Re: Kernel headers changes in F10?
On Mon, 2008-06-02 at 11:55 -0300, Mauro Carvalho Chehab wrote:
Hi David,
On Mon, 02 Jun 2008 11:31:56 +0100
David Woodhouse <dwmw2(a)infradead.org> wrote:
> A few moments with git-annotate showed that the missing ioctls were
> removed in commit 26d507fcfef7f7d0cd2eec874a87169cc121c835ยน by Brandon
> Philips. In their place, we have the following:
>
> /* Deprecated, use V4L2_CID_PAN_RESET and V4L2_CID_TILT_RESET */
> #define V4L2_CID_HCENTER_DEPRECATED (V4L2_CID_BASE+22)
> #define V4L2_CID_VCENTER_DEPRECATED (V4L2_CID_BASE+23)
>
> That seems like a rather dubious change to the user API -- shouldn't we
> ensure that existing software continues to build, but maybe add a
> compile-time or run-time warning for those using the deprecated ioctls?
>
> Mauro, Brendan: unless the removal of those ioctls has been properly
> documented in Documentation/feature-removal-schedule.txt for an
> appropriate amount of time, please could you put them back as they were.
> Changing the userspace API like this isn't acceptable.
IMO, there's nothing to be added at feature-removal-schedule.txt. Those controls
aren't used by any kernel drivers on a long time (I suspect that they were
never used).
The issue is that they are targeted to do exactly what V4L2_CID_PAN_RESET and
V4L2_CID_TILT_RESET do. So, they are duplicated stuff.
I was not aware that an userspace is using those symbols, but, if so, it is
likely to be related to some binary-only or out-of-tree drivers [1].
It was part of the published API, and it was being used by the gstreamer
v4l2 plugin -- which now fails to build.
Yes, of course we'll fix the code in question, but just removing APIs
from the headers and causing source code to break without any prior
warning is not acceptable. That's precisely what
feature-removal-schedule.txt is _for_.
--
dwmw2