Type becomes `@UnknownInitialization` after switch

[cushon]

The following program is accepted by the nullness checker in 3.38.0, and rejected in 3.39.0.

Is this a regression or a deliberate change?

enum E {
  FOO;

  E getIt() {
    return FOO;
  }

  String go() {
    E e = getIt();
    switch (e) {
      case FOO:
        return "foo";
    }
    throw new AssertionError(e);
  }
}

$ ./checker-framework-3.38.0/checker/bin/javac -processor Nullness E.java

$ ./checker-framework-3.39.0/checker/bin/javac -processor Nullness E.javaE.java:14: error: [argument] incompatible argument for parameter detailMessage of AssertionError constructor.
    throw new AssertionError(e);
                             ^
  found   : @UnknownInitialization @Nullable E
  required: @Initialized @Nullable Object
1 error

[smillst]

I improved the handling of switches in 3.39.0, so now dataflow treats throw new AssertionError(e); as dead code and does not analysis it. So the type of e in new AssertionError(e) is not refined and the checker issues a false positive error. I've fixed this in #6246 by making it so that checkers don't issue warnings on dead code. This fix will be in the next release.