Python 3.14: StructuralCoverageTag.__init__() takes 1 positional argument but 2 were given

[nedbat]

The coverage.py nightly Python test runs started failing on Python 3.14:

      def structural_coverage(label: int) -> StructuralCoverageTag:
          try:
              return STRUCTURAL_COVERAGE_CACHE[label]
          except KeyError:
  >           return STRUCTURAL_COVERAGE_CACHE.setdefault(label, StructuralCoverageTag(label))
  E           TypeError: StructuralCoverageTag.__init__() takes 1 positional argument but 2 were given

The other Python versions are fine.

[nedbat]

Note: the deadsnakes nightly 3.14 that I've been using had been broken for a few weeks, so if this is a bug in 3.14, it could have been introduced since July 11th.

[hynek]

The breakage seems to be related to attrs 24.1.0 release.

Based on python-attrs/attrs#1326 (comment) I would guess that removing the dataclass_transform should fix it??

[Zac-HD]

Thanks to both of you for the report and suggestions! I'm looking into this now and should have a fix out soon.

I'm somewhat surprised that this hasn't failed in our CI yet - we're running 3.14-dev, which pyenv is currently building from cpython main. no, this is obvious, we're currently pinned to the prior version of attrs.

I also don't see how our noop decorator could have some runtime effect:

hypothesis/hypothesis-python/src/hypothesis/internal/conjecture/data.py

Lines 71 to 85 in 3c15b34

	if TYPE_CHECKING:
	from typing import TypeAlias
	from typing_extensions import dataclass_transform
	from hypothesis.strategies import SearchStrategy
	from hypothesis.strategies._internal.strategies import Ex
	else:
	TypeAlias = object
	def dataclass_transform():
	def wrapper(tp):
	return tp
	return wrapper

[Zac-HD]

yeah, that's more like it:

(ok, with a repro in hand it's time to fix this!)

[hynek]

it's beautiful 🥲

[Zac-HD]

cd8a71e seems to have fixed it, so my best guess is that it's (probably something in combination with) auto_attribs=True and you might be able to repro from that?

It's after midnight here though, so I'm just going to get the fix released and then go to bed 😅

[hynek]

Yes, I've finally managed to compile Python 3.14 on my Mac and apparently attrs can't find annotation-only fields on 3.14 anymore. You don't have to do anything, I've got a lead, I just want to understand better what's happening.

[Zac-HD]

Ah, there's some other updates I wanted anyway in that PR, so might as well ship it 🙂

[hynek]

24.2.0 is out now, so you should be able to replace the whole thing with a single @attr.frozen or even @attrs.frozen since you depend on 22+ anyways.