Jørgen Schartum Dokken

> We shouldn't turn off re-ordering in serial. > > Can someone bisect to find out when the regression was introduced? @chrisrichardson could it be related to recent changes for...

It is present in 0.6 as well: P=3 ```bash Meshing and creating mixed space: 0.5712962300000015 Collapsing V: 0.7795458059999874 Collapsing Q: 0.0005138619999911498 ``` P=4 ```bash Meshing and creating mixed space: 0.6977521319999767...

> @jorgensd Re: the speedup in parallel - you should do a weak scaling (e.g. double the mesh size with `-np 2`). I think you will then find it is...

> We probably build a sparsity graph for the dofs, which isn't a good use of compute effort for high-order elements. We should probably create a graph of vertices, facets,...

at first glance, this looks like an issue with using different compilers for dolfinx and dolfimx_mpc, on the nanobind/python interface level. I can’t download the logs on my phone, but...

> The debian package dependency mechanism expects incompatibility only with the major version, not the minor version. Perhaps that needs to be tightened up, if the problem really is nanobind...

@drew-parsons is there anything I can do to guide you further?

> ```python > cell = triangle > elem0 = FiniteElement("Lagrange", cell, ...) > mesh0 = Mesh(FiniteElement("Lagrange", cell, ...)) > V0 = FunctionSpace(mesh0, elem0) > > elem1 = FiniteElement("Brezzi-Douglas-Marini", cell, ...)...

> For instance, we can not define `Coefficient` on a `MixedFunctionSpace`; we need to use the (special-cased) `Coefficients` function, which returns a `list` of `Coefficient`s. We thus can not use...

> @jorgensd I do not think merely allowing for `Coefficient(MixedFunctionSpace)` will solve all issues; as long as `MixedFunctionSpace` does not subclass `FunctionSpace` and does not have `ufl_domain`, I think we...