Originally posted by: unrue

Dear IBM developers, 

I'm using xlc 13.1.5 on Power 8 system. I've noted that CHAR_MIN from limits.h is set to 0. 

Under Intel machine otherwise is -128.

I have a code with some cast from float to char. When the float is negative, under IBM the cast return 0 and the program doesn't works well. Is there any particolar reason of such difference?

Is it possible to change CHAR_MIN value?

Thanks. 

Originally posted by: Rafik_Zurob

Hello

In POSIX, the value of CHAR_MIN is tied to whether "char" is the signed or unsigned.  To make it -128, compile with -qchars=signed.  i.e. That would make "char" signed.  XLC's default is that "char" is unsigned.

> cat x.c
#include <limits.h>
#include <stdio.h>

int main() {
  printf("%d\n", (int) CHAR_MIN);
}
> xlc x.c
> ./a.out
0
> xlc x.c -qchars=unsigned
> ./a.out
0
> xlc x.c -qchars=signed
> ./a.out
-128
>

Originally posted by: ThinkOpenly

"-qchars=signed" ("-fsigned-char" for GCC) is a big lever to throw, which could have undesired effects in other code.

It can be argued that since a "char" has an architecturally defined signedness, it is more appropriate to change the variable declarations to the match the desired behavior.  In other words, if the signedness of a char variable matters, declare that variable as "signed char" (or "unsigned char").  (I'm ignoring whether such changes are practical.)

Note that limits.h defines, for "signed char" minimum value:

#  define SCHAR_MIN     (-128)

and then defines CHAR_MIN to that or 0, based on the architecture's default signedness for the "char" type.