[BUG]: test_keyboard_interrupt_during_a_fill_reaches_the_user fails when the test process inherits SIGINT as ignored #3213

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this bug and that I agree to the Code of Conduct

Type of Bug

Something else (test-only failure)

Component

cuda.core

Describe the bug

cuda_core/tests/test_driver_table.py::test_keyboard_interrupt_during_a_fill_reaches_the_user failed on every run (10/10) when pytest was started as a background job of a non-interactive shell:

AssertionError: NOT INTERRUPTED
  WARNING: cuda-bindings could not load the NVRTC: KeyboardInterrupt

A background job of a non-interactive shell inherits SIGINT as SIG_IGN (SigIgn: 0x7 in /proc/<pid>/status). Python installs its KeyboardInterrupt handler only when SIGINT is not ignored at startup, and PyErr_SetInterruptEx does nothing for an ignored signal. The interrupt that the fill re-arms is therefore dropped, and the child prints NOT INTERRUPTED. The subprocess that _run_code starts inherits the same disposition.

How to Reproduce

bash -c '( pytest cuda_core/tests/test_driver_table.py -k keyboard_interrupt & ); wait'

Expected behavior

The test passes however pytest was started; the interrupt it re-arms reaches the child as KeyboardInterrupt.

Proposed fix

Have _INTERRUPTED_CHILD call signal.signal(signal.SIGINT, signal.default_int_handler) before it patches the loader, so the test does not depend on how the runner was started. This is a 6-line change; I can open a PR.

The fix was written with Claude Opus 5.5 and reviewed by the contributor.

Environment

GPU: B300 (cc 10.3, x86_64) | driver 610.57.04 | CUDA 13.4.2 (local CTK) | Python 3.12 | cuda-python 2da9c65a