This month's MSDN magazine contains an article on feature toggles. The subject has been close to my heart in the last few weeks and I have been weighing up whether they would work for our projects or not .
For those, who are unaware of the term, here is a good post by Martin Fowler describing feature toggles and their merits. He is convinced that feature toggles is the way forward and should be used instead of feature branches. Here is another great blog post explains the differences and recommends to use Feature toggles.
I love the idea of having no features branches … makes life easier. However, my take is that feature toggles is not for everyone and every team. For someone like Plural soft who does continuous delivery (and they use feature toggles), the process is simple. Each release, results in some new "features" being added. The process is generally additive with software becoming more "feature rich" and there is control on the release pipeline.
Now, turn our attention to a simple "message broker" kind application that interface with multiple systems and has no UI. The application handles message say M1 from one applications, does something to it and pass on message M2 to another application. Now, let's say there is a change in message interface because the sending application is changing. We start with a feature to handle the new message interface. Since, the change is a few months away, we need to keep supporting the existing interface. In this case, if feature toggling is involved, we would have to create a parallel code path to handle the new message interface and direct to that code path with feature toggle. If it wasn't the case and we were using branching instead, the change in code would have been much simpler. So in essence, we have replaced the complexity of merging by having a more complex code change.
Take another example, this time we have to delete something from the application, let's say a web service from the system. The feature toggle mechanism would require us to modify it to error on invocation when the feature is on. Compare it with the alternative of removing the service altogether.
Similarly, let's consider a windows/web UI application. One of the features is to re-design of the screens. The redesign includes jigging around all the form controls and include some new graphics. With feature toggling approach, we will either have a condition on display of each of these changes or have a new form created altogether, choosing between the two based on toggle value.
These were only some of the scenarios where feature toggle wouldn't essentially simplify things in my opinion. Others might disagree and I would love to listen to them, so please post your comments if you have any.