Skip to content

Instantly share code, notes, and snippets.

@chriseidhof
Created February 18, 2012 16:39
Show Gist options
  • Star 69 You must be signed in to star a gist
  • Fork 6 You must be signed in to fork a gist
  • Save chriseidhof/1860108 to your computer and use it in GitHub Desktop.
Save chriseidhof/1860108 to your computer and use it in GitHub Desktop.
Accessing an API using CoreData's NSIncrementalStore

Accessing an API using CoreData's NSIncrementalStore

Note: the original location of this article is on my blog, however, it is posted here too for better readability.

In this article, we will see how to use Core Data for accessing your API. We will use the Bandcamp API as our running example. I've only been experimenting with this code for a few days, so there might be mistakes in there.

One of the problems with accessing an API is that you typically have API calls everywhere in your code. If the API changes, there are probably multiple spots in your code that you have to change too. Your code knows about the structure of the API results in lots of places, for example, in the Bandcamp API there is a track entity which has the property title. It would be easy to pass the API results around as an NSDictionary and lookup the title key in that dictionary. However, if they would change it to songtitle, you have to find this everywhere in your code.

Another problem is that most APIs are not object oriented. Suppose you have an Album entity that has a to-many relationship with a Track entity: each album can have multiple tracks. In your controller, you will probably have multiple API calls, one for getting the album and another for getting its tracks.

By using a new feature in Core Data we can solve these problems by adding another layer on top of the API which allows us to access the API as if it were an object graph. Entities can be concrete subclasses of NSManagedObject.

First, we will build a regular class that accesses the API and parses the JSON. Then, we will create a CoreData data model that represents the API in an object-oriented way. Finally, we will create an NSIncrementalStore subclass and implement the necessary methods to fetch the entities and relationships.

Step 1: Wrap the API

The first step is to create a simple class that implements your API. Doing this is straightforward, and I will not go into details here. This is the header file for the Bandcamp API:

@interface BandCampAPI : NSObject
+ (NSArray*)apiRequestEntitiesWithName:(NSString*)name 
                             predicate:(NSPredicate*)predicate;
+ (NSArray*)apiDiscographyForBandWithId:(NSString*)bandId;
@end

In summary, you can search for bands, get a band by id, get an album by id, get a track by id. If you request the info for an album, you also get a list of its tracks included in the response. Finally, there is a method for getting the discography of a band.

To find all bands named "Rue Royale" we can do the following API call:

NSPredicate* predicate = [NSPredicate predicateWithFormat:@"name == %@", @"Rue Royale"];
NSArray* bands = [BandCampAPI apiRequestEntitiesWithName:@"Band"
                                               predicate:predicate];

This will return an NSArray with an NSDictionary for each found band. Please have a look at the source to see how it is implemented. The result is as following:

({
    "band_id" = 4246760315;
    name = "Rue Royale";
    "offsite_url" = "http://rueroyalemusic.com";
    subdomain = rueroyale;
    url = "http://rueroyale.bandcamp.com";
})

The discography call looks like this:

NSString* sideditchId = [NSString stringWithFormat:@"2721182224"];
NSArray* albums = [BandCampAPI apiDiscographyForBandWithId:sideditchId];

And the result like this:

({
    "album_id" = 3366378415;
    artist = Sideditch;
    "band_id" = 2721182224;
    downloadable = 1;
    "large_art_url" = "http://f0.bcbits.com/z/70/81/70810089-1.jpg";
    "release_date" = 1267401600;
    "small_art_url" = "http://f0.bcbits.com/z/37/58/3758272301-1.jpg";
    title = "Mary, Me Demo";
    url = "http://sideditch.bandcamp.com/album/mary-me-demo?pk=191";
})

Now that we have access to the raw API, we can continue by making an object-oriented version of the API.

Step 2: Define the model

The next step is to create a new CoreData data model. For each API entity, we create a corresponding CoreData entity (named Band, Album and Track). There are properties for all of the entities, and more importantly: relationships between the entities. For example, the Album entity has a relationship tracks which is a to-many relationship to the Track entity. Creating this data model is exactly the same as creating a normal CoreData data model.

In this project, I've reused all the keys that Bandcamp uses. For example, an album has a key large_art_url, so our entity has a key like that as well. However, this is not necessary. We can name the keys anything we want, we just have to make sure that we convert them in our NSIncrementalStore subclass.

Step 3: implement the NSIncrementalStore methods

Now the hard bit: creating a subclass of NSIncrementalStore. There is a really interesting article by Drew Crawford about how to do this. However, it lacks a concrete example. I created a subclass BandCampIS of NSIncrementalStore. In the documentation, you can see which methods to implement. We will start with executeRequest:withContext:error:. This method is called for multiple purposes, but we will now focus on only one case: when it's called with an NSFetchRequest.

The first argument is of type NSPersistentStoreRequest which is the request we have to act upon. By inspecting its requestType we can turn it into a specific subclass, such as NSFetchRequest. For clarity, the handling code is factored out into a method fetchObjects:withContext, which we will define later.

