Disciplined Meta-Programming

计算机科学家编写操纵程序(如编译器)的程序已经有很长时间了。另一方面，软件工程师使用这一特定技术的速度要慢得多，主要是因为标准设计方法不支持这种技术。这种情况最近开始发生变化，原因有很多，其中包括：1.领域特定语言的日益流行；2.手工编写并行软件既困难又容易出错；3.当前的编程语言提供了很差的“高级”抽象机制。不幸的是，“在野外”遇到的元程序(操纵程序的程序被称为“元程序”)很少像“常规”程序那样结构良好。换句话说，似乎没有好的开发模型来编写这样的程序。这相当不幸，因为元程序允许标准的软件工程概念(如设计选择)成为一流的软件构件。此外，在成熟领域(如科学计算)的情况下，可以有效地捕获和利用高阶结构来极大地缩短源程序以及生成更高效的运行时程序。我们的目标是在静态类型、纯函数元编程的上下文中纠正这种情况。我们将继续我们的成功工作，寻找和发明编写有纪律的、模块化的元程序所需的技术。一种非常成功的方法是采用程序族，并设计适当地捕获这些族的设计选择的非特别抽象。这些抽象构成了家庭成员生成器的核心。通过编写更大的可重用组件库(用于科学计算和游戏)，我们将利用我们正在进行的机械化数学基础工作(数学方案项目)，以及我们在高级元编程技术方面的专业知识，将我们以经验为导向的设计知识提炼成有效的开发方法，该方法可以适当地包含原则性元程序的设计。

Computer scientists have been writing programs which manipulate programs (such as compilers) for a very long time. Software engineers, on the other hand, have been much slower to use this particular technique, mostly because standard design methodologies do not embrace such techniques. This has recently started to change, and there are a variety of reasons for this, amongst which: 1. the increasing popularity of Domain Specific Languages, 2. the fact that writing parallel software by hand is difficult and error-prone, and 3. that current programming languages offer poor "higher-level" abstraction mechanisms. Unfortunately, few meta-programs (as programs which manipulate programs are called) encountered "in the wild" are as well-structured as "regular" programs. In other words, there does not seem to be good development models for writing such programs. This is rather unfortunate, as meta-programs allow standard software engineering concepts (such as design choices) to become first-class software artifacts. Furthermore, in the case of mature domains (like scientific computation), higher-order structure can be effectively captured and leveraged to both drastically shorten source programs as well as to generate more efficient run-time programs. Our objective is to rectify this situation, in the context of statically-typed, purely functional meta-programming. We will continue our successful work on finding and inventing the techniques necessary for writing disciplined and modular meta-programs. One very successful approach has been to take program families, and engineer non ad hoc abstractions which appropriately capture the design choices of those families. These abstractions form the core of generators for the family members. Through writing larger generative libraries of reusable components (for scientific computation and for games), we will leverage our ongoing foundational work on mechanized mathematics (the MathScheme project), as well as our expertise in advanced meta-programming techniques, to distill our experience-driven design knowledge into an effective development methodology which can appropriately encompass the design of principled meta-programs.