Stephen McConnell wrote:
Yannick Menager wrote:
I have been working on some apps that have require a plugin infrastructure.
However merlin doesn't really handle that kind of things very nicely. First there's the lack of selector that already has been discussed, but more, classloading needs to be manually edited to include blocks which might contain such service implementations.
Classloader definitions can be programatically modified by this is complex territory. If (and there is a very big if here) you need to expand a classloader (as opposed to branching a classloader) then you basically have to take the classloader down (which means taking down all of the components and suspending all consumers), modify the classloader metamodel, revalidate and redeploy the classloader, and all of the components. On the otherhand - if the API of the components you want to plug in are already in the classloader - then you talking about branching which is comparitvely drop dead easy (just add a block to the block you want to plug in to - this will result in a new classloader branch automatically).
Yes i'm thinking about branching.... so you have already the block X with the interface Y ... and you request for all implementations of the service Y, in all jars present in the local repository.
What I was thinking, is that it would be nice to implement a functionality where a module could actually request all services X, without indicating the classloading dependencies. Of course in order to do that, merlin would need to have a cache index of which jars contains which services, and then use that to provide the service requestor with a list of available implementations.
This is what the block defintion provides this - it *is* the persistent defintion of the classloader dependencies for a particular deployment scanario.
Yes but at this time it does not keep an index of all the repository jars metadata
What do you guys think ?
I think I need a more complete picture of the definition of plugin (that you have in mind) and how this differs from the plugging in of a block as a child of another block.
The difference, is that you have to *declare* the child of a block. What I'm talking about is that he will automatically add the childs based on the Service I requested ( so it will look in it's index of service implementations, and add all the blocks that implement that service )
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
