Hi, I'm not entirely familiar with this situation but yes, I've seen this bug appear in other libraries such as scikit-* and I think it's mostly related to the way pip handles dependencies (i.e,. they are not marked as installed as the whole transaction is done, which won't happen because well, pyopencl will fail because numpy is not installed).
Cheers! -Santiago. On Thu, Jan 24, 2019 at 03:05:19PM -0200, Goncalo Morgado wrote: > Hey hey :) > > I am fairly new with pyopencl. Still getting the grip of it... > > Although the docs suggest installing pyopencl with conda I have been using > pip, which seems to work quite well, apart from an issue with dependencies. > pip install pyopencl on a freshly created virtual environment will complain > about missing numpy and warn about missing pybind11 and mako. > > Regarding numpy, and if I understand correctly, some effort has been put ( > https://github.com/inducer/pyopencl/blob/master/aksetup_helper.py#L69) to > fix the issue with the class "ExtensionUsingNumpy" as opposed to > "NumpyExtension". > I gave it a try with the latest version in the github repo, and `pip > install .`, but it still complains about missing numpy. > Should I install dependencies manually beforehand? Shouldn't this be > handled in setup.py with setup_requires=[] and install_requires=[]? > From a user perspective it makes a lot of sense to just run `pip install > pyopencl` and have all dependencies installed, similar to what happens when > we install scipy or matplotlib. > > Thanks in advance, > Goncalo > _______________________________________________ > PyOpenCL mailing list -- pyopencl@tiker.net > To unsubscribe send an email to pyopencl-le...@tiker.net
signature.asc
Description: PGP signature
_______________________________________________ PyOpenCL mailing list -- pyopencl@tiker.net To unsubscribe send an email to pyopencl-le...@tiker.net
