Submesh: support tuple subdomain_id

[pbrubeck]

Enhances the valid values for subdomain_id for Submesh, so it matches closely the behaivor of the DirichletBC's subdomain kwarg. This spares users from relabeling the mesh themselves.

• Submesh(mesh, ..., subdomain_id=[1, 2]) takes the union of subdomains 1 and 2 (this works for codim-0 and codim-1)
• Submesh(mesh, ..., subdomain_id=[(1, 2)]) takes the intersection of subdomains 1 and 2 (this works for codim-0 and codim-1)

Fixes #5001

[connorjward]

In #5001 (comment) I've figured out a much nicer way to do this. I'm not sure if we should be adding this very unintuitive API if a better solution exists with only a tiny amount more work.

[pbrubeck]

In #5001 (comment) I've figured out a much nicer way to do this. I'm not sure if we should be adding this very unintuitive API if a better solution exists with only a tiny amount more work.

Even after parsing, we need to implement a mechanism to traverse the expression tree. I don't agree that the amount of work is tiny.

I suggest we keep this PR to support the same format used by DirichletBC. If you want to you can open a PR where you implement the logic to process the parsed tree for both DirichletBC and Submesh, in a backwards compatible way.

[connorjward]

Even after parsing, we need to implement a mechanism to traverse the expression tree. I don't agree that the amount of work is tiny.

I suggest we keep this PR to support the same format used by DirichletBC. If you want to you can open a PR where you implement the logic to process the parsed tree for both DirichletBC and Submesh, in a backwards compatible way.

I disagree that we should be adding a known-to-be-confusing interface to a new a feature just because it's what's already been done. The logic really is very simple. You need:

def myrecursion(expr):
  if isinstance(expr, str | int):
    return dm.getLabel(label_name, expr)
  
  lexpr, op, rexpr = expr
  liset = myrecursion(lexpr)
  riset = myrecursion(rexpr)
  if op == "&":
    return ISIntersect(liset, riset)
  else:
    assert op == "|"
    return ISUnion(liset, riset)

[pbrubeck]

I disagree that we should be adding a known-to-be-confusing interface to a new a feature just because it's what's already been done.

Intuitively, many users would try to pass a list of subdomains, as they already do for DirichletBC. I think we should support this

[connorjward]

I disagree that we should be adding a known-to-be-confusing interface to a new a feature just because it's what's already been done.

Intuitively, many users would try to pass a list of subdomains, as they already do for DirichletBC. I think we should support this

I am very happy with the API for a union. It's the intersection behaviour that I don't like.

[connorjward]

I am not the overlord however and this is subjective. If you can get agreement from other members of the team I will happily concede.