Radix sorting methods have excellent asymptotic performance on string data, for which comparison is not a unit-time operation. Attractive for use in large byte-addressable memories, these methods have nevertheless long been eclipsed by more easily programmed algorithms. Three ways to sort strings by bytes left to right---a stable list sort, a stable two-array sort, and an in-place "American flag" sort---are illustrated with practical C programs. For heavy-duty sorting, all three perform comparably, usually running at least twice as fast as a good quicksort. We recommend American flag sort for general use.

As software systems become more ambitious and extensive, testing them for correctness or performance becomes more challenging, especially when the systems themselves provide a general computing environment, not just a few narrowly defined services. This paper describes some testing experience with S, a large system for data analysis and scientific computing. Included are: 1. comparisons and contrasts with general software testing issues; 2. description of tools and techniques for testing S; 3. examples drawn from recent experience. The paper discusses issues in software testing, with particular reference to testing languages and systems for data analysis. Conversely, some of the examples argue for the importance of flexible techniques for data generation and analysis as support for testing.

Abstract — Multiple Pivot Sort is a new comparisonbased sorting algorithm that has been developed to address shortcomings in current popular comparison-based sorting algorithms. The goal of this work is to perform an extensive empirical study of Multiple Pivot Sort against other established sorting algorithms including Quick Sort, Merge Sort, and Heap Sort. This research presents data to support the hypothesis that Multiple Pivot Sort is an extremely promising new algorithm in a critical field of Computer Science. The empirical study includes comparisons of the various sorts using randomly generated arrays of size 10 data items up to 1 million or more data items. In addition, integer arrays in which the data is in order and in reverse order are used in the study. Data items include integers, strings, and classes. The study keeps track of the number of comparisons, the number of data moves, and CPU time specific to the machine for each sort. Results of the empirical study performed in this research support the claim that Multiple Pivot Sort may well be the fastest sorting algorithm in existence.