dbstudios/analyst

A feature testing library inspired by GitHub's Scientist

2.0.1 2016-01-27 22:41 UTC

Requires

• php: >= 5.6

Requires (Dev)

• phpunit/phpunit: 5.1.*
• satooshi/php-coveralls: 1.0.*

MIT

• Tyler Lartonoix

README

Analyst

A PHP feature testing library inspired by GitHub's Scientist.

Please be aware, this README is a work in progress, and may contain incomplete information.

Installation

$ composer require dbstudios/analyst

How do I analyze?

Let's pretend that you have a large web application, and are in the process of changing the way you verify user permissions when accessing a page.

class MyController {
    public function isAllowed(User $user, $slug) {
        return Experiment::create('user-allowed')
            ->setContext([
                'user' => $user, // context ignores keys, however it's helpful to use keys to "flag" what each variable means
                'slug' => $slug,
            ])
            ->control(function(User $user, $slug) {
                return $user->isAdmin() || $user->getPermissions()->contains('view.page.' . $slug);
            })
            ->candidate(function(User $user, $slug) {
                return $user->getPermissionProvider()->isAccessAllowed($slug);
            })
            ->run();
    }
}

That's it! The run method of Experiment will always return the result from your control. Both your control and your candidate will be run, and Analyst will record what happened to both blocks of code. In the next section, we'll take a look at what you can do with that information.

Publishing Experiment Results

Testing multiple code paths is pretty useless if we don't do anything with the information we get from our experiment. That's where publishing comes into play. All you need to do is implement that publish method in a custom Experiment class.

class MyExperiment extends Experiment {
    private $db;
    private $stmt;

    public function __construct(PDO $db, $name = 'experiment') {
        parent::__construct($name);
        
        $this->db = $db;
        $this->stmt = $db->prepare('insert into experiment_results (name, timestamp, behavior, is_control,
            execution_time, result) values(:experiment, UTC_TIMESTAMP(), :behavior, :control, :duration, :result)');

        $this->stmt->bindValue(':experiment', $name);
    }

    public function publish(Result $result) {
        $behaviors = $result->getCandidates();
        
        array_unshift($candidates, $result->getControl());
        
        $this->db->beginTransaction();
        
        foreach ($behaviors as $behavior) {
            $this->stmt->bindValue(':behavior', $behavior->getName());
            $this->stmt->bindValue(':control', $behavior === $result->getControl(), PDO::PARAM_BOOL);
            $this->stmt->bindValue(':duration', $behavior->getDuration());
            $this->stmt->bindValue(':result', serialize($behavior->getResult());
            
            $this->stmt->execute();
        }
        
        $this->db->commit();
    }
    
    public static function create($db, $name = 'experiment') {
        return new MyExperiment($db, $name);
    }
}

Now, if we were to rewrite our earlier user permission example to use our custom MyExperiment class, it would look like this.

class MyController {
    public function isAllowed(User $user, $slug) {
        return MyExperiment::create($this->getConnection(), 'user-allowed')
            // set up context, control, and candidate...
            ->run();
    }
}

In the above example, we created our own Experiment implementation, named MyExperiment. In it, we made sure we had a database connection (using PHP's PDO class), and wrote data to it using the publish method. Then, when we ran our experiment within our controller example, all we needed to do was provide the extra arguments to the create static method of MyExperiment (which we implemented ourselves to allow for the additional constructor argument).

After every experiment is run, Analyst will always call the publish method of the experiment used to do our analysis. Our options aren't limited to simple database inserts. You might decide to log the data to a file using a library like Monolog, or you might send it off to a cloud service. It's up to you!