- (id)executeRequest:(NSPersistentStoreRequest*)request 
         withContext:(NSManagedObjectContext*)context 
               error:(NSError**)error {
    if(request.requestType == NSFetchRequestType)
    {
        NSFetchRequest *fetchRequest = (NSFetchRequest*) request;
        if (fetchRequest.resultType==NSManagedObjectResultType) {
            return [self fetchObjects:fetchRequest 
                          withContext:context];
        }
    }
    return nil;
}

The fetchObjects:withContext method should return an NSArray containing NSManagedObject items. In the fetchObjects:withContext method, we call the appropriate API method, and get back an NSArray with an NSDictionary for each item. For each item, we create a new NSManagedObjectID and cache the values. Again, this is factored out into a separate method. Then, we call the objectWithID method of NSManagedObjectContext to create an empty NSManagedObject for the item.

- (id)fetchObjects:(NSFetchRequest*)request 
       withContext:(NSManagedObjectContext*)context {
    NSArray* items = [BandCampAPI apiRequestEntitiesWithName:request.entityName 
                                                   predicate:request.predicate];
    return [items map:^(id item) {
        NSManagedObjectID* oid = [self objectIdForNewObjectOfEntity:request.entity 
                                                        cacheValues:item];
        return [context objectWithID:oid];
    }];
}

In the Bandcamp API, each entity can uniquely be identified by its key. For example, a Band entity has the key band_id that uniquely identifies a band. Using the CoreData method newObjectIDForEntity:referenceObject we can create an NSManagedObjectID based on this id. Finally, we cache the values for an entity in the cache instance variable (which is an NSDictionary with NSManagedObjectID as keys and NSDictionary objects as values).

- (NSManagedObjectID*)objectIdForNewObjectOfEntity:(NSEntityDescription*)entityDescription
                                       cacheValues:(NSDictionary*)values {
    NSString* nativeKey = [self nativeKeyForEntityName:entityDescription.name];
    id referenceId = [values objectForKey:nativeKey];
    NSManagedObjectID *objectId = [self newObjectIDForEntity:entityDescription 
                                             referenceObject:referenceId];
    [cache setObject:values forKey:objectId];
    return objectId;
}

Note that when we created the NSManagedObject, the properties were not set. The managed objects only contain their unique ID. Core Data uses faulting when you access the properties, and we have to implement another method to support it: newValuesForObjectWithID:withContext:error:. This method will get called when we access the property of a managed object. Each NSManagedObject is backed by an NSIncrementalStoreNode that holds the values. In a database backend, the NSIncrementalStoreNode would correspond to a database record. In our API, it will be filled with an NSDictionary returned from the API. Note that in the previous method, we already cached this NSDictionary, so we don't need to do an API request:

- (NSIncrementalStoreNode*)newValuesForObjectWithID:(NSManagedObjectID*)objectID 
                                        withContext:(NSManagedObjectContext*)context
                                              error:(NSError**)error {
    NSDictionary* cachedValues = [cache objectForKey:objectID];
    NSIncrementalStoreNode* node = 
        [[NSIncrementalStoreNode alloc] initWithObjectID:objectID
                                              withValues:cachedValues 
                                                 version:1];
    return node;
}

There is one more important method to implement: newValueForRelationship:forObjectWithID:withContext:error:. As you can guess from its name, this is where we lookup the relationships. We look at the relationship source and target entity and name, and call the appropriate API methods to fetch the relationship objects. Again, we cache the API results and return an NSArray with NSManagedObjectID for each result.

We will implement two relationships in this method: discography, which relates a Band to its Albums, and tracks which relates an Album with its Tracks. In this method, we dispatch on the relationship name. In a more complicated situation where multiple relationships with the same name exist you can also inspect the relationship's entities.

- (id)newValueForRelationship:(NSRelationshipDescription*)relationship 
              forObjectWithID:(NSManagedObjectID*)objectID
                  withContext:(NSManagedObjectContext*)context
                        error:(NSError**)error {
    if([relationship.name isEqualToString:@"discography"]) {
        return [self fetchDiscographyForBandWithId:objectID 
                                       albumEntity:relationship.destinationEntity];
    } else if([relationship.name isEqualToString:@"tracks"]) {
        return [self fetchTracksForAlbumWithId:objectID 
                                   trackEntity:relationship.destinationEntity];
    }
    NSLog(@"unknown relatioship: %@", relationship);
    return nil;
}

Note that the return value of the method should be an NSArray with NSManagedObjectIDs, not NSManagedObjects! To give an example, for the discography we do an API call to fetch the raw data, and then create NSManagedObjectIDs for each album returned by the API:

- (NSArray*)fetchDiscographyForBandWithId:(NSManagedObjectID*)objectID
                              albumEntity:(NSEntityDescription*)entity {
    id bandId = [self referenceObjectForObjectID:objectID];            
    NSArray* discographyData = [BandCampAPI apiDiscographyForBandWithId:bandId];
    return [discographyData map:^(id album) {
        return [self objectIdForNewObjectOfEntity:entity cacheValues:album];
        
    }];
}

Usage

Now we can leverage the power of CoreData to access our API. Consumers of our model don't know whether they are accessing an API, an SQLite database or an XML file. It's all abstracted away into our NSIncrementalStore subclass.

