Multiple NSEntityDescriptions Claim NSManagedObject Subclass

Post-automatic-caching

This should not happen anymore with NSPersistent[CloudKit]Container(name: String), since it seems to cache the model automatically now (Swift 5.1, Xcode11, iOS13/MacOS10.15).

Pre-automatic-caching

NSPersistentContainer/NSPersistentCloudKitContainer does have two constructors:

• init(name: String)
• init(name: String, managedObjectModel model: NSManagedObjectModel)

The first is just a convenience initializer calling the second with a model loaded from disk. The trouble is that loading the same NSManagedObjectModel twice from disk inside the same app/test invocation results in the errors above, since every loading of the model results in external registration calls, which print errors once called a second time on the same app/test invocation. And init(name: String) was not smart enough to cache the model.

So if you want to load a container multiple time you have to load the NSManagedObjectModel once and store it in an attribute you then use on every init(name:managedObjectModel:) call.

Example: caching a model

import Foundation
import SwiftUI
import CoreData
import CloudKit

class PersistentContainer {
    private static var _model: NSManagedObjectModel?
    private static func model(name: String) throws -> NSManagedObjectModel {
        if _model == nil {
            _model = try loadModel(name: name, bundle: Bundle.main)
        }
        return _model!
    }
    private static func loadModel(name: String, bundle: Bundle) throws -> NSManagedObjectModel {
        guard let modelURL = bundle.url(forResource: name, withExtension: "momd") else {
            throw CoreDataError.modelURLNotFound(forResourceName: name)
        }

        guard let model = NSManagedObjectModel(contentsOf: modelURL) else {
            throw CoreDataError.modelLoadingFailed(forURL: modelURL)
       }
        return model
    }

    enum CoreDataError: Error {
        case modelURLNotFound(forResourceName: String)
        case modelLoadingFailed(forURL: URL)
    }

    public static func container() throws -> NSPersistentCloudKitContainer {
        let name = "ItmeStore"
        return NSPersistentCloudKitContainer(name: name, managedObjectModel: try model(name: name))
    }
}

Old answer

Loading Core Data is a little bit of magic, where loading a model from disk and using it means it registers for certain types. A second loading tries to register for the type again, which obviously tells you that something registered for the type already.

You can load Core Data only once and cleanup that instance after each test. Cleanup means deleting every object entity and then saving. There is some function which gives you all entities which you can then fetch and delete. Batch delete is not available InMemory though so object-by-managed object it is there.

The (probably simpler) alternative is to load the model once, store it somewhere and reuse that model on every NSPersistentContainer call, it has a constructor to use a given model instead of loading it again from disk.

In the context of unit tests with an in-memory store, you end up with two different models loaded:

• The model loaded in your application by the main Core Data stack
• The model loaded in your unit tests for the in-memory stack.

This causes a problem because apparently + [NSManagedObjectModel entity] looks at all available models to find a matching entity for your NSManagedObject. Since it finds two models, it will complain.

The solution is to insert your object in the context with insertNewObjectForEntityForName:inManagedObjectContext:. This will take in account the context (and as a consequence, the context's model) to look for the entity model and as a consequence limit its search to a single model.

To me it seems to be a bug in the NSManagedObject init(managedObjectContext:) method which seems to rely on +[NSManagedObject entity] rather than relying on the context's model.