Software "design patterns" seek to package proven solutions to design problems in a form that makes it possible to find, adapt, and reuse them. A common claim is that a design based on properly applied patterns will have fewer defects than more ad hoc solutions. This case study analyzes the weekly evolution and maintenance of a large commercial product (C++, 500,000 LOC) over three years, comparing defect rates for classes that participated in selected design patterns to the code at large. We found that there are significant differences in defect rates among the patterns, ranging from 63 percent to 154 percent of the average rate. We developed a new set of tools able to extract design pattern information at a rate of 3/spl times/10/sup 6/ lines of code per hour, with relatively high precision. Based on a qualitative analysis of the code and the nature of the patterns, we conclude that the Observer and Singleton patterns are correlated with larger code structures and, so, can serve as indicators of code that requires special attention. Conversely, code designed with the Factory pattern is more compact and possibly less closely coupled and, consequently, has lower defect numbers. The Template Method pattern was used in both simple and complex situations, leading to no clear tendency.

Three published techniques for image compression by piecewise linear approximation are described and tested on 2D solar photographic UV spectra. The techniques allow the necessary image analysis for these particular images to be performed directly on the compressed data. The compressed/restored images have been evaluated quantitatively by measuring the preservation of Gaussian spectral line features. Tests indicate that conventional polygonization techniques based on invariant error tolerances can only give a limited compression of the present images before serious errors start to occur. Point-to-point adaptive control of the error tolerance through a continuous local activity measure does not achieve the desired stability and robustness. The authors find that introducing adaptivity through image segmentation allows a significant increase in the compression ratio, making these methods very useful for astronomical purposes. An algorithm for automatically finding spectral lines and fitting Gauss curves has also been developed.>