ci: smart-extractor-branches.mjs test failing since PR #669 bulkStore refactor

[jlin53882]

Issue: smart-extractor-branches.mjs test failing since PR #669 bulkStore refactor

Severity: High — CI on master branch is broken since PR #669

Affected: test/smart-extractor-branches.mjs line 497

---

Symptom

core-regression CI job has been failing since PR #669 (a8bb8ec):

AssertionError [ERR_ASSERTION]: The expression evaluated to a falsy value:
  assert.ok(
    multiRoundResult.logs.some((entry) => entry[1].includes("created [preferences] 饮�??�好：�?龙茶")),
  );
    at file:///home/runner/work/memory-lancedb-pro/memory-lancedb-pro/test/smart-extractor-branches.mjs:497

Last passing CI: cf782a2 (Apr 20 02:57)
First failing CI: a8bb8ec (Apr 20 11:20) — PR #669 merged at 11:20:20Z

---

Root Cause

PR #669 refactored src/smart-extractor.ts to use bulkStore() instead of individual store() calls for batch writes. The test smart-extractor-branches.mjs was not updated to reflect this change.

The assertion at line 497 checks for the log message:

"created [preferences] 饮�好：�龙茶"

This message is logged inside storeCandidate() (the old single-write path), but after PR #669, the multi-round scenario now uses bulkStore() which does NOT call storeCandidate() — therefore this specific log line is never emitted in the batch path.

Evidence:

• cf782a2..c52ab2b diff: smart-extractor.ts changes from await this.store.store(...) → createEntries?.push(this.buildStoreEntry(...)) then await this.store.bulkStore(createEntries)
• smart-extractor-branches.mjs was NOT modified in either PR feat: bulkStore batch write + SmartExtractor optimization (Issue #665, #666) #669 or fix: Issue #598 Phase 3 - EmbeddingCache LRU semantics #658
• The test file is identical at cf782a2 and c52ab2b

---

What Needs to Happen

Option A — Fix the test to match new behavior:

1. Update smart-extractor-branches.mjs to track bulkStore calls instead of expecting created [preferences] log messages
2. Or: add a test mode that still logs individual creates during batch operations

Option B — Restore per-item logging in bulkStore:

1. Have bulkStore emit per-entry log lines similar to storeCandidate()
2. Less desirable — defeats the purpose of batch optimization

---

Related Files

• Failing test: test/smart-extractor-branches.mjs (unchanged since cf782a2)
• Breaking commit: a8bb8ec (PR feat: bulkStore batch write + SmartExtractor optimization (Issue #665, #666) #669)
• Broken since: 2026-04-20 11:20 UTC

---

Note

This is NOT caused by PR #678 (fix for Issue #675/#676). PR #678 is mergeable and passes its own tests. This is a pre-existing upstream issue.

[jlin53882]

Thanks for the detailed analysis! I'd like to confirm the fix approach before implementing.

Based on your comment, the preferred fix is:

Remove lines 497-498 in test/smart-extractor-branches.mjs (the created [preferences] log assertion).

The existing assertions at lines 492-493 already verify the correct outcome:

• assert.equal(multiRoundResult.entries.length, 1) → confirms bulkStore wrote something
• assert.equal(multiRoundResult.entries[0].text, "饮品偏好：乌龙茶、茉莉花茶") → confirms the merged content is correct

This approach:

• ✅ Matches your preferred fix (assert batch behavior, not per-item created log)
• ✅ Does not restore per-item logging in bulkStore (no noise added)
• ✅ Uses existing assertions that already cover the bulkStore result
• ⚠️ Note: storeCandidate() is now dead code after feat: bulkStore batch write + SmartExtractor optimization (Issue #665, #666) #669 (never called), but we're not touching it here

Question: Is removing lines 497-498 the correct fix, or should we replace them with a different assertion?

Thanks!