[Sorry for the very long mail. I want to document all that I had in mind & the stuff we discussed. I would recommend all devs to take some time to read this]
I would like to summarize he discussion we had a couple of days back. *The Problems* The problems we are trying to solve are as follows: 1. Trunk & branches structures being completely different 2. Branches containing directories with version numbers 3. It is impossible to move to GitHub with the current structure because of #2 4. It is very easy to break the build by changing already released code. The room for human error is high. 5. Bamboo builds are eternally broken because the build fails at some point & Bamboo cannot continue any further 6. When we branch, the trunk quickly becomes obsolete, and remains in that broken state until the next major platform release. 7. Everybody has to build all components/features, even if those are not related to their products 8. Fixed versions in branches instead of using SNAPSHOT versions. This makes it impossible to upload build artifacts to Maven/Nexus repos. This leads to #7. 9. Impossible to integrate code quality tools such as EraInsight because of #5 *Proposed solution* We have come up with the following solution after much deliberation & thought. Rationale: We started looking at other open source projects out there. We took Axis2 as an example. Axis2 had many dependencies including Axiom, XmlSchema, Woden, WSS4J etc. Those 3rd party dependencies were also developed by some Axis2 contributors, but we never branched all of those together and brought them into the same code branch. We used to start what we used to call a release train, where the upstream code would have to be released first before the downstream code such as Axis2 & Synapse could be released. This way, we never had any of the problems outlined above. If you look at the WSO2 product releases, again, the scenario is not that much different from the Axis2/Synapse releases. Components & features are simply dependencies of the products. In order to get product releases out, we first need releases of those dependencies (components/features). So the proposal is to identify the top level components/features, and first release that code before doing product releases. Each of those components will have a GitHub repo. All active development will be done on the main branch of those components, and will be branched when they are close to the release. Instead of granting commit rights to all, we could only allow the primary developers of those components to commit, and others can send pull requests, which will be merged in by the primary owners after reviewing. These component teams could even have an internal Git repo or clone, and commit to that, run all tests, and when they are satisfied that the code is in a good state to be merged into the main branch, they can send a pull request & merge the changes. We would run Bamboo which would build the components several times a day & deploy them to the Nexus repo. That way, you would not have to build code that is not directly related to what you are working on, which would save 100s of valuable dev hours. In the majority of the cases, the P2 features correspond to one or more related components. So we would keep the feature close to the component. There are some cases where these components & features belong in the product itself. AppFactory components are a good example. Products will also have their own GitHub repos. We will have separate GitHub repos for platform integration tests. Products such as API Manager may opt to have the API Management feature in the product itself. We would have a separate P2-repo GitHub repo which would build & deploy the compatible set of P2 features. We will have Bamboo plans at multiple levels; components, products, platform, p2-repo and so on. We will not change any package structures at this time. We would defer that to Carbon 5 (if necessary). We will get rid of the chunk based release model. Continuous delivery is our ultimate aim & for that to happen, we have to release a compatible set of features. We are going to rely heavily on automation, automated builds & deploys to Maven. The developers will do the majority of the QA (just like the Apache Stratos team is doing now). We will use the Maven release plugin to create releases & upload them to the Maven repo. *Implications to WSO2 code developers* 1. Life becomes easier because you don't have to spend a lot of time building unrelated stuff 2. General development would happen in the main branch. 3. Close to a release, branches would be cut. 4. Once release branches are cut, fixes would be done in the main branch, and pull requests would be sent to the branch & merged in. This will be easy because we no longer have the trunk & branch structure being different. *Implications to users* Users will not see any difference in the components & features because we are not changing packages or binary structure. In fact, we would have a P2 repo with compatible features which all work together. Senaka, Isuruwan & Harshana have already started working on a PoC. Thoughts welcome. Those who were in these discussions, please add anything I may have missed. -- *Afkham Azeez* Director of Architecture; WSO2, Inc.; http://wso2.com Member; Apache Software Foundation; http://www.apache.org/ * <http://www.apache.org/>* *email: **az...@wso2.com* <az...@wso2.com> * cell: +94 77 3320919 blog: **http://blog.afkham.org*<http://blog.afkham.org> *twitter: **http://twitter.com/afkham_azeez*<http://twitter.com/afkham_azeez> * linked-in: **http://lk.linkedin.com/in/afkhamazeez <http://lk.linkedin.com/in/afkhamazeez>* *Lean . Enterprise . Middleware*
_______________________________________________ Architecture mailing list Architecture@wso2.org https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture
