Re: [PATCH] debug: fix BUILD_BUG_ON() for non-constant expressions

[Ingo Molnar <mingo@elte.hu>]

* Linus Torvalds <torvalds@linux-foundation.org> wrote:

> 
> 
> On Sun, 17 Aug 2008, Ingo Molnar wrote:
> > 
> > try the patch below - it only gives this error during build:
> 
> Well, you didn't do it right:
> 
> > +#define __BBO(c)                sizeof(const char[1 - 2*!!(c)])
> > +#define __BBONC(c)              __BBO(!__builtin_constant_p(c))
> > +#define BUILD_BUG_ON_ZERO2(c)    (__BBO(c) - __BBONC(c))
> > +#define BUILD_BUG_ON2(c)         (void)BUILD_BUG_ON_ZERO(c)
> 
> Look at the #define of BUILD_BUG_ON2 a bit more.
> 
> Hint: you're using the _wrong_ BUILD_BUG_ON_ZERO. The old one, not the v2 
> one!

yeah, i already tried various variants earlier today so i really didnt try
that hard with yours. (and i pointed out this mistake in the previous mail)

> That said, with that fixed, there's still something wrong. It does seem 
> like gcc has some very odd interaction there with __builtin_constant_p. 
> Odd.

yeah. I tried various integer arithmetic expressions (which the array trick
relies on) and it didnt work as expected - it's always zero. It only makes
a difference when used in comparisons. (and that's where the kernel uses
__builtin_constant_p quite heavily, and it works fine there.)

Odd indeed.

	Ingo