# Monad Transformers - a Quick Recap

Someone have said that monads are like burrito, if you ever taste one than you can’t imagine live without it.

Monads are a powerful tool. Thanks to them we can abstract over computation. We can make one computation depended on another and if needed fail fast.

But one day the time will come when we have two different monads and we will find out that they don’t compose !

Let’s make some code to visualize the problem. I am going to show two use cases and I will start with the simplest one.

### Case 1

We have two entities : User and Address and two functions retrieving data with the respect of a given predicate

Our goal is to write a function which for a given login returns user’s street name

So far so good - quite simple and classic enterprise task :)

However there are two caveats to this solution worth noting. What happened if there is no such user or the user exists but it has no address ? It is obvious that we will see Null Pointer Exception - sick !

Of course we can filter out those nulls and rewrite functions to be aware of them but as you already know this is also not a good solution. Can we do better ? Yes we can, let’s introduce a context aware of whether value exists or not (Option data type).

But wait below function is not compiling…

It turns out that Future and Option monads do not compose in such a way. For a first look, composition looks very natural in for comprehension, but if we transform it into series of flatMap and map at the end, we will notice that the puzzles don’t feet. If we start with Future than the function passed to flatMap must return a Future. In our case we want to return Option in the middle and based on it return a next Future being a container fo an user’s possible address.

Equipped with this knowledge we can rewrite our function in the following way

Now it compiles and return correct results. But it is not as readable as our first naive attempt. Can we do better ? Ideally we would want to have something like

We need somehow to fuse Future with Option in a smart way to make the composition possible.

#### Fusing Future with Option

We already know that for comprehension deals with flatMap, map, withFilter and foreach. In our case compiler needs only flaMap and map to de sugar for. So let’s introduce a new data type OptionFuture, which wraps Future[Option[A]] and in a proper way handles flatMap in order to compose Future with Option.

Take a little time to better look at OptionFuture data type. First question coming to my mind is - can we make it more abstract ? It turns out that we can abstract over Future very easly. In terms of Future we are calling only two kinds of functions:

• flatMap
• Future.successful

It means that Future can be swapped with Monad.

What about the Option ? Over the Option we are performing pattern matching - so it means that we need to know something about it structure.

And because of that we can’t to abstract over it.

This leads us to the definition of monad transformer for Option and we call it OptionT

#### Monad transformer for Option

OptionT[F[_], A] abstracts over F and A and it only requires that F is a monad. The name of the monad transformer comes from the fact that, in order to implement this wrapper, we need to know what the inner most monad in the stack is - in this case Option. Without this knowledge we can’t compose any two given monads with itself.

A minimal api for monad can be described by following trait

And its instance for Future you can find below.

#### A solution

Putting all pieces together we can finally write

of course we can return directly Future[Option[String]] just by calling value function on the result like

` and that’s it.

#### Final word

At the beginning I said that I have two cases to show, but because the post could be to long to go through without a brake I decided to split it into two pieces. The whole code base used in this post can be found in the following link

More in part 2