The use of software is pervasive in all fields of science. Associated software development efforts may be very large, long-lived, and complex, requiring the commitment of significant resources. However, several authors have argued that the "gap" or "chasm" between software engineering and scientific programming is a serious risk to the production of reliable scientific results, as demonstrated in a number of case studies. This article reviews the research that addresses the gap, exploring how both software engineering and research practice may need to evolve to accommodate the use of software in science.