bug in divergence_result_type

[szy21]

Describe the bug

When calling the following function

function boundary_tendency_momentum(ᶜρ, ᶜuₕ, ρ_flux_uₕ)
    FT = eltype(ᶜρ)
    ᶠgradᵥ = Operators.GradientC2F()
    ᶜdivᵥ_uₕ = Operators.DivergenceF2C(
        top = Operators.SetValue(C3(FT(0)) ⊗ C12(FT(0), FT(0))),
        bottom = Operators.SetValue(ρ_flux_uₕ),
    )
    return @.  ᶜdivᵥ_uₕ((0 * ᶠgradᵥ(ᶜuₕ))) / ᶜρ
end

it results in an error

ERROR: LoadError: MethodError: no method matching divergence_result_type(::Type{ClimaCore.Geometry.AxisTensor{Float32, 2, Tuple{…}, StaticArraysCore.SMatrix{…}}})
The function `divergence_result_type` exists, but no method is defined for this combination of argument types.

Changing @. ᶜdivᵥ_uₕ((0 * ᶠgradᵥ(ᶜuₕ))) / ᶜρ to @. lazy(ᶜdivᵥ_uₕ((0 * ᶠgradᵥ(ᶜuₕ))) / ᶜρ) fixes the issue.

To Reproduce

I'll try to make a minimal working example later.

[charleskawczynski]

I think I ran into this in #2184, and fixed it by defining

@inline divergence_result_type(
    ::Type{AxisTensor{FT, 2, Tuple{A1, A2}, S}},
) where {
    FT,
    A1 <: CovariantAxis{(3,)},
    A2 <: CovariantAxis{(1, 2)},
    S <: SMatrix{1, 2, FT, 2},
} = Covariant12Vector{FT}

(xref).

[szy21]

Oh, thank you!

[charleskawczynski]

Actually, I'm a bit confused how I even found this issue, now that I think about it-- I think I grabbed the example from our test suite. Maybe it was a matter of combining operations? Thoughts, @dennisYatunin?

[dennisYatunin]

You might have tried calling an @. expression without its left-hand side, resulting in an invalid call to divergence_result_type for Axis2Tensors. That's how we found it.

[charleskawczynski]

That would make sense. Could you review #2210?