PyCollector._genfunctions: use already created fixtureinfo

[blueyed]

Function creates a _fixtureinfo already:
https://github.com/pytest-dev/pytest/blob/fed535694/src/_pytest/python.py#L1392-L1395

[blueyed]

NOTE: I've not looked too closely at this, just noticed it via pdb'ing that it got called twice.

[nicoddemus]

I had checked it before approval, but decided to do a double check:

They seem to be identical calls:

pytest/src/_pytest/python.py

Lines 1396 to 1398 in 1dc265e

	fixtureinfo = self.session._fixturemanager.getfixtureinfo(
	self, self.obj, self.cls, funcargs=True
	)

and:

pytest/src/_pytest/python.py

Line 393 in 1dc265e

fixtureinfo = fm.getfixtureinfo(definition, funcobj, cls)

(funcargs is True by default)

FuncFixtureInfo is mostly a data class:

pytest/src/_pytest/fixtures.py

Lines 303 to 339 in 1dc265e

	@attr.s(slots=True)
	class FuncFixtureInfo:
	# original function argument names
	argnames = attr.ib(type=tuple)
	# argnames that function immediately requires. These include argnames +
	# fixture names specified via usefixtures and via autouse=True in fixture
	# definitions.
	initialnames = attr.ib(type=tuple)
	names_closure = attr.ib() # List[str]
	name2fixturedefs = attr.ib() # List[str, List[FixtureDef]]
	def prune_dependency_tree(self):
	"""Recompute names_closure from initialnames and name2fixturedefs
	Can only reduce names_closure, which means that the new closure will
	always be a subset of the old one. The order is preserved.
	This method is needed because direct parametrization may shadow some
	of the fixtures that were included in the originally built dependency
	tree. In this way the dependency tree can get pruned, and the closure
	of argnames may get reduced.
	"""
	closure = set()
	working_set = set(self.initialnames)
	while working_set:
	argname = working_set.pop()
	# argname may be smth not included in the original names_closure,
	# in which case we ignore it. This currently happens with pseudo
	# FixtureDefs which wrap 'get_direct_param_fixture_func(request)'.
	# So they introduce the new dependency 'request' which might have
	# been missing in the original tree (closure).
	if argname not in closure and argname in self.names_closure:
	closure.add(argname)
	if argname in self.name2fixturedefs:
	working_set.update(self.name2fixturedefs[argname][-1].argnames)
	self.names_closure[:] = sorted(closure, key=self.names_closure.index)

and prune_dependency_tree is called here:

pytest/src/_pytest/python.py

Line 415 in 1dc265e

fixtureinfo.prune_dependency_tree()

So I think it is OK, althought the prune_dependency_tree code could be improved (I'm not sure I like the same object being changed over like this).

[blueyed]

@nicoddemus
Thanks for looking into / confirming it.
Let's wait for another approval though.

[RonnyPfannschmidt]

at first glance this looks good to me,
seems like i missed this opportunity when adding function-definition first

[RonnyPfannschmidt]

based on the looks this is fine