[Mazen050]

Summary

Fixes #20226

This PR reduces runtime type-checking overhead in the Python SDK by consolidating two nested DoFn wrappers into a single RuntimeTypeCheckWrapperDoFn.

Previously, input and output checks were handled by separate wrappers. The merged wrapper aims to preserve existing type-checking behavior while eliminating one level of Python call indirection per processed element.

Changes

• Consolidate OutputCheckWrapperDoFn and TypeCheckWrapperDoFn into RuntimeTypeCheckWrapperDoFn.
• Cache the unbound process function while avoiding storing a bound method on the wrapper, preserving stateful DoFn validation.
• Preserve input and output type checking, output-shape validation, and transform error labeling.
• Add regression tests for wrapper application, error reporting, result preservation, and stateful DoFn compatibility.

Tests

• apache_beam/typehints/typecheck_test.py: 18 passed.
• apache_beam/transforms/ptransform_test.py: 202 passed, 4 skipped.

Performance

Compared the original and modified implementations using Beam's existing runtime_type_check_microbenchmark.

Configuration: 20 DoFns and three timing repetitions per configuration.

Elements	Simple types (before → after)	Nested types (before → after)
501	3.22s → 3.06s	3.90s → 3.41s
1,001	5.44s → 5.39s	6.44s → 6.07s
1,501	7.57s → 7.39s	8.93s → 8.83s
2,001	9.91s → 9.46s	11.84s → 11.22s

At 2,001 elements, the measured runtime was approximately 4.5% lower for simple type hints and 5.2% lower for nested type hints in this local run.

These measurements are preliminary and may vary by environment. They demonstrate the results of this benchmark configuration, not a guaranteed improvement for all pipelines.

[github-actions]

Assigning reviewers:

R: @claudevdm for label python.

This pull request likely touches a core component ("core" label). Please review with scrutiny.

Note: If you would like to opt out of this review, comment assign to next reviewer.

Available commands:

• stop reviewer notifications - opt out of the automated review tooling
• remind me after tests pass - tag the comment author after tests pass
• waiting on author - shift the attention set back to the author (any comment or push by the author will return the attention set to the reviewers)

The PR bot will only process comments in the main thread (not review comments).

[codecov]

Codecov Report

❌ Patch coverage is 94.11765% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 79.90%. Comparing base (76f6e6d) to head (72d2196).
⚠️ Report is 219 commits behind head on master.

Files with missing lines	Patch %	Lines
sdks/python/apache_beam/typehints/typecheck.py	94.11%	2 Missing ⚠️

Additional details and impacted files

@@              Coverage Diff              @@
##             master   #40501       +/-   ##
=============================================
+ Coverage     56.16%   79.90%   +23.73%     
=============================================
  Files          1124      488      -636     
  Lines        178336    82542    -95794     
  Branches       1489        0     -1489     
=============================================
- Hits         100170    65957    -34213     
+ Misses        75645    16585    -59060     
+ Partials       2521        0     -2521     

Flag	Coverage Δ
go	?
java	?
python	79.90% <94.11%> (+0.35%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[claudevdm]

I think this looks fine, but not sure it adresses @damccorm comment from previous #39206 (comment)

[claudevdm]

2.75 has been released already. Please move this to future release

[Mazen050]

Done

[claudevdm]

R: @damccorm do you have any concerns with this? I saw a previous attempt you asked for changes #39206 (comment)

[github-actions]

Stopping reviewer notifications for this pull request: review requested by someone other than the bot, ceding control. If you'd like to restart, comment assign set of reviewers

[damccorm]

R: @damccorm do you have any concerns with this? I saw a previous attempt you asked for changes #39206 (comment)

Reviewing now, I think this one is ok - it is not making other non-functional changes, there are just some changes to make sure we correctly preserve additional variables we're passing into the function

Guessing AI caught this distinction both times? But I think it is fine to take it

[claudevdm]

/gemini review

[github-actions]

📋 Review Summary

This PR consolidates OutputCheckWrapperDoFn and TypeCheckWrapperDoFn into a single RuntimeTypeCheckWrapperDoFn to eliminate double DoFn wrapper delegation when runtime_type_check is enabled, and avoids caching a bound method on the wrapper instance so userstate.get_dofn_specs does not see duplicate state/timer specs after serialization. There is one bug where non-generator output type violations get wrapped with _add_label twice, duplicating the Runtime type violation detected within ParDo(...): prefix in the error message.

🔍 General Feedback

• The approach of unwrapping process_fn.__func__ so no types.MethodType attribute is stored on RuntimeTypeCheckWrapperDoFn.__dict__ cleanly prevents userstate.get_dofn_specs from inspecting _process_fn as a separate DoFn method.
• Check the error message formatting on eager (non-generator) output type-check failures, as _type_check_result is currently invoked inside a try ... except TypeCheckError block while type_check_output also wraps TypeCheckError with self._add_label.

---

💡 Tip: Comment /gemini review on this pull request to trigger a new Gemini code review.

[github-actions]

🟠 High: self._type_check_result(...) is called inside the try ... except TypeCheckError as e: raise self._add_label(e) block in both process() (line 133) and wrapper() (line 114), while type_check_output() inside _type_check_result() (lines 174-177) also catches TypeCheckError and raises self._add_label(e).

When process() or wrapper() returns a non-generator iterable (e.g., a list or tuple, which includes beam.Map and list-returning FlatMap/ParDo transforms), for o in transform_results: type_check_output(o) runs synchronously inside _type_check_result. Any output TypeCheckError is wrapped by self._add_label(e) inside type_check_output and then caught and wrapped by self._add_label(e) a second time in process()/wrapper(), producing a duplicated prefix:
Runtime type violation detected within ParDo(<label>): Runtime type violation detected within ParDo(<label>): ...

Moving self._type_check_result(result) into the else: branch (in both process and wrapper) ensures self._add_label(e) is applied only once for both generator and non-generator outputs.

Suggested change

	result = self.dofn.process(*args, **kwargs)

[Mazen050]

I tested this change with the regression test I just pushed. It fails because the TypeCheckError raised in else bypasses the preceding except, so the ParDo label is lost. I'll keep _type_check_result() inside try.

[github-actions]

🟠 High: Pair with moving self._type_check_result out of the try block on line 133 (and similarly in wrapper on lines 114 and 118) so output TypeCheckErrors are not labeled twice.

Suggested change

	else:
	return self._check_type(self._type_check_result(result))

[Mazen050]

The same issue applies here: exceptions raised in else bypass the preceding except. Since type_check_output() doesn't call _add_label() itself, I couldn't reproduce the double-labeling rationale in my current implementation.