Skip to content


Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
branch: master
Fetching contributors…

Cannot retrieve contributors at this time

302 lines (229 sloc) 13.722 kb

Active Resource Kit

What can you do with Active Resource Kit? Active Resource Kit is yet-another RESTful framework. There are others. But Active Resource Kit has a number of distinct features.

  1. It mirrors the Rails Active Resource gem closely. The interface and implementation remain as faithful as an Objective-C implementation can reasonably be to the Ruby-based originals.
  2. It offers a very high-level interface to RESTful resources using Core Data. You can access remote resources just as if they were in a local Core Data store. The implementation uses the new Core Data NSIncrementalStore API to merge the two dissimilar interfaces.
  3. It only has Foundation and Core Data as underlying dependencies. Although it has two immediate dependencies, Active Model Kit and Active Support Kit which fall under the same umbrella framework, Apple's Foundation and Core Data kits form the only external dependencies. The implementation employs only the Foundation framework for network access.
  4. The framework supports various concurrency models when interacting with remote resources: these are the same models offered by Apple's Foundation NSURLConnection class, i.e. delegated URL connections, synchronous loading or queued loading. You can configure according to your requirements on a resource-by-resource basis.
  5. There are no swizzles or other non-standard Objective-C tricks. The framework makes extensive use of C blocks for handling completions for both asynchronous and synchronous interfaces; this simply follows the pattern set by Apple in their URL connection API.

Resources Using Core Data

Setting Up an Active Resource-Based Core Data Stack

This is easy to do. Just follow the usual Core Data-prescribed procedure: load the model, load the coordinator with the model, add the store to the coordinator, and finally attach the coordinator to the context. See example below.

NSBundle *bundle = [NSBundle bundleForClass:[self class]];
NSURL *modelURL = [bundle URLForResource:@"MyCoreDataModel" withExtension:@"momd"];
NSManagedObjectModel *model = [[NSManagedObjectModel alloc] initWithContentsOfURL:modelURL];
NSPersistentStoreCoordinator *coordinator = [[NSPersistentStoreCoordinator alloc] initWithManagedObjectModel:model];

NSError *__autoreleasing error = nil;
NSPersistentStore *store = [coordinator addPersistentStoreWithType:[ARIncrementalStore storeType]
                                                               URL:[NSURL URLWithString:@"http://localhost:3000"]
// <-- error handling goes here

NSManagedObjectContext *context = [[NSManagedObjectContext alloc] initWithConcurrencyType:NSMainQueueConcurrencyType];
[context setPersistentStoreCoordinator:coordinator];
[self setContext:context];

Note that this excerpt uses Automatic Reference Counting, hence the __autoreleasing specifier for the error pointer. Notice the blatant lack of manual auto-releasing.

Accessing Existing Resources

You can then access resources using only Core Data.

NSError *__autoreleasing error = nil;
NSFetchRequest *request = [NSFetchRequest fetchRequestWithEntityName:@"Person"];
NSArray *people = [[self context] executeFetchRequest:request error:&error];
for (NSManagedObject *person in people)
    NSString *name = [person valueForKey:@"name"];
    NSLog(@"person named %@", name);

You ask Core Data for the Person entities. The answer is a collection of managed object representing each Person. You access attributes on the objects using standard Cocoa key-value coding. However, underneath the hood, the Active Resource incremental store has enacted a RESTful GET request at http://localhost:3000/people.json, decoding and caching the active resources at the client side.

At the other side of the connection (assuming your server runs on Rails; it does not need to be Rails but can be any conforming RESTful interface) you will see a GET request in the server log, as follows. Some details elided.

Started GET "/people.json" for at …
Processing by PeopleController#index as JSON
  Person Load (0.2ms)  SELECT "people".* FROM "people"
Completed 200 OK in 5ms (Views: 3.8ms | ActiveRecord: 0.2ms)

Inserting Resources

This just becomes fuss-free:

NSError *__autoreleasing error = nil;
NSManagedObject *person = [NSEntityDescription insertNewObjectForEntityForName:@"Person" inManagedObjectContext:[self context]];
[person setValue:@"Roy Ratcliffe" forKey:@"name"];
BOOL yes = [[self context] save:&error];

And on the server side becomes a familiar POST request:

Started POST "/people.json" for at …
Processing by PeopleController#create as JSON
  Parameters: {"person"=>{"name"=>"Roy Ratcliffe"}}
   (0.1ms)  begin transaction
  SQL (0.4ms)  INSERT INTO "people" ("created_at", "name", "updated_at") VALUES (?, ?, ?)  [["created_at", …], ["name", "Roy Ratcliffe"], ["updated_at", …]]
   (2.3ms)  commit transaction
Completed 201 Created in 5ms (Views: 0.8ms | ActiveRecord: 2.8ms)

Deleting Resources

Again, just very simply:

NSError *__autoreleasing error = nil;
[[self context] deleteObject:person];
BOOL yes = [[self context] save:&error];

And the server responds:

Started DELETE "/people/16.json" for at …
Processing by PeopleController#destroy as JSON
  Parameters: {"id"=>"16"}
  Person Load (0.1ms)  SELECT "people".* FROM "people" WHERE "people"."id" = ? LIMIT 1  [["id", "16"]]
  SQL (0.3ms)  DELETE FROM "people" WHERE "people"."id" = ?  [["id", 16]]
Completed 200 OK in 15ms (ActiveRecord: 12.7ms)

Building Associations

You can conveniently form associations between objects and their remote resources using only the Core Data interface.

The following demonstrates what happens when you instantiate two entities and wire them up entirely at the client side first. Lets say you have a post and comment model; posts have many comments, a one post to many comments association. The following initially creates a post with one comment.

NSManagedObject *post = [NSEntityDescription insertNewObjectForEntityForName:@"Post" inManagedObjectContext:[self context]];
NSManagedObject *comment = [NSEntityDescription insertNewObjectForEntityForName:@"Comment" inManagedObjectContext:[self context]];

// Set up attributes for the post and the comment.
[post setValue:@"De finibus bonorum et malorum" forKey:@"title"];
[post setValue:@"Non eram nescius…" forKey:@"body"];
[comment setValue:@"Quae cum dixisset…" forKey:@"text"];

// Form the one-post-to-many-comments association.
[comment setValue:post forKey:@"post"];

// Send it all to the server.
NSError *__autoreleasing error = nil;
[[self context] save:&error];

It constructs a new post, a new comment and their relationship within the client at first. Then it saves the context in order to transfer the objects and their relationship to the remote server.

Thereafter, you can throw away the comment and refetch it by dereferencing the post's "comments" relationship. The following extract pulls out each text field from the comments based on a given post.

NSMutableArray *comments = [NSMutableArray array];
for (NSManagedObject *comment in [post valueForKey:@"comments"])
    [comments addObject:[comment valueForKey:@"text"]];
[[comments objectAtIndex:0] rangeOfString:@"Quae cum dixisset"].location != NSNotFound;

Resources Using Rails-Style Access

You can also access resources using a Rails-style Active Resource interface. The interface mimics Rails so far as possible, albeit using Objective-C and Cocoa frameworks as the baseline rather than Ruby and the Ruby standard library.

Concurrency and Incremental Stores

By necessity, incremental stores must access remote resources synchronously. The Core Data interface, NSIncrementalStore, implements a pull-style of access to the underlying abstract store, as opposed to push-style. The interface does not let you tell Core Data when remote information arrives; instead, Core Data makes a request and expects its result by reply. There is no scope for running the request asynchronously and telling Core Data when results become available.

Instead, you need to incorporate concurrency within your application at a higher level, e.g. the controller layer. Core Data facilitates such an approach by allowing child contexts to operate in concurrent threads of control, later merging changes to the parent context.

