test(cuda.core): pop the context when the init_cuda leftover warning aborts setup

[juenglin]

Description

Follow-up to #3053. The successful path is unchanged; this makes init_cuda clean up when its setup aborts.

pytest.ini escalates UnreleasedMemoryResourceWarning to an error, and init_cuda raises it during setup, after set_current(). The raise used to happen before the fixture's try, so the finally never ran and the context stayed current. Now everything after set_current() runs inside the try: the CUDA_CORE_TEST_BLOCKING_SYNC flag, the leftover check, and the live-count snapshot. If the warning fires, teardown still synchronizes and pops the context, and it skips the collect comparison that needs the snapshot. A failing set_current() still raises before the try, because nothing has been changed yet that would need undoing.

test_escalated_setup_warning_still_pops_the_context drives that path. It sets the error filter itself, so it does not depend on pytest.ini or command-line warning options. It fails against the previous conftest.py.

The comment on the VA-reservation counter in virtual_memory.cpp now states that memory-pool VA windows are not included: owned pools have their own count, and the device default pool is a borrowed reference that is not counted.

Checklist

• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[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.

[juenglin]

/ok to test d0bce1a

[github-actions]

Doc Preview CI
Preview removed because the pull request was closed or merged.

[juenglin]

/ok to test a32461f

[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: d77f0dd2-4a00-418f-a1ce-6e5cb66b744b

---

📥 Commits

Reviewing files that changed from the base of the PR and between 632bfc4 and a32461f.

---

📒 Files selected for processing (3)

• cuda_core/cuda/core/_cpp/rt/virtual_memory.cpp
• cuda_core/tests/conftest.py
• cuda_core/tests/memory/test_init_cuda_collect_gate.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
  • CUDA test setup now cleans up the active device context when initialization fails, helping prevent test failures from leaving the context in an inconsistent state.
• Other Changes
  • Clarified how live virtual address reservations are counted. No user-facing functionality changes are included in this update.

Walkthrough

The change clarifies which VA reservations the live count includes. It also ensures fixture teardown runs when setup raises and adds a test that checks the CUDA context is cleared.

Changes

CUDA resource accounting and fixture cleanup

Layer / File(s)	Summary
Resource counts and fixture cleanup cuda_core/cuda/core/_cpp/rt/virtual_memory.cpp, cuda_core/tests/conftest.py, cuda_core/tests/memory/test_init_cuda_collect_gate.py	The comment clarifies which VA reservations the live count includes. The fixture skips resource-count checks if setup fails before recording a snapshot, while still synchronizing and unsetting the current device. The test checks that the context is cleared when a setup warning raises as an error.

---

Priority: ⚪ Not assessed

Change: Bug fix

Merge Risk: ⚪ Minimal · up to a3246

No actionable issue is identified; the PR is mergeable after normal checks.

---

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

✨ Finishing Touches 💡 1

🛠️ Fix failing CI checks 💡

• Commit to this branch
• Create a new PR