LatitudeLongitudeGrid may have incorrect metrics for y-Periodic domains

[glwagner]

This code:

using Oceananigans
grid = LatitudeLongitudeGrid(longitude = (215, 215.25),
                             latitude = (50, 50.25),
                             size = (1, 1, 50),
                             halo = (1, 1, 1),
                             z = (-200, 0),
                             topology = (Periodic, Periodic, Bounded))

produces a BoundsError, if run with julia --check-bounds=yes.

The problem is the worksize specified here:

Oceananigans.jl/src/Grids/latitude_longitude_grid.jl

Line 443 in 3b02321

@inline metric_worksize(grid::XRegularLLG) = length(grid.φᵃᶜᵃ) - 1

I also don't like this code because we cannot read it to understand it. For example, what is length(grid.φᵃᶜᵃ)? Is this Ny + Hy, or Ny + Hy + 1? I think it would be better to be explicit.

It appears that for both Periodic and Bounded in y, then we have

length(grid.φᵃᶜᵃ) = 3

Which looks like Ny + 2Hy.

I think both the worksize and the metric size is wrong in general --- I think we have hardcoded a Bounded y-topology. I think the right thing to do is to go in and use total_length to determine both the correct size of the metrics for any topology, as well as the size of the kernel that is launched to compute them.

[glwagner]

A quick fix is to use

 @inline metric_worksize(grid::XRegularLLG)  = length(grid.φᵃᶠᵃ) - 2

[navidcy]

Nice catch!

[navidcy]

What's the metric_worksize?

[glwagner]

I guess it's the size of the kernel that we use to compute the metrics. Basically it has to be the number of faces - 2, beacause we compute metrics using differences.

[glwagner]

Closed by #3450

[navidcy]

Closed by #3450

since #3450 includes other things, perhaps you can point out the source code change or the commit the resolved the issue?