Design Notes


Memory is a major issue on devices running iOS. Such phones and tablets only have either 128, 256 or 512MB of RAM.


Cocoa's Foundation framework supports three distinct URL connections. Active Resource Kit models them using four object classes: one abstract with three corresponding concrete implementation classes.

Class Diagram: Connections

Lazy Getting

Rails makes extensive use of the lazy getter paradigm: attributes remain undefined until you access them for the first time. This is a useful model. It postpones instantiation of dependencies, such as format and connection, until actually needed. Hence clients can easily override before use to customise behaviour.

The cross-platform requirement clashes with lazily-getting for this project however.


Run tests by selecting either the ActiveResourceKitFramework or ActiveResourceKitLibrary targets. Selecting Run » Tests (Cmd+U shortcut) launches the tests.

In the background, the schemes run a Thin server. This assumes that you do not already have a Thin server up and running for the Rails test application. If you do have a server already running in the background, the tests terminate the current background instance in order to set up and prime the fixtures, before launching a new server instance. This happens quite quickly; Thin is a fast-loading web server.

The test launcher script assumes that you have RVM installed as well as the expected version of Ruby; see active-resource-kit-tests/.rvmrc. After installing RVM, install the required Ruby and the required Gems, as follows.

cd active-resource-kit-tests
rvm install ruby-1.9.2-p320
bundle install

Viewing the Test Server Log

You can view the test servers log using the command:

➜  active-resource-kit-tests git:(master) tail -f log/thin.log

Rails Base URL

The kit's test target launches a Rails application in the background. The Xcode schemes run a Thin server using the URL scheme, address and port passed by the RAILS_BASE_URL environment variable. You can find this variable, along with the default RAILS_ENV setting, in the project build settings under User-Defined as follows.

RAILS_BASE_URL = https://localhost:3000
RAILS_ENV = development

Since the default URL scheme specifies https, the test launches Thin with the --ssl option. This enables SSL over HTTP encrypted communication. Changing the build setting to use http rather than https disables the --ssl option. This proves useful when debugging the server-side in tandom, e.g. when you cannot conveniently debug with SSL enabled. Just switch the build setting to RAILS_BASE_URL = http://localhost:3000 (insecure) and launch the Rails app as normal.

Resource Associations

When resources load at the client side, what binds their associations? How can the client resolve foreign keys? To do so, the client needs to identify the active resources.

Using the principle of convention over configuration, ARBase registers its instances by default, and each ARBase retains its AResources. Hence it can resolve the association whenever a new resource appears with an ID matching some existing foreign key. Similarly when a new record comprising an unresolved foreign key loads, ARBase can resolve it against an existing resource.

Incremental Stores

Apple provide a useful Core Data component called NSIncrementalStore, designed for interacting with external stores which do not bring all data into memory the way atomic stores do. Data loads and stores incrementally. Incremental stores let you plug RESTful resources into a standard Core Data stack.

One important drawback exists however. Incremental stores, at the current version, do not accommodate asynchronous network communication. Core Data sends execute-request messages to the store, expecting the store to respond with results immediately on return. You can respond with faults but doing so requires you to know object identities for faulting. Problem is, you cannot execute a fetch request with faulting object identities without some server interaction. Unless the code blocks for synchronous communication, you cannot return with anything else except an error.

Rails Kit Sub-Framework

Active Resource Kit is designed as a sub-framework on Mac OS X, though not so in iOS. Framework requirements mandate the following build setting within the Rails Kit sub-frameworks.

DYLIB_INSTALL_NAME_BASE = "@executable_path/../Frameworks/RailsKit.framework/Versions/Current/Frameworks"

It tells an application to look for the sub-framework at the given location relative to the application binary within the application bundle. This makes an assumption: that you locate the RailsKit.framework within the bundle. The sub-frameworks exist as sub-sub-frameworks within the application bundle. Hence the install-name base path specifies the RailsKit.framework's Frameworks sub-folder.

Jump to Line
Something went wrong with that request. Please try again.