Re: [Uri-review] Proposed scheme registration: sdns (DNS stamps)

I don't typically chime in on these proposals much, but this one strikes me
as odd. I don't understand the point of having a string representation if
the entirety of the string - except the leading scheme name - is base 64
encoded. Well, OK, I get it for a URI scheme where the underlying data is
actually binary, such as the "data" scheme, except even with that, the
"media-type" is not base64 encoded. This scheme doesn't seem to fit that
model.

The previous comment about the "//" is correct. This is *not* a
hierarchical scheme, so it should not have a "//". That is, "../foobar" is
a meaningless relative URI to apply to the proposed sdns scheme.

I took a random example from
https://github.com/DNSCrypt/dnscrypt-resolvers/blob/master/v2/public-resolvers.md

sdns://AgMAAAAAAAAADDM1LjIzMS42OS43NyA-GhoPbFPz6XpJLVcIS1uYBwWe4FerFQWHb9g_2j24OBJkbnMtZ2NwLmFhZmxhbG8ubWUKL2Rucy1xdWVyeQ

When this is expanded, based on the demo <https://dnscrypt.info/stamps/>, I
find that this is:
protocol: DNS-over-HTTP/2
IP:35.231.69.77
host-name: dns-gcp.aaflalo.me
DNSSEC: true
No logs: true
No filter: false
Hashes: 3e1a1a0f6c53f3e97a492d57084b5b9807059ee057ab1505876fd83fda3db838
Makes me wonder why this can't just be represented as:

sdns:dns-over-http2:ip-35.231.69.77:hn-dns-gcp.aaflolo.me:
dnssec:nologs:hashes-3e1a1a0f6c53f3e97a492d57084b5b9807059ee057ab1505876fd83fda3db838

Maybe I don't understand all the design parameters behind the original
proposal? My proposed above form has a whole bunch of up-sides. It is easy
to visually inspect for certain aspects of correctness, it is more robust
in the face of future enhancements, and it is order-insensitive. Maybe my
idea is completely off base, in which case, it would be very helpful to
augment the proposed specification to clarify why the current proposed
opaque form is *required*.

If compatibility with existing tools is required, then there are multiple
paths:
- check for hierarchy use in existing implementions (sdns:// parsed
differently from sdns:...)
- register a under a new scheme, and phase out support for the existing
sdns:// uses

Eric.

On Wed, Oct 24, 2018 at 10:12 AM Frank Denis <ietf=
40dnscrypt.info@dmarc.ietf.org> wrote:

>
> > On 24 Oct 2018, at 18:31, Ted Hardie <ted.ietf@gmail.com> wrote:
> >
> > Hi Frank,
>
>   Hi Ted,
>
>
> > Thanks for sending the proposal to the list.  A couple of high level
> comments.
> >
> > First, the syntax you present uses //, which doesn't seem to be correct
> to me.  If I'm reading the specification correctly, the URL doesn't have a
> hierarchical part and does not rely on an authority section (which is what
> the // delimits in RFC 3986 syntax).
>
>   All the sdns URIs currently in use share a global namespace. However,
> the parameters specifically use URL-safe base64 to avoid slashes, so URIs
> can later:
>
> - Support namespaces
> - Support implementation-specific filters (e.g. `sdns://.../ipv6only)
>
>
> > 2nd, it's not clear in the description of addresses how they would be
> fed into the algorithm you provide in some of the corner cases.  For
> example:
> > "addr is the IP address of the server. It can be an empty string, or
> just a port number."  Is the port number represented with a preceding
> colon?
>
>   A preceding colon is required. The description has been updated to
> clarify this. Thanks!
>
>
> > For IPv6 hosts which use [ and ] as delimeters, is the input URL-encoded
> before being base64 encoded?  Are scopes permitted?
>
>   Scopes are permitted, and are supported by current implementations. The
> document has been updated to reflect this.
>
>   Strings are not encoded individually, besides being preceded by their
> length.
>
>   The entire concatenation of all parameters is eventually base64-encoded.
> The document has been updated to clarify this.
>
>
> > For hostname, the document says  that hostname is the server host name
> which will also be used as a SNI name.  If the hostname uses one of the
> IDNA characters outside the URL-permitted range, is it encoded with URL
> encoding or punycode?
>
>   SNI names in certificates are raw binary strings. Given the context, we
> probably shouldn’t apply any encoding either. The document has been updated
> to clarify this.
>
>
> > As a stylistic note, I think sdns is likely to be confused with the
> existing dns scheme name, and I would suggest using "dns-stamps" for
> additional clarity.
>
>   This is definitely doable, but would be break all existing
> implementations and servers lists,
>
>
> > Thanks again for the opportunity to comment,
>
>   Thanks for your very helpful comments, Ted.
>
>
> _______________________________________________
> Uri-review mailing list
> Uri-review@ietf.org
> https://www.ietf.org/mailman/listinfo/uri-review
>