24 Days of Hackage: pipes

Today’s article is going to be an interesting one… not just because the library that I’ll look at - pipes - is interesting, but it’s also very new to me! However, we’ll see that pipes is a library providing functionality that can otherwise be difficult, providing an API that even us newbies can understand. With that subtle disclaimer out of the way, let’s get going!

Long running input/output in Haskell is unfortunately a hard problem. We’re used to writing code that is modular and composes elegantly, but naïve approaches to dealing with IO quickly turn ugly - producing code that really doesn’t compose at all well. Somewhat even more dishearteningly, if you try and solve these problems you quickly come across ‘enumerators’, ‘iterators’, ‘iteratees’ and all sorts of other concepts. Unless you’re willing to spend some time with these theories, most people are likely left feeling a little deflated, not to mention stuck. This really doesn’t help in a work environment when you’ve got deadlines!

Now there are already a number of solutions on Hackage, but today we’re going to look at pipes, as I feel the theory behind it is extremely compelling, and the provided API is something that doesn’t require hours of academic reading to get results from.

A Christmas List Manager

To dive right in, lets first understand the underlying idea of the library. As a user, you will: build Producers of values, the input to your computation; Consumers that will take values and do something with them, and Pipes which allow you to transform from one form of data to another. Once you’ve built these components, you then compose things together to form a Session, which can be ran to produce your actual application.

The “hello, world” of these libraries is some form of echo, but lets add a Christmas twist. We’re going to build an application which requests a name, and a list of Christmas presents, until the user submits an empty line. Then it will echo all of this back. The first thing we need is a something that produces the name:

name :: Proxy p => () -> Producer p String IO ()
name () = runIdentityP $ do
  lift $ putStr "Ho ho ho! What is your name? "
  lift getLine >>= respond

Our Producer prints out a prompt, and then responds with whatever the user entered. Simple! We can test this in GHCi:

> runProxy $ name
Ho ho ho! What is your name? Oliver

Nothing was returned though, because we didn’t connect a Consumer. A trivial Consumer is provided by pipes in the form of printD. We attach a Consumer to a Producer with the >-> composition operator:

> runProxy $ name >-> printD
Ho ho ho! What is your name? Oliver
"Oliver"

Excellent, printD consumed the String produced by name and then printed it back out. Now, how about a Producer for that stream of presents?

data Present = Present String
  deriving (Show)

presents :: Proxy p => () -> Producer p Present IO ()
presents () = runIdentityP $
    lift (putStrLn "And what presents would you like?") >> go ()
  where
    go = getLineS >-> takeWhileD (not . null) >-> mapD Present

Before we look at what’s going on here, what happens if we try and run this?

Main> runProxy $ presents >-> printD
And what presents would you like?
GameBoy
Present "GameBoy"
Nintendo 64
Present "Nintendo 64"
A Pony
Present "A Pony"

Main>

Well that’s interesting! Everytime we entered a present, we produced a new Present value, which immediately got sent to printD! This shows one of the fundamental principles of pipes - you are working with streams of data - something that I think is a very natural way of programming.

Now, what’s going on in this Producer? Well, first of all we output a basic prompt, and then produce a list of Presents. We do this by first reading a line, and constantly consuming these lines until we encounter one that is empty. We map over all lines that we read, turning them into Present. Again, nothing that you wouldn’t write in Haskell otherwise; pipes doesn’t require you to relearn these basic ideas - they simply extend nicely to pipes.

Now that we have all of this, lets compose everything together to build our final application:

main :: IO ()
main = do
  First (Just name) <- execWriterT $ runProxy $
    raiseK name >-> headD_
  presents <- execWriterT $ runProxy $
    raiseK presents >-> toListD
  putStrLn (name ++ " wants: " ++ show presents)

I’ve used the headD_ utility to take the first value from the name Producer, and then used toListD to consume everything in the presents Producer and convert it to a list. I then output this using putStrLn and show, as normal.

This might seem awfully verbose for what we’ve achieved, and for this example, it is. But even so, the code seems very elegant to me as we’ve split things apart into small, reasonable, pieces of code. For example, we might want to store all of this in a database as it’s entered - every time a present is added it pass through a storeInDatabase pipe, which would store each Present the moment it’s entered. Or we might later expand what a Present consists of, and now we only have to change our Present Producer.

Echo (echo, echo)

Gabriel Gonzalez handed me a little example for showing off pipes, in the form of a pair of applications - a client and a server. This shows how pipes can be used with some slightly more interesting (less contrived!) IO:

server = withSocketsDo $ do
    s <- listenOn (PortNumber 5553)
    (h, _, _) <- accept s
    hSetBuffering h LineBuffering
    runProxy $ hGetLineS h >-> putStrLnD
    hClose h

The server simply opens a socket and listens port 5553, and then consumes lines as they are sent in. The client is equally elegant:

client = withSocketsDo $ do
    h <- connectTo "localhost" (PortNumber 5553)
    hSetBuffering h LineBuffering
    runProxy $ getLineS >-> takeWhileD (/= "quit") >-> hPutStrLnD h
    hClose h

The client connects to the server, and then consumes user input, until the user types “quit”. For every line typed, it is sent directly to the server.

The Rest

As you can see, pipes really doesn’t require you to learn a whole lot of stuff - once you get to grips with runProxy, and composition via >-> and >=>, you are already able to do a lot of stuff. Gabriel has done a fantastic job with documentation on this library too, a practice I really hope other authors will follow. I highly recommend reading the tutorial, which will also show you how pipes can be used for:

pipes is a really nice library, and while it’s still not quite there, it’s well on the way to being a great solution to framing IO computations in Haskell. Thanks Gabriel, and good luck with your future work!


You can contact me via email at ollie@ocharles.org.uk or tweet to me @acid2. I share almost all of my work at GitHub. This post is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.