Dynamic ML without Dynamic Types

We describe a variant of the Standard ML programming language which incorporates a facility for the replacement of modular components during program execution. This useful extension of the language builds upon existing compiler technology which permits the separate compilation of modular units of a Standard ML program. Defined naively, this extension would incur performance overheads due to the need to retain typing information at run-time. Here we explain how this cost can be significantly lessened and in some cases eliminated entirely. The essential technical device which we employ for implementation of our extension to the language is a modification of two-space copying garbage collection.