Make complex number type from typedef float

[dagss]

It would seem natural to be able to do

ctypedef double mydouble
ctypedef mydouble complex mycomplex

Currently, this is not possible (only float, double and long double -- literally -- are before complex).

This requires changes in two areas:

1. Parser currently is hard-coded to not support it
2. Utility code is output to support complex numbers (before typedefs); this won't work if one has to output complex numbers based on typedefs. So rather than having ModuleNode? output type definitions, one should probably make a dependency tree out of the PyrexTypes? and ask them to declare themselves to code in a depth-first manner (PS: Must check whether a type is visited with ==, not is or an instance flag!)

Usecase: numpy.pxd would be slightly cleaner with

ctypedef np.float64_t complex np.complex128_t

...although in practice, double is 64-bit everywhere. So it is mostly for the cleanup.

[dagss]

No way this will make it into 0.13....

[dalcinl]

Is this a feature available in C? See below:

$ cat tmp.c #include <complex.h> typedef float Float; typedef Float complex FloatComplex?;

$ gcc -c tmp.c tmp.c:3: error: two or more data types in declaration specifiers