Front page | perl.perl5.porters |
Postings from December 2008
G'day everyone,
I've been working with the Debian Perl list, who have found an oddity with
how Perl works with $?
With thanks to Niko Tyni:
# perl -le 'system($^X, "-e", q/kill 6,$$; sleep 1/); print "version=$^V
\$?=$? native=${^CHILD_ERROR_NATIVE}"'
Signal 6 is a SIGABRT, which most systems interpret as a request to dump
core. This gives us:
version=v5.10.0 $?=6 native=134
version=5.008008 $?=134 native=
Notice that Perl 5.8 has the expected $? value of 134 (128 for a core dump,
plus 6 for the signal). Perl 5.10 seems to strip the core dump info out of
$?, but provides the original value in ${^CHILD_ERROR_NATIVE}.
At least from my understanding about how $? *should* be working, and the
documentation in 'perldoc -f system', this looks like a bug. Any code
(including my own) that relies upon testing $? & 128 to detect core dumps is
going to fail.
Unfortunately I'm still travelling, so I don't have a Unix machine with 5.10
handy to test this. May I ask:
a) If anyone else can produce the same results on a
non-Debian system.
b) If this is considered a bug. (If not, we need to
update perlfunc.pod)
c) If I can rely upon 5.10 setting ${^CHILD_ERROR_NATIVE} to
what I consider to be the "correct" value (so I don't have
code that works around the problem on 5.10.0, but breaks
on 5.10.1).
If this is considered a bug, I can file a report in RT.
Many thanks to Gregor Hermann, Jonas Smedegaard, Jonathan Yu, and Niko Tyni
for tracking down this problem.
All the very best,
Paul
--
Paul Fenwick <pjf@perltraining.com.au> | http://perltraining.com.au/
Director of Training | Ph: +61 3 9354 6001
Perl Training Australia | Fax: +61 3 9354 2681
Thread Next