Have you thought about breaking out the jpa related code base so that you only need to branch those bits? You could then have e42 and e43 plugins, included by jpa e42 and e43 features that are declared as optional in the main feature. In that case, all the plugins and features have the same version and the correct child feature gets installed based on the wtp version. I'm not sure about the update process (I worry you'd have to uninstall the previous jpa feature prior to installing the new feature). ------------------------------ Any input is welcome. ---------- Forwarded message ---------- From: Fred Bricon <fbricon(a)gmail.com&gt; Date: Thu, Jan 10, 2013 at 4:29 PM Subject: Best way to handle 2 code streams To: M2E-WTP Developers mailing list <m2e-wtp-dev(a)eclipse.org&gt; Hi, first of all, let me wish you all a very Happy New Year. I hope it'll see m2e-wtp graduate from the eclipse incubator next June ;-) During last call, we decided m2e-wtp would provide a JPA configurator for both the Kepler and pre-Kepler streams, due to Dali moving its provisional API, meaning we now need to maintain two branches of code. So I've been trying to think at the best strategy that would lead to the minimum maintenance overhead while allowing to provide new cool features for everybody. 1/ have 2 different JPA features in 1 git repo : Somethink like org.eclipse.m2e.wtp.jpa.feature (for kepler and onward) and org.eclipse.m2e.wtp.jpa.e42.feature (pre-kepler). Both features would be built and appear in the same update site (version 0.17.0). This gives us all the code easily accessible for maintenance (from my point of view) as we wouldn't have to switch between branches in order to apply fixes in both JPA configurators. The biggest problem though, is we wouldn't be able to provide a way to upgrade from jpa.e42 to jpa. There's already a p2.inf hack allowing the upgrade from the jboss tools jpa feature to the eclipse one, that hack prevents us from adding the jpa (juno) to jpa (kepler) upgrade path (Tycho breaks seeing the same IU being built twice). 2/ Have 2 m2e-wtp versions in 2 branches of a git repo We'd provide 2 separate update sites: * a "pre-kepler" p2 repo with m2e-wtp 0.17.0, built from a git 0.17.0 branch * a kepler repo with m2e-wtp 0.18.0/1.0, built from master. This would be easier to build, from a tycho perspective, and we'd still be able to upgrade from juno to kepler AND from jboss tools. But it would bring a *huge* overhead in non-JPA maintenance. 3/ Have JPA in a dedicated git repo, use specific qualifiers : We'd provide 2 separate update sites: * a "pre-kepler" aggregated p2 repo with m2e-wtp 0.17.0, built from master and a jpa.0.17.0-e42.timestamp site, built from the e42 branch of the m2e-wtp-jpa git repo, * a kepler aggegated p2 repo with m2e-wtp 0.17.0, built from master and a jpa.0.17.0-e43.timestamp site, built from the master branch of the m2e-wtp-jpa git repo, It requires 2 tycho builds and more manual hacking of the aggregated sites, but would be easier to maintain, from a code perspective. And we'd still be able to upgrade from 0.17.0-e42 to 0.17.0-e43. I think I like #3 better (but I haven't tested the approach works yet). What do you think? Fred Bricon -- "Have you tried turning it off and on again" - The IT Crowd -- "Have you tried turning it off and on again" - The IT Crowd _______________________________________________ jbosstools-dev mailing list jbosstools-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/jbosstools-dev

Yep. That's what I was thinking. As I said, I'm not sure about the upgrade process. My concern is that once one of the child features is installed, you may not be able to update at all (i.e. a wtp upgrade would balk because the e42 plugin isn't compatible). That said, users probably won't be using the update mechanism when updating the platform, so you should be good.

Good catch. Sorry. Although....What if the containing feature depended on dali? The user wouldn't be able to install it without dali, and the correct plugins would be installed if they did. WDYT?

That said, users probably won't be using the update mechanism when updating the platform, so you should be good.