Re: [RFC/PATCH] v4l2-compliance: Reject invalid ioctl error codes

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

Hi Hans,

On Monday 24 December 2012 10:24:48 Hans Verkuil wrote:
> On Sun December 23 2012 23:24:04 Laurent Pinchart wrote:
> > The recent uvcvideo regression that broke pulseaudio/KDE (see commit
> > 9c016d61097cc39427a2f5025bdd97ac633d26a6 in the mainline kernel) was
> > caused by the uvcvideo driver returning a -ENOENT error code to
> > userspace by mistake.
> > 
> > To make sure such regressions will be caught before reaching users, test
> > ioctl error codes to make sure they're valid.
> 
> I don't like this change. Error codes should be checked in the test for
> the actual ioctl.
> 
> Apparently it is QUERYCTRL that is returning the wrong error code in uvc,
> but looking at the code in v4l2-test-controls.cpp it is already checking
> for ENOTTY or EINVAL and returning a failure if it is a different error
> code. So why is that not triggered in this case?

I've just checked that, the missing control class issue made the control tests 
stop early before hitting the wrong return value. I guess that's a good reason 
to fix *all* compliance errors...

We can drop this patch.

-- 
Regards,

Laurent Pinchart