This Week on perl5-porters (28 July / 3 August 2003)

This Week on perl5-porters (28 July / 3 August 2003)

This week will undoubtedly be known to the future generations as the two-release-candidate-week. Be the first to read about it. And don't miss the other interesting parts : this week's summary is full of action, suspense and bug fixes.

Release Candidates

Jarkko released perl 5.8.1 RC3 and 5.8.1 RC4.

RC3 announces the return of the hash randomisation : it's again turned on by default.

In RC4 several modules were updated, notably MakeMaker and CPAN. The main problem with RC3 was that the CPAN module was forcing to use Module::Signature (to verify cryptographically signed modules), and this feature wasn't tested enough. Moreover Module::Signature is not part of the core. Autrijus Tang is working on better portability and integration for Module::Signature, so it may become part of the core some day.

There was also some courteous fight about whether deprecating vstrings was a good thing or not in 5.8.1, regarding backward compatibility. It's apparently not finished yet (but Jarkko is until now keeping the deprecation warnings turned on.)

Notes from the P5P BOF, and perl 5.6.2

Elizabeth Mattijsen posted the notes taken at the P5P meeting in Paris, after YAPC::EU. Go read it, that's an interesting complement to this summary.

One of the points discussed was the importance of a 5.6.2 release -- the conclusion being that it's more important to release 5.6.2 than Sarathy to be able to fix all his bugs. Therefore, Sarathy transferred the 5.6 pumpkin to Rafael, who in turn posted plans for 5.6.2 (and perhaps 5.6.3). A new branch was created in the perforce repository ; it's available like bleadperl and maintperl via rsync (pathnames being /perl-5.6.2 and /perl-5.6.2-diffs for the patches) and via the APC. The goal of this new release is mainly to fix the known problems to compile it on recent systems ; for example, Configure seems to produce no makefile on Mac OS X with a case-insensitive filesystem.

To packagers

The DESTDIR variable is now honored by make install, to specify a prefix to all installation paths. This should please RPM package maintainers, thanks to Michael Schroeder. See the INSTALL file for details.

STDOUT reopened

Chris Nandor complains that opening a file for reading when STDOUT has been just closed emits a warning : Filehandle STDOUT opened only for input. That's because the newly opened filehandle gets the file descriptor number of STDOUT (which is 1). As the file descriptors are actually allocated by the C-level system call API, and as STDOUT is always the fd 1 by definition, there is no way to change this ; however, the condition on which the warning is produced could be changed, if we could agree on it.

Still working on MakeMaker

Steve Hay and Michael Schwern were trying to solve a MakeMaker regression that prevented libapreq to build correctly on Windows XP with VC++ 6.0. Apparently the problem comes from a hackish workaround for a now-solved MakeMaker bug in the libapreq sources (MakeMaker wasn't handling recursive makes correctly.) Michael fixed other bugs as well. As of this writing, the latest release of MakeMaker on CPAN is 6.13^W14^W^W^W 6.15.

In Brief

Dave Mitchell found and fixed a utf8+regex regression bug in 5.8.1-RC2. Abigail found another regex regression bug (#23171), that was fixed by Hugo.

When use File::Glob ':glob' is enabled, calling twice the glob angle-bracket operator may cause a segfault or a memory corruption error (bug #23185). Marcus Holland-Moritz found out that two incompatible internals APIs were mistakenly used together, and provided a fix.

Michael Schwern notices that perldoc CPAN gives the doc for the cpan script on Mac OS X, due to the case-insensitive filesystem. To get the docs for the module, it's necessary to tell explicitly perldoc

Ed Avis finds that localization of $_ fails when it's aliased to $1. Bug #23141.

Stephen McCamant continues to improve B::Deparse.

Enache Adrian proposed tests for B::Bytecode and B::ByteLoader. They weren't integrated into maintperl, though, due to potential instabilities.

Enache also fixed bug #23207, about chained calls to ^= not behaving properly.

Abe Timmerman released Test::Smoke 1.18.

Look, a spec for the META.yml file ! (it may be a bit outdated though.)

About this summary

This week's summary was, as usual, written by Rafael Garcia-Suarez. Weekly summaries are published on and on a mailing list, which subscription address is Comments, corrections, additions, and suggestions are welcome.