dataclasses requiring InitVar/ClassVar at runtime

[jonyscathe]

As per https://docs.python.org/3/library/dataclasses.html#class-variables and https://docs.python.org/3/library/dataclasses.html#init-only-variables dataclass() inspects the type of a field in order to treat those types in some special way.

Code like their example (with imports added):

from dataclasses import InitVar, dataclass

@dataclass
class C:
    i: int
    j: int = None
    database: InitVar[DatabaseType] = None

    def __post_init__(self, database):
        if self.j is None and database is not None:
            self.j = database.lookup('j')

c = C(10, database=my_database)

Will throw TC002 on the import of InitVar.
But if InitVar is moved into a type checking block then initializing C will fail because database won't exist in the post_init function.

I imagine the only way to handle this (other than my current solution of a noqa anywhere I use InitVar) will to add exceptions for InitVar and ClassVar if dataclass is also imported or something along those lines ?

[sondrelg]

I agree, we should add handling for this 👍

Do you think it makes sense to just ignore InitVar, or are there cases where dataclasses will evaluate other annotations?

Would you be interested in creating a fix for this?

[jonyscathe]

From the dataclasses documentation it appears that it will evaluate InitVar or ClassVar annotations.
I can try making a fix for this.

[sondrelg]

I started looking at this, but I'm no longer able to replicate the issue after v2.

v2 changes the plugin behaviour so that we won't flag imports like from dataclasses import InitVar if there's already one or more other imports from the module made that cannot be guarded. This means, since from dataclasses import dataclass will always exist in the same file, the InitVar import should never be flagged. The only way to mess up is if you did something like this:

from my_app.some_file import dataclass  # bad import
from dataclasses import InitVar

But this doesn't seem worth handling.

The typing module was also completely whitelisted in v2 since we assume from typing import TYPE_CHECKING will be needed in most scopes.

Are you able to still replicate the issue? If not, I think we might consider this resolved, accidentally 🙂

[sondrelg]

I'll close this for now. Please let me know if you still think there's an issue 👍

[jonyscathe]

Sorry I was sick yesterday and wasn't at my computer.
I checked this morning, and yes, it looks like this has been fixed 👍

[sondrelg]

No worries! Thanks for checking 🙂