MotionModel: Models, Relations, and Validation for RubyMotion

MotionModel is a DSL for cases where Core Data is too heavy to lift but you are still intending to work with your data, its types, and its relations. It also provides for data validation and actually quite a bit more.

File	Module	Description
ext.rb	N/A	Core Extensions that provide a few Rails-like niceties. Nothing new here, moving on...
model.rb	MotionModel::Model	You should read about it in "What Model Can Do". Model is the raison d'etre and the centerpiece of MotionModel.
validatable.rb	MotionModel::Validatable	Provides a basic validation framework for any arbitrary class. You can also create custom validations to suit your app's unique needs.
input_helpers	MotionModel::InputHelpers	Helps hook a collection up to a data form, populate the form, and retrieve the data afterwards. Note: MotionModel supports Formotion for input handling as well as these input helpers.
formotion.rb	MotionModel::Formotion	Provides an interface between MotionModel and Formotion
transaction.rb	MotionModel::Model::Transactions	Provides transaction support for model modifications

MotionModel is MIT licensed, which means you can pretty much do whatever you like with it. See the LICENSE file in this project.

• Getting Going
• What Model Can Do
• Model Data Types
• Validation Methods
• Model Instances and Unique IDs
• Using MotionModel
• Transactions and Undo/Cancel
• Notifications
• Core Extensions
• Formotion Support
• Problems/Comments
• Submissions/Patches

Changes for Existing Users to Be Aware Of

Please see the CHANGELOG for update on changes.

Version 0.3.8 to 0.4.0 is a minor version bump, not a patch version. Upgrading to 0.4.0 will break existing code. To update your code, simply insert the following line:

class ModelWithAdapter
  include MotionModel::Model
  include MotionModel::ArrayModelAdapter # <== Here!

  columns :name
end

This change lays the foundation for using other persistence adapters. If you don't want to update all your models, install the gem:

$ gem install motion_model -v 0.3.8

or if you are using bundler:

gem motion_model, "0.3.8"

Version 0.3.8 was the last that did not separate the model and persistence concerns.

Getting Going

If you are using Bundler, put this in your Gemfile:

gem motion_model

then do:

bundle install

If you are not using Bundler:

gem install motion_model

then put this in your Rakefile after requiring motion/project:

require 'motion_model'

If you want to use Bundler from master, put this in your Gemfile:

gem 'motion_model', :git => '[email protected]:sxross/MotionModel.git'

Note that in the above construct, Ruby 1.8.x hash keys are used. That's because Apple's System Ruby is 1.8.7 and won't recognize keen new 1.9.x hash syntax.

What MotionModel Can Do

You can define your models and their schemas in Ruby. For example:

class Task
  include MotionModel::Model
  include MotionModel::ArrayModelAdapter

  columns :name        => :string,
          :long_name   => :string,
          :due_date    => :date
end

class MyCoolController
  def some_method
    @task = Task.create :name => 'walk the dog',
                :long_name    => 'get plenty of exercise. pick up the poop',
                :due_date     => '2012-09-15'
   end
end

Side note: The original documentation on this used description for the column that is now long_name. It turns out Apple reserves description so MotionModel saves you the trouble of finding that particular bug by not allowing you to use it for a column name.

Models support default values, so if you specify your model like this, you get defaults:

class Task
  include MotionModel::Model
  include MotionModel::ArrayModelAdapter

  columns :name     => :string,
          :due_date => {:type => :date, :default => '2012-09-15'}
end

You can also include the Validatable module to get field validation. For example:

class Task
  include MotionModel::Model
  include MotionModel::ArrayModelAdapter
  include MotionModel::Validatable

  columns :name        => :string,
          :long_name   => :string,
          :due_date    => :date
  validates :name => :presence => true
end

class MyCoolController
  def some_method
    @task = Task.new :name  => 'walk the dog',
                 :long_name => 'get plenty of exercise. pick up the poop',
                 :due_date  => '2012-09-15'

    show_scary_warning unless @task.valid?
  end
end

Important Note: Type casting occurs at initialization and on assignment. That means If you have a field type int, it will be changed from a string to an integer when you initialize the object of your class type or when you assign to the integer field in your class.

a_task = Task.create(:name => 'joe-bob', :due_date => '2012-09-15')     # due_date is cast to NSDate

a_task.due_date = '2012-09-19'    # due_date is cast to NSDate

Model Data Types

Currently supported types are:

• :string
• :text
• :boolean, :bool
• :int, :integer
• :float, :double
• :date
• :array

You are really not encouraged to stuff big things in your models, which is why a blob type is not implemented. The smaller your data, the less overhead involved in saving/loading.

Special Columns

The two column names, created_at and updated_at will be adjusted automatically if they are declared. They need to be of type :date. The created_at column will be set only when the object is created (i.e., on first save). The updated_at column will change every time the object is saved.

