Download the PHP package serenitylabs/phatcats without Composer

Functional PHP

Contents

• About
• Running Tests
• Type Classes
• Types
  • LinkedList
  • Maybe
    • Accessing the Value
    • Maybe as Functor
    • Maybe as Monad
    • Maybe as Monoid
    • Converting
  • Either
  • MaybeT
  • Validation
  • AssociativeArray

About

This library is designed to give PHP developers some Category-Theory-like facilities available in a language like Haskell.

Running Tests

Natively

This project uses Composer for dependency management and PHPUnit for unit testing. First, install the dependencies (including dev dependencies):

$ composer install --dev

PHPUnit requires access to autoloading and the autoloading files must first be generated by composer with the following command:

$ composer dump-autoload

Then, run the unit tests with

$ ./vendor/bin/phpunit

Using Docker

If you have docker installed you can run tests with:

$ make test

Type Classes

The following type classes are supported:

• SemiGroup
• Monoid
• Functor
• Monad
• Applicative

Note that not all types support all type classes. For more information on type classes check out the wikipedia page.

A Tale of Two Types of Type Class

The Typeclass directory contains the type classes that this library implements. They are implemented as interfaces that contain the methods of the type class. For example this is the SemiGroup type class:

The above interface can be implemented to combine two objects of any type. For example, here's an implementation for the Maybe type (this gives the same behavior as the Haskell instance):

But you may also notice the ObjectTypeclass directory. This directory contains a set of traits that mirror the type class interfaces. It turns out that in a majority of cases type class functions can be implemented as methods on the target type. Therefore, most of the type classes have counterparts in the ObjectTypeclass directory directory. These object type classes are prefixed with Object. For example, here's ObjectSemiGroup, the object version of the SemiGroup type class:

The first argument is a value whose type should be the same as the type of this and will be appended to it. Also note the second SemiGroup parameter. This is used to combine the values contained within the objects that are being appended. This is the same value that was passed as a constructor argument in the implementation of the MaybeSemiGroup class above.

One example of a type class function that cannot be implemented as a target object method is pure from the Applicative type class. This is the main reason that the type classes in the Typeclass directory exist; otherwise, they might not be needed!

Types

This library supports the following types:

• AssociativeArray
• Attempt
• Either
• LinkedList
• Maybe
• MaybeT
• Validation

Note that not all of these types have instances of the above type classes. Likewise there are instances of type classes that do not have a corresponding type (e.g., there are instances of Monoid for both ints and strings but no corresponding library types).

LinkedList

The LinkedList type is an abstract data type implementing a typical linked list data structure.

Creating

LinkedList's should be created using an instance of the LinkedListFactory class. For example, you can create an empty list:

You can create a LinkedList from a PHP array like so:

You can also easily create a range of values. This works exactly the same as the standard library function range.

LinkedList Monoid

LinkedLists can be appended:

LinkedList Functor

As expected, LinkedLists are functors. Simply pass a function of one argument to the map method:

LinkedList Monad

They are also monads:

LinkedList Applicative

The list applicative may be a little unintuitive if you've never seen it before. It allows you to apply each function in a list of functions to each of a list of arguments. An example may make it a bit clearer.

The __invoke magic method can also be used to achieve the same result:

You can even call a LinkedList of no-argument functions:

LinkedList Traversable

The traverse method of the Traversable type class is another method that at first may seem a little strange but is actually quite useful. traverse takes as its first argument a function that takes an element of the LinkedList and returns some monad. As its second argument it takes an instance of that same monad. The return value of traverse is an instance of the monad wrapping a LinkedList containing the values that were wrapped in monads returned by the passed-in function. That was a mouthful but it's more intuitive when seen in an example. First, let's define a function that returns a Maybe:

Then we'll traverse a list of integers with that function:

traverse is useful for when you want to map over a LinkedList but the result of doing so would give you a LinkedList of some monad. Using traverse inverts the LinkedList and the monad.

But note what happens in this example if one of the calls to $divideTwelveBy returns Nothing:

The Traversable type class also has a method sequence that is useful for the situation when you already have a LinkedList of some monad:

LinkedList Foldable

LinkedLists can also be folded in various ways. Here's the classic example of summing a list of integers:

Or multiplying the elements of the same list:

There's also foldRight which can be used, among other things, to concatenate two LinkedLists:

Then there's fold which takes some monoid as it's argument. The idea with fold is that it presumes that the elements are all some monoid and combines each of them by appending them all together. The monoid argument is needed in the case that the LinkedList is empty.

foldMap is similar to fold but takes as a second parameter a function that converts each element to a monoid and then appends them. Here we have a LinkedList of strings; not a LinkedList of Maybe strings as above. The $toMonoid function converts them before appending them.

LinkedList Collection

LinkedList also implements the Collection trait. All operations work as expected and we will not describe them in detail except to give an example of filtering a LinkedList.

Maybe

Maybe is intended to be used to represent the situation where there is the possibility of having an absense of a value. Typically, you would use Maybe when you might ordinarily return a null value from a function. Sometimes Maybe is also used to represent an error condition.

