On Sep 24, 2008, at 3:06 AM, David Jencks wrote:
On Sep 23, 2008, at 8:28 PM, Kevan Miller wrote:
On Sep 23, 2008, at 1:17 PM, David Jencks wrote:
I'd like framework to build the entire working framework server,
thus demonstrating that it really is a complete server. I think
it would work with the following changes:
- move at least the car-maven-plugin into framework, perhaps all
of buildsupport
- move the jsr88 classloader into framework
- move the framework plugingroup into framework
- move the assemblies geronimo-boilderplate and geronimo-framework
into framework
This is somewhat related to https://issues.apache.org/jira/browse/GERONIMO-4258
Thoughts? Objections? Comments?
Personally, seems a bit forced... I'm not sure I understand your
motivations, and if I do understand, not sure that I really agree
with them. Why exactly is "demonstrating that it really is a
complete server" necessary? Or good?
Can you be a bit more precise about what you mean by 'jsr88
classloader'?
Personally, I kind of like that a 'plugingroup' can be found in
'trunk/plugingroups'. Similar feelings for assemblies and
buildsupport.
My sense is that you want 'framework' to be a standalone entity, as
if it were separately releasable -- which we could do (although I
don't see much advantage in that, at the moment...). Until we're
ready to truly split it apart, I'm not sure there's much gain by
going half-way.
My idea about the geronimo servers is that we have a base, the
framework server, that can't do anything except be extended, and
then we have bunches of functionality, such as the jetty web
container, that can be added to it. I want this to be not just a
theoretical hand waving marketing statement but actually reflected
in the code. To me this means that there should be a clearly
delineated way to build the framework server with essentially
nothing else. Currently this is impractical. You'd have to build a
few framework modules, go build the car-maven-plugin, build a few
framework configs, build a plugins/classloader plugin, build the
rest of the framework modules, build the rest of the framework
configs, build the framework plugingroup, build the assembly
boilerplate, and finally the framework assembly. IMNSHO this is
ridiculous and implies that geronimo is not very agile or easy to
deal with, and certainly not easily extensible for other purposes.
Someone who wants to build some kind of special purpose server -
such as the plugin-farm-node - has to build all of geronimo to get
to the few parts they actually want.
Well, you asked for thoughts and comments... ;-)
IMHO this *build-time* structure is "ridiculous" only if the above
framework server functionality cannot be achieved at *run-time*. I'd
say that what we have and what we can demonstrate is much more than
"theoretical hand waving marketing statement". And "actually reflected
in the code" would be more accurately described as "actually reflected
in our source directory structure". Assuming the runtime aspects are
close to right, then becomes a question of some balance of the
following aspects:
build experience -- where do we normally run builds from? how many
build commands are required? how long does a build take?
maven artifact structure -- what's the organization of our released
artifacts in maven?
release process -- how many separately released artifacts will go into
a server release? how many versioned artifacts do we need to keep
track of?
etc...
The jsr88 classloader is plugins/classloaders/geronimo-javaee-
deployment_1.1MR3_spec. It now has one of our extension classes in
it as well as the jsr88 spec.
I have no problem having framework/plugingroups for the framework
plugin group, framework/buildsupport for the maven plugin(s) and
framework/assembly for the geronimo-framework assembly.
I think this also moves us closer to what I thought was an agreed
goal of releasing the plugins separately from the framework and
assembled servers.
Personally, it's not a big goal of mine. I think it introduces
complexity to the community, without adding much benefit to the
community. I do want the basic structure to be there -- allowing
additional plugins / assemblies to be built and separately released,
if we so desire. However, doesn't mean that I'm eager to break our
server release into a number of parts, holding separate release votes,
and tracking some number of separately versioned artifacts.
--kevan
