GH-51228: [Python] Raise instead of crashing on an invalid registry

[1fanwang]

Rationale for this change

Passing an invalid registry to a PyArrow tabular call or UDF registration can terminate Python with SIGSEGV. Applications should receive TypeError instead.

Fixes #51228.

What changes are included in this PR?

The tabular call and shared registration path validate the registry before reading its native pointer. Registration covers scalar, vector, aggregate, and tabular functions. Default and explicit FunctionRegistry instances remain accepted.

Are these changes tested?

Separate native processes reproduced all four registration crashes with a 20-second timeout. The rebuilt compute extension raises TypeError, and registration plus lookup still works with default and explicit registries. The regression tests run in-process.

Raw logs

$ python -c "import pyarrow as pa; import pyarrow.compute as pc; pc.register_scalar_function(func=lambda context: None, function_name='invalid_registry', function_doc={'summary': '', 'description': ''}, in_types={}, out_type=pa.struct([]), func_registry=1)"
Before, native Arrow 26.0.0-SNAPSHOT:
returncode=-11
After:
TypeError: func_registry must be a FunctionRegistry
returncode=1

$ cd python
$ python -m pytest pyarrow/tests/test_compute.py -k 'register_function_rejects_invalid_registry or call_tabular_function_rejects_invalid_registry or list_functions' -q
......                                                                   [100%]
6 passed, 603 deselected, 1 warning in 0.38s

Are there any user-facing changes?

Invalid registry arguments raise TypeError instead of terminating Python. Valid registry arguments retain their behavior.

New Contributor's Guide |
Contributing Overview |
AI-generated Code Guidance

• GitHub Issue: [Python] call_tabular_function() segfaults for some wrong typed func_registry values #51228

[github-actions]

⚠️ GitHub issue #51228 has been automatically assigned in GitHub to PR creator.

[Copilot]

🟡 Changes recommended

The new subprocess-based test should be marked with @pytest.mark.processes so it can be skipped on platforms/configurations where process creation is disabled.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Fixes a crash in the PyArrow compute bindings by validating the optional func_registry argument passed to pyarrow.compute.call_tabular_function() before performing the Cython cast, so invalid types raise a catchable TypeError instead of segfaulting.

Changes:

• Add a runtime type check for func_registry in call_tabular_function() and raise TypeError on invalid values.
• Add a regression test that runs the call in a subprocess to assert the failure mode is a Python exception (not a process crash).

File summaries

File	Description
python/pyarrow/_compute.pyx	Adds func_registry type validation before casting to a native registry pointer.
python/pyarrow/tests/test_compute.py	Adds a subprocess-based regression test to ensure invalid registries raise TypeError rather than crashing.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 1
• Review effort level: Lite

---

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

[Copilot]

🟢 Approval recommended

The fix is narrowly scoped, matches the stated user-facing behavior, and is covered by a regression test that would fail under the prior segfaulting implementation.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 0 new
• Review effort level: Lite

[AlenkaF]

Would it make sense to also add a check in _register_user_defined_function as every register_* function uses that?

[Copilot]

🟡 Changes recommended

The added regression test should be executed in a subprocess (as described) to remain safe if the segfault regresses and to avoid crashing the test runner.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 1
• Review effort level: Lite

[AlenkaF]

Two things left:

• comment GH-51228: [Python] Raise instead of crashing on an invalid registry #51247 (review) is not yet addressed
• PR description needs an update on the testing section.

[Copilot]

🟢 Approval recommended

The fix is covered by regression tests and prevents the reported process crash.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 0 new
• Review effort level: Lite

[github-actions]

⚠️ GitHub issue #51228 has been automatically assigned in GitHub to PR creator.

[1fanwang]

Done in 388ac7b.