Sunday, January 27, 2013

When should you start using a source control?

My eyes were immediately drawn to this thread on Ars Technica: "When should I make the first commit to source control?", with answers from Stack Exchange users.

I'm from a team at IBM that is building a source control. So getting my opinion is a bit like asking your stock broker what he or she thinks about the merits of investing on the stock market. Of course I'm convinced the source control should be a transparent part of your development - you should not even have to think about it while you design and edit your code. It should just be there when you're in a bad state and you want to return to a know good one. Aka, you should not even ask yourself when to make the first commit...

A source control doesn't have to be in the way of your dev work. It knows when you save files, and in many IDEs and products out there, it even knows what task or defect you are working on. It can work in the background, safely pushing your changes from your local drive to a remote machine, so you can crash your hard drive or pursue your work from a separate machine in the evening.

A source control is now a basic feature like syntax highlighting, code assist or refactoring. It's a given that whatever you've done, you can review and roll back in time, fork and try something else. Suspend your current work, resume some other on-going work. Work with a buddie, with a team. And when you can do all that without thinking about having to use a source control, then we've succeeded.

The first tip I give about using a source control is to check-in frequently. By that I mean storing your changes frequently in the source control, which isn't the same as giving every one of your changes to your team. Many source controls have a staged approach - you can version control your own changes in a private way then control when you find them useful and good enough to be shared with others.

In the 8 years of RTC Source Control, we tested different source control workflows. In the early days, we were really keen on something called auto check-in. In 1.0, that feature was on by default. You save a file. Bing. That change is safely backed up on the RTC server, in your private workspace. Some of us were wondering - why bother asking the user when to manually check-in? Why not doing it automatically for every file that is modified? It turned out to be a very divisive matter. 
  • Some really thought that was the way of the future - just have every change you do automatically be replicated in your backup on the server. 
  • And others really wanted to stay in control as to when their changes leave their local drive and become part of the source control's memory.
I was initially a strong proponent of the auto check-in workflow. I always thought the ones opposing it were a bit like the mathematician Gauss as described by Abel: "he is like the fox, who effaces his tracks in the sand with his tail". Like you don't want others to know all the wrong paths you've explored before you actually perfected the fix for a defect... That's too bad, because e.g. RTC Source Control has a way to highlight the initial and final versions in a change set and make the intermediate versions mostly hidden in regular operations. 

So why did we turn off auto check-in by default? Because many users were hit by some of its drawbacks.
  • Network latency when you work over a poor wifi connection at the airport
  • Big mess when a mistake you do in your IDE refactors 1000 files.
  • Your history is filled with meaningless changes (if you use auto-complete of change sets) or with huge change sets if you forget to help the tool and complete your change sets from time to time
I did experience all these annoyances myself and was convinced that it wasn't appropriate for a majority of users in its current form. It's fantastic for those users who like the convenience and remember to complete their change sets at appropriate stable moments. I don't use auto check-in anymore. But I frequently check-in my changes so that I can easily go back to a previous good state, and in RTC we make it easy with a simple "Check-in all" button. I only deliver my changes to the stream used by my team when it's ready - after a green personal build for example.

If the source control you use makes a commit sound like a complicated, slow and cumbersome task to perform, you're likely missing the greatest strength of a source control. It's invisible to you when things work, and it's there when you need it. So, in conclusion, next time you create a project, put it under version control right away... Version control is good for you even if you don't intend to share with other peers right away.