Difftime code

[Paul Eggert]

"Clive D.W. Feather" <clive at demon.net> writes:

> No credit?

Sure.  How about adding this comment?

                       ** Thanks to Clive D.W. Feather for detailed technical
                       ** advice about hosts with padding bits.

>> #ifndef HAVE_STDINT_H
>> #define HAVE_STDINT_H		(199901 <= __STDC_VERSION__)
>
> 19910L (can you say "16 bits"?)

No, 16 bits is a red herring.  The expression is used only in
preprocessor contexts, so arithmetic must be at least 32 bits (64 with
C99) and the "L" is unnecessary.  And even if the expression were used
outside of preprocessor contexts, __STDC_VERSION__ can't be negative
so the expression would still work.

Anyway, I'll put the "L" in (if only to forestall pedantic bug reports :-).

>> 	if ((TYPE_BIT(time_t) <= LDBL_MANT_DIG
>> 	     && (TYPE_BIT(time_t) == LDBL_MANT_DIG
>> 		 || (TYPE_SIGNED(time_t) && UINTMAX_MAX / 2 < INTMAX_MAX)))
>
> Why write this last test like this?
> UINTMAX_MAX and INTMAX_MAX are both going to be numbers of the form 2**N-1.
> Furthermore, UINTMAX_MAX >= INTMAX_MAX. So that last one can be done as
> UINTMAX_MAX == INTMAX_MAX.

It's just a style thing, as the two expressions are logically
equivalent under the constraints that you mention.  "UINTMAX_MAX / 2 <
INTMAX_MAX" is more closely related to the constraints that were in my
head when I wrote the code for the case where uintmax has padding bits
and time_t is signed.