[Bug] Wrong field order with typed fields in dspy.Signature

[malusamayo]

What happened?

When adding type annotations to some of the output fields, the final parsed field orders are wrong:

class ExampleSignature(dspy.Signature):
    """Evaluate the text as "good" or "bad"."""
    
    text = dspy.InputField(desc="The text to evaluate.")
    output = dspy.OutputField(desc="The evaluation of the text.")
    pass_evaluation: bool = dspy.OutputField(desc="The evaluation result, True if the text is good, False otherwise.") 

print(ExampleSignature.fields)

In the output, the orders of two output fields are swapped:

{'text': FieldInfo(annotation=str, required=True, json_schema_extra={'desc': 'The text to evaluate.', '__dspy_field_type': 'input', 'prefix': 'Text:'}),
 'pass_evaluation': FieldInfo(annotation=bool, required=True, json_schema_extra={'desc': 'The evaluation result, True if the text is good, False otherwise.', '__dspy_field_type': 'output', 'prefix': 'Pass Evaluation:'}),
 'output': FieldInfo(annotation=str, required=True, json_schema_extra={'desc': 'The evaluation of the text.', '__dspy_field_type': 'output', 'prefix': 'Output:'})}

This only happens when output fields are partially typed. If all output fields are typed, the order is correct.

Steps to reproduce

See above.

DSPy version

2.6.0

[okhat]

Yes, I believe it's a bug upstream in pydantic as far as @isaacbmiller discovered