Allow explicitly passing a dynamic repo to Ecto.Adapters.SQL.Sandbox.checkout/1

[Gigitsu]

As discussed in this Elixir Forum post, I encountered issues when testing with a dynamic repo.

This PR allows explicitly passing a dynamic repo to both Ecto.Adapters.SQL.Sandbox.checkout/1 and Ecto.Adapters.SQL.Sandbox.start_owner!/1.

With this change, I can change my DataCase setup as follows:

def setup_sandbox(tags) do
  pid = SQL.Sandbox.start_owner!(Hello.Repo.get_dynamic_repo(), shared: not tags[:async])
  on_exit(fn -> SQL.Sandbox.stop_owner(pid) end)
end

This enables proper testing of dynamic repos without manual connection management.

[Gigitsu]

Integration tests are failing because a test expects an UndefinedFunctionError with the message ~r"function UnknownRepo.get_dynamic_repo/0 is undefined".

Could there be a better way to check if an atom represents a module?

I could try using Kernel.function_exported?/3 but this won't fix the failing test.

[josevalim]

I assume your code is a bit more complex than this, right?

def setup_sandbox(tags) do
  pid = SQL.Sandbox.start_owner!(Hello.Repo.get_dynamic_repo(), shared: not tags[:async])
  on_exit(fn -> SQL.Sandbox.stop_owner(pid) end)
end

How does your code look like?

[Gigitsu]

I assume your code is a bit more complex than this, right?

Yes, it's not much more complex, but it involves some code that might be internal to Ecto.

How does your code look like?

I ended up with this minimal setup:

  def setup_sandbox(tags) do
    pid = SQL.Sandbox.start_owner!(repo_pid(), shared: not tags[:async])
    on_exit(fn -> SQL.Sandbox.stop_owner(pid) end)
  end

  defp repo_pid do
    GenServer.whereis(Hello.Repo.get_dynamic_repo())
  end

We could also introduce a new get_dynamic_repo_pid() to Ecto.Repo instead of this, but I don’t really like the idea of introducing a function solely for this case.

[josevalim]

The code that you pasted above is completely fine to me and only rely on public APIs. That’s definitely what I would recommend in this case.

[Gigitsu]

Ok, that seems fair to me.
Honestly, I thought this would be an easier PR... I didn’t expect it to break some tests.
Sorry about that, and thank you!

[josevalim]

No need to be sorry, thank you for giving it a shot!