Looking ahead
Niclas Hedhman
niclas at hedhman.org
Fri Jun 24 01:25:04 EDT 2005
<<sent to Steve privately, but since no response, I post on public list>>
Hi,
I have been playing around a lot with OSGi over the last week or so, and I am
convinced it is much better to drop the runtime platform altogether. I am not
sure whether you feel that to be a good idea, but I do. OSGi has proper
specification, several independent implementations, 5 years on the market,
plenty of additional support in tools and documentation, many useful bundles
(also non-useable ones), and so forth.
We should re-position ourselves and end up with better market value, than a
runtime platform noone wants to use.
1. Build systems should become better and smarter. Put our efforts there.
2. FT already showed how much can be moved to build time, so why not go the
full step. AFAICT, FT runtime is now not much more than a classloading
manager, with some service deps and config handling.
3. We should continue that path, and allow for generation against one or more
runtime platforms.
4. I suggest OSGi is the primary one support, because it handles the
classloading issue, and allow me to swap stuff in and out over time without
restarts (provided the bundles are well-behaved).
I have been experimenting with Metro classic on OSGi, and have some initial
good results; The dpml-console running as 4 co-operating bundles that can
execute commands against one or many Metro instances running as OSGi bundles.
Each of the dpml-console bundles can be reloaded/upgraded independently. The
next two steps is to expose the OSGi services as Metro services and vice
versa. I expect to complete that this weekend.
This provides a fairly good migration path for ScanCoin away from "classic"
without a lot of re-writing in one bang. Standard OSGi bundles will ease
their codebase a bit as well.
Furthermore, OSGi handles registrations of protocols. So, I will create an
artifact protocol handler based on "good'ol" Transit (not the new Starship
Enterprise you are constructing, since I think it is bloated beyond
recognition and no longer "Do one thing and do it well." ).
If you find this to be a reasonable way forward, then great. Otherwise, I
think it is time for us to part, and you will need to sort out infrastructure
fairly soon.
WDYT?
Cheers
Niclas
More information about the dev-dpml
mailing list