Regression in `ChangeType` recipe

[BoykoAlex]

Consider the following unit test:

    @Test
    void testOkWithTopLevelType1() {
        rewriteRun( //language=java
          spec -> spec.recipe(new ChangeType("java.util.Map$Entry", "java.util.List", true)),
          java("""
               import java.util.Map;

               public class TestController {

                   public Map.Entry respond() {
                      return null;
                   }
               }
               """,
            """
               import java.util.List;

               public class TestController {

                   public List respond() {
                      return null;
                   }
               }
            """)
        );
    }

This test fails because actual results are:

public class TestController {

    public java.util.List respond() {
       return null;
    }
}

New type name appears fully qualified. I've traced this dow to this change

The change in the J class:

        public boolean isFullyQualifiedClassReference(String className) {
            if (getName().getFieldType() == null && getName().getType() instanceof JavaType.FullyQualified &&
                !(getName().getType() instanceof JavaType.Unknown) &&
                TypeUtils.fullyQualifiedNamesAreEqual(((JavaType.FullyQualified) getName().getType()).getFullyQualifiedName(), className)) {
                return true;
            } else if (!className.contains(".")) {
                return false;
            }
            return isFullyQualifiedClassReference(this, TypeUtils.toFullyQualifiedName(className), className.length());
        }

If the first if block is commented out then it works. There is some kind of mistake there it seems... I don't see where this - the field access node content is used there... just its type is used.

[knutwannheden]

Still work in progress, but I started a PR here:

• ChangeType: Use unqualified name if possible #5080.

[knutwannheden]

I wonder if the recipe should try to keep fully qualified types as fully qualified types. In your example that isn't the case, as it is "just" a reference to a nested type.