ShortenFullyQualifiedTypeReferences conflicts with Types defined in parent class

[mbruggmann]

What version of OpenRewrite are you using?

rewrite-java 8.31.1

What is the smallest, simplest way to reproduce the problem?

I can reproduce the issue using this test case:

  @Test
  void leaveFullTypeReferenceForSuperclassClash() {
    rewriteRun(
        spec -> spec.recipe(new ShortenFullyQualifiedTypeReferences()),
        srcMainJava(
            java(
                """
                package com.helloworld;

                public interface Parent {
                  class Set {}
                }"""),
            java(
                """
                package com.helloworld;

                public class Main implements Parent {
                  public void get() {
                    final java.util.Set set = null;
                  }
                }""")));
  }

What did you expect to see?

Because a class definition in the parent interface with the same type name shadows the import, the fully qualified type reference is still needed. I would expect the recipe to bail out in this case.

What did you see instead?

  package com.helloworld;
  
  import java.util.Set;
  
  public class Main implements Parent {
    public void get() {
      final Set set = null;  // <-- this variable now became Parent.Set instead of java.util.Set
    }
  }

Are you interested in contributing a fix to OpenRewrite?

Short on time at the moment, possibly later.

[knutwannheden]

This looks tricky to fix. The problem is that we in the type attribution only have the reference from a nested class to the outer class, but not the inverse. This is something we should look into to allow this recipe to be implemented correctly.

[svaningelgem]

I encountered a very similar issue in which we had something along these lines:

import org.example.v1.UnrelatedType1;
import org.example.v1.SomethingElse2;
import org.example.v360.Type1;
import org.example.v360.Type2;
import org.example.v360.Type3;

[ .. and then a lot of methods using things like org.example.v1.Type3 / 4 / ... ]

Let's assume both v1 & v360 have Role defined.

However, we had (in this case) a test defined like:

        when(roleMapper.mapRole(any(java.util.ArrayList.class))).thenReturn(new ArrayList<org.example.v1.Role>());

As you can spot, Role is not imported, but referenced directly. (as were a lot of other things -- bad I know)

Autoconversion made it into:

import org.example.v1.*;
import org.example.v360.*;

        when(roleMapper.mapRole(any(ArrayList.class))).thenReturn(new ArrayList<Role>());

But now Role is not uniquely defined any longer.

I think this is a play between ShortenFullyQualifiedTypeReferences and org.openrewrite.staticanalysis.CodeCleanup or org.openrewrite.java.OrderImports.

If this is sufficiently different from OP's issue, I'll copy/paste this explanation into a new ticket. Just tell me if I need to do so.