Andreas Garnæs

How to test Async OCaml code

In the previous blog post, we used Core and Async to write a tiny library for talking to Memcached using the binary protocol. I wanted to write tests for the library in a readable and succint manner to ensure correctness – the type system cannot ensure binary data is parsed correctly after all. The regular go-to tool for testing OCaml code is OUnit, but this doesn’t work well with Async. As I couldn’t find a suitable library, I decided to write something myself.

The Goal

My goal was to write a tiny testing framework with the following features:

Inspired by Command, I also wanted write the tests in a DSL-like manner. This is an example of the end goal:

Async_suite.(
  create_suite
    ~before_all:(fun () ->
      Memcached.connect "localhost" 11212
    )
    ~before:ident
    ~after:(fun cache ->
      Memcached.flush cache
    )
    ~after_all:(fun cache ->
      Memcached.close cache
    )
  
  |> test "setting a value" (fun cache ->
    Bool.assert_equal (Memcached.set cache "foo" "bar") true
  )

  |> test "getting a value" (fun cache ->
    String.assert_equal (Memcached.get cache "mykey") "myvalue"
  )

  (* add more tests here ...
  |> test "foo" (fun cache ->
    ...
  )
  *)

  |> run
)

Here we’re using Local Opens to avoid prefixing function names with the module name all the time, i.e. Async_suite.(f x y) is the same as Async_suite.f x y. We’re also using the pipe function x |> f, which is the same as f x.

The Implementation

The type of a test suite is parameterized over the argument for the before_all and after_all functions, and the argument for the before and after functions:

(* the type of a single test *)
type 'a test = string * ('a -> unit Deferred.t)

(* test functions should raise AssertionFailure if assertions fail *)
exception AssertionFailure of string

(* the type of a test suite *)
type ('suite_params, 'params) t = {
  before_all : unit -> 'suite_params Deferred.t;
  before     : 'suite_params -> 'params Deferred.t;
  after      : 'params -> unit Deferred.t;
  after_all  : 'suite_params -> unit Deferred.t;
  tests      : 'params test list
}

Note that a test function is expected to return a unit Deferred.t that resolves when the test is completed and should raise AssertionFailure if an assertion fails. For brevity I’ll write ('suite_params, 'params) t as ('a, 'b) t henceforth.

Creating the test suite and adding tests is straight forward:

(* Create a empty test suite with before- and after-hooks *)
let create ~before_all         (* unit ->   'a Deferred.t *)
           ~before             (* 'a   ->   'b Deferred.t *)
           ~after              (* 'b   -> unit Deferred.t *)
           ~after_all =        (* 'a   -> unit Deferred.t *)
  {
    before_all;
    before;
    after;
    after_all;
    tests = []
  }

(* Add a test function to a test suite                  *)
(* test : string -> 'b test -> ('a, 'b) t -> ('a, 'b) t *)
let test name f t = { t with tests = (name, f)::t.tests }

Now we’re just missing functionality to run the tests:

(* run_test_with_suite : ('a, 'b) t -> 'a -> 'b test -> unit Deferred.t *)
let test_with_hooks t suite_params (name, test) = 
    t.before suite_params                              >>= fun params ->
    try_with ~extract_exn:true (fun () -> test params) >>= fun result ->
    t.after params                                     >>| fun () ->
    match result with
    | Ok () ->
        printf "%s: PASS\n" name
    | Error (AssertionFailure s) ->
        printf "%s: FAIL\n%s\n" name s
    | Error exn ->
        printf "%s: ERROR\n%s\n" name (Exn.to_string exn)

(* run : ('a, 'b) t -> unit Deferred.t *)
let run t =
  t.before_all () >>= fun suite_params ->
  let tests = List.rev t.tests in
  let run_test = test_with_hooks t suite_params in
  Deferred.List.iter ~how:`Sequential tests ~f:run_test >>= fun () ->
  t.after_all suite_params

In brief, iterate sequentially over each test function while making sure to call the before- and after-handlers at appropriate times. Test functions are expected to raise AssertionFailure if the test fails, which is then caught and printed – otherwise the test is deemed a pass.

Writing Assertions

Raising AssertionFailure can be done writing plain OCaml code, but I’ve found it very helpful to define some assertion-functions myself. The following code makes it easy to assert on values of type Deferred.Or_error.t (doc), which is used in the Memcached-client, and get helpful output on failures:

(* signature for assertable modules *)
module type Assertable = sig
  type t
  val equal : t -> t -> bool
  val to_string : t -> string
end

(* AssertionError will be raised if an unexpected exception is raised       *)
exception AssertionError of string

(* MakeAsserter-functor will return a module with a suitable assert_equals- *)
(* function given an Assertable                                             *)
module MakeAsserter (M : Assertable) = struct
  let assert_equal actual_dfd expected =
    actual_dfd >>= function
    | Error e ->
        raise (AssertionError (Core.Error.to_string_hum e))
    | Ok actual when M.equal actual expected ->
        return ()
    | Ok actual ->
        let err = sprintf "Expected %s was %s" (to_string expected)
                                               (to_string actual)   in
        raise (AssertionFailure err)
end

(* Make asserters for common types *)
module Int    = MakeAsserter(Int)
module Int64  = MakeAsserter(Int64)
module Bool   = MakeAsserter(Bool)
module String = MakeAsserter(String)

The signature Assertable ensures we can compare and print values of some type t. This signature is coincidentally fulfilled by all the common data types in Core. The functor MakeAsserter allows us to create asserters for any Assertable with suitable behavior.

These asserters make it easy to write concise assertions:

String.assert_equal (Memcached.get  cache "mykey") "foo"
Int64.assert_equal  (Memcached.incr cache "count" 12L) 42L

Wrap up

In this blog post we wrote a tiny library for testing code that uses Async. The resulting library makes it easy to write concise tests with life-cycle hooks and helpful test output. I wrote this out of a need to test Async code and couldn’t find a suitable test library, but if I missed something, please write a comment and enlighten me.

If you like this post, please vote on Hacker News.

comments powered by Disqus