Interactive code snippets not yet available for SoH 2.0, see our Status of of School of Haskell 2.0 blog post

Episode 4 - Recipe Puppy


This is part of The Pragmatic Haskeller series.

Speaking to the world

Now we have our webapp that can read json from the outside world and store them inside MongoDB. But during my daily job what I usually need to do is to talk to some REST service and get, manipulate and store some arbitrary JSON. Fortunately for us, Haskell and its rich, high-quality libraries ecosystem makes the process a breeze.

The "hard" way

Remember what I told you about this series? We have to reason as pragmatic programmers, choosing the tools which seem more appropriate for the task at stake. During my initial exploration of the library space, I landed on HTTP. It makes the process of making GET requests as simple as:

module Main where

import Network.HTTP
import Control.Applicative

get :: String -> IO String
get url = do
    response <- simpleHTTP $ getRequest url
    getResponseBody response

main = take 204 <$> get "" >>= print

Alas, HTTP does not support SSL out of the box (a quick glimpse to the dependency list for this library would be enough):

module Main where

import Network.HTTP
import Control.Applicative

get :: String -> IO String
get url = do
    response <- simpleHTTP $ getRequest url
    getResponseBody response

-- show
main = take 20 <$> get "" >>= print
-- /show

Now, here is where pragmatism comes into play; if you are planning to do just "plain" http requests and don't need SSL at all, you can stick with HTTP and live long and happy. In case you have more demanding use case scenarios, you have two options:

  1. Use an Haskell library like HsOpenSSL or tsl to integrate HTTP with SSL.

  2. Use a library which supports SSL out of the box, like http-conduit

Fred asks: Why not use something like http-streams?

Answer: even though I'm a big fan of io-streams per se, I decided not to use http-streams. It looks very promising, but I've found one little wart: it depends on HsOpenSSL. This brings two drawbacks on the table:

  1. HsOpenSSL's author "wants to encourage you to use and improve the tls package instead as long as possible. The only problem is that the tls package has not received as much review as OpenSSL from cryptography specialists yet, thus we can't assume it's secure enough."

  2. Using HsOpenSSL makes the process of writing secured http connection a bit too verbose, in my humble opinion. This is an excerpt from http-streams documentation:

import OpenSSL (withOpenSSL)

main :: IO ()
main = withOpenSSL $ do
    ctx <- baselineContextSSL
    c <- openConnectionSSL ctx "" 443
    closeConnection c

As we'll see, using http-conduit will make the process a no-brainer. Said that, don't take my words as a judgment over the library quality. I like it very much and I hope to see a bit of boilerplate scrapped in the future.


This library took some design choices for you, using tsl under the hood. It's extremely easy to use, so easy that I'll include here an extra snippet, non included in the Github lesson, but that I hope will be useful as a real world example. After all, I couldn't be in peace of mind publishing an episode which claims to be some sort of "solution for real world problems" if I was limiting the scope to GET requests. Hardly in production you'll do just plain GET requests.

Getting some JSON from Foursquare

Imagine the scenario; you need to fetch some JSON from Foursquare (which talks through SSL) and manipulate this JSON to extract some information. To be honest, during the first episode I didn't tell you all the truth about aeson. There are two features which are simply outstanding:

  • Generic derivation. Using GHC's Generics you can let aeson derive automatically the marshalling/unmarshalling boilerplate code. Try this at home:

{-# LANGUAGE DeriveGeneric #-}
{-# LANGUAGE OverloadedStrings #-}
module Main where

import Data.Aeson
import qualified Data.ByteString.Lazy.Char8 as BL

import GHC.Generics (Generic)

data Person = Person { name :: String,
                       age  :: Int} deriving (Show, Generic)

instance FromJSON Person
instance ToJSON Person

rawData = BL.pack "{\"name\": \"John\", \"age\": 35}"
main = print $ (decode rawData  :: Maybe Person)

Pretty impressive, isn't it?

  • Manipulation of the "raw" AST. As written in aeson's documentation, "sometimes you want to work with JSON data directly, without first converting it to a custom data type. The Value type, which is an instance of FromJSON, is used to represent an arbitrary JSON AST (abstract syntax tree)."

We can use this awesome functionality to extract the number of checkins for a FS' venue, with the following code:

{-# LANGUAGE OverloadedStrings #-}
module Main where

import Data.Monoid
import Data.Aeson
import Network.HTTP.Conduit

type FsVenueId = String

apiUrl :: String
apiUrl = ""

requestUrl :: String
requestUrl = "?oauth_token=" <>

requestBuilder :: FsVenueId -> String
requestBuilder vid = apiUrl <> vid <> requestUrl

getVenue :: FsVenueId -> IO (Maybe Value)
getVenue vid = do
    rawJson <- simpleHttp $ requestBuilder vid
    return (decode rawJson :: Maybe Value)

main = do
  response <- getVenue "40a55d80f964a52020f31ee3"
  case response of
    (Just v) -> print . take 400 . show $ v
    Nothing  -> print "Failed to fetch venue."

Et voilà! Now we have a pretty generic data structure, and the task to write a simple function to find the field "checkinsCount" inside the Object "stat" is left as an exercise for the reader. Highlighted you can see how easy it was to talk to FS using SSL; yes, just a one liner, which the function simpleHttp.

Calling Recipe Puppy from our webapp

Recipe Puppy is a nice REST service I've found googling for a web service with minimal authentication overhead and, of course, cooking-related. Let's suppose we want to retrive recipes based on one ingredient contained within, well, it turns out that all it takes is to call this REST link:

Find recipes which contain onion

All we need, then, is a small wrapper which allows us to call our underlying web service. We'll be using Recipe Puppy for its simplicity, but at this point you should know we can make complex calls thanks to HTTP or http-conduit. The choice I've made was to segregate our routes into a separate file, and then appending the routes back to the main routes function (inside Site.hs):

{-# LANGUAGE OverloadedStrings #-}

module Pragmatic.Server.RecipePuppy where

import Pragmatic.Server.Application
import Data.ByteString
import Snap hiding (get)
import Data.Monoid
import Data.Text (Text)
import qualified Data.Text as T
import qualified Data.Text.Encoding as T
import qualified Network.HTTP as H

apiUrl :: Text
apiUrl = ""

puppyRoutes :: [(ByteString, AppHandler ())]
puppyRoutes = [("/puppy/search/:ingredient", searchByIngredient)]

get :: String -> IO String
get url = do
    response <- H.simpleHTTP $ H.getRequest url
    H.getResponseBody response

searchByIngredient :: AppHandler ()
searchByIngredient = do
    i <- getParam "ingredient"
    let ingredient = maybe "" T.decodeUtf8 i
    output <- liftIO $ get (T.unpack $ apiUrl <> "?i=" <> ingredient)
    writeText (T.pack output)

Just some comments:

  • I used HTTP, but http-conduit would have been a good choice too
  • I've used the maybe function to yield an empty string in case i was Nothing, or applying the function decodeUtf8 in case it was not.

Done! Now if we append the puppyRoutes list to the "main" one inside Site.hs, we can navigate to http://localhost:8000/puppy/search/garlic and see some JSON!

External References

Refer to the official documentations, as always.

The code

Grab the code here. The example is self contained, just cabal-install it!

Next Time

It's time to build our small DSL for describing recipes! Stay tuned!