You have faced an edge case here. To clarify what happens let's define the initial conditions:
- you have a Java (or Groovy) class that gets executed inside JVM
- you have two Groovy classes that get loaded outside of the JVM
The problem you have described does not exist if you put these two Groovy classes inside the same path you execute your Java class from - in this case IDE takes care to compile these Groovy classes and put them to the classpath of a JVM that gets started to run your Java test class.
But this is not your case and you are trying to load these two Groovy classes outside the running JVM using GroovyClassLoader (which extends URLClassLoader btw). I will try to explain in the simplest possible words what happened that adding field of type MyOuter does not throw any compilation error, but MyOuter.MyInner does.
When you execute:
Class<?> clazz = groovyScriptEngine.getGroovyClassLoader().loadClass("MyClass");
Groovy class loader goes to script file lookup part, because it was not able to find MyClass in the current classpath. This is the part responsible for it:
// at this point the loading from a parent loader failed
// and we want to recompile if needed.
if (lookupScriptFiles) {
// try groovy file
try {
// check if recompilation already happened.
final Class classCacheEntry = getClassCacheEntry(name);
if (classCacheEntry != cls) return classCacheEntry;
URL source = resourceLoader.loadGroovySource(name);
// if recompilation fails, we want cls==null
Class oldClass = cls;
cls = null;
cls = recompile(source, name, oldClass);
} catch (IOException ioe) {
last = new ClassNotFoundException("IOException while opening groovy source: " + name, ioe);
} finally {
if (cls == null) {
removeClassCacheEntry(name);
} else {
setClassCacheEntry(cls);
}
}
}
Source: src/main/groovy/lang/GroovyClassLoader.java#L733-L753
Here URL source = resourceLoader.loadGroovySource(name); it loads the full file URL to the source file and here cls = recompile(source, name, oldClass); it executes class compilation.
There are several phases involved in Groovy class compilation. One of them is Phase.SEMANTIC_ANALYSIS which analyses class fields and their types for instance. At this point ClassCodeVisitorSupport executes visitClass(ClassNode node) for MyClass class and following line
node.visitContents(this);
starts class contents processing. If we take a look at the source code of this method:
public void visitContents(GroovyClassVisitor visitor) {
// now let's visit the contents of the class
for (PropertyNode pn : getProperties()) {
visitor.visitProperty(pn);
}
for (FieldNode fn : getFields()) {
visitor.visitField(fn);
}
for (ConstructorNode cn : getDeclaredConstructors()) {
visitor.visitConstructor(cn);
}
for (MethodNode mn : getMethods()) {
visitor.visitMethod(mn);
}
}
Source: src/main/org/codehaus/groovy/ast/ClassNode.java#L1066-L108
we will see that it analyses and process class properties, fields, constructors and methods. At this phase it resolves all types defined for these elements. It sees that there are two properties m1 and m2 with types MyOuter and MyOuter.MyInner accordingly, and it executes visitor.visitProperty(pn); for them. This method executes the one we are looking for - resolve()
private boolean resolve(ClassNode type, boolean testModuleImports, boolean testDefaultImports, boolean testStaticInnerClasses) {
resolveGenericsTypes(type.getGenericsTypes());
if (type.isResolved() || type.isPrimaryClassNode()) return true;
if (type.isArray()) {
ClassNode element = type.getComponentType();
boolean resolved = resolve(element, testModuleImports, testDefaultImports, testStaticInnerClasses);
if (resolved) {
ClassNode cn = element.makeArray();
type.setRedirect(cn);
}
return resolved;
}
// test if vanilla name is current class name
if (currentClass == type) return true;
String typeName = type.getName();
if (genericParameterNames.get(typeName) != null) {
GenericsType gt = genericParameterNames.get(typeName);
type.setRedirect(gt.getType());
type.setGenericsTypes(new GenericsType[]{ gt });
type.setGenericsPlaceHolder(true);
return true;
}
if (currentClass.getNameWithoutPackage().equals(typeName)) {
type.setRedirect(currentClass);
return true;
}
return resolveNestedClass(type) ||
resolveFromModule(type, testModuleImports) ||
resolveFromCompileUnit(type) ||
resolveFromDefaultImports(type, testDefaultImports) ||
resolveFromStaticInnerClasses(type, testStaticInnerClasses) ||
resolveToOuter(type);
}
Source: src/main/org/codehaus/groovy/control/ResolveVisitor.java#L343-L378
