| Over the past decade, there have been substantial advances in software development techniques. However, these innovations still cannot cope with the ever increasing complexity of modern software systems, which makes software development an expensive and error-prone process. Deficiencies during the development process can lead to serious quality problems that hamper both the initial development as well as subsequent maintenance activities. One approach for overcoming these challenges is to develop advanced program synthesis techniques to increase the overall software dependability, improve software development productivity, and reduce the development cost and time.;In this Dissertation, we propose an advanced AI planning-based program synthesis framework, in which Model-Driven Development (MDD), AI planning, and Component-based Software Development (CBSD) techniques are seamlessly integrated. This framework enables both static and behavioral aspects of the system to be automatically generated. Specifically, MDD facilitates the generation of static aspects of the system through transformations among the Platform Independent Model (PIM), the Platform Specific Model (PSM), and the final code. AI planning and CBSD are used to generate the behavioral aspects of the system. This is achieved by automatically generating the glue code needed to meet a given specification by assembling the system from existing components.;In the advanced AI planning-based program synthesis framework, the reasoning capability of the synthesizer is based on automated AI planning by mapping plans to programs. However, existing AI planning techniques are either not very powerful or are not very efficient and scalable. This Dissertation presents a powerful and efficient AI planner, FIP (Fast Iterative Planner), built upon Graphplan's intrinsic features to rapidly produce program-like plans with conditional and loop constructs.;Based on FIP, AI planning can be integrated with service patterns, a CBSD technique that captures the typical usages of the underlying services as well as their interactions. A planning domain model for service patterns is formulated to enable service patterns and pattern operators to be translated into composition rules to guide the planning process. Then, the connection between the AI planning-based synthesis subsystem and MDD is established by formulating system behaviors as AI planning problems.;This research has a significant impact on automated software engineering by raising the level of abstraction and increasing the level of reuse in system construction, thereby increasing the systems reliability, improving development productivity, and reducing the development cost. The experimental evaluations show that most of the system can be automatically generated through this framework and the quality of the generated program is comparable to manually developed programs. |
