Best practices for defining Ecto associations

Offhand, here’s a few:

• filtering with the where option
• composing associations with the through option
• Repo.preload etc

Counter-question: what’s the advantage to doing this? It seems mostly harmless - foreign keys don’t change names very much, reducing the risk of duplication - but it’s still repeating things that don’t need to be repeated…

This one is quite handy as well: Ecto.assoc

I find associations useful when preloading data and doing joins:

defmodule Account do
  schema do
    has_many :users, User
  end
end

defmodule User do
  schema do
    belongs_to :account, Account
  end
end

and then:

user = Repo.get(User, 1) |> Repo.preload(:account)

or

from user in User,
  join: account in assoc(user, :account),
  where: account.foo = ...

• Answer 1: Yes. But I prefer using them.
• Answer 2: Yes. I think it’s better to maintain association at schema level. Generally, migrations are broken code snippets, they can’t give you a panoramic view.
• Answer 3: Yes. Read more in following content.

Suppose A has many B:

• When you just want to get B, use Ecto.assoc.
• When you want A and related B, use Ecto.preload or Ecto.Query.preload.
• When :join is required, try assoc which make use of the associations defined in schema. Then, generally, you don’t need to use :on.

If you don’t define associations at schema level, you can’t use all of these good stuff.