Given the the "reintegrate" merge is said to be an automatically generated two-URL merge, it would be useful if Subversion client programs had access to the component parts -- that is:
* do the validity checks * determine which two source URLs (especially the "parent branch at last sync'd rev" one) * run the standard two-URL merge function That would enable the client to do things like print information after the second step about what two-URL merge it is going to run, so that the user can better understand what is happening. (In the 'showing-merge-info' branch I currently have modified the libsvn_client function merge_reintegrate_locked() to print a bit of such info, but libraries are not allowed to print directly and sending this info through some sort of callback mechanism would be entirely the wrong way to go.) Currently merge_reintegrate_locked() calls merge_cousins_and_supplement_mergeinfo() to do the actual two-URL merge. That function is a subroutine of the standard two-URL merge function, and I'm not quite sure how it fits in but it appears to be responsible for doing a two-URL merge in which one source is a direct ancestor of the other, whereas the full two-URL merge as implemented in merge_locked() can also do a diff between two URLs that have a common ancestor, forming a Y-shaped ancestry graph. In such a case it calls merge_cousins_and_supplement_mergeinfo() to do one branch of the Y, and then calls do_merge() to do the other branch of the Y. I'm not sure yet whether merge_cousins_and_supplement_mergeinfo() should become public functionality and the reintegrate code should call it, or whether the reintegrate code should call the full two-URL merge. Thoughts? - Julian
