Re: [Add] Comparative DoH Discovery DNS RR Types

[Martin Thomson <[e-mail]>]

On Tue, Jun 30, 2020, at 12:14, Christian Huitema wrote:
> I heard EKR make two arguments for using CNAME: not just middlebox
> compatibility, but also software simplicity. The latter is interesting,
> and might be worth explaining a bit. The way I understand it, Mozilla
> would rather use a well known API like getaddrinfo(), rather than a more
> capable API that supports different record types. I can guess that this
> is a software simplicity/complexity trade-off. I suppose that the more
> capable API tend to be implemented slightly differently on different
> OSes, which leads to more complex code with a variety of #ifdef to
> accommodate the different platform, more tests, and probably more bugs.
> But I may be getting it wrong, and I would rather hear what Mozilla
> actually thinks...

I'm not super close to this, but my understanding is that we now have two resolution paths: one native and one for DoH.  On the DoH path, we have greater ability to manage new RRtypes and queries.  However, this specific case cannot follow that path.  As this has to be cross-platform, there is a non-trivial amount of plumbing required to support doing anything other than the getaddrinfo() call.  Our abstractions are pretty old and have never had to grow the ability grow features like the ability to handle new RRtypes.  In other words, Christian, your guess is spot on.

As this is (currently) driven by a need for expediency, and CNAME works, it wasn't worth investing a whole lot of effort in finding something better, especially when it is clear that the matter was not settled.  If presented with an IETF consensus position on the topic, that would likely be a different story.