Discussion questions from CS 7390 lecture

Date: 2/9/96

Presented by: John Stasko

Topic: Hybrid Systems 2

• In UWPI, to overcome the difficulty of data type inference in a buggy program using the behavior pattern matching method, will the name of the variables be another source of inference?
• In Lens, will it be helpful to provide more complex data display mechanisms, such as stacks, queues, objects which can be wrapped into the variables in the program?
• The animation/visualization capabilities provided by lens seem to be useful for certain types of programs, but I'm not quite convinced that it would be generally useful. Would it be possible to build a system along the lines of Lens that would be directly applicable to real workd software development?
• UWPI can only work well when wekk known ADTs are implemented and can be recognized. Its extensibility seems to be limited to additional commonly used structures that are nearly always implemented similarly. Given this restriction, wouldn't it make sense to build a package something like the Standard template Library (full of ADTs) that would have visuali illsutration design incorporated into it?
• It seems that the biggest advantage of this system is the fact that users don't have to program the algorithm illustrations. This feature is very necessary for a debugging system. Would it be possible to take the next step and provide a robust way of dealing with serious programmer errors in order to make UWPI a useful debugging system?
• How important is it to have context when doing an algorithm animation? I'm thinking of the Towers of Hanoi when I ask this: most animations of it depict disks on poles being moved around. Wouldn't it be just as effective to have 3 arrays, all depicted horizontally, and illustrate the bars moving? It seems that thios would be much easier to automate.
• Have any of the developers of the algorithm animation systems that "develop their own animations" using the algorithm code, done any studies to show how well these systems work at constructing an animation? That is, other than just saying it works well.
• How can the UWPI system create animation when there are bugs in the algorithm? It seems to me that the code would have to be bug-free and implemented in a way that UWPI could understand.
• The Lens system works in connection with dbx. Automated tools not asking for more coding are really useful. What effort, for which result should be put in that direction, in opposition to effort to display appropriate information?
• UWPI: Going from Prolog to Common Lisp is seen as an improvement. Is Lisp really the ultimate programming language for this type of application? I am not sure.
• UWPI seems to be targetted to "old" (legacy) code too much: it seems that today programmers are more likely to use compiler support for abstract data types than to use the FORTRAN-style coding (ie, being stuck with just arrays and pointers). Do you think this approach will be more or less limited to re-engineering, or is it more general?
• In choosing the operations to support in Lens, a bunch of XTango programs were analyzed. Are we sure that the frequently used XTango elements were frequently used because they are the most useful? Perhaps it's just that these elements were the easiest to code?
• Lens: There's a lot of talk about direct manipulation but I didn't see one good example of this. It seems to be misused at times: "instantiate objects graphically through direct manipulation", but the user has to type in the name of an object to display it.
• Isn't the limited variable name syntax a problem for Lens? For example, you can't name any node in a tree (graph) unless you have an explicit pointer to it. This implies extra programming which really should not be necessary.
• Another problem is the lack of implicit layout. Suppose you have an n node graph where n is determined at run-time. The programmer would have to come up with some layout algorithm to visualize the graph.
• Doesn't the limited programming constructs defeat the purpose of UWPI? I thought the idea was to lessen the burden on the programmer (illustrator) but a) there are no pointers or records, b) the language is only a subset of Pascal, c) the compiler cannot possibly infer good visual representations of complex algorithms and interpret the use and meaning of all the data structures. This means that the code and algorithms have to be greatly modified to conform to UWPI's limited capabilities.