[Dhairya-007]

behave.model had no stubs at all, so context.table — the object almost every step implementation that uses a Gherkin data table touches — was Incomplete, and everything reached through it was Any:

@given("a table of users")
def step_impl(context: Context) -> None:
    for row in context.table:        # Any
        name = row["name"]           # Any

This adds behave/model.pyi covering Table and Row, plus behave/model_core.pyi covering Replayable, which Table derives from. Both keep the module-level __getattr__ escape hatch, so the rest of behave.model is untouched and the package stays a partial stub.

Context is wired up to the new types:

attribute	was	now	assigned from
table	Incomplete | None	Table | None	Step.table in Step.run
active_outline	Incomplete	Row | None	scenario._row in ScenarioOutline.run

Details worth a second look

Signatures were read off behave 1.3.3 and checked against the runtime. The cases that needed care:

• Table.__getitem__ is overloaded — table[0] gives a Row, but table[0:2] gives list[Row], since it delegates to self.rows[index].
• Row.items() returns a zip object, not a list or a dict view.
• Row.as_dict() returns an OrderedDict.
• Row.get is overloaded on its default, like Mapping.get.
• Table.assert_equals is NoReturn — it asserts the comparison and then unconditionally raises NotImplementedError.
• Replayable.type is None on the base and a string on subclasses, so it is ClassVar[str | None] there and ClassVar[str] on Table.
• Table.from_data returns Table, not Self — one branch constructs Table directly rather than cls.
• Row.from_dict's **kwargs are spelled out as the keyword-only line and comments that Row.__init__ actually accepts.

Row.comments stays Incomplete | None: nothing in behave constructs a Row with it, and there is no Comment type in the package, so I did not want to guess.

Validation

• stubtest: Success: no issues found in 84 modules
• mypy --strict: a representative step implementation reading headings, iterating rows, and using named, indexed and sliced access reported 14 Any errors before this change and none after. Every assert_type for str, int, list[str], list[Row] and OrderedDict[str, str] now holds.

[Dhairya-007]

Hii @srittau , requesting for a review and merge

[srittau]

Thanks, two comments below.

[srittau]

Mapping too narrow (there is an explicit isinstance check):

Suggested change

	@classmethod
	def from_dicts(cls, headings: list[str], table_data: Sequence[Mapping[str, str]], line: int | None = None) -> Self: ...
	@classmethod
	def from_data(
	cls, headings: list[str], table_data: Sequence[Mapping[str, str] | _RowData] | None = None, line: int = 0
	) -> Table: ...
	@classmethod
	def from_dicts(cls, headings: list[str], table_data: Sequence[dict[str, str]], line: int | None = None) -> Self: ...
	@classmethod
	def from_data(
	cls, headings: list[str], table_data: Sequence[dict[str, str] | _RowData] | None = None, line: int = 0
	) -> Table: ...

[Dhairya-007]

Done in a2f1e41 — both now take Sequence[dict[str, str]] / Sequence[dict[str, str] | _RowData].

from_dicts raises TypeError("%r (expected: List[dict]") when table_data[0] is not a dict, and from_data takes its else-branch and passes the data straight to Table(headings, rows=table_data), so a non-dict Mapping is rejected by one and silently mishandled by the other.

Left Row.from_dict on Mapping: it only does row_data.get(name), no isinstance check.

[srittau]

Use Never from typing_extensions. NoReturn is deprecated.

[Dhairya-007]

Done in b7eb5c6 — Never from typing_extensions, and NoReturn is dropped from the typing import.

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[Dhairya-007]

Hii @srittau , good for re-review