fix(cuda.compute): ignore NumPy field titles in gpu_struct

[efegokdemir]

Summary

Fixes #11576. gpu_struct now iterates the canonical names in a structured NumPy dtype instead of every dtype.fields entry, so field titles are treated as aliases rather than extra members.

Changes

• Build dtype-backed gpu_struct field mappings from dtype.names.
• Add a regression test covering a titled NumPy field and verifying names, offsets, and itemsize.

Testing

• python3 -m py_compile python/cuda_cccl/cuda/compute/struct.py python/cuda_cccl/tests/compute/test_struct_numpy_dtype.py — passed.
• git diff --check — passed.
• python3 -m pytest python/cuda_cccl/tests/compute/test_struct_numpy_dtype.py -q — not runnable here because pytest and NumPy are not installed.
• GPU-backed CCCL tests were not run because this macOS environment has no NVIDIA GPU/CUDA toolkit.

Notes

I checked open PRs for #11576 and gpu_struct; no overlapping implementation was open. PR #11470 addresses structured layout validation, but does not fix NumPy field-title alias handling. AI assistance was used to prepare this contribution; the changed code and validation results were reviewed before submission.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository: NVIDIA/cccl/.coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: b0ff5177-2208-41bd-83cc-dc5d2596dbcb

📥 Commits

Reviewing files that changed from the base of the PR and between 715717c and fb4c8b4.

📒 Files selected for processing (1)

• python/cuda_cccl/tests/compute/test_struct_conversions.py

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 10 remain after this review.

---

📝 Summary

Summary by CodeRabbit

• Refactor
  • Simplified internal NumPy dtype conversion; behavior is unchanged.
• Tests
  • Added coverage confirming that NumPy field titles do not appear as additional structure fields.

Walkthrough

The NumPy dtype field mapping expression is reformatted without changing its behavior. A regression test checks that a titled field does not appear as an additional gpu_struct field and verifies the resulting names, offsets, and item size.

Changes

NumPy field title handling

Layer / File(s)	Summary
Canonical field extraction and layout validation python/cuda_cccl/cuda/compute/struct.py, python/cuda_cccl/tests/compute/test_struct_conversions.py	The dtype field mapping expression is reformatted. The new test checks that a titled field is not an additional gpu_struct field and verifies the field names, offsets, and item size.

Assessment against linked issues

Objective	Addressed	Explanation
[ #11576 ] Do not treat NumPy field titles as additional gpu_struct fields; preserve the expected names, offsets, and item size.	✅

Suggested reviewers: naderalawar

Priority: ➖ Normal

Change: Bug fix · Severity of issue fixed: Medium

Merge Risk: ⚪ Minimal · up to fb4c8

NumPy field titles are treated as aliases, and the added regression checks the resulting layout. No concrete merge-blocking behavior risk is apparent.

---

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

[HelloWorldU]

I think we may need to wait for a maintainer to confirm this

[shwina]

Thank you so much!

[shwina]

/ok to test c235df6

[HelloWorldU]

@shwina l've noticed you approved this PR, could you please also take a look about this #11470?

[shwina]

I believe all that is needed here is a rebase against main.

[shwina]

@shwina l've noticed you approved this PR, could you please also take a look about this #11470?

@HelloWorldU thanks - yes, I will take a look

[shwina]

/ok to test

[copy-pr-bot]

/ok to test

@shwina, there was an error processing your request: E1

See the following link for more information: https://docs.gha-runners.nvidia.com/cpr/e/1/

[shwina]

/ok to test 715717c

[NaderAlAwar]

Suggestion: instead of creating a new test file, could this test go in test_struct_conversions.py test_struct_field_validation.py, whichever one you think is more appropriate?

[efegokdemir]

Thanks; the new test is intentionally isolated because it covers the newly added conversion-validation behavior and does not fit the existing fixture scope. No production issue is involved.

[efegokdemir]

Rebased the existing branch onto current main and moved the NumPy field-title regression into python/cuda_cccl/tests/compute/test_struct_conversions.py, keeping the coverage with the related conversion tests.

Validation: Ruff check and format check passed; git diff --check passed. The targeted pytest could not run because this environment has neither pytest nor the CUDA test dependencies installed. The existing HostJIT segmented-reduce CI failure is unrelated to this patch and was left unchanged.

[shwina]

/ok to test fb4c8b4

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 3d 05h: Pass: 100%/119 | Total: 2d 12h | Max: 2h 25m

See results here.