Simon, Regarding bullet A, this is fine if the set of contributions is static. What happens as contributions are added and removed (particularly contributions that contain definitions)? I think the spec is silent about what should happen.
Raymond, Are you suggesting that the user should have to explicitly import definitions from other user (non-system) contributions? As far as I know all definitions that have been contributed to the domain are implicitly visible to all contributions. On Fri, Jan 9, 2009 at 12:20 PM, Simon Laws <[email protected]>wrote: > > > On Fri, Jan 9, 2009 at 4:46 PM, Raymond Feng <[email protected]> wrote: > >> Hi, >> >> I had an offline chat with Mike Edwards on a related SCA spec question a >> few weeks ago. He came up an idea to use the "system contribution" to >> contain the global definitions in the SCA domain. The definitions from SCA >> specs and Tuscany extensions will be loaded by Tuscany runtime and added to >> the "system contribution". The "system contribution" has the exports to make >> them visible to application contributions. >> >> I quite like his idea and I think it's worth to explore. It's similar to >> the OSGi idea to have a system bundle that exports system packages. >> >> Let's say we have an SCA composite application which requires 2 >> contributions: C1 and C2. The "system contribution" is C0. Then the >> composite application will have access to intents and policySets in C1, C2 >> and C0 if necessary imports/exports are declared. By default, we can assume >> the import to SCA spec and Tuscany extension namespace is allowed. >> >> Thanks, >> Raymond >> >> *From:* Simon Laws <[email protected]> >> *Sent:* Friday, January 09, 2009 7:56 AM >> *To:* tuscany-dev <[email protected]> >> *Subject:* [1.x][2.x][DISCUSS] definitions.xml processing >> >> In 1.x, and by implication 2.x, the definitions.xml processing is >> problematic in a number of ways. We have started to talk about revisiting >> the policy model [1] and although we haven't done that yet in 2.x I'm >> trying to get my mind around how definitions.xml files are processed by >> Tuscany. There seem to be a number of gaps, e.g. TUSCANY-2499 [2]. >> >> A) Resolve definitions gloabally given the set of contributions available. >> I think the ideal order of processing for a stand-alone node should be. >> >> - read all defintiions.xml files from extensions that the runtime loads >> - read all contributions (which in turn will read any definitions.xml >> files that appear in contributions) >> - resolve all the definitions that have been read (aggregate all the >> definitions models together and resolve them) >> - resolve all of the contributions >> >> I believe the contribution processing has been split up already in 2.x so >> this change should be straightforward there. I have a patch for 1.x. It >> relies on exposing interfaces on the contribution service to allow read and >> resolve to be called independently. This does solve 2499 so I'll attach the >> patch there for people to look at. I haven't commited directly as there >> workspace contribution processing is already in 1.x but I'm not sure yet >> what the implications of switching across to that would be. >> >> An interesting problem I uncovered is that the JAAS policy itself dependes >> on contribution artifacts (java classes) so you have to resolve the >> definitions in the context of the contributions that have previously been >> read. This feels a little odd to me. >> >> B) Global access to the definitions artifacts >> >> In 1.x the definitions model is owned by the RuntimeBootStrapper and is >> made available to the model through the resolver structure. It seems that it >> is not easy to use the resolver approach in some places, e,g the JMS binding >> where its gets data from the definitions.xml binding structure. We need to >> review this and see how access should be provided to the definitions.xml >> model. >> >> C) Retrieve domain definitions from the domain >> >> The problem described in A) raises it's head in the domain also. Here >> though the solution is not so straighforward as the domain level defintions >> model must be subsequently provided to all nodes (assuming that I understand >> the specs correctly of course). >> >> While there are some TODO comments in the code suggesting that defintions >> dependencies should be expressed through contribution import/export >> statements I don't think this is true as the spec considers defintiions.xml >> to be a domain level resource while Tuscany just happens to allow you to >> include them in a contribution. >> >> We could take the aggreated definitions.xml artifacts and create a special >> contribution and provide this to all nodes. Alternatively we could provide a >> specific endpoint from which to retrieve the defintiions.xml artifacts. The >> first approach would seem to be the simpler of the two. >> >> Regards >> >> Simon >> >> [1] http://cwiki.apache.org/confluence/display/TUSCANYWIKI/Policy >> [2] http://issues.apache.org/jira/browse/TUSCANY-2499 >> > > Yep, I'm with your there. Seems like the most straightforward way to > represent it. > > Simon >
