Re: Video4Linux header files

On Saturday 19 May 2007 21:14, Luigi Rizzo wrote:
> On Sat, May 19, 2007 at 05:30:06PM +0200, Hans Petter Selasky wrote:
> > Hi,
> >
> > The Video4Linux header files are dual BSD/GNU licensed! How nice! I want
> > to put these into the kernel source tree. For example under
> > /usr/src/sys/sys .
> >
> > The reason is that I want to implement a slightly modified version of the
> > V4L so that I can have a well known Linux USB Webcam driver working on my
> > FreeBSD using my new USB stack.
> >
> > Does anyone here have any comments on this ?
>
> the short answer is that in the short term you better rely on a port
> e.g. v4lcompat
>

Ok.

> The long answer is this:
> There was a long discussion back in january about importing the
> headers, and in the end the discussion somewhat stalled due to far
> too many intertwined issues and the existence of a partial workaround
> in ports/v4lcompat.
>
> As i recall/see it the problems were the following:
> - licensing (at the time it was unclear; now this is solved);
> - many v4l*-related programs are broken in various ways, typically
>   assuming that the existence of the v4l2 headers guarantees that the
>   OS supports the v4l2 api, so the program doesn't even try to check/use
>   v4l1 or something else at runtime.
>   At the moment, several of these ports get away with the problem by
>   ignoring it or removing v4l* support. Adding the headers in the base
>   system (which by itself is a desirable goal) requires working with
>   the maintainers to fix the build on the various os versions
>   (being ports not versioned, this could be slightly tricky).
> - having v4l2 support makes no good to anyone now, since there is
>   no v4l2 driver available, and many linux v4l2 drivers use
> USB2-isochronous transfers which are not supported on the current usb
> stack;

> - i don't know how problematic is this, but v4l2 headers seem to use
>   unnamed unions which, last time i tried, conflict with the compiler
>   setting used to build the kernel. While this is possibly an orthogonal
>   problem which we may have to address at some point (as unnamed unions
>   seem to be a common paradigm in linux headers), it is yet another hurdle.

Unnamed unions are not so good. I suggest that we terminate unnamed unions 
with an "u".

>
> I suppose a suitable plan would be the following, but i have not had
> the time to work on it:
>
> 1. add a v4lcompat build-dependency on the ports potentially affected,
>    making sure they still build correctly using v4l1
> 2. add v4l1 headers in the base system (possibly changing the v4lcompat
>    port to not install the headers if they are already in the base system);

I want to put a V4L2.0 "videodev.h" under /usr/src/sys/sys/ with some 
modifications. What do you think about that? 

> 3. modify the vl4compat port to add the v4l2 headers, again fixing
>    any breakage in the builds
> 4. as in step 2, add the v4l2 headers to the system, changing v4lcompat
>    as needed where the v4l* headers are in the base OS
>

Does anyone know which ports use the current Linux compat layer port and the 
v4lcompat port ?

I think it is smarter to look forward.  I'm going for V4L2.0 support in the 
FreeBSD kernel with some modifications. It looks much more coherent than the 
V4L1.0 support.

--HPS

PS: The Linux compat port is broken on FreeBSD-current, because "msleep" is 
being re-defined.