On Thu, Apr 2, 2026 at 6:48 PM Eric Rescorla <[email protected]> wrote:

> On Thu, Apr 2, 2026 at 3:25 PM David Benjamin <[email protected]>
> wrote:
>
>> I think trying to cite IANA registries is a distraction. IANA registries
>> are *not* enough. To meaningfully define a codepoint, you need to say
>> what the codepoint means. That's why our IANA registries have document
>> references. Moreover, to be well-defined, it needs to be a TLS-specific
>> document, not just the underlying crypto, to write the small handful of
>> TLS-specific definitions needed to glue things together.
>>
>> Now, *those* documents are candidates for citing, and our registries *do*
>> sometimes reference drafts, as they do here. One *could* argue that
>> people should just be OK citing drafts. But I cannot *entirely* blame
>> folks for being reticent to cite drafts when the IETF spends so much energy
>> to emphasize that drafts are not done. They have expiry, there is no
>> indication whether people expect to define a new one with different
>> semantics, etc. For things that the IETF *does* intend to publish, we
>> have all spent quite a lot of energy making sure people downstream of us
>> know that it's not "official" until it's published. We all did this because
>> it's harmful for standards work if people freeze intermediate snapshots
>> that aren't done.
>>
>
> ISTM that the "freezing" here happened at the time of code
> point assignment.
>

Not quite. Assigning the codepoint means that 0x0201 forever means
draft-ietf-tls-mlkem-07's ML-KEM-768 construction. It does not mean that we
won't publish draft-ietf-tls-mlkem-08 to define 0x0203 for a slightly
different ML-KEM-768 construction.

Now, there is barely anything (but still something) to define
in draft-ietf-tls-mlkem, so this is obviously not going to happen here.
But, in general, we have lots of things to fiddle with when defining
things. TLS 1.3 and ECH went through a number of codepoints, each of which
meant particular drafts, but we were not ready to call it done for several
codepoints.


> We can't have it both ways. Either we tell everyone that drafts are
>> perfectly fair game to implement, with no stopgap against people
>> inadvertently ossifying things that aren't yet done, or there must be ways
>> for drafts to progress into a state that signals they are done. That
>> durable state doesn't have to be an RFC, but it's the only one the IETF has
>> defined right now.
>>
>> One could say, ah, the problem is the IETF needs to define a non-RFC end
>> state, and that *that's* not TLS WG's business. But it *is* very much
>> the TLS WG's business that we're worrying about this at all, because we've
>> managed to spend a disproportionate amount of time waffling on technically
>> trivial documents.
>>
>
> I don't think that this analysis is really correct.
>
> Stepping back from this document, we have roughly two categories of
> algorithms (and other extensibility points):
>
> A. Those which have the support [0] of the TLS WG.
> B. Those which do not have the support of the TLS WG but some set
>    of people nevertheless wants ot use.
>
> Type A algorithms get published as RFCs.
>
> The current registry structure was designed to allow people with Type
> B algorithms to nevertheless reserve code points without the TLS WG
> getting involved, either on the basis of some non-RFC specification or
> by persuading the ISE to publish their RFC. The reason we did this is
> not because we want to promote the use of those algorithms but rather
> because we know people will use them anyway and we want to avoid
> contamination of the code point space.
>
> Now, sometimes, as in this case, we get people who want to use an
> algorithm ask the TLS WG to publish an RFC because they don't want to
> use the code point without an RFC. However, the reason these documents
> aren't being published as RFCs is because there's not consensus that
> have consensus promote their use, so I don't think publishing them as
> an RFC so that others can more readily use them is a very strong
> argument. If people find it somewhat harder to use algorithms for
> which there isn't consensus why is that bad?
>
> With that said, if we want to make a shift like that contemplated
> draft-barnes-tls-this-could-have-been-an-email, such that even
> documents which the WG does feel positively about and otherwise could
> have gotten consensus to publish as RFCs are just I-Dsor emails, then
> we perhaps do in fact need some permanent definition of the algorithm,
> but that doesn't seem to be the case here.
>

What's missing here is that publication can potentially signal one of two
things:
- Whether the TLS WG promotes its use (though we also have the Recommended
column)
- Whether the community of folks who care about that scheme are done with it

RFCs, in different contexts, are used for both, which unavoidably causes
some tension.

-Ekr
>
> [0] By support I mean "enough people in favor that they can be
> published in the IETF".
>
>
>
>
>
>
>
_______________________________________________
TLS mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to