Fix strict none errors in the test subpackage.

[gvanrossum]

This shows several different approaches to dealing with #2199 for p[i].arg.

@ddfisher Thoughts? Maybe we can also commit a mypy.ini containing

[mypy]
show_none_errors = False
[mypy-*mypy/test/*]
show_none_errors = True

[gvanrossum]

This refactor was needed because program_text is previously inferred as str, and that's correct. So this is technically a case of reusing the same variable name with a different type (#1174). [UPDATE: This comment should be one line up.]

[ddfisher]

Should this really be the empty string, or is this an error condition?

[gvanrossum]

IIUC the '' is unreachable -- when id is 'file' arg is never None. A different rewrite would be

arg = p[i].arg
assert arg is not None

[gvanrossum]

Hm, I rewrote it using the assert, then realized that arg may be None if someone puts [file] in a testcase. That's nonsense. In general the parsing of test cases is pretty crummy, so I don't consider it my job to fix it.

[ddfisher]

Agreed. Having an assertion failure in that case seems fine to me.

[ddfisher]

Not sure we can add that mypy.ini, because suppressing Optional errors suppresses other kinds of errors too...

[gvanrossum]

Adding those flags to mypy.ini should be fine, since they don't have any effect unless you also pass --strict-optional. I guess we should add another mypy run to runtests.py, where it runs mypy --strict-optional mypy and expects no errors.

[ddfisher]

LGTM!