Re: macro FIXNUM_OVERFLOW_P in lisp.h is valid ?

[Stefan Monnier]

>>>>> "Eli" == Eli Zaretskii <address@hidden> writes:

>> From: Andreas Schwab <address@hidden>
>> Date: Fri, 23 Oct 2009 22:57:07 +0200
>> Cc: address@hidden
>> 
>> >     #define FIXNUM_OVERFLOW_P(i) \
>> >       ((EMACS_INT)(i) > MOST_POSITIVE_FIXNUM \
>> >        || (EMACS_INT) (i) < MOST_NEGATIVE_FIXNUM)
>> >
>> > I think FIXNUM_OVERFLOW_P is problematic.
>> 
>> Thanks for the report.  The right fix is to remove the cast, so that the
>> compiler will promote the operands to the appropriate common type.

> I think removing the cast will cause the compiler to whine on 64-bit
> platforms about comparison being always right or wrong when the
> argument is narrower than a 64-bit long.  IIRC, that's why the cast
> was introduced in the first place.

Actually, when the cast was added in 2001, it came with the following
commit-log-comment:

  (FIXNUM_OVERFLOW_P): Cast I to EMACS_INT in comparisons
  in case I is of some unsigned type, in which case
  MOST_NEGATIVE_FIXNUM will be converted to unsigned, and the
  comparison becomes bogus.


-- Stefan