Incorrect control flow analysis causes statement subsequent to a switch statement to be flagged unreachable under some circumstances

[iloveeclipse]

javac from Java 21 compiles the code below fine.
ecj starting from a589ad2 reports "Unreachable code" compilation error on return.

public class SwitchTest {
    String unreachableCode(String arg) {
        String result;
        switch (arg) {
            case "a" -> {
                result = "A";
            }
            default -> throw new RuntimeException(arg);
        }
        return result;      // <- ecj reports "Unreachable code"
    }
}

Both compiler agree that the code below is fine:

public class SwitchTest {
    String unreachableCode(String arg) {
        String result;
        switch (arg) {
            case "a" -> {
                result = "A";
                break;       // <- this makes ecj happy
            }
            default -> throw new RuntimeException(arg);
        }
        return result;
    }
}

For the "old style" switch both ecj and javac agree that the return is unreachable:

public class SwitchTest {
    String unreachableCode(String arg) {
        String result;
        switch (arg) {
            case "a" : {
                result = "A";
            }
            default : throw new RuntimeException(arg);
        }
        return result;
    }
}

I assume ecj has a regression with a589ad2, but I haven't checked what JLS says here.
@srikanth-sankaran : please check.

[srikanth-sankaran]

Thanks for the report. I can see where the problem originates from. The commit comment says: Eliminate the hack to inject synthetic BreakStatement in label rule switch blocks. Enhance the code generator to handle these explicitly.

To remove the earlier kludgy implementation, it is not enough that the code generator is enhanced to handle these explicitly - we should also delink flow analysis's dependence on the earlier kludge.

PR will follow shortly.