While looking into #46461, it occurred to me that type-checking the following code will panic:

package p

type T[U interface{ M() T[U, int] }] int

type X T[int]

The reason for this is that, while we get an error for the invalid instantiation T[U, int] while type checking U's constraint, when instantiating T[int] we try to substitute in T[U, int] (at which point we won't have a Checker) (edit: this ended up not being as complicated as initially feared).

Working on a fix. We can perhaps skip substituting *Named types with invalid underlying.

CC @griesemer