fix: keep query embedding for fine search

Author: fancyboi999

src/memos/memories/textual/tree_text_memory/retrieve/searcher.py

@@ -320,10 +320,13 @@ def _parse_task(
         )
 
         query = parsed_goal.rephrased_query or query
-        # if goal has extra memories, embed them too
-        if parsed_goal.memories:
-            embed_texts = list(dict.fromkeys([query, *parsed_goal.memories]))
-            query_embedding = self.embedder.embed(embed_texts)
+        embed_texts = [
+            text.strip()
+            for text in [query, *parsed_goal.memories]
+            if isinstance(text, str) and text.strip()
+        ]
+        if embed_texts:
+            query_embedding = self.embedder.embed(list(dict.fromkeys(embed_texts)))
         return parsed_goal, query_embedding, context, query
 
     @timed

tests/memories/textual/test_tree_searcher.py

@@ -3,6 +3,7 @@
 import pytest
 
 from memos.memories.textual.item import TextualMemoryItem, TreeNodeTextualMemoryMetadata
+from memos.memories.textual.tree_text_memory.retrieve.retrieval_mid_structs import ParsedTaskGoal
 from memos.memories.textual.tree_text_memory.retrieve.searcher import Searcher
 from memos.reranker.base import BaseReranker
 
@@ -24,13 +25,14 @@ def mock_searcher():
     return s
 
 
-def make_item(content: str, score: float):
+def make_item(content: str, score: float, memory_type: str = "WorkingMemory"):
     # Simulate a TextualMemoryItem with usage list for update test
     return (
         TextualMemoryItem(
             memory=content,
             metadata=TreeNodeTextualMemoryMetadata(
                 embedding=[0.1] * 5,
+                memory_type=memory_type,
                 usage=[],
             ),
         ),
@@ -104,6 +106,34 @@ def test_searcher_fine_mode_triggers_reasoner(mock_searcher):
     assert len(result) == 1
 
 
+def test_fine_search_embeds_query_when_parser_returns_no_memory_expansions(mock_searcher):
+    query = "我喜欢什么"
+    user_memory = make_item("我喜欢草莓", 0.9, memory_type="UserMemory")[0]
+
+    mock_searcher.task_goal_parser.parse.return_value = ParsedTaskGoal(
+        keys=["喜欢", "偏好", "兴趣"],
+        tags=["personal preference", "user interest", "taste"],
+        memories=[],
+        rephrased_query="",
+    )
+    mock_searcher.embedder.embed.return_value = [[0.1] * 5]
+
+    def retrieve_side_effect(*args, **kwargs):
+        if kwargs.get("memory_scope") == "UserMemory":
+            assert kwargs["query_embedding"] == [[0.1] * 5]
+            return [user_memory]
+        return []
+
+    mock_searcher.graph_retriever.retrieve.side_effect = retrieve_side_effect
+    mock_searcher.reranker.rerank.return_value = [(user_memory, 0.9)]
+
+    result = mock_searcher.search(query=query, top_k=1, mode="fine", memory_type="UserMemory")
+
+    mock_searcher.embedder.embed.assert_called_once_with([query])
+    assert len(result) == 1
+    assert result[0].memory == "我喜欢草莓"
+
+
 def test_searcher_respects_memory_type(mock_searcher):
     parsed_goal = MagicMock()
     parsed_goal.memories = ["Something"]