This Week on perl5-porters (15-21 September 2003)

This Week on perl5-porters (15-21 September 2003)

Those weeks, the perl 5 porters are in maintenance mood. Or in maintenance mode, if you prefer. Read about the progress made in the 5.8.x and 5.6.x branches.

A strange error case in bleadperl

Stas Bekman posted a code snippet that produces the fatal error Modification of a read-only value attempted with bleadperl, but not with perl versions <= 5.8.1. (Filed as bug #23803.) It can be reduced to :

    for("a") { for $x (1,2) { local $_="b"; s/(.*)/+$1/ } }

This was the consequence of a previous bug fix. Rafael explained that in perl 5.9.0, any attempt at changing the value of a localized readonly magic scalar will fail with this error. Before, it used to silently do nothing, and the assignment $_="b" was a no-op : this could lead to difficult and well-hidden bugs. In bleadperl, readonlyness of magical scalars is now preserved when localisation occurs.

But, will you say, where is the magic in the above snippet ? In fact, it appears that the substitution s/(.*)/+$1/ adds some magic to the scalar stored in $_. This kind of magic is used internally by the regular expression engine. Thus, one can say that the behaviour reported by Stas is an unfortunate side-effect of the implementation of regular expressions.

Configure and gcc 3.4

Enache Adrian, watchful on the portability front, forwards a mail from the gcc development mailing list. A change in gcc's behaviour in the 3.4 development branch leads to breaking perl's Configure script ; more precisely, it breaks the way Configure tests for the existence of a function in the standard library.

This was followed by a scary C wizardry thread. Enache provided a fix.

In Brief

Jarkko Hietaniemi released several snapshots of perl 5.8.1. The next release candidate approaches. Extensive tests have been done with mod_perl 1 and 2, and this uncovered several problems, both in mod_perl and in perl.

Speaking about snapshots, Rafael released one of perl 5.6.2. It's OK on a large number of platforms ; adjustments are needed on Mac OS X and Cygwin at least. Rafael incremented the version number, so it reports itself as being 5.6.2.

The bulk of the discussion focused on portability on various platforms -- mainly the non-Unix ones, VMS, OS/2, Windows and Cygwin. Peter Prymmer reported that perl now looks good on VMS 7.3-1, with no regression test failing.

Ilya Zakharevich and Yves Orton reported a few bugs in Test::Harness.

Stas Bekman noticed that with perl 5.8.1, if a socket gets closed by the server while printing to it, the client gets a SIGPIPE. It wasn't happening in perl 5.8.0 or earlier. (This is considered a feature, not a bug.)

Michael Schwern released version 2.03, which fixes an obscure (and ancient) bug on attribute inheritance, when an intermediate class has no fields.

About this summary

This week's summary was produced by Rafael Garcia-Suarez. Summaries are published weekly on and on a mailing list, which subscription address is Corrections and comments are welcome.