optimize plain search

tests/test_search.py

@@ -1,9 +1,12 @@
 from pathlib import Path
+from typing import Any
 
 import pytest
 
+import corpus_mcp.database as database_module
 from corpus_mcp.database import (
     CorpusError,
+    SEARCH_CANDIDATE_MULTIPLIER,
     build_database,
     read_document_excerpt,
     search_database,
@@ -50,8 +53,8 @@ def test_search_returns_separate_excerpts_from_a_large_document(tmp_path: Path)
     assert excerpts[1]["chunk_end_char"] <= len(content)
 
 
-def test_large_document_does_not_hide_other_matching_documents(
-    tmp_path: Path,
+def test_common_term_search_uses_a_bounded_candidate_pool(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
 ) -> None:
     source = tmp_path / "source"
     source.mkdir()
@@ -59,13 +62,28 @@ def test_large_document_does_not_hide_other_matching_documents(
     (source / "small.txt").write_text("common term", encoding="utf-8")
     database = tmp_path / "corpus.sqlite"
     build_database(source, database)
+    connection = database_module._connect_read_only(database)
+    executed: list[tuple[str, tuple[Any, ...]]] = []
 
-    result = search_database(database, "common", limit=2)
+    class RecordingConnection:
+        def execute(self, statement: str, parameters: tuple[Any, ...] = ()) -> Any:
+            executed.append((statement, parameters))
+            return connection.execute(statement, parameters)
 
-    assert {match["path"] for match in result["matches"]} == {
-        "large.txt",
-        "small.txt",
-    }
+        def close(self) -> None:
+            connection.close()
+
+    monkeypatch.setattr(
+        database_module,
+        "_connect_read_only",
+        lambda _database: RecordingConnection(),
+    )
+
+    search_database(database, "common", limit=2, excerpts_per_document=3)
+
+    assert len(executed) == 1
+    assert "GROUP BY" not in executed[0][0]
+    assert executed[0][1][-1] == 2 * 3 * SEARCH_CANDIDATE_MULTIPLIER
 
 
 def test_read_document_excerpt_is_bounded_and_navigable(