Cython, remove numpy dependency in setup.py - numpy

Question

In order to build a Cython extension using numpy, one has to add numpy.get_include() in "setup.py":
from setuptools import setup, Extension
setup(
...
include_dirs = [numpy.get_include()]
)
Otherwise, one gets this error:
fatal error: numpy/arrayobject.h: No such file or directory
even if one builds the Extension directly from the precompiled .c file.
The obvious solution is to install numpy beforehand.
But setup.py install to install a package is the usual way to install all the package dependencies (including numpy). Is there any way to get rid of the dependency to numpy when distributing the package, i.e. so that people don't have to install numpy before running setup.py install ?
I was thinking about including the numpy .h files into the distribution, but I am afraid of version clashes with an existing numpy version of the user.
Maybe try to import, and if I get an ImportError, include the packaged .h files? Any standard/simpler way?
Edit: or is there any way to force the installation of numpy before setup() runs?

Answer 1

It may be ugly but I did this to force the install/upgrade:
from pkg_resources import parse_version
## Install numpy if it is not found or too old
try:
import numpy
if parse_version(numpy.__version__) < parse_version('1.10'):
print("numpy {} was found but is too old. Upgrading.".format(numpy.__version__))
raise ImportError
print("Numpy was found. Build extensions.")
except ImportError:
print("Building Cython extensions requires numpy. Installing numpy.")
import pip
pip_args = ['install', numpy_req]
pip.main(pip_args)
import numpy