DiscretizationCollection constructor should do no work

[inducer]

This is too fat. The constructor itself should just be an attribute setter. The actual work should be done by a "maker function", like make_discretization_collection. This is a long-term thing that'll hopefully make it easier to move towards moving geometry.

x-ref inducer/meshmode#123

[thomasgibson]

What's the cleanest way to do this without breaking a whole bunch of stuff? We need to support dcoll = DiscretizationCollection(...)? Should we introduce a function alias? Or should we take the opportunity to use PRs like illinois-ceesd/mirgecom#360 to make sure this get's taken care of? ATM our main worry is mirgecom in terms of compatibility.

[thomasgibson]

Actually, I think we should do this after EagerDiscretization is completely removed from mirgecom. After taking a cursory look at what is entailed, I think this will not be an easy refactoring.

In order to remove all work from DiscretizationCollection.__init__(...), you need to split out the following:

grudge/grudge/discretization.py

Line 191 in fc24878

def _set_up_distributed_communication(self, mpi_communicator, array_context):

grudge/grudge/discretization.py

Line 300 in fc24878

def connection_from_dds(self, from_dd, to_dd):

grudge/grudge/discretization.py

Line 259 in fc24878

def discr_from_dd(self, dd):

and several other related private method (the ones I listed are the really big ones)

[inducer]

My thought is that this won't be cataclysmic. Most of the work could be moved to functions that the constructor calls while complaining, whereas the canonical call sites would be in make_discretization_collection or some such. I don't know why connection_from_dd or discr_from_dd would need to move though.