Re: [Lurker-users] Debugging signal 11 program faults

[Wesley W. T. <we...@te...>]

On Wed, Apr 02, 2003 at 04:42:58PM +0800, Federico Sevilla III wrote:
> <stuff about the filesystem>

This is irrelevant; don't change your filesystems. I am positive it is a bug
in lurker from the new mbox added, and I believe it has now been corrected.

> The Lurker database I was previously using was built when the system was
> running ext3, and seemed to be ok. Recently I purged the database to add
> some "back-issues" of the linux-xfs mailing list to the archives, and
> now cannot rebuild the database. Lurker keeps getting a signal 11
> program fault, traps it, and syncs the database to disk. This continues
> to happen when I rerun lurkerd, whether I purge the database or let it
> resume where it left off.

It sounds like one of the back-issue messages you added causes lurker to
segfault; probably corrupt in a way I don't handle.

> I am using the lurkerd_0.1f-5 Debian package. I rebuilt the package with
> only one change: I added "-g" to the CFLAGS to allow gdb to do tracing.
> I could not run lurkerd from gdb, because it would just exit normally
> without doing anything. So I started lurkerd the standard Debian way,
> which makes it run as the www-data user, then attached to this process
> manually using gdb, also as the www-data user.

This is because lurker detaches at startup; add the -n option:
gdb$ run -n

> First I got a lot of messages[1] about the attachment process which
> seemed to work fine, although there was a permission denied error at
> mbox.c line 220 for which I have a backtrace[2]. I made gdb continue,
> and it took awhile, seemingly doing its real work. Then I hit the
> SIGSEGV[3], at line 172 of indexer.c, for which a backtrace is also
> available[4].

This is useful; looking into it... My line 172 is empty. :-)

I presume it is the:
	my_indexer_traverse(in, body->nested.msg->body);
line?

Sounds like there is no body->nested->msg->body.
Which means that the email has a corrupt embedded email.
I have added a check to ignore messages in this case.

Please try the latest cvs; I believe I have now fixed this bug.

-- 
Wesley W. Terpstra <we...@te...>