Description of problem: Headers in site-packages/ /numpy/core/include/numpy should probably be installed to /usr/include/python-version/numpy/ Version-Release number of selected component (if applicable): How reproducible: Steps to Reproduce: 1. 2. 3. Actual results: Expected results: Additional info:
How much would code using these headers have to change in order to work with this?
I don't know. What code is using these headers and expecting to find it in the original location? I forgot the original motivation for this request, but I suspect it was because I couldn't build scipy without this change.
For the moment, I'm going to close this wontfix, since neither the impacts or benefits are known at the moment. I have no problem moving them with proper justification though, so feel free to reopen if there's a compelling reason to do so.
(In reply to comment #3) > For the moment, I'm going to close this wontfix, since neither the impacts or > benefits are known at the moment. I have no problem moving them with proper > justification though, so feel free to reopen if there's a compelling reason to > do so. Reopening. When using cython and 'cimport numpy as np' cython wants to have that headers in /usr/include/ and therefore it would be great to have them there. For now I just copied the header files manually :( (In reply to comment #1) > How much would code using these headers have to change in order to work with > this? Cython uses a setup.py file, see [1] to this would need to be changed upstream completely. Because upstream didn't do that before, I assume, they use a distro, in which the headers are in /usr/include/ Ubuntu symlinks from /usr/include/numpy -> /usr/share/pyshared/numpy/core/include/numpy [2] [1] http://docs.cython.org/src/userguide/source_files_and_compilation.html#basic-setup-py [2] http://packages.ubuntu.com/search?searchon=contents&keywords=numpy&mode=filename&suite=karmic&arch=any
Would a symlink for a directory suffice here?
I don't see why not, although I think for Fedora this is a bit unconventional.
Looks like the same subject was discussed here: http://permalink.gmane.org/gmane.comp.python.numeric.general/36425
Fixed and built in rawhide.
What is the decision then? Install headers? Install symlinks? Do neither?
1 symlink.
IMHO, this would prevent the future possibility of installing multiple versions of numpy. CMake can automatically find the correct include path --- it internally calls `numpy.get_include()` to get the path.