"""Tests for hybrid search (vector + full-text)."""
import pytest
from semantic_code_mcp.models import Chunk, ChunkType, ChunkWithEmbedding
from semantic_code_mcp.storage.lancedb import LanceDBConnection, LanceDBVectorStore
@pytest.fixture
def store_with_chunks(lance_connection: LanceDBConnection) -> LanceDBVectorStore:
"""Create a store with sample chunks for testing hybrid search."""
store = LanceDBVectorStore(lance_connection)
# Create embeddings that are actually different (not parallel)
# Using varied values to get different cosine similarities
emb_low = [0.1 if i % 2 == 0 else -0.1 for i in range(384)]
emb_high = [0.9 if i % 2 == 0 else 0.8 for i in range(384)]
emb_mid = [0.5 if i % 3 == 0 else 0.3 for i in range(384)]
chunks = [
# Chunk with "duration_ms" - should be found by keyword search
ChunkWithEmbedding(
chunk=Chunk(
file_path="/project/indexer.py",
line_start=100,
line_end=120,
content='log.debug("timing", duration_ms=round(time.time() - t0))',
chunk_type=ChunkType.METHOD,
name="index",
),
embedding=emb_low, # Low similarity to query
),
# Chunk semantically about timing but no "duration_ms" keyword
ChunkWithEmbedding(
chunk=Chunk(
file_path="/project/utils.py",
line_start=1,
line_end=10,
content="def measure_elapsed_time(): return time.perf_counter()",
chunk_type=ChunkType.FUNCTION,
name="measure_elapsed_time",
),
embedding=emb_high, # High similarity to query
),
# Chunk with "log.debug" but different topic
ChunkWithEmbedding(
chunk=Chunk(
file_path="/project/cache.py",
line_start=50,
line_end=60,
content='log.debug("cache_saved", files_count=len(files))',
chunk_type=ChunkType.METHOD,
name="_save",
),
embedding=emb_mid,
),
]
store.add_chunks(chunks)
return store
class TestHybridSearch:
"""Tests for hybrid search functionality."""
def test_vector_only_search_misses_keyword_match(self, store_with_chunks: LanceDBVectorStore):
"""Pure vector search may miss results with exact keyword matches."""
# Query embedding similar to "measure_elapsed_time" chunk (emb_high pattern)
query_embedding = [0.9 if i % 2 == 0 else 0.8 for i in range(384)]
results = store_with_chunks.search(query_embedding, limit=2)
# Vector search finds semantically similar, not keyword matches
assert len(results) > 0
# The "duration_ms" chunk has low embedding similarity, may not appear
names = [r.name for r in results]
assert "measure_elapsed_time" in names
def test_fts_search_finds_exact_keyword(self, store_with_chunks: LanceDBVectorStore):
"""Full-text search finds chunks containing exact keywords."""
results = store_with_chunks.search_fts("duration_ms", limit=5)
assert len(results) >= 1
assert any("duration_ms" in r.content for r in results)
def test_fts_search_returns_empty_for_no_match(self, store_with_chunks: LanceDBVectorStore):
"""FTS returns empty list when no matches found."""
results = store_with_chunks.search_fts("nonexistent_keyword_xyz", limit=5)
assert results == []
def test_hybrid_search_combines_both(self, store_with_chunks: LanceDBVectorStore):
"""Hybrid search finds both semantic and keyword matches."""
query_embedding = [
0.9 if i % 2 == 0 else 0.8 for i in range(384)
] # Similar to measure_elapsed_time
results = store_with_chunks.search_hybrid(
query_embedding=query_embedding,
query_text="duration_ms",
limit=5,
)
# Should find the keyword match in results
assert "duration_ms" in " ".join(r.content for r in results)
def test_hybrid_search_weight_adjustable(self, store_with_chunks: LanceDBVectorStore):
"""Can adjust weight between vector and FTS search."""
query_embedding = [0.9 if i % 2 == 0 else 0.8 for i in range(384)]
# Heavy FTS weight should prioritize keyword matches
results_fts_heavy = store_with_chunks.search_hybrid(
query_embedding=query_embedding,
query_text="duration_ms",
limit=5,
vector_weight=0.3, # 30% vector, 70% FTS
)
# Heavy vector weight should prioritize semantic matches
results_vec_heavy = store_with_chunks.search_hybrid(
query_embedding=query_embedding,
query_text="duration_ms",
limit=5,
vector_weight=0.9, # 90% vector, 10% FTS
)
# Both should return results, but ordering may differ
assert len(results_fts_heavy) > 0
assert len(results_vec_heavy) > 0
