Re: [PATCH 1/4] the CMPLX macros must be usable in initializations of static variables

[Szabolcs Nagy <nsz@port70.net>]

* Jens Gustedt <Jens.Gustedt@inria.fr> [2014-11-25 15:49:39 +0100]:
> diff --git a/include/complex.h b/include/complex.h
> index 13a45c5..3e14e04 100644
> --- a/include/complex.h
> +++ b/include/complex.h
> @@ -112,8 +112,7 @@ long double creall(long double complex);
>  #define cimagf(x) __CIMAG(x, float)
>  #define cimagl(x) __CIMAG(x, long double)
>  
> -#define __CMPLX(x, y, t) \
> -	((union { _Complex t __z; t __xy[2]; }){.__xy = {(x),(y)}}.__z)

hm the compound literal is not a constant expression

that is sad

> +#define __CMPLX(x, y, t) ((t)(x) + _Complex_I*(t)(y))
>  

this is only correct if the compiler supports annex g kind of
imaginary type, otherwise it is incorrect for infinites

_Complex_I*INFINITY == (0 + i)*(inf + i*0) == nan + i*inf

so imaginary inf will turn into nan real part

this is what gcc (and i assume clang) does

>  #define CMPLX(x, y) __CMPLX(x, y, double)
>  #define CMPLXF(x, y) __CMPLX(x, y, float)
> -- 
> 1.9.1