Re: Call for Consensus: Frame size (to address #553)

On Jul 14, 2014, at 3:02 PM, Poul-Henning Kamp <phk@phk.freebsd.dk> wrote:

> In message <78AC8C33-6851-4EA2-93D2-1CFD37020F9F@gmail.com>, Yoav Nir writes:
> 
>>> And what exactly would the SHOULD level requirement make anybody do
>>> differently anyway ?
>> 
>> It means that if you are creating a client (or server) that supports
>> HTTP/2, then you SHOULD be able to work with a server (or client) that
>> advertises a 256-byte max-framesize. Otherwise we get two perfectly
>> legitimate implementations that fail to interoperate. That's a bad
>> thing.
> 
> That SHOULD does not make any sense to me.
> 
> Do you want us to also include SHOULD requirements for 257, 258,
> 512, 1024, 1536, 2048 ... ?
> 
> Once you realize that 256 is not magic, your SHOULD requirement
> reduces to:
> 
> 	HTTP/2 implementations SHOULD be able to work with any other
> 	HTTP/2 implementation.
> 
> ...which I really hope we don't need to tell people.

We should tell them how.

> I also personally have a hard time seeing what I would code differently
> to interoperate with a peer announcing 256 as opposed to 16k or 1MB
> max frame size:
> 
> 	frame_size = min(his_settings, my_preference);
> 
> What am I missing ?

Down-thread, Jeff asked "So am I to read this as a client might advertise a max frame size of 256 bytes and then request a 2GB file?”, and you answered, "Yes. And the server is free to return 418 or react in any other way” 

I’m saying that if whenever a client announces 256 some servers will send an error code, we have a problem - we don’t have interoperability.

Yoav

Received on Monday, 14 July 2014 13:03:03 UTC