To give an example, here's how you can find a band using Core Data:

NSEntityDescription *entityDescription = [NSEntityDescription
                                          entityForName:@"Band" inManagedObjectContext:moc];

NSPredicate* predicate = [NSPredicate predicateWithFormat:@"name == %@", @"Rue Royale"];
NSFetchRequest *fetchRequest = [[NSFetchRequest alloc] init];
fetchRequest.entity = entityDescription;
fetchRequest.predicate = predicate;

NSArray *results = [moc executeFetchRequest:fetchRequest error:nil];
Band* band = [results lastObject];

This code will make the appropriate API request and return an object of type Band. The Band class is generated from our CoreData model, and all of the properties are accessible as Objective-C properties.

When you then want to find the bands albums, it's as simple as doing:

for(Album* album in band.discography) {
  NSLog(@"album title: %@", album.title);
}

The nice thing about this code that there are no implementation details which bleed through. We could change the backend to an SQLite store and the code won't break. Additionally, the code is more type-safe: if we mistype a property name (for example, discograhpy instead of discography) we get a compiler error.

Next steps / Problems

Because this is all very new and not too well documented, I might have made a couple of mistakes. I would be really interested in hearing about it if you do spot something, and will update this post accordingly.

All the API calls and Core Data calls are done in a synchronous way. This is not a good idea in production code, as it will block the main thread. I'm experimenting myself with how to deal with that, and don't have a single answer yet. The comments on Drew's article are really helpful.

Finally, we implemented a readonly API. By implementing some more things in our NSIncrementalStore subclass we can add support for changing, saving, deleting and creating objects.

I can imagine it would be really interesting to write a subclass of NSIncrementalStore that can deal with CouchDB or Parse. Implementing a backend would then be as simple as defining your CoreData data model and initializing the class and you're up and running.

This is the first long technical blogpost I've written, and I would love to hear your thoughts on it. Especially parts that are not clear or written in a bad way. Please email me with your thoughts.

Running the example code

To run the example code, clone the project from github and open it in XCode. I just used a standard iOS template, and pressing 'Run' will not do much. The documentation is in the tests: open IncrementalStoreTestTests.m to see how to use the code. You can run the tests by pressing Cmd+U or Product > Test.

References

Project on Github
SealedAbstract
NSIncrementalStore Programming Guide
NSIncrementalStore Class Reference
Core Data talks
Core Data Programming Guide

@shpakovski
Copy link

shpakovski commented Feb 20, 2012

Chris, you've written a brilliant companion to the Drew's article. Thanks to your tutorial and real code, NSIncrementalStore has got even less miracle API :) Thank you.

@chriseidhof
Copy link
Author

chriseidhof commented Feb 20, 2012 via email

@thinktopdown
Copy link

Nice work. Thanks for the writeup. I'm working on my own NSIncrementalStore.

@chriseidhof
Copy link
Author

If the objects are already in memory, it doesn't access the api, I think. But there's no other caching mechanism other than what Core Data already does for you.

@chriseidhof
Copy link
Author

chriseidhof commented May 14, 2012 via email

@dreamwieber
Copy link

This is really nice, thanks for putting time into it.

It looks like you're fetching all of the attributes and then caching them when you initially hit the API. Doesn't that cancel out the benefit of faulting? I'm guessing in your particular case, the second API call would be more expensive than simply caching the data on the first hit, which may be why you're doing it that way. Perhaps I'm misunderstanding the cache flow.

@chriseidhof
Copy link
Author

Thanks!

You're understanding it exactly right =).

@mrjackdavis
Copy link

Hi love the post,
Have you figured out what NSStoreUUIDKey is used for? And how to generate one?

@chriseidhof
Copy link
Author

@mrjackdavis, other than what's in the NSPersistentStoreCoordinator docs, no.

@mbinna
Copy link

mbinna commented Aug 29, 2012

Great post. Thanks for sharing!

@dreamwieber
Copy link

In the example above you use an NSDictionary to cache object data - perhaps an NSCache would be better suited?

@dreamwieber
Copy link

After spending more time with this, just wanted to underline that this is the best example out there of how to implement an NSIncrementalStore – concise, and easily grokable. Really nice work with this.

@colemancda
Copy link

I read this a long time ago and that Store design stuck with me even though I couldn't remember who recommended it. I'm glad I happened to re-find this text. Its a great design for Core Data stores and I implement it in my projects.

@cvasilak
Copy link

cvasilak commented Aug 3, 2014

thank you @chriseidhof, that's really an interesting article and a welcome addition to the NSIncrementalStore land, which lacks proper documentation.

Looking forward for your next series with supporting full CRUD as well as your technique to overcome the main thread limitation.

keep up the good work!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment