Re: UnBlocking Blocks

[Stefano Mazzocchi]

Nicola Ken Barozzi wrote:

Stefano Mazzocchi wrote:

Antonio Gallardo wrote:

I think it is time to start thinking in a plug-in technology for Cocoon.

Yes, it's time, but our avalon container is not good enough for that

Let's put down what is lacking, and what the intermediate goals are.

Avalon will do this now:
 - release Fortress
 - release Merlin extends Fortress
 - work on Avalon 5

Merlin is the thing that can give us real blocks.

But I'm wide open to suggestions on how to move this forward.

This is basically what I have proposed at the last live-meeting:
IMHO blocks need to be done in a slow and steady manner, because we are walking on new ground and I don' twant CocoonHEAD to break for big changes.
Avalon will soon release Fortress, that is a ECM successor.

Now, blocks are needed for two reasons:

1 - separate components development and deployment from Cocoon
2 - dynamic loading, polimorphic usage and wizbang super extra inheritance

I see the first part much easier to accomplish than the second, and we could get to that point much easier.

True, I see wisdom in your thinking.

Only a few details... see below.

Step 1 is the last iteration of the .jar blocks.
Step 2 is the .cob system.

What do we need to get to step 1, that will alleviate problems a lot?

 - loading of Avalon components from the block jars
 - loading of Cocoon components from the block jars
 - way of defining blocks in the sitemap

this should be future compatible with the planned cob design, otherwise we'll break things again.

- automatic download of blocks

I imagine a

components
blocks

section, where I can specify the block to use, and the default download location, and the Cocoon components to load.

Something like:

<map:blocks repository="blah">
<map:block name="batik" version="..."/>
<map:block name="fop" version="..."/>
</map:blocks>

and

<map:serializer mime-type="image/png"
name="svg2png"
src="o.a.c.s.SVGSerializer"
block="batik"/>

This way we just tell the component to be loaded from a specific block.

As far as classes go (avalon and cocoon components), don't forget that we already have unique identifiers, so the sitemap should not include that information but it should be somewhere else, and should be treated by the classloader as a way to know where to get stuff.

The automatic download of blocks is not a problem. Krysalis Ruper already is able to download things and has a nice version specification system, so that part is solved.

Uh, cool, didn't know that.

What remains is the automatic load from the block jar, and how to tell the sitemap to do it.

As I said, I don't think the sitemap should be any different from that is now. It should be the classloader's concern to know where to get those classes.

Given that Fortress is ECM2, and given that we are going to release soon, if we need some feature we can ask if it can support it and move to Fortress.

This will give us breathing room for Merlin and proper block implementation.

I'm +1 on implementing blocks incrementally, but only if this doesn't require to break things later on (like atting the block="" attribute first and then remove it later)

what do you think?

--
Stefano Mazzocchi <[EMAIL PROTECTED]>
--------------------------------------------------------------------




---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, email: [EMAIL PROTECTED]