Improve meshing

[duarte-jfs]

The present changes are meant to correct the meshing along lines, and prevent overrefinement, giving users more control on the mesh sizes.

The addition of the

gmsh.model.mesh.field.setNumbers(
                n + 1,
                "CurvesList",
                meshtracker.get_gmsh_xy_lines_from_label(label),
            )

Allows us to see a significant change on the mesh due to changes in the resolution definitions of LineStrings, and the second part follows from the gmsh manual, in chapter 1:

Note that when the element size is fully specified by a mesh size field, it is thus often desirable
to set
Mesh.MeshSizeFromPoints = 0;
Mesh.MeshSizeFromCurvature = 0;
Mesh.MeshSizeExtendFromBoundary = 0;
to prevent over-refinement inside an entity due to small mesh sizes on its boundary.

This will allow us to go from this kind of meshing when distance=0:

To this:

Which can be desirable.

Disclaimer: I'm not sure how much of a mess the commit history of my fork is. If it's too much trouble I'd be happy to close this PR and you guys can just make the necessary changes, if you see this would benefit the package as a whole.

[HelgeGehring]

Looks great! I looked over the last days also in this kind of refinement :)

We wanna migrate longterm to using meshwell for the meshing instead of the meshing in femwell. But as it'll take quite a bit of work to switch, it makes sense to keep maintaining the meshing within femwell until then :)

@simbilod could you have a look? To me it seems as if we can just merge it (whereas I don't know if the last three lines might be redundant (?))

[duarte-jfs]

I agree. I tried to use meshwell, but found it easier to work with the mesh_from_OrderedDict for now as I was more used to it. The transition should be gradual to allow users to adapt

[simbilod]

Looks good!

[HelgeGehring]

maybe we should just think of a implementation of mesh_from_OrderedDict which uses meshwell? 🤔 then noone would notice the transition? :D

[HelgeGehring]

@simbilod didn't you put those options already somewhere? Or is that only in meshwell?

[HelgeGehring]

@duarte-jfs is there a reason why you add the last three lines again?

it's already done in (and the two following lines)

femwell/femwell/mesh/mesh.py

Line 409 in 9ce5de4

gmsh.model.mesh.MeshSizeFromPoints = 0

[duarte-jfs]

Hi, sorry for getting back to you so late. I just looked at this again. To be honest, I don't understand it either. But the behaviour is different if you incliude the lines as I did. No clue why

[duarte-jfs]

But it is also seems that gmsh.model.mesh.MeshSizeFromPoints = 0 is not equivalent to gmsh.option.setNumber("Mesh.MeshSizeFromPoints", 0) (and likewise for the other two lines). I tried to use them interchangeably and the behaviour is different

[HelgeGehring]

Ok, then I'm totally fine with just merging it :)

[simbilod]

Yes this will also add the mesh constraint on the curves and not just the surface (interior)

Meshwell has much more advanced resolution settings now, we should finish the migration #52

[simbilod]

Yes this will also add the mesh constraint on the curves and not just the surface (interior)

Meshwell has much more advanced resolution settings now, we should finish the migration #52