Fix #2417: split_continuous_references leaves references merged when comma styles are mixed

src/memos/mem_os/utils/reference_utils.py

@@ -1,13 +1,21 @@
+import re
+
 from memos.memories.textual.item import (
     TextualMemoryItem,
 )
 
 
+_REFERENCE_SEPARATOR_RE = re.compile(r",\s*")
+
+
 def split_continuous_references(text: str) -> str:
     """
     Split continuous reference tags into individual reference tags.
 
-    Converts patterns like [1:92ff35fb, 4:bfe6f044] to [1:92ff35fb] [4:bfe6f044]
+    Converts patterns like [1:92ff35fb, 4:bfe6f044] to [1:92ff35fb][4:bfe6f044].
+    Every comma between the brackets is treated as a separator regardless of
+    the whitespace that follows it, so mixed styles inside one tag (e.g.
+    ``[1:a,2:b, 3:c]``) all split correctly.
 
     Only processes text if:
     1. '[' appears exactly once
@@ -41,10 +49,11 @@ def split_continuous_references(text: str) -> str:
     # Check if there's a comma between brackets
     if "," not in content_between_brackets:
         return text
-    text = text.replace(content_between_brackets, content_between_brackets.replace(", ", "]["))
-    text = text.replace(content_between_brackets, content_between_brackets.replace(",", "]["))
-
-    return text
+    # Split on every comma (with optional trailing whitespace) in a single pass.
+    # Sequential str.replace calls fail when the tag mixes ", " and "," styles
+    # because the first pass mutates the substring the second pass looks for.
+    split_content = _REFERENCE_SEPARATOR_RE.sub("][", content_between_brackets)
+    return text.replace(content_between_brackets, split_content)
 
 
 def process_streaming_references_complete(text_buffer: str) -> tuple[str, str]:

tests/mem_os/test_reference_utils.py

@@ -0,0 +1,72 @@
+"""
+Test suite for src/memos/mem_os/utils/reference_utils.py
+
+Focus: split_continuous_references handles every comma-based
+separator style (with or without trailing whitespace, mixed styles
+inside a single tag). Related issue: #2417
+"""
+
+import pytest
+
+from memos.mem_os.utils.reference_utils import split_continuous_references
+
+
+class TestSplitContinuousReferences:
+    """Behavioral tests for split_continuous_references."""
+
+    def test_splits_comma_space_separator(self):
+        assert split_continuous_references("[1:aaa, 4:bbb, 7:ccc]") == "[1:aaa][4:bbb][7:ccc]"
+
+    def test_splits_bare_comma_separator(self):
+        assert split_continuous_references("[1:aaa,4:bbb,7:ccc]") == "[1:aaa][4:bbb][7:ccc]"
+
+    def test_splits_mixed_comma_styles_bare_then_space(self):
+        # Regression for #2417: bare comma followed by ", " style
+        # previously left the first pair merged.
+        assert (
+            split_continuous_references("[1:92ff35fb,4:bfe6f044, 7:abcd1234]")
+            == "[1:92ff35fb][4:bfe6f044][7:abcd1234]"
+        )
+
+    def test_splits_mixed_comma_styles_space_then_bare(self):
+        assert split_continuous_references("[1:aaa, 4:bbb,7:ccc]") == "[1:aaa][4:bbb][7:ccc]"
+
+    def test_splits_comma_with_multiple_spaces(self):
+        assert split_continuous_references("[1:aaa,  4:bbb,\t7:ccc]") == "[1:aaa][4:bbb][7:ccc]"
+
+    def test_single_reference_unchanged(self):
+        assert split_continuous_references("[1:aaa]") == "[1:aaa]"
+
+    def test_empty_string_returned_as_is(self):
+        assert split_continuous_references("") == ""
+
+    def test_text_without_brackets_unchanged(self):
+        assert split_continuous_references("no brackets here") == "no brackets here"
+
+    def test_text_with_multiple_bracket_pairs_unchanged(self):
+        original = "[1:aaa] and [2:bbb, 3:ccc]"
+        assert split_continuous_references(original) == original
+
+    def test_text_without_comma_between_brackets_unchanged(self):
+        assert split_continuous_references("hello [1:aaa] world") == "hello [1:aaa] world"
+
+    def test_prefix_and_suffix_preserved(self):
+        assert (
+            split_continuous_references("prefix [1:aaa, 4:bbb] suffix")
+            == "prefix [1:aaa][4:bbb] suffix"
+        )
+
+    def test_reversed_brackets_returned_unchanged(self):
+        assert split_continuous_references("]1:aaa,2:bbb[") == "]1:aaa,2:bbb["
+
+    @pytest.mark.parametrize(
+        "raw,expected",
+        [
+            ("[1:x,2:y]", "[1:x][2:y]"),
+            ("[1:x, 2:y]", "[1:x][2:y]"),
+            ("[1:x,  2:y]", "[1:x][2:y]"),
+            ("[1:x,\t2:y]", "[1:x][2:y]"),
+        ],
+    )
+    def test_parametrized_trailing_whitespace_variants(self, raw, expected):
+        assert split_continuous_references(raw) == expected