The part about performance seems surprising to me. Sure, it's slower, but shouldn't the baseline be something like: first run the old borrow checker, and only r

The part about performance seems surprising to me. Sure, it's slower, but shouldn't the baseline be something like: first run the old borrow checker, and only run the new one if it finds a problem? Given that the checking is a function at a time, it seems like that shouldn't even slow down failing cases by all that much. Then you can optimize that by dividing traditional failures into "nope, that's just wrong" vs "might possibly pass Polonius". Polonius Alpha accepts a superset of the programs that the old NLL one accepted, right? Not the greatest for code maintenance, but it's hard to imagine this producing a "2-3x regression" (the worst case observed). And it's not something I"d recommend if it didn't sound like speed is the gating factor for releasing this. (I doubt I'm the first to think of this, so there are probably Very Good Raisins behind the decision to not do this.)