Re: MEDIA_IOC_G_TOPOLOGY and pad indices

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Hans,

On Sunday, 4 February 2018 15:06:42 EET Hans Verkuil wrote:
> Hi Mauro,
> 
> I'm working on adding proper compliance tests for the MC but I think
> something is missing in the G_TOPOLOGY ioctl w.r.t. pads.
> 
> In several v4l-subdev ioctls you need to pass the pad. There the pad is an
> index for the corresponding entity. I.e. an entity has 3 pads, so the pad
> argument is [0-2].
> 
> The G_TOPOLOGY ioctl returns a pad ID, which is > 0x01000000. I can't use
> that in the v4l-subdev ioctls, so how do I translate that to a pad index in
> my application?
> 
> It seems to be a missing feature in the API. I assume this information is
> available in the core, so then I would add a field to struct media_v2_pad
> with the pad index for the entity.
> 
> Next time we add new public API features I want to see compliance tests
> before accepting it. It's much too easy to overlook something, either in
> the design or in a driver or in the documentation, so this is really,
> really needed IMHO.

I agree with you, and would even like to go one step beyond by requiring an 
implementation in a real use case, not just in a compliance or test tool.

On the topic of the G_TOPOLOGY API, it's pretty clear it was merged too 
hastily. We could try to fix it, but given all the issues we haven't solved 
yet, I believe a new version of the API would be better.

-- 
Regards,

Laurent Pinchart