preprocessor bug in gcc 2.95.2[.1]

[Zack Weinberg]

On Wed, Mar 07, 2001 at 11:18:39PM -0800, Laurent Montaron wrote:
> 
> Interesting problem seen on on Intel 32-bit platforms such as cygwin32
> or i386-*-svr4 with GCC 2.95.2 and 2.95.2.1 .
> 
> The preprocessor evaluates the expression "(~0UL) == (0xFFFFFFFFUL)" as
> false, whereas it is true on IA-32, where "unsigned long int" is a
> 32-bit quantity.
> 
> As a result, the following piece of code (from /usr/include/string.h in
> IBM DYNIX/ptx v4.6) doesn't preprocess properly on Intel platforms:
> 
> #if ((~0UL) == 0xffffffffUL)
> typedef unsigned int    size_t;         /* ILP32 size_t */
> #else
> typedef unsigned long   size_t;         /* LP64 size_t */
> #endif /* ((~0UL) == 0xffffffffUL) */
> 
> Amusingly, the GNU preprocessor evaluates "(~0UL) == (0xFFFFFFFFFFFFFFFFUL)"
> as true, which doesn't make much sense to me.

C99 requires that preprocessor arithmetic be carried out in intmax_t,
which is 64 bits on most targets supported by GCC.

zw