A little Haskell cheat sheet - Functor vs Applicative vs Monad
When learning about Haskell (from the really good book
“Learn You a Haskell for Great Good”
by Miran Lipovača) I came across the four typeclasses from the title -
They all seem to be very cool and powerful, but after reading through all of them it was a bit hard to wrap my mind around it - what is exactly the difference between them, when should I prefer one over another?
This is why I wrote this small cheat sheet - mostly for myself to better remember and understand it, but if it helps somebody else as well even better!
So let’s dig in:
If we have
map for lists and its obvious how useful it is,
why wouldn’t we have the same thing for any other type of data structure, like e.g. tree?
map is for lists,
fmap is for any data structure.
More formally, from Data.Functor Hackage doc page:
Functors: uniform action over a parametrized type, generalizing the
mapfunction on lists.
Functor is for mapping over parametrized types, such as
Maybe, trees, …
Functor typeclass consists of one function only,
fmap works like this:
- input 1:
(a -> b)- function that does the mapping (transformation)
- input 2:
f a- value with a context (“wrapped” value)
f b- transformed value with the same type of context
<$> is just the infix version of
Cool stuff it can do
Mapping over anything, either a predefined data structure or the one you created! How cool is that?
When to use it
When we have a value (e.g.
Just 10) or data
structure (e.g. tree) that provides context along with the data, and we want to transform that
data while preserving the type of context.
When not to use it
With functor, we can not change the type of the given context - e.g. after applying
a transformation the list will still be a list,
Maybe, will still be a
In the process of finding cool examples.
- Show mapping over a function (e.g. parametrized newtype, like Parser in Real World Haskell).
- Explain what does
(+) <$> (+3)do
Sources and additional reading
- LYAH Functor explanation
- Functor design pattern - seems really cool and in-depth but have not read it yet
- Map function in a context to the value in a context
- Can be chained together
- We can apply “normal” function to the values with context - “lifting” a function
What if we wanted to apply “normal” function to the values with the context? E.g. what if we had
Maybe values and wanted to use
(+) on them? Given
Just 3 and
Just 5 we’d expect to
Just 8 as a result, while given
Just 3 and
Nothing we’d expect
This is exactly what
Applicative type class helps us achieve.
There is probably some more motivation but this is what I have for now.
Applicative instances must also be
Functor instances. Given that, with
Applicative we can
apply function in a context to the value in a context.
We can see that
<*> is very similar to
<$>, only that given function also has a context.
is a method that enables us to bring some value in a default context (e.g. a “normal” function we
want to use on values with the context).
Here we can see how we “lifted” a function
(+) to work with
Maybe values. We have to start with
<$> first since we are starting with a “normal” function without context, so
(+) <$> Just 3
Just (3+) (or we could have started with
pure (+) <*> Just 3).
After that we can just add all the other arguments by chaining them with
I am not sure how well did I get the motivation, but here is what I have for now: when dealing with values with a context, monad is the most general tool for it. When we have values with context and they are flowing from one method to another, somebody always has to “handle” the context.
In order to avoid doing it manually (which would increase the complexity a lot, and its a repetitive process), the monads step in - with them, we can “pretend” we are dealing with “normal” functions while they appropriately handle the context for us.
When dealing with values with context,
Monad typeclass helps us by automatically handling the
context for us.
returnis just like
Applicative, putting value in a default, minimal context.
(>>=), pronounced “bind”, is the main
Monadfunction. It allows us to “feed” a value with a context into a function that takes a pure value but returns also a value with a context.
(>>)is used when we want to “execute” one monad before another, without actually caring for its result (because it probably changes some state, has some side-effect).
fail- apparently never used explicitly in code, but by Haskell for some special cases. Did not see it in action so far.
- Maybe example -> Show how would handling context look like without monad
- Parsec example
- IO example
I could write about the situations that drove the inception of the above concepts. E.g.:
- First we just had functions that returned a single, simple value.
- But soon, we also needed functions that returned multiple different values (e.g. because they can fail)
- The values with context were created to address that need.
- But then we needed to handle such types of values throughout the code -> we needed new tools to help us with that.