p5pRT
commented
16 years ago From William.Ricker@FMR.COM
Created by a273121@fmr.com
Summary -
A uninitialized undef Value in while($mightbeundef ... ){... results in warning that have strange line number\, in some cases.
User ID --
a273121@fmr.com on Linux = William.Ricker@FMR.com more generally = is also known as use.perl.org/~n1vux \, Boston.pm type person
Version + BugLevel -
BOTH 5.10-RC1 (Linux) + 5.8.8/5.9.5 (Windows)\, possibly others.
"Medium" may be over-high for a misleading warning that 5.10 *has* improved. However\, this Bug applies to 5.8.8 as well\, where the better messages in 5.10 doesn't help figure out the bad line number\, so more serious on 5.8.8. I'd be happy with LOW on 5.10 and MEDIUM on 5.8.8.
Figured we should send it in during 5.10-RC1 so you have a chance to get the fix into 5.10 if it's easy\, since we can reproduce it in RC1.
Speculation -
This may be related to code-motion optimization in loop\, as it uses the line# of last statement of body as line# for control expression many lines before in source. (Simplest versions of code get correct line #.)
Background / detail -
A user adopting a complex non-CPAN published module (COBOLIO.pm) reported a strange uninitialized value warning for a line unrelated to any variable.
Inserting !defined in loop control expression of the loop of which the line numer given was last executable statement made the warning go away.
(This *will* [has been?] be reported to module author separately as a module bug. This report is about the *bad* line# on the warning.)
Using locally built 5.9.5 (Windows) or 5.10.0_RC1 (Linux) gives a better warning\, with name of variable\, great progress\, but still not right line number.
REPRODUCTION TESTCASE --
Our first attempt to remove all other bits of the module to make a simple test case failed; the simplest test case warning warned the "right" line. But we found something just simple enough.
Here's the simple reproduction that tricks it into appearing. The following should warn on use at line 12\, warns at 14.
[perl-5.10.0-RC1]$ cat bad2.pl #! bleedperl/bin/perl
use strict; use warnings;
our @recLevel= (); our $currLevel=9; our $level=3;
sub foo { my (@elsewhere\, $c); # temp while($currLevel > $level) { # 12 $c = $currLevel = pop @recLevel; # 13 push @elsewhere\, $c; # 14 } # 15
}
foo();===EOF===
[perl-5.10.0-RC1]$ perl510 bad2.pl Use of uninitialized value $currLevel in numeric gt (>) at bad2.pl line 14.
Note there is no $currLevel in line #14. (If not using $c temp\, there would be and it would be MORE confusing!)
IMPACT UNDER 5.8.8 --
If any other uses of '>' occur in loop\, the 5.8.8 version of this message is
Use of uninitialized value in numeric gt (>) at bad2.pl line 14.
Which is even more misleading. This confusion was such that our user downloaded the tarball for 5.9.5 to get the 5.10 better message\, which solved the conundrum. Hence I think this is long term more important for a 5.8.x maintenance release; if it is triaged to miss 5.10.0 and is fixed in 5.8.9 and ported to 5.10.1\, that's completely understandable\, we just want the p5p's to have the option.
Migrated from rt.perl.org#47632 (status was 'open')
Searchable as RT47632$