fix -Werror regressions with gcc 4.8 on x86_64

[Paul Eggert]

On 01/06/2015 08:35 AM, Richard Henderson wrote:
> I do wonder about the public header file using an expression like
>
>    __ASSERT_VOID_CAST (0 && (expr))
>
> Side effects are not evaluated, but identifiers referenced in the expression
> still get marked as "used", etc.

Just as a heads-up, the GNU Emacs internals have a macro 'eassert' that 
does that, and it works well; we prefer it to 'assert'. There's also a 
macro 'eassume' that uses __builtin_unreachable; we tried having eassert 
do that too, but it led to too many compile-time false-alarms.

However, the C Standard doesn't allow eassert's implementation 'assert', 
as it requires that if NDEBUG is defined, then 'assert' must be defined 
via '#define assert(x) ((void)0)', which means weird uses like 
'assert(***)' have to work.

Perhaps the public header file could use '0 && (expr)' only when in 
standards-conforming mode?