A DESCRIPTION OF THE PROBLEM :
The switch expression is claimed to be inexhaustive, even though it covers every case.

We have to consider all the combinations of parameters c,d of record R_D.

If we only expand c twice, we get either R_A(R_A(...),...) or R_A(R_B(),...), as c is of type R_A and R_A's first parameter a is of type I_A. Since R_A and R_B are the only two records that extend I_A, this is all we have to consider.

R_A(R_A... is trivially covered in case 1.

R_A(R_B is more complex, it is only paired up with R_A(R_B in case 2, but it would have to be paired up with every possible implementation of I_A (as that is the type of parameter d).

R_A and R_B implement I_A. R_B is covered in case 4. (var a can be R_A(R_B )
For R_A we have the same two possibilities as before, namely R_A(R_B and R_A(R_A.

Both of them are covered in cases 2 and 3.

With that the switch statement covers all possible values and is exhaustive

STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
Run the source code with javac 24.

EXPECTED VERSUS ACTUAL BEHAVIOR :
EXPECTED -
The compiler should not issue an error for a switch expression that does not cover all possible input values.
ACTUAL -
It reports an error

---------- BEGIN SOURCE ----------
class TestClass {
    sealed interface I_A {
    }

    sealed interface I_B {
    }

    record R_A(I_A a, I_B b) implements I_A {
    }

    record R_B() implements I_A {
    }

    record R_D(R_A c, I_A d) implements I_B {
    }

    public static void main(String args[]) {

        I_B v_a = null;
        Integer v_b = switch (v_a) {
            case R_D(R_A(R_A(var a, var b), var c), var d) -> 1;

            case R_D(R_A(R_B(), var b), R_A(R_B(), var x)) -> 2;

            case R_D(var a, R_A(R_A(var b, var c), var d)) -> 3;
            case R_D(var a, R_B()) -> 4;
        };
    }

};
---------- END SOURCE ----------

FREQUENCY : always