Re: [TLS] request for early IANA assignment for ALPN

[Trevor Perrin <[e-mail]>]

On Wed, Mar 27, 2013 at 3:23 PM, Michael D'Errico <[e-mail]> wrote:
> Trevor Perrin wrote:
>>
>> Hmm.... so we'd have to do all interop testing and trial deployments
>> based on an "ExperimentalExtension".  Then - once everything is
>> working and RFCs are assigned - make a major incompatible protocol
>> change to a different extension format.
>
>
> The extension_data would not change.  The only difference would be in
> locating it.

Sure.  But it's still a code change *after* all the testing and
deployment that went into the Internet-Draft.


>> I see the costs and risks here, but not the benefit.
>
>
> The problem with using a range of temporary extension numbers comes when
> you run out and have to reuse one.

I wasn't proposing a separate range of temporary extension numbers.

I was seconding Nikos' proposal that there should be an easy way for
implementors of an Internet-Draft to temporarily reserve "real"
numbers while doing testing and trial deployment.

If the Internet-Draft advances to RFC, those numbers can be enshrined
by IANA.  Otherwise, they're recycled.


>  Somewhere a server that hasn't been
> updated in years will be running code like this:
>
>     switch (extension_number) {
>        case 0x0000:     process_sni_extension(); break;
>
>        case 0xABCD:     // experimental ALPN extension number - REMOVE!
>        case 0x00xx:     // IANA-assigned ALPN extension number
>                         process_alpn_extension(); break;
>     }
>
> If 0xABCD is ever used for any other experimental extension, it could
> cause this server to start acting funny.

We're proposing the opposite of this - instead of having different
code paths for I-D vs RFC, there should be one code path (i.e the IANA
should have assigned 0xABCD, above).


> The problem outlined above, plus it allows for non-IETF-sanctioned
> experiments such as NPN/SPDY.  The wire format of an extension could
> change during the experiment - simply add "v2" to the name instead of
> requiring a new number assignment via rough consensus at the next IETF
> meeting resulting in a wiki update.

If early implementors of an I-D need to make a wire format change, I
think they should upgrade their implementations and keep using the
same temporarily-reserved extension number.


Trevor