geronimo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Rick McGuire <>
Subject Re: Bean validation tck failures and JNDI contexts
Date Thu, 21 Oct 2010 18:40:09 GMT
On 10/21/2010 2:33 PM, David Jencks wrote:
> yikes...
> I didn't study this in detail recently but I wonder if we should add an xbean RuntimeNamingException
that wraps the NamingException and have ContextFederation.getFederatedBindings() unwrap only
this runtime exception?
I'm rebuilding Geronimo now to pick up the changes, but here's what I 
added to getFederatedBindings():

                     // don't overwrite existing bindings
                     if  (!bindings.containsKey(bindingName)) {
                         try  {
                         }catch  (RuntimeException  e) {
                             // if this is a wrapped NamingException, unwrap and throw the
                             Throwable  cause  =e.getCause();
                             if  (cause  !=null  &&  cause  instanceof  NamingException
 ) {
                                 throw  (NamingException)cause;
                             // Wrap this into a RuntimeException.
                             throw  (NamingException)new  NamingException("Could retrieve
bound instance"  +name).initCause(e);

This unwraps and rethrows a NamingException if it is there and wraps any 
other RuntimeException inside of a NamingException so that only 
NamingExceptions are seen down stream.  I think this should give the 
behavior we need to fix the TCK problems.


