Summary: | 博士 === 國立交通大學 === 資訊工程學系 === 86 === Software reuse is an effective way to improve software
productivity and quality. The process of software reuse can be
divided into four phases: abstraction, selection,
specialization, and integration. In this thesis, the research
includestwo parts: one is component retrieval and the other is
component integration. Theyare aimed to improve the
effectiveness in the selection and integration phases,
respectively. For component retrieval, current software
libraries provide all userswith the same search/display order.
The retrieved/display results are not suitablefor users for
their abilities well. For component integration, current web
databaseapplication programs may consist of hundreds of files
and the application logic is distributed on the files. When a
programmer modifies some program files of an application on the
integration phase, the program affected should be updated
simultaneously to keep the functional consistency. The
identification of the affected programs directly influences the
effectiveness of program integration. In this thesis, a
Software Reuse Framework (SRF) is presented to improve
componentretrieval based on a (built-in) hierarchical thesaurus.
The classification process ofa code library was made in a semi-
automatic way. SRF is a domain-independent frameworkthat can be
adapted to various libraries. SRF also provides four search
levels to provide users of different skill levels with
correspondent search/display order. Forthe component
integration, a model called HED model is presented to meet the
prior requirement of web database applications. HED model is
composed of three diagrams, hyperlink diagrams, entity-
relationship diagrams, and data-flow diagram, to representthe
logic of application programs. Based on the HED model, the
programs files affectedby a program modification can be
identified precisely via the structure and database analyses.
In addition, two sample SRF systems* are implemented, one for
Smalltalk-80 and the other for MFC 4.0 (MicrosoftR Foundation
Class) libraries. Some information collected during classifying
these two libraries shows that the classification effort is
reducedvia SRF. Another experiment is designed to evaluate the
feasibility of SRF search process. The result reveals that the
SRF search process achieve 33.2% speedup on the average. On the
other hand, an analysis tool is implemented based on the HED
model. The functionalities of this tool are presented in detail.
|