[OpenID] Backwards Compatibility

[Martin Atkins]

Martin Atkins wrote:
> David Recordon wrote:
>> Hey Carsten,
>> I agree with you.  It's time to make sure that 1.1 is clearly 
>> deprecated and that everyone implements 2.1 once it is completed.
>>
> 
> Is there some compelling reason not to make 2.1 a superset of how 2.0 is 
> implemented today? (That is to say, not a superset of the spec as 
> written, but a superset of the parts of it that are implemented and the 
> errata.)
> 
> I think in an ideal world every existing, well-behaved 2.0 
> implementation would already be a fully-compliant 2.1 implementation. 
> This implies a research-driven approach involving the studying of 
> existing implementations; there should be a high barrier to specifying 
> anything that disagrees with an existing, well-behaved implementation.
> 

To complete this thought, I think a pre-requisite for work on 2.1 is a 
comprehensive test suite for 2.0 and a harness to run the tests against 
popular implementations.

This will allow changes made for 2.1 to be regression tested against 
existing implementations. For example, we could make an implementation 
that doesn't do the DH association step and see which implementations 
that breaks.