Using Immutable Types for Debugging and Parallelism

在许多编程语言中，尤其是那些具有“函数式”编程风格的编程语言中，绝大多数对象在创建后不会被修改。ML语言的不同寻常之处在于，编译时类型检查器可以提前区分可以修改的对象和不可以修改的对象。实现可以通过几种方式使用这种关于不变性的高级信息：通过优化编译器，它简化了别名计算；通过垃圾回收器，它可以提高引用的局部性；通过并行化编译器，它简化了数据共享；以及通过调试器，它允许高效地重放程序执行。由于ML是少数几种在编译时提供不变性保证的命令式语言之一，因此对于不变性的用法还没有太多的研究。有许多方法可以将编译时保证转换为运行时节省；我们将研究其中的几种方法。

In many programming languages, especially those with a "functional" style of programming, the vast majority of objects are not modified after they are created. The ML language is unusual in that the compile-time typechecker can distinguish in advance objects that may be modified from objects that may not. This advance information about immutability can be used by implementations in several ways: by optimizing compilers, where it simplifies aliasing calculations; by garbage collectors, where it can improve locality of reference; by parallelizing compilers, where it simplifies sharing of data; and by debuggers, where it allows efficient replay of program execution. Since ML is one of the few imperative languages that provides compile-time guarantees of immutability, there has not been much research into the uses of immutability. There are many ways that compile-time guarantees can be translated into run-time savings; several of these will be studied.