We intentionally separate oid types from hash types; a hash is a generic
hunk of bytes, an object id has meaning and backs an object on disk.  As
a result of this separation, we need a 1:1 mapping.