fix(tools): correct args validation in langchain tool wrapper (#761)

Fixed logic error in MelleaTool.from_langchain() where the condition
`if args is not None or len(args) != 0:` was always True, causing
false positive warnings on every tool call.

Changed to `if args:` which correctly checks for non-empty tuples,
since *args is always a tuple (never None) in Python.

Added test_from_langchain_args_handling() to verify:
- No warning when called with kwargs only (normal case)
- Warning logged when positional args passed (edge case)

Closes: #760

Signed-off-by: Mark Sturdevant <mark.sturdevant@ibm.com>

Author: markstur

mellea/backends/tools.py

@@ -95,7 +95,7 @@ def from_langchain(cls, tool: Any) -> "MelleaTool":
 
                 def parameter_remapper(*args, **kwargs):
                     """Langchain tools expect their first argument to be 'tool_input'."""
-                    if args is not None or len(args) != 0:
+                    if args:
                         # This shouldn't happen. Our ModelToolCall.call_func actually passes in everything as kwargs.
                         FancyLogger.get_logger().warning(
                             f"ignoring unexpected args while calling langchain tool ({tool_name}): ({args})"

test/backends/test_mellea_tool.py

@@ -99,6 +99,30 @@ def test_from_langchain():
     assert t.run(input=2) == "2"
 
 
+def test_from_langchain_args_handling(caplog):
+    """Test that langchain tools handle args correctly.
+
+    Verifies that:
+    1. Tools work correctly when called with kwargs only (normal case)
+    2. A warning is logged if positional args are passed (shouldn't happen)
+    """
+    t = MelleaTool.from_langchain(langchain_tool)
+
+    # Normal case: kwargs only (no warning expected)
+    caplog.clear()
+    result = t.run(input=5)
+    assert result == "5"
+    assert "ignoring unexpected args" not in caplog.text
+
+    # Edge case: positional args passed (warning expected)
+    # This shouldn't happen in practice since ModelToolCall.call_func uses **kwargs
+    caplog.clear()
+    result = t.run(42, input=5)  # Pass both positional and keyword args
+    assert result == "5"  # Should still work, using kwargs
+    assert "ignoring unexpected args" in caplog.text
+    assert "langchain_tool" in caplog.text
+
+
 @pytest.mark.qualitative
 @pytest.mark.ollama
 @pytest.mark.llm