Enable cudf.pandas test_numpy_ufuncs_basic via tolerant Index comparison

[galipremsagar]

Description

The cudf.pandas pandas-tests plugin xfailed 77 tests/indexes/test_numpy_compat.py::test_numpy_ufuncs_basic[...] parametrizations. 75 of them were not real failures: the GPU results are numerically correct and only differ from NumPy's CPU results by ~1 ULP for transcendental ufuncs (sin, cos, tan, exp, log, sinh, ...). They were xfailed only because pandas._testing.assert_index_equal compares Index values exactly (check_exact=True) by default, so the last-bit drift trips the assertion even though the values agree to rtol=1e-5.

This PR lets those tests run and pass instead of being xfailed:

• Adds a node-id-scoped, autouse fixture relax_exact_index_compare that, only for node ids matching TOLERANT_INDEX_COMPARE_SUBSTRINGS (currently test_numpy_ufuncs_basic), patches tm.assert_index_equal to inject check_exact=False (NumPy/pandas default rtol/atol) and restores it afterward. It reuses the existing replace_kwargs helper and is scoped by node id, so no other test's comparison is affected.
• Drops the 75 now-passing entries from NODEIDS_THAT_FAIL.

Two entries are intentionally kept xfailed because they are not a tolerance issue: complex64-log2 and complex128-log2. There, log2(0+0j) yields -inf+nanj on the GPU (cupy) vs -inf+0j on the CPU (NumPy) — a structural nan-vs-0 mismatch in the imaginary part that no rtol/atol can bridge. Their xfail reason is updated to say so.

Safety / scope

• The relaxation only loosens value comparison; dtype correctness is still enforced — test_numpy_ufuncs_basic makes its own explicit assert result.dtype == ... checks, which run independently of assert_index_equal.
• The fixture is gated on the test node id, so the rest of the pandas-test suite continues to compare exactly.

Verification

Running test_numpy_ufuncs_basic with the updated plugin: 746 passed, 2 xfailed, 0 failures, 0 XPASS. The tolerance margin is comfortable (~100× for float32 ULP drift, far larger for complex128), so it should hold across CI hardware.

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: 23eb506f-764d-4129-b61f-cd46f1ad28ae

📥 Commits

Reviewing files that changed from the base of the PR and between 974d4b0 and cbcc9ce.

📒 Files selected for processing (1)

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

---

📝 Walkthrough

Summary by CodeRabbit

• Chores
  • Updated the testing framework to relax exact index equality checks for a targeted set of test cases, improving reliability for otherwise rigid comparisons.
  • Refined test discrepancy/xfail explanations for log2 scenarios, clarifying that mismatches come from structural NaN vs 0 differences rather than tolerance drift.
  • Removed outdated flaky or version-sensitive dispatch reason entries for nullable NumPy ufunc test cases.

Walkthrough

The pandas testing plugin gains a NODEIDS_TOLERANT_INDEX_COMPARE constant and marker registration to apply a tolerant_index_compare pytest marker to matching test node ids. An autouse fixture checks for this marker and temporarily monkeypatches pandas._testing.tm.assert_index_equal with check_exact=False and rtol=1e-6 for marked tests. Two log2-on-complex-input xfail reason strings are updated with explicit GPU-vs-CPU NaN-vs-0 descriptions, and a block of stale flaky nullable-ufunc entries is removed.

Changes

pandas testing plugin – tolerance fixture and xfail mapping cleanup

Layer / File(s)	Summary
Marker registration and index-comparison fixture python/cudf/cudf/pandas/scripts/pandas-testing-plugin.py	Adds NODEIDS_TOLERANT_INDEX_COMPARE constant and registers tolerant_index_compare marker via pytest_configure. pytest_collection_modifyitems applies the marker to items whose nodeid contains configured substrings. An autouse relax_exact_index_compare fixture monkeypatches pandas._testing.tm.assert_index_equal with check_exact=False and rtol=1e-6 for marked tests, restoring the original in a finally block.
xfail reason mapping cleanup python/cudf/cudf/pandas/scripts/pandas-testing-plugin.py	Updates two test_numpy_ufuncs_basic log2-on-complex-input xfail entries with explicit GPU-vs-CPU NaN-vs-0 mismatch descriptions. Removes a block of stale "Flaky/version-sensitive cudf.pandas dispatch" reason entries for nullable ufunc parameterizations in test_numpy_compat.py.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• rapidsai/cudf#22643: Both PRs modify python/cudf/cudf/pandas/scripts/pandas-testing-plugin.py, specifically the pytest_collection_modifyitems hook for test marker and configuration logic.

Suggested labels

improvement

Suggested reviewers

• brandon-b-miller
• TomAugspurger
• Matt711
• bdice
• vyasr

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately describes the main change: enabling cudf.pandas tests by introducing tolerant index comparison.
Description check	✅ Passed	The description is comprehensive and clearly related to the changeset, explaining the motivation, implementation, and verification of the tolerance-based index comparison feature.
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