Object, Message, and Performance: How They Coexist in Self

For example, last year you might have written a routine to sort an array of numbers. It was a method defined for arrays that works by sending the less-than (<) message to the numbers. Today, you might need to sort an array of strings. In a pure object-oriented language, you could just call the same sort method. The old sort method would still work, because the objects in the array (now strings) respond to the < message. The code run for < is decided at runtime according to the type of the receiver, so the sort method works for any object that implements a < method. When the same code can be used for different types of objects, it is said to be polymorphic. With this kind of polymorphism, you do not have to explicitly parameterize the sort routine when you write it--it just works! This quality of unanticipated reusability may be one of the reasons programmers feel empowered by pure object-oriented languages.

However, unlike structured programming, pure object-oriented programming cannot be implemented efficiently with traditional compilation techniques because traditional optimizations rely on static declarations of representation types. In this paper, we will present the novel implementation techniques that recapture much of the efficiency that would seem to be lost in a pure object-oriented language. For many of the bench marks we have measured, these techniques have provided a fivefold speedup, enabling SELF programs to come within a factor of two or three of optimized C.

To get the PostScript file, click here.