Asset sync test stability under parallel runs

• Status: Accepted
• Date: 2025-12-24
• Context:
  • cargo llvm-cov runs tests in parallel and surfaced a flaky asset_sync test.
  • The temp directory helper used timestamp-based names that could collide under parallel execution.
  • CI requires just ci (including coverage) to pass reliably without intermittent failures.
• Decision:
  • Replace the time-based temp directory naming with a process id + atomic counter.
  • Retry on AlreadyExists to ensure unique per-test directories without new dependencies.
• Consequences:
  • Asset sync tests are deterministic under parallel runners and coverage instrumentation.
  • No new crates or runtime behavior changes.
• Follow-up:
  • None.

Motivation

• Remove flaky coverage failures caused by temporary directory collisions in asset_sync tests.

Design notes

• Use a static AtomicUsize counter plus std::process::id() to generate unique temp roots.
• Loop on AlreadyExists without introducing external dependencies.

Test coverage summary

• just ci (fmt, lint, udeps, audit, deny, ui-build, test, cov).

Observability updates

• None.

Risk & rollback plan

• Risk: low; change is test-only.
• Rollback: revert the temp directory helper to its previous implementation.

Dependency rationale

• No new dependencies.