Project

data_works

0.0
No commit activity in last 3 years
No release in over 3 years
DataWorks makes it easier to work with FactoryBot in the context of a complex data model.
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
 Dependencies

Development

~> 1.5.0
~> 1.9
>= 0
>= 0
~> 10.4
>= 0

Runtime

 Project Readme

DataWorks

DataWorks makes it easier to work with FactoryGirl in the context of a complex data model.

DataWorks has these benefits:

  • Reduces test data bloat.
  • Improves test clarity.
  • Improves test correctness.
  • Reduces refactoring cost when changing the data model.
  • Improves test performance.

Contents

  • The Problem That DataWorks Solves
  • How DataWorks Solves This Problem
    • Approach
    • Implementation
    • Benefits
  • Usage
    • Basic Example
    • Referencing Objects
    • Adding Multiple Objects
    • Avoiding Object Reuse
    • Overriding Attributes
    • Better Variable Names
    • Visualizing the Object Graph
    • Setup and Configuration
      • Configuration
      • Subclassing DataWorks::Base
  • Issues

The Problem That DataWorks Solves

Consider the following data model:

            ,----------,
            |          |
      .----(1) County (1)----.
      |     |          |     |
      |     '----------'     |
      |                      |
      |                      |
 ,---(*)----,           ,---(*)----,
 |          |           |          |
 |  Person (*)---------(1) School  |
 |          |           |          |
 '----------'           '----------'
  • A Person must live in a County.
  • A Person can optionally attend a School.
  • A School must belong to a County.

Let's say we need to test the method School#number_of_people. So we factory two people and a school and add the people to the school. Then we test that number_of_people returns two. The problem is that our object graph would look like this:

  ,----------,      ,----------,      ,----------,
  |          |      |          |      |          |
  |  County  |      |  County  |      |  County  |
  |          |      |          |      |          |
  '----.-----'      '----.-----'      '----.-----'
       |                 |                 |
       |                 |                 |
  ,----'-----,      ,----'-----,      ,----'-----,
  |          |      |          |      |          |
  |  Person  |      |  Person  |------|  School  |
  |          |      |          |      |          |
  '----.-----'      '----------'      '----.-----'
       |                                   |
       `-----------------------------------'

Each Person object and the School object would have their own individual county because the factories don't know that they should all be the same county. Now this probably won't affect the accuracy of a simple method like number_of_people, but in the case of more complex business logic, having an incorrect object graph could cause the test to fail even though the code under test is correct. So we need to make sure our factory objects reflect a valid state of the system.

The takeaway is that we need to manually factory a County object and pass it to the Person and School factories to ensure that only one County object exists.

Now consider a more complex data model:

              ,-----------,
              |           |
              |   Region  |
              |           |
              '----(1)----'
                    |
                    |
                    |
              ,----(*)----,
              |           |
        .----(1)  State  (1)----.
        |     |           |     |
        |     '-----------'     |
        |                       |
        |                       |
  ,----(*)---,            ,----(*)---,        ,------------------,
  |          |            |          |        |                  |
  |   Town   |      .----(1) County (1)------(1) SchoolDistrict  |
  |          |      |     |          |        |                  |
  '----(1)---'      |     '-----.----'        '-------(1)--------'
        |           |                                  |
        |    ,------'                                  |
        |    |                                         |
  ,----(*)--(*)                                   ,---(*)----,
  |          |                                    |          |
  |  Person (*)----------------------------------(1) School  |
  |          |                                    |          |
  '----------'                                    '----------'
  • A Person must live in a County.
  • A Person must live in a Town.
  • Towns can span Counties.
  • A Town must be in a State.
  • A County must be in a State.
  • A State must be in a Region.
  • A County has one and only one SchoolDistrict.
  • A SchoolDistrict has many Schools.
  • A Person can optionally attend a School.

Let's say we need to test the method School#number_of_people. So we factory a couple of Person models and a School model. But if that's all we did, we'll have a proliferation of separate parent models, such as three separate County objects, five separate State objects, and five separate Region objects.

In order to craft accurate test data, we're forced to manage this manually. We need to factory all the parent objects and pass them down to their child factories. So even though we are mostly caring about how a School model interacts with a Person model, we are forced to factory five different parent models.

How DataWorks Solves This Problem

Approach

DataWorks solves this problem by following one simple rule: by default, reuse objects that already exist.

When using FactoryGirl, a factory only concerns itself with its particular neighborhood of the data model. Since FactoryGirl factories do not know the big picture, their default is not to reuse existing objects but create new ones. This results in the proliferation of unwanted objects.

DataWorks knows the big picture of your data model and assumes that you are wanting to build a consistently connected network of models. So it will reuse objects that already exist unless explicitly told otherwise.

Implementation

DataWorks exists as a layer on top of FactoryGirl. DataWorks assumes that you have FactoryGirl factories for each model and that they create valid objects. DataWorks always uses a create strategy with FactoryGirl.

