Ecto changeset generated with phx.gen.context does not require id for update

Generally this is handled by fetching the structs from the database via Repo.get etc; you don’t normally cast :id in changesets.

Oh ok I see. I’m in a situation where if a query does not return anything, I am simply passed an empty struct so it’s a bit different

It looks You are checking the db before creating, or updating?

You might have race condition, and some other methods deals with this in a safer way.

I’m not really checking the database, I am just retrieving the struct to modify like @al2o3cr said. It’s just that my (simplified) workflow it’s like this: Repo.get(id) |> Map.put(something) |> changeset() |> Repo.update(), but instead of failing on Repo.get!() when there is no object I return an empty struct, and I fail on Repo.update() because the struct has no PK.

Instead of changing the changeset function I plan to call Repo.insert if there is no PK instead of Repo.update

This will fail because Repo.get/3 returns nil when there’s no result and you can’t Map.put/3 on nil

This is called an “upsert” and ecto has support for it as described here.

This will fail because Repo.get/3 returns nil when there’s no result and you can’t Map.put/3 on nil

I don’t exactly use Repo.get, it was only a simplification.

This is called an “upsert” and ecto has support for it as described here.

Many thanks, I’ll look into it!

If your pipeline can fail, You should use with…

with %{} = struct <- Repo.get(id),
  data <- Map.put(struct),
  changeset <- changeset(data),
  any <- Repo.update(changeset) do

  ...
else
  ...
end