Add typing to RangeSet to avoid typing errors when iterating

[carl-aa]

Summary

Currently you are able to iterate over RangeSet instances in a ConcreteModel (e.g. for x in y where y is a RangeSet instance), but I get typing warnings saying Expected type `collections.iterable`, got `RangeSet` instead. It would be helpful to expose type hints for RangeSet so that these type warnings could be avoided.

Rationale

It would make development in type aware environments like PyCharm more ergonomic.

Description

I'm not exactly sure what's missing to make the iteration error go away. I would need to research typing issues more.

Additional information

The Set instances work just fine, so it may be easy to compare the two classes to see what change is needed.

[jsiirola]

I think I know what is going on, but I am not convinced that there is an easy fix (nor that the fix would be "correct").

The challenge is that not all RangeSet instances are iterable.

RangeSet() is a factory that actually creates either a FiniteScalarRangeSet (e.g., from RangeSet(5)) or an InfiniteScalarRangeSet (e.g., from RangeSet(1, 5, 0), or the Reals set -- which is a special instance of RangeSet(ranges=(NumericRange(None, None, 0),))).

Non-finite sets are (obviously) not iterable.

Set works out because of a bit of a fluke: Set inherits from IndexedComponent (RangeSet does not support indexing, and therefore only inherits from Component). The type hints don't complain about Set because they see the IndexedComponent.__iter__ method (which is defined for iterating over the index that defines the IndexedComponent -- borrowing from the MutableMapping model). This isn't a huge deal, as all "real" Set objects (e.g., OrderedScalarSet, SortedScalarSet, FiniteScalarSet, etc) are all finite sets (and will support iteration) and override the __iter__ method.

[carl-aa]

Ahh, I did not know about the infinite case. In my particular case, I simply switched over to using Set, since it didn't seem to matter for my modeling use case.

[jsiirola]

Non-finite range sets are a pretty niche topic (they are mostly used for defining domains). A reasonable solution might be to break away from the Factory pattern and force users to explicitly create InfiniteRangeSet objects. That way RangeSet could become an alias for FiniteRangeSet, and type hints could make sense.

...there might be some issues around ctypes, but that should be resolvable. It actually may help motivate something I have wanted to do for a while now (switch RangeSet's ctype over to Set). I am interested in hearing others' thoughts on the issue...