Skip to content

Codebase intelligence for AI and humans: code health scores, auto-generated docs, git analytics, dead code detection, and architectural decisions via MCP.

fix(search): score hybrid symbols on content words when there is no identifier (#3677)

Yunus Ozen25413649c8eb
diff --git a/packages/server/src/repowise/server/mcp_server/tool_search.py b/packages/server/src/repowise/server/mcp_server/tool_search.py
index a0bdef1ce..748c4929e 100644
--- a/packages/server/src/repowise/server/mcp_server/tool_search.py
+++ b/packages/server/src/repowise/server/mcp_server/tool_search.py
@@ -80,6 +80,7 @@ from repowise.server.mcp_server._query_shape import (
_symbol_matches_name,
path_tokens,
)
+from repowise.server.mcp_server._query_terms import content_terms
from repowise.server.mcp_server._references import path_identity, symbol_identity
from repowise.server.mcp_server._retrieval_rank import (
boost_named_paths,
@@ -1085,8 +1086,12 @@ async def _structured_search(
symbol_query = canonical_symbol[1]
if mode == "hybrid":
_idents = _embedded_identifiers(query, names)
- if _idents:
- symbol_query = " ".join(_idents)
+ # Without an identifier, score on the content words so stopwords such
+ # as "on" can't match symbols like ``test_basic_auth_on_session``. A
+ # query made only of stopwords keeps the raw text.
+ symbol_query = (
+ " ".join(_idents) if _idents else " ".join(content_terms(query)) or query
+ )
# Only the hybrid window collapses symbol rows, so only it over-fetches
# them; path hits are file pages, already one row per file. Symbol mode
diff --git a/tests/unit/server/mcp/test_search.py b/tests/unit/server/mcp/test_search.py
index 462d2ad9c..3cbad2fdf 100644
--- a/tests/unit/server/mcp/test_search.py
+++ b/tests/unit/server/mcp/test_search.py
@@ -1194,6 +1194,51 @@ class TestExactMatchSignal:
await search_codebase("AuthService", mode="symbol")
assert seen == ["AuthService"]
+ @pytest.mark.asyncio
+ async def test_hybrid_without_identifier_scores_symbols_on_content_words(
+ self, setup_mcp, monkeypatch
+ ):
+ # Regression for #3593: with no identifier in the question the scorer
+ # got the raw prose, so "on" matched test_basic_auth_on_session.
+ from repowise.server.mcp_server import search_codebase, tool_search
+
+ seen: list[str] = []
+ real = tool_search.search_symbols_single
+
+ async def spy(ctx, query, limit, **kwargs):
+ seen.append(query)
+ return await real(ctx, query, limit, **kwargs)
+
+ monkeypatch.setattr(tool_search, "search_symbols_single", spy)
+
+ await search_codebase("retry on connection failure", mode="hybrid")
+ assert seen, "symbol scorer was not called"
+ assert "on" not in seen[0].split(), seen
+ assert {"retry", "connection", "failure"} <= set(seen[0].split()), seen
+
+ @pytest.mark.asyncio
+ async def test_hybrid_stopword_does_not_outrank_the_content_match(self, setup_mcp):
+ from repowise.server.mcp_server import search_codebase
+
+ await _seed_symbol("test_basic_auth_on_session")
+ await _seed_symbol("RetryTransport")
+
+ result = await search_codebase("retry on connection failure", mode="hybrid", limit=10)
+ names = [r.get("name") for r in result["results"] if r.get("type") == "symbol"]
+ assert "RetryTransport" in names, names
+ if "test_basic_auth_on_session" in names:
+ assert names.index("RetryTransport") < names.index("test_basic_auth_on_session")
+
+ @pytest.mark.asyncio
+ async def test_symbol_mode_still_matches_stopword_identifiers(self, setup_mcp):
+ from repowise.server.mcp_server import search_codebase
+
+ await _seed_symbol("get_on_change")
+
+ result = await search_codebase("get_on_change", mode="symbol")
+ names = [r.get("name") for r in result["results"]]
+ assert "get_on_change" in names, names
+
@pytest.mark.asyncio
async def test_exact_hit_sets_true_and_no_fuzzy_note(self, setup_mcp):
from repowise.server.mcp_server import search_codebase