Trouble installing from source

[zaccharieramzi]

I use a virtual environment with python3.5 and I am trying to install odl form source using information from the docs (the following example requires a working virtualenv and git):

mkdir tests
cd tests
git clone https://github.com/odlgroup/odl
cd odl
virtualenv -p /usr/bin/python3.5 venv
source venv/bin/activate
pip install -e .

However I get the following error (summarized here to the main point):

ImportError: No module named 'numpy'

It required me installing the needed packages as listed here (pip install numpy future packaging scipy) to get it to work. Maybe there is something mixed up with the setup.

[kohr-h]

Hi @zaccharieramzi, thanks for reporting. This is a flaw of setuptools, which tries to import the module to determine the version before dependencies have been installed, see here: pypa/setuptools#1724

We should update our docs, however, to make the -e . installation work again as documented.

[kohr-h]

From the answer I got at the referenced issue in the setuptools repo, it seems as though the issue lies quite deep and won't be fixed. So we have to come up with something ourselves.
I got a pointer to a documentation page that lists several options for "single-sourcing" version numbers. Most of them are horrible, but the last one with setuptools_scm seems interesting. It would only add a setup dependency, not a runtime one, so I don't think it would be too bad. Looking into it.