Incorrect behaviour when moving parent mesh of `VertexOnlyMesh`

[leo-collins]

Suppose we have a VertexOnlyMesh immersed in a parent mesh. Changing mesh.coordinates doesn't reimmerse the VOM as expected. For example,

from firedrake import *

points = [[0.5, 0.5]]
mesh = UnitSquareMesh(10, 10)
V = FunctionSpace(mesh, "CG", 1)
x, y = SpatialCoordinate(mesh)
f = Function(V).interpolate(x + y)

vom = VertexOnlyMesh(mesh, points)
P0DG = FunctionSpace(vom, "DG", 0)
f_at_points = assemble(interpolate(f, P0DG))

assert np.allclose(f_at_points.dat.data, [1.0])

mesh.coordinates.dat.data_wo[:, 0] += 1.0
f = Function(V).interpolate(x + y)

f_at_points_after_move = assemble(interpolate(f, P0DG))

assert np.allclose(f_at_points_after_move.dat.data, [2.0])

coord = vom.topology_dm.getField("DMSwarmPIC_coor")
assert np.allclose(coord[0], [0.5, 0.5])
vom.topology_dm.restoreField("DMSwarmPIC_coor")

f_at_points_after_move contains f evaluated at the point [0.5,0.5] which isn't in the mesh. I think this is because the value at the point is calculated from the global number of the cell and the local reference coordinates of the point, neither of which are affected by changing the mesh coordinates. But clearly this calculation makes no sense.

Moving the mesh is equivalent to moving the points in the opposite way, so a fix for this would require implementing moving particles.

[connorjward]

Instead of doing the full-on fix of having moving particles is there something that could be done to invalidate the VOM? Maybe it could store the mesh coordinates dat version?

[leo-collins]

Yeah I think a check inside interpolate somewhere that throws an error if the coordinates have changed is probably sufficient for now.