fix(evidence): retain referenced vector generations

This commit is contained in:
2026-07-12 05:34:47 +02:00
parent 3a9808f126
commit 92b1d3ccc3
7 changed files with 127 additions and 28 deletions
@@ -119,6 +119,8 @@ def test_real_pgvector_corpus_job_lifecycle(tmp_path, persistent_pgvector):
source_root = tmp_path / "sources"
source_root.mkdir()
kept = source_root / "kept.md"
stable = source_root / "stable.md"
stable.write_text("unchanged dependency evidence", encoding="utf-8")
removed = source_root / "removed.md"
removed.write_text("removed evidence generation zero", encoding="utf-8")
@@ -208,6 +210,7 @@ def test_real_pgvector_corpus_job_lifecycle(tmp_path, persistent_pgvector):
assert active_pack
assert any("active fourth generation" in result.content for result in active_pack)
assert all(removed_document.content not in result.content for result in active_pack)
assert any("unchanged dependency evidence" in result.content for result in active_pack)
manifest = CorpusStore(tmp_path / "corpus").active_manifest()
assert resolve_evidence_file(
CorpusStore(tmp_path / "corpus"), removed_document_id,
@@ -229,7 +232,9 @@ def test_real_pgvector_corpus_job_lifecycle(tmp_path, persistent_pgvector):
recreated_hits = ActiveEvidenceSearcher(
CorpusStore(tmp_path / "corpus"), EvidenceDelegate(recreated)
).search(query, top_n=2, kinds=["evidence"])
assert recreated_hits and all(hit.metadata["vector_generation"] == resumed.generation for hit in recreated_hits)
active_dependencies = set(manifest.metadata["document_generations"].values())
assert recreated_hits
assert all(hit.metadata["vector_generation"] in active_dependencies for hit in recreated_hits)
orphan = "gen:" + "f" * 32
recreated.upsert("evidence", [VectorWriteRecord(
@@ -243,7 +248,8 @@ def test_real_pgvector_corpus_job_lifecycle(tmp_path, persistent_pgvector):
final_pipeline = _pipeline(tmp_path, source_root, recreated)
report = final_pipeline.gc(workspace_root=tmp_path)
assert report["evicted"] == [orphan]
expected = set(generations[-2:])
assert set(CorpusStore(tmp_path / "corpus").list_generations()) == expected
assert set(recreated.list_evidence_generations("evidence")) == expected
expected_fs = set(generations[-2:])
assert set(CorpusStore(tmp_path / "corpus").list_generations()) == expected_fs
expected_vectors = expected_fs | {generations[0]}
assert set(recreated.list_evidence_generations("evidence")) == expected_vectors
assert final_pipeline.gc(workspace_root=tmp_path)["evicted"] == []