A question of long-running and disruptive branches
- by Matt Enright
We are about to begin prototyping a new application that will share some existing infrastructure assemblies with an existing application, and also involve a significant subset of the existing domain model.
Parts of the domain model will likely undergo some serious changes for this new application, and the endgame for all of this, once the new application has been fully specified and is launch-ready is that we would like to re-unify the models of the two applications (as well as share a database, link functionality, etc.), but for the duration of development, prototyping, etc, we will be using a separate database so that we can change things without worrying about impact to development or use of the existing application. Since it is a prototype, there will be a pretty long window during which serious changes or rearchitecturing can occur as product management experiments with different workflows, different customer bases are surveyed, and we try and keep up.
We have already made a Subversion branch, so as to not impact concurrent development on the mature application, and are toying with 2 potential ways of moving forward with this:
Use the svn branch as the sole mechanism of separation. Make our changes to the existing domain models, and evaluate their impact on the existing application (and make requisite changes to ProjectA) when we have established that our long-running side branch is stable enough for re-entry to trunk.
"Fork" the shared code (temporarily): Copy ProjectA.Entities to NewProject.Entities, and treat all of the NewProject code as self-contained. When all of the perturbations around the model have died down and we feel satisfied, manually re-integrate the changes (as granular or sweeping as warranted) back into ProjectA.Entities, updating ProjectA to use the improved models at each step (this can take place either before or after the subversion merge has occurred). The subversion merge will then not handle recombination of any of the heavy changes here. Note: the "fork" method only applies to the code we see significant changes in store for, and whose modification will break ProjectA - shared infrastructure stuff for example, we would just modify in place (on our branch) and let the merge sort out.
Development is hard, go shopping.
Naturally, after not coming to an agreement, we're turning it over to the oracle of power that is SO. Any experience with any of these methods, pain points to watch out for, something new entirely?
