On Oct 22, 2009, at 10:58 PM, Ivan wrote:
Thanks, David, I opened a JIRA 4925 to track it.
I will temporarily enable it, so that we would not have any compile error in the plugins. After the main integration work has been done, we could have time to consider whether we still need it, or we have a better way to implement it.
About generating the bundle meta data in the deployment process, I am interested that how it would be implemented in the deployer, use BND or ASM tool ?
There is probably a better solution, but so far we've been ok with a simple calculation of the packages needed by the gbeans plus dynamic import package *. I wrote some code in ExecutableConfigurationUtil that collects the classes needed for the gbeans in the plan and saves the list of packages to a file, then the ArchiveCarMojo reads this and uses it to construct a manifest. Since IMO we should only be using packed plugins in 3.0 we should move all the archive functionality into the deployer somewhere, including figuring out the manifest needed. Very likely the archiveing code can just go into ExecutableConfigurationUtil.
2009/10/23 David Jencks <email@example.com>
To me, this issue of getting runtime modifications of running plugins to persist or work at all is a very low priority item. I think there is a very very good chance that once much of the server is running we will step back look at what we have and decide to completely change how this works, or if we allow it. I think there are much more pressing issues such as the fact that the osgi manifest info is put into plugins by the car-maven-plugin rather than a deployer, so that the only way to build a plugin or deploy an application is through a maven build and assembling a custom server. So, I would comment out or disable any functionality that requires EditableConfigurationManager and file a jira so we don't forget about it.
However, as long as this doesn't constrain the basic architecture in any way, I have no problem with anyone working on it.
On Oct 22, 2009, at 7:50 AM, Ivan wrote:
Do you have more comment on it ? Currently, I think that EditableConfigurationManager is a requried function, and I am sure that something is need to improve for it, Such as, it is better that we could have an attribute to mark it as "not started", because while we stop the connector in the console, after we restarting the server, the connector will start again IIRC.
2009/10/22 Ivan <firstname.lastname@example.org>
The ActiveMQ plugin also depends on it.
Basically, since we allow the user to add new gbeans in the config.xml file for an existed configuration, we may need the API to do it programically.
2009/10/22 Rick McGuire <email@example.com>
Well, to start with, both the jetty and tomcat plugins are using it, and not just for running the tests.
David Jencks wrote:
I would rather make the server work without this functionality. I'm really not sure its appropriate to modify plugins in this way. Can we catalog the places this functionality is absolutely required before we put it back in?
On Oct 21, 2009, at 6:06 AM, Ivan wrote:
If no objection, I will try to recover those functions tomorrow, including the help methods in the ConfigurationUtil. 2009/10/21 Ivan <firstname.lastname@example.org <mailto:email@example.com>>
Seems that EditableConfigurationManager support is removed (at
least temporarily) in the Geronimo kernel. I am not sure that
adding the gbeans to the existed configuration in the runtime is
a good idea. But, IIRC, some plugins depend on it to add gbeans
I would suggest to recover it, the staff I could see is to change
some codes to adapt the OSGI environment in its implemetation.
2009/10/21 Rick McGuire <firstname.lastname@example.org
The Tomcat unit tests are making use of
ConfigurationUtil.getEditableConfigurationManager, which has
been commented out. Is that method no longer applicable, or
does it still require some work? Is
getConfigurationManager() an workable replacement?