Re: [PATCH v3 1/2] media: exynos4-is: Properly report _MPLANE caps

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

Hi Sylwester,

On Fri, Apr 12, 2019 at 01:14:38PM +0200, Sylwester Nawrocki wrote:
> On 4/12/19 12:32, Hans Verkuil wrote:
> > On 4/12/19 12:16 PM, Sylwester Nawrocki wrote:
> >> On 4/12/19 11:20, Boris Brezillon wrote:
> >>> The fimc-isp-video.c and fimc-lite.c were missing the
> >>> V4L2_CAP_VIDEO_CAPTURE_MPLANE flag when reporting device caps.
> >>
> >> Th omission was intentional, as these video nodes cannot be used
> >> by "standard" V4L2 capture applications. Some sub-devices need to 
> >> be configured first in order to use these video nodes. It was 
> >> agreed to not set these capabilities in the driver, instead user
> >> space library, e.g. libv4l2 with driver-specific plugin handling
> >> the media controller and v4l2 subdev calls would set the capability
> >> for the applications. Has something changed?
> > 
> > I remember that this was discussed, but nobody actually followed that
> > guideline. Just do:
> >
> > git grep -l V4L2_CAP_VIDEO_CAPTURE drivers/media/platform
> > 
> > It really doesn't make sense to leave it out since userspace has to know if
> > it is single or multiplanar, and if you don't set this capability, then how
> > can it tell?
> 
> Presumably you can't by only querying the video device, this would require 
> MC and driver-specific user space. 
> 
> > I always thought we need a separate CAP or some other way of signaling this,
> > but the few times I proposed something it was always shot down.
> > 
> > So to be consistent with the other MC drivers you should set this MPLANE cap
> > as well in the fimc driver.
> 
> I'm not against the $subject patch, rather the opposite. Especially if 
> it is a part of reasonable API updates. But removing the capabilities was 
> actually a requirement to get the driver merged, not a guideline.
> And was rather painful then, now the driver is probably not in use any more
> so I don't really mind.

As this driver is quite old I assume there's little interest in
investing in it, but is there any chance Samsung would be interested in
contributing to libcamera for newer generations ? :-)

-- 
Regards,

Laurent Pinchart