Re: [DNSOP] When to Clarify - was Re: Ambiguous standards suck (was Re: draft-jabley-dnsop-ordered-answers)

["Joe Abley" <[e-mail]>]

On 6 Nov 2015, at 5:08, Edward Lewis wrote:

> 3. Running code trumps written documents.
>
> If interoperability is achieved with insufficient original text, the text
> is evidently sufficient.

I'll note that the discussion that kicked all this off was an observed failure to interoperate in deployed code, and whether that code was following the specification or not was a matter of some debate depending on how 1035 was interpreted.

I agree that changing the specification such that the majority of deployed code is suddenly non-compliant is a recipe for disaster. Let's not do that.

I originally proposed that all sections should be unordered, and that nobody should reject a message on the basis that it should be otherwise. It was pointed out to me that this would make several widely-deployed implementations non-compliant, so I changed the proposal such that the answer section was considered an ordered set, and all other sections unordered.

There were several examples given at the microphone the other day about additional known bad reactions in deployed code relating to OPT and TSIG in the additional section. I would argue that those should also be documented, since I think the proposal should align itself with the steps required to make new code interoperate with deployed code.


Joe