"Static methods are death to testability" - alternatives for alternative constructors? - php

Question

It is being said that "static methods are death to testability". If that is so, what is a viable alternative pattern for the below?
class User {
private $phone,
$status = 'default',
$created,
$modified;
public function __construct($phone) {
$this->phone = $phone;
$this->created = new DateTime;
$this->modified = new DateTime;
}
public static function getByPhone(PDO $pdo, $phone) {
$stmt = $pdo->prepare('SELECT * FROM `users` WHERE `phone` = :phone');
$stmt->execute(compact('phone'));
if (!$stmt->rowCount()) {
return false;
}
$record = $stmt->fetch(PDO::FETCH_ASSOC);
$user = new self($record['phone']);
$user->status = $record['status'];
$user->created = new DateTime($record['created']);
$user->modified = new DateTime($record['modified']);
return $user;
}
public function save(PDO $pdo) {
$stmt = $pdo->prepare(
'INSERT INTO `users` (`phone`, `status`, `created`, `modified`)
VALUES (:phone, :status, :created, :modified)
ON DUPLICATE KEY UPDATE `status` = :status,
`modified` = :modified');
$data = array(
'phone' => $this->phone,
'status' => $this->status,
'created' => $this->created->format('Y-m-d H:i:s'),
'modified' => date('Y-m-d H:i:s')
);
return $stmt->execute($data);
}
...
}
This is just a cut down example. The class has a few more methods and properties and there's more validation when writing to the database etc. The guiding design principle behind this class is that it models a user as an object. Some of the object's properties cannot be modified after it has been created, like the phone number (which acts as a primary id), the date the user was created and so on. Other properties can only be altered according to strict business rules, which all have strictly validating setters and getters.
The object does not represent a database record per se, the database is only seen as one possible form of permanent storage. As such, a database connector is not stored in the object but rather needs to be injected every time the object needs to interact with the database.
When a new user is created, this looks like:
$user = new User('[phone]');
When an existing user is restored from permanent storage, that looks like:
$pdo = new PDO('...');
$user = User::getByPhone($pdo, '[phone]');
If I were to take the "death to testability" line seriously, this is supposedly bad. I am perfectly able to test this object though, since it is fully dependency injected and the static methods have no state. How could I do this any differently and avoid the use of static methods? Or rather, what exactly argues against static in this case? What makes this particular use of static methods so hard to test?

Answer 1

This is mostly a summary of (my perspective of) the chat that ensued between me and #zerkms:
The point of contention is actually this:
public function doSomething($id) {
$user = User::getByPhone($this->pdo, $id);
// do something with user
return $someData;
}
This makes it hard to test doSomething since it hardcodes the User class, which may or may not have a lot of dependencies. But this is in fact the same as instantiating the object using a non-static method:
public function doSomething($id) {
$user = new User;
$user->initializeFromDb($this->pdo, $id);
// do something with user
return $someData;
}
We're not using a static method, but it's still unmockable. Actually, it got worse.
The answer is to use a factory:
public function doSomething($id) {
$user = $this->UserFactory->byPhone($id);
// do something with user
return $someData;
}
Now the factory can be dependency injected and mocked and the User class is no longer hardcoded. You may or may not think this overkill, but it certainly improves mockability.
That does not change the fact though that this factory may very well instantiate the actual user object using a static method:
public function byPhone($id) {
return User::getByPhone($this->db, $id);
}
There's no difference between using a static method or a regular constructor here.
$user = new User($db, $id);
$user = User::getByPhone($db, $id);
Both expressions return an instance of User and both "hardcode" the User class. Which simply needs to happen at some point anyway.
For my use case, a static constructor method makes the most sense for the object. And as was demonstrated, static methods are not the problem. Where to call them is the point of contention, not that they exist at all. And I have yet to see a convincing argument for not using static constructors, since they can be wrapped in a factory, which alleviates any problem with mockability, the same as it does for regular object instantiation.

