Phoenix Ecto Guidelines

Associations

Always preload associations in queries when they'll be accessed in templates or downstream code:

# Good - preload when you need message.user.email
Repo.all(from m in Message, preload: [:user])

# Bad - will raise or N+1 query
messages = Repo.all(Message)
messages |> Enum.map(& &1.user.email)  # Ecto.Association.NotLoaded

Schema Fields

Ecto.Schema fields always use :string type, even for database :text columns:

# Correct
field :description, :string  # Even if DB column is TEXT

# Wrong - :text is not a valid Ecto type
field :description, :text

Changeset Field Access

Use Ecto.Changeset.get_field/2 to access changeset fields—never pattern match or access directly:

# Good
Ecto.Changeset.get_field(changeset, :email)

# Bad - doesn't account for changes vs data
changeset.changes.email
changeset.data.email

Validation Gotcha

Ecto.Changeset.validate_number/2 does not support the :allow_nil option. By default, Ecto validations only run when:

1. A change for the field exists
2. The change value is not nil

So :allow_nil is never needed.

Security: Mass Assignment

Fields set programmatically (like user_id) must never be listed in cast calls. Set them explicitly:

# Good - user_id set explicitly, not in cast
def create_changeset(post, attrs, user) do
  post
  |> cast(attrs, [:title, :body])
  |> put_change(:user_id, user.id)
end

# Bad - allows user to set user_id via attrs
def create_changeset(post, attrs) do
  post
  |> cast(attrs, [:title, :body, :user_id])
end

Seeds Files

Always import supporting modules in seeds.exs:

import Ecto.Query
alias MyApp.Repo
alias MyApp.Accounts.User