pixel (pinterface) wrote,

The Terrible, Horrible, No-Good, Very-Bad Things I Do To PHP (Or: Greenspunning in PHP)

Controller Methods

What we call a “controller method”, at least, is simply a function which is called on the web side via AJAX and returns some value back (generally either JSON or a snippet of HTML).

When I started at my current employer, controller methods looked like this:

class MyController extends Controller {
  function f() {
    if (!$this->required('foo', 'bar', 'baz')) return;
    if (!is_numeric($this->requestvars['foo'])) return;
    // lots of code
    if (isset($this->requestvars['quux'])) {
      // do something with quux
    }
    // more code
    return $someString;
  }
}

Aside from being hideous, this has a number of glaring problems:

  1. The error handling is terrible.
  2. Validation is hard, and thus incredibly easy to screw up or forget entirely.
  3. Even something that should be easy―merely figuring out how to call the method―requires reading and understanding the entire method.

This just won’t do.

Surely it would be much nicer if controller methods could simply be defined like a regular function. Fortunately, PHP offers some manner of reflexive capabilities, meaning we can ask it what arguments a function takes. We can then match up GET/POST parameters by name, and send the function the proper arguments.

In other words, we can define the function more like:

class MyController extends Controller {
  function f($foo, $bar, $baz, $quux = null) {
    // lots of code
    if (isset($quux)) {
      // do something with quux
    }
    // more code
    return new Response\HTML($someString);
  }
}

And have it actually work. That’s much nicer. Now, we can call the method from PHP as easily as we call it from JavaScript, and we don’t have to read the entire function to figure out what arguments it takes.

(The astute reader will also notice I’ve moved to returning an object, so the response has a type. This is super-handy, because now it’s easy to ensure we send the apropriate content-type, enabling the JS side to do more intelligent things with it.)

Of course, this only tells us which arguments it takes, and whether they’re optional or required. We still need easier data validation. PHP provides type hints, but they only work for classes. Or do they?

Type Hints

In a brazen display of potentially ill-advised hackery (our code is a little more involved, but that should give you the general idea), I added an error handler that enables us to define non-class types to validate things.

So now we can do this:

class MyController extends Controller {
  function f(
    int $foo,
    string $bar,
    string $baz,
    int $quux = null
  ) {
    // lots of code
    if (isset($quux)) {
      // do something with quux
    }
    // more code
    return new Response\HTML($someString);
  }
}

And all the machinery ensures that by the time f() is executing, $foo looks like an integer, as does $quux if it was provided.

Now the caller of the code can readily know what the value of the variables should look like, and the programmer of the function doesn’t really have an excuse for not picking a type because it’s so easy.

Of course, this isn’t sufficient yet either. For instance, if I’d like to be able to pass a date into the controller, it has to be a string. Then the writer of the controller has to convert it to an appropriate class. Surely it’d be much nicer if the author of the controller method could say “I want a DateTime object”, which would be automagically converted from a specially-formatted string sent by the client.

Type Conversion via Typehints

Because PHP provides references via the backtrace mechanism, we can modify the parameters a function was called with.

class MyController extends Controller {
  function f(
    int $foo,
    string $bar,
    DateTime $baz,
    int $quux = null
  ) {
    // lots of code
    if (isset($quux)) {
      // do something with quux
    }
    // more code
    return new Response\HTML($someString);
  }
}

So while $baz might be POSTed as baz=2014-08-16, what f() gets is a PHP DateTime object representing that date. Due to the implementation mechanism, even something as simple as:

$mycontroller->f(1, “bar”, “2014-08-16”);

Will result in $baz being a DateTime object inside f().

Caveat

There is an unfortunate caveat, and I have yet to figure out if it’s a quirk of the way I implemented things, or a quirk in the way PHP is implemented, but optional arguments do not change. That is, SomeClass $var = null will result in $var still being a string. func_get_args() will contain the altered value, however.

Multiple Inheritance and Method Combinations

PHP is a single inheritance language. Traits add some ability to build mixins, which is super-handy, but has some annoying restrictions. Particularly around calling methods―in particular, you can’t define a method in a trait, override it in a class which uses a trait, and then call the trait method from the class method. At least, not easily and generally.

Plus there’s no concept of method combinations. It’d be really handy to be able to say “hey, add this stuff to the return value” (e.g., by appending to an array) and have it just happen, rather than having to know how to combine your stuff with the parent method’s stuff.

