Use __uintptr_t in place of unsigned long in UNCONST() macros.

[brooksdavis]

The C language only guarantees that pointers can be reconstituted after being cast to an integral type if that type if that type is (u)intptr_t.

[jmmv]

(I took the original UNCONST implementation from the original NetBSD headers so didn't question it much.)

But about your change: if the cast should be against uintptr_t, why does the code say __uintptr_t? Note that the tests are failing.

[brooksdavis]

__uintptr_t is define by the compiler and thus is more appropriate for a normal UNCONST macro in a system header since it has no dependencies. In practice uintptr_t should be fine and might be more appropriate here.

I see about 30 test failures with and without this change on FreeBSD 10.2. In Travis CI, it looks like there's an explicit exit 2 in the script somewhere which is triggering failures here.

[jmmv]

You have to scroll up (before the dump of the config.log contents) to see the actual build error.

[brooksdavis]

It looks like the previous versions were working due to header pollution on FreeBSD (both clang and gcc5 built without warnings without stdint.h included).