Occasionally I like the idea of producing some sort of dynamic instrumentation code (I privately think of it as "shitgrind") that detects dynamic code that is d

Occasionally I like the idea of producing some sort of dynamic instrumentation code (I privately think of it as "shitgrind") that detects dynamic code that is doing redundant work and flags it. Of course, this doesn't tell you that the code is useless (i.e. I might have to initialize something that subsequently isn't used down some deep and not-anticipatable-at-compile-time set of branches), but it might provide hints that I'm doing a lot of writes that never match up with reads, or that the writes/calculations that I'm doing are going into places that already have that value. It would be incredibly slow but there are plenty of uses I can imagine where you could detect performance problems in a toy version of the problem (maybe running 100-1000x slower) that are the same problems that you would have in the real version of the problem at full scale. Would be a fun project, IMO.