Answer 2

As long as OP asked about general issue, and not asked how to improve his particular code - I'll try to answer using some abstract and tiny classes:
Well, it is not more difficult to test static methods themselves, but it is more difficult to test the methods that use static methods.
Let's see the difference on small example.
Let's say we have a class
class A
{
public static function weird()
{
return 'some things that depends on 3rd party resource, like Facebook API';
}
}
It does some work that requires setting up additional environment (specifying API keys in this case) and internet connection to FB API services. It will take some time to test this method (just because of network and API lags), but it is definitely easy enough to test it.
Now, we implement a class that uses A::weird() method:
class TestMe
{
public function methodYouNeedToTest()
{
$data = A::weird();
return 'do something with $data and return';
}
}
For now - we cannot test TestMe::methodYouNeedToTest() without additional steps required to make A::weird() worked. Yes, instead of testing methodYouNeedToTest we also need to do things that are not directly related to this class, but to another.
If we followed another way from the very begin:
class B implements IDataSource
{
public function weird()
{
return 'some things that depends on 3rd party resource, like Facebook API';
}
}
you see - the key difference here is that we implemented the IDataSource interface and made method normal, not static. For now we could rewrite our code above in this way:
class TestMe
{
public function methodYouNeedToTest(IDataSource $ds)
{
$data = $ds->weird();
return 'do something with $data and return';
}
}
And now we don't rely on specific implementation but we do on an interface. And now we can easily mock datasource.
Such abstractions help keeping our tests focusing more on the testing itself rather than on creating necessary environment.
Such steps helps us to have our unit tests fast. While we still could have acceptance, load and functional tests (but it is another story) that test that our application works as expected

Answer 3

Static methods are only "death to testability" if they depend on state. If you avoid writing such methods to begin with (which you should), then this issue simply goes away.
The Math.abs() example given is one of a good use of a static method. It does not depend on state, therefor it is super easily tested.
That said, whether or not you think static methods should be used is another story. Some people dislike their seemingly procedural nature. I agree with those who say that OOP is a tool, not a goal. If writing "proper" OO code doesn't make sense for a particular situation (e.g. Math.abs()), then don't do it. I promise the boogey man won't eat your application just because you used a static method. :-)

Answer 4

As mentioned in comments, I'd implement a repository pattern for this case.
For example, User would be a simple model with read-only properties
class User {
private $phone,
$status = 'default',
$created,
$modified;
public function __construct($phone) {
$this->setPhone($phone);
$this->created = new DateTime;
$this->modified = new DateTime;
}
private function setPhone($phone) {
// validate phone here
$this->phone = $phone;
}
public function getPhone() {
return $this->phone;
}
public function getCreated() {
return $this->created;
}
public function getModified() {
return $this->modified;
}
}
Your repository interface could then look like this
interface UserRepository {
/**
* #return User
*/
public function findByPhone($phone);
public function save(User $user);
}
A concrete implementation of this interface could look something like this
class DbUserRepository implements UserRepository {
private $pdo;
public function __construct(PDO $pdo) {
$this->pdo = $pdo;
}
public function findByPhone($phone) {
// query db and get results, return null for not found, etc
$user = new User($phone);
// example setting the created date
$reflectionClass = new ReflectionClass('User');
$reflectionProperty = $reflectionClass->getProperty('created');
$reflectionProperty->setAccessible(true);
$created = new DateTime($res['created']); // create from DB value (simplified)
$reflectionProperty->setValue($user, $created);
return $user;
}
public function save(User $user) {
// prepare statement and fetch values from model getters
// execute statement, return result, throw errors as exceptions, etc
}
}
The cool thing here is that you can implement many different repositories, all with different persistence strategies (XML, test data, etc)

