavior with invalid input in relation to its documentation. Eirik Berg Hanssen adds this useful hex check pattern:

eval { use warnings "FATAL"; hex(s/^\s+|\s+$//rg) }

Perl #126432, reported by Peter J. Holzer, suggests inconsistencies with autovivification. Dave Mitchell commented on the ticket, explaining how perl understands the code used in the ticket.

Resolved bugs

• Perl #126309: x operator on list causes segfaults, confuses valgrind.
• Perl #117341: av_undef's POD is confusing.
• Perl #126181: \c inside (?[]) causes panics and unexpected behavior.
• Perl #126178: /(?i/ and similar should raise an error.
• Perl #126253: /.{1}??/ should be an error.

Discussion

Due to the new parallel Win32 dmake feature, a race condition failure was discovered. POSIX's Makefile.PL file was throwing an error (swallowed by make_ext.pl), and a POSIX XS binary was created with missing subs. Tests were failing despite the POSIX binary loading.

The error in the Makefile.PL was traced to POSIX being built before version was, and ExtUtils::MakeMaker's Pure Perl fallback required core XS modules aren't available in miniperl.

A number of fixes were proposed by Bulk88 and Steve Hay, the simplest of which was eventually committed.

The entire discussions can be read on the thread.

Jarkko Hietaniemi has been running Coverity and found a few issues with POSIX. Jarkko has been asking about specific file descriptor usage. Tony Cook mentions the known implications in how they're used and Jarkko suggests possibly cleaning it anyway.

Michael Felt reports a smoker not working on an AIX machine due to IPv6 tests.

Karl Williamson explains problems with the implemented \s{wb} and \s{sb} in perl 5.22, and suggests to include fixes in 5.22.1 despite perl versions policies. So far he received positive feedback on this suggestion.

Felipe Gasper had garnished more attention in his question on Perl #126403 about optimizing the read call buffer size within the PerlIO layer. The ticket contains some background on the buffer size thanks to Christian Hansen and Craig A. Berry, including a mention of Leon Timmermans' PerlIO::buffersize and even a first patch by Tony Cook.