Mixed topology assemble scalar and gmsh api + fix GMSH tests on AlmaLinux

[jorgensd]

Attempt to resolve: #4213.
Seems like running pytests in xdist mode requires all gmsh models to have unique names. This has now been added.

Adds support for:

• dolfinx.fem.assemble_scalar on mixed topology meshes.

• Second order prism and pyramids (the ones currently supported are not the ones we use for Lagrange elements, as they are missing the middle node of the quadrilateral faces, similar to serendipity).

• Split tests in two:

  • simplex meshes tested for order 1,2,3
  • mixed meshes tested for order 1 and 2

• Supports reading in meshes with model_to_mesh for mixed meshes. No meshtags are read in as we do not have a way of constructing these yet.

• Mixed meshes with order > 2 is not supported, as create_entity_permutations doesn't work on mixed meshes.

Gemini Pro was used to create the 3D mixed mesh with two cylinders on top of each other. All other coding was done by me with no LLM.

The test now uses all 4 cell types for the 3D test (tetra green, pyramid orange, hex red, wedge blue):

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated:
https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated: https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

Resolved by giving each gmsh model a name that is unique for parametrized tests.

[jhale]

Thanks for fixing this and also finishing a cool new feature!

My view is that gmsh.initialize() and gmsh.finalize() are not properly cleaning up global state between pytest function calls on the same process and so we should probably add this to all gmsh-related tests.

[jorgensd]

Attempt to resolve: #4213.
Seems like running pytests in xdist mode requires all gmsh models to have unique names. This has now been added.

Adds support for:

• dolfinx.fem.assemble_scalar on mixed topology meshes.

• Second order prism and pyramids (the ones currently supported are not the ones we use for Lagrange elements, as they are missing the middle node of the quadrilateral faces, similar to serendipity).

• Split tests in two:

  • simplex meshes tested for order 1,2,3
  • mixed meshes tested for order 1 and 2

• Supports reading in meshes with model_to_mesh for mixed meshes. No meshtags are read in as we do not have a way of constructing these yet.

• Mixed meshes with order > 2 is not supported, as create_entity_permutations doesn't work on mixed meshes.

Gemini Pro was used to create the 3D mixed mesh with two cylinders on top of each other. All other coding was done by me with no LLM.

The test now uses all 4 cell types for the 3D test (tetra green, pyramid orange, hex red, wedge blue):

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated:
https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated: https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

Resolved by giving each gmsh model a name that is unique for parametrized tests.

[jhale]

Thanks for fixing this and also finishing a cool new feature!

My view is that gmsh.initialize() and gmsh.finalize() are not properly cleaning up global state between pytest function calls on the same process and so we should probably add this to all gmsh-related tests.

[jorgensd]

Attempt to resolve: #4213.
Seems like running pytests in xdist mode requires all gmsh models to have unique names. This has now been added.

Adds support for:

• dolfinx.fem.assemble_scalar on mixed topology meshes.

• Second order prism and pyramids (the ones currently supported are not the ones we use for Lagrange elements, as they are missing the middle node of the quadrilateral faces, similar to serendipity).

• Split tests in two:

  • simplex meshes tested for order 1,2,3
  • mixed meshes tested for order 1 and 2

• Supports reading in meshes with model_to_mesh for mixed meshes. No meshtags are read in as we do not have a way of constructing these yet.

• Mixed meshes with order > 2 is not supported, as create_entity_permutations doesn't work on mixed meshes.

Gemini Pro was used to create the 3D mixed mesh with two cylinders on top of each other. All other coding was done by me with no LLM.

The test now uses all 4 cell types for the 3D test (tetra green, pyramid orange, hex red, wedge blue):

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated:
https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated: https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

Resolved by giving each gmsh model a name that is unique for parametrized tests.

[jhale]

Thanks for fixing this and also finishing a cool new feature!

My view is that gmsh.initialize() and gmsh.finalize() are not properly cleaning up global state between pytest function calls on the same process and so we should probably add this to all gmsh-related tests.

[jorgensd]

Attempt to resolve: #4213.
Seems like running pytests in xdist mode requires all gmsh models to have unique names. This has now been added.

Adds support for:

• dolfinx.fem.assemble_scalar on mixed topology meshes.

• Second order prism and pyramids (the ones currently supported are not the ones we use for Lagrange elements, as they are missing the middle node of the quadrilateral faces, similar to serendipity).

• Split tests in two:

  • simplex meshes tested for order 1,2,3
  • mixed meshes tested for order 1 and 2

• Supports reading in meshes with model_to_mesh for mixed meshes. No meshtags are read in as we do not have a way of constructing these yet.

• Mixed meshes with order > 2 is not supported, as create_entity_permutations doesn't work on mixed meshes.

Gemini Pro was used to create the 3D mixed mesh with two cylinders on top of each other. All other coding was done by me with no LLM.

The test now uses all 4 cell types for the 3D test (tetra green, pyramid orange, hex red, wedge blue):

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated:
https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated: https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

Resolved by giving each gmsh model a name that is unique for parametrized tests.

[jhale]

Thanks for fixing this and also finishing a cool new feature!

My view is that gmsh.initialize() and gmsh.finalize() are not properly cleaning up global state between pytest function calls on the same process and so we should probably add this to all gmsh-related tests.

