Codebase intelligence for AI and humans: code health scores, auto-generated docs, git analytics, dead code detection, and architectural decisions via MCP.
Python aicode-complexitycode-healthcode-intelligencecode-qualitydead-codedeveloper-toolsdocumentationgit-analyticsmcpopen-sourcepythonrefactoringstatic-analysistechnical-debt
@ (#266)
Raghav Chamadiyae9f3fb22295f
diff --git a/packages/core/src/repowise/core/persistence/vector_store/lancedb_store.py b/packages/core/src/repowise/core/persistence/vector_store/lancedb_store.pyindex 86354aec1..84f1f415f 100644--- a/packages/core/src/repowise/core/persistence/vector_store/lancedb_store.py+++ b/packages/core/src/repowise/core/persistence/vector_store/lancedb_store.py@@ -47,11 +47,32 @@ class LanceDBVectorStore(VectorStore):else:self._table = None # will be created on first upsert+ @staticmethod+ def _existing_vector_dim(schema) -> int | None:+ """Return the fixed-length dimension of the ``vector`` field, or None.++ Returns None when the field is absent or not a fixed-size list (in+ which case we can't meaningfully compare dimensions).+ """+ try:+ field = schema.field("vector")+ except KeyError:+ return None+ list_size = getattr(field.type, "list_size", None)+ # pyarrow uses ``list_size == -1`` for variable-length lists.+ if isinstance(list_size, int) and list_size > 0:+ return list_size+ return None+async def _ensure_table(self, sample_vector: list[float]) -> None:- """Create the LanceDB table if it does not exist yet."""- if self._table is not None:- return+ """Create the LanceDB table if it does not exist yet.+ If a table already exists but its vector dimension differs from the+ current embedder's output (e.g. the embedder was switched from+ ``mock`` (dim 8) to ``openai`` (dim 1536) between reindexes), the stale+ table is dropped and recreated. Otherwise every write would fail deep+ inside LanceDB with an opaque IO error that never mentions dimensions.+ """try:import pyarrow as pa # type: ignore[import]except ImportError as exc:@@ -61,6 +82,15 @@ class LanceDBVectorStore(VectorStore):) from excdim = len(sample_vector)++ if self._table is not None:+ existing_dim = self._existing_vector_dim(await self._table.schema())+ if existing_dim is None or existing_dim == dim:+ return+ # Embedder changed dimensions — the old vectors are unusable.+ await self._db.drop_table(self._table_name) # type: ignore[union-attr]+ self._table = None+schema = pa.schema([pa.field("page_id", pa.string()),diff --git a/tests/unit/persistence/test_vector_store.py b/tests/unit/persistence/test_vector_store.pyindex 17a8620f2..6c42cc147 100644--- a/tests/unit/persistence/test_vector_store.py+++ b/tests/unit/persistence/test_vector_store.py@@ -354,3 +354,56 @@ async def test_lancedb_embed_batch(tmp_path, mock_embedder):assert ids == {"p1", "p2"}finally:await store.close()+++class _FixedDimEmbedder:+ """Deterministic embedder with a configurable output dimension."""++ def __init__(self, dimensions: int) -> None:+ self.dimensions = dimensions++ async def embed(self, texts: list[str]) -> list[list[float]]:+ out: list[list[float]] = []+ for i, _ in enumerate(texts):+ vec = [0.0] * self.dimensions+ vec[i % self.dimensions] = 1.0 # unit-length+ out.append(vec)+ return out+++async def test_lancedb_reindex_recreates_table_on_dimension_change(tmp_path, mock_embedder):+ """Switching embedders (mock dim 8 → dim 1536) must recreate the table.++ Regression test for #234: the stale wiki_pages table kept the old vector+ schema, so every write failed with an opaque LanceDB IO error instead of+ being transparently rebuilt for the new embedder.+ """+ lancedb = pytest.importorskip("lancedb") # noqa: F841+ from repowise.core.persistence.vector_store import LanceDBVectorStore++ db_path = str(tmp_path / "lance")++ # First index with the 8-dim mock embedder.+ store = LanceDBVectorStore(db_path, mock_embedder)+ try:+ await store.embed_and_upsert(+ "p1", "old content", {"target_path": "a.py", "content": "old content"}+ )+ finally:+ await store.close()++ # Reindex against the same path with a different-dimension embedder.+ big_embedder = _FixedDimEmbedder(1536)+ store = LanceDBVectorStore(db_path, big_embedder)+ try:+ await store.embed_and_upsert(+ "p2", "new content", {"target_path": "b.py", "content": "new content"}+ )+ # The old 8-dim row is gone; the table now holds only the new write.+ ids = await store.list_page_ids()+ assert ids == {"p2"}+ results = await store.search("new content", limit=5)+ assert {r.page_id for r in results} == {"p2"}+ finally:+ await store.close()