Re: [Tinycc-devel] Tinycc issues

[Dominic Mazzoni]

Henrik Nordstrom wrote:
> On Tue, 15 Mar 2005, Dominic Mazzoni wrote:
>
> > 1. NaN in floating-point constants:
> >
> >   It is sometimes appropriate for a user-defined numerical function to
> >   return NaN.  Normally I do this by defining a constant in my code, like
> >   this:
> >
> >     float NAN_FLOAT = (float)(0.0 / 0.0);
>
> This is undefined. May be either a quiet NaN or a signalling NaN 
> depending on the floating point implementation.

That's true, although for all practical purposes all modern CPUs use IEEE 
floats, which is well-defined, and all modern POSIX systems let you 
control whether to use quiet or signalling NaNs...

> You should use the NAN macro if available.
>
> Example (Linux):
>
> #define _ISOC99_SOURCE
> #include <math.h>
>
> int main(int argc, char **argv)
> {
>     printf("%f\n", NAN);
> }
>
> If NAN is not defined (it's a macro) then the floating point 
> implementation does not support quiet NaNs.

Thanks, that will work fine.

However, the crash in tinycc should still be fixed.  :)

- Dominic

> Regards
> Henrik
>
>
> _______________________________________________
> Tinycc-devel mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/tinycc-devel