restore/1 exposes restoration as complete before it may actually be complete

[liamwhite]

The implementation of restore/1 calls the server:

patch/lib/patch/mock/server.ex

Lines 119 to 125 in 69e1a36

	@doc """
	Restores a module to its original state.
	"""
	@spec restore(module :: module()) :: :ok
	def restore(module) do
	call(module, :restore, fn -> :ok end)
	end

which triggers the mock server to shutdown:

patch/lib/patch/mock/server.ex

Lines 185 to 187 in 69e1a36

	def handle_call(:restore, _from, state) do
	{:stop, {:shutdown, {:restore, do_restore(state)}}, :ok, state}
	end

patch/lib/patch/mock/server.ex

Lines 207 to 217 in 69e1a36

	@spec call(module :: module(), message :: term(), default :: term()) :: term()
	defp call(module, message, default) do
	server = Naming.server(module)
	try do
	Freezer.get(GenServer).call(server, message)
	catch
	:exit, {:noproc, _} ->
	default.()
	end
	end

GenServer will then reply to the call while termination is in progress.

If you do something silly in your test like

  def reset_spy do
    restore(MyModule)
    spy(MyModule)
  end

, this will almost guarantee a test flake when spy/1 sees the old pid that is in the process of terminating, and then not create a new Mock server. So the new spy will not work.

The proper fix for this would be to make sure restore waits until the mock server is fully terminated.