If you are a dynamically typed programmer you probably are not super familiar with type signatures. (Perhaps even if you are an avid TypeScript user!)
In the following example, I will try to demonstrate with a practical example why we would like to use type signatures. The example is in Haskell, but the syntax is easy enough.
Example: the motion of a particle
Suppose we have a particle moving over a straight line. Let's say its position at time $t$ is given by the function $x(t)$.
Consider now an interval of time \(\Delta t = t_1 - t_0\) for an initial time \(t_0\). Then the average velocity of the particle over the interval of time is given by the function:
$$v_{t_0}{(t_1)}=\frac{x(t_1)-x(t_0)}{t_1 - t_0}$$
Let's see how we can express this in Haskell.
averageVelocity::Time->Time->PositionFunction->Velocity
averageVelocity t0 t1 x = (x t1 - x t0) / (t1 - t0)
Note that x is a function of some PositionFunction type. And is being given as an argument to averageVelocity.
But the type of x could also be:
x::Time->Positon
Therefore, averageVelocity may have the type signature
averageVelocity::Time->Time->(Time->Position)->Velocity
But note that "time", "position" and "velocity" are just numbers. Inside the computer, they are just float or double. So why would we not instead do this:
averageVelocity::float->float->(float->float)->float
Type synonyms
Furthermore, in Haskell we can define type synonyms:
type R = float
type Time = R
type Position = R
type Velocity = R
type PositionFunction = Time->Position
averageVelocity::Time->Time->PositionFunction->Velocity
This makes the code much clearer. And whenever there is an error, the compiler will tell us something about the type involved.
The meaning of function signatures
Let's try something more interesting. Usually, when modeling the motion of a particle we are interested in its instantaneous velocity. Not the average over an interval. We need to model the derivative of the position function for this.
$$\frac{dx(t)}{dt}=v(t)= \lim_{\Delta t \rightarrow 0} \frac{x(t+\Delta t/2)-x(t-\Delta t/2)}{\Delta t}$$
Note that the left side of the equation takes as input for the derivative operator. Which is a function. We may write \(D(x(t)) = v(t)\) if we want to be more explicit.
If you are keen-eyed you may have noticed that I treated operators as if they were functions. That is because they are! As an example, the "addition" operator \(+\) is just a function \(+: \mathbb{R} \times \mathbb{R} \rightarrow \mathbb{R}\). So for two numbers $a, b$ their sum is given by \(a+b=c\) which is just syntactic sugar for \(+(a,b) = c\). Because \(+\) is just a function!
Let's attempt to write the type signature for a derivative in Haskell!
type Derivative = (R->R)->R->R
We are expressing that the type Derivative is a function type, for a function that takes a function with signature R->R as input and a function with signature R->R as output.
However, look at the following:
evaluateDerivative::R->Derivative
evaluateDerivative dt x t = (x (t+dt/2) - x (t+dt/2)) / dt
This may seem wrong to you. Isn't evaluateDerivative supposed to take a single value of type R and return a function of type Derivative?
Yes and No. We can rewrite the signature as:
evaluateDerivative::R->R->R->R->R
This way, believe it or not, we are telling the compiler that whatever fits that signature is a valid parameter. Functions and values are the same kind of thing in Haskell.
So evaluateDerivative as we defined above takes a very small interval dt (in place for taking the limit), a function x::R->R, and a time value t. And returns the result of evaluating \(\frac{dx(t)}{dt}\).
Yet. If this is so, then we have at least three functions with quite different meanings:
differential One that takes a single interval \(\Delta t\) of type R as input and outputs a function $D$ with signature (R->R)->(R->R) which takes a position function $x(t)$ and outputs its derivative $v(t)$. That is \(D(x) = \frac{dx}{dt}\). This is a "generic" differentiation function.
evaluateDerivative A function that takes three arguments \(\Delta t\),$x(t)$, \(t_i\) and outputs the result of evaluating the derivative of $x$ at time \(t_i\). That is, \(v(t_i)\), a single numeric value that has type R.
getDerivative A function that takes two arguments \(\Delta t\) and $x(t)$ of type R and R->R, respectively. And returns the derivative of $x$, with type R->R.
All three of them have the exact same code. The only difference is whether we use the type signature and name to remind us of what they do.
You may now be wondering how on earth can three different functions have the exact same code? The Haskell compiler can by default curry a function when not all parameters are provided. That is, it does partial application whenever needed.
So...?
They are all the same. We get all of that versatility just from the types (and partial application).
Keep in mind that code, functions, objects, variables and so are just data to the computer. The difference is only how we think about it.
Having a type system allows you to think about your programs in terms of higher abstractions.
That may (or may not) be what you want or need... As with anything in software, it depends. There are always tradeoffs!
:D
