Fix class/def line reporting and ignoring for 3.8+.

[brandtbucher]

Fixes #3871. This change fixes a few things concerning class/function definition line numbers.

For users running mypy with Python 3.8+:

• Function definitions are now reported on the correct line, rather than the line of the first decorator plus the number of decorators.
• Class definitions are now reported on the correct line, rather than the line of the first decorator.
• These changes are fully backward compatible with respect to # type: ignore lines. In other words, existing comments will not need to be moved. This is accomplished by defining an ignore "scope" just like we added to expressions with Expression-scoped ignores in Python 3.8. #6648. See the recent discussion at Function line number information is not always correct #3871 for reasoning.

For other users:

• Class definition lines are reported by using the same decorator-offset estimate as functions. This is both more accurate, and necessary to get 15 or so tests to pass for both pre- and post-3.8 versions. This seems fine, since there doesn't appear to be a compelling reason why it was different in the first place.
• This change is also backward-compatible with existing ignores, as above.

About 15 tests had to have their error messages / nodes moved down one line, and 4 new regression tests have been added to check-38.test.

[brandtbucher]

Bump. Hopefully this doesn't get too stale from all of the new sprint PRs!

[gvanrossum]

Alas, it's already stale. Could you merge the conflicts?

[brandtbucher]

@gvanrossum Merged, tests are passing.

[gvanrossum]

This is getting close, but I have a couple of questions, since this looks like it will be fairly invasive...

[gvanrossum]

This comment confused me. The key thing to note here is that we want to point lineno to the line that has the def keyword, while we want to point end_lineno to the line that would have been used for Python 3.7 and before. And we must compute end_lineno (i.e. the "old start line") using the crufty way given here since the old way of computing it (i.e. n.lineno + len(n.decorator_list) won't have the same result in 3.8.

Sorry that this is still so confusing. The code is right!

[gvanrossum]

Are you going to update these comments? I hope so!

[brandtbucher]

I'll try to reword this to be more concise. Maybe something like:

# Set end_lineno to the old pre-3.8 lineno, in order to keep
# existing "# type: ignore" comments working:

[gvanrossum]

Sounds good -- also for the other location!

[gvanrossum]

Do you even need this? It seems this was already set on line 432 above.

[brandtbucher]

Nice catch. I believe that's left over from an earlier iteration of this change.

[gvanrossum]

Similar comment (except I wasn't really confused here, since I had just figured out the previous one :-).

[brandtbucher]

As above:

# Set end_lineno to the old mypy 0.700 lineno, in order to keep
# existing "# type: ignore" comments working:

[brandtbucher]

New changes are passing, ready for another review.

[gvanrossum]

Thanks! Congratulations on getting this thorny PR in. Excellent work.

[brandtbucher]

Thanks, that really means a lot coming from you. Always happy to help!