You must be in the hot seat over there. It must be frustrating.

On Thu, 2006-11-23 at 13:46 +0100, Oliver Schäfer wrote:
> Hello again,
> 
> I've got a problem, I need a quick solution for this problem, as users are
> getting more and more impatient.
> I will have to disable ScmBug, meaning I can't continue debugging if there's
> no solution at the horizon.
> 
> Any suggestions?

The educated guess I could make is that svnlook in our pre-commit hook
is causing problems in Subversion over Apache. You are indicating that
sometimes the integration works, and sometimes it doesn't. I'm wondering
if a slow or fast response from Apache+WebDAV (I don't know much about
them) is causing this.

My feeling is that people haven't experimented with writing hooks enough
to perhaps reveal some bugs in Subversion, and you may have hit some
weird one. Scmbug is causing it, definitely, but it's probably
uncovering a bug in Subversion. We don't try to do anything fancy, or
hard with svnlook. I like to believe we followed the Subversion
guidelines in using it, which are summed up in: don't modify a
transaction. We don't. We only inspect it, and svnlook is supposed to be
read-only. 

For example, in Subversion.pm, add this line at the top in
prepare_activity_verify_or_commit:

exit 0;

and see if the commits go through. Exiting with a zero there would mean
that svnlook is not called at all.

Perhaps running svnlook within Apache uncovers a timing bug, or locking
bug, or something that's still causing this. Have you asked the people
in the Subversion mailing list ? 



_______________________________________________
scmbug-users mailing list
[email protected]
http://lists.mkgnu.net/cgi-bin/mailman/listinfo/scmbug-users

Reply via email to