Mocking is evil. At the very notion of that, all the citizens of Nounlandia, are screaming for my blood, shouting ‘Heresy! Heresy!’. I think I can hear the sound of sharpening stones on pitchfork tines even as I write this. Mocking is fundamentally evil. It encourages you to write bad, poorly factored code, not-very-functional code. It encourages you to avoid standing up as much of your system as possible during integration testing. Finally, all too often at the end of all that mocking all you end up with are some really well tested Mocks nested N! levels. In other words, mocking produces a bunch of useless code that costs money and time to maintain.

Well, smart guy, I can already hear you saying, ‘If that’s true how do you test functions? Huh? Tell me that!’. Well the answer to that is the same as the reply to most other questions like this: By writing, well factored, idiomatic code. In this case, you need to separate out your side effect containing code from your side effect free code, then create unit tests around your side effect free code.

Let’s take an example to get your mind around the solution. Let’s say that we want to get the result of the command git describe --tag --always and check to see if it is a git ref or something else. Well, git refs are hex numbers so we can get away with making a little function that tests that ref. Let’s do this in Erlang since I just got done writing something similar not half an hour ago. Let’s take the naive approach first.

-module(foo).

-export([get_git_ref/0]).

-spec get_git_ref() -> ok | {error, no_ref}.
get_git_ref() ->
  PossibleRef = re:replace(os:cmd("git describe --tags --always"), "\\s+",
                           "",
                           [global]),
  case re:run(PossibleRef, "^[0-9a-fA-F]+$")) of
    {match, _} ->
      ok;
    nomatch ->
      {error, noref}
  end.

Obviously, this isn’t testable without having a git repo. So our unit tests aren’t possible unless we mock os:cmd/1 right? No! No! No!. We can make this testable bit of refactoring. Let’s make a non-naive version.

-module(foo).

-export([get_git_ref/0]).

-spec get_git_ref() -> ok | {error, no_ref}.
get_git_ref() ->
  is_ref(trim(os:cmd("git describe --tags --always"))).

is_ref(Ref) ->
  case re:run(PossibleRef, "^[0-9a-fA-F]+$")) of
    {match, _} ->
      ok;
    nomatch ->
      {error, noref}
  end.

trim(Input) ->
  re:replace(Input, "\\s+", "", [global]).

Notice here that we localize our side-effects to a small function that calls other functions to do the majority of the work. Every single one of those other functions is side effect free. Our code goes from being a side effect dependent mess to well factored testable code. Well factored, testable code is your goal when it comes to code that has side effects. Taking this approach will vastly reduce the need for Mocks.

Your next complaint will be ‘Well this still isn’t testable! We don’t export is_ref/1 and trim/1, so how can you test it?’. This issue of private functions leads us to the next tenant of avoiding Mocks. Your tests should live near the code that they test. If you have code that tests functions in a module than those tests should be in that same module. In Erlang, we would do it as follows, but the principle applies in any language.

-module(foo).

-export([get_git_ref/0]).

-spec get_git_ref() -> ok | {error, no_ref}.
get_git_ref() ->
  is_ref(trim(os:cmd("git describe --tags --always"))).

is_ref(Ref) ->
  case re:run(PossibleRef, "^[0-9a-fA-F]+$")) of
    {match, _} ->
      ok;
    nomatch ->
      {error, noref}
  end.

trim(Input) ->
  re:replace(Input, "\\s+", "", [global]).

-ifdef(DEV_ONLY).
-include_lib("eunit/include/eunit.hrl").

is_ref_test() ->
   ?assertMatch(ok, "af19665");
   ?assertMatch({error, "foobarbaz"}, "foobarbaz").

-endif.

Why should your tests be near the code it tests? For the same reason documentation should be near the things it documents. Tests are a form of documentation, perhaps the primary form of documentation. That implies that when the code changes the tests that test that code change as well. Otherwise, they tend to get out of sync and cause a lot of flipping back and forth between code and tests when changing/reading/maintaining anything. You would think this would be a no-brainer, but I have had people argue vociferously against this.

So now we have tests, but I can hear one last complaint echoing through the halls of the great Temple of Imperativeness. ‘You aren’t testing the functions that contain the side effects!!! You fail!’. Nice try, but not quite. While we don’t try to unit-test the functions that contain side effects. This doesn’t mean that you don’t try to test them. You need to stand your system up during testing. That way you can test the functionality your system and not how well you have mocked it. Figuring that out is very hard. If you plan that from the beginning of the project, it makes standing things up much easier. In either case, you still have to do it. Otherwise, your tests are useless.

If you don’t see another post from me, you will know the forces of Nounlandia has breached my fortress citadel and has taken me prisoner. Even now they are probably torturing me by forcing me to write EJBs Java and Groovy using Eclipse on Windows. Pity me. Pity me greatly. ⤧  Next post Problems with the Peruvian Startup Environment ⤧  Previous post Team Development with Github