Fix generic inheritance for dataclass init methods

[natemcmaster]

Description

Fixes#7520

Updates the dataclasses plugin. Instead of directly copying attribute type along the MRO, this first resolves typevar in the context of the subtype.

Test Plan

Added 4 new test cases to cover various generic inheritance scenarios.

[natemcmaster]

One of the test jobs failed (python3.5.1 with mypyc). The errors appear unrelated to my changes, and that same job passed when I ran the same commit hash - https://travis-ci.org/github/natemcmaster/mypy/builds/722519864. Is this failure due to flaky tests?

[JelleZijlstra]

@natemcmaster it's likely due to a recent setuptools release that broke stuff; I submitted a PR that will hopefully fix CI.

[natemcmaster]

@JelleZijlstra thanks! I merged latest updates from master into my PR. CI is passing now on this PR.

[natemcmaster]

Friendly bump for review :)

[natemcmaster]

Hello there, checking in again. I merged from master and pushed to my branch to ensure this PR is still merge-conflict free and passing tests. Any chance I can get a review? Thanks!

[nathan-chappell]

Any status on the review? I have a super lame generic dataclass inheritance chain mess that's waiting on it. And I'm sure you don't have anything better to do...

[natemcmaster]

I haven't heard anything from the repo maintainers :-/

[hauntsaninja]

Thanks for this (and #9383)!

Sorry for the delay here, mypy's just been really short on reviewer bandwidth / I've shied away from reviewing these because I'm not familiar with these plugins. But the change make sense, your tests look good and mypy_primer doesn't surface any complaints.

[natemcmaster]

Thanks for merging this and the other PR, @hauntsaninja! I totally understand the delay. I used to be a maintainer on a big open source project myself :). Thanks for following up on this!

[binishkaspar]

@natemcmaster Fix works very well. However the following code raises INTERNAL ERROR,

DRIVE_TYPE=TypeVar('DRIVE_TYPE', bound='Driver')
# This WORKS# DRIVER_STATE_TYPE = TypeVar('DRIVER_STATE_TYPE', bound='DriverState')@dataclassclassDriverState(Generic[DRIVE_TYPE]):
driver: DRIVE_TYPEpid: str=''# This doesn't workDRIVER_STATE_TYPE=TypeVar('DRIVER_STATE_TYPE', bound='DriverState')
@dataclassclassDriver(Generic[DRIVER_STATE_TYPE]):
config: DRIVER_STATE_TYPE@dataclassclassESState(DriverState['ESDriver']):
special_name: str='ES Driver'@dataclassclassESDriver(Driver[
ESState
]):
defname(self) ->None:
print(self.config.special_name)

Is it because supertype not recursively expanded?

[natemcmaster]

@binishkaspar can you open a new issue? I'm not sure of the expected behavior of TypeVar with an open generic bound.