[jorgensd]

Attempt to resolve: #4213.
Seems like running pytests in xdist mode requires all gmsh models to have unique names. This has now been added.

Adds support for:

• dolfinx.fem.assemble_scalar on mixed topology meshes.

• Second order prism and pyramids (the ones currently supported are not the ones we use for Lagrange elements, as they are missing the middle node of the quadrilateral faces, similar to serendipity).

• Split tests in two:

  • simplex meshes tested for order 1,2,3
  • mixed meshes tested for order 1 and 2

• Supports reading in meshes with model_to_mesh for mixed meshes. No meshtags are read in as we do not have a way of constructing these yet.

• Mixed meshes with order > 2 is not supported, as create_entity_permutations doesn't work on mixed meshes.

Gemini Pro was used to create the 3D mixed mesh with two cylinders on top of each other. All other coding was done by me with no LLM.

The test now uses all 4 cell types for the 3D test (tetra green, pyramid orange, hex red, wedge blue):

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated:
https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated: https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

Resolved by giving each gmsh model a name that is unique for parametrized tests.

[jhale]

Thanks for fixing this and also finishing a cool new feature!

My view is that gmsh.initialize() and gmsh.finalize() are not properly cleaning up global state between pytest function calls on the same process and so we should probably add this to all gmsh-related tests.

[jorgensd]

Attempt to resolve: #4213.
Seems like running pytests in xdist mode requires all gmsh models to have unique names. This has now been added.

Adds support for:

• dolfinx.fem.assemble_scalar on mixed topology meshes.

• Second order prism and pyramids (the ones currently supported are not the ones we use for Lagrange elements, as they are missing the middle node of the quadrilateral faces, similar to serendipity).

• Split tests in two:

  • simplex meshes tested for order 1,2,3
  • mixed meshes tested for order 1 and 2

• Supports reading in meshes with model_to_mesh for mixed meshes. No meshtags are read in as we do not have a way of constructing these yet.

• Mixed meshes with order > 2 is not supported, as create_entity_permutations doesn't work on mixed meshes.

Gemini Pro was used to create the 3D mixed mesh with two cylinders on top of each other. All other coding was done by me with no LLM.

The test now uses all 4 cell types for the 3D test (tetra green, pyramid orange, hex red, wedge blue):

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated:
https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated: https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

Resolved by giving each gmsh model a name that is unique for parametrized tests.

[jhale]

Thanks for fixing this and also finishing a cool new feature!

My view is that gmsh.initialize() and gmsh.finalize() are not properly cleaning up global state between pytest function calls on the same process and so we should probably add this to all gmsh-related tests.

[jorgensd]

Attempt to resolve: #4213.
Seems like running pytests in xdist mode requires all gmsh models to have unique names. This has now been added.

Adds support for:

• dolfinx.fem.assemble_scalar on mixed topology meshes.

• Second order prism and pyramids (the ones currently supported are not the ones we use for Lagrange elements, as they are missing the middle node of the quadrilateral faces, similar to serendipity).

• Split tests in two:

  • simplex meshes tested for order 1,2,3
  • mixed meshes tested for order 1 and 2

• Supports reading in meshes with model_to_mesh for mixed meshes. No meshtags are read in as we do not have a way of constructing these yet.

• Mixed meshes with order > 2 is not supported, as create_entity_permutations doesn't work on mixed meshes.

Gemini Pro was used to create the 3D mixed mesh with two cylinders on top of each other. All other coding was done by me with no LLM.

The test now uses all 4 cell types for the 3D test (tetra green, pyramid orange, hex red, wedge blue):

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated:
https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated: https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

Resolved by giving each gmsh model a name that is unique for parametrized tests.

[jhale]

Thanks for fixing this and also finishing a cool new feature!

My view is that gmsh.initialize() and gmsh.finalize() are not properly cleaning up global state between pytest function calls on the same process and so we should probably add this to all gmsh-related tests.

[jorgensd]

Attempt to resolve: #4213.
Seems like running pytests in xdist mode requires all gmsh models to have unique names. This has now been added.

Adds support for:

• dolfinx.fem.assemble_scalar on mixed topology meshes.

• Second order prism and pyramids (the ones currently supported are not the ones we use for Lagrange elements, as they are missing the middle node of the quadrilateral faces, similar to serendipity).

• Split tests in two:

  • simplex meshes tested for order 1,2,3
  • mixed meshes tested for order 1 and 2

• Supports reading in meshes with model_to_mesh for mixed meshes. No meshtags are read in as we do not have a way of constructing these yet.

• Mixed meshes with order > 2 is not supported, as create_entity_permutations doesn't work on mixed meshes.

Gemini Pro was used to create the 3D mixed mesh with two cylinders on top of each other. All other coding was done by me with no LLM.

The test now uses all 4 cell types for the 3D test (tetra green, pyramid orange, hex red, wedge blue):

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated:
https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

[jorgensd]

@jhale this still occasionally crashes on alma linux. Now I'm not sure why, as our API should support the meshes generated: https://github.com/FEniCS/dolfinx/actions/runs/25905433471/job/76137968460

Resolved by giving each gmsh model a name that is unique for parametrized tests.

[jhale]

Thanks for fixing this and also finishing a cool new feature!

My view is that gmsh.initialize() and gmsh.finalize() are not properly cleaning up global state between pytest function calls on the same process and so we should probably add this to all gmsh-related tests.