Validation Methods

To use validations in your model, declare your model as follows:

class MyValidatableModel
  include MotionModel::Model
  include MotionModel::ArrayModelAdapter
  include MotionModel::Validatable

  # All other model-y stuff here
end

Here are some sample validations:

validate :field_name, :presence => true
validate :field_name, :length => 5..8 # specify a range
validate :field_name, :email
validate :field_name, :format

The framework is sufficiently flexible that you can add in custom validators like so:

module MotionModel
  module Validatable
    def validate_foo(field, value, setting)
      # do whatever you need to make sure that the value
      # denoted by *value* for the field corresponds to
      # whatever is passed in setting.
    end
  end
end

validate  :my_field, :foo => 42

In the above example, your new validate_foo method will get the arguments pretty much as you expect. The value of the last hash is passed intact via the settings argument.

You are responsible for adding an error message using:

add_message(field, "incorrect value foo #{the_foo} -- should be something else.")

You must return true from your validator if the value passes validation otherwise false.

An important note about save once you include Validatable, you have two flavors of save:

Method	Meaning
save(options)	Just saves the data if it is valid (passes validations) or if you have specified :validate => false
save!	Saves the data if it is valid, otherwise raises a MotionModel::Validatable::RecordInvalid exception

Model Instances and Unique IDs

It is assumed that models can be created from an external source (JSON from a Web application or NSCoder from the device) or simply be a stand-alone data store. To identify rows properly, the model tracks a special field called :id. If it's already present, it's left alone. If it's missing, then it is created for you. Each row id is guaranteed to be unique, so you can use this when communicating with a server or syncing your rowset to a UITableView.

Using MotionModel

• Your data in a model is accessed in a very ActiveRecord (or Railsey) way. This should make transitioning from Rails or any ORM that follows the ActiveRecord pattern pretty easy. Some of the finder syntactic sugar is similar to that of Sequel or DataMapper.

  • Finders are implemented using chaining. Here is an examples:

    @tasks = Task.where(:assigned_to).eq('bob').and(:location).contains('seattle')
    @tasks.all.each { |task| do_something_with(task) }

    You can use a block with find:

    @tasks = Task.find{|task| task.name =~ /dog/i && task.assigned_to == 'Bob'}

    Note that finders always return a proxy (FinderQuery). You must use first, last, or all to get useful results.

    @tasks = Task.where(:owner).eq('jim')   # => A FinderQuery.
    @tasks.all                              # => An array of matching results.
    @tasks.first                            # => The first result

    You can perform ordering using either a field name or block syntax. Here's an example:

    @tasks = Task.order(:name).all                                  # Get tasks ordered ascending by :name
    @tasks = Task.order{|one, two| two.details <=> one.details}.all # Get tasks ordered descending by :details

    You can implement some aggregate functions using map/reduce:

  @task.all.map{|task| task.number_of_items}.reduce(:+)                # implements sum
  @task.all.map{|task| task.number_of_items}.reduce(:+) / @task.count  #implements average

• Serialization is part of MotionModel. So, in your AppDelegate you might do something like this:

    @tasks = Task.deserialize_from_file('tasks.dat')

  and of course on the "save" side:

    Task.serialize_to_file('tasks.dat')
  end

  After the first serialize or deserialize, your model will remember the file name so you can call these methods without the filename argument.

  Implementation note: that the this serialization of any arbitrarily complex set of relations is automatically handled by NSCoder provided you conform to the coding protocol (which MotionModel does). When you declare your columns, MotionModel understands how to serialize your data so you need take no specific action.

  Warning: As of this release, persistence will serialize only one model at a time and not your entire data store.

  • Relations

  class Task
    include MotionModel::Model
    include MotionModel::ArrayModelAdapter
    columns     :name => :string
    has_many    :assignees
  end
  
  class Assignee
    include MotionModel::Model
    include MotionModel::ArrayModelAdapter
    columns     :assignee_name => :string
    belongs_to  :task
  end
  
  # Create a task, then create an assignee as a
  # related object on that task
  a_task = Task.create(:name => "Walk the Dog")
  a_task.assignees.create(:assignee_name => "Howard")
  
  # See? It works.
  a_task.assignees.assignee_name      # => "Howard"
  Task.first.assignees.assignee_name  # => "Howard"
  
  # Create another assignee but don't save
  # Add to assignees collection. Both objects
  # are saved.
  another_assignee = Assignee.new(:name => "Douglas")
  a_task.assignees << another_assignee  # adds to relation and saves both objects
  
  # The count of assignees accurately reflects current state
  a_task.assignees.count              # => 2
  
  # And backreference access through belongs_to works.
  Assignee.first.task.name            # => "Walk the Dog"

There are four ways to delete objects from your data store:

