Currently the MIR type checker has some kind of hacky code to figure out what sorts of well-formedness constraints are present on a constant. The values of constants now are "post normalization" and don't give us ready access to what we need. As the comment from the code says:
// FIXME(#46702) -- We need some way to get the predicates
// associated with the "pre-evaluated" form of the
// constant. For example, consider that the constant
// may have associated constant projections (`<Foo as
// Trait<'a, 'b>>::SOME_CONST`) that impose
// constraints on `'a` and `'b`. These constraints
// would be lost if we just look at the normalized
// value.
Currently the MIR type checker has some kind of hacky code to figure out what sorts of well-formedness constraints are present on a constant. The values of constants now are "post normalization" and don't give us ready access to what we need. As the comment from the code says: