pytest.approx considers boolean numeric types

[The-Compiler]

From the docs:

You can also use approx to compare nonnumeric types, or dicts and sequences containing nonnumeric types, in which case it falls back to strict equality. This can be useful for comparing dicts and sequences that can contain optional values:

>>> {"required": 1.0000005, "optional": None} == approx({"required": 1, "optional": None})
True
>>> [None, 1.0000005] == approx([None,1])
True
>>> ["foo", 1.0000005] == approx([None,1])
False

However, that seems to treat booleans as numerical values too:

import pytest

def test_approx():
    d = {"value": 0.3, "active": True}
    expected = {"value": 0.1 + 0.2, "active": False}
    assert d == pytest.approx(expected)

results in:

AssertionError: assert {'active': True, 'value': 0.3} == approx({'value': 0.30000000000000004 ± 3.0e-07, 'active': False ± 1.0e-12})

(on Python 3.9.7 and pytest 6.2.5). Note the False ± 1.0e-12. While this is technically correct, it doesn't seem like a very reasonable way of comparing booleans.

cc @jvansanten who seems to have implemented this originally in #7710.

[jvansanten]

I'm not so sure what to do about this. The idea of #7710 was to use approximate comparison whenever (expected-obtained) was allowed by the target types, only falling back to equality for non arithmetic types. int-bool is allowed (in contrast to Enum-Enum), so True == approx(False, abs=2) is technically correct as you point out.

One straightforward solution to special-case the equality fallback, so approximate comparisons are used for all subclasses of Decimal and Complex, but not for bool.