[Numpy-tickets] [NumPy] #522: Compiling extension modules/libraries should use individual compilers

[NumPy]

#522: Compiling extension modules/libraries should use individual compilers
-----------------------------+----------------------------------------------
 Reporter:  pearu            |       Owner:  pearu
     Type:  task             |      Status:  new  
 Priority:  normal           |   Milestone:  1.1  
Component:  numpy.distutils  |     Version:  devel
 Severity:  normal           |    Keywords:       
-----------------------------+----------------------------------------------
 When a setup.py defines two extension modules, one is pure C and the other
 one is C++,
 then C++ compiler is used to compile both extension modules. This feature
 is harmless
 (in the world of C/C++) and inherited from Python std distutils.

 When dealing with extensions with Fortran sources
 then this feature should be changed: compiling and linking extension
 modules should
 use "minimal" compiler combination. Note that if there exists an extension
 module which
 uses Fortran codes then also all other extension modules, no matter if
 they contain only
 C or C++ sources, are linked with a Fortran compiler.

 This will cause problems when setup.py defines C++ based and Fortran based
 extension
 modules at the same time. Similar problems may appear when there is a
 choice to made
 between F77 or F90 compilers.

 So, I am planning to fix this by using "minimal" compiler set for each
 extension module
 based on the language keyword.

-- 
Ticket URL: <http://projects.scipy.org/scipy/numpy/ticket/522>
NumPy <http://projects.scipy.org/scipy/numpy>
The fundamental package needed for scientific computing with Python.