Patch: FYI: add another newInstance test

Patch: FYI: add another newInstance test

[Tom Tromey]

Twisti said on irc that the xalan problem is caused by something in
newInstance.

I added a quick test case for this ... but the test already passes on
jamvm, gij, and the JDK.

So perhaps I'm misunderstanding what the bug is actually supposed to
be.  Twisti, any comments?

Tom

2006-02-28  Tom Tromey  <tromey@redhat.com>

	* gnu/testlet/java/lang/Class/newInstance.java (test6): New class.
	(test): Updated comment.  Run test6 test.

Index: gnu/testlet/java/lang/Class/newInstance.java
===================================================================
RCS file: /cvs/mauve/mauve/gnu/testlet/java/lang/Class/newInstance.java,v
retrieving revision 1.5
diff -u -r1.5 newInstance.java
--- gnu/testlet/java/lang/Class/newInstance.java	22 Dec 2005 12:46:02 -0000	1.5
+++ gnu/testlet/java/lang/Class/newInstance.java	28 Feb 2006 18:27:45 -0000
@@ -23,6 +23,8 @@
 
 package gnu.testlet.java.lang.Class;
 
+import java.io.IOException;
+
 import gnu.testlet.Testlet;
 import gnu.testlet.TestHarness;
 
@@ -135,6 +137,29 @@
     }
   }
 
+  public static class test6
+  {
+    public test6() throws IOException
+    {
+      throw new IOException("hi bob");
+    }
+    
+    static void check(TestHarness harness)
+    {
+      boolean ok = false;
+      try
+        {
+    	  test6.class.newInstance();
+        }
+      catch (Throwable t)
+        {
+    	  harness.debug(t);
+    	  ok = t instanceof IOException;
+        }
+      harness.check(ok);
+    }
+  }
+
   public void test(TestHarness harness)
   {
     test1.check(harness);
@@ -150,9 +175,11 @@
     // Just see to it that the following is legal.
     new test5();
     // If new test5() is legal, why should test5.class.newInstance()
-    // throw IllegalAccessException?  But the runtime access control
-    // is somewhat different from the compile time access control.
-    // So it may be OK that Sun's JDK throws IllegalAccessException here.
+    // throw IllegalAccessException?  The reason that it is different is
+    // that 'new test5()' will call a compiler-generated accessor
+    // constructor.  This accessor has package-private access and an
+    // extra argument (to differentiate it from the user-written
+    // constructor).
     checkFail(harness, test5.class);
 
     checkSuccess(harness, test1.inner.class);
@@ -169,6 +196,8 @@
         harness.debug(x);
         harness.fail("test configuration failure");
       }
+    
+    test6.check(harness);
   }
 
   static void checkSuccess(TestHarness harness, Class c)

Re: Patch: FYI: add another newInstance test

[Christian Thalinger]

On Tue, 2006-02-28 at 11:26 -0700, Tom Tromey wrote:
> I added a quick test case for this ... but the test already passes on
> jamvm, gij, and the JDK.
> 
> So perhaps I'm misunderstanding what the bug is actually supposed to
> be.  Twisti, any comments?

The bug is/was actually in ResourceBundle.tryBundle.  We didn't catch
enough exceptions.  Jeroen's patch fixed it and xalan does work now.
Probably we should add an according ResourceBundle test to mauve.

TWISTI