For the counterpoint from the late great Joe Armstrong,
> Rewrite from scratch with zero external dependencies - always the fastest way (or at least the most fun way) - it increases understanding. [1]
I don't have a permanent viewpoint on this, I kind of bounce back and forth. I definitely think that whole rewrites are important for getting the module boundaries right, and I think it's important to have seen a problem from a bunch of different perspectives, ideally some of them really bad and borderline outrageous, before you commit to one solution. (“This problem is NP-complete, you just want to brute force it?!” -- yeah it turned out that in practice n was no greater than 9 and you could in fact easily hardcode an upper limit and try all possibilities to find the best one. But even if you don't get lucky, the fact that ridiculous ideas came to mind suggests that all the good ideas were on the table at least once.)
Right now I am more inclined to enjoy rewrites because I am working on a project which itself was a rewrite, and I would like to rewrite it back to something more like what it used to be, hah. If I do get my way you can bet I will turn coat, “we can't rewrite this now, don't you know how expensive rewriting is?!”
