shim classes generated incorrectly in 1.8.1 (was fine in 1.8.0) for some static inner-class usages

I suspect this was fixed as part of GROOVY-4939. Any chance you could try a snapshot, otherwise a better example would be good because the one I just created according to above appears to work in trunk. Thanks.

I'm, trying to run the Groovy build and it barfs foe me: some plexus classCastException issue, and I'm timing out on this.

I'll wait for a formal release of 1.8.2 and report then what it feels like.

The uploaded 1.8.2-SNAPSHOT (dated 30th July) still exhibits the same defect.

With the following test, I couldn't reproduce the problem experienced in Gradle, neither with trunk nor 1.8.x:

/**
 * GROOVY-4965: shim classes generated incorrectly in 1.8.1 (was fine in 1.8.0) for some static inner-class usages
 * 
 * @author Guillaume Laforge
 */
class InterfaceInnerEnumUsedAsGenericsParamTest extends StringSourcesStubTestCase {

    Map<String, String> provideSources() {
        [
                'org/gradle/api/file/FileCollection.java': '''
                    package org.gradle.api.file;

                    public interface FileCollection {
                        enum AntType {
                            MatchingTask, FileSet, ResourceCollection
                        }
                    }
                ''',

                'org/gradle/api/tasks/AntBuilderAwareUtil.groovy': '''
                    package org.gradle.api.tasks

                    import org.gradle.api.file.FileCollection

                    class AntBuilderAwareUtil {
                        static def assertSetContains(FileCollection set, Set<String> filenames, Iterable<FileCollection.AntType> types, boolean generic = true) {
                            true
                        }
                    }
                '''
        ]
    }

    void verifyStubs() {
        println stubJavaSourceFor('org.gradle.api.tasks.AntBuilderAwareUtil')
    }
}

I've managed to build trunk using ant, commenting out the hanging tests, ignoring the maven-deploy stage, then doing a manual install later.

The bug is still there with a TRUNK checkout from moments ago.

Can someone direct me to the class that writes shim classnames? I'll add a .replace("$",".") where appropriate, and report back.

Hi Paul, you can look at org.codehaus.groovy.tools.javac.JavaStubGenerator.
And you can look at the printParams() method, which is most probably where the action takes place – if it's in the types of the parameters of your methods that the problem arises.
Ideally, it would be good to have a reproducable test case like the one I pasted in my previous comment, so that we know it won't happen again once fixed.

Thanks Guillaume,

Line 589 or so, genericsTypes[i].toString() should be genericsTypes[i].toString().replace("$",".")
Either that or reuse/recurse into printType(..) as the same replacement is done elsewhere in the code, but not for this particular path.

I can confirm, that my issue with live client code (hence it's not so easy to share) disappeared.

Paul, you made me delete a very long comment You are obviously right that printType should be reused here. Can you make a patch that works for you Paul?

Sounds like y'all are on the cusp of a change that doesn't lower test coverage

However, answering the questions posed in the deleted comment:

genericsTypes[i].getType().isGenericsGenericsPlaceHolder() is false.
genericsTypes[i].toString() looks like so my.package.MyClass$MyInnerClass

Here's the one-liner diff, that aims at reuse of existing logic (that does the right thing). Tested of course:

Index: src/main/org/codehaus/groovy/tools/javac/JavaStubGenerator.java
===================================================================
--- src/main/org/codehaus/groovy/tools/javac/JavaStubGenerator.java    (revision 22747)
+++ src/main/org/codehaus/groovy/tools/javac/JavaStubGenerator.java    (working copy)
@@ -589,7 +589,7 @@
         out.print('<');
         for (int i = 0; i < genericsTypes.length; i++) {
             if (i != 0) out.print(", ");
-            out.print(genericsTypes[i].toString());
+            printTypeName(out, genericsTypes[i].getType());
         }
         out.print('>');
     }

I have been unable to replicate an error case yet but the latest snapshot (I believe) should have the "replace" patch in place - the "printTypeName" patch broke some other behavior so I didn't go that way. It would be interesting to see if the latest snapshot fixes your issue.

close this issue. i moved my inner class to outer.

Just waiting feedback from Peter to see if this is still a problem for Gradle. If not I will close the issue.

I eventually worked around it like Paul (converted the nested class to a top-level class). If you are interested, the groovy-1.8 branch of Gradle is available on GitHub.

Fix to replace '$' with '.' was added but no test case could reproduce the original problem. Assuming fixed.