Blackify

[trexfeathers]

This PR adds Black and support for pre-commit git hooks, using SciTools/iris#3518 as a guide where appropriate.

Had to move conda_requirements.txt since Black was treating it differently when it was in the repo root, and I was unable to exclude it via a custom exclude entry in pyproject.toml - weird.

Black is currently returning errors for README.md and pyproject.toml, but this is the case for SciTools/iris as well and doesn't cause problems - we don't need these formatting and it seems simpler to not need a custom exclude entry at all.

[stephenworsley]

This seems different to the way Iris is set up to test for black compliance
https://github.com/SciTools/iris/blob/d16f6767d4d775e2afd691de9c6514782ba2a90d/.travis.yml#L108-L113
In particular, it looks as though black --check is being run for iris. Is this test equivalent? If there is a difference, is there a reason for that difference?

[trexfeathers]

I was hoping to leverage the extra elegance of pytest (which Iris doesn't use) by using the pytest-black plugin. Unfortunately, while pytest-black is under active development, it is not kept up to date on Anaconda, which has resulted in the deprecation error in Travis.

I am unsure whether the correct action should be to explicitly pip install pytest-black, or to invoke black the 'old-fashioned' way like Iris does. Thoughts?

[trexfeathers]

... just trying the pip solution on Travis as a POC ...

[stephenworsley]

I do like the idea of using the latest features of pytest, since we are kind of using this repo to "test drive" pytest anyway. I guess there is a good case for making changes where pytest allows a more elegant solution.

[trexfeathers]

OK well that worked at least, and I've included a TODO to get rid of the pip line once conda gets a later pytest-black version.
Gonna take a break now. I await your verdict!

[stephenworsley]

It's probably worth noting that in Iris, black is pinned to version 19.10b0 https://github.com/SciTools/iris/blob/d16f6767d4d775e2afd691de9c6514782ba2a90d/requirements/test.txt#L4.
This approach means that Iris-ugrid will be tested against a different version of black than Iris. I don't think this should cause massive issues, but it might be worth thinking about.

[trexfeathers]

Have not replicated Iris' line removing .gitignore - testing has shown that Black is successfully skipping .gitignore automatically.

[stephenworsley]

I think Iris also has the line - export INSTALL_DIR=$(pwd) within the script section rather than the install section. I don't know if this is relevant or not.

[trexfeathers]

It's present here too, but further up - line 10

[stephenworsley]

The current travis run seems to give "No Path provided. Nothing to do 😴" despite passing. I wonder if - export INSTALL_DIR=$(pwd) is only applying to the contex of the install: block of code and should be moved down to below line 91 where the script: section begins.

[trexfeathers]

It seemed safer to declare it a second time. As you've alluded to, it seems that environment variables are independent between the code blocks.

[stephenworsley]

Well, "No Python files are present to be formatted. Nothing to do 😴" seems like an improvement at least.

[trexfeathers]

From local testing, I'm expecting:
All done! ✨ 🍰 ✨

[trexfeathers]

The mystery now is why I get 3 passes locally, but on Travis it doesn't like __init__.py...

[stephenworsley]

From what I can tell, the only thing that needs changing now is a newline at the end of the __init__.py file, and I suppose it's good to know it's picking that up. I'd say this is good to go once that's done.

[trexfeathers]

After some trial and error this is now ready for your review again @stephenworsley. Thanks for your patience.

[stephenworsley]

Looks good now, I'm happy to merge this.