outer.new Inner(){} in static method throws IndexOutOfBoundsException

[jeffrey-easyesi]

package nullawaytest;
class Outer {
    class Inner {}
    static Inner f(Outer outer) { return outer.new Inner() {}; }
}

Using JDK 21 and NullAway 0.12.6:

Outer.java:4: error: An unhandled exception was thrown by the Error Prone static analysis plugin.
    static Inner f(Outer outer) { return outer.new Inner() {}; }
                                               ^
     Please report this at https://github.com/google/error-prone/issues/new and include the following:
  
     error-prone version: 2.37.0
     BugPattern: NullAway
     Stack Trace:
     java.lang.IndexOutOfBoundsException: Index: 0, Size: 0
        at jdk.compiler/com.sun.tools.javac.util.List.get(List.java:476)
        at com.uber.nullaway.NullAway.handleInvocation(NullAway.java:1986)
        at com.uber.nullaway.NullAway.matchNewClass(NullAway.java:453)
        at com.google.errorprone.scanner.ErrorProneScanner.processMatchers(ErrorProneScanner.java:509)
        ...

Analysis

When the outer.new Inner(){} syntax is used, the constructor symbol has an extra formal parameter.
b3646ef added some special-case code to work around this issue, but it does not take effect when in a static method (methodSymbol.enclClass().isInner() returns false - I think the anonymous subclass of Inner` is not itself considered an inner class)

Interestingly, this bug does not occur when Inner's constructor has explicit parameters, because in that case, matchNewClass uses the superclass's constructor instead:

NullAway/nullaway/src/main/java/com/uber/nullaway/NullAway.java

Line 445 in 649f25a

if (tree.getClassBody() != null && actualParams.size() > 0) {

    if (tree.getClassBody() != null && actualParams.size() > 0) {
      // passing parameters to constructor of anonymous class
      // this constructor just invokes the constructor of the superclass, and
      // in the AST does not have the parameter nullability annotations from the superclass.
      // so, treat as if the superclass constructor is being invoked directly
      // see https://github.com/uber/NullAway/issues/102
      methodSymbol = getSymbolOfSuperConstructor(methodSymbol, state);
    }

So I think you could just remove the actualParams.size() > 0 check, and this would handle all cases, making the special-case in handleInvocation unnecessary.

Vaguely-related minor bug I noticed while looking at this code

matchNewClass also should check that tree.getEnclosingExpression(), if it exists, is non-nullable. outer.new ... will throw NPE if outer is null.

[msridhar]

Thank you for the detailed report and diagnosis! #1191 fixes this issue pretty much exactly as you suggested (along with the related minor bug)