Usage of ndims(mesh) at many avoidable places

[DanielDoehring]

We often dispatch functions based on the spatial dimensionality of the mesh, see for instance

Trixi.jl/src/callbacks_stage/positivity_zhang_shu_dg1d.jl

Lines 8 to 9 in 288d41d

	function limiter_zhang_shu!(u, threshold::Real, variable,
	mesh::AbstractMesh{1}, equations, dg::DGSEM, cache)

Nevetheless, in these functions the dimensionality of the mesh is still often queried:

Trixi.jl/src/callbacks_stage/positivity_zhang_shu_dg1d.jl

Line 30 in 288d41d

u_mean = u_mean / 2^ndims(mesh)

While the overhead for this is hopefully negligible due to optimizations, this is (in my opinion) bad style, as this code suggests some variability in the method which is not there.

[ranocha]

It should be compiled away since it just returns a compile-time constant. In the specific case you mentioned, it makes it much clearer how the magic numbers appear there. For example, it would be hard for me to see how to generalize it if I just saw / 4 in the 2D case.

Having said that, it would be great if you could make some concrete suggestions on how to clarify and improve the code.

[DanielDoehring]

I would suggest to put the dimensionality right there and leave the usage of this function for the places where the mesh is actually generic.

If this leads to confusion, one could think about adding comments (in the example above, there is for instance already a comment explaining the 2.)

[DanielDoehring]

See discussion in #2081