I see this was pushed already. Good, quick work guys!
Yes, there are a couple other RMI test failures caused by aggressive GC of
loggers (or possibly other objects). I know the Hotspot folks have filed a
couple already. But we should keep an eye out for them.
s'marks
On 12/4/13 7:44 AM, Seán Coffey wrote:
Recent jdk8 builds seem to be more ambitious on the GC front. It turns out that
we've no strong reference for the Logger being created in this testcase and it's
collected before use.
Verified that test now passes.
suggested fix :
diff -r 28ca338366ff test/sun/rmi/runtime/Log/checkLogging/CheckLogging.java
--- a/test/sun/rmi/runtime/Log/checkLogging/CheckLogging.java Mon Nov 25
20:22:23 2013 -0800
+++ b/test/sun/rmi/runtime/Log/checkLogging/CheckLogging.java Wed Dec 04
15:22:42 2013 +0000
@@ -76,6 +76,7 @@
public class CheckLogging {
private static int REGISTRY_PORT = -1;
private static String LOCATION;
+ private static Logger logger;
private static final ByteArrayOutputStream clientCallOut =
new ByteArrayOutputStream();
@@ -89,8 +90,8 @@
System.err.println("set default stream");
LogStream.setDefaultStream(new PrintStream(clientCallOut));
} else {
- Logger.getLogger("sun.rmi.client.call").
- addHandler(new InternalStreamHandler(clientCallOut));
+ logger = Logger.getLogger("sun.rmi.client.call");
+ logger.addHandler(new InternalStreamHandler(clientCallOut));
}
}
regards,
Sean.
