Re: C99 _Complex and C++ complex<> interoperability.

[Ed Smith-Rowland <3dw4rd at verizon dot net>]

Howard Hinnant wrote:
> On Feb 22, 2007, at 1:25 PM, Smith-Rowland, Edward M wrote:
>
> > My questions are:
> >  Is this behavior standard or is this a gcc extension?
> >  If this is a gcc extension where is it documented?
> >
> > Also, if this behavior is nonstandard is anyone thinking about
> > standardizing it?
> > Finally, the circle would be complete if we could assign from
> > std::complex to _Complex.
> > Any ideas?  There is a private functions that return _Complex from
> > std::complex.
>
> Fwiw here are some relevant links:
>
> http://www.open-std.org/jtc1/sc22/wg21/docs/lwg-active.html#387
>
> The above link is a C++ committee library working group defect report 
> on this subject.  It is in an open state meaning that no decision has 
> been made.
>
> Below is a link to a paper which further discusses (really just 
> summarizes) the issue:
>
> http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2004/n1589.html
>
> In a nutshell, there is support on the C++ committee for making the 
> C++ and C99 complex types layout compatible.  I'm not aware of support 
> for making them API compatible.  Currently C++ will not parse the 
> syntax for C99 complex types and I know of know proposal to do so.
Actually, g++ happily accepts a mixture of things like 
"std::complex<float>" and "double _Complex" as the code attached to the 
first mail demonstrates.  I will attempt to see if this compiles on 
Visual Studio 2005.

It actually doesn't surprise me that these types could appear side by 
side - "double _Complex" is a built in type like double as far as C99 is 
concerned and of course std::complex<> is just another type.  I think 
that for C and C++ compilers that share a common codebase this is not a 
stretch in practice.

I was a little more surprised that C++ complex vars could take 
assignments from C99 complex and the C++ ctors could take C99 input.

I guess I'll come out and say it: I think the feature of having C++ 
standard complex variables accept input from C99 complex numbers is a 
good idea in the interest of interoperability.  Getting the C++ powers 
to accept _Complex as a new keyword might be a bit of a challenge though 
;-).  But it would make using C libraries within C++ easier.

If this was an accident or an experiment in gcc that will never be 
standard I still think we should put it up as an extension in the docs.
>   However if the complex types were layout compatible one could link 
> C99 and C++ sources and have them refer to the very same complex data 
> structures, even passing them back and forth (just referenced with 
> different syntax in the different translation units).
I really also would add a strong vote to mandate the complex layout - 
contiguous real complex - as standard.
I think I speak for at least some people in the GSL and other scientific 
communities too.
> -Howard

As long as I'm voicing my opinion on such matters (FWIW) I would go for 
std::complex<T>::set_real(T) and std::complex<T>::set_imag(T) if only 
lvalue functions are a tad untraditional in C++.

Ed