Coerce stream_ptr to int in view_as_dlpack

[mdboom]

Coerce stream_ptr to int in view_as_dlpack

Fixes the nightly numba-cuda failure that started after #2902
(https://github.com/NVIDIA/cuda-python/actions/runs/39190777771/job/117862954433).

Problem

#2902 set Cython's annotation_typing=False. Until then, the
stream_ptr: int | None annotations on StridedMemoryView made Cython convert
any int-convertible object to a plain int on entry. numba-cuda passes its
CUstream(0) wrapper, which is truthy and converts to 0. Without the
annotation typing, the wrapper reached view_as_dlpack unchanged, and
int(stream_ptr) if stream_ptr else None took the truthy branch. The result
was numpy_array.__dlpack__(stream=0) instead of stream=None, and NumPy
raised RuntimeError: NumPy only supports stream=None.

The stream_ptr == -1 opt-out check had the same problem: a wrapper around
-1 was never recognized and was forwarded as stream=-1.

Fix

view_as_dlpack now converts stream_ptr with int() once, up front, so every
later check sees an int. This matches what view_as_cai already does. The
fix is explicit rather than a per-function @cython.annotation_typing(True),
so it does not depend on the build-wide directive.

Audit for similar regressions from #2902

I went through every int / int | None annotated parameter in cuda_core
that #2902 un-typed. stream_ptr was the only place that makes a decision on
the raw object before converting it. The other sites store the value, check it
with is None, or convert it directly to a C type (<CUdeviceptr>dptr,
size_t, unsigned int, ...), and those conversions accept any __index__
object with or without annotation typing.

Test

test_cpu_dlpack_accepts_int_convertible_stream_ptr in
cuda_core/tests/test_utils_dlpack.py needs neither numba nor a GPU, so it
runs in the regular (non-nightly) suite.

• A small stand-in stream wrapper is int-convertible with no __bool__, so it
  is always truthy, like CUstream.
• A CPU DLPack source wraps a NumPy array and records the stream argument it
  receives in __dlpack__. NumPy then enforces stream=None as in the
  nightly failure.
• The test is parametrized over from_dlpack, from_any_interface and the
  deprecated constructor, and over wrapper values 0 and -1. It asserts that
  the exporter received stream=None and that a view was produced.

All six cases fail without the fix and pass with it.

🤖 Generated with Claude Code

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Repository: NVIDIA/cuda-python/.coderabbit.yaml
• Review profile: CHILL
• Plan: Enterprise
• Run ID: bc26c87c-7eef-4c47-bfa8-0d9abcb98b02

📥 Commits

Reviewing files that changed from the base of the PR and between 9d903cd and c72fa87.

📒 Files selected for processing: 2

• cuda_core/cuda/core/_memoryview.pyx
• cuda_core/tests/test_utils_dlpack.py

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 11 remain after this review.

📝 Summary

Summary by CodeRabbit

• Bug Fixes
  • Improved compatibility when creating views from DLPack objects using stream handles that can be converted to integers.
  • Applied consistent stream-handle handling across the available DLPack import methods, including for CPU-backed objects.

view_as_dlpack now converts each non-None stream pointer to an integer before validation and DLPack export. Tests cover handles that convert to 0 or -1 across three view-construction paths.

Changes: DLPack stream handling

Layer / File(s)	Summary
Stream handle normalization and export tests cuda_core/cuda/core/_memoryview.pyx, cuda_core/tests/test_utils_dlpack.py	view_as_dlpack converts non-None stream handles before its checks and DLPack call. Tests verify that the CPU exporter receives stream=None for handles that convert to 0 or -1.

Suggested reviewers: rluo8

Priority: ➖ Normal

Change: Bug fix

Merge Risk: ⚪ Minimal · up to c72fa

The change addresses wrapped stream handles and adds coverage for both sentinel values across the construction paths. No specific remaining issue warrants delaying the merge.

- [ ] Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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

[github-actions]

Doc Preview CI
🚀 View preview at https://nvidia.github.io/cuda-python/pr-preview/pr-3154/
https://nvidia.github.io/cuda-python/pr-preview/pr-3154/cuda-core/
https://nvidia.github.io/cuda-python/pr-preview/pr-3154/cuda-bindings/
https://nvidia.github.io/cuda-python/pr-preview/pr-3154/cuda-pathfinder/
Preview will be ready when the GitHub Pages deployment is complete.

[jpascucci-nv]

It took me a little while to figure out. if stream_ptr = int(CUstream(0)) ends up 0, it's going to trigger the else of line 1065, so it calls it with None. Ok.