Re: [PATCH] [RFC] Interface compatibility example - for comment.

[Paul Durrant <xadimgnik@xxxxxxxxx>]

On 07/09/2022 05:23, Martin Harvey wrote:
> This patch is an example of implementing *client-side* interface
> compatibility, in addition to the existing server-side mechanism.
>
> Two-way interface compatiblity would reduce the number of cases
> where we need to increment the Device-ID, which is likely to
> reduce the frequency of upgrade problems, and/or lost IP configuration.
>
> Further details on exact compatibility and upgrade cases solved by this
> pending (tests currently in progress).
As you know, I'm still opposed to this idea and this patch, while I am 
sure it is functional, is just a small part of what is a fundamental shift.
If we decide to go down this route then we'll want to do away with the 
whole PDO revision scheme as it stands. It also seems inefficient to 
have a child just walk backwards down interface version numbers until it 
finds a match; we really ought to be able to have some property of the 
PDO indicate what interface versions are available.
Then there is also the pass-thru aspect of interfaces; a child may 
acquire an interface from any ancestor (and potentially from a filter), 
which may make things more complicated.
As a PoC I think I'd at least want to see a stack of XENBUS, XENVIF and 
XENNET being able to deal with two compatibility in several different 
combinations.
  Paul