This Week on perl5-porters (24-30 May 2004)
This week, you'll read in this summary more about the uninitialized
warning plans for 5.8.($n+1), some XS tricks, intriguing bugs, and the
different types of UIDs.
Uninitialized warnings in maint
Nicholas Clark thinks about a way to backport in maintperl the enhanced
bleadperl warning, "Use of uninitialized value $foo at...". However,
they're not 'stable' enough to be enable the extra information by
default. He lists a couple of solutions (add a class of warnings, or a
new pragma, or a global flag.) Rafael favors a global flag, for example
a (version-specific) PERL_5_10_ISH environment variable, while Nicholas
envisions some idiom "use more 'future'".
http://groups.google.com/groups?selm=20040524161941.GG1383%40plum.flirble.org
Undefined scalars in XS code
Stas Bekman asks about a wrong XS idiom, namely testing the definedness
of an SV by comparing its address with &PL_sv_undef; this works when the
undef() literal value is passed to the XS code, but not when a random
undefined scalar is used. Gisle Aas points out that SvOK() should be
used instead; Stas produces a documentation patch.
http://groups.google.com/groups?selm=40B59C7E.6090902%40stason.org
FileCache problems
Alan Burlison finds that FileCache (a little-known core module, meant
to cache a large number of open filehandles) resets some signal handlers
when reopening some files which have been closed in the mean time (to
avoid exceeding the limit on the allowed amount of filehandles on the
system.) Jerrad Pierce answers that this is by design, for the case the
filehandle to be reopened is a pipe to a command. Alan disagrees with
this approach (and finds that using FileCache with a pipe is a weird
idea anyway) and suggests to remove the %SIG assignments.
http://groups.google.com/groups?selm=40B1D409.9090109%40sun.com
More UTF-8 vs taint
Stas Bekman filed bug #29841, about utf8::decode() not working under
-T. Dave Mitchell diagnoses the symptoms, but doesn't know what is the
right fix.
http://groups.google.com/groups?selm=rt-3.0.9-29841-88083.1.56655856690257%40perl.org
UIDs and GIDs
Paul Fenwick announces that he's working on a module (Proc::UID) to
manipulate Unix user ids: the famous real and effective UIDs, which have
their own Perl variable counterparts, as well as the saved UID and the
filesystem UID. But since perl caches the results of getuid()/getgid()
in internal interpreter variables, the contents of the perl variables
$<, $>, $( and $) cannot be always believed (they might have been set
via an XS module). Thus Paul asks whether this cache is really a wise
implementation.
http://groups.google.com/groups?selm=40B402C7.8060209%40perltraining.com.au
Paul also wonders why saved UIDs aren't accessible natively from perl
like real and effective UIDs. Rafael gives some ideas about adding an
API for them.
http://groups.google.com/groups?selm=40B815C7.3050104%40perltraining.com.au
In Brief
Nicholas released a maintenance snapshot of perl 5.8.x, and adds: "The
plan is for 5.8.5 code freeze at midnight (GMT) on 30th June, with RC1
soon after."
http://groups.google.com/groups?selm=20040530225534.GN1147%40plum.flirble.org
Sam Vilain finds a macro in the glibc (makedev()) that isn't correctly
processed by h2xs (bug #29969). This was fixed by Wolfgang Laun.
Lionel Cons finds an incompatibility between the innards of Carp and
the Safe module (bug #29851). Rafael proposed a patch.
Andy Lauder reported that in some cases, perl seems to perform a seek()
before a close() (bug #29883). This is in fact a behaviour of Solaris
that is reproducible with a C program, so perl isn't at fault; Alan
Burlison gives a rule of thumb: "you shouldn't use exit() in a child
process that doesn't exec(), you should use _exit()."
Using threads::shared prevents the profiler Devel::DProf from
working (bug #29939).
The autouse bug (#29708) reported last week was fixed by Dave
Mitchell.
Encode 2.01 was released.
About this summary
This summary was written by Rafael Garcia-Suarez. Weekly summaries are
published on http://use.perl.org/ and posted on a mailing list, which
subscription address is [EMAIL PROTECTED] Comments and
corrections welcome.
