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): scan slash-containing literal queries (#4230)

Vamsi-258143acf0625d3b
diff --git a/packages/server/src/repowise/server/mcp_server/_line_hits.py b/packages/server/src/repowise/server/mcp_server/_line_hits.py
index 03b089f25..cddd1b8b3 100644
--- a/packages/server/src/repowise/server/mcp_server/_line_hits.py
+++ b/packages/server/src/repowise/server/mcp_server/_line_hits.py
@@ -95,6 +95,8 @@ def _literal(query: str) -> tuple[str, bool] | None:
return m.group(2), False
if len(q) >= 2 and _TOKEN.match(q) and not q.endswith("."):
return q, True
+ if len(q) >= 2 and "/" in q and q.strip("./") and not any(char.isspace() for char in q):
+ return q, False
if _is_code_shaped(q):
return q, False
return None
@@ -419,9 +421,13 @@ async def attach_line_hits(
async def _attach(
response: dict, query: str, mode: str, names: Container[str] | None, repo: str | None
) -> None:
- if repo == "all" or mode == "path":
+ if repo == "all":
return
literal = _literal(query)
+ if mode == "path" and (
+ response.get("results") or literal is None or "/" not in literal[0]
+ ):
+ return
candidates = _identifier_candidates(query, mode, names)
if literal and not literal[1] and not _QUOTED.match(query.strip()):
# Code punctuation names the text itself: an identifier inside it
diff --git a/tests/unit/server/mcp/test_search_line_hits.py b/tests/unit/server/mcp/test_search_line_hits.py
index cc1b2428f..c3f513051 100644
--- a/tests/unit/server/mcp/test_search_line_hits.py
+++ b/tests/unit/server/mcp/test_search_line_hits.py
@@ -9,7 +9,7 @@ from datetime import UTC, datetime
import pytest
-from repowise.core.persistence.models import GraphEdge, GraphNode
+from repowise.core.persistence.models import GraphEdge, GraphNode, Page
from repowise.server.mcp_server import _edit_sites, _line_hits, _meta, _state
from repowise.server.mcp_server._budget import enforce_response_budget
from repowise.server.mcp_server.tool_search import search_codebase
@@ -26,6 +26,7 @@ _FILES = {
" print('hello there')\n"
" return backend_of('m')\n"
"x = 147821\n"
+ 'USER_AGENT = "acme-cli/0.1.0"\n'
),
}
@@ -137,6 +138,60 @@ async def test_literal_number_is_scanned_with_definition_first(tree):
assert out["complete"] is True
+async def test_slash_literal_falls_back_from_path_mode(tree):
+ out = await search_codebase("acme-cli/0.1.0")
+
+ assert _rows(out) == [("app/widget.py", 8, "definition")]
+ assert out["complete"] is True
+
+
+async def test_slash_literal_is_scanned_in_hybrid_mode(tree):
+ out = await search_codebase("acme-cli/0.1.0", mode="hybrid")
+
+ assert _rows(out) == [("app/widget.py", 8, "definition")]
+ assert out["complete"] is True
+
+
+async def test_matching_path_query_does_not_add_line_hits(tree, session, setup_mcp):
+ # Path mode matches indexed file pages, so the file needs one to be found.
+ session.add(
+ Page(
+ id="file_page:app/widget.py",
+ repository_id=setup_mcp,
+ page_type="file_page",
+ title="Widget",
+ content="# Widget",
+ target_path="app/widget.py",
+ source_hash="widget",
+ model_name="mock",
+ provider_name="mock",
+ generation_level=2,
+ confidence=0.8,
+ freshness_status="fresh",
+ metadata_json="{}",
+ created_at=_NOW,
+ updated_at=_NOW,
+ )
+ )
+ await session.commit()
+ out = await search_codebase("app/widget.py", mode="path")
+
+ assert any(result["path"] == "app/widget.py" for result in out["results"])
+ assert "lines" not in out
+
+
[email protected]("query", ["SomeName", "./", "../"])
+async def test_unmatched_path_query_only_falls_back_for_slash_literals(tree, query):
+ out = await search_codebase(query, mode="path")
+
+ assert out["results"] == []
+ assert "lines" not in out
+
+
@pytest.mark.asyncio
async def test_quoted_string_is_scanned_as_a_substring(tree):
out = await search_codebase('"hello there"')