JS assessment: A test-driven approach to assessing JS skills

Great idea. I do PHP training, and one of the things I did in some classes is to give failing unit tests to students and have them make them work. I noticed a couple of consistent points when I'd do this:

1. How some students came up with interesting approaches I wouldn't have initially thought up.

2. How many students wouldn't get the tests to pass. They'd do something part way, which would have looked 'good enough' if I'd just described the problem. But when running via a test runner, it failed.

I like this idea but I think the design needs to be pretty drastically rethought. Node.js is not really the right medium for this sort of test, although it might function as a first barrier for screening the interested; "you must be able to follow readmes to install software" etc.

What I'm more concerned is the idea that this is apparently structured like an automated test. If you used it that way, then here is a function which would pass the first test file:

    fn = function (a, b) { 
        switch (b) {
            case 3:         return 2;
            case undefined: return 10;
            case 2:         return '1 3 4'.split(" ");
            case 10:        return [0, 0, 0, 0, 10];
            case 'z':       return '1 2 z 3 4'.split(" ");
            default:        return '1 2 3 4 a b c'.split(" ");
        }
    };

It was built by just looking at the "expect" calls, finding the input to the function and the output from it, and duplicating that output.

That's obviously not what you want to test. You really want to say, "you give me an object, and I'm going to generate a couple random test vectors and make sure you do the right thing with those vectors." You shouldn't hardcode the result, but the right algorithms, so that someone else's solutions can be measured against the right way to do things. And then you need to ask people not to modify the right solutions, but to duplicate them.

I agree, I also thought it wasn't clear if we were supposed to be writing tests, or code that was to be tested. Instead of modifying the test cases have them fill out a functions file and then run the random tests against those.

Isn't this what is known as currying[1]?

[1] http://en.wikipedia.org/wiki/Currying

I definitely want to add more tests to the set -- if you have ideas for other tests, please open an issue or a pull request on the repo!