This Week on perl5-porters (14-20 April 2003)

This Week on perl5-porters (14-20 April 2003)

Not a good week to stop smoking ! Test-wise, that is. In this week's summary, read about configuration changes, language proposals, and the usual amount of bug fixes.

Binary error messages

Marc Lehmann remarks that the current maintperl may produce error messages that contain a null character (bug #21962), with the one-liner :

    ./perl -e '&{""}'

Abhijit Menon-Sen finds out that perl, trying to access a symbol that has an empty name, sets explicitly the length of this name to 1, which seems obviously wrong, and thus fixes the bug by suppressing this behaviour.

Long Doubles

There was a long thread, mainly involving Jarkko and Merijn -- the Configure guys -- about Configure's support for long doubles. A patch for long double support on Solaris (by Alan Burlison) was integrated. An important question was raised, which was : if we ask Configure to enable long double support in perl, (or, similarly, 64 bit integers), and if Configure fails to detect support for it on the platform, should it abort ? The answer, for now, is yes. As a consequence, some holes appear in the smoke matrices, corresponding to now-disallowed configs.

Ties, Fetches and Storabilization

The patch that Dave Mitchell sent last week, to allow nested FETCHes, has been applied. The benchmarks give varying results, which apparently are due, as Dave says, to the usual cache noise effects, which have been encountered before. So this patch probably doesn't change significantly the (already low) performance of ties.

Speaking about ties, Storable now handles self-tied objects (bug #17040, fixed), thanks to Abhijit. The question of serializing tied objects is raised again ; David Nicol proposes the convention to use SERIALIZE and DESERIALIZE special methods. Slaven Rezic points out that Storable uses already the STORABLE_thaw and STORABLE_freeze methods, if they're available.

More on cross-compilation

Redvers Davies posted refactored patches to add support for cross-compilation to perl. Jarkko is considering merging them into perl 5.8.1. Vadim Konovalov reminds us about the module, created during the cross-compilation process for WinCE, which is basically a substitution for when cross-compiling (either perl itself, or perl modules).

More black smoke

Nicholas Clark is struggling with strange smoke failures on BSD. Unreproducibly, for some reasons, Makefile.PL expansion is going wrong under crontab.

In Brief

Nathan Torkington notices that reading from a magic scalar with PerlIO::scalar sometimes causes coredumps. This buglet is fixed by Enache Adrian.

Jarkko Hietaniemi finished to add a new function PerlIO::get_layers(), to retrieve the I/O layers for a given filehandle. Dan Kogai is excited about it.

Bug #21979 demonstrates that, with an ithread-enabled perl, modifying an alias to a read-only item of an argument list of a subroutine works, while it's expected to fail.

Arthur Bergman and Dave Mitchell are struggling with pads, ithreads, cloning, and closures, and finally fix an obscure bug I don't understand.

Shlomi Fish posted a proposal to add three new keywords, class, up and this, to manipulate namespaces. He didn't get positive feedback.

About this summary

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