The approach that has worked for me is: 1. Develop against trunk 2. Branch for release 3. When a defect is reported in that release, fix it in trunk if it manif

The approach that has worked for me is: 1. Develop against trunk 2. Branch for release 3. When a defect is reported in that release, fix it in trunk if it manifests there and then merge the commit to that release branch. This will prevent regression in the next branch. 3b. If the defect only manifests in the release branch, fix it there, and then skip merging from release-branch back into trunk. 4. Dis-allow any new feature work to be merged from trunk to the release branch after it is cut. Only defect work. I don't like the idea of merging from a release branch back into the trunk. I see branches as things that are cut, potentially hardened, and then discarded.