This Week on perl5-porters (27 October / 2 November 2003)

This Week on perl5-porters (27 October / 2 November 2003)

The big news of the week is of course the first release candidate of perl 5.8.2, the problems it solves, and the new problems it causes.


Nicholas Clark, the new maint pumpking (if you have missed previous episodes), released perl 5.8.2-RC1.
    first perldelta draft :

Alexey Tourbin found a problem with this latest version : an incompatibility is caused by a patch to committed at the beginning of the month. This affects BerkeleyDB's test suite, but can arguably be caused by BerkeleyDB's reliance on the specific behaviour so far of UNIVERSAL::import(), that was inherited by subclasses. Paul Marquess is going to fix the next release of BerkeleyDB. Nicholas backed the change out of maint for 5.8.2.

Another compatibility problem was caused in mod_perl by a patch by Rick Delaney that changed the behaviour of require() regarding files that one try to load again after a first unsuccessful attempt. This one also was backed out of maint.

The adventures of binary compat

There was also a long, technical discussion about mod_perl 2 and the new hashing scheme in perl 5.8.2 (rehash hashes only if a bucket contains too many entries), that was introduced for binary compatibility. These threads involved also security considerations and tests for potential attacks against this hashing schemes. I'll shamelessly refer the interested parties to the archives :

Meanwhile, Jan Dubois spotted another potential binary incompatibility between 5.8.0 and 5.8.1, the problem being related to the reentrant APIs. His proposed change doesn't solve all issues (XS modules compiled with a threaded 5.8.1 and that use the reentrant system calls will have problems) but it makes the situation better, as it restores the compatibility between 5.8.0 and 5.8.2. Details to be supplied in the next iteration of perldelta.

Threads and -D

Stas Bekman produces a segfault by using the -Dm debug flag on a threaded perl, and suggests a patch. Rafael says that spotting problems with perl's debugging flags could be done by running perl's test suite with PERL5OPT=-D... (at least for the tests that aren't run under taint mode).

Attribute::Handlers and UNIVERSAL

Michael Jacob notices that Attribute::Handlers is a bit cavalier with UNIVERSAL, pushing itself into @UNIVERSAL::ISA. Thus, all classes now can() AUTOLOAD(), DESTROY() and import(). To fix this, he suggests to override UNIVERSAL::can() to reject inherited methods. But Damian Conway says he hasn't a good solution for this problem.

Setting group and user ids

Alan Burlison remarks that assigning to $) and then to $> doesn't produce the same result than assigning to the list ($),$>) (at least on Solaris.) Rick Delaney sends a patch, but points out that it's hard to write tests for this.

Selected bugs

Abigail reported (bug #24313) that the numeric constant 0 can't be overloaded via overload::constant('integer') like other integers. This was fixed by Rick Delaney.

Ton Hospel produces a segfault by attempting to modify an array slice with -= while the array is being assigned to. Michael Jacob explains why this happens, but there's no fix in sight. (bug #24342.)

Ton Hospel also files bug #24346, about the substr() lvalues. He notices that he's able, when operating on a substring, to pull in characters from the rest of the said string (i.e. from outside the substr() window.) As the docs are unclear, he can't say he's sure it's a bug, but that's at least a potentially dangerous behaviour. Graham Barr proposes a patch.

David Dyck has some problems with a snippet of code where he's eval()'ing a string that contains a scalar variable in an END block (bug #24296). Dave Mitchell explains what's happening and provides some insight about how closures work.

Yitzchak Scott-Thoennes demonstrates that list assignment to a hash in list context has some problems (bug #24380).

In Brief

Arthur Bergman considers deprecating the SvPVX macro.

Jeff Pinyan has a question about how the regular expression engine treats terms that match an empty string in an alternation. Mark-Jason Dominus provides some answers.

ExtUtils::MakeMaker 6.18 is coming soon, announces Michael Schwern, listing the new improvements and features :

Nicholas explicits his maint patching policy.

About this summary

This week was summarized by Rafael Garcia-Suarez. Summaries appear weekly on and on a mailing list, which subscription address is Corrections and comments are welcome.