Re: gnulib warnings when compiling Faddeeva.cc

[Steven G. Johnson]

On 12/9/12 2:01 PM, Júlio Hoffimann wrote:
> Yes, John explained it to me, there is a trick with _GL_CXXALIAS_SYS and
> _GL_CXXALIASWARN. The first macro creates an alias to the system
> implementation and the second produces the warning whenever we use the
> function without the gnulib namespace.
>
> What about the copysign() then? It's a C++11 feature.

copysign is also specified by POSIX.

But it's not available on Windows, where MS uses _copysign instead.

I'll have an updated version of my Faddeeva code posted shortly that 
only uses copysign in C++11, if HAVE_COPYSIGN is #defined in config.h, 
or if __linux is defined, and also fixes a couple of other small porting 
problems.

> By the way:
> http://en.cppreference.com/w/cpp/numeric/math/erf
> http://en.cppreference.com/w/cpp/numeric/math/erfc

Yes; I use std::erf and std::erfc for real arguments if HAVE_ERF and 
HAVE_ERFC are defined, or in C++11.  The standard doesn't provide the 
complex version, however.