In practice, scientific software is always work-in-progress in general, people who write it are in many cases novices and software remains unattended after the results have been published. Although scientists should learn to write good programs, we shouldn't expect them to write the best software; a lot of science is not engineering.