Remove operator from reactor

[jakehader]

There are only a handful of uses of r.o. it would be great to get rid of these, as:

• it is unclear why a reactor should have an operator
• reactor construction is muddled by being incomplete until the reactor has an operator
• many instances are only used for getting a cs
• removing the r.o would reduce the number of circular references r.o -> o.r -> r.o ...
• in almost all circumstances, a reactor and an operator are already in scope.

[john-science]

Yeah, it's never made sense to me that the Reactor object should have an Operator on it.

It really seems like accidental design, not purposeful design.

[john-science]

Here are all of the usages of this I could find in ARMI (though, of course, any number of plugins might be using this "feature"):

• [ ] high-level

  armi/interfaces.py:  self.o = r.o if r else None
  armi/mpiActions.py:  self.r.o = self.o

• [ ] operators

  armi/operators/operator.py:  r.o = self
  armi/operators/operator.py:  self.r.o = self
  armi/operators/operator.py:  self.r.o = None

• [ ] neutronics

  armi/physics/neutronics/globalFlux/globalFluxInterface.py:  daysIntoCycle = sum(self.r.o...
  armi/physics/neutronics/macroXSGenerationInterface.py:  sum([self.r.o.burnSteps[i] ...

• [ ] converters

  armi/reactor/converters/geometryConverters.py:  self._o = self._sourceReactor.o
  armi/reactor/converters/parameterSweeps/generalParameterSweepConverters.py:  fh = r.o.getInterface("fuelHandler")

• [ ] uniform mesh (recently added back in by PR #961

  newReactor.o = (sourceReactor.o)

• [X] uniform mesh

  armi/reactor/converters/uniformMesh.py:  newReactor = Reactor(sourceReactor.o.cs.caseTitle, bp)
  armi/reactor/converters/uniformMesh.py:  coreDesign.construct(sourceReactor.o.cs, bp, newReactor, loadAssems=False)
  armi/reactor/converters/uniformMesh.py:  _geometryConverter = newReactor.core.growToFullCore(sourceReactor.o.cs)

• [x] report plotting

  armi/utils/reportPlotting.py:  dbi = reactor.o.getInterface("database")
  armi/utils/reportPlotting.py:  dd = reactor.o.getInterface("ductDistortion")
  armi/utils/reportPlotting.py:  history = reactor.o.getInterface("history")

So, a really cursory read of the situation makes it look like the things that would need to change are:

• The UniformMesh should keep a copy of the CaseSettings, and not the operator.
• Clean up of reportPlotting.py
• paramSweepConverters needs to get the FuelHandlerInterface somehow differently.
• The neutronics interfaces above need to get steps/stepLength information from a different source.
• There may be some unwinding of Interfaces and MPI stuff; but it might that is only complicated BECAUSE of this problem.

[jakehader]

I think this should be open because it was auto closed when the other PR was merged. Thanks for the effort on this @john-science! I hope there aren't too many downstream impacts

[john-science]

@jakehader So far, there are no downstream impacts. I only did the removed the really low-hanging fruit versions of r.o (that I could fit into my day).

What's left is some higher-level stuff that will have possible downstream impacts. It appears that sometimes people use r.o.getInterface("xxx"), and squashing all usages of those is the trick. There aren't TOO many places people do this, but it's certainly some fast-and-loose code that will take some untangling to improve the code flow.

TLDR; this is just a start. Thanks!