If we decide that exact error message parity is not needed these can and should be removed as they complicate the tokenizer.