Re: [RFC] replace assorted ASSERT()s by something officially sanctioned

[Jeff Garzik <jgarzik@pobox.com>]

Dean Nelson wrote:
> It doesn't appear that an officially 'sanctioned' version of ASSERT() or
> an ASSERT()-like macro exists.
> 
> And by the proliferation of its use in the linux 2.6 kernel (I saw over
> 3000 references to it), it would seem that BUG_ON() does not satisfy all
> of the requirements of the community.
> 
> One problem with BUG_ON() is that it is always enabled. And even though
> the compiler does a good job of minimizing the impact of the conditional
> expression, there are times when the conditional check requires the
> accessing of a cacheline that would not get accessed had the BUG_ON() not
> been enabled. And if that cacheline were one that is hotly contended for,
> one's performance can be adversely affected.
> 
> For debugging purposes it would be nice to have a version of BUG_ON() that
> was only enabled if DEBUG was set. This is what appears to be behind the use
> of the ASSERT()-like macros.
> 
> As an example of what I have in mind, I've included the following quilt
> patch.
> 
> Thanks,
> Dean
> 
> 
> Index: linux/include/asm-i386/bug.h
> ===================================================================
> --- linux.orig/include/asm-i386/bug.h
> +++ linux/include/asm-i386/bug.h
> @@ -21,6 +21,12 @@
>  
>  #define BUG_ON(condition) do { if (unlikely((condition)!=0)) BUG(); } while(0)
>  
> +#ifdef DEBUG
> +#define DBUG_ON(condition)	BUG_ON(condition)
> +#else
> +#define DBUG_ON(condition)
> +#endif

This won't work as it often needs to be driver-granular.  Also, 
WARN_ON() is often the closer implementation of assert(), than BUG_ON()

	Jeff