core[patch]: stringify tool non-content blocks (#24626)

Slightly breaking bugfix. Shouldn't cause too many issues since no
models would be able to handle non-content block ToolMessage.content
anyways.

libs/core/langchain_core/tools.py

@@ -1485,14 +1485,7 @@ def _format_output(
     content: Any, artifact: Any, tool_call_id: Optional[str], name: str, status: str
 ) -> Union[ToolMessage, Any]:
     if tool_call_id:
-        # NOTE: This will fail to stringify lists which aren't actually content blocks
-        # but whose first element happens to be a string or dict. Tools should avoid
-        # returning such contents.
-        if not isinstance(content, str) and not (
-            isinstance(content, list)
-            and content
-            and isinstance(content[0], (str, dict))
-        ):
+        if not _is_message_content_type(content):
             content = _stringify(content)
         return ToolMessage(
             content,
@@ -1505,6 +1498,26 @@ def _format_output(
         return content
 
 
+def _is_message_content_type(obj: Any) -> bool:
+    """Check for OpenAI or Anthropic format tool message content."""
+    if isinstance(obj, str):
+        return True
+    elif isinstance(obj, list) and all(_is_message_content_block(e) for e in obj):
+        return True
+    else:
+        return False
+
+
+def _is_message_content_block(obj: Any) -> bool:
+    """Check for OpenAI or Anthropic format tool message content blocks."""
+    if isinstance(obj, str):
+        return True
+    elif isinstance(obj, dict):
+        return obj.get("type", None) in ("text", "image_url", "image", "json")
+    else:
+        return False
+
+
 def _stringify(content: Any) -> str:
     try:
         return json.dumps(content)

libs/core/tests/unit_tests/test_tools.py

@@ -32,6 +32,8 @@
     StructuredTool,
     Tool,
     ToolException,
+    _is_message_content_block,
+    _is_message_content_type,
     tool,
 )
 from langchain_core.utils.function_calling import convert_to_openai_function
@@ -1623,3 +1625,47 @@ def foo(a: int, b: str) -> str:
         "title": pydantic_model.__name__,
         "type": "object",
     }
+
+
+valid_tool_result_blocks = [
+    "foo",
+    {"type": "text", "text": "foo"},
+    {"type": "text", "blah": "foo"},  # note, only 'type' key is currently checked
+    {"type": "image_url", "image_url": {}},  # openai format
+    {
+        "type": "image",
+        "source": {
+            "type": "base64",
+            "media_type": "image/jpeg",
+            "data": "123",
+        },
+    },  # anthropic format
+    {"type": "json", "json": {}},  # bedrock format
+]
+invalid_tool_result_blocks = [
+    {"text": "foo"},  # missing type
+    {"results": "foo"},  # not content blocks
+]
+
+
+@pytest.mark.parametrize(
+    ("obj", "expected"),
+    [
+        *([[block, True] for block in valid_tool_result_blocks]),
+        *([[block, False] for block in invalid_tool_result_blocks]),
+    ],
+)
+def test__is_message_content_block(obj: Any, expected: bool) -> None:
+    assert _is_message_content_block(obj) is expected
+
+
+@pytest.mark.parametrize(
+    ("obj", "expected"),
+    [
+        ["foo", True],
+        [valid_tool_result_blocks, True],
+        [invalid_tool_result_blocks, False],
+    ],
+)
+def test__is_message_content_type(obj: Any, expected: bool) -> None:
+    assert _is_message_content_type(obj) is expected