Remove Undefined from stubs

[JamesGuthrie]

Part of #680

[JukkaL]

Cool, thanks for the PR! A lot of "Python evaluation" tests (that use full stubs) are failing -- it could be an problem with builtins.py, typing.py or another stub file that is imported in many test cases. The stack traces are probably caused by a mypy bug -- I'll look at them.

By the way, are you working on other parts of #680? If you are not working on test cases, I'm probably going to start updating them to not use Undefined.

[JamesGuthrie]

Huh, strange. It seems as though the travis.sh script runs more tests than python3 tests.py, so I didn't notice those failures. I will fix the following things:

• It does look as though the tests will have to be updated because I removed Undefined = object() from the typing.py stub. I will put that definition back in typing.py so that the tests pass, we can remove it from the stubs when it's really no longer needed.
• There are also other bugs that I introduced into the stubs which I'm seeing now thanks to the tests which are run in travis.sh. I will fix those and re-do this PR.
• It looks as though ... is not valid python2, so I should remove x = ... # type: int in the 2.7 typing.py stub.

I haven't worked on anything else for #680, so I'm quite happy for you to continue working on it - I would like to get back to #638 at some point :)

[JukkaL]

I removed most uses of Undefined in the rest of the codebase (outside stubs). Once this PR is in, we should be pretty close to getting rid of Undefined for good.

[JamesGuthrie]

These commits close #680 and include the latest version of typing.py from ambv/typehinting.

[JukkaL]

Great, almost there!

I did a quick pass and found a few issues:

In stubs/2.7/difflib.pyi:

... a: Sequence[_T] = b: Sequence[_T], ... is wrong (near SequenceMatcher)

In stubs/3.2/_io.pyi:

_IOBase.seek has a missing =... default arg value:

-    def seek(self, offset, whence=Undefined): pass
+    def seek(self, offset, whence): pass

I found a few other instances where x=Undefined is replaced with just x. Probably worth manually checking all instances of =Undefined in the diff.

Also, the best replacement for an x=Undefined argument is x=... (with a literal ellipsis). This is slightly better than x=None, as sometimes None is not a valid argument value.

[JamesGuthrie]

I fixed the two cases you mentioned above and went through all of the functions which were modified, ensuring that def f(x=Undefined) was replaced with def f(x=...), I managed to catch some other issues so I think it should all be good now.