This method gets executed for both MyOuter and MyOuter.MyInner classes. It is worth mentioning that class resolving mechanism only checks if given class is available in the classpath and it does not load or parse any classes. That is why MyOuter gets recognized when this method reaches resolveToOuter(type). If we take a quick look at its source code we will understand why it works for this class:
private boolean resolveToOuter(ClassNode type) {
String name = type.getName();
// We do not need to check instances of LowerCaseClass
// to be a Class, because unless there was an import for
// for this we do not lookup these cases. This was a decision
// made on the mailing list. To ensure we will not visit this
// method again we set a NO_CLASS for this name
if (type instanceof LowerCaseClass) {
classNodeResolver.cacheClass(name, ClassNodeResolver.NO_CLASS);
return false;
}
if (currentClass.getModule().hasPackageName() && name.indexOf('.') == -1) return false;
LookupResult lr = null;
lr = classNodeResolver.resolveName(name, compilationUnit);
if (lr!=null) {
if (lr.isSourceUnit()) {
SourceUnit su = lr.getSourceUnit();
currentClass.getCompileUnit().addClassNodeToCompile(type, su);
} else {
type.setRedirect(lr.getClassNode());
}
return true;
}
return false;
}
Source: src/main/org/codehaus/groovy/control/ResolveVisitor.java#L725-L751
When Groovy class loader tries to resolve MyOuter type name it reaches
lr = classNodeResolver.resolveName(name, compilationUnit);
which locates script with a name MyOuter.groovy and it creates a SourceUnit object associated with this script file name. It is simply something like saying "OK, this class is not in my classpath at the moment, but there is a source file I can see that once compiled it will provide a valid type of name MyOuter". This is why it finally reaches:
currentClass.getCompileUnit().addClassNodeToCompile(type, su);
where currentClass is an object associated with MyClass type - it adds this source unit to MyClass compilation unit, so it gets compiled with the MyClass class. And this is the point where resolving
MyOuter m1
class property ends.
In the next step it picks MyOuter.MyInner m2 property and it tries to resolve its type. Keep in mind - MyOuter got resolved correctly, but it didn't get loaded to the classpath, so it's static inner class does not exist in any scope, yet. It goes through the same resolving strategies as MyOuter, but any of them works for MyOuter.MyInner class. And this is why ResolveVisitor.resolveOrFail() eventually throws this compilation exception.
Workaround
OK, so we know what happens, but is there anything we can do about it? Luckily, there is a workaround for this problem. You can run your program and load MyClass successfully only if you load MyOuter class to Groovy script engine first:
import java.io.File;
import java.net.MalformedURLException;
import java.net.URL;
import groovy.util.GroovyScriptEngine;
public class TestGroovyScriptEngine {
public static void main(String[] args) throws MalformedURLException, ClassNotFoundException {
final File myGroovySourceDir = new File("C:/MyGroovySourceDir");
final URL[] urls = { myGroovySourceDir.toURL() };
GroovyScriptEngine groovyScriptEngine = new GroovyScriptEngine(urls,
Thread.currentThread().getContextClassLoader());
groovyScriptEngine.getGroovyClassLoader().loadClass("MyOuter");
Class<?> clazz = groovyScriptEngine.getGroovyClassLoader().loadClass("MyClass");
}
}
Why does it work? Well, semantic analysis of MyOuter class does not cause any problems, because all types are known at this stage. This is why loading MyOuter class succeeds and it results in Groovy script engine instance knows what MyOuter and MyOuter.MyInner types are. So when you next load MyClass from the same Groovy script engine it will apply different resolving strategy - it will find both classes available to the current compilation unit and it wont have to resolve MyOuter class based on its Groovy script file.
Debugging
If you want to examine this use case better it is worth to run a debugger and see analyze what happens at the runtime. You can create a breakpoint at line 357 of ResolveVisitor.java file for instance, to see described scenario in action. Keep in mind one thing though - resolveFromDefaultImports(type, testDefaultImports) will try to lookup MyClass and MyOuter classes by applying default packages like java.util, java.io, groovy.lang etc. This resolve strategy kicks in before resolveToOuter(type) so you have to patiently jump through them. But it is worth it to see and get a better understanding about how things work. Hope it helps!
