Fix masked dtype equality in cudf.pandas so real_dtype == proxy_dtype holds

[galipremsagar]

Description

Fixes the test_reductions_2d_axis0 failures for masked dtypes (Boolean/Int*/UInt*) in the cudf.pandas pandas test suite.

Root cause

The test asserts dtype == expected.dtype, where:

• dtype is a real pandas dtype pulled from pandas' internal NUMPY_INT_TO_DTYPE table, and
• expected.dtype is a cudf.pandas proxy dtype.

This evaluates real_dtype.__eq__(proxy_dtype). pandas' ExtensionDtype.__eq__ does isinstance(other, type(self)). Since the proxy only subclasses the abstract ExtensionDtype (not the concrete pd.Int64Dtype), the isinstance check fails and equality returns False — even though the two dtypes are semantically identical. Note the asymmetry: proxy == real already worked (it dispatches through the proxy's own __eq__); only real == proxy failed.

Fix

Expose the concrete slow type as __class__ on each masked dtype proxy. isinstance consults __class__, so isinstance(proxy, pd.Int64Dtype) becomes True, equality proceeds, and since these masked dtypes have empty _metadata (()) it reduces to exactly that type check. type(proxy) is unaffected (it bypasses __class__), so the proxy machinery that keys off the real type is undisturbed.

Why this approach

• Fix equality on the proxy, not the data. The real == proxy asymmetry occurs anywhere pandas internals hold real dtype instances (NUMPY_INT_TO_DTYPE, cast.py, the dtype registry, ...) and compare them against proxy-derived dtypes. Fixing the proxy repairs every such site at once; patching individual call sites would be whack-a-mole.
• Proxying NUMPY_INT_TO_DTYPE is not viable. It's a plain dict whose values are real dtype instances built during pandas' own module execution; the module accelerator wraps namespace attributes, not the interior of data containers, so it is (and stays) un-proxied.
• Scoped to dtype proxies only — not a blanket __class__ spoof. These dtypes are immutable, have empty _metadata, and are never reconstructed via self.__class__(...), so spoofing __class__ is safe. Doing it globally for stateful proxies (Series/DataFrame/arrays) would be dangerous: reconstruction via self.__class__(...) would yield native pandas objects, and proxy detection / copy / pickle all rely on the true class.

With equality fixed, all 28 previously-xfailed test_reductions_2d_axis0 parametrizations now pass, so their entries are removed from NODEIDS_THAT_FAIL.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: df6aa889-7be3-4982-a7e5-8fc6c9935129

📥 Commits

Reviewing files that changed from the base of the PR and between afd5497 and 4db5c14.

📒 Files selected for processing (1)

• python/cudf/cudf/pandas/scripts/pandas-testing-plugin.py

🚧 Files skipped from review as they are similar to previous changes (1)

• python/cudf/cudf/pandas/scripts/pandas-testing-plugin.py

---

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes
  • Fixed dtype equality comparisons between cuDF and pandas for nullable/masked Boolean, Integer (signed/unsigned) and Float dtypes.
  • Addressed tracked test failures for multi-dimensional reductions on masked/nullable arrays, improving compatibility and test stability.

Walkthrough

Proxy pandas dtypes in cuDF now expose the matching pandas dtype class via additional_attributes["__class__"] for Boolean, integer/unsigned integer, and float proxies. Corresponding test expectation entries for masked-array reductions were removed from the pandas-testing-plugin.

Changes

Dtype Proxy Equality Compatibility

Layer / File(s)	Summary
Dtype proxy __class__ attribute mapping python/cudf/cudf/pandas/_wrappers/pandas.py	BooleanDtype, signed/unsigned integer proxies (Int8/16/32/64Dtype, UInt8/16/32/64Dtype), and float proxies (Float32Dtype, Float64Dtype) add additional_attributes["__class__"] pointing to the corresponding pandas dtype types (e.g., pd.Int8Dtype, pd.Float32Dtype). Inline docstrings explain pandas' ExtensionDtype.__eq__ using isinstance(other, type(self)) checks.
Remove now-passing test expectations python/cudf/cudf/pandas/scripts/pandas-testing-plugin.py	Removed previously recorded failing expectation entries for tests/extension/test_masked.py::TestMaskedArrays::test_reductions_2d_axis0[...] that mapped to "AssertionError".

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Suggested reviewers

• vyasr
• bdice
• rjzamora

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main fix: enabling equality checks between real pandas dtypes and their cudf.pandas proxy equivalents for masked dtypes.
Description check	✅ Passed	The description thoroughly explains the root cause (isinstance check asymmetry), the fix (exposing concrete dtype via class), and rationale, all directly related to the changeset.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[galipremsagar]

/merge