[ 
https://issues.apache.org/jira/browse/GROOVY-12363?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18112039#comment-18112039
 ] 

ASF GitHub Bot commented on GROOVY-12363:
-----------------------------------------

Copilot commented on code in PR #2887:
URL: https://github.com/apache/groovy/pull/2887#discussion_r3943607395


##########
src/main/java/org/codehaus/groovy/transform/sc/StaticCompilationVisitor.java:
##########
@@ -268,14 +268,27 @@ public void visitMethodCallExpression(final 
MethodCallExpression call) {
      */
     @Override
     public void visitConstructorCallExpression(final ConstructorCallExpression 
call) {
-        super.visitConstructorCallExpression(call);
-
-        if (call.isUsingAnonymousInnerClass() && 
call.getType().getNodeMetaData(StaticTypeCheckingVisitor.class) != null) {
+        if (call.isUsingAnonymousInnerClass()) {
             ClassNode anonType = call.getType();
-            anonType.putNodeMetaData(STATIC_COMPILE_NODE, 
anonType.getEnclosingMethod().getNodeMetaData(STATIC_COMPILE_NODE));
-            anonType.putNodeMetaData(WriterControllerFactory.class, 
anonType.getOuterClass().getNodeMetaData(WriterControllerFactory.class));
+            if (anonType.getNodeMetaData(STATIC_COMPILE_NODE) == null) {
+                // GROOVY-6925: an anonymous inner class follows its enclosing 
method
+                // (visitClass has already decided for the inner classes of an 
annotated
+                // class). GROOVY-12363: decide before the type checker visits 
the body
+                // below, otherwise its methods are checked while still 
counting as
+                // dynamic (no direct call targets, no 
SUPER_MOP_METHOD_REQUIRED) and yet
+                // the class is generated as statically compiled, which omits 
the super$
+                // MOP bridges a dynamic super call needs.
+                MethodNode enclosingMethod = anonType.getEnclosingMethod();
+                boolean isSC = enclosingMethod != null
+                        ? isStaticallyCompiled(enclosingMethod)
+                        : isStaticallyCompiled(getEnclosingDeclaration());
+                anonType.putNodeMetaData(STATIC_COMPILE_NODE, isSC);
+                anonType.putNodeMetaData(WriterControllerFactory.class, 
anonType.getOuterClass().getNodeMetaData(WriterControllerFactory.class));
+            }

Review Comment:
   This null guard can preserve an earlier `false` class-level decision even 
when the anonymous class is inside a more-specific `@CompileStatic` method. For 
example, `@CompileDynamic class C { @CompileStatic Runnable m() { new 
Runnable() { void run() { ... } } } }` is supported (the most-specific 
annotation wins), but the outer-class visit records 
`STATIC_COMPILE_NODE=false`; this method-level visit then refuses to change it 
before checking the body, recreating the dynamic-body/static-writer mismatch. 
Recompute the value from the enclosing method whenever this constructor is 
actually visited; skipped methods do not reach this hook.





