> Do we need a new test for conversion of string defaults?

I think we should also update the documentation of the "default" keyword argument (and/or the "type" argument) to remove any ambiguity and make the behavior more clear.

Maybe one way to view what is happening (and maybe one way to describe it in the documentation) is that if a default value is a string, the parser treats it as a default for the command-line *argument*, otherwise it treats it as a default for the final *attribute*.

Lastly, can we also add the test from one of my first comments above to check that "double conversion" of strings doesn't take place?