Should use of SortBy enums or properties work gracefully if they don't exist?

[jdegenstein]

a = Cylinder(1,1)
a.edges().filter_by(GeomType.CIRCLE).sort_by(Edge.radius) # works
a.edges().filter_by(GeomType.CIRCLE).sort_by(SortBy.RADIUS) # works
a.edges().sort_by(Edge.radius) # fails "ValueError: Shape could not be reduced to a circle"
a.edges().sort_by(SortBy.RADIUS) # fails "ValueError: Shape could not be reduced to a circle"

Should the latter two cases return the same as the former two cases? (i.e. an implicit filter)

[jwagenet]

To expand my initial thoughts:

part.edges().filter_by(GeomType.CIRCLE).group_by(Edge.radius) requires the GeomType filter to avoid a ValueError kickback from OCP for an incompatible GeomType in a mixed-type list of edges. ShapeList methods like group_by(Edge.radius) or filter_by(lambda e: e.radius == 8) could handle the errors so the GeomType filter isn't necessary or the properties themselves dont throw the error.

If handling within the properties, returning None seems safest.

If handling in operators, filter can could treat error cases as the False case.
For sort/group its a little more complicated as returning a 'zero' value eg Line().radius == 0 really right in this case. Returning None wouldn't play well with sort, unless error/None values are implicitly filtered. None can work as a key in GroupBy, but it would need to be excluded from list slicing.

[gumyr]

I'm not sure that making sort_by into an implicit filter_by is very intuitive (and it probably breaks some of the Python guidelines). I don't think any sort should have the property of changing the length of the list.