Putting Core Data on the Map

Core Data is a powerful framework for all kinds of data persistence, and its NSFetchedResultsController is a key class in many an app. However, its API – especially its delegate protocol – is aimed mostly at table views, and can be a little difficult to connect to other UI classes.

In this post, I’ll walk through the process of hooking up a Core Data stack to an MKMapView. While mostly straightforward to display NSManagedObject instances on a map, there are a few tricks to building a solid app and keeping your map data up to date.

The Setup

The genesis of this article was in trying to maintain an MKAnnotation for each managed object fetched by a given FRC. (For the sake of argument, let’s call the managed objects here Points.) Right off the bat, this puts a few constraints on the problem:

  • MKMapView instances maintain only a flat array of annotations. I didn’t need to worry about keeping Points in sections; in fact, the source FRC could be unsectioned.
  • The latitude and longitude of each annotation needed to come from the Point in question. I wanted to persist the location – and, if possible, the title – of each Point in Core Data.
  • Because of how users could modify Points, I needed the map to be able to respond to any kind of FRC update – inserts, deletes, and updates.

The Core Data setup was fairly straightforward: I wanted Point to store latitude, longitude, and title, so I added two numeric and one string attribute to the Point entity. A quick class-generation pass later, and I was in business, ready to tackle the map.

extension Point: MKAnnotation

Right off the bat, it seemed like the easiest thing to do was to make the Point instances themselves conform to MKAnnotation. They already had latitude, longitude, and title; all that was left was to merge the former into a coordinate property. This wound up taking the following form:

extension Point: MKAnnotation {
    public var coordinate: CLLocationCoordinate2D {
        // latitude and longitude are optional NSNumbers
        guard let latitude = latitude, let longitude = longitude else {
            return kCLLocationCoordinate2DInvalid
        }
        
        let latDegrees = CLLocationDegrees(latitude.doubleValue)
        let longDegrees = CLLocationDegrees(longitude.doubleValue)
        return CLLocationCoordinate2D(latitude: latDegrees, longitude: longDegrees)
    }
}

Then, in the UIViewController managing the map, it was straightforward to create an NSFetchedResultsController fetching Points; become its delegate; and populate the map with its fetchedObjects, all of which could be passed straight to addAnnotation(_:).

This seemed promising: Point instances appeared on the map on launch. What’s more, with a small implementation of a single NSFetchedResultsControllerDelegate method, the map could dynamically update its pin annotations with Core Data changes:

func controller(_ controller: NSFetchedResultsController<NSFetchRequestResult>, didChange anObject: Any, at indexPath: IndexPath?, for type: NSFetchedResultsChangeType, newIndexPath: IndexPath?) {
    guard let point = anObject as? Point else {
        preconditionFailure("All changes observed in the map view controller should be for Point instances")
    }
    
    switch type {
    case .insert:
        mapView.addAnnotation(point)
        
    case .delete:
        mapView.removeAnnotation(point)
        
    case .update:
        mapView.removeAnnotation(point)
        mapView.addAnnotation(point)
        
    case .move:
        // N.B. The fetched results controller was set up with a single sort descriptor that produced a consistent ordering for its fetched Point instances.
        fatalError("How did we move a Point? We have a stable sort.")
    }
}

On the surface, this seemed like it would work well, even if updates were a little heavy-handed. (I’d like to explore a way of leaving annotations alone on Core Data update if the change didn’t affect a visible property of the pin annotation, but that’s a topic for another day.) Most importantly, we’re not creating any additional objects to keep track of – the Point instances themselves serve as MKAnnotations here.

The Sticking Point

In testing, though, this soured quickly: a critical operation in everyday use was deleting a Point from the database. However, telling the map view to remove a Point annotation wasn’t quite as prompt as it should have been. Annotations would remain on the map for a few seconds, or in rare cases much longer.

At first, this problem had all the symptoms of a threading issue. Manipulating views off the main thread is a common error, and can produce long-delayed view updates (or crashes, in more recent iOS releases). However, setting breakpoints in the FRC delegate method revealed that it was always called on the main thread – which made sense, given that the FRC’s managed object context was a main-thread context.

Further debugging led me in circles, and I began considering more complex approaches to the problem. At one point, the app in question was maintaining a set of model-like MKAnnotation-conforming objects, and manually mapping between them and the source Points for most map operations – a complex situation that was rapidly growing unmaintainable. It wasn’t until I went back to the MKAnnotation documentation that I found the answer.

A Key Observation

In the “Discussion” section of the docs for MKAnnotation.coordinate, there exists this critical qualifier:

Your implementation of this property must be key-value observing (KVO) compliant. For more information on how to implement support for KVO, see Key-Value Observing Programming Guide.

As soon as I noticed this, everything clicked: the map was likely using KVO to watch for changes in an annotation’s coordinate, in order to update its location. Deleting a Point changed its coordinate, but didn’t post the right KVO notifications, so the map’s display would lag until it noticed the change in annotation through some other means.

Thankfully, KVO has a mechanism meant for precisely this situation. When one property’s value is derived from others, implementing a specifically-named class method can express that dependency, allowing the KVO machinery to notify observers of changes flowing from the original property or properties.

The format for this method’s name is keyPathsForValuesAffecting<KeyPath>(), and it needs to return a set of strings. In our specific case, we can add this method alongside the implementation of coordinate in the Point extension as:

extension Point: MKAnnotation {
    var coordinate: CLLocationCoordinate2D: { /* same implementation as above */ }

    class func keyPathsForValuesAffectingCoordinate() -> Set<String> {
        return Set<String>([ #keyPath(location.latitude), #keyPath(location.longitude) ])
    }
}

This tells the KVO system that changes to location.latitude and location.longitude will both affect the value of coordinate. Any changes coming up the Core Data stack to those two properties – such as deleting a Point – will trigger KVO notifications for any observers of coordinate, including the MKMapView.

After this change, deleting a Point (using a separate UI, or just by poking the managed object context directly) caused its annotation to disappear from the MKMapView immediately. From there, all that’s left is a little more UI polish, and we’re ready to ship an app that can show Core Data objects on a map!