On JDK 8, the Set/Map/List.of(...) MethodHandles are null at class load time, making the invokeExact branch unreachable in tests. On JDK 9+, the null-guard branch is never hit. Neither branch achieves full coverage under a single JDK version.

Approach

Extract the MethodHandle dispatch logic from each public factory method into a package-private helper that accepts the handle as an explicit parameter. The public method becomes a one-liner delegate.

Before:

public static <E> Set<E> ofSet() {
    if (of0MethodHandle == null) {
        return emptySet();
    }
    try {
        return (Set<E>) of0MethodHandle.invokeExact();
    } catch (Throwable e) {
        return emptySet();
    }
}

After:

public static <E> Set<E> ofSet() {
    return ofSet0(of0MethodHandle);
}

static <E> Set<E> ofSet0(MethodHandle methodHandle) {
    if (methodHandle == null) {
        return emptySet();
    }
    try {
        return (Set<E>) methodHandle.invokeExact();
    } catch (Throwable e) {
        return emptySet();
    }
}

Tests can now exercise both branches unconditionally by injecting null or a type-incompatible handle:

@Test
void testOfSet0WithNull() {
    assertEquals(emptySet(), ofSet0(null));
}

@Test
void testOfSet0WithWrongMethodHandle() {
    MethodHandle wrongMethodHandle = findPublicStatic(Collections.class, "emptySet");
    assertEquals(emptySet(), ofSet0(wrongMethodHandle));
}

Changes

• Sets.java — extracted ofSet0–ofSet10 and ofSetElements helpers
• Maps.java — extracted ofMap0–ofMap10 and ofMapEntries helpers
• Lists.java — extracted ofList0–ofList10 and ofListElements helpers
• SetsTest / MapsTest / ListsTest — added null + wrong-handle tests for every helper (114 tests total)