Re: [#153] PUSH_PROMISE headers

On 3/07/2013 8:24 a.m., Mike Bishop wrote:
> Did we have consensus here on whether PUSH_PROMISE in the Implementation Draft contains *all* request headers or simply the ones that have changed from the original request?
>
> The more I think about it, I'm inclined to say "all" since the server can drop any headers it knows it won't be looking at if it wants, and headers that are the same will just get compressed out after they've been pushed back once.

Except that they are being sent over the response side of the connection 
so they are affecting the server send context and client receive 
context. They are also likely to die again quickly under the relatively 
large bunch of response HEADERS blocks which are following - both the 
main response and the pushed ones.

I'm sightly in favour of "all" as well despite that problem because when 
sending over multiple hops the request headers may be altered. The 
server omiting header Foo because a proxy delivered Foo:hello when the 
client sent Foo:bar will cause big potential issues with both security 
validation and caching. Ensuring those functionalities work reliably 
outweighs some sub-optimal performance IMO.

Amos

Received on Tuesday, 2 July 2013 21:43:14 UTC