[
https://issues.apache.org/jira/browse/DBCP-358?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13027957#comment-13027957
]
Joerg Schaible commented on DBCP-358:
-------------------------------------
This behavior is unfortunately used in the PoolableManagedConnection w/
TransactionRegistry. When a new connection is created in the pool, the
unwrapped connection is registered (and used as key in the internal
WeakHashMap). Later on the wrapped connection is used to unregister again (or
to lookup the XAResource as done in
org.apache.commons.dbcp.managed.TestBasicManagedDataSource.testReallyClose()).
As workaround it is possible to wrap the registered connection if it is not
wrapped yet:
{noformat}
Index: src/java/org/apache/commons/dbcp/managed/TransactionRegistry.java
===================================================================
--- src/java/org/apache/commons/dbcp/managed/TransactionRegistry.java
(revision 1098813)
+++ src/java/org/apache/commons/dbcp/managed/TransactionRegistry.java
(working copy)
@@ -28,7 +28,9 @@
import javax.transaction.TransactionManager;
import javax.transaction.xa.XAResource;
+import org.apache.commons.dbcp.DelegatingConnection;
+
/**
* TransactionRegistry tracks Connections and XAResources in a transacted
environment for a single XAConnectionFactory.
* </p>
@@ -62,7 +64,10 @@
public synchronized void registerConnection(Connection connection,
XAResource xaResource) {
if (connection == null) throw new NullPointerException("connection is
null");
if (xaResource == null) throw new NullPointerException("xaResource is
null");
- xaResources.put(connection, xaResource);
+ Connection key = connection instanceof DelegatingConnection
+ ? connection
+ : new DelegatingConnection(connection);
+ xaResources.put(key, xaResource);
}
/**
{noformat}
This approach will also solve DBCP-356 (another workaround is to use a HasMap
instead of a WeakHashMap for the XAResources).
> Equals implementations in DelegatingXxx classes are not symmetric
> -----------------------------------------------------------------
>
> Key: DBCP-358
> URL: https://issues.apache.org/jira/browse/DBCP-358
> Project: Commons Dbcp
> Issue Type: Bug
> Affects Versions: 1.2, 1.2.2, 1.3, 1.4
> Reporter: Phil Steitz
> Fix For: 1.3.1, 1.4.1
>
>
> For reasons unclear to me, DelegatingConnection, DelegatingStatement,
> PoolGuardConnectionWrappers and other DBCP classes implement equals so that
> the wrapping class is considered equal to its innermost delegate JDBC object.
> This makes equals asymmetric when applied to a wrapper and its wrapped JDBC
> object - wrapper.equals(delegate) returns true, but delegate.equals(wrapper)
> will in general return false.
> I am pretty sure that DBCP itself does not rely on this bugged behavior, so I
> am inclined to fix it, making equals an equivalence relation on wrapper
> instances, with two considered equal iff their innermost delegates are equal.
> I can't imagine use cases where the bugged behavior is required. Can anyone
> else?
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
