A slightly complicated test setup. I wanted to both make sure I captured
the semantics of Python and also the fact that the kinds of global flow
we expect to see are indeed present.
The code is executable, and prints out both when the execution reaches
certain files, and also what values are assigned to the various
attributes that are referenced throughout the program. These values are
validated in the test as well.
My original version used introspection to avoid referencing attributes
directly (thus enabling better error diagnostics), but unfortunately
that made it so that the model couldn't follow what was going on.
The current setup is a bit clunky (and Python's scoping rules makes it
especially so -- cf. the explicit calls to `globals` and `locals`), but
I think it does the job okay.