[SciPy-user] Syntax for assigning to a complex matrix via weave?

[Fernando Perez]

eric jones wrote:
>>From the paper that discusses a superior complex<T> implementation:
> 
>     """
>     <talks about implementation>
>     This proposal is already implemented and used in the GNU 
>     Implementation of the C++ standard library.
>     """
> 
> I guess we could simply use this implementation in weave.  The only
> issue is that it is GPLed and so it can't just be dropped into the
> distrbution.  A second option is just to overload the real() and imag()
> methods to return T& instead of T in a weave specific implementation of
> complex.  Thoughts?  Is it really worth breaking with the standard
> interface?

I'd say -0.5.  Breaking standards is in most cases a road I'd rather not 
travel, if nothing else because of how it can cause subtle problems in the 
future.  This seems to be under discussion by the C++ standards people, so it 
might be worth chiming in with a voice pushing for a sane solution in the 
standard itself.

On the other hand, the current situation is ugly enough that having a sane 
syntax which allows one to simply do:

complex_array(i,j,k).real() = something;
complex_array(i,j,k).imag() = something_else;

would definitely be nice (even better if the parens weren't necessary, but 
that's _really_ breaking the standard).

I can't understand why they didn't settle on complex numbers being by default 
equivalent to struct { <type> real; <type> imag; }.  This would address a 
large majority of the usage cases, enable interoperability with every 
Fortran/C code out there, and one could always write a PolarComplex class for 
the situations where a native polar representation is truly necessary.

Oh well...

Cheers,

f.