On Monday 04 September 2006 14:59, Peter Kriens wrote:
> Can this model be interesting for a new repository?
I have not checked the details yet, but I am sure it is good work... ;o)
Now, Raffael ain't that much into OSGi (yet), so his Ivy Reop is a lot outside
the OSGi world.
In respect to the repository, I think the approach is generally not that
great. I have stopped believing in central repositories. They don't scale
well.
Instead, I would like to see a repository model based on Google instead. For
instance;
* Each publisher has their own http server with the bundles in any format
they wish.
* Each publisher puts an "index file" on to that http server, and the index
file contains a bunch of meta-info;
- bundles available, bundle MD5 checksum, specs and categories those
bundles belong to,
- packages provided/required
- services provided/required (optional)
- public key signature of the publisher,
- names, licensing, addresses and so forth,
- A unique string, for instance G8o73408gfo2TOFGRH68tbhVoo2378etg8weou
* The clients do a simple Google search for
- the unique string and possibly
- category and/or
- specification identifiers and/or
- publisher public key and/or
- any other meta data stuck into it.
The search result will hit on the index files which are downloaded and
parsed.
The clients can then implement features around security, caching, directory
and other things.
IMHO, this allows repositories to scale, be responsive to new releases and
control are transfered away from the repository owner to the publisher
(what/when things are published) and consumer (what/who to trust)
respectively.
The whole concept could be easily be adapted for all Java jars, not only OSGi,
but let's take one thing at a time.
Existing mega repositories, such as http://www.ibiblio.org/maven2 can easily
be fitted with auto-generated index file.
The Oscar Bundle Repository and possibly the OSGi Bundle Repository are
probably already close to the above, they only need to stick in the Google
friendly bits, and we can go off and write clients to do the real work. The
difference is, I don't need anything from OSGi Alliance to publish my work.
Cheers
Niclas
_______________________________________________
general mailing list
[email protected]
http://lists.ops4j.org/mailman/listinfo/general
