fix: amdsmi import when libamd_smi is missing (macOS / no ROCm)

[davidberenstein1957]

Summary

On systems where the amdsmi Python package is installed but ROCm / libamd_smi.so is not available (common on macOS), importing CodeCarbon could crash with KeyError during import amdsmi inside gpu_amd.py, breaking the CLI and any code path that loads GPU support.

This change broadens the existing AttributeError handler to also catch OSError and KeyError, sets amdsmi = None and AMDSMI_AVAILABLE = False, and logs a clear warning so AMD GPU metrics are skipped instead of failing import.

Test plan

• On macOS (or any host without ROCm) with amdsmi installed: codecarbon --help and python -c "import codecarbon" succeed with a warning, no traceback.
• On a ROCm system with working amdsmi: AMD GPU metrics still work as before.

Made with Cursor

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 88.90%. Comparing base (afa3ac8) to head (ba1565f).

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1201   +/-   ##
=======================================
  Coverage   88.90%   88.90%           
=======================================
  Files          45       45           
  Lines        4301     4301           
=======================================
  Hits         3824     3824           
  Misses        477      477           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[benoit-cty]

@copilot can you improve coverage by updating the tests ?

[Copilot]

@copilot can you improve coverage by updating the tests ?

Added a new TestAmdsmiImportFailures class (commit 241e498) with 4 tests covering the import-time exception handling in gpu_amd.py:

• test_oserror_sets_amdsmi_unavailable — verifies OSError (e.g. missing libamd_smi.so) sets amdsmi=None and AMDSMI_AVAILABLE=False
• test_keyerror_sets_amdsmi_unavailable — same for KeyError (macOS / no ROCm)
• test_attributeerror_sets_amdsmi_unavailable — same for AttributeError (pre-existing case)
• test_import_failure_logs_warning — verifies all three failure types emit a logger warning

Each test uses builtins.__import__ patching + sys.modules manipulation to simulate a fresh module import that raises the target exception, then restores the original state.

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• get.geojs.io
  • Triggering command: /usr/bin/python python -m pytest tests/test_gpu_amd.py tests/test_gpu.py tests/test_gpu_nvidia.py tests/test_unsupported_gpu.py -v ader (dns block)
• ipinfo.io
  • Triggering command: /usr/bin/python python -m pytest tests/test_gpu_amd.py tests/test_gpu.py tests/test_gpu_nvidia.py tests/test_unsupported_gpu.py -v ader (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)