TST: Handle new numpy 2.0.dev string representation

[pllim]

• Revert CI: use numpy printoptions in testing #211

Example log: https://github.com/scientific-python/pytest-doctestplus/actions/runs/5704149430/job/15457469475

____________________________ [doctest] module2.foo _____________________________
EXAMPLE LOCATION UNKNOWN, not showing all tests of that example
??? >>> foo(1, 2)
Expected:
    3.0
Got:
    np.float64(3.0)

module2.py:None: DocTestFailure

[bsipocz]

Shall we do the easy thing with the legacy printoption, or you rather suggest to make a new directive that does this flexibly? (I suppose we need to do the printoption either way to fix the CI, but for downstream, it maybe useful to have a directive? Or should they all just update the reprs?

cc @seberg

[pllim]

I have mixed feeling about the new printout. This extra np.float64(...) really messes up the WCS print over at astropy.wcs. I wish there is a way to opt out permanently (not just temporarily).

[drammock]

@pllim IDK how astropy.wcs works, but would it work to redefine the __print__ and/or __repr__ definitions to cast to python numeric types before printing (leaving the underlying object(s) as a numpy dtype)?

[pllim]

Re: WCS -- Maybe but it is just a small portion of things that would be similarly affected. For us downstream that don't see the point of extra stuff in printouts, this seems like busywork that we don't have time for. I think one of the WCS printout code is at https://github.com/astropy/astropy/blob/8ab8208a4df5c636054f07c763006324d245ddbd/astropy/wcs/wcs.py#L3116

[pllim]

But that said, I'll check out what Marten has worked out at astropy/astropy#15065 though so far he didn't touch WCS.

[seberg]

There is a !r there, if you know you got a numerical value, using !s should be fine? I don't care about making it more permanent, although i would care a bit about making printoption saner (e.g. context safe).

[pllim]

I thought we ignored it but it came back last night (https://github.com/scientific-python/pytest-doctestplus/actions/runs/5915248667/job/16041319959). So I took the liberty to relax the numpy-dev version check directly on main at 91104d8 since it is Sunday morning and I don't have much time. FYI.

[bsipocz]

@pllim - I don't think it's necessary to push hot fixes like this, as 1) the failing cron only an effect in our internal testing 2) we still don't provide a solution for downstream.

Besides, leaving an explicit dev version would be helpful while one is in the middle of the dev cycle.

[bsipocz]

Do we still need this? I mean the numpy <2 support window is fast approaching, and it doesn't look like there is a huge interest in implementing the dual approach.

[pllim]

Did we ever revert #211 ? I think some action is still needed if not?

[pllim]

Still here:

pytest-doctestplus/tests/conftest.py

Lines 9 to 12 in c1f124c

	# Keep this until we require numpy to be >=2.0 or there is a directive in doctestplus
	# to support multiple ways of repr
	if Version(np.__version__) >= Version("2.0.dev"):
	np.set_printoptions(legacy="1.25")

[bsipocz]

Ahh, yes, sure we will need changes to support >2+, my understanding was that this issue was about supporting both types of representations.