In computer science, a side-effect is a property of a programming language function that it modifies some state other than its return value. For example, a function might modify a global or "static" variable, modify one of its arguments, write data to a display or file, or read some data from other side-effecting functions. Side-effects often make a program's behavior more difficult to understand.

Imperative programming is known for employing side effect to make programs function. Functional programming in turn is known for its minimization of side effects.

Terminology

A function that uses side-effects is referred to as referentially opaque, and one that doesn't is called referentially transparent. For simplicity's sake, we say that a referentially transparent function is one that, given the same parameters, will always return the same result.

Example

As an example, let's use two functions, one which is referentially opaque, and the other which is referentially transparent:

 globalValue = 0;
 integer function rq(integer x)
 begin
   globalValue = globalValue + 1;
   return x + globalValue;
 end

integer function rt(integer x) begin return x + 1; end

Now, rt is referentially transparent, which means that rt(x) = rt(x) as long as x is the same value. For instance, rt(6) = rt(6) = 7, rt(4) = rt(3+1) = 5, and so on. However, we can't say any such thing for rq because it uses a global value which it modifies.

So, how is this a bad thing? Well let's say we want to do some reasoning about the following chunk of code:

 integer p = rq(x) + rq(y) * (rq(x) - rq(x));

Now, right off-hand, one would be tempted to simplify this line of code to:

 integer p = rq(x) + rq(y) * (0) = 
 integer p = rq(x) + 0           = 
 integer p = rq(x);

However, this will not work for rq() because rq(x) does not equal rq(x)! Remember, that the return value of rq is based on a global value which isn't passed in and which gets modified all over the place. This goes against common sense since anything minus itself should be zero.

This however will work for rt, because it is a referentially transparent function.

Therefore we can reason about our code which will lead to more robust programs, the possibility of finding bugs that we couldn't hope to find by testing, and even the possibility of seeing opportunities for optimization.