2018-05-13
Getting zippy with PHP arrays
Introduction
zip
- not to be confused with the archive format - is a useful merging operation that is a native feature in many languages including Python, Ruby and functional languages such as Elixir and Haskell.
Exact behaviour varies by language but, generally speaking, zip
combines multiple input sequences into an output sequence of values grouped by their corresponding index.
Examples
This is best illustrated by example. Python’s native zip()
function combines one or more lists and groups their values into a list of tuples; e.g:
zip([1, 2], ['a', 'b'], ['A', 'B'])
# [
# (1, 'a', 'A'),
# (2, 'b', 'B')
# ]
Elixir - which has two slightly different implementations, as illustrated below - also zips lists into tuples; e.g:
# Enum.zip/1 accepts an iterable
# containing an arbitrary but
# finite number of iterables
Enum.zip [[1, 2], [:a, :b], [:A, :B]]
# [
# {1, :a, :A},
# {2, :b, :B}
# ]
# Enum.zip/2 accepts exactly two iterables
Enum.zip [1, 2], [:a, :b]
# [
# {1, :a},
# {2, :b}
# ]
Ruby takes a predictably OOP approach to combine values into an array of arrays; e.g:
[1, 2].zip [:a, :b], [:A, :B]
# [
# [1, :a, :A],
# [2, :b, :B]
# ]
JavaScript doesn’t have native zip
functionality (or tuples) but utility libraries such as Lodash provide the functionality to combine values into an array of arrays; e.g:
_.zip([1, 2], ['a', 'b'], ['A', 'B']);
// [
// [1, 'a', 'A'],
// [2, 'b', 'B']
// ]
And in PHP?
PHP doesn’t have a native zip()
function either, but a lesser-known feature of array_map()
approximates its behaviour. As stated in the documentation:
An interesting use of this function is to construct an array of arrays, which can be easily performed by using NULL as the name of the callback function
Interesting indeed! array_map()
is normally used to apply a callback to one or more arrays, but if you pass null
as its first argument, and pass it two or more arrays - more on this presently - it will implicitly “zip” their values; e.g:
array_map(null, [1, 2], ['a', 'b'], ['A', 'B']);
// [
// [1, 'a', 'A'],
// [2, 'b', 'B'],
// ]
A PHP implementation
Let’s use the behaviour of array_map()
to create our own zip()
function. Using the variadic syntax (...
) introduced in PHP 5.6, we can elegantly unpack variadic arguments, and add some type hinting for good measure:
function zip(array ...$arrays): array
{
return array_map(null, ...$arrays);
}
zip([1, 2], ['a', 'b'], ['A', 'B']);
// [
// [1, 'a', 'A'],
// [2, 'b', 'B'],
// ]
A better PHP implementation
However, as implied above, there’s a catch… passing only one array to array_map()
just returns the same output as input:
array_map(null, [1, 2]);
// [1, 2] *not* [[1], [2]]
This is not consistent with the behaviour of zip
in other languages, so let’s rectify that:
function zip(array $first, array ...$rest): array
{
return $rest ? array_map(null, $first, ...$rest)
: array_chunk($first, 1);
}
Again we are using PHP’s variadic syntax to accept and unpack a dynamic number of arguments, while retaining array
type checking on those arguments. When only one array is passed to zip()
the variadic $rest
argument is an empty array, which which we can use to assert a condition.
And, if only one array is present, we instead use PHP’s built in array_chunk()
to format the output into an array of single element sub-arrays as required.
This gives us the following behaviour:
zip([]);
// []
zip([1, 2]);
// [
// [1],
// [2],
// ]
zip([1, 2], ['a', 'b']);
// [
// [1, 'a'],
// [2, 'b'],
// ]
What about arrays with different lengths?
This is where things get interesting. As mentioned at the beginning, implementations vary. Our implementation does what I’ll call a “full” zip, where the values are zipped according to the longest input, and non-existent values are substituted with null
:
zip([1, 2], ['a'], ['A', 'B', 'C']));
// [
// [1, 'a', 'A'],
// [2, null, 'B'],
// [null, null, 'C']
// ]
This is the same behaviour as Lodash’s zip implementation. Ruby is similar - adding nil
values - but stops when the called array’s length is reached. Elixir and Python stop zipping at the length of the shortest list. Maybe an interesting exercise for the reader to figure out an elegant way to emulate this behaviour? :)