Bruce Evans bde at zeta.org.au
Sun Nov 16 04:36:48 PST 2003

On Sun, 16 Nov 2003, Jacques A. Vidrine wrote:

> On Sun, Nov 16, 2003 at 04:20:10AM -0600, Jacques A. Vidrine wrote:
> >     /* How can this be implemented correctly? */
> >     int range_error(long n, time_t t)
> >     {
> > 	    return (long)(t = n) == n;
> >     }
> Hrmp.  Because time_t is probably signed, technically this can cause
> `undefined behavior' if the range of `long' is more than the range of
> `time_t' (e.g. on alpha). *sigh*

Actually, it's implementation-defined if time_t is integral (doesn't
matter if it is signed or unsigned) (and the value is not representable).
It's only undefined if time_t is a floating type.

> All I really want to do is correct a parsing bug and at the same time
> eliminate a warning so that I can set WARNS?=1 in libc before the code
> freeze.

You can safely assume that it won't change to floating before the code
freeze :-).

I think (t = n) would cause compiler warnings at higher WARNS levels
if time_t were unsigned.  `t == n' certainly would.
`(long)(time_t)n == n' could be used (this is like the above except it
doesn't use a temporary variable.  However, the cast to long breaks the
warning about the bug that if n is -1L and time_t is unsigned long, then
the comarison will succeed on 2's complement machines although time_t
cannot represent -1.


More information about the freebsd-arch mailing list