Lars Bendix:
Towards Artificially Intelligent Management of Software Development Tasks,
at the 12th Summerschool "Programming '87", Varna, Bulgaria, May 25, 1987.

Abstract:
During the life cycle of a large software system, two problems are often found: dealing with a large number of versions of a component and composing the system in different ways and using different components. These problems occur both during the initial development and during the maintenance. Solving these problems is often a simple and trivial task that, however, requires a lot of clerical work and accuracy. Indeed, for their simplicity, these tasks can be fully automated.

In this talk we first give a thorough description of the two domains/tasks we are addressing, i. e. System Version Control (SVC) for dealing with the number of versions, and Configuration Management (CM) for composing the system. This is done by giving a detailed survey of two among the standard tools in this area: SCCS for SVC, and Make for CM. The survey will stress the main problems in each area.

Then we discuss the ways to integrate the two tasks/tools, as well with each other as with the rest of the tasks/tools of a Software Development Environment. In this discussion we will focus on two main points: how to exploit the available knowledge in an optimal way, and how to increase the amount of available knowledge. This is done by first giving a brief survey of some of the efforts that have been done to clarify the models behind the two tasks and to integrate them.

Finally, we will present our proposal for a complete knowledge integration in a Software Development Environment and the relevant issues that we have identified:

• How an environment/project database can be used as the central repository of knowledge.
• Which information to put into this database.
• How to use AI techniques to reason about the knowledge (active databases, planners).
• Generality and genericity of problems.
• The underlying models for the tasks.