On Feb 1, 2008 7:34 PM, C. Michael Pilato <cmpilato_at_collab.net> wrote:
>
> David Glasser wrote:
> > On Feb 1, 2008 4:31 PM, David Glasser <glasser_at_davidglasser.net> wrote:
> >> This revision caused the segfault described in
> >> http://svn.haxx.se/dev/archive-2008-02/0064.shtml.
> >>
> >> (Ah, binary search.)
> >>
> >> (Ah, pools.)
> >
> > I've tracked it down a little more, to svn_client_merge_peg3:
> >
> >>> @@ -4801,6 +4833,7 @@
> >>> const char *wc_repos_root, *source_repos_root;
> >>> svn_opt_revision_t working_rev;
> >>> svn_ra_session_t *ra_session;
> >>> + apr_pool_t *sesspool;
> >>>
> >>> /* No ranges to merge? No problem. */
> >>> if (ranges_to_merge->nelts == 0)
> >>> @@ -4828,9 +4861,10 @@
> >>> &working_rev, adm_access, ctx, pool));
> >>>
> >>> /* Open an RA session to our source URL, and determine its root URL. */
> >>> + sesspool = svn_pool_create(pool);
> >>> SVN_ERR(svn_client__open_ra_session_internal(&ra_session,
> >>> URL, NULL, NULL, NULL,
> >>> - FALSE, FALSE, ctx, pool));
> >>> + FALSE, TRUE, ctx, sesspool));
> >>> SVN_ERR(svn_ra_get_repos_root(ra_session, &source_repos_root, pool));
> >>>
> >>> /* Normalize our merge sources. */
> >>> @@ -4838,6 +4872,9 @@
> >>> source_repos_root, peg_revision,
> >>> ranges_to_merge, ra_session, ctx, pool));
> >>>
> >>> + /* We're done with our little RA session. */
> >>> + svn_pool_destroy(sesspool);
> >>> +
> >>> /* Do the real merge! (We say with confidence that our merge
> >>> sources are both ancestral and related.) */
> >>> SVN_ERR(do_merge(merge_sources, target_wcpath, entry, adm_access,
> >
> > Swapping this svn_pool_destroy with the do_merge that follows it saves
> > the day. Perhaps merge_sources contains something allocated in
> > sesspool?
>
> This might be another of those instances where the repos_root's pool is
> prematurely destroyed. In another of a series of "let's design a
> non-obvious API" decisions, svn_ra_get_repos_root() apparently returns its
> value allocated in the session pool, *not* in the apparently unused
> passed-in pool (like svn_ra_get_uuid does). I can almost guarantee that
> this bit me in this change, and likely in other changes I've made, too.
That was what I thought too at first, but after hours of debugging I'm
less convinced. I think it has to do with Neon stuff.
--
David Glasser | glasser@davidglasser.net | http://www.davidglasser.net/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe_at_subversion.tigris.org
For additional commands, e-mail: dev-help_at_subversion.tigris.org
Received on 2008-02-02 06:09:12 CET