Answer 5

I think the citation you give has a good point but takes too hard a line.
Your static method is what he calls a "leaf" method. In this case I think you are fine, as long as your static method doesn't have any external dependencies.
The alternative is a data mapper, an object which is aware of the relationship between User and how it is stored in the database. Example:
class UserDBMapper {
protected $pdo;
protected $userclass;
function __construct(PDO $pdo, $userclass) {
$this->db = $db;
// Note we can even dependency-inject a User class name if it obeys the interface that UserMapper expects.
// You can formalize this requirement with instanceof, interface_exists() etc if you are really keen...
$this->userclass = $userclass;
}
function getByPhone($phone) {
// fetches users from $pdo
$stmt = $this->db->query(...);
$userinfo = $stmt->fetch()....
// creates an intermediary structure that can be used to create a User object
// could even just be an array with all the data types converted, e.g. your DateTimes.
$userargs = array(
'name' => $userinfo['name'],
'created' => $userinfo['created'],
// etc
);
// Now pass this structure to the $userclass, which should know how to create itself from $userargs
return new $this->userclass($userargs);
}
function save($userobj) {
// save method goes in the Mapper, too. The mapper knows how to "serialize" a User to the DB.
// User objects should not have find/save methods, instead do:
// $usermapper->save($userobj);
}
}
This is a very powerful pattern (e.g., you need no longer have 1-1 type<->table, instance<->row correspondence like the Active Record pattern requires), and you can completely change your serialization method without altering your domain objects at all. It should also be obvious how much easier the mapper is to test. But in many cases this pattern is also over-engineered and more than you need. After all, most websites use the much simpler Active Record pattern.

Answer 6

Firstly, the DateTime class was a good (tricky) class to choose because it is a horrible class. All of its important work is done in the constructor and there is no way to set the date/time after it has been constructed. This requires us to have an objectGenerator that can build the DateTime object at the correct time. We can still manage this without calling new in the User class though.
I have kept things very simple to solve the problem at hand, but they can be extended easily to handle an arbitrarily complex problem.
Here is a simple objectGenerator to remove the coupling you get with new.
class ObjectGenerator {
public function getNew($className) {
return new $className;
}
}
Now we inject all of the dependencies into the constructor. The constructor shouldn't do real work, only set up the object.
class User {
private $phone,
$status = 'default',
$created,
$modified,
$pdo,
$objectGenerator;
public function __construct(PDO $pdo, $objectGenerator) {
$this->pdo = $pdo;
$this->objectGenerator = $objectGenerator;
$this->created = $this->objectGenerator->getNew('DateTime');
}
public function createNew() {
$this->phone = '';
$this->status = 'default';
$this->created = $this->objectGenerator->getNew('DateTime');
}
public function selectByPhone($phone) {
$stmt = $this->pdo->prepare('SELECT * FROM `users` WHERE `phone` = :phone');
$stmt->execute(compact('phone'));
if (!$stmt->rowCount()) {
return false;
}
$record = $stmt->fetch(PDO::FETCH_ASSOC);
$this->phone = $record['phone'];
$this->status = $record['status'];
$this->created = $record['created'];
$this->modified = $record['modified'];
}
public function setPhone($phone) {
$this->phone = $phone;
}
public function setStatus($status) {
$this->status = $status;
}
public function save() {
$stmt = $this->pdo->prepare(
'INSERT INTO `users` (`phone`, `status`, `created`, `modified`)
VALUES (:phone, :status, :created, :modified)
ON DUPLICATE KEY UPDATE `status` = :status,
`modified` = :modified');
$modified = $this->objectGenerator->getNew('DateTime');
$data = array(
'phone' => $this->phone,
'status' => $this->status,
'created' => $this->created->format('Y-m-d H:i:s'),
'modified' => $modified->format('Y-m-d H:i:s')
);
return $stmt->execute($data);
}
}
Usage:
$objectGenerator = new ObjectGenerator();
$pdo = new PDO();
// OR
$pdo = $objectGenerator->getNew('PDO');
$user = new User($pdo, $objectGenerator);
$user->setPhone('123456789');
$user->save();
$user->selectByPhone('5555555');
$user->setPhone('5552222');
$user->save();
So, no new or static in the user class. Try testing both solutions. The test code is a pleasure to write without the calls to new. All classes that use the User will also be easy to test without static calls to it.
The differences in test code are:
new/static - Require a stub for each new or static call to stop the unit from reaching outside of itself.
dependency injection - Mock objects can be injected. It is painless.