Hmm.  It seems to me this is an edge case as far as backward compatibility goes.  On the one hand it does seem like an RFC violation (ie: bug), on the other hand there could be user programs depending on the current behavior.  I'm inclined toward treating it as a bug, but I can see it being argued the other way.  The fact that the file is designed for interoperability and other tools do take the wildcard first argues in favor of treating it as a bug, I think.