Re: Server Push and Content Negotiation

[Mark Nottingham <[e-mail]>]

> On 25 Aug 2016, at 12:36 AM, Patrick McManus <[e-mail]> wrote:
> 
> [implementer hat]
> 
> On Wed, Aug 24, 2016 at 12:26 AM, Mark Nottingham <[e-mail]> wrote:
> What do people think -- is this worth specifying? How are implementations currently doing this?
> 
> push is one way - negotiation doesn't make any sense and yet the alternative to negotiation is some kind of sniffing. sniffing is awful and is even harder here where you don't have a request to sniff - you'd have to do some whole connection guess.
> 
> how would you push some feature that relies on conneg today - like webp or brotli?

The server has to make an informed guess about client support. One way would be to look at the initiating request headers; as I think a few people have said, they're rarely variant. Worst case, try a push and look for an error code saying it wasn't a good encoding (see other thread).


> I think as a client realistically you have to wait for the response headers and check them out to see if its an acceptable push. Whatever is in the request headers isn't really enough for that - which is a bit of a change from what I currently do.
> 
> fwiw, this was one of the reasons gzip was a mandatory-to-implement CE in spdy and probably should have stayed. It meant you could reliably push it. Doing so now is really just an assumption that gzip is a defacto standard.
> 
> Or this all might be another argument about the excessive complexity of push - maybe that should be a document :)
> 

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