Now that perl 5.8.0 is out and flies by itself, the 5.9 development track begins. Learn about what may go (or not) in perl 5.9. Plus the usual amount of bugs.
Hugo van der Sanden posted his views on the general directions of perl 5.9 development. Shortly, he listed (with varying degrees of importance or probability) : improving perl's speed ; cleaning the sources ; converging with Perl 6 (with a possible new perl6ish pragma, to use some of the incompatible Perl 6 constructs or deprecations) ; providing multiple Perl distributions containing a various amount of modules.
There was also some handwaving (which I initiated) about having some kind of Perl 5 to Perl 6 or Parrot translator.
This would probably involve improving the B:: backend framework if this is the way to go.
http://groups.google.com/groups?threadm=200207301220.g6UCKtr07589%40crypt.compulink.co.uk
(I now give the message links to google, so you can navigate the whole thread. I found this more convenient.)
Pseudo-hashes and the old 5005 threads will be removed. Michael G Schwern provided a first patch to cut off pseudo-hashes, that doesn't handle the necessary changes to the core modules, esp. fields.pm and base.pm.
Hugo also wondered if non-PerlIO perls should be deprecated. This decision should be based on some real-world feedback -- are there platforms that have problems with PerlIO ?
http://groups.google.com/groups?selm=200207301220.g6UCKvQ07607%40crypt.compulink.co.uk
As 5.8.0 is out, some problems arise when the CPAN shell is used to install modules that have a dual life (on CPAN and in the core.) Andreas Koenig has updated the CPAN indexer so that it indexes perl distribution only for packages that have no separate lives on CPAN. All packages that have dual lives have been also re-indexed ; this should solve current problems.
One of the specificities of the implementation of threads on Linux is that threads get different PIDs. So Elizabeth Mattijsen asked for a way to get this PID from inside Perl, because $$ is always the process number of thread 0.
It turned out that POSIX::getpid() simply returns $$, and that $$ is is a read-only scalar, set at startup- and at fork-time. But the system call getpid() returns different values for different threads (on Linux).
It was agreed that, to be portable (and POSIX-compliant), $$ should return the same value across all threads -- the current behavior is thus correct (but undocumented.) However Perl's built-in function getppid() always performs the underlying C call, and return different values from different threads. This should be fixed.
http://groups.google.com/groups?threadm=4.2.0.58.20020801191058.02b8a440%40mickey.dijkmat.nl
Somebody reported that writing %:: = "" with warnings enabled leads to immediate segfault (bug #15479). Elizabeth, trying to write something like our ${""} : foo = 1, reported another core dump case (this should be a syntax error) (bug #15898). This lead to a small thread about the right place for regression tests for nonsense code like this. The conclusion is that they should go in the test file that seems the more appropriate for it (in those cases, respectively tests for symbol table hashes and for attributes.)
Michael G Schwern sent a list of thoughts and ideas that were discussed at TPC :
What new modules should be included in 5.10, and based on which criteria ? The simple criterion he proposes is : Will this module help users install more modules? He suggests notably to borgify CPANPLUS, a simplified LWP, Archive::Tar, Archive::Zip, Module::Build (this implies YAML), and Inline.
http://groups.google.com/groups?threadm=20020801160521.GB1064%40ool-18b93024.dyn.optonline.net
About Inline, and related to the QA effort, Schwern suggests to test the core API and internals with Inline::C. Continuing on Inline, he emits also the idea that XS core modules may be ported to an Inline::C implementation. Brian Ingerson (author of Inline) acknowledges that Inline::C can be used at perl build-time, but he thinks that the only thing that needs to be installed on a user machine is enough of Inline.pm to invoke Dynaloader. About 30 lines of code. He adds that this is the only part of Inline that he would like to see distributed with 5.9.
http://groups.google.com/groups?threadm=20020801161356.GD1064%40ool-18b93024.dyn.optonline.net
Finally, Schwern suggests to eliminate the changelogs and the old perldelta manpages from the perl distribution (those are pretty huge and can be distributed separately), and to provide an alternate bzip2 tarball of perl.
Philip Hazel reported a couple of bugs about the \C regexp anchor. (Bugs #15763 and #15774).
Ilya Zakharevich reported that list assignment with common variables on both sides (e.g. ($a,$b) = ($b,$a)) don't work with aliased variables (e.g. ($a,$b) = ($c,$d) where $c is an alias to $b and $d an alias to $a.) Bug #15667.
Craig Berry noticed that extension building against an installed Perl on VMS is broken in 5.8.0. He provided a patch, and said also that in lieu of the patch, folks can either build extensions against an uninstalled Perl or simply copy the missing files manually.
Elizabeth Mattijsen reported a memory leak in threads::shared. Hopefully this problem will only need a patch to threads::shared, that can be released to CPAN. (Bug #15893.)
Bleadperl now reports its version as 5.9.0, and Hugo began to apply patches. We're waiting for the first snapshot...
This summary brought to you by Rafael Garcia-Suarez, from Lyon, France. -- Nota. I'll be on vacation (and off-line) for two weeks in August : approximately from 12th to 25th. So there will be no summary for those weeks. If you find those summaries helpful, and if you want to become a superhero, take over the report for two little weeks! Please drop me a line at rgarciasuarez at free.fr. (If you find those summaries helpful but don't want to take them over, go send money to YAS. Perhaps they'll hire an interim writer.)
This report is also available via a mailing list, which subscription address is perl5-summary-subscribe@perl.org.