235045 - collectstats.pl crashes on bugzilla running against CVS code

Status: RESOLVED INVALID

(Bugzilla :: Reporting/Charting, defect)

Description

[corporate_gadfly]

User-Agent:       
Build Identifier: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.6b) Gecko/20031216 Firebird/0.7+

First time reporter, so please be gentle.

I'm running bugzilla CVS on a Gentoo Linux (kernel 2.4.24) machine. My perl
version is 5.8.3.

Last 3 nights I've gotten the following from my nightly cron job which runs
collectstats.pl

free(): invalid pointer 0x873ccf4!

Reproducible: Always
Steps to Reproduce:
Running collectstats.pl manually gave me a Segmentation fault.

Reading some of the other bug reports on collectstats.pl I happened to notice
the --regenerate option. Running with --regenerate gives me:

Regenerating -All- [100.0%] - 00:00:06
Regenerating SWS [100.0%] - 00:00:05
Not a HASH reference at /www/htdocs/bugzilla/collectstats.pl line 471.
Attempt to free unreferenced scalar: SV 0x86a9b68.
Attempt to free unreferenced scalar: SV 0x8773f10.
Segmentation fault

Successively running collectstats.pl with no flags give me just plain
Segmentation faults.

Anything else I can report?
Actual Results:  
Segmentation fault

Expected Results:  
Complete the task with no output.

Comment 1

[timeless]

you might try valgrind, you'll want a deep stack trace
VALGRIND_OPTS='--error-limit=no --num-callers=20 -v --leak-check=yes
--show-reachable=yes --suppressions=/mnt/ibm/mozhack/xpcom.supp'
is what I use, you can drop the leakcheck/showreachable bits, they're for when a
program runs too completion :)

but most likely you'll need to report this upstream
it's possible that gentoo's compilation of something resulted in a bad behavior,
not very likely.
more likely some c/c++ module (possibly perl, although i'd bet on a binary
component) is buggy (mozilla seamonkey is like that).

you'll want to get a stack trace, valgrind should give you a chance to attach
with gdb when you get to the double free, you could do that, although the
valgrind stack traces might be sufficient.

Comment 2

[corporate_gadfly]

Attachment: valgrind log

Thanks in advance for your patience. Not familiar with valgrind, so I might
need some hand holding.

I used the following command line:

valgrind --error-limit=no --num-callers=20 -v ./collectstats.pl

Find attached the output log.

Comment 3

[Gervase Markham [:gerv]]

CCing timeless, who is dealing with this issue.

Gerv

Comment 4

[timeless]

*shrug*, can you test a normal perl binary (same conditions), also perhaps you
should read about how to file a bug about perl. i'm out of my league.

Comment 5

[Bradley Baetz (:bbaetz)]

Yeah, this is a perl bug not a bugzilla bug. Could be due to anything; try
reporting it into gentoo's tracking system as a perl bug.

Reopen if this happens in stock perl on a stock system, and theres a workarround
we can put in the code.