On Thu, 17 Nov 2011, David Daney wrote:
> A counter argument would be:
>
> There are hundreds of places in the kernel where dummy definitions are
> selected by !CONFIG_* so that we can do:
>
> if (test_something()) {
> do_one_thing();
> } else {
> do_the_other_thing();
> }
>
>
> Rather than:
>
> #ifdef CONFIG_SOMETHING
> if (test_something()) {
> do_one_thing();
> } else
> #else
> {
> do_the_other_thing();
> }
>
>
>
> We even do this all over the place with dummy definitions selected by
> CONFIG_HUGETLB_PAGE, What exactly makes HPAGE_MASK special and not the
> hundreds of other similar situations?
>
Dummy functions that return 0 when a feature isn't enabled isn't the
problem, that's very convenient. Definitions of constants are a
completely separate story because they can be easily used outside the
required context and cause brekage. What happens if you reference a
variable in a function that is declarated in a different function? Does
the compiler put a BUG() in there and let you explode at runtime? This is
just absurd, every other arch has done the necessary declarations in their
own header files and everything works fine because there's a clear config
dependency on using HPAGE_*. Adding dummy definitions to panic at runtime
is irresponsibly to an extreme.