Benefits

Here's how DataWorks gives us the following benefits:

  • Reduces test data bloat. The amount of test data code that you have to write is significantly reduced.
  • Improves test clarity. DataWorks allows you to factory only the objects that are relevant to what's being tested. Having to factory up a bunch of indirectly-related objects distracts from what the test is trying to communicate.
  • Improves test correctness. Manually preventing the duplication of parent factories is more error-prone than having DataWorks do it automatically.
  • Reduces refactoring cost when changing the data model. If you refactor the data model high up in the parent chain, you aren't forced to touch practically every test that has factoried test data, since DataWorks allows for implied, not explicit, parents.
  • Improves test performance. Because intermediate models aren't being created, there is a small performance benefit.

Usage

Basic Example

describe "School#number_of_people" do
  before do
    data = TheDataWorks.new
    @school = data.add_school
    data.add_person(school: @school)
    data.add_person(school: @school)
  end

  it "returns the correct number of people attending that school" do
    expect( @school.number_of_people ).to eq( 2 )
  end
end
  • Always start by creating a new DataWorks object. This starts DataWorks off with a blank slate of objects.
  • Use add_[model_name] to create a new object.
  • The necessary parent objects (SchoolDistrict, Town, County, etc.) are implied.

Referencing Objects

  • Any object that's created through DataWorks can be referenced by calling [model_name][number] to DataWorks.
  • Objects are numbered in the order they are created.
  • the_[model_name] is a synonym for [model_name]1 that can aid test readability.

Here's a test that puts two people in one school and three in another:

describe "School#number_of_people" do
  before do
    data = TheDataWorks.new
    data.add_school
    data.add_school
    data.add_person(school: data.school1)
    data.add_person(school: data.school1)
    data.add_person(school: data.school2)
    data.add_person(school: data.school2)
    data.add_person(school: data.school2)
    @school = data.school1
  end

  it "returns the correct number of people attending that school" do
    expect( @school.number_of_people ).to eq( 2 )
  end
end

If we wanted to check the number of people in the school district, we can access the SchoolDistrict created behind the scenes like this:

data.school_district1.number_of_people

Or we could use the synonym:

data.the_school_district.number_of_people

Adding Multiple Objects

You can factory many objects at once by passing an integer count as the first argument to add_

DataWorks treats any call to add_ with an integer as the first argument as an attempt to create multiple objects. For clarity and readability DataWorks allows you to use the plural name of the models being created. This is not required

Given

data = TheDataWorks.new
data.add_schools(10)

and

data.add_school(10)

Will both create 10 school records.

Plural model names are converted to singular using the ActiveSupport inflections module. If you need to specify any additional inflection rules you can add them in your test configuration. For reference here is the example given in the Rails documentation.

ActiveSupport::Inflector.inflections(:en) do |inflect|
  inflect.plural /^(ox)$/i, '\1\2en'
  inflect.singular /^(ox)en/i, '\1'

  inflect.irregular 'octopus', 'octopi'

  inflect.uncountable 'equipment'
end

Avoiding Object Reuse

Let's say that we want to factory two Schools and want to avoid having them automatically share a SchoolDistrict. We can do this by simply manually creating separate SchoolDistrict objects and explicitly setting them when we create the Schools:

data = TheDataWorks.new
data.add_school_districts(2)
data.add_school(school_district: data.school_district1)
data.add_school(school_district: data.school_district2)

The case above works find for records which only need to be different in one association but what if we want to generate a chain of associations or need to create many record which all need to share certain associations?

Given the data model described above say that we want to be able to create two schools but have them each belong to a different state with the two states belonging to the same region. DataWorks can restrict the possible 'parentage' of models using the #set_restriction method.

massachusetts = data.add_state(name: 'Massachusetts')
vermont       = data.add_state(name: 'Vermont')

# We now have two states so lets create those schools...

mass_school = data.add_school # this will cause a SchoolDistrict and County to be created

data.set_restriction(for_model: :state, to: vermont)

vermont_school = data.add_school # this will also cause a SchoolDistrict and County to be created

The return value of #set_restriction is the parent object passed as to:.

Had we not used the set_restriction method above we would have needed to set the vermont model hierarchy up manually like so.

vermont_county   = data.add_county
vermont_district = data.add_school_district(county: vermont_county)
vermont_school   = data.add_school(school_district: vermont_district)

The #set_restriction method also accepts a block. When the block form is used the restriction applies only to those records added within the block.

The return value of the block form of #set_restriction is the last statement evaluated in the block.

DataWorks also provides a #set_current_default method which allows you to set the current default record to use when automatically associating models.

Say we want to create multiple towns which belong to two states. We can do so like this

massachusetts = data.add_state(name: 'Massachusetts')
vermont       = data.add_state(name: 'Vermont')

data.add_towns(3) # this will create 3 towns all belonging to massachusetts

# massachusetts was the default above since it was created first. Now we
# want vermont to be the default state for 3 more towns

data.set_current_default(to: vermont, for_model: :state)

data.add_towns(3) # this will create 3 towns all belonging to vermont

Both set_current_default and set_restriction have corresponding clear_X_for methods for removing the default/restricting record.

