Pytest.mark values in test_oneAxisSolarArrayPoint.py lead to singularities

[bbercoviciUspace]

Hi all,

I think there's a need for a quick fix in test_oneAxisSolarArrayPoint.py. The pytest decorators are preceded by the following statement

# The 'ang' array spans the interval from 0 to pi. 0 and pi are excluded because 
# the code is less accurate around those points; it still provides accurate results at 1e-6
ang = np.linspace(0, np.pi, 5, endpoint=True)
ang = list(ang)

@pytest.mark.parametrize("alpha", ang)
@pytest.mark.parametrize("delta", ang)

The comment contradicts the use of endpoint=True. The latter results in several unit tests failing, especially when ang is equal to pi (the error is then around 1e-8 which is greater than the specified 1e-9).

I think the module's logic is sound (as far as I am concerned, I just want this module's unit test to pass and am not planning on using it for now).

@rcalaon , what are your thoughts about this ? Thanks =)

[rcalaon]

Hi! We are aware of this very odd bug. It only shows up occasionally, not consistently. For some reason, the numerical precision fails when the delta input to the test function is PI. This is very odd because the test only fails on the CI server, whereas when I test the module on physical Windows or Mac machines, it passes up to a precision of 1e-15.

While we investigate this bug further, I'd suggest to just re-run the jobs that failed. It will eventually pass. We are working on a permanent fix!

[bbercoviciUspace]

Thanks for the quick response ! I simply switched endpoint to False which will do the trick just fine for now =)

[schaubh]

Riccardo has been relaxing the test constraints as they were overly aggressive. That should correct for this issue.

[rcalaon]

Unfortunately, this is not correct. The error keeps showing up with a repeatable accuracy of 2e-08, which keeps causing the test to fail. While we could just loosen the accuracy to 1e-07 and make the error disappear, this bug is very odd and I was exchanging thoughts with Pat to see if we can isolate the issue and fix it.