[john-the-dev]

Closes

Resolves the remaining entry in shell-ci-known-failures.txt (tracked in PR #1948).

Summary

Two independent root causes for CI failures in tests/sync-workspace.test.sh:

1. Lock contention (tests 8, 9, 10)
The default lock dir /tmp/sync-workspace.lock.d is shared between test invocations and the live sync-workspace cron. When the cron fires mid-test it steals the lock; subsequent test calls exit immediately with "another instance is running, skipping" causing assertion failures.

Fix: add SUTANDO_SYNC_LOCK_DIR env hook to sync-workspace.sh (same pattern as SUTANDO_WS_ID_OVERRIDE, SUTANDO_TEST_MODE) and set it to a temp-dir-scoped path in the test COMMON_ENV.

2. Test 27 init guard (silent set-e abort)
Test 27 creates a pre-wsId workspace (flat host/mighost branch) to test the migration path. But it doesn't write the .git/info/exclude marker that _assert_sync_initialized() requires. The script exits 1, set -euo pipefail in the test aborts silently (no FAIL printed), and tests 27–28 never run — so the test exits with fail=1 from this abort.

Fix: write # Generated by sync-workspace.sh — do not edit by hand. to .git/info/exclude in the test 27 setup subshell. A real pre-wsId workspace would have this marker since sync-workspace.sh has always written it.

Checklist

• Confirmed no other open PR closes the same issue
• Git author email matches CLA-signed email
• Single concern per PR — two fixes but both are for the same test file / same CI entry
• Confirmed failures reproduce on upstream/main before fix
• All 88 assertions pass locally after fix

Test plan

bash tests/sync-workspace.test.sh
# Total: 88 — pass: 88, fail: 0

Run several times to confirm no timing-dependent failures (lock isolation eliminates the race).

[github-actions]

Coverage Gate

✅ No Python changes — nothing to gate (bar: 95% on changed lines).

[john-the-dev]

Related: PR #1997 (fix(sync-workspace): allow pre-wsId flat branch past init guard) takes a complementary approach to the test 27 fix here — it adds a third sentinel in _assert_sync_initialized() (flat host/<hostname> HEAD passes) rather than writing the exclude marker in the test setup. Both work; they fix the same test 27 failure via different mechanisms.

The lock contention fix (SUTANDO_SYNC_LOCK_DIR env hook in sync-workspace.sh + isolated lock dir in test COMMON_ENV) is unique to this PR and not covered by #1997.

Suggested merge order: #1997 first, then this PR — or drop the test-27 exclude-marker hunk from this PR if #1997 lands first (the lock contention fix stands on its own).

[qingyun-wu]

Reviewed — LGTM. Parameterizing LOCK_DIR (default unchanged) is the minimal correct fix for test cross-talk, and the test-27 change is right: a legitimately-initialized pre-wsId workspace would carry the "Generated by sync-workspace.sh" marker in .git/info/exclude, so writing it in the fixture is simulating reality, not dodging the guard. CI green.

[github-actions]

@cla-assistant check