It's probably fair. They are certainly missing some things you might expect,
that Antoine ran into…
If we can find a way to create a similar extension capability but implemented
differently, it would be good. However what they offer is too useful to pass up.
You often have a situation where you want to create the effect of having a
group of beans for a multiple configurations of something. Injection of
InjectionPoint can go some way to solving this, but suffers from three key
limitations. Let's first take an example problem domain which I know well -
caches, specifically Infinispan.
You obviously want to be able to configure multiple caches in an application,
as they might well have different characteristics, such as eviction policy,
persistent storage, and so on.
Infinispan offers a number of caching classes associated with a cache - a Cache
interface, and an AdvancedCache interface, as well as the Configuration for the
cache. We want to be able to inject any of these objects for each cache
configured. e.g.
@Inject @Cache("cache1") Cache cache;
@Inject @Cache("cache1") AdvancedCache cache;
@Inject @Cache("cache1") Configuration cache;
@Inject @Cache("cache2") Cache cache;
@Inject @Cache("cache2") AdvancedCache cache;
@Inject @Cache("cache2") Configuration cache;
The first problem we can see here is that we've lost type safety. This is quite
easy to fix, simply by having the user create an annotation per cache, which
perhaps could be meta-annotated with the name of cache. We now have (truncated
for brevity!)
@Inject @Cache1 Cache cache;
However, now let's assume we are running in JavaSE, and we need to don't have
something like JNDI to look up the CacheManager in, every time we want to
access a cache. We need to make the beans that hold the cache reference
application scoped. This is the first of the key problems I identified above.
We could solve this by saying that the cache manager is stored in application
scope, and the cache is looked up each time, but it's also reasonable to assume
that there can be multiple cache managers in an application.
The second problem, is that we really still need a way to attach a
configuration to a cache. A producer method is an ideal way to do this (produce
the configuration needed for the cache, so that CDI can pass it to Infinispan
at the right point), but we somehow need to associate this producer method
through, and have CDI know how to call this. Qualifiers of course solve this.
Finally, we of course want this properly validated at startup, and we do know
the entire system at startup.
I hope that outlines some of the problems we wanted to solve with generic
beans. Anyone have a better idea how to solve this? I'm all ears :-D
On 4 Mar 2012, at 20:01, Mark Struberg wrote:
> I don't think it's correct to call them buggy. It's just that it might be
> _very_ hard to provide the same behaviour over all our supported containers.
> But we will hit those kind of compat problems sooner or later anyway and will
> need to find a way to deal with them.
>
>
> LieGrue,
> strub
>
>
>
> ----- Original Message -----
>> From: Antoine Sabot-Durand <[email protected]>
>> To: [email protected]
>> Cc:
>> Sent: Sunday, March 4, 2012 8:38 PM
>> Subject: Re: [DISCUSS] DELTASPIKE-14 GenericBeans
>>
>> T hank you John to launch this subject. I've been very busy since january
>> and
>> didn't found time to launch the subject. To be totally honest I thought I
>> was the only one interested in them.
>>
>> Now regarding Generic beans in Solder :
>>
>> - documentation is quite inaccurate
>> - they are bugy : I didn't had bug, but it seems that some their tests
>> don't pass
>> - I read some wrong information about them : you can't create beans in
>> another scope that the generic bean definition.
>>
>> I'll prepare a description of how I use them in Seam Social to ease
>> extension of the framework and the issue I encounter using them.
>>
>> regards,
>>
>>
>> Antoine SABOT-DURAND
>>
>>
>> Le 4 mars 2012 à 18:27, Jason Porter a écrit :
>>
>>> I think they're really powerful, but we may need to do some rewrite to
>> make sure it works correctly in a modular container.
>>>
>>> Sent from my iPhone
>>>
>>> On Mar 4, 2012, at 8:52, "John D. Ament"
>> <[email protected]> wrote:
>>>
>>>> Hi All
>>>>
>>>> I would like to begin discussing the use of Generic Beans from Solder
>>>> (currently this issue is assigned to Antoine, but I have some bandwidth
>> and
>>>> offered to help him here). This feature is used to configure a set of
>>>> related beans that require shared components, while still allowing
>> scopes
>>>> to be provided. This is useful when trying to make legacy
>> libraries/APIs
>>>> CDI capable. The following are the API components required for
>>>> GenericBeans:
>>>>
>>>> - @GenericType(Class<?> clazz) - defines the type of
>> configuration for the
>>>> generic. This annotation is placed on another annotation, as defined
>> by
>>>> the application developer or framework author to support how
>> configuration
>>>> is resolved. This will look for a matching bean of the given type and
>>>> resolve it based on the annotation that this is assigned to.
>>>> - @Generic - when using the manager type, defines an expected injection
>>>> point for a generic bean.
>>>> - @GenericConfiguration(Class<?> clazz) - defines the
>> relationship between
>>>> generic objects.
>>>> - @ApplyScope - indicates that the produced object should inherit the
>> scope
>>>> of the configuration.
>>>>
>>>>
>>>> The examples in the Solder documentation describe this in depth:
>>>>
>> http://docs.jboss.org/seam/3/3.1.0.Final/reference/en-US/html/solder-genericbeans.html
>>>>
>>>> Thoughts/questions on the feature?
>>>>
>>>> john
>>
