[GEM-dev] DV support uses deprecated header file

[Johannes M Zmoelnig]

guenter geiger wrote:
> On Tue, 12 Oct 2004, Johannes M Zmoelnig wrote:
> 
>>so we just have to force people to have kernel-headers installed.
>>i think we can assume, that most linux will have the current
>>kernel-headers in /usr/src/linux (?)
> 
> 
> If it would be that simple. Normally the interface to the kernel drivers get
> included into /usr/include/linux hierarchy if they are stable.

> Now, as the kernel is like a library and we don't know of the ioctl
> API changes, you have to compile gem against a special kernel version, and
> theoretically you would have to recompile everytime you upgrade the
> kernel.

correct me if i am wrong (i don't know anything about this)
ioctl()-commands are defined(!) numbers that tell the driver (which *is* 
kernel-dependant) to do something.
if we do "ioctl(DV1394_IOC_WAIT_FRAMES, ...)" this is the same as doing 
"ioctl(8969, ...)"
so if the API changes, the ioctl() will fail!

thus compiling against dv1394.h (either from a special kernel or 
provided with Gem) could always break when upgrading the kernel!
but using the kernel-headers will at least enable us to compile against 
the current kernel (which a gem-included file will not!) and make it 
work again.

however, if they really change the API (e.g. rename DV1394_WAIT_FRAMES 
to DV1394_IOC_WAIT_FRAMES) we will have a problem.
is this likely to change ??


> In practice it might be that the interface doesn't change, and you can
> use the kernel header file. This means, in this case you sort of
> "inofficially freeze" the API (by including the corresponding header file
> in your sources, just for convenience).
> In this case you have to be aware that strange things could happen.

ah, i think this is, what i have just tried to explain.

the official stable kernel-headers (/usr/include/linux) are versioned 
(on debian) as 2.5.99
this does not look very "stable" to me
as i read the /usr/share/doc/kernel-source-2.6.7/README.headers.gz file, 
the /usr/include/linux is mainly for libc6


> Then, as another wild guess, probably the libraw1394 can be used
> for the task ?

maybe (??)

but why do we have a kernel-module for dv1394 if we cannot/mustnot use it ?


mfg.as.dr
IOhannes