I recently received a question from a reader who asked how to sort a large file. My solution was based on the qsort() algorithm, which is the only sorting algorithm defined by ANSI C. Another reader who saw my suggested solution quipped that qsort() was inadequate for that particular task because it’s slower than “insertion sort.” Although technically he was correct, he failed to consider the following crucial factors: the application was to be activated as a batch process only once a month. Therefore, its speed wasn’t paramount. Furthermore, any sorting algorithm other than qsort() must be written from scratch, with the likelihood of introducing accidental bugs and inefficiencies due to programmers’ lack of experience. Worse yet, a homemade algorithm could turn into a maintenance problem in the future and might not be portable. Overall, the suggestion to avoid qsort() and use another algorithm was a clear case of premature optimization.
Too many programmers and designers waste precious time on premature optimization. This is why in many computer sciences courses and books on code optimization, the first rule is “don’t!” In real world programming, performance usually isn’t the most important factor—correctness, development time and costs, and software’s stability are at a higher priority. Only when all these goals have been accomplished can you consider optimization, if and only if it’s truly necessary.