WildcardSuperType has variable size zero

Issue #276 resolved
Jesper Öqvist created an issue

ExtendJ 8.1.0-27-gcd7effa Java SE 5

An expression that has a type like ? super Integer will be treated as having size zero by ExtendJ. This can cause issues in generated bytecode. For example, when calling a method that returns type ? super X, as in the following program:

// Test that executable bytecode is generated when discarding a method
// result from a method that returns some wildcard-super type.
// .result: EXEC_PASS
public class Test {
  public static void main(String[] args) {
    Container<? super Integer> con = new Container<Integer>(123);
    for (int i = 0; i < 1; ++i) {
      con.get();
    }
  }
}

class Container<T> {
  private final T value;

  public Container(T value) {
    this.value = value;
  }

  T get() {
    return value;
  }
}

In this case, there should be a POP instruction after the call to con.get(). However, ExtendJ does not generate a POP which causes a stack height mismatch when branching back to the start of the for-statement. This leads to a verification error when trying to run the compiled program:

    [junit] [FAIL] runTest[run/generics/container_05](tests.extendj.TestJava7)
    [junit] Error output files differ expected:<[]> but was:<[Exception in thread "main" java.lang.VerifyError: (class: Test, method: main signature: ([Ljava/lang/String;)V) Inconsistent stack height 1 != 0
    [junit]     at java.lang.Class.getDeclaredMethods0(Native Method)
    [junit]     at java.lang.Class.privateGetDeclaredMethods(Class.java:2575)
    [junit]     at java.lang.Class.getMethod0(Class.java:2816)
    [junit]     at java.lang.Class.getMethod(Class.java:1665)
    [junit]     at sun.launcher.LauncherHelper.getMainMethod(LauncherHelper.java:418)
    [junit]     at sun.launcher.LauncherHelper.checkAndLoadMain(LauncherHelper.java:410)]>

Comments (1)

  1. Jesper Öqvist reporter

    Fix error in TypeDecl.variableSize()

    This changes TypeDecl.variableSize() to have the default value 1.

    This fixes an error where some types that should have size 1 were treated as having size 0, which could lead to missing pop instructions (and other errors) in generated bytecode.

    fixes #276 (bitbucket)

    → <<cset 1cac6f166899>>

  2. Log in to comment