While I’m sad to say I don’t have this working generally across any class, I have managed to get it working for a particular base class where it’s most useful to our codebase. Subclasses and traits can define certain methods, and when called, the class heirarchy will be automatically walked and the results of calling each method in the heirarchy will be combined.

trait BobsJams {
  static function BobsJams_getAdditionalJams() {
    return [ new CranberryJam(), new StrawberryJam() ];
  }
}

trait JimsJams {
  static function JimsJams_getAdditionalJams() {
    return [ new BlackberryJam() ];
  }
}

class Jams {
  function getJams() {
    return (new MethodCombinator([], 'array_merge'))
      ->execute(new ReflectionClass(get_called_class()), "getAdditionalJams");
  }
}

class FewJams extends Jams {
  static function getAdditionalJams() {
    return [ new PineappleJam() ];
  }
}

class LotsOJams extends FewJams {
  use BobsJams;
  use JimsJams;

  static function getAdditionalJams() {
    return [ new OrangeJam() ];
  }
}

(new LotsOJams())->getJams();
// => [ OrangeJam, CranberryJam, StrawberryJam, BlackberryJam, PineappleJam ]

(The somewhat annoying prefix on the traits’ method names is to avoid forcing users of a trait to deal with name collisions.)

Naturally, all the magic of the getJams() method is hidden away in the MethodCombinator class, but it just walks the class hierarchy―traits included―using the C3 Linearization algorithm, calls those methods, and then combines them all using the combinator function (in this case, array_merge).

This, as you might imagine, greatly simplifies some code.

Oh, but you’re not impressed by shoehorning some level of multiple inheritance into a singly-inherited language? Fine, how about…

Context-Sensitive Object Behavior

Web code tends to be live, while mobile code is harshly asynchronous (as in: still needs to function when you have no signal, and then do something reasonable with data changes when you do have signal again), so what we care about changes between our Mobile API and our Web code, and yet we’d still like to share the basic structure of any given piece of data so we don’t have to write things twice or keep twice as much in our heads.

Heavily inspired by Pascal Costanza’s Context-Oriented-Programming, we define our data structures something like this:

class MyThing extends Struct {
  public $partA;
  public $userID;
  // ...
  function getAdditionalDefaultContextualComponents() {
    return [ new MyThingWebUI(), new MyThingMobileAPI() ];
  }
}

class MyThingWebUI extends Contextual {
  public $isReadOnly;
  // ...
  function getApplicableLayer() { return "WebUI"; }
}

class MyThingMobileAPI extends Contextual {
  public $partB;
  // ...
  function getApplicableLayer() { return "MobileAPI"; }
}

The two Contextual subclasses define things that are only available within particular contexts (layers). Thus, within the context of WebUI, MyThing appears from the outside to look like:

{
  "partA": "foo",
  "userID": 12,
  "isReadOnly": false,
}

But within the Mobile API, that same $myThing object looks like:

{
  "partA": "foo",
  "userID": 12,
  "partB": "bar",
}

In addition to adding new properties, each layer can also exclude keys from JSON serialization, add aliases for keys (thus allowing mobiles to send/fetch data using old_key, when we rename something to new_key), and probably a few other things I’m forgetting.

Conclusion

PHP is remarkably malleable. error_handlers can be used as a poor-man’s handler-bind (unlike exceptions, they run before the stack is unwound, but you’re stuck dispatching on regular expressions if you want more than one); scalar type hints can be provided as a library; and traits can be abused to provide a level of multiple inheritance well beyond what was intended. While this malleability is certainly handy, I miss writing code in a language that doesn’t require jumping through hoops to provide what feel like basic facilities. But I’m also incredibly glad I can draw from the well of ideas in Common Lisp and bring some of that into the lives of developers with less exposure to the fantastic facilities Lisp provides.

Bonus!

My employer is desperate for user feedback, and as such is offering a free eight week trial. So if you want to poke at stuff and mock me when things don’t work very well (my core areas are nutritional analysis for recipes and food-related search results), that’s a thing you can do.

If you're outside the US, I should warn you that we have a number of known bugs and shortcomings you're much more likely to hit (we use a US-based product database; searching for things outside ASCII doesn't work due to MySQL having columns marked as the wrong charset; and there's a lot of weirdness around time because most user times end up stored as unix timestamps). The two bugs will be fixed eventually, but since they're complicated and as the US is our target market they're not exactly at the top of the list.

Don't you just love it when people move to a new blog? I'd 301 redirect you if I could, but since I can't you'll have to click through to read comment count unavailable comments or leave your own.

Tags: lisp, php, programming, work
Comments for this post were disabled by the author