Infinite precision intermediate arithmetic: how much would break?

This is, btw, essentially what C does, except the "infinite" precision is just to long int, and narrowing coercions back are implicit.

Realistically, this proposal hinges on the ability of the compiler to prove non-overflow in "typical uses", and that would require importing a lot of machinery to the front end from LLVM, be difficult to write a specification for, and would probably fail often enough that people would get massively frustrated when their i += 1 doesn't work and has to be rewritten as i = (i + 1).unwrap() or something. Proving anything about basic arithmetic is, of course, undecidable in the fully general case and NP-complete in somewhat less general cases; the optimizer gets away with it because it works strictly on a best effort basis. But surface language semantics cannot be best-effort.