migrate to cirrus-ci and support mpl 3.4.1

[bjlittle]

The PR migrates the repo from travis-ci to cirrus-ci.

Additionally, it also:

• adds a pre-commit git hook
• rationalises the package requirements into requirements.yml
• flake8 lints the code
• blackifys the code
• adds the require phash images to support mpl 3.4.1

i.e., generally shows some love to test-iris-imagehash 👍

[bjlittle]

The additional image files are a collection of the usual culprits:

• minor anti-aliasing differences
• font and tick changes
• axes pixel changes
• pixel banding between contour levels i.e., the transition from one contour colour level to the next
• some contour rendering differences, but minor broad-brush pixel shifting and differences

Nothing that I'd constitute as a macroscopic visual change i.e., changes the scientific interpretation by the viewer. To the eye, the changes are imperceptible.

[bjlittle]

Ping @jamesp

[jamesp]

Go mambaforge! 👍

[bjlittle]

It seems somewhat total overkill for this repo... but, what the heck 😉

It's a reasonable pattern to adopt and borrow from the collaboration with https://github.com/pletzer/mint/blob/master/.cirrus.yml

If the glove fits... 😉

[jamesp]

awesome 🚀

[trexfeathers]

It seems somewhat total overkill for this repo

I whole-heartedly agree. Was expecting this to a 5min review tops 🙄

[bjlittle]

@trexfeathers My point about overkill, is that the minimum you need to do is quite a lot relative to the content of the repo.

I have added extra "goodness", but that is simply plumbing for best practice i.e., black, flake8 and pre-commit all of which are common patterns elsewhere through out SciTools

[jamesp]

Ok, ci looks reasonable to me and the automated tests are passing which is good 👍

I've done some spot checks of the image hashes you're including and they match what I get with iris master HEAD and matplotlib 3.4.1.

The only one that looks materially different on my instance is: result-iris.tests.test_mapping.TestLimitedAreaCube.test_grid.0.png

This corresponds to https://github.com/SciTools/test-iris-imagehash/blob/a980a1b1fcbf1b6a747c74b959264a2e9c413062/images/v4/fa1585e885ea7a1785fa7a157a177a017a1585e817a885ea85e86a1785fa7a17.png in this PR

extract of imagerepo.json below suggests this grid should have colour structure:
"iris.tests.test_mapping.TestLimitedAreaCube.test_grid.0": [
"https://scitools.github.io/test-iris-imagehash/images/v4/bf80e2b1c17f1d0ac4f7c8d739a637202749699b6bb3ce3666e4b048944d9d89.png",
"https://scitools.github.io/test-iris-imagehash/images/v4/bf80e2f1c17f1d0ac457c8d619a637213749699b6bb34e3666e4b04e944d9d89.png",
"https://scitools.github.io/test-iris-imagehash/images/v4/ea05392995bac6d691ce3f21666569d86a96c6360ee195cb91e8ce54953b313b.png",
"https://scitools.github.io/test-iris-imagehash/images/v4/ea05392995bac6d691ea3f21666569d86a97c6320ee195cb91e8ce559539391b.png"
]

[bjlittle]

@jamesp I suggest we bank this as an issue on iris to investigate further - it's clearly a mpl-ism rather than iris. Perhaps a genuine regression, hence the big hamming distance (clearly I was "Accept"-tastic for the tests)

That would allow us to move forwards here. What do you think?

[jamesp]

Yes agree, this doesn't seem a blocker to merging.

[bjlittle]

Thinking about this it would be nice to capture or integrate the output of idiff here so that the job of the review is much easier.

I'll have a think... 🤔

[bjlittle]

Thinking about this it would be nice to capture or integrate the output of idiff here so that the job of the review is much easier.

I'll have a think... 🤔

[jamesp]

@bjlittle @trexfeathers To be clear, I'm happy for this to be merged, but currently am not marked as a reviewer nor repo owner so cannot press the button

[bjlittle]

@jamesp I've added the SciTools/admins group with Admin rights on this repo (sorry, I thought that was already setup)

You should now be good to go 👍

[bjlittle]

@jamesp Awesome, thanks 🥳