Refactoring table view data source and delegate methods

Tired of MVC? But MVVM and VIPER don’t feel quite right? There are reasons for it. Code dependence. Data dependence. Unidirectional flow of control. Just to name a few issues for these alternatives. If you just master the simple VIP cycle in Clean Swift, you can isolate your code into small and testable components. You can solve all of these problems, and have a beautiful, consistent architecture for your iOS apps.

A lot has been said about the massive view controller problem and how to slim it down. Specifically, you’ve most likely seen how to move the table view data source and delegate methods out of the view controller. The result is fewer lines of code in the view controller.

In this post, we are going to look at some code and different approaches to see if it’s really such a good idea. If not, what should we do instead? Hopefully, you’ll come away with new insights on what you want to do with your own apps.

You can find the source code for this sample project on GitHub.

Approach 1: Do nothing

You can do nothing. You can just leave the table view data source and delegate methods in the view controller.

Let’s look at a really simple app that lists some ice cream flavors.

The following IceCreamListViewController is very typical. It implements the following UITableViewDataSource methods:

  • numberOfSectionsInTableView(_:)
  • tableView(_:numberOfRowsInSection:)
  • tableView(_:cellForRowAtIndexPath:indexPath:)

The IceCreamListViewController looks like the following if you choose to do nothing:

The IceCreamStore is just a very simple Swift class that privately keeps track of some ice cream flavors. It has just one allFlavors() method that returns these flavors. It is there so that you don’t have to store the flavors directly in the view controller.

The following is the full implementation of IceCreamStore:

There is nothing wrong with this code. But the three UITableViewDataSource methods reside in the IceCreamListViewController.

Let’s try to move them out of there.

Approach 2: Refactor the table view data source methods to a separate class

One very common technique that you’ve probably seen many times elsewhere is to move the UITableViewDataSource methods into its own class.

The following IceCreamListDataSource class conforms to the UITableViewDataSource protocol and implements the three methods we’re interested in. The implementation detail is exactly the same. We’re just moving code around, not changing behavior.

A couple side notes:

  • IceCreamListDataSource inherits from NSObject so that we don’t have to fiddle with the @objc and @class keywords because UITableViewDataSource is an Objective-C protocol.

  • The override keyword is only needed for IceCreamListViewController because it is derived from UITableViewController which already conforms to UITableViewDataSource (i.e. already implements these data source methods, albeit maybe empty and meant to be overridden). This means IceCreamListViewController also inherits these data source methods. So you need to override them. But IceCreamListDataSource doesn’t inherit any data source methods so you just need to provide, not override, the implementation.

Now you’re left with a rather empty IceCreamListViewController. You just need to tell the table view to use the new data source by writing tableView.dataSource = dataSource in the viewDidLoad() method.

Approach 3: Move the table view data source methods to a Swift extension

If the goal is to reduce code in the view controller, what if we just move the data source methods out of the view controller and into a Swift extension.

Once again, the IceCreamListViewController is very empty, and the extension looks identical. The data source method implementation hasn’t changed either.

Note that Swift extension does not allow us to define stored properties. So we have to use a computed property that returns an instance of IceCreamStore(), like it used to before when defined as a constant.

Comparison of the three approaches

Let’s compare the pros and cons of the 3 approaches that you’ve seen based on:

  • Refactoring
  • Reusability
  • Ease of use
  • Maintainability

Refactoring

The do nothing approach is arguably the easiest, because you don’t have to do anything. But your view controller will increase in size over time because of new features and edge cases. When that happens, you’ll end up with a massive view controller.

Before we dismiss this approach, let’s take a deeper look though. When your view controller increases in size, what kind of stuff gets added. Is it more view controlling logic for your labels, text fields, pickers, and other custom views you may have? Or is it business logic that shouldn’t really be there?

If it’s business logic, you should really move that out of the view controller, instead of the view controlling logic such as table view data source and delegate methods.

If you have a lot of view controlling logic, you then think about how to move them out of there, using either new classes, categories, or extensions.

I think a lot of developers think about solving the massive view controller problem in the opposite direction. To get something working, they put everything in the view controller. When it turns messy, they then think about what easy things they can pull out to slim it down.

I believe we should go about this in the other way. My general rule of thumb is, by default, everything is out of the view controller. When you add code, think about what kind of code you are adding. If it’s business logic, put them in the interactor. If it’s view controlling logic, put them in the view controller.

Instead of default-in-selective-out, it should be default-out-selective-in.

Reusability

The class approach may seem beneficial at first. Your view controller has fewer lines of code. Your new data source class can potentially be reused.

So let’s talk about reusability. The table view data source and delegate methods are very easy to write. How much values does it really add if you spend a lot of time and effort to come up with a generic data source so that you can reuse it in another UITableViewController subclass?

In this simple example, not much. But in a real app, you are likely to have more multiple and/or more complex table views. They may get the data synchronously and asynchronously. They may format the data differently. They may have different section and row logic. Some sections may need to be hidden and some rows are need to be dynamic. Who knows what the UI/UX designers can come up with in their dreams?

