cache: write_cache is not atomic; interrupted write leaves a corrupt cache file

[eneskemalergin]

Summary

path.write_text(json.dumps(data), encoding="utf-8") is not atomic. If the process is interrupted (SIGINT, SIGKILL, power loss) mid-write, the file is left truncated or malformed.

Impact

On the next run, read_cache detects the JSONDecodeError, deletes the file, and returns None — the network fetch is retried correctly. No data loss or incorrect result; only an unnecessary extra API call.

Proposed fix

Write-then-rename pattern (atomic on POSIX):

import os

tmp = path.with_suffix(".tmp")
tmp.write_text(json.dumps(data, ensure_ascii=False), encoding="utf-8")
os.replace(tmp, path)  # atomic rename on POSIX; best-effort on Windows

Notes

Low priority for Phase 1 since recovery is already in place. Recommended for a hardening pass before Phase 2 when batch/concurrent auditing is considered.