Test package installation in temporary virtual environment

[matt-graham]

Resolves #516

Creates a virtual environment in temporary package directory and tests package installation in there rather than installing in parent environment tests are run from.

Calling pip install command by directly referencing Python executable in created virtual environment seems to work in terms of ensuring package is installed in site-packages of temporary virtual environment rather than parent environment (at least I no longer see a cookiecutter-test entry when running pip list after running tests) and it seemed non trivial to source the virtual environment script with subprocess.run.

[matt-graham]

I don't think the link checker check fail is anything to do with changes in this PR as it doesn't change or add any URLs. It looks like it might be due to

[304] https://discord.com/ | Redirect (304 Not Modified): Not Modified

flagging https://discord.com/ as a redirect, potentially because of the trailing slash?

[samcunliffe]

Looks great!

2 ± 1 things come to mind:

• Did you look at pytest-venv? It's unfortunately not receiving so much dev (last commit ~2 years ago). But it (or something similar) might cut down on some of the boilerplate?
• Alternatively we could plop all of this into a fixture (i.e. do as they do in the pytest-venv plugin but with our own code.
• The whole package generation test... um is there a smarter way to do it (I appreciate that is less of a comment for this bugfix and more of a general unhelpful musing).

[matt-graham]

• Did you look at pytest-venv? It's unfortunately not receiving so much dev (last commit ~2 years ago). But it (or something similar) might cut down on some of the boilerplate?
• Alternatively we could plop all of this into a fixture (i.e. do as they do in the pytest-venv plugin but with our own code.

I hadn't come across pytest-venv and my cursory searching for pytest temporary virtual environment didn't throw it up, but it does look like it does exactly what we want including dealing with the Windows specific directory layout that caught me out. While the lack of recent commits is a bit of a flag, it is a very minimal package so can imagine there isn't too much that can break and so need maintenance. I'd be happy to add it as a test dependency rather than reinventing the wheel ourselves. Any opinions to the contrary?

[paddyroddy]

The whole package generation test... um is there a smarter way to do it

There must be. I do think it's currently causing more problems than it has solved.

[Saransh-cpp]

There is https://github.com/hackebrot/pytest-cookies, but it is unmaintained as well (tho used for testing in https://github.com/osl-incubator/scicookie).

If we end up migrating to copier, we can use https://github.com/12rambau/pytest-copie.

[matt-graham]

Updated to use pytest_venv fixture rather than manual solution. mypy was complaining that

error: Cannot find implementation or library stub for module named "pytest_venv"  [import-not-found]

so I added a manual override to ignore import. Alternative would be not to add a type annotation for venv argument and add a noqa comment to stop Ruff complaining about this.

[matt-graham]

Link checker flaking again (tried one re-run) this time due to

[503] https://coveralls.io/pricing | Network error: Service Unavailable

[paddyroddy]

Link checker flaking again (tried one re-run) this time due to

[503] https://coveralls.io/pricing | Network error: Service Unavailable

Coveralls is currently down

[samcunliffe]

Nice!

I'd say the new dependency is worth it for the one-liner environmentisation of our messy tests. (I think I've invented "environmentisation".)

[paddyroddy]

Nice, pytest_venv looks much neater than previous iteration

[Saransh-cpp]

Nice! 🚀