A Static Evaluation of Code Completion by Large Language Models