On Wednesday 11 August 2004 04:22, Sylvain Wallez wrote:
> Although this is very good from a robustness point of view, I don't feel
> at ease with this as it requires to somehow manage dependencies twice:
> in the model where they have to be formally expressed, and in the code
> where actual lookup occurs. Again, depencency injection (as opposed to
> lookup) avoids this problem.
Looking at the actual differences;
public AbcComponent( org.hedhman.niclas.SomeDependency some )
{
m_SomeDependency = some;
}
/** @avalon.dependency type="org.hedhman.niclas.SomeDependency"
* key = "some"
*/
public void service(ServiceManager man )
{
m_SomeDependency = (SomeDependency) man.lookup( "some" );
}
If there is proper SoII in CDI, the SomeDependency is an interface, and that
leaves room for more than one implementation, in which case both Merlin and
Pico/Nano needs additional meta/code to declare exactly which implementation
to be used.
If there is only one implementation available, Merlin will use that only, as I
believe Pico/Nano will also figure out.
So, the current differences are not as great as some may think. Merlin has
come a long way compared to ECM, but I agree that there are room for further
improvements.
> You said that Merlin is going to support injection, which IMO a very
> good thing.
We have taken a few steps in that direction already. So called "custom
contexts" can already be injected via constructors, as can the traditional
Avalon lifecycle objects, such as Logger, Context, ServiceManager,
Configuration and Parameters.
The next step ain't that far away, we are just struggling with some semantics,
especially surrounding the possibility of more than a single constructor,
which Pico says is a no-no, but could be a necessity.
Another 'issue' that CDI imposes is how to deal with aggregations and possibly
custom configuration/parameterization objects, which Pico (unlike Type2) will
need.
All and all, I am fairly positive about the continuation of the Merlin
convergence with other IoC styles, especially if Merlin becomes the Metro
TLP, in which case there no longer exist any 'religious reasons' to stick
with the Avalon-way only.
Cheers
Niclas
--
+------//-------------------+
/ http://www.bali.ac /
/ http://niclas.hedhman.org /
+------//-------------------+
