Are we still creating standards based on "BIND does this"? :p
On 28-11-16 13:57, Tony Finch wrote:
Matthijs Mekking <matth...@pletterpet.nl> wrote:
1. In case of a DNS responder selecting one or a subset of the RRsets at the
QNAME, The draft does not give clear guidance on which RRset(s) to pick.
The code in BIND just picks an arbitrary RRset, without making any effort
to be clever.
It makes an ANY query to the back end, then adds the first RRset to the
response. The data from the back end doesn't have a deterministic order.
Doing better than this would (I think) require additional methods in the
rdataset API to get the size of the RRset. Unless we use a dumb heuristic
like smallest RR TYPE number :-)
Fair enough for reasoning not to use a determinative mechanism, but in
that case such considerations should be added to the document IMO.
Best regards,
Matthijs
2. People expressed that they would like to see ANY over TCP stick to the
original (undefined) behavior, that is to return all RRsets at the QNAME.
BIND does this.
One more nit comment, I would like to see explicitly called out in section 7
that "DNS implementations are free to not return all RRSIGS *in case
QTYPE=RRSIG*".
BIND also does this.
Tony.
_______________________________________________
DNSOP mailing list
DNSOP@ietf.org
https://www.ietf.org/mailman/listinfo/dnsop
