On Thursday 19 June 2003 10:54, cmpilato@collab.net wrote:
> Christian Daudt <csd_ob@daudt.org> writes:
> > Hi,
> > I'm starting to roll out subversion for a number of people in my group
> > and I've run into many of them getting coredumps on checkouts (of course
> > I never got one...). Anyways, I've tracked it down to parsing this piece
> > of xml being received: <S:open-directory rev=\"-1\">
> > And with that, start_element in fetch.c (in libsvn_ra_dav) is running
> > this piece of code:
> > name = get_attr(atts, "name");
> > /* ### verify we got it. punt on error. */
> > svn_stringbuf_set(rb->namestr, name);
>
> While I appreciate the patch attempt (omitted for brevity), ultimately
> I think the problem should be solved a little more directly. That is,
> we should figure out why we getting an open-directory with no "name"
> attribute. The code obviously thinks this case shouldn't occur.
>
I agree completely - that's why I said my patch was only addressing the
consequence, not the cause . But the fact is that, regardless of the problem
that caused the code to call stringbuf_set with null (which remains to be
fixed) , svn_stringbuf_set shouldn't coredump because it received a value it
doesn't like. It should either set the string to empty (which is what my
patch did) or return an error and then get the callers to test for that
error.
csd
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu Jun 19 20:07:38 2003