[ProtonDev-sys]

Why

History saves used a fixed history.tsv.tmp neighbor. A successful save could consume a pre-existing file at that path, and a failed save could delete it. Two running instances also shared the staging file. CSV staging appended a GUID to the full destination filename, causing otherwise valid long filenames to exceed Windows' component limit.

Changes

• Share an owned, buffered atomic-file writer between history and CSV export. Each attempt creates an exclusive, short .taskmgr-{GUID}.tmp sibling.
• Stream history fields through a bounded 64 KiB buffer, preserving the existing UTF-8 escaping, LF records and full unsigned timestamp range.
• Check writes, flush and close before replacing the destination. Preserve the first error and clean up only a staging file this attempt successfully created.
• Keep large CSV writes bounded without copying an entire large field through the small buffer.

Verification

The previous history writer was extracted and reproduced consuming a sentinel neighbor on both successful and failed saves. The replacement passes local strict-warning GCC/UBSan checks: 31 history fixture checks and 15 atomic-file boundary/failure checks. Independent review covered ownership, destruction order, write failure handling and exact history serialization.

New native fixtures exercise pre-existing neighbors, locked destinations, directory and missing-parent failures, abandoned staging cleanup, two simultaneous staging files, concurrent complete-snapshot saves, full-width unsigned timestamps and a valid long CSV filename. Concurrent saves preserve one complete winning snapshot; this does not add history merging between instances.

The corrected head passed Windows verification run 38209554455: 12/12 CTests, all 21 suite cases, Python tests on both platforms, release guards, production checks and packaging. The initial diagnostic-only shadowing warning was corrected without relaxing /WX.

The retained artifact matches GitHub SHA-256 02fd1620c36853d68fa7c95ebdd49068e6c582080668f46b629b9a4881280047. All 31 diagnostic JSON reports pass. All 25 UI reports (7 CTest and 18 suite) have sortingPassed=true and an empty failure field; the history and CSV fixtures run unconditionally in that gate. This verifies the long filename, complete concurrent snapshots, exact serialized bytes, destination preservation and cleanup on Windows. Individual successful fixture-assertion counts are not emitted, so the source labels are not presented as additional independent runtime test counts.

This patch is based on the fully verified combined source merged in #39; the paired performance experiment was not repeated for this persistence-only change.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 4ea2f834-7b9d-444c-b1f7-f552d6bbe04c

---

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

---

_{Comment @coderabbitai help to get the list of available commands.}