If you were to make this generic data source reusable for all these situations, it can very well turn into a massive data source itself! It’ll be harder and harder to read and understand, making it more difficult to use by all your UITableViewController subclasses, by other developers on your team who are less familiar with the data source.

If it’s easier to write them out every time than to make a generic data source class, why do we care so much about making it reusable? On the contrary, is DRY really that important if the data source code is trivial to write? Which one is easier to use? Which one do most developers already know very well?

I’ll take readability over reusability.

Ease of use

When it comes to the effort you make to create this generic data source class. Does it take more time to do nothing or do something? It’s obvious in the short term. But over the long term, you better do something now before your view controller becomes massive. That’s why you are reading, right?

So, after throwing the do nothing approach out the window, should we use the class approach or the extension approach?

The class approach requires you to manually set the table view data source to your new custom data source. If you forget, weird things happen. You also need to make sure all developers on your team know about this data source, know how to use it, know to set it in viewDidLoad(). BTW, don’t you already have tons of stuff in viewDidLoad()? Your new data source class also needs to inherit from NSObject or use the @objc keyword. That’s just fact of life for Objective-C interoperability. Oh, you also need to get rid of that override keyword?

The extension approach is just straight copy-and-paste. Yes, you do need to make your data source a computed property because Swift extension does not allow stored properties. Mind you, Objective-C category does not allow stored properties either.

I’ll take extension over class.

Maintainability

In a real app, your custom data source class is likely to be more complex than what I’ve shown above, as we discussed about reusability. When you need to make the data source work for one more view controller, it’s going to be harder to add this new feature. Or, when the data source has bugs, it’s also going to be harder to fix.

But if you do nothing or use an extension, even though it’s not as DRY, it’s going to be easy to modify. After all, it’s not meant to be reused. It only needs to work for this one view controller. It doesn’t need to work for all your view controllers in your app, or even across other apps.

Suppose you don’t build this custom data source on your own. And you use a popular data source solution from GitHub. Does the author have time to fix issues, merge pull requests, and provide support? What if the data source break when new versions of Xcode and Swift come out? How soon will it get back working again? Can you afford to wait?

Free software is free software. You aren’t paying the author to do all this free work. Suppose the author is so gracious and do a reasonable amount of support. What if he changes his mind and loses interest in further maintain the data source repo? He may move on to other more interesting and rewarding projects.

Apple, as a company, is much more likely to maintain UITableViewDataSource and UITableViewDelegate protocols as long as the iOS platform is not deprecated. When new versions of Xcode and Swift come out, they are guaranteed not to be broken.

Apple’s support lasts longer than any individual’s.

Conclusion

I found refactoring table view data source and delegate methods from the view controller into a separate class is just not that useful. It requires more effort. Reusability is overrated. A super generic, reusable data source class is harder to get it right and maintain.

Often, it’s very easy for us developers to get overly creative and fall into the trap of making something cool. But it’s also important to be pragmatic when we refactor.

My preferred solution is:

  • For a simple view controller, where there is only one table view, leaving the data source and delegate methods right in the view controller isn’t bad.
  • For a complex view controller, where there are maybe, 4 or 5, sibling or nested, table views, separating each into its own Swift extension or Objective-C category may often be enough.
  • If you truly have a need for a generic table view data source, I recommend watching the WWDC 2014 – Advanced User Interfaces with Collection Views session. Just make sure the time-vs-benefit tradeoff is worthwhile.

The key is to move all business logic out of the view controller and into the interactor while keeping the view controlling logic in the view controller. I believe that is more important in slimming down a massive view controller than refactoring out table view data source and delegate methods.

The source code used in this post is on GitHub.

Raymond
I've been developing in iOS since the iPhone debuted, jumped on Swift when it was announced. Writing well-tested apps with a clean architecture has been my goal.

22 Comments

  1. Excellent points.

    I know that is one of the things I continually have to work on myself, it’s just so easy to come up with a “creative” approach and when you start working with it, keep adding the “one more thing” to it, until it actually becomes more confusing and complex then what it replaced.

    Your statement “Often, it’s very easy for us developers to get overly creative and fall into the trap of making something cool. But it’s also important to be pragmatic when we refactor.” really hits home!

    1. Hi Darren,

      Thanks for the nice words!

      One thing I found that helps me avoid being overly creative is to just stick to the Clean Swift system. When I need to add a piece of code, I ask myself – Is this business logic, presentation logic, or display logic? Then I just follow the VIP cycle. Everything just falls into place naturally.

  2. Wonderful analysis. :]

    In general, I really like this approach and follow it in my own apps.

    One caveat, however, is situations when the table view data requires multiple representations: perhaps there’s a “detailed” and “simple” way the data needs to be shown.

    In this case, I would prefer pulling out the data source into a class, and most likely, I would create two data source classes, one to represent each way of showing the data.

    However, I’d treat these data sources as not reusable. As you mention, use cases are bound to diverge between different view controllers over time- after all, they’re different view controllers ;] – and I don’t even attempt to make a data source class shareable.

    Thanks for sharing this post!

    1. Hi Joshua,

      Thank you. And you made a really good point!

