[kitten] SCRAM-SHA512 and SCRAM-SHA3

[Simon Josefsson <[e-mail]>]

Hi

I'm seeing push on implementers to add support for these variants, and I
noticed new drafts were published recently.  I want to repeat some
earlier concerns.  I believe the cost of having these two mechanisms as
standard mechanisms in the ecosystem costs more than any advantages we
would get out of them.  There is still no cryptographic attack on
HMAC-MD5, yet alone the HMAC-SHA1 or HMAC-SHA256 that are used in
SCRAM-SHA1 and SCRAM-SHA256 that we are still seeing deployment of.
Adding SCRAM-SHA512/SHA3 variants create additional requirements on
hashed password database formats and APIs, since they are not compatible
with SCRAM-SHA1 and SCRAM-SHA256.  Parametrization of security protocols
and algorithms are generally a bad idea as it adds complexity which
reduce security.  There is the negotiation interop problem if a server
has one credential but not the other for a subset of users.  If some
people are using these variants, I would agree that having them
documented is useful.  Then I believe the category should be
Informational rather than standards track, and warnings about the
problems should be added.

/Simon