test(helpers): keep the EGL device probe from importing pyglet.gl

[Andy-Jost]

Summary

Follow-up to #3020. The narrowed catch imports pyglet.gl.lib.MissingFunctionException inside select_headless_egl_device_for_cuda, which runs before _configure_pyglet_headless() sets pyglet.options["headless_device"]. Importing pyglet.gl creates pyglet's shadow window (pyglet/gl/__init__.py imports pyglet.window at the bottom). In headless mode HeadlessDisplay.__init__ reads headless_device at that moment, and pyglet.display.get_display() returns the cached display afterwards, so the later headless_device assignment has no effect. The headless tests run in the shadow window's context, so on the multi-GPU systems #2865 fixed, the GL context lands on EGL device 0 again and cuGraphicsGLRegister* fails with the original CUDA_ERROR_INVALID_DEVICE. Before #3020 the probe imported only pyglet.libs.egl, which does not import pyglet.gl.

CI cannot detect this: no runner has an EGL device order that differs from the CUDA device order.

Changes

• cuda_python_test_helpers/graphics.py: drop the pyglet.gl.lib import from the probe and match MissingFunctionException by module and name through a small helper, the same pattern is_gl_context_unavailable uses. Everything else still propagates. The module docstring now states the rule.
• cuda_core/tests/test_graphics.py: add test_egl_device_probe_does_not_import_pyglet_gl, which runs the probe in a subprocess with headless set and asserts pyglet.gl is not in sys.modules.

Verification

Reproducible without a GPU on any Linux host with a libEGL (shown here with Mesa):

import pyglet
pyglet.options["headless"] = True
pyglet.options["headless_device"] = 99  # sentinel
from pyglet.gl.lib import MissingFunctionException
# ValueError: Invalid EGL device id: 99, raised from HeadlessDisplay.__init__ during the import

With the default headless_device, the import leaves a ShadowWindow and a cached HeadlessDisplay behind; the fixed probe leaves pyglet.gl unimported. The multi-GPU case itself was not re-run on hardware with a divergent device order.

Related Work

• [cuda.core] test_graphics.py fails on headless multi-GPU Linux: EGL device does not match current CUDA device #2864, cuda.core: match headless EGL device to the current CUDA device in test_graphics.py #2865 (the fix this restores)
• test(helpers): narrow EGL device probe catch to MissingFunctionException #3020 (the change that introduced the import)

🤖 Generated with Claude Code

[github-actions]

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

[coderabbitai]

📝 Summary

Summary by CodeRabbit

• Bug Fixes
  • Improved headless EGL device detection so it can handle missing EGL functions without importing graphics code that may create an unintended shadow window.
  • Other EGL query errors continue to be reported rather than silently ignored.
• Tests
  • Added coverage to verify the headless EGL probe runs without importing the graphics module that can trigger the shadow window.

Walkthrough

The EGL device probe no longer imports pyglet.gl to identify pyglet’s missing-function exception. A Linux/EGL-gated subprocess test checks that the probe leaves pyglet.gl unloaded.

Changes

Headless EGL probe

Layer / File(s)	Summary
Exception handling and import-safety test cuda_python_test_helpers/cuda_python_test_helpers/graphics.py, cuda_core/tests/test_graphics.py	The probe recognizes MissingFunctionException by its type metadata, returns None for that exception, and re-raises other exceptions. A subprocess test probes EGL device 0 and checks that pyglet.gl was not imported.

Suggested reviewers: rparolin

Priority: ⬇️ Low

Change: Bug fix

Merge Risk: 🔵 Low · up to 6dd84

The EGL regression test can stall on a system where device probing hangs. Add a subprocess timeout before merging, or accept this bounded test-run risk.

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

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

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

• Configuration used: Repository: NVIDIA/cuda-python/.coderabbit.yaml
• Review profile: CHILL
• Plan: Enterprise
• Run ID: cfa2f253-9dee-4a61-9ef2-0aa48edc8019

📥 Commits

Reviewing files that changed from the base of the PR and between ea58e1b and 6dd84a1.

📒 Files selected for processing (2)

• cuda_core/tests/test_graphics.py
• cuda_python_test_helpers/cuda_python_test_helpers/graphics.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.

[coderabbitai]

🩺 Stability & Availability | 🟡 Minor | ⚡ Quick win

suggestion: Bound the EGL probe subprocess. If EGL initialization or the probe blocks, subprocess.run waits indefinitely and stalls the test run. Set a timeout and report subprocess.TimeoutExpired as a test failure. Based on learnings: flag process-spawning calls without a timeout because a hung child can block indefinitely.

🧰 Tools

🪛 ast-grep (0.45.3)

[error] 104-106: Command coming from incoming request
Context: subprocess.run( # noqa: S603 - trusted argv: this interpreter
[sys.executable, "-c", code], capture_output=True, text=True, env=env
)
Note: [CWE-78] Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection').

(subprocess-from-request)

Source: Learnings