Requirements and Promises of Automatic Programming

Of course, these trends are not as clear-cut as their descriptions might suggest. Indeed, many systems sacrifice a little bit of every objective. 

In this last introductory section to automatic programming, we list a few requirements for research in program synthesis, and outline the promises of this research. Finally, pointers to introductory literature are given. 

there is a great need for incorporating knowledge into automatic programming systems, in order to overcome the old sins of general problem solving. There are essentially three kinds of useful synthesis knowledge  

Furthermore, there is a need for efficiency knowledge (useful for algorithm transformation and implementation), but this carries us beyond our focus on actual synthesis. As usual, there is the knowledge acquisition bottleneck, but current machine learning techniques (see Chapter 3) seem a promising step towards overcoming this. 

there is a need for evolutionary specifications, where one deliberately omits some functionalities in a first specification, and then relies on replay facilities [Wile 83] when adding the missing functionalities in subsequent versions of the specification. This form of exploratory programming promises an answer to the endemic difficulty of providing partially correct and complete formal specifications. 

---