> david jencks
> On Oct 21, 2010, at 11:10 AM, Rick McGuire wrote:
>> On 10/21/2010 12:58 PM, David Jencks wrote:
>>> umm...
>>> doing a naming lookup should only throw NamingExceptions.... so I think what
should happen is that the reference should wrap the ValidationException in a NamingException.
 Will this mess up the tck?
>> This seems like the correct thing to do, but unfortunately, it didn't work.  The
problem is in the xbean ContextUtil$ReadOnlyBinding.getObject() method:
>>         public  Object  getObject() {
>>             try  {
>>                 return  resolve(value,getName(),null,context);
>>             }catch  (NamingException  e) {
>>                 throw  new  RuntimeException(e);
>>             }
>>         }
>> This is catching the NamingException and turning it into a RuntimeException.  This
doesn't seem correct to me, but the Binding interface it implements does not throw any checked
exceptions, so this is the only way this error can get reflected back.
>> Looking at the call chain, I'm thinking that ContextFederation.getFederatedBindings()
should catch RuntimeExceptions and and either turn them into NamingExceptions or just ignore
the exception like it does with the NotContextExceptions.
>> Rick
>>> thanks
>>> david jencks
>>> On Oct 21, 2010, at 9:34 AM, David Jencks wrote:
>>>> Hi Rick,
>>>> The code that does the lookup is here in the jetty integration code: (GeronimoWebAppContext
near line 104)
>>>>         try {
>>>>             javax.naming.Context ctx = integrationContext.getComponentContext();
>>>>             Object validatorFactory = ctx.lookup("comp/ValidatorFactory");
>>>>             setAttribute("javax.faces.validator.beanValidator.ValidatorFactory",
>>>>         } catch (NamingException e) {
>>>>             // ignore.  We just don't set the property if it's not available.
>>>>         }
>>>> I suspect it used to pass because we were only using default validatory factories
so we could always create one.  Either that, or we used to throw a NamingException when we
failed (the code you quote catches a naming exception....).
>>>> I wonder if a better solution would be to also catch and ignore a ValidationException
>>>> thanks
>>>> david jencks
>>>> On Oct 21, 2010, at 7:55 AM, Rick McGuire wrote:
>>>>> I played around with different solutions and finally came up with something
that fixes the problem.  Unfortunately, I'm not sure what I did is legitimate or not.  The
root problem here is the naming reference implementations were throwing ValidationExceptions
for any failures with creating a ValidatorFactory.  This probably was the behavior that should
be implemented, but unfortunately, the getFederatedBindings() processing was triggering the
resolution of these objects and the resulting exceptions were causing deploy failures.  The
test cases in question were testing the very conditions that triggered the exceptions.  The
exception was raised, but at deploy time, resulting in a test case failure.
>>>>> I managed to fix this by having the reference objects we bind into jndi
catch the exceptions and just return null.  Everything is passing in the TCK now, but I'm
not sure returning null is the correct thing to do here.
>>>>> I'm not really sure how we every were passing 100% in the container with
the original code.  I would have thought that if the same sequence of calls were getting made
to resolve the provider, then some of the same failures would have been seen.  I'm going to
hold off on committing my changes until I get some feedback on this.
>>>>> Rick
>>>>> On 10/21/2010 7:48 AM, Rick McGuire wrote:
>>>>>> We're down to 13 bean validation failures in the tck now, but these
failures are a little puzzling.   The tests in error are all giving deploy failures, with
the root cause being an exception triggered by getFederatedBindings():
>>>>>> java.lang.RuntimeException: javax.naming.NamingException: Validator
[Root exception is javax.validation.ValidationException: Unable to find suitable provider:
class org.hibernate.jsr303.tck.common.TCKValidationProvider]
>>>>>>        at org.apache.xbean.naming.context.ContextUtil$ReadOnlyBinding.getObject(
>>>>>>        at org.apache.xbean.naming.context.ContextFederation.getFederatedBindings(
>>>>>>        at org.apache.xbean.naming.context.AbstractFederatedContext.getBindings(
>>>>>>        at org.apache.xbean.naming.context.AbstractFederatedContext.getBinding(
>>>>>>        at org.apache.xbean.naming.context.AbstractContext.lookup(
>>>>>>        at org.apache.xbean.naming.context.AbstractContext.lookup(
>>>>>>        at org.apache.geronimo.jetty8.handler.GeronimoWebAppContext.<init>(
>>>>>>        at org.apache.geronimo.jetty8.WebAppContextWrapper.<init>(
>>>>>>        at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native
>>>>>>        at sun.reflect.NativeConstructorAccessorImpl.newInstance(
>>>>>>        at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(
>>>>>>        at java.lang.reflect.Constructor.newInstance(
>>>>>>        at org.apache.xbean.recipe.ReflectionUtil$ConstructorFactory.create(
>>>>>>        at org.apache.xbean.recipe.ObjectRecipe.internalCreate(
>>>>>>        at org.apache.xbean.recipe.AbstractRecipe.create(
>>>>>>        at org.apache.xbean.recipe.AbstractRecipe.create(
>>>>>>        at org.apache.geronimo.gbean.runtime.GBeanInstance.createInstance(
>>>>>>        at org.apache.geronimo.gbean.runtime.GBeanInstanceState.attemptFullStart(
>>>>>>        at org.apache.geronimo.gbean.runtime.GBeanInstanceState.start(
>>>>>> The root cause of this failure is an exception in DefaultValidatorReference.getContent():
>>>>>>    @Override
>>>>>>    public  Object  getContent()throws  NamingException  {
>>>>>>        ValidatorFactory  factory  =null;
>>>>>>                try  {
>>>>>>            factory  = (ValidatorFactory)new  InitialContext().lookup("java:comp/ValidatorFactory");
>>>>>>        }catch(NamingException  e) {
>>>>>>            factory  =Validation.buildDefaultValidatorFactory();
>>>>>>        }
>>>>>>                return  factory.getValidator();
>>>>>>    }
>>>>>> The root cause of this failure is the NamingException on the .lookup()
call.  Since this occurs during the building of the federated context, I suspect the initial
context is not initialized correctly at this phase.  There's a bit of a chicken-and-egg problem
here.  The buildDefaultValidatorFactory() call is failing because the incorrect thread context
classloader is getting used to resolve the provider.
>>>>>> The puzzling piece to me is why this process is making the getContent()
calls in the first place.  Since this binding will create a new instance each time it is requested,
either A) an instance is getting created needlessly and thrown away or B) this instance is
ending up bound to the JNDI context as a one-off, which would be an incorrect result.
>>>>>> I think I can fix this by making the DefaultValidatorReference look
up the ValidatorFactoryGBean to obtain the factory used to create the ValidatorInstance rather
than doing a jndi lookup, but I want to verify that the lookup occurring at this point is
the correct behavior and there's not a better solution available.
>>>>>> Rick

View raw message