In <1926730(_dot_)1121811235(_at_)[192(_dot_)168(_dot_)0(_dot_)2]> Greg Connor
<gconnor(_at_)nekodojo(_dot_)org> writes:
wayne wrote:
I would say "yes". Section 4.5 doesn't say "equivalent", it
says "identical".
I may be missing something obvious, but why are we checking records
against each other again? Is there something wrong with choosing one
or the other and just running with it?
I think that leaving it up to implementations to choose which, of two
inconsistent records, to use is A Bad Idea.
The text about how to handle the SPF RR type is largely unchanged from
the MARID drafts, where it was first introduced. This was worked out
between MarkL and unnamed "DNS gurus", with additional feedback from
everyone else. In particular, the marid-protocol draft said that they
MUST be identical. It didn't say what to do if they are not. In some
cleanup, I added text to section 4.5 to say that if they aren't
identical, to return PermError. I remember this being discussed, at
least briefly, but I haven't tracked down the thread.
I can see that we want to strongly recommend that they be the same (to
the point of saying they MUST be) but should we force all parsers to
check and compare both?
Doing a straight compare on the records is very cheap.
But, actually, the spec only says that you have to compare both if you
fetch both. Those that choose to only fetch one type are not required
to do the comparison.
-wayne