• object.delete # just deletes the object and ignores all relations
• object.destroy # deletes the object and honors any cascading declarations
• Class.delete_all # just deletes all objects of this class and ignores all relations
• Class.destroy_all # deletes all objects of this class and honors any cascading declarations

The key to how the destroy variants work in how the relation is declared. You can declare:

  class Task
    include MotionModel::Model
    include MotionModel::ArrayModelAdapter
    columns     :name => :string
    has_many    :assignees
  end

and assignees will not be considered when deleting Tasks. However, by modifying the has_many,

has_many    :assignees, :dependent => :destroy

When you destroy an object, all of the objects related to it, and only those related to that object, are also destroyed. So, if you call task.destroy and there are 5 assignees related to that task, they will also be destroyed. Any other assignees are left untouched.

You can also specify:

has_many    :assignees, :dependent => :delete

The difference here is that the cascade stops as the assignees are deleted so anything related to the assignees remains intact.

Note: This syntax is modeled on the Rails :dependent => :destroy options in ActiveRecord.

Hook Methods

During a save or delete operation, hook methods are called to allow you a chance to modify the object at that point. These hook methods are:

before_save(sender)
after_save(sender)
before_delete(sender)
after_delete(sender)

MotionModel makes no distinction between destroy and delete when calling hook methods, as it only calls them when the actual object is deleted. In a destroy operation, during the cascading delete, the delete hooks are called (again) at the point of object deletion.

Note that the method signatures may be different from previous implementations. No longer can you declare a hook method without the sender argument.

Finally, contrasting hook methods with notifications, the hook methods before_save and after_save are called before the save operation begins and after it completes. However, the notification (covered below) is only issued after the save operation. However... the notification understands whether the operation was a save or update. Rule of thumb: If you want to catch an operation before it begins, use the hook. If you just want to know about it when it happens, use the notification.

The delete hooks happen around the delete operation and, again, allow you the option to mess with the object before you allow the process to go forward (pretty much, the before_delete hook does this).

IMPORTANT: Returning false in a before hook stops the rest of the operation. So, for example, you could prevent the deletion of the last admin by writing something like this:

def before_delete(sender)
  return false if sender.find(:privilege_level).eq('admin').count < 2
end

Transactions and Undo/Cancel

MotionModel is not ActiveRecord. MotionModel is not a database-backed mapper. The bottom line is that when you change a field in a model, even if you don't save it, you are partying on the central object store. In part, this is because Ruby copies objects by reference, so when you do a find, you get a reference to the object in the central object store.

The upshot of this is that MotionModel can be wicked fast because it isn't moving much more than pointers around in memory when you do assignments. However, it can be surprising if you are used to a database-backed mapper.

You could easily build an app and never run across a problem with this, but in the case where you present a dialog with a cancel button, you will need a way to back out. Here's how:

# in your form presentation view...
include MotionModel::Model::Transactions

person.transaction do
  result = do_something_that_changes_person
  person.rollback unless result
end

def do_something_that_changes_person
  # stuff
  return it_worked
end

You can have nested transactions and each has its own context so you don't wind up rolling back to the wrong state. However, everything that you wrap in a transaction must be wrapped in the transaction block. That means you need to have some outer calling method that can wrap a series of delegated changes. Explained differently, you can't start a transaction, have a delegate method handle a cancel button click and roll back the transaction from inside the delegate method. When the block is exited, the transaction context is removed.

Notifications

Notifications are issued on object save, update, and delete. They work like this:

def viewDidAppear(animated)
  super
  # other stuff here to set up your view

  NSNotificationCenter.defaultCenter.addObserver(self, selector:'dataDidChange:',
                                                           name:'MotionModelDataDidChangeNotification',
                                                         object:nil)
end

def viewWillDisappear(animated)
  super
  NSNotificationCenter.defaultCenter.removeObserver self
end

# ... more stuff ...

def dataDidChange(notification)
  # code to update or refresh your view based on the object passed back
  # and the userInfo. userInfo keys are:
  #   action
  #     'add'
  #     'update'
  #     'delete'
end

In your dataDidChange notification handler, you can respond to the MotionModelDataDidChangeNotification notification any way you like, but in the instance of a tableView, you might want to use the id of the object passed back to locate the correct row in the table and act upon it instead of doing a wholesale reloadData.

Note that if you do a delete_all, no notifications are issued because there is no single object on which to report. You pretty much know what you need to do: Refresh your view.

