Re: Fastest minimum-width integer types and limits macros

[David Brown <david at westcontrol dot com>]

On 30/09/2011 08:05, Gunther Nikl wrote:
> Ian Lance Taylor schrieb:
> > Gunther Nikl<gnikl@users.sourceforge.net>  writes:
> >
> > > Is there a special requirement about the value of the limits
> > > macros for fastest minimum-width integers? Eg. would it be legal
> > > to use the limits of a a "signed char" if int_fast8_t is defined
> > > in terms of an int? Or is it required to use the proper limit of
> > > int?
> >
> > I don't understand what you mean.  Can you show us some example code
> > that you are uncertain about?
>
> Here is an example for int_fast8_t:
>
>    typedef signed char int_fast8_t;
>    #define INT_FAST8_MIN INT8_MIN
>    #define INT_FAST8_MAX INT8_MAX
>
> Now if int_fast8_t uses "int" as underlying type like this:
>
>    typedef int int_fast8_t;
>
> which value is its MIN/MAX macros supposed to have? Is it
> allowed to use the same "char" limits as above or do I need
> to use
>
>    #define INT_FAST8_MIN INT32_MIN
>    #define INT_FAST8_MAX INT32_MAX

You mean "#define INT_FAST8_MIN INT_MIN" in this case - you would only 
use "INT32_MIN" if you had "typedef int32_t int_fast8_t".  If you are 
going to write your own <stdint.h>, try to be accurate - otherwise 
you'll end up with something that looks standard and portable, but is 
not quite correct and will cause problems if it is reused on other 
platforms.

I believe it is standard practice that INT_FAST8_MIN and similar limits 
match those of the underlying type, even though it would be legal to 
make it smaller.  Similarly with INT_LEAST8_MIN.  That seems to be the 
case on the various compilers I have lying around.



> in that case?
>
> So this is not about user code but an implementation question.
>
> Thank you,
> Gunther Nikl