Programming Interactive Scripts for the Web

Matthias Felleisen

The programmer of an interactive CGI script (or servlet) faces a serious design constraint. When a script needs to query a consumer, it must produce a Web page and terminate. When the consumer submits a response, the server will invoke another script. In addition, a consumer can bookmark generated pages, switch back and forth between such pages, clone browser windows, and submit responses repeatedly and simultaneously. All of this requires the programmer to break a simple interactive program into many separate scripts and to communicate data between these scripts in an unnatural manner.

My team and I have developed two approaches to overcome this problem. The first relies on the implementation of a special-purpose Web server that supports a scripting language with constructs for consumer interactions. Using these new constructs, a programmer can develop and test an ordinary interactive program, and the server takes care of the interactions between the program and the consumer. The second implements these new constructs as extensions to existing programming languages using well-established program transformation techniques from the functional world. Programmers can also use this second approach directly, without any tool support.

We have tested both approaches. In this talk, I will demonstrate the problem, explain our two approaches, and discuss the advantages and disadvantages of each.

Joint work with: Paul Graunke, Robert Findler, Shriram Krishnamurthi