Re: [PATCH] new ioctl type checking causes gcc warning

[Arnd Bergmann <arnd@arndb.de>]

On Saturday 13 September 2003 01:43, Kevin P. Fleming wrote:

> After working on this some more this afternoon, I realize now that
> it's much better to have the typechecking in place than not, even for
> userspace. Maybe the best solution is to still leave the typechecking
> (don't wrap it in #ifdef __KERNEL__), and just
>
> #ifdef size_t

This doesn't work, because size_t is a typedef, not a macro.

> extern size_t __invalid_size_argument_for_IOC;
> #else
> extern unsigned int __invalid_size_argument_for_IOC;
> #endif
>
> Would the type specification of this non-existent variable ever
> actually effect the generated code? If not, then even putting this
> #ifdef in is overkill.

No, but as Andreas pointed out earlier, doing non-optimized builds
with the _IOC_TYPECHECK macro in place always results in link
errors, even for correct code. Since we know that the kernel
is always built with -O2, '#ifdef __KERNEL__' is sufficient
here.

The type checking this in user space is not necessary, because 
the point of the check is only to keep people from adding *new*
invalid ioctl numbers and doing the check for the kernel does that.
However, the old numbers need to be kept for a long time and there
is no point in breaking user applications that use established
interfaces.

	Arnd <><