Selector regression in an inheritance context? about param

[jbednar]

When instantiate=True leads to copying the Parameter object, can we make it do a copy (not a deepcopy) of any mutable slot values (or all slot values, which should have the same result)? That seems cheap enough and preserves object identity for the items in the Selector while avoiding this problem.

The only mutable types I can see in slot values are all containers (_objects and names here, plus attribs for Composite, columns potentially on DataFrame, and search_paths on Path), so maybe we could shallow copy only containers, since arbitrary class instances aren't necessarily safe to copy if some Parameter did have a slot with such an item. For current Parameter types I think shallow copying everything, all mutable values, and containers would have the same result, so it may not matter much.

I'm pretty sure that for all those cases, sharing the contents of the container across instantiated Parameter objects would be seen as a bug and not expected behavior. I.e., if the user wanted that Parameter instantiated, they will want independent containers, because if they wanted them all to share the same values they wouldn't have instantiated the parameters.

[maximlt]

When instantiate=True leads to copying the Parameter object, can we make it do a copy (not a deepcopy) of any mutable slot values (or all slot values, which should have the same result)?

Yes there's nothing that prevents us from doing that when instantiate=True. However that doesn't resolve neither this issue nor #746 as they deal with the Selector Parameter which is set by default with instantiate=False, right?