In message <20160809232819(_dot_)1291(_dot_)qmail(_at_)ary(_dot_)lan>, "John
Levine" writes:
Obviously, taste and correctness matter.
It still won't be a good idea to say "The reserved bit must be zero on
send and must be ignored on receive," arguing "Well, we don't want to
use MUST because some implementations don't do that so it can't be
normative."
I'd write "The reserved bit is set to zero on send and is ignored on
receive" and save the command terms for things where one might think
that there was a reason to do something else.
RFC2671/RFC6891 has
Z
Set to zero by senders and ignored by receivers, unless modified
in a subsequent specification.
which resulted in 2% of deployed nameservers just copying reserved
bits to the reply or 3% of nameservers not answering because a
reserved bit is set.
https://ednscomp.isc.org/compliance/ts/alexa.flagsfail.html
A bit more emphasis may have changed the result.
At least the TLD servers no longer copy the reserved bits. The
last server doing that was just upgraded a week or so ago.
https://ednscomp.isc.org/compliance/ts/tld.flagsfail.html
Mark
The point of lower case keywords shouldn't be to allow people to be
sloppy and to avoid normative text to make a false consensus easier.
This SHOULD be about writing clearer RFCs and not having to contort
language when should and must are perfectly good non-normative things to
say.
Yup.
R's,
John
--
Mark Andrews, ISC
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742 INTERNET: marka(_at_)isc(_dot_)org