Open tommbendall opened 4 years ago
Thanks, this is an issue in how we stash "shared" data on function spaces. The introduction of variants is not safe since the variant is not taken into account in the cache key. Perhaps we should cache on the finat element instead in this case...
Unassigning myself, but just noting that this is still an issue (although geometric boundary nodes have been removed).
If two instances of
FunctionSpace
are made withFiniteElement
s that are similar but have a differentvariant
, I think the twoFunctionSpace
s aren't completely distinguished. For instance, they will have the same set of boundary nodes (even if these should be different).Example:
produces the output
but we expect these to differ as the equispaced elements should have basis functions which are zero on the boundaries. If the order of building function spaces is swapped, i.e.
then the output is