NumPy dependency is lacking in the requirements.txt

Hey, Marco!

First of all, great work! vedo is amazing and it eases the pain to use VTK. Thanks!

I’m opening this issue due to a very minor problem I have faced. I tried vedo from CLI and it complained of:

ModuleNotFoundError: No module named 'numpy'

Since you mentioned in the README.md that vedo is based on VTK and numpy, maybe including numpy as a requirement would make sense. What do you think? Could I submit a PR?

Cheers! Diego

Steps to reproduce the problem

Below I described exactly what I did to make it reproducible if you want to check (or vedo interested users):

OS: Ubunutu 18.04.4

Procedures: I installed vedo in the development mode. It means that I cloned your repo and executed in an isolated env:

$ virtualenv .vedo
$ source .vedo/bin/activate
$ pip install -r requirements.txt
$ python setup.py install

Then I ran vedo with its command-line interface:

$ vedo pore_network.vtu

The .vtu file is attached here.

[marcomusy]

Hi Diego, thanks a lot for drawing my attention to this and for the explainations! i just made a quick check (hope it makes sense) by adding the numpy in the install_requires then doing pip install . -U did not mess up with the conda environment… so I suggest we add it in the setup.py only!

[volpatto]

I’m not sure if ‘numpy’ has to be added in the setup.py only or also in the requirements.txt

Hmm… this is a good question. As pointed by PyPA, install_requires should contain only the very essential dependencies, while requirements.txt can be more extensive, maybe when some dependency is useful for a subset of use cases. In my understanding (please correct me if I’m wrong), numpy is essential for vedo, since it can compromise a big feature of vedo (CLI usage). So, in this case, I think that it would be appropriate to include numpy in both. Well, in either way, vedo would have only 2 dependencies, it’s still tidy and clean. Are you OK with that? Please feel free to disagree 😃

(some time ago i remember that trying to do pip -U also updated the conda’s numpy version which was unintended… but i’m not expert of pip!)

Hmm! Yes, if you have conda’s python activated, and then run pip -U, it probably will mess with Python packages inside the conda env. But, IMHO, this is OK. Actually I think that to mix conda and pip to manage packages can be a source of problems (oh, boy… I have faced hard times, I don’t recommend it), this is a not recommended practice AFAIK. Better choose only one and stick together with it until the end 😉