Use docker to pin dependencies

[mpharrigan]

No description provided.

[fdmalone]

What's the minimum python version we support now?

[mpharrigan]

Instead of abusing the strategy matrix, I moved the "least supported" test to a new job pytest-minimum. There, I bumped the minimum python version for shady reasons:

• The "minimum" environment uses cirq 0.15, which supports python 3.8 so we could run the job on python 3.8
• however, the way the job works is first you install the full, modern, pinned environment which requires python 3.9 because cirq 1.2 requires python 3.9
• After the full environment, we install the older version of cirq. But that step would fail with the old python

The "correct" fix would be to pin a minimal environment. The correct way to do this would be to have another Dockerfile that uses python 3.8

[mpharrigan]

I have unbumped the minimum python version by doing a correct fix

[fdmalone]

What is the workflow for regenerating new requirement files now? Should we run this script on our local machine and add?

[mpharrigan]

If you run re-pip-compile-in-docker.sh on a workstation with docker installed, it should just work ™️

You can also run the python script outside of docker but its output may vary depending on the environment you run it from

[fdmalone]

More of a discussion question, but it looks like we're moving to only testing python3.10, which seems fine, should we add 3.9 / 3.11?

[fdmalone]

But yeah, this is generally a much simpler way of managing environments.

[mpharrigan]

ok, for better or for worse I have changed my pip-compile driver script to be able to support multiple "platforms" where each platform is a collection of environment (pinned requirement) files. Within a platform they're all consistent.

So here: the default envs/ directory has all the normal stuff with the latest cirq whereas max_compat/ envs all have cirq constrained to be 1.0.0

furthermore: the python version of the docker image is now parameterized so we generate the envs/ default envs on python 3.10 and the max_compat/ envs on python 3.8

[fdmalone]

@mpharrigan is this good to go?

[mpharrigan]

Sorry, I'll have to revisit this and give it the once-over to make sure it's good to go

[fdmalone]

I need to regenerate as in #843 for black. I'll close #843 in favour of this.

[fdmalone]

Assuming the macos runners are down for the moment.