🚀 go-pugleaf

> #2)  GCC fails to detect the error in the following code, even when the
>      -pedantic-errors option is used.
> 
> unsigned long overflow = LONG_MAX + 3; /* ERROR */

I presume this is being called erroneous because of the constraint:

#  Each constant expression shall evaluate to a constant that
#  is in the range of representable values for its type.

in section 3.4 (ANSI) / 6.4 (ISO).  However, 3.4/6.4 also says:

#  The semantic rules for evaluation of a constant expression
#  are the same as for non-constant expressions.

This presumably includes the assertion in 3.3/6.3:

#  If an "exception" occurs during the evaluation of an expression
#  (that is, if the result is not mathematically defined or not in
#  the range of representable values for its type), the behavior
#  is undefined.

Since the constraint on the constant-expression applies to the *result*
of evaluating it, but the undefined behavior due to overflow applies
*during* the evaluation, I think "undefined behavior" wins.  That is,
the compiler is allowed to ignore the overflow and compute some value
for the constant-expression that is a valid long value.  Having done
so, it has satisfied the constraint and need not report an error.

It may be asked what, in that case, the words "in the range of
representable values for its type" are doing in 3.4/6.4.  And my
answer is that I don't know; the Rationale is silent on the point.
It seems to me that the words are entirely redundant in this context.
Maybe they were an attempt to specify that the expression had to
evaluate without any exceptions occurring, but if so, I think the
attempt was unsuccessful.
-- 
Mark Brader, SoftQuad Inc., Toronto, utzoo!sq!msb, msb@sq.com
#define	MSB(type)	(~(((unsigned type)-1)>>1))

This article is in the public domain.