BUG: assert_equal/assert_array_equal now handle NaN in object arrays (gh-9023) - #32661

PR summary

numpy.testing.assert_equal and assert_array_equal raise a spurious AssertionError when comparing dtype=object arrays containing NaN values in matching positions, even though the two arrays are otherwise identical:

>>> import numpy as np
>>> from numpy.testing import assert_equal
>>> a = np.array([1, 2, np.nan], dtype=object)
>>> assert_equal(a, a)
AssertionError:
Arrays are not equal
(mismatch 33.33333333333333%)

This closes gh-9023. assert_equal/assert_array_equal already treat matching-position NaNs as equal for numeric dtypes (via a NaN-position check in assert_array_compare), but that check is gated by an isnumber(x) and isnumber(y) condition that excludes dtype=object arrays entirely, so dtype=object arrays fall through to a plain == comparison, and nan != nan under IEEE-754 semantics then triggers a false mismatch.

This PR adds an isobject(x) and isobject(y) branch to assert_array_compare, parallel to the existing branches for
numeric, time, and vstring dtypes. Since np.isnan doesn't support dtype=object, NaN-detection for this branch uses each element's self-inequality (v != v), guarded with a try/except so non-comparable objects don't raise.

Scope note: assert_allclose on dtype=object arrays has a related but distinct failure (TypeError from isfinite not
supporting dtype=object, reported in gh-25496). That failure occurs even on object arrays with no NaN at all, since it comes from np.isclose's internal use of isfinite, not from the NaN-position gating this PR touches. Fixing it would require changes to np.isclose itself or a fallback path in assert_allclose's compare(), which is out of scope here — I've left it for gh-25496.

A regression test (test_object_array_nan_items) is added, verifying both that matching-NaN object arrays compare equal and that a genuine mismatch is still correctly caught.

First time contributor introduction

Started contributing to NumPy a few days ago, and main aim it to learn more, enhance my knowledge and to help team to clear their issue triaging backlog.

AI Disclosure

No AI tool is used.