GH-51301: Fix deprecation warnings in the tests with numpy/pandas nightly

[AlenkaF]

Rationale for this change

Test warnings accumulated in our CI.

What changes are included in this PR?

The warnings are fixed or filtered if the deprecated functionality is still being tested.

Are these changes tested?

Yes.

Are there any user-facing changes?

No. Fixing test warnings in our CI.

Was AI used for this PR?

In accordance to the AI generation guidelines, please disclose below whether and how AI was used in this PR.

PR code and description written by:

• Human
• AI

Reviewed before submission by:

• Human
• AI
• Not reviewed

• GitHub Issue: [Python] Fix deprecation warnings in the tests with numpy/pandas nightly #51301

[github-actions]

⚠️ GitHub issue #51301 has been automatically assigned in GitHub to PR creator.

[AlenkaF]

Opened a PR though we have one already opened: #51309.
Might be wrong, but the process seems very similar to other reviews where quite a lot of time is needed to communicate with the contributor (or the contributor's agent) what would be good to change. Decided to fix this quickly by myself.

[AlenkaF]

Warnings gone:
https://github.com/apache/arrow/actions/runs/35588604946/job/106297664397?pr=51404#step:6:4899

vs:
https://github.com/apache/arrow/actions/runs/35589580522/job/106300673910#step:6:4897

from the same AMD64 Conda Python 3.14 Pandas latest job.

@jorisvandenbossche if you have time for review.

[Copilot]

Copilot review overview

🟡 Changes recommended

Warning failures can leak allocated DLPack tensors in both export paths.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 2

Open (2)

• Prevent tensor leak when warning emission raises · New
• Prevent exported tensor leak when warning emission raises · New

What changed in this PR

This PR updates tests and DLPack handling to address NumPy/Pandas deprecation warnings while preserving legacy behavior coverage.

Changes:

• Updated deprecated NumPy and pandas test usage.
• Filtered intentional deprecation warnings.
• Adjusted legacy DLPack export warning handling and tests.

File	Summary
python/​pyarrow/​tests/​test_pandas.py	Uses explicit timedelta units.
python/​pyarrow/​tests/​test_dlpack.py	Updates warning filters and versioned DLPack coverage.
python/​pyarrow/​tests/​test_compute.py	Updates timestamp construction and warning filters.
python/​pyarrow/​tests/​test_array.py	Handles NumPy generic-unit deprecations.
python/​pyarrow/​tensor.pxi	Adjusts legacy DLPack export warning handling.
python/​pyarrow/​array.pxi	Adjusts legacy DLPack export warning handling.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[AlenkaF]

Converting to draft as I need to think about Copilot's comment, looks right from a quick read.
I also need to rebase as I included some changes that have been fixed by #51307

[AlenkaF]

I decided to remove changes connected to the depr warning in __dlpack__:
The solution would need or check if warning is turned into an error and calling a deleter, or constructing a PyCapsule before emitting a warning. Both seem to be things that need some discussion and a fix in a separate issue I think.

Currently the CI is not showing any deprecation warnings in out tests: https://github.com/apache/arrow/actions/runs/35728469810/job/106747835234?pr=51404#step:6:4240

[Copilot]

Copilot review overview

🟡 Changes recommended

Unresolved NumPy compatibility issues affect DLPack tests and nightly generic-unit warning coverage.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

• Versioned DLPack tests fail on supported NumPy 2.0 · New

Resolved since last review (2)

• Prevent exported tensor leak when warning emission raises
• Prevent tensor leak when warning emission raises

[jorisvandenbossche]

@github-actions crossbow submit test-conda-python-3.14-pandas-nightly-numpy-nightly

[github-actions]

Revision: 84a8518

Submitted crossbow builds: ursacomputing/crossbow @ actions-0a1f3fcf86

Task	Status
test-conda-python-3.14-pandas-nightly-numpy-nightly

[jorisvandenbossche]

If leaving the DLPack depr question for a separate issue, I would maybe also leave out any of the dlpack-related changes here.

[AlenkaF]

Reverted the rest dlpack changes: a148c13
and created a new issue: #51481

[jorisvandenbossche]

@AlenkaF thanks for the PR! I triggered a nightly pandas/numpy crossbow build, because it is there that I checked for the warnings (it might be that some of those were also showing up in the main CI builds, though)

[Copilot]

Copilot review overview

🟡 Changes recommended

Three moderate review issues remain unresolved.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

• Use 2.5.0.dev0 threshold for NumPy nightly deprecation guard · New

Resolved since last review (1)

• Versioned DLPack tests fail on supported NumPy 2.0

[AlenkaF]

@github-actions crossbow submit test-conda-python-3.14-pandas-nightly-numpy-nightly

[Copilot]

Copilot review overview

🔵 Needs a closer look

Unresolved warnings and reduced generic timedelta test coverage remain.

Review effort: Lite
Findings: None

Resolved since last review (1)

• Preserve strict frequency checks in shared round-trip helper

[github-actions]

Revision: 8020466

Submitted crossbow builds: ursacomputing/crossbow @ actions-d4a8f679ef

Task	Status
test-conda-python-3.14-pandas-nightly-numpy-nightly

[Copilot]

Copilot review overview

🔵 Needs a closer look

DLPack warning handling remains incomplete, and the NumPy prerelease version guards need correction.

Review effort: Lite
Findings: None

[AlenkaF]

I have run 3 different version of the test_does_not_mutate_timedelta_dtype on this diff:

-  if (type == NPY_DATETIME || type == NPY_TIMEDELTA) {
+  if (type == NPY_DATETIME) {

which should prodice a bug and the test should fail. That does
happen for the test on main:

expected = np.dtype('m8')

assert np.dtype(np.timedelta64) == expected

df = pd.DataFrame({"a": [np.timedelta64()]})
t = pa.Table.from_pandas(df)
t.to_pandas()

assert np.dtype(np.timedelta64) == expected

and the older version of the test on this PR:

expected = np.dtype('m8')

assert np.dtype(np.timedelta64) == expected

df = pd.DataFrame({"a": [np.timedelta64(0, "s")]})
t = pa.Table.from_pandas(df)
t.to_pandas()

assert np.dtype(np.timedelta64) == expected

but the bug can not be reproduced with this change:

expected = np.dtype('<m8[s]')

assert np.dtype(np.timedelta64(0, "s")) == expected

df = pd.DataFrame({"a": [np.timedelta64(0, "s")]})
t = pa.Table.from_pandas(df)
t.to_pandas()

assert np.dtype(np.timedelta64(0, "s")) == expected

as @tadeja opened up here: #51404 (comment).
I reverted the change in 40997f0 so we do keep testing the buggy behavior is fixed.

[AlenkaF]

@github-actions crossbow submit test-conda-python-3.14-pandas-nightly-numpy-nightly

[github-actions]

Revision: 40997f0

Submitted crossbow builds: ursacomputing/crossbow @ actions-7566b946d6

Task	Status
test-conda-python-3.14-pandas-nightly-numpy-nightly

[AlenkaF]

I think I have addressed all the comments, the logs are clean: https://github.com/ursacomputing/crossbow/actions/runs/37419973097/job/112126927373#step:7:4398 with only dlpack fixes needed. I will work on that next.