On Tue, 16 Apr 2002 08:27, Greg KH wrote:
> But to shorten this message, user space should _not_ be including any
> kernel space headers, end of story. See Linus's postings on lkml about
> this topic.
http://uwsg.iu.edu/hypermail/linux/kernel/0007.3/0587.html for the interested.
However ioctl() is a special case (aka, the story continues). For exactly the
same reasons that Linus pointed out: the user space apps have to match the
underlying support.
Only difference is: underlying support for ioctl() is not glibc - it is the
running kernel.
Here is an example I came across recently, in the evdev support.
#define EVIOCGID _IOR('E', 0x02, short[4])
Now, you have no real chance of figuring out what the values in argp will be.
So I proposed a change:
#define EVIOCGID _IOR('E', 0x02, struct input_devinfo)
Now, do you think I can reasonably update my kernel to make this change, and
use the old kernel space headers? Maybe in some cases (including this one),
but not in general.
It is exactly the same situation that Linus identified, except that it happens
in the kernel, instead of the libs.
> Because of this I don't really see what you want to change here. What
> would be changed in include/linux/hiddev.h that would help out
> user space or any of the *BSD people?
Remove anything that is linux unique - use standard types. I will make a
patch, later.
Brad
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