Maybe is impelemented as an abstract class with two concrete sub-classes: Just and Nothing. But you cannot instantiate these sub-classes directly; you must use static creation methods defined in the Maybe class. If you want to put a regular value in a Maybe context, use the fromValue() static method like so:

After the above code executes, $maybeInt will be an instance of Just, unless $myInt was null in which case $maybeInt will be an instance of Nothing. If you want to represent the absence of value, use the nothing() static method:

Accessing the Wrapped Value

You may want to get direct access to the value wrapped in a Maybe. This only makes sense if you have a default value that can be used in the case that your Maybe is Nothing. In Haskell you would use the fromMaybe function to do this. Here, you do this by calling the getOrElse method like so:

Maybe as Functor

Another common desire is to simply apply a regular function to the value wrapped in the Maybe and to have the returned value wrapped back up in another Maybe. A datatype used in this manner is known as a Functor.

The following code shows how you could convert the string "apples" to uppercase while it's contained in a Maybe:

But if $a had been an instance of Nothing, then the result would have been Nothing() and the strtoupper function would never have been run. You can also chain maps:

There are a couple of things to note here. First, map takes a callable. In PHP callables take several forms but one of them is a string and in the first call to map, we passed in the string version of a built-in PHP function. In the second case we pass in an anonymous function, also a callable. See PHP's documentation on callable for more info. If calling a function using a string seems strange, good; it is strange! :)

Second, callables passed into map must be functions of one argument and that argument will be the value wrapped in the Maybe. Third, the value returned by this function will automatically be wrapped back up in a Maybe. So the result of calling map on a Maybe is again a Maybe which is what allows us to chain calls to map this way.

Maybe as Monad

It's not uncommon that the function you want to apply to the value wrapped in the Maybe itself returns a Maybe. You can't simply use the map method in this case. To demonstrate why, let's first create a function that returns Maybe. A classic example is the function head() which returns the first element of an array. Strangely, PHP does not have such a function and the recommended approach is not straitforward as evidenced by this StackOverflow answer: http://stackoverflow.com/a/3771228. But even if you use the convoluted solution described there, you still have to deal with a possible NULL value being returned in the case of an empty array.

The following function hides the complexity of getting the first element of an array and returns a Maybe type so that we don't need to deal with NULLs:

When given a non-empty array, the above function will return Just($v) where $v is the first value of the array argument. It will return Nothing in all other cases. See this file for examples of using this function.

To see why we can't use this function with map() let's expand on the example we used above but instead of starting off with a string wrapped in a Maybe, we have an array of string:

As you can see we're left with a Just inside of a Just and this is almost certainly not what you're going to want, in general. To fix this, we simple need to use the flatMap method instead:

A datatype used in this way is called a Monad.

Maybe as Monoid

You may find yourself in a situation where you want to combine several Maybes into one Maybe. In Haskell you would do this the the mappend function or the (<>) operator Here, you can do this using the append method. The following code shows how this works.

Converting

It is extremely common to want to convert your Maybe into some other type. In Haskell you would use the maybe function to achieve this. This library does not have such a function but you can use other techniques to achieve the same result. For example, you may want to convert a Maybe into an HTTP response. You could use PHP's instanceof operator like so:

A slightly better but equivalent way is to use the provided isNothing() method:

The recommended way to convert a Maybe to something else is to use the Visitor Pattern. You create a Maybe visitor by creating a class that implements the MaybeVisitor interface like so:

And you do the conversion by creating an instance of this visitor and passing it to the accept method of the Maybe:

And that's it!

Either

The Either datatype is used to represent one of two possible values. Either is very similar in functionality to Maybe - So similar in fact that I'm not going to go into detail on its use since it would be almost identical to what was presented for Maybe. But I will note the differences here.

Where Maybe is used to signal a possible lack of a value, Either is often used to signal the possibility of an error (though it is more general than that). The two sub-classes of the Either abstract class are the rather unintuitively named Left and Right. This is because the Either type is actually more general than simply indicating an error; it can be used to return any two possibilities. We use Either in this library for no other reason than because it's what Haskell does.

The Right subclass is used to signal a successful calculation. You create an instance like so:

The Left subclass is used to signal an error and you create an instance like so:

Notice that we created a Left by passing a string containing an error message. This is a common way to use Either for signalling error but Left can contain any type and we could have just as correctly (and possibly more clearly) passed in a custom error or exception class.

These are the only significant differences with Maybe.

MaybeT

TBD

AssociativeArray

AssociativeArray is a simple wrapper around PHP's native array so array/list processing methods could be added to it. Currently it only contains an implementation for the Traversable trait.

To create an instance of an AssociativeArray, one simply calls the constructor:

AssociativeArray Traversable

There are two methods in the Traversable trait: traverse() and sequence(). sequence() is the simpler of the two so we'll start with that. In the case of Maybe, sequence() is most-commonly used to convert an array of Maybe to a Maybe of array as follows:

Note that an instance of an Applicative must be passed to the sequence() method. This is an unfortunate consquence of dynamic typing where the type of the objects contained in the array is not known in the case of an empty array.

The traverse() method is similar except that it gives you the opportunity to run a function on each value in the array. To demonstrate this, we first define a function that returns an Either type (see Either):

Then we use that function in a call to traverse():

Note that both sequence() and traverse() have the characteristic that if one or more elements in the array is Nothing (in the case of sequence()) or if the function passed in to traverse() evaluates to Nothing (in the case of traverse()), then the result is also Nothing. Let's show this by looking at a slightly modified version of the above sequence() example: