Hi Laurent, On Fri, Jul 05, 2019 at 11:04:24AM +0300, Laurent Pinchart wrote: ... > > A reasonable compromise would be to add a Kconfig option that allows > > enabling MC. This way you can provide backwards compatibility and allow > > making use of the full potential of the hardware. That's also why hardware > > that may be part of a non-trivial MC pipeline should start with MC-enabled > > so we wouldn't run into this. > > I really don't like this, as it introduces additional complexity. My > recommendation is to go for an MC-centric approach. Going for a video > node-centric approach is really shooting oneself in the foot regarding > future extensions. But that being said, if there's a strong desire to go > for foot self-shooting, the way to go is explained above. Well, there's nothing that can be done anymore as this has already happened: this is an existing driver in the mainline kernel. Unless you have a time machine of some sort, of course. :-) The choice is now really between breaking existing applications (plain V4L2) and supporting new functionality (MC-centric), so if you need both, I don't really see another choice than a Kconfig option. On the other hand, if we know there are no existing users that could not support the MC-centric view of the device, we could just change the driver API and forget the Kconfig option. It'd be much more simple that way indeed. But I don'k know what's the case. -- Regards, Sakari Ailus sakari.ailus@xxxxxxxxxxxxxxx
