Announcing Bouncer, a Validation Library for Clojure Apps
Today I’m releasing bouncer, which was extracted from a project I’ve been working on.
It’s a validation library for Clojure apps and it lets you write code like this:
1 2 3 4 5
If you’d like to see more examples and a detailed guide check out the github repository. The README should get you started.
This post however isn’t only about announcing bouncer. It’s also about the motivation and implementation details behind it.
There are a couple of Clojure validation libraries already out there so why would I write a new one? Well…
Writing Clojure is fun! (who knew? :P)
Because I believe this problem can be solved more elegantly with the use of Monads
If you’ve been following me for a while, you’ll know that I spent most of 2012 deepening my knowledge about functional programming.
In that journey, the unavoidable subject of monads came about - and it was both interesting and enlightening enough that made me write a whole series of posts about it.
After learning what they are and then thinking about the validation problem for a while, I couldn’t help but notice that the problem had a lot in common with the State Monad.
In order to explain how the two relate, I’ll have to digress for a moment. It’ll all make sense in the end - or so I hope
In pure functional languages, such as Haskell, functions can’t have side effects. These include performing IO, changing global variables or launching missiles.
Because of that, functions in Haskell are pure: if you repeatedly call a function f with the same argument
x over time you will always get the same result back.
Pure functions are not a feature of Haskell though. We, too, can write pure functions if we wish:
double above is pure. If we call it with 10, we can be sure the result will always be 20.
This leaves us with a question though: If we were to write our programs with pure functions only, how would we perform computations that need to carry state - state that needs to change over time - around?
A good example of such computation is generating random numbers.
Most programming languages provide generators capable of creating random numbers on demand. Using them is usually trivial. Here’s an example in Java:
1 2 3 4
nextDouble above is obviously not pure. Multiple invocations of it with the same argument - in this case, none - returns different results.
nextDouble is keeping some sort of global state between function calls.
This is where the State Monad comes in. It allows such functions to remain pure.
The State Monad
The State Monad provides a way to abstract state from the function that needs to operate on it.
Sounds confusing? Hopefully an example will clear things up.
Let’s have a look at the clojure function
1 2 3 4
rand suffers from the same problem as
nextDouble we saw above. It keeps it’s own state that is shared across calls, therefore being an impure function.
Now let’s write a pure version of
rand. We’ll call it
1 2 3 4 5 6 7 8 9 10
This is interesting.
pure-rand now takes a generator as an argument and returns a two-element vector containing the random number itself - the result we’re actually interested in - and the generator that was passed in.
Recall however that the generator returned, albeit the same object, is in a new sate.
By re-writing the function like this, we’ve regained purity, as demonstrated below:
1 2 3 4 5
As you can see, as long as we provide the same generator - the same argument - we get the same result back.
If you’re wondering why I’m returning a two element vector from our little function, the answer lies in the State Monad implementation as found in the algo.monads library.
From its docstring:
State Monad: Monad describing stateful computations. The monadic values have the structure:
(fn [old-state] [result new-state]).
It expects a function that receives its old state and returns the result and the new state - these are called monadic values.
By designing the function to follow this contract, we can leverage the
domonad macro - think of it as syntactic sugar for working with monads:
1 2 3 4 5 6 7 8 9 10 11
In the example above, we’re using our
pure-rand function in the context of the State Monad to generate 3 random numbers - based on some initial state - and returning them as a vector.
As we’ve seen before the result is itself in a vector alongside the new state.
This is where the State Monad and the validation problem meet:
In bouncer, each validation function is designed to be compatible with the State Monad, just like
It receives an initial state - at first, the map to be validated - and returns a vector with the map of errors and the new state: the original map augmented with any errors from previous validators.
The end result, should one or more validations fail, is a map with all errors that might have happened, plus our new state.
Now if you head to the github repository and read the examples by keeping the State Monad and the above explanation in mind, the similarities should be obvious.
As I mentioned in the beginning of the article, there are other validation libraries for Clojure and at the time of this writing they have more features than bouncer - by all means have a look at them.
However I will keep maintaining bouncer for a couple of reasons:
- That’s what I’m using in my current side project
- It takes a fundamentally different implementation approach that is in itself worthy of exploration
- If nothing else, this is yet another example of where Monads can be useful.
As usual, let me know your thoughts.