Leaking or not

[icarus1111]

Hi,

I'm having difficulties with the Login servlet example which can be found in the manual 15.3 Calling a bean from a servlet.
When I run this example with a profiler I see that for each servlet request one org.jboss.weld.resolution.ResolvableBuilder$ResolvableImpl and one org.jboss.weld.resolution.TypeSafeResolver$Key instance is created. However, when I force the Garbage Collector to cleanup these items, they won't be cleaned up after all. Am I doing something wrong? In our application we have a simular situation which leads eventually to an OufOfMemoryError.
I'm using Glassfish 3.0.1 (build 22) with WELD-000900 1.0.1 (SP3).

Hope someone can help me out.

Regards,
Ramon

[swd847]

Thanks for reporting this, there is a leak.

The fix should be merged into trunk in the next few days. If you want it right now it is available at http://github.com/swd847/core/tree/WELD-681

[icarus1111]

That's great!
However, we are using Glassfish 3.0.1, which (AFAIK) doesn't support Weld 1.1 beta (yet).
Do you think it is safe that I patch Weld 1.0.1 by applying the Weld 1.1 beta fix modifications to the Weld 1.0.1 sources?

Regards,
Ramon

[swd847]

I think that should be fine, all that is required is to fix the equals and hashcode method on TypeSafeResolver.Key

[icarus1111]

It seems that it is bit too complex for me to fix it in 1.0.1. Stuart, maybe you can help me out? This is the TypeSafeResolver.equals implementation in 1.0.1:

      @Override
      public boolean equals(Object obj)
      {
         if (obj instanceof Resolvable)
         {
            Resolvable that = (Resolvable) obj;
            return this.getResolvable().getTypes().equals(that.getTypes()) && this.getResolvable().getQualifiers().equals(that.getQualifiers());
         }
         else
         {
            return false;
         }
      }

...and the TypeSafeResolver.hashcode implementation:

      @Override
      public int hashCode()
      {
         int result = 17;
         result = 31 * result + this.getResolvable().getTypes().hashCode();
         result = 31 * result + this.getResolvable().getQualifiers().hashCode();
         return result;
      }

I cannot find a clear match with the current beta code. Maybe you can give me some pointers how to patch this code?

Regards,
Ramon

[swd847]

I think it is probably this bit:

@Override
      public boolean equals(Object obj)
      {
         if (obj instanceof Resolvable)
         {
            Resolvable that = (Resolvable) obj;
            return this.getResolvable().getTypeClosure().equals(that.getTypeClosure()) && this.getResolvable().getQualifiers().equals(that.getQualifiers());
         }
         else
         {
            return false;
         }
      }

should be something like:

@Override
      public boolean equals(Object obj)
      {
         if (obj instanceof Key)
         {
            Key that = (Key) obj;
            return this.getResolvable().getTypeClosure().equals(that.getResolvable().getTypeClosure()) && this.getResolvable().getQualifiers().equals(that.getResolvable().getQualifiers());
         }
         else
         {
            return false;
         }
      }

[icarus1111]

Great!

So no hashcode modifications needed?
Resolvable.getTypeClosure() is not defined in 1.0.1, should I replace it with getTypes?

Thanks a lot,
Ramon

[swd847]

I think the hash code is fine in that version, and you should replace getTypeClosure() with getTypes().

[icarus1111]

It looks like the leak is gone, Stuart, thanks a lot!

So, for those interested in this fix for Weld 1.0.1: replace TypeSafeResolver.Key.equals(Object obj) with:

      public boolean equals(Object obj)
      {
         if (obj instanceof Key)
         {
            Key that = (Key) obj;
            return this.getResolvable().getTypes().equals(that.getResolvable().getTypes()) && this.getResolvable().getQualifiers().equals(that.getResolvable().getQualifiers());
         }
         else
         {
            return false;
         }
      }