Re: Support for gzip at the server #424

On 2014-03-10 09:59, Martin Thomson wrote:
> Discussion offline leads me to conclude that doing this would be bad idea.
>
> The basic problem is with 2->1.1 translation at intermediaries.  While
> the HTTP/2 request might include a Content-Length, the gateway is
> going to be forced to decompress and buffer the entire request body
> before forwarding to a 1.1 server.

I'm not sure it is forced to. We could make it required for origin 
servers, and let intermediaries forward as-is.

In that case clients would still have to handle errors, but at least the 
chances of getting server support would be greater.

> This doesn't seem like a good outcome.  Maybe this is something to
> defer to the version of HTTP/x that ships when most of the world is
> already using HTTP/2.

In that case the same problem would come up with HTTP/3->HTTP/2 
translation...


Best regards, Julian

Received on Monday, 10 March 2014 09:14:36 UTC