Once, when doing code review on a TypeScript project in my work, I came across several instances when a colleague of mine used a
for loop, even though it wasn’t necessary (i.e. a more readable declarative equivalent was available). In a joke I stated that we should impose a “for-loop tax” for every loop used unnecessarily.
It made me think however, why so many people tend to go for the longer and more error prone solution with the loop and I got to the following conclusion: Nearly every (mainly) imperative programming language course/book I ever took/read (be it Pascal and C# on high school or C/C++ and Wolfram Mathematica in college) contained a section like
“These are the loops available, this is how you write them and this is how you use them to solve these basic problems”.
We start with the simplest of problems:
Given an array of numbers, return the sum of its elements.
If you attended similar courses as me, you surely recognise this code. Create a temporary variable, initialise it with zero and using a for loop iterate over the array returning the final value of the variable. There are some problems though:
ithat is strictly less than the length of the array (not less than or equal). This is prone to errors especially if your work in many languages at the same time.
The solution that remedies both those problems is to use the reduce function (in other languages also called fold or aggregate). In a single expression we iterate over each of the array elements adding them together (stating the sum’s default and initial value is zero). Notice there is no mention of the iteration bounds, it just guarantees that it will go over all the elements from first to last.
This is another common problem, let’s illustrate it with this example:
Given the array of prices, return a new array with the prices n % lower.
The loop-based way to do this is pretty similar to the sum code. There is one additional problem though: in order to not destroy the input array, we must clone it first and then transform the values in the new array. This can easily be forgotten introducing a potentially unwanted side effect in the application.
The cloning problem can be avoided altogether using the map function. For a given array it returns a new array where each element is the corresponding element in the original array transformed using the provided function (in our case multiplied by the discount multiplier).
Another common situation where loops are used is when generating linear ranges as an input for further transformations. A classic example is:
Return an array of the first n squares
This is a problem that can be solved very badly when using loops. The first naïve solution suffers from the problem that it pushes a new element to an array every iteration. This expands the array and may cause it to reallocate in memory being slow (benchmark).
The second approach instantiates the array of correct size beforehand avoiding this problem, but we can easily make a mistake when assigning the current value (see the
result[i — 1]expression on line 12).
The final use case of loop I want to discus here is invoking a method with side effects more than once. As Edsger Dijkstra famously said:
Two or more, use a for
The simplest example to illustrate this case is:
Console.log the string “Hello world” n times
However, I would strongly recommend to abstract this into a helper function to restrict the loop to a single place. As we can see in the examples to the left (both calling the provided function with numbers from 1 to n), the iterative version is shorter and simpler to write. Also the “loop-free” version would cause a stack overflow on environments without Tail Call Optimisation.
On four elementary situations, we described how to use declarative style to replace loops and therefore make our code shorter and less error-prone.
Do you use loops? Do you disagree with any of the solutions? Comment please!