Update to mypy 0.991 for compatible-mypy & CI #1260
Conversation
intgr
force-pushed
the
update-to-mypy-0.991
branch
from
0d42a91 to
8aaaee9
Compare
intgr
force-pushed
the
update-to-mypy-0.991
branch
from
8aaaee9 to
075caf5
Compare
intgr
force-pushed
the
update-to-mypy-0.991
branch
from
84993cf to
32f18cd
Compare
intgr
force-pushed
the
update-to-mypy-0.991
branch
from
3b8cbb6 to
9115891
Compare
| warn_unused_ignores = True | ||
| warn_redundant_casts = True | ||
| warn_unused_configs = True | ||
| warn_unreachable = True | ||
| disallow_untyped_defs = true | ||
| disallow_incomplete_defs = true | ||
| show_error_codes = False |
There was a problem hiding this comment.
show_error_codes default changed in mypy 0.990.
I'm reverting to the old behavior, because so many .yml test files have mypy error messages without the error code.
There was a problem hiding this comment.
Should we try to work towards show_error_codes = True? It feels reasonable that tests are granular enough to fail if a specific error code changes.
It could be an option to try to resolve this over time. Best thing would probably be to have a linter that disallows type: ignore i.e. without code in the test suite. I suppose that could be tracked in an issue.
| warn_unused_ignores = True | ||
| warn_redundant_casts = True | ||
| warn_unused_configs = True | ||
| warn_unreachable = True | ||
| disallow_untyped_defs = true | ||
| disallow_incomplete_defs = true | ||
| show_error_codes = False | ||
| disable_error_code = empty-body |
There was a problem hiding this comment.
Replaced warn_no_return = False with disable_error_code = empty-body, the latter is less dangerous as it now only applies to empty functions.
| @@ -309,6 +311,8 @@ def get_type_analyze_hook(self, fullname: str) -> Optional[Callable[[AnalyzeType | |||
| "django_stubs_ext.annotations.WithAnnotations", | |||
| ): | |||
| return partial(handle_annotated_type, django_context=self.django_context) | |||
| else: | |||
| return None | |||
There was a problem hiding this comment.
This was added because I removed warn_no_return = False mypy setting.
|
I really want this for the next release, but it has been difficult to get anyone to review this. Maybe @ngnpope? |
| warn_unused_ignores = True | ||
| warn_redundant_casts = True | ||
| warn_unused_configs = True | ||
| warn_unreachable = True | ||
| disallow_untyped_defs = true | ||
| disallow_incomplete_defs = true | ||
| show_error_codes = False |
There was a problem hiding this comment.
Should we try to work towards show_error_codes = True? It feels reasonable that tests are granular enough to fail if a specific error code changes.
It could be an option to try to resolve this over time. Best thing would probably be to have a linter that disallows type: ignore i.e. without code in the test suite. I suppose that could be tracked in an issue.
|
Awesome, thanks!
For adding But in the It would make the already long |
Ah, right, didn't think about that it was the reverse case that was the issue. I suppose if we'd prefer matching on code instead of a complete message one could go with |
Sorry, I do want to help out, but I have to squeeze it in around everything else... 🙂 Me right now: 🤹🏻
I don't think that would work so well in most cases because the messages are often formatted with the type values which we then would not detect changes for... |
No worries :) |
Made a few tweaks that were needed to get our CI passing with mypy 0.991.
There is still a new regression withThanks to @flaeppe for the fix!mypy_django_pluginthat I haven't investigated, which happens with Django 3.2 only. Created an issue for it: #1261 Disabled testing with Django 2.2 and 3.2 in CI for now.