Surfaces can compute more than they should

[unalmis]

We have things in the list of variables that we say we can compute on certain objects, but shouldn't be able. One could be misled that a quantity not typically associated with their surface was computed somehow by integrating on the boundary, or just by sampling the given surface.

If a quantity cannot be computed on an object, we should just not have a compute function for that object instead of setting dummy zero values.

Stetting dummy values causes bugs. For example n_rho defined as $\nabla \rho / \Vert \nabla \rho \Vert$ should not be defined on a ZernikeRZToroidalSection. But we say it is, and give it a dummy value of 0. Now, the area of that cross-section is well-defined, however we compute that area using n_rho, and since that dependency is a dummy for this surface, it shouldn't give the correct value.

[unalmis]

Remove @pytest.mark.xfail(reason="GitHub issue 1127.") from tests/test_configuration.py::TestGetSurfaces::test_get_zeta_surface once resolved.

[dpanici]

Fix for this is to remove the Surface level implementations of the compute quantities in question, and implement them only for the subclasses that actually make sense

[unalmis]

Note discussed in dev meeting: if redesigning to propagate nan for quantities that can't be computed, then make sure surface_integral class doesn't mask out all nan quantities. Just mask out nan on the axis.

[dpanici]

Would resolution of this issue make sense to happen on #1094 ? @unalmis

[unalmis]

Either there or separate is fine

[dpanici]

I think I will do it separate for now. That branch still has some things that have to be fixed it seems and it requires this to be done anyways

[dpanici]

I have a PR I think addressing this now. Basically,

• For "desc.geometry.surface.FourierRZToroidalSurface" , the radial derivative is undefined (is a constant rho surface so has no notion of how things change off-surface), so any $\partial_{\rho}$ quantity is now undefined
  • also has its own implementation of n_rho which has no axis limit defined (avoiding errors from the data index when it tried to build n_rho dependencies for the surface, and saw that the surface has no radial deriv defined which are needed for the axis limit of n_rho

"desc.geometry.surface.ZernikeRZToroidalSection" , the toroidal derivative is undefined (is a constant zeta surface so has no notion of how things change off-surface), so any $\partial_{\zeta}$ quantity is now undefined

• I also added a differente perimeter(z) implementation that is actually at constant zeta, not constant phi, as it makes no sense to ask for the perimeter at constant phi of a ZernikeRZToroidalSection that maybe has nonzero omega, as it is defined only at constant zeta

"desc.geometry.core.Surface" parametrization is only used for quantities well-defined by both of the above subclasses (basically only things that need no derivs or theta derivs only)