Overriding Attributes

DataWorks simply delegates to the factory corresponding to the model, so it will use the default attributes specified there. If you pass attributes to DataWorks, it will pass them on to the factory:

data = TheDataWorks.new
data.add_school_district(name: 'Washington', rural: true)

Better Variable Names

Let's say you don't like calling your test data by number and want more meaningful names. Just use plain old Ruby variables to make your tests clearer:

data = TheDataWorks.new
franklin = data.add_school_district(name: 'Franklin')
greenville = data.add_school_district(name: 'Greenville')
data.add_school(school_district: franklin)
data.add_school(school_district: greenville)

Visualizing the Object Graph

If you want to debug your test data and explicitly see the object graph that DataWorks created for you, you can use the visualize method which will render an object graph for you and automatically open it:

data = TheDataWorks.new
data.add_school_district
data.add_school
data.visualize

If an object labeled "unmanaged" appears in the object graph it means that an object was factoried outside of DataWorks. This can happen if your base factories create children (not parent) associations by default. Note that the object graph cannot show the entire set of all objects that were created, just those created through DataWorks and some of the associated objects that may have been created outside of DataWorks and attached to a DataWorks-managed factory object.

Setup and Configuration

DataWorks expects the following:

  • You must have factories for all of your models, and these factories must produce valid (persistable) models.
  • None of these basic factories should create any child objects by default, as this does not give DataWorks a chance to manage them.
  • You must configure DataWorks in your spec_helper.rb by calling DataWorks.configure. This is where you teach DataWorks about your data model relationships.
  • You must have a class that subclasses DataWorks::Base. Use this in your tests; do not use DataWorks::Base directly.
  • You must tell DataWorks if you have any associations with special names (the belongs_to or has_many do not match the model names). You can do this by passing in a hash instead of a symbol for the parent model name: { :association_name => :model_name }

Configuration

Necessary Parents

In your spec_helper.rb file, put the following:

DataWorks.configure do |config|
  config.necessary_parents = {
    classroom:             [:school, :grade],
    district:              [ ],
    event:                 [:schedule, :school],
    failure:               [:service_schedule_set],
    grade:                 [ ],
    iep_service:           [:service, :student],
    schedule:              [:service_schedule_set],
    scheduled_service:     [{:schedulable => :event}, :student, :iep_service],
    school:                [:district],
    service:               [:district, :service_type],
    service_schedule_set:  [:district, :service],
    service_type:          [ ],
    student:               [:school, :classroom],
  }
end

config.necessary_parents is where you tell DataWorks which other factories must be created when you create a particular factory. Because FactoryGirl causes a proliferation of extra objects, DataWorks does not allow FactoryGirl to create necessary associated objects. Instead, DataWorks will create the necessary parent objects and pass them down into the factory, ensuring that too many parent objects do not get created.

This must list all of the models.

You must keep this list up to date when you change your data model. To aid with this, whenever a belongs_to relationship changes in your code, DataWorks raises an error that reminds you that you need to update this configuration to match the new data model.

Polymorphic Parents

The :scheduled_services portion of the configuration above demonstrates DataWorks' support for polymorphic associations. Rather than a symbol for the necessary parent a polymorphic class must have a hash, of a single key/value pair, which names the polymorphic relationship and indicates which possible parent should be created by default. Note that other possible parent objects can be passed as arguments when the child object is being created as with any other classes.

Autocreated Children

DataWorks can handle models that autocreate a child object:

DataWorks.configure do |config|
  config.necessary_parents = {
    ...
  }

  config.autocreated_children = {
    city: [:city_location]
  }
end

For example, let's say that every time a City model is created, it must have a CityLocation model, so there is logic in the City model that autocreates a CityLocation object. Note that when a model does this, DataWorks has no way of knowing about this autocreated child object and so it is not managed by DataWorks and could end up being a zombie object that could break your tests. So by explicitly listing out the names of the models that get autocreated, this gives DataWorks a chance to remove the zombie object and hook up a DataWorks-aware object in its place.

Note that this is only for the situation where a model autocreates a single child model, not multiple.

Blessing

Once you're satisfied the configuration is accurate, run

$ rake data_works:bless

And DataWorks will no longer complain (until you the next time you change a belongs_to relationship).

Subclassing DataWorks::Base

Subclass DataWorks::Base and call it whatever you want, such as TheDataWorks. spec/support is a good place to put this.

Inside of this class you can put some convenience functions. For example, if you find yourself frequently factorying two types of objects together, you can add a convenience function here.

Issues

  • DataWorks does not yet work with namespaced models.
  • DataWorks does not support factory traits
  • DataWorks does not allow associations where the foreign key name is not the same as the class name
  • The visualization component uses respond_to in verifying associations, which is not the most robust method if has_many is missing or with :through associations.

Credit

DataWorks was developed by Wyatt Greene and Luke Inglis at The District Management Group.