OK, as promised, I whipped up some crater numbers. The result: 14 crates (0.6%) use constants that fall outside the “intersection” semantics (iow, constants whose types are not ints, floats, or tuples thereof). I plan to analyze the uses of those 14 a bit more but have not had time.
Here is the report and the branch that I tested. The interesting thing in this report is the total regressions – this measurement was done in such a way that constants which appeared in dependencies generated warnings, but constants appearing in the “main crate” being tested reported errors, so we can accurately assess the total number of crates that will need to be changed.