Re: [DNSOP] raising the bar: requiring implementations

[Phillip Hallam-Baker <[e-mail]>]

On Wed, Mar 28, 2018 at 2:45 PM, Paul Vixie <[e-mail]> wrote:

>
> i'm in general agreement with each of the assertions made at each layer of
> quoting above, but i have two quibbles.
>
> first, they aren't reference implementations. not even BIND, which for
> many years i called a reference implementation, is not one. a reference
> implementation is a special kind of beast, it's something that if you don't
> interoperate with it, you are in the wrong. we have a specification, and we
> judge the quality of that specification by the ease with which
> interoperable non-reference implementations can be made.
>
> second, i think it's 2018, and  can require that at least one of the
> demonstrated interoperable implementations be source-available. (not open
> source; we don't care about license, only transparency.)


​Quite, a reference implementation is not a production implementation. In
fact it may well not even be standards compliant because the most important
parts of an implementation to test are what happens when messages are not
as expected.

Production implementations should be forgiving in what they accept and
conservative in what they generate. Reference implementations should be the
opposite.

I generate my deployment and reference implementations from the same source
but they are not the same thing.