GLL/GL on simplices with recursive points

[pbrubeck]

No description provided.

[rckirby]

This all seems eminently reasonable. I had a few comments. Major point is whether we want to re-implement recursivenodes or create a dependency. It should be a simple install procedure and would give us more features and let us incrementally get rid of old code like for orthogonal bases! But before we embark on a major undertaking like that, we should probably see where elements, etc, wind up with ongoing UFL work.

[rckirby]

I still think the discussion of using vs recreating recursivenodes is worth having, but it's probably best not to create a new dependency at this point?

[tisaac]

I still think the discussion of using vs recreating recursivenodes is worth having, but it's probably best not to create a new dependency at this point?

You can compare the approaches in other projects.

• The construction is intentionally simple and I had expected reimplementation to be the most common choice. This is what basix does.

• At Andreas Klöckner's prodding, I kept the transitive requirements lightweight -- only numpy is required to generate the nodes -- so he has recursivenodes as a dependency in meshmode. The package is on an MIT license. I have not has reason to test the package with the latest python (its minimum requirement is 3.6), but am happy to address issues if they come up.