[erlang-questions] test driven development

Joe Armstrong <>
Sun Nov 25 13:58:01 CET 2007


I think a lot of Erlang programmers have accidentally being using test
driven development for a long time without knowing it. This is because
the test mechanism
is so lightweight that it goes unnoticed...

There are actually two interpretations of a pattern:

   - a test
  - an unpacking operation

If there is any possibility that an '=' can fail then the '=' can be
viewed as a test.

Suppose I want to start by writing unit tests before I write my code.

I'll illustrate this by writing a function nth
that returns the n'th element of a list.

I can start by writing a test case:

test()->
    b = nth(2,[a,b,c]).

Because = can fail here this serves as a test.
This can be viewed as an extremely light-weight unit test

Now I write my nth function and test it

..

As I proceed I add more tests:

test() ->
    b = nth(2,[a,b,c]),
    {'EXIT',_} = (catch, nth(-1, [a,b])),
    {'EXIT',_} = (catch nth(3,[a,b])),
    hooray.

This is three unit test, but no big deal.

I often write a function test/0 like this (that returns hooray) and
build this into my
edit-compile-test cycle - when all my module say hooray then I'm happy.

Functions can be viewed as having in-build unit test code in many places:

    foo(...) ->
          ...
          {ok, Socket} = gen_tcp:connect(....)
          ...

is a test AND an unpacking operation boiled into one.

    Suppose I write

         ...
         gen_tcp:send(Socket, Data)
        ...

     This involves no testing, but adding a test is easy

        ...
        ok = gen_tcp:send(Socket, Data)
       ....

     adds a test - this has the added benefit of causing the program
to fail as soon as the
error is detected. "fail early" is *always* a good idea since it
minimizes the chance
of causing additional errors.

     Even without equality the way functions are written provides a
form of assertion checking.

     foo(a) -> 1;
     foo(b) -> 2.

asserts that the argument to foo/1 is a or b. So ideas of assertions
and "design by contract" are pretty much implicit in all Erlang code.

It's only after the Erlang book was published that I realized that this way of
interpreting code was not widespread - this is perhapses something further books
and articles might like to emphasis.







On Nov 25, 2007 3:41 AM, Andrew Arrow <> wrote:
> I've just finished the Joe Armstrong book (coming from a Java and Ruby
> background) and I noticed there was nothing in the book about test
> driven development.
>
> Is the idea that Erlang forces you to write such bug free code there
> are no need for tests?  Or was this simply beyond the scope of the
> book?
>
> Do professional Erlang programmers create lots of unit and functional
> tests like Java and Ruby programmers do?  If so, is there a way to run
> all the tests like there is with junit/testng or Test::Unit in ruby?
> _______________________________________________
> erlang-questions mailing list
> 
> http://www.erlang.org/mailman/listinfo/erlang-questions
>



More information about the erlang-questions mailing list