chore: defensive improvements for robustness and best practices

[zhengdaqi]

Summary

Five defensive improvements that harden edge-case handling and follow Python best practices. All changes are non-breaking and only affect error paths or rare edge cases.

1. Fix potential UnboundLocalError in write_memory_output()

benchmark_name was only assigned inside if isinstance(benchmark_cfg, dict). If benchmark_cfg is not a dict, the variable is used uninitialized on the next line. Fixed by initializing benchmark_name = "" before the conditional.

2. Fix __main__ entry point

main() is a synchronous function that internally calls asyncio.run() for async subcommands. The __main__ block wrapped it in another asyncio.run(main()), which would raise ValueError after main() completes (since it returns None, not a coroutine).

3. Fix benchmark key mapping producing ragged lists

When some data records are missing a mapped key, the original list comprehension [item[key] for item in data if key in item] skips those records entirely. This produces lists of different lengths for different aliases, breaking downstream zip-based processing. Changed to item.get(key) to maintain alignment (returns None for missing keys).

4. Replace bare except: with except Exception:

Four bare except: clauses in custom.py catch SystemExit and KeyboardInterrupt in addition to regular exceptions. This can prevent clean process termination. Changed to except Exception: per PEP 8 guidelines.

5. Fix file handle leak in server.py

yaml.safe_dump(data, out_path.open("w")) creates a file handle without a with statement. If safe_dump raises an exception, the handle may not be closed promptly. Wrapped with with statement for proper resource cleanup.

Test Plan

• Run pytest to verify no regressions
• These are defensive fixes for edge cases; primary functionality is unaffected

Made with Cursor