Fix inconsistency with set equality in approx_eq

[ACE07-Sev]

closes #6376 .

[pavoljuhas]

Please fix for unsortable sets and make sure the tests exercise some equal, but different-order sets; see inline comments for more details.

[pavoljuhas]

Suggested change

	if isinstance(val, set):
	if isinstance(val, (frozenset, set)):

And the same below.

[pavoljuhas]

This will throw TypeError for set sa0 = {"a", 0}, but such set should be approx_eq to itself - approx_eq(sa0, sa0) is True.

Please add a test for unsortable sets, ie, that asserts cirq.approx_eq({"a", 0}, {"a", 0}) is true as expected.

Suggested change

	val = sorted(val)
	try:
	val = sorted(val)
	except TypeError:
	return NotImplemented

and the same for other below.

You will also need to change the caller to correctly handle a NotImplemented value here -

Cirq/cirq-core/cirq/protocols/approximate_equality_protocol.py

Lines 116 to 118 in 078af27

	# If the values are iterable, try comparing recursively on items.
	if isinstance(val, Iterable) and isinstance(other, Iterable):
	return _approx_eq_iterables(val, other, atol=atol)

[pavoljuhas]

This does not test the new code as shown by coverage error. set and frozenset use the same hash and are constructed from the same sequence so this test works without sorting too:

for n in range(10, 20):
    assert list(frozenset(cirq.LineQubit.range(n))) == list(frozenset({*cirq.LineQubit.range(n)}))

To test the new code, you need to find sets for which the iteration sequence changes with insertion order:

found_differently_ordered_sets = False
for i in range(20):
    for j in range(i + 1, 20):
        sij = {cirq.q(i), cirq.q(j)}
        sji = {cirq.q(j), cirq.q(i)}
        if list(sij) != list(sji):
            found_differently_ordered_sets = True
            assert cirq.approx_eq(sij, sji)
            assert cirq.approx_eq(frozenset(sij), sji)

assert found_differently_ordered_sets

[pavoljuhas]

Please revert all formatting changes that convert single to double quotes. They make it harder to review the code and also make git-blame less useful.

We may decide to standardize double quotes later, but that would need its own PR (which can be added to .git-blame-ignore-revs).

[ACE07-Sev]

Sure thing, I accidentally called ruff on it. I'll revert them in my commit.