> Anonymous inner class declared in a @CompileStatic method: super call fails 
> with MissingMethodException (no MOP bridge generated)
> ---------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: GROOVY-12363
>                 URL: https://issues.apache.org/jira/browse/GROOVY-12363
>             Project: Groovy
>          Issue Type: Bug
>            Reporter: Paul King
>            Priority: Major
>
> An anonymous inner class declared inside a method annotated with 
> {{@CompileStatic}} (the enclosing *class* is not annotated) that calls 
> {{super.someMethod(...)}} fails at run time:
> {noformat}
> groovy.lang.MissingMethodException: No signature of method: publish for 
> class: java.util.logging.StreamHandler is applicable for argument types: 
> (java.util.logging.LogRecord) values: [java.util.logging.LogRecord@783a467b]
> Possible solutions: publish(java.util.logging.LogRecord), 
> publish(java.util.logging.LogRecord)
>       at 
> org.codehaus.groovy.runtime.ScriptBytecodeAdapter.unwrap(ScriptBytecodeAdapter.java:85)
>       at 
> org.codehaus.groovy.runtime.ScriptBytecodeAdapter.invokeMethodOnSuperN(ScriptBytecodeAdapter.java:156)
>       at Holders$1.publish(Repro.groovy:6)
> {noformat}
> h3. Reproducer
> {code:groovy}
> import java.util.logging.*
> class Holders {
>     @groovy.transform.CompileStatic
>     static Logger csLogger() {
>         def l = Logger.getLogger('cs'); l.useParentHandlers = false
>         l.addHandler(new StreamHandler(System.out, new SimpleFormatter()) { 
> void publish(LogRecord r) { super.publish(r); flush() } })
>         l
>     }
>     static Logger dynLogger() {   // identical, no annotation
>         def l = Logger.getLogger('dyn'); l.useParentHandlers = false
>         l.addHandler(new StreamHandler(System.out, new SimpleFormatter()) { 
> void publish(LogRecord r) { super.publish(r); flush() } })
>         l
>     }
> }
> Holders.dynLogger().info('works')
> Holders.csLogger().info('fails')   // MissingMethodException
> {code}
> Putting {{@CompileStatic}} on the class {{Holders}} instead of the method 
> makes it work (the super call then compiles to {{invokespecial}}). The JDK 
> superclass is not essential; any {{super}} call from such an anonymous class 
> takes the same path.
> h3. Root cause
> The two anonymous classes compile to *identical* bytecode for the super call, 
> {{ScriptBytecodeAdapter.invokeMethodOnSuperN(Holders$1.class, this, 
> "publish", args)}}, i.e. a dynamic MOP super call. That call only works if 
> the class carries the compiler-generated {{super$N$publish}} bridge, which 
> the MOP uses to reach the superclass implementation 
> ({{MetaMethodIndex.methodsForSuper}}). {{javap}} shows the difference:
> {noformat}
> Holders$1 (in @CompileStatic method):  public void publish(LogRecord)
> Holders$2 (in dynamic method):         public void publish(LogRecord)
>                                        public void super$3$publish(LogRecord)
> {noformat}
> With no bridge, {{MetaClassImpl.getSuperMethodWithCaching}} finds 
> {{methodsForSuper == null}} and the call ends in {{invokeMissingMethod}}.
> The inconsistency comes from the order in which {{StaticCompilationVisitor}} 
> marks things:
> # {{visitClass(outer)}} marks every inner class, including the anonymous one, 
> with {{STATIC_COMPILE_NODE = false}} because the outer class is not 
> statically compiled (the {{isSC}} computation over 
> {{node.getInnerClasses()}}).
> # The type checker then visits the anonymous class body from the constructor 
> call. At that point {{isStaticallyCompiled(publish)}} resolves through the 
> declaring class to {{false}}, so the {{super.publish(r)}} call gets no 
> {{DIRECT_METHOD_CALL_TARGET}} and no {{SUPER_MOP_METHOD_REQUIRED}} entry is 
> recorded.
> # Only afterwards does {{visitConstructorCallExpression}} flip the anonymous 
> class to {{STATIC_COMPILE_NODE = true}} (copied from the enclosing method) 
> and give it the static {{WriterControllerFactory}}.
> # At class generation the class is therefore treated as statically compiled: 
> {{StaticTypesWriterController}} uses {{StaticCompilationMopWriter}}, which 
> emits bridges only for the (empty) {{SUPER_MOP_METHOD_REQUIRED}} list, while 
> the method body, lacking a direct target, falls back to the dynamic 
> {{invokeMethodOnSuperN}}.
> So the class is compiled as "static" for the purpose of omitting MOP bridges 
> and as "dynamic" for the purpose of the call itself.
> h3. Expected
> Either outcome is acceptable as long as both halves agree: the anonymous 
> class body should be statically compiled like the enclosing method (matching 
> the class-level {{@CompileStatic}} behaviour and giving an 
> {{invokespecial}}), or, if it stays dynamic, the ordinary {{MopWriter}} must 
> still generate the {{super$}} bridges. Marking the anonymous class from the 
> enclosing method's flag *before* its body is type-checked (in 
> {{visitConstructorCallExpression}} ahead of 
> {{super.visitConstructorCallExpression}}, or when {{visitClass}} marks inner 
> classes) would give the first.
> Verified on Groovy 6.0.0-SNAPSHOT master and 5.1.2 on JDK 21, and inside a 
> GraalVM 25.2.4 native image where it was first noticed.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to