This is implemented as a notification and not a delegate so you can dispatch something like a remote synch operation but still be confident you will be updating the UI only on the main thread. MotionModel does not currently send notification messages that differentiate by class, so if your UI presents Tasks and you get a notification that an Assignee has changed:

class Task
  include MotionModel::Model
  include MotionModel::ArrayModelAdapter
  has_many :assignees
  # etc
end

class Assignee
  include MotionModel::Model
  include MotionModel::ArrayModelAdapter
  belongs_to :task
  # etc
end

# ...

task = Task.create :name => 'Walk the dog'  # Triggers notification with a task object
task.assignees.create :name => 'Adam'       # Triggers notification with an assignee object

# ...

# We set up observers for `MotionModelDataDidChangeNotification` someplace and:
def dataDidChange(notification)
if notification.object is_a?(Task)
  # Update our UI
else
  # This notification is not for us because
  # We don't display anything other than tasks
end

The above example implies you are only presenting, say, a list of tasks in the current view. If, however, you are presenting a list of tasks along with their assignees and the assignees could change as a result of a background sync, then your code could and should recognize the change to assignee objects.

Core Extensions

• String#humanize
• String#titleize
• String#empty?
• String#singularize
• String#pluralize
• NilClass#empty?
• Array#empty?
• Hash#empty?
• Symbol#titleize

Also in the extensions is a Debug class to log stuff to the console. It uses NSLog so you will have a separate copy in your application log. This may be preferable to puts just because it's easier to spot in your code and it gives you the exact level and file/line number of the info/warning/error in your console output:

• Debug.info(message)
• Debug.warning(message)
• Debug.error(message)
• Debug.silence / Debug.resume to turn on and off logging
• Debug.colorize (true/false) for pretty console display

Finally, there is an inflector singleton class based around the one Rails has implemented. You don't need to dig around in this class too much, as its core functionality is exposed through two methods:

String#singularize String#pluralize

These work, with the caveats that 1) The inflector is English-language based; 2) Irregular nouns are not handled; 3) Singularizing a singular or pluralizing a plural makes for good cocktail-party stuff, but in code, it mangles things pretty badly.

You may want to get into customizing your inflections using:

• Inflector.inflections.singular(rule, replacement)
• Inflector.inflections.plural(rule, replacement)
• Inflector.inflections.irregular(rule, replacement)

These allow you to add to the list of rules the inflector uses when processing singularize and pluralize. For each singular rule, you will probably want to add a plural one. Note that order matters for rules, so if your inflection is getting chewed up in one of the baked-in inflections, you may have to use Inflector.inflections.reset to empty them all out and build your own.

Of particular note is Inflector.inflections.irregular. This is for words that defy regular rules such as 'man' => 'men' or 'person' => 'people'. Again, a reversing rule is required for both singularize and pluralize to work properly.

Formotion Support

MotionModel now has support for the cool Formotion gem. Note that the Formotion project on GitHub appears to be way ahead of the gem on Rubygems, so you might want to build it yourself if you want the latest gee-whiz features (like :picker_type, as I've shown in this example).

This feature is extremely experimental, but here's how it works:

class Event
  include MotionModel::Model
  include MotionModel::Formotion  # <== Formotion support

  columns :name => :string,
          :date => {:type => :date, :formotion => {:picker_type => :date_time}},
          :location => :string
end

This declares the class. The only difference is that you include MotionModel::Formotion. If you want to pass additional information on to Formotion, simply include it in the :formotion hash as shown above.

MotionModel has sensible defaults for each type supported, so any field of :date type will default to a date picker in the Formotion form. However, if you want it to be a string for some reason, just pass in:

:date => {:type => :date, :formotion => {:type => :string}}

To initialize a form from a model in your controller:

@form = Formotion::Form.new(@event.to_formotion('event details'))
@form_controller = MyFormController.alloc.initWithForm(@form)

The magic is in: MotionModel::Model#to_formotion(section_header).

The auto_date fields created_at and updated_at are not sent to Formotion by default. If you want them sent to Formotion, set the second argument to true. E.g.,

@form = Formotion::Form.new(@event.to_formotion('event details', true))

On the flip side you do something like this in your Formotion submit handler:

@event.from_formotion!(data)

This performs sets on each field. You'll, of course, want to check your validations before dismissing the form.

Problems/Comments

Please raise an issue on GitHub if you find something that doesn't work, some syntax that smells, etc.

If you want to stay on the bleeding edge, clone yourself a copy (or better yet, fork one).

Then be sure references to motion_model are commented out or removed from your Gemfile and/or Rakefile and put this in your Rakefile:

require "~/github/local/MotionModel/lib/motion_model.rb"

The ~/github/local is where I cloned it, but you can put it anyplace. Next, make sure you are following the project on GitHub so you know when there are changes.

Submissions/Patches

Obviously, the ideal patch request is really a pull request from your own fork, complete with passing specs.

Really, for a bug report, even a failing spec or some proposed code is fine. I really want to make this a decent tool for RubyMotion developers who need a straightforward data modeling and persistence framework.