Reverse merges of inheritable revision ranges when the merge target has a
missing child with no explicit mergeinfo currently results in inaccurate
mergeinfo, since we have no way to mark that child as *still* having the
reversed range -- See http://svn.haxx.se/dev/archive-2007-07/0038.shtml for
details. Later in that thread cmpilato suggests negated revision ranges as a
way to solve this problem. I stated back then that I think this can work, and
still think that, but will now dive into it in detail to see if any problems
shake out...
I'm setting this as P3 for now as it strikes me as a bit of an edge case...does
that sound reasonable?