      For example, you have a segmented control that switches either the data to be shown, or the look, or different sections. I think having multiple data sources is a perfectly fine solution. The other option is to litter the data source and delegate methods with a lot of conditionals.

      And yes, data sources are very easy to build. Most of the time, you have an array of some data. And you just need to pull stuff out and display them. Trying to make them reusable in other view controllers almost always seem a bit outstretched.

  3. Hello Raymond,

    2 questions about the extension approach.

    I use separate nibs to create cells, where do you register its identifiers if not in the main view controller?
    I am failing to see how to use your dataStore approach without static data.

    For example, using your templates when my view controller gets called with displaySomeData, and it needs to display it on a tableView. How do I abstract this from the main view controller?

    Thank you!

    1. Hi Miguel,

      You can just register your cells in the table view controller.

      The data store can be anything. I use static data for simplicity sake. In practice, you may get the data from Core Data or over the network. Just make sure these details are contained in the data store object and abstracted away from the rest of your app.

      If you follow the VIP cycle and use the request, response, and view model objects, things will already be isolated.

      Does it help?

  4. Hi Raymond,

    Thanks for this very informative post. Learned a great deal from you. I’d like to hear your thoughts on this: what about the case when the data source is being dynamically updated? How do you suggest breaking things up?

    For example, let’s say the data source is a list of GPS locations, which is continuously provided by CLLocation. Let’s also say there will be some additional logics to be done on this data source. Would you leave these logics in the subclassed Data Source file? Or would you move that logic to another separate file?

    1. Hi LTran,

      If your app needs to be notified of continuous GPS updates, you can have a LocationManager that receives these updates and invokes the completion handler to notify your interactor. Your interactor can invoke your presenter multiple times (i.e. output.presentLocationUpdates()).

      Does this work?

  5. Great Article on Refracting Table View.
    Third Approach shows one of the great feature of Swift which can be implemented in refracting View Controller for Clean-Swift .

  6. Hi Raymond,

    Great post. You make a really great point about ‘default-out’ when it comes to view controllers.

    It’s something which will perhaps help me better understand the true architecture of my apps if every time I’m writing a line of code I’m asking’ “Should this belong in the view controller?”.

    Thanks!

  7. I’d argue that the extensions approach (option 3) is functionally no different than the “do nothing” approach (option 1).

    What is the problem we’re trying to solve? Is it that the tableView has become a god object, or is it that the source file itself is too large to maintain? In either case, option 3 doesn’t solve either problem. Using extension doesn’t eliminate the god object problem being that the new functionality still lives within the tableView itself. Also, your extension are still within the same source file, which does nothing to solve the second problem. Adding extensions that exist within the same file is functionally equivalent to adding spacing and comments to your existing tableView. What would be different if you simply adding more spaces below line 11 in option 1?

    In many applications, the data source (just like the model) rarely changes as often as behavior. That being the case, it’s probably better to factor it out into it’s own class and concentrate on the code that actually does need to change. Keeping it without the same source file simply adds unnecessary complexity that the developer doesn’t need to view.

    1. I wondered that too while I read it, but is it like food off a plate?
      I can eat my meat from my potatoes and veges separately, or I can mash ’em up and quaff it all down together. My mum goes pale with disgust, but I counter “..it all ends up together anyway”.

      However microbiologists will tell us that ingestion order is important for gut flora to work well. So maybe Outclassing vs Extension invokes different Memory Handling by the compiler. Because of Oo we don’t get concerned at what happens at bare metal, but is that a mistake?

      I’m tossing up whether to use Core Data as opposed to a dictionary of rather large and complex arrays as a datasource. Ultimately stuff is held in RAM like food in my belly, but I always think about balancing processing power ( battery life ) of computed properties vs speed of access against likelyhood of storage space available at the time.

      Great article Raymond, not only for its content, but to show decision processes in the design.

  8. Hey Raymond!

    Very nice article 🙂 However, I’m really in doubt here. Going for an extension over a class would make the datasource harder to test, wouldn’t it?

    Thanks!

    1. Hi Matheus,

      A Swift extension used in this way simply group related methods together. The methods still belong to the same class. Why would it make it hard to test?

  9. Hi

    Good tutorial. However, i have one question.

    how can i implement didSelect Method if i separate the datasource into separate file. And if i want to add a selector function to a button in the cell?

    1. Hi Awais,

      There is very little benefit in separating data source and delegate methods into its own class. If your view controller does become massive, first consider moving these methods to an extension. IBAction selector methods are the same. It doesn’t matter whether they are in the same file or not.

  10. Thanks Raymond.

    I did not know that “It doesn’t matter whether they are in the same file or not.”

    So i moved approx 600 lOC in the separate file myVCExtension.

    Thanks for the reply. It really helped me.

  11. Hi,
    With approach 2,I have one button in Cell,So I want to go to next View Controller on click on the button.How can I manage the button click?

  12. Where should I make a customTableViewCell class for working with the prototype cell content view elements in clean swift.

    before this i was working in a new swift cocoa touch class file with inheriting the TableViewCell class

Leave a Comment

Your email address will not be published. Required fields are marked *