Engineers Detail Extensive Efforts to Rewrite Facebook iOS App

One of the biggest advantages we've gained from building on native iOS has been the ability to make the app fast. Now, when you scroll through your news feed on the new Facebook for iOS, you'll notice that it feels much faster than before. One way we have achieved this is by re-balancing where we perform certain tasks. For example, in iOS, the main thread drives the UI and handles touch events, so the more work we do on the main thread, the slower the app feels. Instead, we take care to perform computationally expensive tasks in the background. This means all our networking activity, JSON parsing, NSManagedObject creation, and saving to disk never touches the main thread.

To give another example, we use Core Text to lay out many of our strings, but layout calculations can quickly become a bottleneck. With our new iOS app, when we download new content, we asynchronously calculate the sizes for all these strings, cache our CTFramesetters (which can be expensive to create), and then use all these calculations later when we present the story into our UITableView.

Finally, when you start Facebook for iOS, you want to see your news feed, not a loading spinner. To provide the best experience possible, we now show previously-cached content immediately. But this introduces a new problem: If you have a lot of stories in your news feed, UITableView throws a small spanner in the works by calling the delegate method -tableView:heightForRowAtIndexPath: for each story in your news feed in order to work out how tall to make its scrollbar. This would result in the app loading all the story data from disk and calculating the entire story layout solely to return the height of the story, meaning startup would get progressively slower as you accumulate more stories.

The solution to this particular problem has two main parts. Firstly, when we do our initial asynchronous layout calculations, we also store the height of the story in Core Data. In doing so, we completely avoid layout calculation in -tableView:heightForRowAtIndexPath:. Secondly, we've split up our "story" model object. We only fetch the story heights (and a few other things) from disk on startup. Later, we fetch the rest of the story data, and any more layout calculations we have to do are all performed asynchronously.