Face coordinates

[lijas]

I working on an example with contact, where it is nice to have some functionality for getting the coordinates of a face.

Up until this point, there has not really been a need for this in Ferrite since we have used FaceValues which uses all the coords/shape functions of the cell.

I am redefining the function getcoordinates(grid::AbstractGrid, face::FaceIndex) , but a doubt anyone has been using this.

for faceindex in contactfaces
    getcoordinates!(facecoords, grid, faceindex)
    ...
end

Note: draft PR, so I will add tests etc :)

[KnutAM]

Nice!
For ref: #664 deals with the names of these functions.
getcoordinates!(cellcoords, grid, faceindex) can be used for looping over faces and reiniting facevalues. (But this is quite unexpected in my opinion that it returns the cell coordinates, hence the renaming suggestions in the linked pr)

[fredrikekre]

Don't you want faces(cell) here instead?

[lijas]

faces only returns the vertices/corner-nodes of the cell (even for QuadraticHexahedrons etc). So we would need to update faces(cells) in that case...

[termi-official]

I think we should not touch these definitions (i.e. faces, because they already give the expected description of the corresponding topological entity and this is what we actually need in most cases.

[termi-official]

I should also note that AFAIK there is no hard guarantee that the returned tumple of facedof_indices must match the exact ordering of the face nodes.

[KnutAM]

Would be nice to have a good interface for getting the dofs on a face, this is also used in this function:

Ferrite.jl/src/Dofs/ConstraintHandler.jl

Line 981 in c6c185c

function _add!(ch::ConstraintHandler, pdbc::PeriodicDirichlet, interpolation::Interpolation,

(when I needed it once I would have wanted such an interface, e.g. get_face_dofs[!]([dofinds], dh, faceindex))

[lijas]

@termi-official I dont know what makes most sense: faces(QuadraticHexhedron) returning only the 4 corner nodes or all 9 nodes on the face.

But I think the cell-ordering has to follow facedof_indices right?, otherwise the dof-distrubtion would be wrong I think.

[termi-official]

Note that the purpose of facedof_indices is different from facedof_interior_indices. We only need the latter during serial dof distribution. The latter is just a query of "give me every dof in the face", which is a cherry pick of what I need in the distributed assembly PR. I think it makes sense to somehow enforce that the order of the dofs is aligned with the subentity order on the corresponding cell, but I need to check exactly which invariants I will require for the dof synchronization in the new API. Does this clear up the confusion?

[termi-official]

Not sure if I understand the first one correctly. faces(QuadraticHexhedron) should return the 6 faces with each face described by its corners. This is everything that we need to derive the relevant topological information as well as orientation information. The higher order nodes give you "just" the extra geometric information about how the element deforms on this specific face. But yea, we should implement some additional API to query higher order information for the geometry, even if it is just by enforcing a strict correspondence between the element ad the default interpolation (which I think Kim did vote against previously).

[AbdAlazezAhmed]

Just a reminder (not sure if it's needed here), Ferrite.facedof_indices is not defined for discontinuous interpolations. In case it's needed, Ferrite.dirichlet_facedof_indices was added in #729 . It might sound out of place here tho.