IssueTracker

Ok, there are multiple issues related to this.

Basically multiple enum optimizations assume the result of toString() on an enum field is the field name. This is invalid in the context of library being minified, and then compiled with another program.

After discussing about this problem, we concluded that:

Without any specific keep rules, the following main method should always be correct, assuming toString() is not overriden and despite multiple compilation (shrunk lib and shrunk app using shrunk lib):

enum MyEnum {FIRST, SECOND};

class Main {
  public static void main (String[] args) {  
  System.out.print(MyEnum.valueOf(MyEnum.FIRST.toString()));
}

That is not the case at this point, and that should be fixed.

However, the following main method should be correct only if there is a keep rule on the enum fields:

enum MyEnum {FIRST, SECOND};

class Main {
  public static void main (String[] args) {  
  System.out.print(MyEnum.valueOf("FIRST"));
}

Since valueOf uses internally reflection on the field name. That is already not working, and that is the expected behavior.

assert "FIRST".equals(FIRST.toString())

is also not valid code, unless the fields are kept.

Some application use for this the following keep rule to keep the enum fields:

-keepclassmembers enum * {
          <fields>;
           public static **[] values();
           public static ** valueOf(java.lang.String);
          }

We can also note that the same problem is present for both the methods name and toString.

In this case, the issue is slightly more nuanced. The enum usage is entirely within the library. The enum is not exposed to the application. Additionally, the library was compiled with an older version of r8, which didn't have enum.name inlining. When the app is now compiled with a newer version of r8, it inlines the usage even in library classes. So, the keep rule in the application code doesn't matter. So, other than using an older version of r8, the only other option left is to tell the library developers to recompile the library with a newer r8. This may not always be possible, depending the on the library author.

To go back to your example, in the method:

public String findMessage(String key) {
        if (key == null) {
            return null;
        }else if (key.contains(MyEnum.FIRST.name())) {
            return MyEnum.FIRST.getMessage();
        } else if (key.contains(MyEnum.SECOND.name())) {
            return MyEnum.SECOND.getMessage();
        } else if (key.contains(MyEnum.THIRD.name())) {
            return MyEnum.THIRD.getMessage();
        } else {
            return "No match found";
        }
    }

The result of MyEnum.FIRST.name() is not guaranteed to be FIRST, since the implementation of name() internally uses reflection on MyEnum to return the field name, and there is no keep rule on MyEnum fields, so that field name can be changed to anything by R8. So the library implementor has to insert something like:

-keep enum MyEnum { <fields>; }

or

-keepclassmembers enum * { <fields>; }

for its library code to work.

If that code used to work in a version of R8, it's just coincidental due to other keep rules or unexploited optimization potential.

Sorry Guptan, I got confused. Your code is supposed to be working.

We have a bug in the Enum#toString / Enum#name optimizations. I'm working on a fix.

Thank you Clement!

Backports to Studio 4.0 and 4.1: https://googleplex-android-review.git.corp.google.com/c/platform/prebuilts/r8/+/12108071 https://googleplex-android-review.git.corp.google.com/c/platform/prebuilts/r8/+/12108069

It will get into Android studio 4.0/4.1/4.2, timeline depends on various release schedules, and some reviewers are on vacation, so I can't really give an estimate.