geronimo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From David Jencks <>
Subject Re: module builders/deployers in plugins
Date Wed, 12 Mar 2008 17:54:30 GMT

On Mar 12, 2008, at 10:45 AM, Jarek Gawor wrote:

>>>>  2) The AdminObjectRefBuilder is always trying to process _all_
>>>>  resource-env-ref entries (AdminObjectRefBuilder is just an
>>>> example in
>>>>  this case). However, as things evolve (Java EE 5 -> 6) and new
>>>>  resource env. types are added, it does not scale to keep updating
>>>> the
>>>>  AdminObjectRefBuilder to handle these new types. So I think it  
>>>> would
>>>>  be nice to install a new builder that would handle these new types
>>>>  only. But that would require communicating with the
>>>>  AdminObjectRefBuilder and somehow telling it to ignore the new
>>>> types.
>>>>  The question is, what would be the best way of doing it? Or  
>>>> maybe we
>>>>  need a different way of processing the DDs?
>>  This is kind of a tricky problem and I don't know the best way to
>>  handle it.  I doubt you want to get into rewriting the entire way we
>>  resolve resource-env-refs at this point -- let me know if you do --
>>  so I would look into exactly what is keeping the current
>>  AdminObjectRefBuilder from handling your new kinds of refs and
>>  whether there is an easy way to exclude them.
>>  Some possibilities....
>>  - the AdminObjectRefBuilder.AdminObjectRefProcessor picks out the
>>  annotations we will look at.  If the new stuff can only be referred
>>  to as @Resource you can have the AdminObjectRefProcessor skip the  
>> new
>>  kinds.
>>  - basically the AdminObjectRefBuilder works by finding a gbean that
>>  corresponds to the name information and type information supplied,
>>  around line 373.  We could expand the choices for how we look for
>>  target gbeans.
>>  - we might be able to make the NamingBuilderCollection ordered and
>>  have the builders not try to deal with questionable items that have
>>  already been processed.
> I tried the last option. Specifically, I modified the
> AdminObjectRefBuilder to ignore elements that were already added to
> the jndi context map (assuming the NamingBuilderCollection would be
> ordered and that my builder would execute before the
> AdminObjectRefBuilder). However, since both Builders process the same
> type of elements in the DDs the NamingBuilderCollection did not like
> that and I got the following exception:
> 3:01:30,421 ERROR [ProxyCollection] Listener threw exception
> java.lang.IllegalArgumentException: Duplicate builderSpecQNames in
> builder set: QNameSet+(+resource-env-ref@ 
> javaee,
> +resource-env-ref@ and duplicate
> builderPlanQNames in builder set :
> QNameSet+(+resource-env-ref@ 
> naming-1.2)
> So this option would work only if I disabled this check or returned
> some other qnames for builderSpecQNames and builderPlanQNames in my
> builder.
> Also, I could create a new switching builder for processing the
> resource-env-ref elements and have it just basically redirect the
> calls to the AdminObjectRefBuilder and my builder (to avoid the above
> error) and combine it with the last option you described.

<mumbles things about hoist by own petard, too clever by half, etc etc>

I think you can avoid this problem by cheating and not registering  
any qnames in your builder.  This might be sort of plausible since  
its not the "main" builder for the qnames.  However we might want to  
rethink the entire collision avoidance theory altogether.

david jencks

> Jarek

View raw message