Clarification re: installation

[bhrutledge]

The README suggests pip install pip-tools, but it's not clear what environment that should be run in. Some of the options that I see:

1. System Python (e.g., distributed with OS): generally not recommended
2. Third-party Python (e.g., Homebrew, python.org, pyenv, Anaconda)
3. Virtual environment dedicated to pip-tools
4. Each project's virtual environment

In my case, I chose option 3, using pipsi to streamline adding it to my PATH. I've installed other tools like fabric and cookiecutter in a similar manner, whereas I install things like flake8 using option 4.

I think option 1 or 2 would be safer with pip install --user.

[vphilippon]

Considering the pip-sync command, it needs to be installed in the same virtualenv that you want to sync, which is each project virtualenv, because pip-sync will simply perform packages install/uninstall in the current active virtual environment.
That means option 4.

I don't know much about pipsi, but if tools installed with pipsi are available when in another virtualenv (like a given project's virtualenv), then it could work (or cause horrible unforeseen side-effect). If you want to give us you're feedback on how the pip-sync command work in these condition, it could be good to know, and document if needed.

At the moment, I think that option 4 is the only fully valid option that will work with pip-tools as a whole (not only pip-compile). That should indeed be clarified in the doc.

[bhrutledge]

I've been using pip-sync successfully with pipsi, which I think makes sense based on your description. But, I'm happy to update the docs to suggest option 4, as that seems the easiest and least surprising.

On that note: is there any suggestion for including pip-tools as a development requirement in setup.py or a .in file?

[vphilippon]

@bhrutledge It might not seem like it, but that's a pretty loaded question right there. It points back to the broader question "how do I deal with dev requirements in Python", and there's not a single viewpoint on this in the community, AFAIK.

Although, I'm not an expert on all the fancy options available in setup.py using setuptools, but I would be very careful about including pip-tools somewhere in a setup.py. I would deal with pip-tools the same way you deal with pip. If there was a "dev_requires" section in setup.py, it could be the right spot, although it would need some testing to make sure everything goes smoothely.

Personally, I keep my dev packages in a separate dev_requirements.in/.txt. But in the end, it comes down to using a way that fits your need, as long as you're aware that this is part of your "python environment tooling", like pip is.