zeta_B computation is sometimes wrong

[unalmis]

Sometimes when I rerun the test that compares everything is computed the same as master up to floating point errors, I get significant errors on zeta_B.

Not equal to tolerance rtol=1e-10, atol=1e-10
Parameterization: desc.magnetic_fields._core.OmnigenousField. Name: zeta_B.
Mismatched elements: 270 / 297 (90.9%)
Max absolute difference: 11.42397329
Max relative difference: 34.0466585
 x: array([ 0.      ,  0.550532,  1.076765,  1.557329,  1.976363,  2.325429,
        2.604526,  2.822094,  2.993993,  0.      ,  0.550532,  1.076765,
        1.557329,  1.976363,  2.325429,  2.604526,  2.822094,  2.993993,...
 y: array([0.      , 0.550532, 1.076765, 1.557329, 1.976363, 2.325429,
       2.604526, 2.822094, 2.993993, 0.      , 0.550532, 1.076765,
       1.557329, 1.976363, 2.325429, 2.604526, 2.822094, 2.993993,...

This error is not always reproducible. However, it has even occurred on new branches, even when the file that the results are compared to was regenerated immediately prior to running the test.

Therefore, either

1. zeta_B is not deterministic or subject to a numerical instability (silly)
2. Pickle is messing something up when the file is dumped
3. Some issue from changes to this test in Move basis change to main compute function #1027 ?

[f0uriest]

Could they be the same modulo 2pi/NFP?

[YigitElma]

Is this test_compute_everthing? #1027 only added xyz computation to the test, it didn't change logic or etc.

[unalmis]

Could they be the same modulo 2pi/NFP?

I checked: no. This is reproducible on my machine. Repeated calls to the test give 40/60 pass,fail split respectively, sample size 8. It is consistent in that when the failure occurs it is always the same result. I'm now leaning towards option 1.

Is this test_compute_everthing? #1027 only added xyz computation to the test, it didn't change logic or etc.

Yes it looked fine when we merged. I think it was around time failure starts occurring.

@register_compute_fun(
    name="zeta_B",
    label="\\zeta_{B}",
    units="rad",
    units_long="radians",
    description="Boozer toroidal angular coordinate",
    dim=1,
    params=[],
    transforms={},
    profiles=[],
    coordinates="rtz",
    data=["zeta", "nu"],
    # default parameterization is equilibrium
)
def _zeta_B(params, transforms, profiles, data, **kwargs):
    data["zeta_B"] = data["zeta"] + data["nu"]
    return data

....
# in data_index.py
_class_inheritance = {
    ...
    "desc.magnetic_fields._core.OmnigenousField": [], # in that case shouldn't this include equilibrium or add parameteriation to above?
}