Towards Generating Functionally Correct Code Edits from Natural Language Issue Descriptions