Do the damn thing from the scratch and do it right,
then replace the old steaming pile of junk with it
They all fail miserably, leaving nothing but a smoking crater and piles of code that is never to be used.
Turns out all those millions of lines of c code written over a period of 25 years actually do quite a lot.
Unfortunately, often, you'll draw exaggerated conclusions from past failures and over-engineer things ("All problems in computer science can be solved by another level of indirection."). You might also end up creating new design problems, e.g. a bottleneck in an area that the old design just happened to avoid. And even if you do get everything right design-wise, you'll always end up creating new bugs along the way.
(Obviously, none of this is a new insight, the pros and cons of rewrite from scratch have been discussed back and forth forever.)
But the opposite may have some weight too. I think I read somewhere that Google codebase had been rewritten from scratch. Maybe it was early enough, because I suppose rewriting it today would be painful. Are there any other successful rewrite from scratch stories? Maybe vim?