TCH003 false-positive when using dataclasses.KW_ONLY

[flying-sheep]

Seen in Ruff 0.5.7

In the following code, from __future__ import annotations causes the KW_ONLY to become a string.
The dataclasses module can deal with that, but only if the object isn’t in a if TYPE_CHECKING block.
Ruff supports this for things like ClassVar¹, i.e. if a ClassVar is used in a dataclass, Ruff knows the from typing import ClassVar needs to come out of the if TYPE_CHECKING block. However the same doesn’t seem to be true here.

from __future__ import annotations

from dataclasses import KW_ONLY, dataclass

@dataclass
class Test:
    a: int
    _: KW_ONLY
    b: str

---

¹This is probably the place that should handle KW_ONLY

ruff/crates/ruff_linter/src/rules/flake8_type_checking/helpers.rs

Lines 108 to 139 in 82a3e69

	/// Returns `true` if an annotation will be inspected at runtime by the `dataclasses` module.
	///
	/// Specifically, detects whether an annotation is to either `dataclasses.InitVar` or
	/// `typing.ClassVar` within a `@dataclass` class definition.
	///
	/// See: <https://docs.python.org/3/library/dataclasses.html#init-only-variables>
	pub(crate) fn is_dataclass_meta_annotation(annotation: &Expr, semantic: &SemanticModel) -> bool {
	if !semantic.seen_module(Modules::DATACLASSES) {
	return false;
	}
	// Determine whether the assignment is in a `@dataclass` class definition.
	if let ScopeKind::Class(class_def) = semantic.current_scope().kind {
	if class_def.decorator_list.iter().any(|decorator| {
	semantic
	.resolve_qualified_name(map_callable(&decorator.expression))
	.is_some_and(|qualified_name| {
	matches!(qualified_name.segments(), ["dataclasses", "dataclass"])
	})
	}) {
	// Determine whether the annotation is `typing.ClassVar` or `dataclasses.InitVar`.
	return semantic
	.resolve_qualified_name(map_subscript(annotation))
	.is_some_and(|qualified_name| {
	matches!(qualified_name.segments(), ["dataclasses", "InitVar"])
	|| semantic.match_typing_qualified_name(&qualified_name, "ClassVar")
	});
	}
	}
	false
	}

[charliermarsh]

Wow TIL, I’ve never seen that feature before.

[charliermarsh]

Ah, this requires:

[tool.ruff.flake8-type-checking]
strict = true

So probably somewhat rare (but should be fixed).

[flying-sheep]

Thank you!

[flying-sheep]

Btw.: I always do

[tool.ruff.lint.flake8-type-checking]
exempt-modules = []
strict = true

because that way I can use new typing features without having to pay attention that they are imported from typing_extensions or in a if TYPE_CHECKING block for backwards compatibility. They just always are in there.