Thousands of [WARNING] Component returned which is not the same manager.

I did some tests by changing the comments. This message is definately coming from the ClassicSingletonComponentManager.java not the KeepAlive.....java file.

This seems to be more than a nuisance now. When this is occuring, the time to build my project triples from 17 mins to about 48 mins. Possibly because the count isn't decremented and somethins is timing out?

Two comments:

1. The code looks good but maybe it's a complicated case of "Double-Checked Locking is Broken" (http://www.cs.umd.edu/~pugh/java/memoryModel/DoubleCheckedLocking.html)

To verify this, move the synchronize to the method declaration instead of creating your own lock object and run your tests.

2. I'm wary of the comment in dispose(). Add a check in there for connectionCount == 0 to make sure no one calls this method out of order.

Maybe you should convert dispose() into a dud (just throws Exception) and do the real dispose in a private method?

I observed this today on a laptop with a single processor and no hyperthreading. The warning this time was on the OperatingSystemProfileActivation component. Of note is that we don't use either the JDK or OS profile activation anywhere in our build. The only profile activation we use is default and property.

We get this on an offline OpenEJB build too. Not quite every time, seems like every other time.

$ svn co http://svn.apache.org/repos/asf/incubator/openejb/trunk/openejb3
$ cd openejb3
$ mvn -o clean install -Dassemble -Dmaven.test.skip=true | tee build.log
$ grep -c 'Component returned which is not the same manager' build.log
20594

So about 20594 times.

Super annoying.

This is still occuring in Maven 2.0.5 (no surprise since it seems to be using the same plexus rev)

just saw the thread on maven user list.

• as aaron, i wonder what is the need of the lock object. just use synchronize. The parent class is not using synchronization anyway.

• what if the problem was caused by someone calling release() on the same component more than once ? Has someone tried to print the value of singleton in the same warning message ?

I am having trouble replicating this issue to try and trace it.

The OpenEJB example above does not yeild any such messages here.

Any other examples of how to replicate it?

At least continuum-1.0.3 spits them out and also my application does.

Fixed \o/ be happy

version alpha-19-SNAPSHOT is current, so as of alpha-19 the releases should have this addressed.

Am using maven 2.1.0 on a win xp dual core. These messages showed up again. Used mvn archetype:create to create a project with jar packaging. Just ran mvn install after creating the project.

Am also seeing these messages showing up intermittently using maven 2.2.1 on a win xp dual core, during normal mvn install operations. The line appears 100's of times sometimes:

[WARNING] Component returned which is not the same manager. Ignored. component=org.apache.maven.profiles.activation.JdkPrefixProfileActivator@491a491a

Also occuring for me on xp, dual core, mvn 2.1.0 (as well as 2.2.1):

[WARNING] Component returned which is not the same manager. Ignored. component=org.apache.maven.profiles.activation.AlwaysOnProfileActivator@db40db4

After a quick check it seems like the root cause for this is that component managers are looked up by the managed object's hashCode, and in most cases a default hashCode() implementation is used which returns object's physical address, so there is always a possibility that the same physical address is reused for different objects.

Example: address 23e823e8 being reused for 2 different components:

Created singleton org.codehaus.plexus.logging.console.ConsoleLoggerManager@23e823e8 in org.codehaus.plexus.component.manager.ClassicSingletonComponentManager@4cf04cf0, Thread[main,5,main]
Deleting singleton org.codehaus.plexus.logging.console.ConsoleLoggerManager@23e823e8 in org.codehaus.plexus.component.manager.ClassicSingletonComponentManager@4cf04cf0, Thread[main,5,main]
Created singleton org.apache.maven.profiles.activation.JdkPrefixProfileActivator@23e823e8 in org.codehaus.plexus.component.manager.ClassicSingletonComponentManager@5c905c90, Thread[main,5,main]
[WARNING] Component returned which is not the same manager. Ignored. component=org.apache.maven.profiles.activation.JdkPrefixProfileActivator@23e823e8, singleton=org.codehaus.plexus.logging.console.ConsoleLoggerManager@23e823e8 in org.codehaus.plexus.component.manager.ClassicSingletonComponentManager@4cf04cf0, Thread[main,5,main]

The root cause probably lies somewhere in DefaultComponentManagerManager.findComponentManagerByComponentInstance() (module plexus-container 1.0-alpha-9-stable-1).

There may be other places that use the flawed hashCode-based lookup algorithms.

Attaching a patch that will change component manager lookup from hashCode to hashCode AND class type (also not perfect but seems to work for now).

Can a committer please clarify the status of this? Is there a new bug that appeared in 2.1.0 which alpha-19 release doesn't fix.
Thanks.

@Neale, this is an old bug which will not be fixed because Maven itself has moved away from Plexus and is now using Guice for dependency injection.