[4.8 Regression] gjavah throws an exception

[4.8 Regression] gjavah throws an exception

[Andrew Haley]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55716

This bug is caused by http://cvs.savannah.gnu.org/viewvc/classpath/tools/gnu/classpath/tools/javah/Main.java?root=classpath&r1=1.13&r2=1.14

Automagic handling of inner classes was added, but in a bogus way.
If a class has inner classes, those inner classes are searched on
the classpath.  This is wrong if the argument to javah is a filename:
the inner class may not exist on the classpath, and even if they do
they will be the wrong versions.

IMO, if given filenames gjavah should not search for inner classes
at all.  This patch does so.

I have another patch that finds the correct inner class files for a
a class, but this is better.  As a whitespace diff it is simply

Index: Main.java
===================================================================
--- Main.java	(revision 194553)
+++ Main.java	(working copy)
@@ -370,6 +370,8 @@
         results.put(filename, klass);
         parsed.add(item.toString());

+        if (! (item instanceof File))
+          {
         // Check to see if there are inner classes to also parse
         Iterator<?> innerClasses = klass.innerClasses.iterator();
         HashSet<Object> innerNames = new HashSet<Object>();
@@ -381,6 +383,7 @@
           }
         results.putAll(parseClasses(innerNames.iterator()));
       }
+      }
     return results;
   }

Andrew.


2012-12-17  Andrew Haley  <aph@redhat.com>

	* tools/gnu/classpath/tools/javah/Main.java (parseClasses): Dont
	scan inner classes if our item is a file.

Index: Main.java
===================================================================
--- Main.java	(revision 194553)
+++ Main.java	(working copy)
@@ -370,16 +370,19 @@
         results.put(filename, klass);
         parsed.add(item.toString());

-        // Check to see if there are inner classes to also parse
-        Iterator<?> innerClasses = klass.innerClasses.iterator();
-        HashSet<Object> innerNames = new HashSet<Object>();
-        while (innerClasses.hasNext())
+        if (! (item instanceof File))
           {
-            String innerName = ((InnerClassNode) innerClasses.next()).name;
-            if (!parsed.contains(innerName))
-              innerNames.add(innerName);
+            // Check to see if there are inner classes to also parse
+            Iterator<?> innerClasses = klass.innerClasses.iterator();
+            HashSet<Object> innerNames = new HashSet<Object>();
+            while (innerClasses.hasNext())
+              {
+                String innerName = ((InnerClassNode) innerClasses.next()).name;
+                if (!parsed.contains(innerName))
+                  innerNames.add(innerName);
+              }
+            results.putAll(parseClasses(innerNames.iterator()));
           }
-        results.putAll(parseClasses(innerNames.iterator()));
       }
     return results;
   }

Re: [cp-patches] [4.8 Regression] gjavah throws an exception

[Mark Wielaard]

On Mon, 2012-12-17 at 15:39 +0000, Andrew Haley wrote:
> If a class has inner classes, those inner classes are searched on
> the classpath.  This is wrong if the argument to javah is a filename:
> the inner class may not exist on the classpath, and even if they do
> they will be the wrong versions.
> 
> IMO, if given filenames gjavah should not search for inner classes
> at all.  This patch does so.
> [...]
> 2012-12-17  Andrew Haley  <aph@redhat.com>
> 
> 	* tools/gnu/classpath/tools/javah/Main.java (parseClasses): Dont
> 	scan inner classes if our item is a file.

That looks right to me. As you said on irc, no other javah
implementation seems to support "raw class files" as input. Others only
seem to accept class names. So there is no "best practice" to follow in
this case. Not trying to search for inner classes when a class file is
provided seems a good choice since at the bytecode level there really
isn't such a thing as an inner class (just funnily named classes).

Thanks,

Mark