This Week on perl5-porters (10-16 November 2003)

This Week on perl5-porters (10-16 November 2003)

For some reason, (maybe the approach of mod_perl 2), one of the main topics this week was interpreter cloning and the problems it may cause. But the release of perl 5.6.2 is an event that is not to be neglected. Read on.


The latest episode in our maintenance series is perl 5.6.2, which was released saturday. After the RC1 last week, some remaining bugs were shaken out, and some of them solved. The announcement on P5P is here :

Perl 5.6.3 will eventually be a more advanced release, including a load of bugfixes and module upgrades (while 5.6.2 is merely oriented toward fixing compilation issues.)

Cloning problems

Bug #24463 reports that a DBI test is failing under perl 5.8.2-ithreads built with debugging capabilities (-DDEBUGGING). This is due to a regression in the cloning of interpreters, that was introduced to fix a segfault when using the -D switch to perl. The problem here is that the new interpreter that is being cloned hasn't yet perlio layers, and thus can't use print debug messages ; but trying to set the context back at the perlio layers of the pristine interpreter leads to this breakage.

Stas Bekman reported that Config::myconfig() is prone to fail with a fatal error under ithreads, with perl 5.8.[12], due to its internal use of a :unique variable (made readonly when a new thread is started). Elizabeth provides a fix. (It appears that mod_perl 2 is a great tool to find thread-related bugs.)

Stas reported also another problem with Scalar::Util::weaken() : clones of weak scalars are messed up. This happens since perl 5.8.1.

Following to another question by Stas about the CLONE method, Elizabeth introduced her module Thread::Bless.

Empty subroutines

Elizabeth Mattijsen notices that using an empty subroutine as a method may produce segfault in some specific cases with perl 5.8.2. Enache Adrian provides a fix.

C++ bindings

Chip Salzenberg reports as bug #24468 that the headers of perl 5.8.2, with threading turned on, can't be compiled from within C++ sources. This is a problem for C++ applications that embed perl. Jan Dubois suggests a fix, which Chip is going to try.

In Brief

Nicholas Clark notices that creating threads in BEGIN blocks is unsupported. He disapproves this constraint.

Nicholas also found and fixed obscure bugs with hashes and UTF8 keys. He proposes to refactor some hash-handling code.

Alan Burlison notices that the MANIFEST.SKIP distributed with perl 5.8.2 (as part of ExtUtils::Makemaker) is unfortunately broken. Rafael takes the blame.

Nick Ing-Simmons suggests to investigate using mapfiles on Linux for the compilation of perl. Alan Burlison nods.

A version of the IRIX compiler can't build perl 5.8.1 or 5.8.2 (bug #24471). Rafael suggests it's a bug in the C preprocessor.

New modules

Andy Lester released Test::Harness 2.36. This version includes a command-line utility, prove, that can run individual tests. Its inclusion into the core still causes a few unresolved problems.

Michael Schwern released ExtUtils::MakeMaker 6.21, mostly to fix building modules that call dir_target() or completely rewrite top_targets().

About this summary

This summary was brought to you by Rafael Garcia-Suarez. Weekly summaries are available on and via a mailing list, which subscription address is Feedback (and patches) welcome.