I’ve started to test my application and immediately run into the mock debate. Just done reading:

The above suggests using explicit contracts and config variables to manage environment-specific dependencies initially, and then goes on to suggest the simpler method of injecting dependency directly.

However, my current method is a little different. I am using explicit contracts and delegation such as:

config :myapp, request_client: MyApp.Tesla.Request

defmodule MyApp.Request do
  @client Application.get_env(:myapp, :request_client)

  defdelegate make, to: @client

  @callback make(etc) :: etc
end

defmodule MyApp.Tesla.Request do
  @behaviour MyApp.Request

  use Tesla
  plug Tesla.Middleware.Retry, delay: 1000, max_retries:: 3

  def make do
    # perform request, return response
  end
end

This gives me a common interface and call MyApp.Request.make, while also giving me the ability to switch out the client with configuration, which I will use during testing.

Does this look sensible? Is there something inherently wrong with the above approach or, in your opinion, an improvement?

The second question is about testing MyApp.Tesla.Request – perhaps I can switch out the client I use in testing, but surely I still need to test that my Tesla request is configured/running properly.