Re: I-D Action: draft-ietf-httpbis-client-hints-03.txt

[Mark Nottingham <[e-mail]>]

> On 24 Dec 2016, at 2:14 am, Mark Nottingham <[e-mail]> wrote:
> 
> 
>> On 13 Dec. 2016, at 7:31 pm, Martin Thomson <[e-mail]> wrote:
>> 
>> On 6 December 2016 at 08:51, Ilya Grigorik <[e-mail]> wrote:
>>> Background: https://github.com/httpwg/http-extensions/issues/168
>>> 
>>> We didn't intend Save-Data to be a list.. The goal was to allow attributes
>>> on the single token defined in CH. I'll defer to Mark on Key.
>> 
>> 
>> Whether you have a list or not, Save-Data has some odd semantics when
>> it comes to extension.  How do I model the following variants?
>> 
>> Save-Data: on
>> Save-Data: on;video-only
>> Save-Data: on;exclude-video
>> Save-Data: on;compress-video-more
>> 
>> Assuming that the extensions are not understood, we end up in an odd
>> place.  I see several conclusions:
>> 
>> 1. unknown extensions are ignored
>> 2. unknown extensions can only increase the amount of compression; and
>> are ignored otherwise (see compress-video-more)
>> 3. unknown extensions cause the entire clause to be ignored
>> 
>> Option 1 seems to be what we have right now, but that doesn't give a
>> client any way to request selective compression (the
>> video-only/not-video example).
>> 
>> Option 2 might allow a client to request more aggressive compression,
>> but it does nothing to address the case where a client wants video
>> compression, but would prefer not to have images degraded by
>> compression.
>> 
>> Option 3 might be combined with a comma-separated list to allow
>> clients to provide a fallback strategy.
>> 
>> There's also probably something more complex as well.
> 
> I think there's also an Option 4: extensions can only be defined by a sd-token, when it is defined. I.e., this extension syntax is for the use of future sd-tokens, not for arbitrary extension independent of them.
> 
> That's what I'd assumed when we defined the syntax; regardless of the intent, we should write something down about how the extension point is used.

Is this workable for everyone?


--
Mark Nottingham   https://www.mnot.net/