Preload associations with cast_embed

Questions / Help
, ,
1

Hi :wave:

I had this piece of (working) code:

    items =
      products
      |> Enum.map(fn product ->
        params = %{product_id: product.id}

        Snapshot.new_item_changeset(
          %Snapshot.Item{
            product: product
          },
          params
        )
      end)

That later created a form with:

    form =
      Ecto.Changeset.put_embed(
        Ecto.Changeset.change(%Snapshot{}),
        :items,
        items
      )
      |> to_form

That I was rendering as:

    <.form for={@form} phx-change="validate" phx-submit="save">
      <.inputs_for :let={snapshot} field={@form[:items]}>
        <div><%= snapshot.data.product.description %></div>

However, since I didn’t want to repeat that Ecto.Changeset.put_embed( I wanted to move that logic to the schema as:

  def changeset(snapshot, params \\ %{}) do
    snapshot
    |> cast(params, [])
    |> cast_embed(:items, with: &new_item_changeset/2)
  end

  def new_item_changeset(item, params \\ %{}) do
    item
    |> cast(params, [:quantity, :product_id, :rate_id])
    |> validate_number(:quantity, greater_than_or_equal_to: 0)
  end

But if I move the original code to:

    items =
      products
      |> Enum.map(fn product ->
        %{product_id: product.id}
      end)

And then just:

    form =
      Snapshot.changeset(%Snapshot{}, %{items: items})
      |> to_form

That works, however the template doesn’t have a product loaded ending up in the following error:

key :description not found in: #Ecto.Association.NotLoaded<association :product is not loaded>

It would be easy to fix this with a Repo.preload but since it’s an embedded schema I am not sure where to do that.

2

So first you assigned a product and later you only assigned an ID and then you lost the product?

3

Thanks for your reply @i-n-g-m-a-r,

when you say that I “assigned” something you mean that I have a list of products and I just map it to get the product.id?

I was doing it like that because in the past I was using the product when creating the Snapshot.Item{} (you can see that in the first snippet), however, now that I want to use Snapshot.changeset to create the embedded version I don’t know how to set it or preload it.

4

I mean assigning a value to a variable, like so: product: product.
Before you were loading product data into a changeset.
Now you seem to disregard everything but the product ID.
So it makes sense you would end up with an ID only.
Changesets don’t do any (pre)loading.
Their function is to guarantee valid data.

5

Thanks again for your response @i-n-g-m-a-r.

What do you think would be the best approach to solve this issue then? Should I send a products map to the template and use the changeset product_id to read the product info from the map?

It seems there are ways to solve this, I am just not sure what would be the “most correct” (if that even exists :sweat_smile:) in Phoenix’s world. I thought that even when the changeset wouldn’t take the product into consideration and just use the product_id relation that was the proper way to do it.

Any insight would be welcome!

6

Actually when I started using ecto it was not easy for me to fully grasp the power of changesets.
The concept is very powerful, you just need some experience to “get it”.

I’m not the right guy to ask about “correctness”.
Personally I use a macro to build schema’s automatically inserting changeset functions and all kinds of pre and post data transformation hooks.
I don’t think anyone considers that to be “correct”.

I think this part is a good starting point:
Snapshot.changeset(%Snapshot{}, %{items: items})

I would focus first on fetching items.
Making sure every item is a certain struct.
Validated using changesets.

Then when you want to load a list of structs into a list of another type of structs, you can first (recursively) transform the list into a list of “params” (a nested map) and then load the params into a changeset for every item.
This should “just” work, no need for put_embed.

Also when you have a list of structs, you could load them into a schema embedding a list of structs (embeds_many).
Then you would have a (guaranteed) “type” instead of (just) a list.
In order to embed schema’s that are stored as records in a database you can create schemaless “meta” models.
I do this all the time.
Repo.all |> to_meta |> to_form
Submit…
params |> to_meta |> validate |> to_schema |> Repo.update_all

I would want my interface to look like this:
Snapshot.changeset(params)

Or like this:
Snapshot.changeset(stored_struct, params)

7

Thanks for the proposals.

I finally fixed my initial issue by also assigning to the socket a map of product id to product:

%{"1" => %xxx.Products.Product{

Then on the form I am showing it as follows:

<div><%= @products[snapshot.params["product_id"] |> to_string].description %></div>