This Week on perl5-porters (21-27 April 2003)

This Week on perl5-porters (21-27 April 2003)

This week's summary presents a nice variety of language issues. Read about some new errors, documentation patches, bug closing and configuration.

Don't assign to %::

Dave Mitchell proposed to change assignments to %:: to be a compile-time error (currently, this may cause crashes, and a non-warned user may fall in this trap). Enache Adrian objected that this doesn't prevent all nasty things that may be done with the main stash, and thus was unnecessary bloat. Arthur Bergman proposes to mark %:: as read-only instead. Hugo van der Sanden feels this patch is actually trying to hide other problems.

Use of freed value in iteration

It's a well-known fact that you can crash perl by modifying an array over which a for loop iterates. Dave Mitchell changed this into a new fatal run-time error, Use of freed value in iteration.

Localization of lvalues

Dave Mitchell (who got bored and was browsing the bug database) reminds the crowd that local($x='bar') is currently equivalent to $x='bar' (i.e. the local() is silently ignored). This is not the right thing ; it should either produce some error, or localize $x, before or after the assignment occurs. The general opinion is to disallow it, purely and simply. Rafael, noticing that other lvalues can be localized (e.g. pos or substr, although not very reliably sometimes), thinks this should be fixed to work. Hugo, however, says that the most we can do is to add a deprecation warning.

Slice auto-extending

But Dave didn't stop there. He also asked about whether splice() should extend the array it operates on (bug #1832). The consensus was to fix the docs, which are unclear.

Restricted hashes strike back

Jarkko asks for someone to solve the remaining problems about restricted hashes -- that replace pseudohashes in the implementation of the fields pragma in the perls to be released. And this someone will be ... Dave Mitchell ! (More precisely, the problem is with list assignment to a restricted hash.) Dave sends a patch that modifies the way a restricted hash is cleared : all keys are set to placeholders. So you can now assign a list to a restricted hash ( %h = (key => 1) ). However, if this list contains a disallowed key, %h is left empty rather than with its original contents.

Documentation effort

Casey West announces that he's going to do something useful for documentation, and posts a list of goals. Notably, he's willing to coordinate and encourage efforts. Then, proving it, he posted a large number of documentation patches, closing a large number of bugs.

Meanwhile, Nick Ing-Simmons agrees that the PerlIO / open / binmode / perliol docs need to be cleaned up so that they fit more nicely together.

Duplicating the DATA filehandle

Josh Purinton notices (bug #22010) that reading from a dup'ed DATA filehandle, by doing open(F,"<&DATA"), in fact doesn't read anything. (Further investigation indicates that this only occurs when the environment variable PERLIO is set at stdio.) Nick Ing-Simmons points out that when the dup is done, the file pointer of DATA is at EOF, and that dup'ing DATA is not going to copy the DATA buffer.

Lists in scalar context

Bug #22027 is merely a question, asked by Mike Stok : the value of $x, after this statement :

    $x = (1,2,3,());

is undef. Is this the right thing to do, or should the final parentheses be ignored, hence setting $x to 3 ? Graham Barr explains that a list in a scalar context evaluates each element in a scalar context before any flattening is done.


Nick Ing-Simmons asks whether sfio support can be removed. Andy Dougherty and Stas Bekman point out that it's used on some mod_perl installations, because under mod_perl STDOUT is tied (unless perl is compiled to use the sfio library), and formats don't work with a tied STDOUT. Mark Mielke considers that writing a CGI and expecting it to work under mod_perl is naive anyway.

prebind on Mac OS X

Nathan Torkington asks about the use of the -prebind linker option on Mac OS X. Dan Kogai tells him to use the update_prebinding command, which should optimize the loading of XS modules (and other dynamic libraries). Wilfredo Sánchez adds that adding -prebind to the linker flags can't hurt.


Dave Mitchell fixed a case of memory corruption with the goto &sub construct.

Abe Timmerman released Test-Smoke v1.17, with loads of improvements.

Peter Scott produces a mysterious case (filed as bug #21999) where $1 is modified by an unsuccessful regex match, inside a for loop.

Several people walked through the bug database, closing bugs to be closed, or proposing fixes for old bugs. Thanks to Robert, Ask, Arthur, Dave, Casey, and probably others.

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.