The most thorny issue here, with respect to B/C, seems the sha1 field. That can't not change semantics: if we use the revision sha1, it does not equal sha1(text) anymore, and the assumption that revisions with the same main slot have the same sha1 does not necessarily hold. If we use the main slot sha1, the assumption that same sha1 means same content does not hold. The first approach seems less bad, but this is something people running old dump processing scripts will have to be mindful of. (And the main slot sha1 should probably be in there, somewhere.)
