Add some tests based on coverage report: Part 2

[ilevkivskyi]

This is a continuation of #373

Here, however, a (very small) change to typing.py was necessary. The two PRs boost coverage to virtually 100% (if one does not count passes in abstract methods and other trivial things).

[JelleZijlstra]

Why doesn't this check impl_type and type_checker?

[ilevkivskyi]

This is an internal class, we know that if name is the same, then impl_type and type_checker must also be the same.

[ilevkivskyi]

OK, It looks like the simplified version depends on changes in #383.
I will leave it like this until #383 is (hopefully) merged.

@gvanrossum Do you mind if we make GenericMeta code a bit more complex for 4x memory and 2x speed optimization? (this came up on python-dev some time ago)

[gvanrossum]

I don't mind in principle, but I still haven't found the energy to review your code. I hope you can find someone else on python-dev to help you review your types changes if Inada doesn't feel comfortable.

[ilevkivskyi]

@gvanrossum
I have merged discussed PRs, but there is one thing that needs to be fixed before merging this into CPython: I added a test module mod.py some time ago (it was required to verify that cross-module type caching works right, a bug was reported). I will fix import for this module, and then it will be necessary to copy it to /Lib/test (where ann_module.py etc live). I will let you know, when everything is ready (probably later today).

[gvanrossum]

OK. Since the move to git is currently happening I can't merge anything into CPython anyway right now, so there's no particular hurry. I wonder if git can cherry-pick across repos, that would be amazing. Otherwise I guess it will have to be a series of PRs to the new cpython repo (see https://paper.dropbox.com/doc/CPython-workflow-changes-mx1k8G6M0rg5JLy80F1r6 ).

[gvanrossum]

One more thing, "mod.py" is a rather non-descript name. Can you come up with something better?

[ilevkivskyi]

Can you come up with something better?

Yes, exactly, this is among the things I am going to do today.

[ilevkivskyi]

@gvanrossum I fixed the remaining issue. As I understand the current workflow, a PR should me made to master, and after it is merged, the corresponding squash-commit should be cherry-picked to other branches (3.5 and 3.6 in this case).
I have made a PR to CPython master python/cpython#28

By the way git can in principle cherry-pick between repos, but I tried this and it feels too complicated, because typing and cpython have quite different directory structure (it is quite easy however to cherry-pick a single commit between repos).

[gvanrossum]

Thanks! Godspeed.