There can be cases when both the changesets which we are merging are not
descendants of the merge base. In those cases dirtyc1 and dirtyc2 both will be
true. The existing code assumes that either of them will be true always but that
is not a right assumption.
I found this while working with content-divergence resolution. In
content-divergence resolution, we use the common predecessor as the merge base
for resolving content divergence and there it can be possible that the merge
base is not the descendant of both the content-divergence changsets.
The code in content-divergence which does this is at: