Update tmpdir usage to tmp_path and enable no:legacypath

[braingram]

Usage of tmpdir in ci_watson prevents packages that use ci_watson from using the -p no:legacypath pytest option to enforce tmp_path instead of tmpdir usage.

Although _jail is clearly marked with a leading underscore it is used in downstream:
https://github.com/search?q=repo%3Aspacetelescope%2Fjwst%20_jail&type=code
https://github.com/search?q=repo%3Aspacetelescope%2Fromancal%20_jail&type=code

This PR updates this package to use tmp_path instead of tmpdir and enables the -p no:legacypath option.

The failure of the slow-devdeps CI job also occurs on main and appears related to numpy 2.0.

[pllim]

LGTM but @jhunkeler wrote this so maybe he can confirm.

[pllim]

tox.ini needs to be updated to make devdeps pass. Add this. GitHub won't let me do it for some reason.

devdeps: pyerfa>=0.0.dev0

[jdavies-st]

You may want to just yield the tmp_path instead of its string form?

That is what is done here

https://github.com/astropy/astroquery/blob/4d2311745752ffc0133e439053dd4046e4a8294b/astroquery/conftest.py#L70-L78

in a similar fixture I wrote based on _jail.

This would allow one to be sure one is pathlib compatible, though it might break a few things downstream. Though perhaps the breaking is good?

Or one could leave this fixture unchanged and make a properly-named fixtured called tmp_cwd that can be used that yields a pathlib.Path object, and can be transitioned to at anytime by any package that uses ci-watson.

[braingram]

Thanks for the question and comments.

I added the str to keep this consistent for the downstream (as you mentioned changing it to a Path might break things).

Given the naming (_jail) I would have assumed this was not intended for "public" use. So replacing it with a public fixture (or having the downstream libraries implement their own for their needs) seems reasonable.

To use -p no:legacyapth here (to keep this test suite updated) _jail would need to be updated (as is done in this PR).

[pllim]

In this package, this fixture is also used for HSTCAL. So unless you feel like patching up hstcal and acstools test suite, backward compatibility is a must.

[jdavies-st]

Definitely not! Let's keep backwards compatibility. 😅

[pllim]

Oops sorry, also have to add this to PIP_EXTRA_INDEX_URL

https://pypi.anaconda.org/liberfa/simple

[braingram]

Oops sorry, also have to add this to PIP_EXTRA_INDEX_URL

https://pypi.anaconda.org/liberfa/simple

Thanks for your help! CI is now green.

I'm not really seeing any numpy usage (grepping finds no imports of numpy) in this package so maybe numpy 2.0 testing isn't needed? That's certainly outside the scope of this PR.

[pllim]

Does CRDS use numpy/astropy?

ci_watson/setup.cfg

Line 30 in 2fc02e8

crds

Also, if you want to test against astropy-dev, you should probably also pull in numpy-dev, but you don't have to. Your call.

[pllim]

Thanks, all!