Greg Stein wrote on Wed, Apr 25, 2012 at 15:25:59 -0400:
> On Apr 25, 2012 8:16 AM, <rhuij...@apache.org> wrote:
> >...
> > +++ subversion/trunk/subversion/libsvn_subr/sqlite.c Wed Apr 25 12:16:12
> 2012
> > @@ -718,8 +718,23 @@ close_apr(void *data)
> > for (i = 0; i < db->nbr_statements; i++)
> > {
> > if (db->prepared_stmts[i])
> > - err = svn_error_compose_create(
> > + {
> > + if (db->prepared_stmts[i]->needs_reset)
> > + {
> > +#ifdef SVN_DEBUG
> > + const char *stmt_text = db->statement_strings[i];
> > + stmt_text = stmt_text; /* Provide value for debugger */
> > +
> > + SVN_ERR_MALFUNCTION_NO_RETURN();
> > +#else
> > + err = svn_error_compose_create(
> > + err,
> > + svn_sqlite__reset(db->prepared_stmts[i]));
> > +#endif
>
> I think the reset should be outside the #ifdef. Couldn't the app install an
> "ignore" malfunction handler and fall through? If so, we want the statement
> reset.
No. SVN_ERR_MALFUNCTION_NO_RETURN() doesn't return. (It abort()s if
the handler returns)
