Core » Changesets

Changesets are an advanced abstraction for making changes in your database. They work on top of commands, and provide additional data mapping functionality and have support for associating data.

Built-in changesets support all core command types, you can also define custom changeset classes and connect them to custom commands.

Working with changesets

You can get a changeset object via Relation#changeset interface. A changeset object wraps input data, and may optionally convert it into a representation that's compatible with your database schema.

Assuming you have a users relation available:

:create example

users.changeset(:create, name: "Jane").commit
=> {:id=>1, :name=>"Jane"}

:update example

users.by_pk(4).changeset(:update, name: "Jane Doe").commit
=> {:id=>4, :name=>"Jane Doe"}

Checking diffs

Update changesets check the difference between the original tuple and new data. If there's no diff, an update changeset will not execute its command

:delete example

users.by_pk(4).changeset(:delete).commit
=> {:id=>4, :name=>"Jane Doe"}

users.by_pk(4).changeset(:delete).commit
# => nil

Restricting relations for changesets

In the examples above, we used Relation#by_pk method, this is a built-in method which restricts a relation by its primary key; however, you can use any method that's available, including native adapter query methods.

Changeset Mapping

Changesets have an extendible data-pipe mechanism available via Changeset.map (for preconfigured mapping) and Changeset#map (for on-demand run-time mapping).

Changeset mappings support all transformation functions from transproc project, and in addition to that we have:

  • :add_timestamps–sets created_at and updated_at timestamps (don't forget to add those fields to the table in case of using rom-sql)
  • :touch–sets updated_at timestamp

Pre-configured mapping

If you want to process data before sending them to be persisted, you can define a custom Changeset class and specify your own mapping. Let's say we have a nested hash with address key but we store it as a flat structure with address attributes having address_* prefix:

class NewUserChangeset < ROM::Changeset::Create
  map do
    unwrap :address, prefix: true
  end
end

Then we can ask users relation for your changeset:

user_data = { name: 'Jane', address: { city: 'NYC', street: 'Street 1' } }

changeset = users.changeset(NewUserChangeset, user_data)

changeset.to_h
# { name: 'Jane', address_city: 'NYC', address_street: 'Street 1' }

changeset.commit

Custom mapping block

If you don't want to use built-in transformations, simply configure a mapping and pass tuple argument to the map block:

class NewUserChangeset < ROM::Changeset::Create
  map do |tuple|
    tuple.merge(created_on: Date.today)
  end
end

user_data = { name: 'Jane' }

changeset = users.changeset(NewUserChangeset, user_data)

changeset.to_h
# { name: 'Jane', created_on: <Date: 2017-01-21 ((2457775j,0s,0n),+0s,2299161j)> }

user_repo.create(changeset)
# => #<ROM::Struct[User] id=1 name="Jane" created_on=2017-01-21>

Custom mapping blocks are executed in the context of your changeset objects, which means you have access to changeset's state

On-demand mapping

There are situations where you would like to perform an additional mapping but adding a special changeset class would be an overkill. That's why it's possible to apply additional mappings at run-time without having to use a custom changeset class. To do this simply use Changeset#map method:

changeset = users
  .changeset(:create, name: 'Joe', email: 'joe@doe.org')
  .map(:add_timestamps)

changeset.commit(changeset)
# => #<ROM::Struct[User] id=1 name="Joe" email="joe@doe.org" created_at=2016-07-22 14:45:02 +0200 updated_at=2016-07-22 14:45:02 +0200>

Associating data

Changesets can be associated with each other using Changeset#associate method, which will automatically set foreign keys for you, based on schema associations.

Let's define :users relation that has many :tasks:

class Users < ROM::Relation[:sql]
  schema(infer: true) do
    associations do
      has_many :tasks
    end
  end
end

class Tasks < ROM::Relation[:sql]
  schema(infer: true) do
    associations do
      belongs_to :user
    end
  end
end

With associations established in the schema, we can easily associate data using changesets and commit them in a transaction:

task = tasks.transaction do
  user = users.changeset(:create, name: 'Jane')

  new_task = tasks.changeset(:create, title: 'Task One').associate(user)

  new_task.commit
end

task
# {:id=>1, :user_id=>1, :title=>"Task One"}

Association name

Notice that associate method can accept a rom struct and it will try to infer association name from it. If this fails because you have an aliased association then pass association name explicitly as the second argument, ie: associate(user, :author)

Learn more