[0z5a]

Summary

check-error-example-status.sh runs under set -euo pipefail, and its contains() piped printf into grep -qx. grep -q exits at the first match. printf can then die of SIGPIPE, and pipefail turns the match into a miss.

On a GH200 (aarch64) host, just check-guards failed intermittently with:

error: error_enum_shared_pointer_layout is not in ERROR_EXAMPLES in smoketest.sh

That name is in the array. The name sits early in the list, so grep exits while printf still has lines to write.

This PR compares in a bash loop instead.

Changes

• cuda-oxide/scripts/check-error-example-status.sh: contains() compares in a bash loop instead of piping printf into grep -qx.

Testing

Host	Original guard, failures	Fixed guard, failures
GH200 (aarch64), 50 runs each	13	0
B300 (x86_64), 20 runs of the original	0	—

Removing error_enum_shared_pointer_layout from ERROR_EXAMPLES still makes the fixed guard fail with that message.

GH200 (cc 9.0, aarch64) | driver 595.58 | CUDA 13.2 | rustc 1.99.1 | cuda-rust fc747196

check-oxide-artifacts-parity.sh:55 has the same grep … | grep -q shape under pipefail, but its producer writes at most four lines, and I did not see it fail.

• just -f cuda-oxide/Justfile check passes -- fmt-check, clippy and check-intrinsics pass on GH200; the test stage did not fit on that host's disk, and this change touches only the guard script, run directly above
• cargo oxide run <example> -- not applicable
• New example added (if applicable) -- not applicable

Checklist

• All commits signed off (git commit -s)
• SPDX headers on new source files -- no new files

[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.