Consider using pipenv for managing packages

[patcon]

Any thoughts on using this tool?
https://docs.pipenv.org/

To explain, I'm having some issues getting the python processing libraries working in an alpine linux docker container. I feel like being a little looser on the specific version constraints in a Pipfile might help people navigate issues like this across systems.

Specifically, there seem to be some off conditions where pandas version X has trouble in docker, and where certain packages must be installed in a specific order -- the alpha order in the requirements.txt doesn't seem to work. Anyhow, I'll sort the specifics elsewhere, but wondering if there's any opposition to using pipenv to generate real dep resolution lockfiles.

Thanks for any consideration

[danvk]

pipenv looks great! Thanks for the pointer. I'd welcome a change to move our requirements.txt file into the new world of Python dependency management.

One distinction that would be nice to draw is between the Python packages that are required for the geocoding work and those that are required for running the site. (This could be dev. vs. prod.) In fact, there should only be Python dependencies for serving the site if you run your own API server (oldtoronto/devserver.py).

[patcon]

🙌 pipenv unfortunately only has 2 groups -- packages and dev-packages. but these might work for us. Alternatively, we could just have more clear division between the packages used, and have separate pipenv/requirements files

[danvk]

I think those two groups should be enough: dev=geocoding pipeline, prod=anything required for serving (currently nothing!).