I just want to add that for Geronimo plugins, it's not very useful to
have the car-maven-plugin write the environment into the deployment
plan. More often that not, you want the source tree to contain a
fully valid and correct deployment plan, so you can easily redeploy
the module (e.g. on the command line or hot deploy directory) without
going through the plugin mechanics. (This is, while you're actively
developing and testing the new G plugin.) That means that it's not
acceptable for the source tree to contain a plan with an empty
environment and rely on the car-maven-plugin to generate an output
plan with a valid environment. I'd be happy if there was a flag
saying "don't write environment", and that would eliminate the need
for a G plugin build to use the "scope=provided" in the plugin build
POM (currently all the artifacts in the environment are listed with
scope=provided so the build is aware of them but they are not written
into the environment).
Anyway, I'm going to restate the problem I think you're describing
just to make sure we're talking about the same thing. I think the
main problem with the car-maven-plugin is that is starts a
mini-Geronimo server to do the deployment. Therefore, it needs all
the core Geronimo CARs in your local repository so it can use them to
start the server. Those are the ones that Maven does not download on
your behalf (therefore the comment that it only works if you've built
G on that machine before). It seems like you ought to be able to list
the deployers you care about as well as the dependency CARs/JARs, and
Maven should download those and anything they depend on transitively,
so it has everything it needs in order to run the server. Normally
I'd say the car-maven-plugin should just list all that stuff as
dependencies, but a single version of the car-maven-plugin can
potentially work across several Geronimo versions, so I guess all the
Geronimo CARs need to be resolved dynamically. The main problem is,
if we're on Geronimo 2.0 and you configure the car-maven-plugin for
the Geronimo 1.2 GBean deployer, and that depends on
*/j2ee-system/*/car then it's likely to incorrectly resolve that to
the 2.0 j2ee-system CAR, so we need some way to tell the
car-maven-plugin to resolve all Geronimo libs to version X (and all
OpenEJB libs to version Y, and all TranQL libs to version Z, and
etc.). Is that the same issue you're describing?
FWIW, several of the SF plugins have a POM that lists all the stuff
needed to build a plugin on G 1.1.1, and I'll probably be coming up
with a 1.2 version in the relatively near future. If you include that
as a module in your build it will produce an empty JAR, but it will
cause all the needed JARs and CARs to be downloaded into your local M2
repo so the car-maven-plugin works as expected.
Thanks,
Aaron
On 12/14/06, Guillaume Nodet <[EMAIL PROTECTED]> wrote:
I'm not completely aware of what the car plugin do, but I think this is
pretty close to what we have done for servicemix.
I think we should be able to get back transitive dependencies on jars,
by rebuilding the full dependency graph, and excluding artifacts that
are already included by referenced cars. Let me explain:
If the following is the dependency graph (we are trying to build carA)
- carA
- jarB
-jarC
- jarD
- carE
- jarC
it is possible to load this full graph and prune it to obtain a list of *real*
dependencies: jarB, jarD, carE. jarC would be removed, because it already
contains jarC. To do this, we rely only on the poms of the dependencies.
We have already done that in servicemix maven plugin, so there's no reason
it could not be done for Geronimo ;)
I'm not sure about the provided scope. Actually, the servicemix plugin handle
this scope the same way as the car plugin, but the main difference is
that we don't
have the need of deployers, so things are easier.
On 12/14/06, David Jencks <[EMAIL PROTECTED]> wrote:
> Ever since we got close to having an m2 build we've been having
> problems with the car plugin overloading the maven scopes to mean
> something in geronimo. This has caused endless problems, the latest
> of which is that in order to be able to use the car plugin outside
> geronimo, you have to have built geronimo on that machine so all the
> geronimo bits you need are sure to be there. In more detail this is
> caused by the use of the provided scope to mean "don't stick this
> dependency in the environment". So, if you mark a builder car as
> provided so it won't get stuck in the environment, maven won't
> download all the bits it needs to run.
>
> This is all caused by our attempt to build the environment plan
> element from the pom. The main reason I wanted to do that was to be
> sure that if you mentioned something in the environment, maven would
> make sure it was available. So, maybe it's time to take a more
> direct approach:
>
> How about changing the plugin so it doesn't generate the environment
> element, but instead verifies that all the environment dependencies
> are mentioned in the pom? Among other things this will give easier
> control over which versions are included in the environment element.
>
> The other possible solutions to this problem I have thought of are:
>
> -- get maven to have an extensible scope system or allow dependency
> annotations like you could do in maven 1. I think there's no chance
> of this happening before geronimo 3, but I could be wrong.
>
> -- copy the environment dependencies into the car plugin
> configuration. I think this would end up being harder to deal with
> than validating the environment against the pom.
>
> -- write a maven project that just has dependencies on all geronimo
> artifacts included in any geronimo assembly, so you can build it and
> all the bits will be around for the car plugin to use.
>
>
> Thanks to gnodet for pointing out this problem and discussing
> possible solutions.
>
> Any thoughts?
>
> thanks
> david jencks
>
>
--
Cheers,
Guillaume Nodet
