On 9 Oct 2015, at 10:54, Paul Wouters wrote:
On Fri, 9 Oct 2015, Joe Abley wrote:
In a fit of zeal I wrote up what I thought was a reasonable
clarification to 1034/1035 with respect to the ordering of RRSets
within sections of a response to a DNS QUERY, prompted by the
discussions on this list in August, to which maybe this link is a
useful pointer:
Mark and Paul gave me some opinions as I was writing this up, that I
may or may not have represented accurately in the text. I think the
advice is reasonable, but thoughts from the throng as to (a) whether
this was worth writing down and (b) whether what I wrote is nonsense
would be appreciated.
I find it strange that you suggest ordering matters for the Answer
section but not the Authoritative section. It seems that we will just
get more assumptions from code that order matters there too, and 10
years from now we are writting this document again, but for the
Authoritative section.
The principal reason that ordering seems empirically to matter in the
answer section is that that's the section that stub resolvers care
about. It's really the stub-recursive interaction that matters, here.
With a mind towards running code, if we specified that order mattered in
the additional section we'd essentially be making all major
implementations non-compliant in some way. There's no indication that
different ordering strategies in the additional section have any impact
on deployed resolvers (and perhaps we can help ensure that none come to
light in the future by making the specification clear).
For responses without DO=1 I don't know of a case where we see more than
one RRSet in the authority section, so ordering is a no-op there in any
practical sense.
Joe
_______________________________________________
DNSOP mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dnsop
