Re: svn commit: r26813 - trunk/subversion/libsvn_client

On Thu, 27 Sep 2007, pburba@tigris.org wrote:
...
> Follow-up to r26803, fix some potential segfaults.
>
> * subversion/libsvn_client/merge.c
> (drive_merge_report_editor): Be consistent with our assumption that
> drive_merge_report_editor may be NULL and check for that.
> (svn_client_merge_peg3): Initialize children_with_mergeinfo.
...
> --- trunk/subversion/libsvn_client/merge.c (original)
> +++ trunk/subversion/libsvn_client/merge.c Thu Sep 27 12:38:58 2007
> @@ -1623,7 +1623,7 @@
>
> if (merge_b->target_has_dummy_merge_range)
> default_start = end;
> - else if (children_with_mergeinfo)
> + else if (children_with_mergeinfo && children_with_mergeinfo->nelts > 0)
> {
> svn_client__merge_path_t *target_merge_path_t;
> target_merge_path_t = APR_ARRAY_IDX(children_with_mergeinfo, 0,
> @@ -3797,6 +3797,8 @@
> }
> else
> {
> + children_with_mergeinfo =
> + apr_array_make(pool, 0, sizeof(svn_client__merge_path_t *));
> SVN_ERR(do_merge(URL1,
> range.start,
> URL2,

Hey Paul, why was this second case necesssary (e.g. where was the
segfault)? drive_merge_report_editor() seems to handle a NULL hash
okay, as does cleanup_noop_merge(). I didn't catch any other
consumers of this hash. It'd be nice to drop the NULL initialization
when declaring children_with_mergeinfo in svn_client_merge3(); it's
already absent in svn_client_merge_peg3().

Thanks, Dan