Pair programming is the concept of two developers working together, in real-time, on the same code base on the same device. There are some variations and techniques that vary from organisation to organisation around pair programming, but to me, it boils down to the concept of “Two brains are better than one, depending on the problem”. A spin-off is mob programming, where the whole team works on the same thing.
Pair programming is ideal when onboarding a new developer into a code base by solving a problem, like building a new feature or ﬁxing a bug. In day-to-day work, pair programming is helpful in capturing bugs that a single developer wouldn’t had noticed. It’s also great for reasoning about high level strategic parts of the code together with another developer.
This post isn’t about the pros and cons of pair programming (there are a ton of material to digest about that online).
But what about remote work and pair programming? Since the physical aspect of traditional pairing is one of the main pros, it has been hard to replicate when all developers are working remote.
At work, we started a project where we needed to start off with a whole new code base. The task at hand involved rewriting an existing web frontend in React views, built on a functional reactive software pattern. We also had to integrate it against our existing Meteor backend, for data loading and user authentication. Finally, it needed to talk with an in-house WebRTC service.
So we had a few unknowns:
- How do we handle TypeScript compilation in the existing code base?
- How to intergrate new React views in the Meteor frontend?
- How to drive the frontend state from the Meteor backend?
- How to manage frontend state and WebRTC connections with the functional reactive pattern?
I had an inkling on how to solve 1) and 2). I had very little experience in 3) and 4). A colleague, who was a big driver behind the functional reactive pattern, had expertise in 3) and 4). But due to our remote setup at the company, we couldn’t sit down for a hack day and ﬂesh things out together.
The problems with remote and pair programming are thus:
- Not being in the same physical space — tricker to have high bandwidth communication.
- Cannot work on the same computer.
- Cannot with ease see what the other developer is seeing.
“Okay, Johan”, you say. “You can remedy at least 2) and 3) in that list with remote control applications and screensharing”. Yes you can! That’s precisely what this post is about ✨
What if I told you that you can pair program like you’re pair writing on a Google Doc, right from inside your code editor?
Well, you can.
Me and my colleague landed on these two tools:
- VS Code
- Slack voice call
Co-operating on the code
This was the deal breaker for us: how to co-operate on the code base without introducing old school solutions like remote control software? We wanted this to feel like your own editor, even you were remote.
Luckily, the developers behind the popular editors Atom and VS Code have worked on this.
We’ve only used VS Code, so I’m going to cover that only in this post.
VS Code’s Live Share works like this:
- One developer hosts a session on their computer, by installing the extension.
- The developer hits “Share” in the editor’s bottom status bar. VS Code will ask them to auth with GitHub or Windows Live (yea, right…).
- Once that’s done, VS Code puts a link in the clipboard, and the developer can send it to co-workers. Hitting the “Share” button again brings up a menu to manage the sharing session.
You can now:
- Share a server
- Share a terminal
When your co-developer joins, they can navigate your workspace right from within their own editor.
Talking along the way
The solution here is to set up a voice or video call on the side of VS Code or Atom. It’s really great to be able to talk alongside working with the code with your colleague.
Just because you’re working remote doesn’t have to mean that you’re “on your own” with tricky code problems. It’s totally doable to pair program when you’re stuck, or need guidance in a new system or programming language. Existing solutions include screen sharing and remote control and all kinds of boring setups. Having code sharing right from within your editor is really great, which makes it easier to focus on the code and collaboration. The fact that you are working remote won’t matter when your colleague is just a link away from your editor.
Some features are still lacking in the VS Code extension to make it feel “just like my own editor”. The most annoying one is that the guest isn’t able to jump between ﬁles. And I’d love a nicer way to highlight “Here I am!” from within the editor, since you’re probably gonna jump around in the project structure.
- On the tech side, I’m not sure exactly how VS Code has built the code sharing architecture. There’s no open source repository for the project, except for a documentation repo. That’s too bad… The Atom folks have built their system in WebRTC, where the initial handshake goes to their servers, but the actual code sharing goes over Peer-to-peer connect to avoid any middle men.
- We experienced some weird bug in VS Code at the time of writing where the text editor would freak out when one of us hit
cmd+zto undo a text input. The editor would go haywire and delete most of the code written to that point. But then it was magically restored.