I have a processing pipeline, and an ecto struct that gets passed around between stages. In each stage’s entry and exit, status field in the ecto struct gets updated and notification is sent to subscribers:
This is very different from queries and mutations where my Resolver could return {:ok, my_ecto_struct} and some magic happens behind the scene where fields are resolved from Ecto into Absinthe Schema’s object before it’s returned to GraphQL client in the form of Json.
So how do I get an absinthe resolution for my_ecto_struct so I can publish/3 it?
One of the reasons I need some translation from ecto to absinthe is because not all fields match between ecto schema and absinthe schema, for example, absinthe has user that is resolved via UserResolver using ecto schema’s user_id, and signature that is resolved from absinthe context and is not present in ecto struct.
I was thinking of a custom resolver for subscriptions:
object :my_struct_subscriptions do
@desc "Subscribes to status changes"
field :my_struct_status_changed, non_null(:my_absinthe_object) do
arg(:id, non_null(:id))
config(&MyStructResolver.sub_config/2)
trigger([:status_changed], topic: fn _ -> "status_changed" end)
resolve(&MyStructResolver.resolve_status_change/3) # <-------
end
end
and then in the Resolver,
def resolve_status_change(value, args, _absinthe_resolution) do
value
|> Map.from_struct()
|> Map.delete(:__meta__)
|> Map.put(:user, User.get(value.user_id))
|> Map.put(:status, StatusResolver.get_status(value))
|> Map.put(:signature, SignatureUtils.generate(value))
else
nil
end
{:ok, value}
end
But seems like publish/3 is used mostly for notifying on a handful of fields and not a large object, and I would hate to have to change this bit of code every time schema changes and making sure that every field that requires special handling and can’t be passed in verbatim is resolved separately. I wish there was a utility/helper function to resolve ecto struct into Absinthe object just like the framework already does it for mutations and queries…
Seems like publish/3 was not meant for sending an entire object, but rather a few select fields. So I abandoned the idea of resolving an Ecto struct and just sending a few fields that change as the object passes through the processing pipeline - like status and a few datetime fields.
