Hey @zachdaniel , here is a quick update regarding this issue:
I created a small resource to make troubleshot easier:
defmodule Pacman.Markets.Entity2 do
@moduledoc false
use Ash.Resource,
data_layer: AshPostgres.DataLayer
attributes do
uuid_primary_key :id
end
postgres do
table "entities"
repo Pacman.Repo
end
identities do
identity :unique_id, [:id]
end
end
When I run the take_snapshot code, I get:
iex(87)> snapshot = Pacman.Markets |> AshPostgres.MigrationGenerator.take_snapshots(Pacman.Repo, [Pacman.Markets.Entity2]) |> Enum.at(0)
%{
attributes: [
%{
default: "fragment(\"uuid_generate_v4()\")",
size: nil,
type: :uuid,
source: :id,
references: nil,
primary_key?: true,
allow_nil?: false,
generated?: false
}
],
table: "entities",
hash: "8123997EA2DF402FDFB3E8145AF8666DCD0894EEF02C235DE06B42D47AF3457A",
repo: Pacman.Repo,
schema: nil,
identities: [
%{
name: :unique_id,
keys: [:id],
index_name: "entities_unique_id_index",
all_tenants?: false,
base_filter: nil
}
],
base_filter: nil,
multitenancy: %{global: nil, attribute: nil, strategy: nil},
check_constraints: [],
custom_indexes: [],
custom_statements: [],
has_create_action: false
}
If I just run the rest of the code, I will get the error shown above, to make it work I needed to do the following changes:
- Clean the
attributes field array since I don’t need it and it will break;
- Replace the
repo field value from atom to string;
- In the
identities field, I changed the name field value from atom to string;
- In the
identities field, I changed the keys field values from atom to string;
After these changes, my snapshot changed to the following one:
%{
attributes: [],
table: "entities",
hash: "8123997EA2DF402FDFB3E8145AF8666DCD0894EEF02C235DE06B42D47AF3457A",
repo: "Pacman.Repo", # CHANGED
schema: nil,
identities: [
%{
name: "unique_id", # CHANGED
keys: ["id"], # CHANGED
index_name: "entities_unique_id_index",
all_tenants?: false,
base_filter: nil
}
],
base_filter: nil,
multitenancy: %{global: nil, attribute: nil, strategy: nil},
check_constraints: [],
custom_indexes: [],
custom_statements: [],
has_create_action: false
}
After these changes, the rest of the code works great:
iex(90)> snapshot_without_indices = %{snapshot | custom_indexes: [], identities: []}
%{
attributes: [],
table: "entities",
hash: "8123997EA2DF402FDFB3E8145AF8666DCD0894EEF02C235DE06B42D47AF3457A",
repo: "Pacman.Repo",
schema: nil,
identities: [],
base_filter: nil,
multitenancy: %{global: nil, attribute: nil, strategy: nil},
check_constraints: [],
custom_indexes: [],
custom_statements: [],
has_create_action: false
}
iex(91)>
nil
iex(92)> operations = AshPostgres.MigrationGenerator.get_operations_from_snapshots([snapshot], [snapshot_without_indices])
[
%AshPostgres.MigrationGenerator.Operation.RemoveUniqueIndex{
identity: %{
name: :unique_id,
keys: [:id],
index_name: "entities_unique_id_index",
all_tenants?: false,
base_filter: nil
},
schema: nil,
table: "entities",
multitenancy: %{global: nil, attribute: nil, strategy: nil},
old_multitenancy: %{global: nil, attribute: nil, strategy: nil},
no_phase: true
}
]
iex(93)>
nil
iex(94)> {down, up} = AshPostgres.MigrationGenerator.build_up_and_down(operations)
{"drop_if_exists unique_index(:entities, [:id], name: \"entities_unique_id_index\")\n",
"create unique_index(:entities, [:id], name: \"entities_unique_id_index\")\n"}
So, the problem is that `AshPostgres.MigrationGenerator expects that some values are string and try to convert it to an atom, but in this case they are already an atom.
An easy solution for that is just to add the following function to AshPostgres.MigrationGenerator:
defp maybe_to_atom(value) when is_atom(value), do: value
defp maybe_to_atom(value), do: String.to_atom(value)
And replace all String.to_atom calls in that module to use that function.
After doing that the original snapshot starts working great.
Would you accept an PR with that change?
I have no idea if this is really what you’ll end up wanting, but we do have some tooling around this kind of thing. What you can do is this:
