Doing a good job of constant prop means doing a thorough job of implementing the best algorithm in the best paper. And to do that, you need to build a form of SSA. And to do that you need to compute dominators. And to do that, you need to build an appropriate form of the CFG. Don't need a PhD to do all those things (not inventing anything new here, just using other peoples' beautiful results), but it's still plenty of work to build a fast & thorough implementation, with lots of testing required.
Can't possibly be economical; but of course, the experience will teach you a lot.
Can't possibly be economical; but of course, the experience will teach you a lot.