[BUG]: locate_dofs_topological segmentation (core dumped)

[mantledparamountic]

Summarize the issue

I don't have time to write a MWE. I use FEniCSx v 0.1 (latest stable at the moment).
I have a mesh coming from Gmsh that I read with the dolfinx api.
I know how to fix the problem, so this isn't urgent, but still, I don't believe FEniCSx should core dump without leaving any trace of the error.

The code:

        dim = mesh.topology.dim
        volt_boundary_dofs = locate_dofs_topological(ME.sub(1), dim-1, bottom_facets)

where bottom_facets is worth facet_markers.find(12). This code is fine.

But if I accidentally put a left_dofs_temp = locate_dofs_topological(ME.sub(0), mesh.topology.dim-1, bottom_facets) instead of the bottom_facets... then the program core dumps.

How to reproduce the bug

No MWE at the moment.

Minimal Example (Python)

Output (Python)

zsh: segmentation fault (core dumped)  /home/user/.conda/envs/fenicsx_0dot1_-env/bin/python

Version

main branch

DOLFINx git commit

No response

Installation

No response

Additional information

No response

[jorgensd]

But if I accidentally put a left_dofs_temp = locate_dofs_topological(ME.sub(0), mesh.topology.dim-1, bottom_facets) instead of the bottom_facets... then the program core dumps.

I don't quite understand the difference between the working and segfaulting solution. What does bottom_facets in the second one come from?

[mantledparamountic]

Sorry for my delay, dokken.
What I mean is that I get the core dumped problem when I do something like:

volt_boundary_dofs = locate_dofs_topological(ME.sub(1), dim-1, locate_dofs_topological(ME.sub(0), mesh.topology.dim-1, bottom_facets))

Of course, it is a "stupid" thing to do (happened to me by inattention), and easy to debug. But still, I don't think a core dump is a good "error".

[jorgensd]

Sorry for my delay, dokken. What I mean is that I get the core dumped problem when I do something like:

volt_boundary_dofs = locate_dofs_topological(ME.sub(1), dim-1, locate_dofs_topological(ME.sub(0), mesh.topology.dim-1, bottom_facets))

Of course, it is a "stupid" thing to do (happened to me by inattention), and easy to debug. But still, I don't think a core dump is a good "error".

We could add a check on the max value in entities being smaller than the num_nodes in e_to_c in:

dolfinx/cpp/dolfinx/fem/DirichletBC.cpp

Line 58 in 0dffbe0

for (std::int32_t e : entities)

@jhale @chrisrichardson @garth-wells what do we think about this?
I agree that segfault is not a great error here that is easy to debug.
We could either do:

std::int32_t max_entity = std::ranges::max(entities);
if (e_to_c.num_nodes() < max_entity)
{
  throw std::runtime_error(
        "Input entity " + std::to_string(max_entity) + " is bigger than the number of entities on this process (" + std::to_string(e_to_c.num_nodes()) + ")".);
}

or

  std::int32_t num_entities_local = e_to_c.num_nodes();
  for (std::int32_t e : entities)
  {
   if (num_entities_local <= e)
   throw std::runtime_error(.....);

    // Get first attached cell
    ...
  }

[garth-wells]

std::int32_t num_entities_local = e_to_c.num_nodes();
for (std::int32_t e : entities)
{
  if (!(0 <= e < num_entities_local))
    throw std::runtime_error(.....);

  // Get first attached cell
  ...
}

would be ok in this case since it won't have a measurable performance impact.