I'm checking this in.
Lillian and others have run into problems with our URLClassLoader not
handling the Class-Path attribute correctly. I caused this regression
when I implemented index.list handling.
I tracked the problem down to 2 bugs. One is that we weren't clearing
the index map if we didn't see an index. Fixing this exposed another
bug, namely that sometimes a Class-Path attribute in a jar will
include the jar itself (this occurs in the jfreechart demo, for
instance), yielding infinite recursion.
Tom
2006-06-14 Tom Tromey <[EMAIL PROTECTED]>
* gnu/java/net/loader/JarURLLoader.java (initialize): Skip our own
jar.
* gnu/java/net/IndexListParser.java (IndexListParser): Call clearAll
when index entry does not exist.
Index: gnu/java/net/IndexListParser.java
===================================================================
RCS file: /cvsroot/classpath/classpath/gnu/java/net/IndexListParser.java,v
retrieving revision 1.3
diff -u -r1.3 IndexListParser.java
--- gnu/java/net/IndexListParser.java 19 May 2006 20:58:59 -0000 1.3
+++ gnu/java/net/IndexListParser.java 14 Jun 2006 14:27:29 -0000
@@ -125,7 +125,11 @@
br.close();
}
- // else INDEX.LIST does not exist
+ else
+ {
+ // INDEX.LIST does not exist
+ clearAll();
+ }
}
catch (Exception ex)
{
Index: gnu/java/net/loader/JarURLLoader.java
===================================================================
RCS file: /cvsroot/classpath/classpath/gnu/java/net/loader/JarURLLoader.java,v
retrieving revision 1.4
diff -u -r1.4 JarURLLoader.java
--- gnu/java/net/loader/JarURLLoader.java 21 May 2006 11:35:08 -0000
1.4
+++ gnu/java/net/loader/JarURLLoader.java 14 Jun 2006 14:27:29 -0000
@@ -135,6 +135,11 @@
try
{
URL subURL = new URL(baseURL, e);
+ // We've seen at least one jar file whose Class-Path
+ // attribute includes the original jar. If we process
+ // that normally we end up with infinite recursion.
+ if (subURL.equals(baseURL))
+ continue;
JarURLLoader subLoader = new JarURLLoader(classloader,
cache, factory,
subURL, subURL);
