Re: svn commit: r27939 - in trunk/subversion: libsvn_client tests/cmdline

On Tue, 20 Nov 2007, C. Michael Pilato wrote:

> Daniel L. Rall wrote:
> > On Mon, 19 Nov 2007, cmpilato@tigris.org wrote:
> >
> >> Author: cmpilato
> >> Date: Mon Nov 19 20:28:57 2007
> >> New Revision: 27939
> >>
> >> Log:
> >> Allow disabling of mergeinfo considerations with the --ignore-ancestry
> >> option to 'svn merge'. This allows folks to effectively force merges
> >> that mergeinfo would otherwise prevent without sacrificing the
> >> recording -- and therefore, future consideration -- of the merge.
> >>
> >> * subversion/libsvn_client/merge.c
> >> (notification_receiver): Add some comments.
> >> (filter_reflected_revisions, filter_merged_revisions,
> >> populate_remaining_ranges): Add note about only calling this
> >> function when honoring mergeinfo.
> >> (calculate_remaining_ranges): Lose 'same_repos' argument, and add a
> >> note about only calling this function when honoring mergeinfo.
> >> (populate_remaining_ranges): Add 'honor_mergeinfo' parameter, and if
> >> it isn't set, populate the remaining ranges with the whole
> >> requested range. Update call to calculate_remaining_ranges().
> >> (drive_merge_report_editor, do_directory_merge, do_file_merge): Grow
> >> awareness of merge-tracking-disabled invocations.
> > ...
> >> @@ -3494,6 +3530,12 @@
> >> apr_pool_t *subpool;
> >> svn_boolean_t is_rollback = (revision1 > revision2);
> >> const char *primary_url = is_rollback ? url1 : url2;
> >> + svn_boolean_t honor_mergeinfo = (merge_b->sources_related
> >> + && merge_b->same_repos
> >> + && (! merge_b->ignore_ancestry));
> >> + svn_boolean_t record_mergeinfo = (merge_b->sources_related
> >> + && merge_b->same_repos
> >> + && (! merge_b->dry_run));
> >
> > Mike, shouldn't sources_related (currently) imply same_repos?
>
> I would have thought so, but apparently not. We have a regression test
> aimed specifically at supporting merges from one repository into a working
> copy of another repository altogether.

We currently allow merging from one repository into another's WC (e.g. to
support vendor branches), but we can't consider -- and so also shouldn't
record -- mergeinfo in these cases, because we don't record a repository URL,
only a path relative to the current repository.