RE: ALPN ID for HTTP/2 interop

On 14/11/2014 1:59 a.m., Tatsuhiro Tsujikawa wrote:
> Hi,
> 
> We've been doing interop using h2-14 ALPN ID for a while. Some 
> client/server now only advertise h2-15, others do both h2-14 and 
> h2-15. Since draft-15 is binary compatible to h2-14, for code 
> simplicity and interop, it would be better to stick to h2-14, unless 
> we introduce binary incompatible changes in the future draft.
> 
> Thoughts?

This has affected work I have been performing in the last week, principally between Chrome Canary and nghttp2 (but presumably any client/server combo would be susceptible).

I tend to agree with Amos, in general behavioural differences between versions could lead to interop issues. Being strict with advertising support for what is truly implemented builds greater confidence  "in-band".

In cases where software is capable of supporting multiple versions, can that not be supported in the ALPN negotiation? In the h2-14/h2-15 case, if the differences truly are minimal the branching complexity within the software is minimal.

Received on Thursday, 13 November 2014 14:46:17 UTC