Added np.get_include() to Cython autowrapper, fixes #8847. #8848
Conversation
|
If you look at
The one in the past didn't, but it also didn't work. How would we support arrays without it? |
| "\n" | ||
| "setup(\n" | ||
| " cmdclass = {{'build_ext': build_ext}},\n" | ||
| " ext_modules = [Extension({ext_args}, extra_compile_args=['-std=c99'])]\n" | ||
| " ext_modules = [Extension({ext_args}, extra_compile_args=['-std=c99'])],\n" | ||
| " include_dirs = [np.get_include()],\n" |
There was a problem hiding this comment.
Maybe the numpy lines of the template should also depend on _need_numpy? Assuming that that bit of code works the way I intended, this would allow scalar expressions to be compiled without NumPy being installed.
There was a problem hiding this comment.
Ok, I didn't notice _need_numpy. I'll make it dependent on that.
- Made CodeWrapper a new style class. - Moved _need_numpy into __init__ so it doesn't pollute other instances. - Optionally include numpy include directories in Cython setup.py.
moorepants
force-pushed
the
issue-8847
branch
from
fad508c
to
1615390
Compare
|
I verified this works on Mac OSX. @jcrist once tests pass and you approve I'll merge. |
|
Looks good to me. |
Added np.get_include() to Cython autowrapper, fixes #8847.
@jcrist This fixes this issue for MacOSX (and probably anyone that has the NumPy headers in an odd place). But after doing this, I was wondering why we have numpy calls in the CythonWrapper. Seems like you should be able to Cythonize sympy expressions without having numpy installed. Did the CythonWrapper always require numpy? Maybe we really need two Cython wrappers, one with the numpy headers and another without.