Hello,
if I understood the problem, I think it can be easily fixed without using an
inline function.
For example:
#ifdef __GNUC__
#define SCM_MAKE_CHAR(__x) \
({ scm_t_int32 __t = (scm_t_int32)(__x); \
(__t < 0) \
? SCM_MAKE_ITAG8 ((scm_t_bits) (unsigned char) (__x), scm_tc8_char) \
: SCM_MAKE_ITAG8 ((scm_t_bits) (__x), scm_tc8_char); \
__t; \
})
#else
#define SCM_MAKE_CHAR(x) \
((scm_t_int32) (x) < 0 \
? SCM_MAKE_ITAG8 ((scm_t_bits) (unsigned char) (x), scm_tc8_char) \
: SCM_MAKE_ITAG8 ((scm_t_bits) (x), scm_tc8_char))
#endif
solves the problem.
Actually a much more efficient solution specifically designed for GCC may use
__builtin_types_compatible_p() and __builtin_choose_expr() intrinsic functions,
but this will force you to support GCC 3.x or newer.
Something like:
#if defined __GNUC__ && __GNUC__ >= 3
...
#elif ...
will need to be written.
Sincerely,
Carlo Bramini.
---------- Initial Header -----------
>From : [email protected]
To : [email protected]
Cc :
Date : Mon, 17 Aug 2009 17:33:03 +0200
Subject : Re: `SCM_MAKE_CHAR ()' signedness issue
> Mike Gran <[email protected]> writes:
>
> > On my system I ran a test with SCM_MAKE_CHAR as a macro, an an inline,
> > and as a never inlined function. I ran ./check-guile twice for each.
>
> You're cheating! ;-)
>
> The test suite does lots of things besides calling `SCM_MAKE_CHAR ()',
> so I'm not sure if it's a good benchmark.
>
> I'm fairly confident that for such a small piece of code inlining is
> always a good idea.
>
> Thanks,
> Ludo'.
>
