Keynote I: The language, optimizer, and tools mess

For many years, new languages promised higher performance in addition to other desirable characteristics such as higher productivity, better static checking, better reusability and composability, etc. However, in practice, virtually no new languages win in performance over historic languages like C and Fortran. Why? Higher level semantics were supposed to provide additional information so that optimizers could perform better analyses and so that conservative analyses could be less conservative.