fix #11797: be more lenient on SequenceLike approx

[RonnyPfannschmidt]

this needs a validation as it allows partially implemented sequences

closes #11797

this changes the recursive map of approx to accept all sequences (just like approx)
i would like to gather feedback form @nicoddemus and @Zac-HD on whether to accept the recursion in general,
or if we ought to limit it by changing the toplevel call in ApproxSequenceLike to a listcomp operating only on the elements of the sequence

[Zac-HD]

I'd go with the limited form only; we already treat lists and tuples as equal if their contents are approx-equal, so extending that to any sequence-like makes sense to me.

We'll just want to be careful that the repr shows the original thing, and that self-containing sequences don't cause an infinite loop (they do now for list/tuple, but adding new types makes it more likely imo).

[RonnyPfannschmidt]

@Zac-HD the repr from ApproxSequenceLike currently shows a list of approx in the repr instead of the original object

as per

pytest/src/_pytest/python_api.py

Lines 315 to 320 in a3a2ffc

	def __repr__(self) -> str:
	seq_type = type(self.expected)
	if seq_type not in (tuple, list):
	seq_type = list
	return "approx({!r})".format(
	seq_type(self._approx_scalar(x) for x in self.expected)

altering that part of the repr is something i consider out of scope for this fix

[nicoddemus]

Overall this looks good!

if we ought to limit it by changing the toplevel call in ApproxSequenceLike to a listcomp operating only on the elements of the sequence

Which top level call you mean?

[nicoddemus]

altering that part of the repr is something i consider out of scope for this fix

Indeed, that's not so simple as might seem at first glance, because we cannot assume we can create arbitrary classes just by passing a sequence... a user might be using a custom class which is sequence-like, but we cannot assume we know/can create a new one from a sequence.

[RonnyPfannschmidt]

@nicoddemus the call to the recursive map function - we have a coice between keeping certain custom types or not in its insides

im currently not sure whether the caller should map over the sequence or the recursion should expand

[thorink]

Is there any update on this PR or anything I could test or where I could contribute?

[nicoddemus]

Suggested change

	Ensure approx for SequenceLike objects doesn't wrap those sequences in a scalar.
	Ensure :func:`pytest.approx` for :class:`Sequence <collections.abc.Sequence>`-like objects does not wrap those sequences in a scalar.

Or:

Suggested change

	Ensure approx for SequenceLike objects doesn't wrap those sequences in a scalar.
	:func:`pytest.approx` now correctly handles :class:`Sequence <collections.abc.Sequence>`-like objects.

But then this feels more like an improvement than a bugfix? 🤔

[RonnyPfannschmidt]

I'll recategorize and rebase as soon as I get to it again