┌───────┐

│ Okapi │

└───────┘

Okapi is a micro framework for HTTP servers.

Okapi is NOT recommended for production use at this time.

Please feel free to reach out to me on GitHub if you have any questions/criticisms/ideas or would like to contribute to this project.


Getting Started

  1. Use the command stack new <project-name> to create a new Haskell project
  2. Add the okapi library to your project's dependencies

Introduction

There are two ways to implement Servers in Okapi.

The recommended way to implement a Server in Okapi is via Endpoints:

-- | Define Endpoints using an Applicative eDSL
myEndpoint = Endpoint
  { method = GET
  , path = do
      Path.static "index"
      magicNumber <- Path.param @Int
      pure magicNumber
  , query = do
      x <- Query.param @Int "x"
      y <- Query.option 10 $ Query.param @Int "y"
      pure (x, y)
  , body = do
      foo <- Body.json @Value
      pure foo
  , headers = pure ()
  , responder = do
      itsOk <- Responder.json @Int status200 do
        addSecretNumber <- AddHeader.using @Int "X-SECRET"
        pure addSecretNumber
      pure itsOk
  }

An alternative, more concise way of defining a Server in Okapi is via Matchpoints:

-- | Define Matchpoint patterns using PatternSynonyms,
--   ViewPatterns, and the same eDSL used for Endpoints
pattern GetUsers optF <- Matchpoint
  GET
  ["users"]
  _
  (Body.eval $ Body.optional $ Body.json @Filter -> Ok optF)
  _

pattern AddUser user <- Matchpoint
  POST                                     -- Method
  ["users"]                                -- Path
  _                                        -- Query
  (Body.eval $ Body.json @User -> Ok user) -- Body
  _                                        -- Headers

pattern GetUsersByID userID <- Matchpoint
  GET
  (Path.eval $ Path.static "users" *> Path.param @UserID -> Ok userID)
  _
  ""
  _

-- | Servers are just contextful functions from a Request to a Response
type Server m = Request -> m Response

myServer :: Server IO
myServer = \case
  GetUser -> do
    ...
  GetUserByID userID -> do
    ...
  AddUser user -> do
    ...
  _ -> do
    ...

-- | Run your Server using Warp
main = Warp.run 3000 $ instantiate id myServer

The advantadge of using Endpoints over Matchpoints is that Okapi can automatically generate Specifications and Clients for a Server implemented with Endpoints, but not a Server implemented with Matchpoints.

On the flip side, a Server implemented with Matchpoints will be more concise than the same Server implemented with Endpoints.


Endpoint

An Endpoint is an executable specification representing a single Operation that can be taken against your API.

An Endpoint has 6 fields.

data Endpoint p q h b r = Endpoint
  { method :: StdMethod             -- (1)
  , path :: Path.Script p           -- (2)
  , query :: Query.Script q         -- (3)
  , body :: Body.Script b           -- (4)
  , headers :: Headers.Script h     -- (5)
  , responder :: Responder.Script r -- (6)
  }

The method field is a simple value, but the other fields point to what's called a Script. Scripts represent Okapi's DSL for extracting and parsing data from Requests. There's a specific type of Script for each part of a Request.

The type parameter of a Script represents the type of value it returns.

Therefore, the concrete type of an Endpoint is determined by the return types of the Scripts that are used to construct the Endpoint.

All the different Script types are Applicatives, but not all are lawful Applicatives.

Since all Script types are Applicatives, we can use the Applicative typeclass methods to write our Scripts. Here's an example of a Query Script.

data Filter = Filter
  { color :: Text
  , categoryID :: Int
  , isOnSale :: Maybe ()
  }

myQueryScript :: Query.Script Filter
myQueryScript = Filter
  <$> Query.param "color"
  <*> Query.param "category"
  <*> (Query.optional $ Query.flag "sale")

If you have the -XApplicativeDo language extension turned on, you can also write your Scripts using do syntax.

We recommend using -XApplicativeDo in conjuction with the -XRecordWildCards language extension if you're not comfortable with using the Applicative operators. Here's the same Query Script we defined above, but with these language extensions turned on.

{-# LANGUAGE ApplicativeDo #-}
{-# LANGUAGE RecordWildCards #-}

myQueryScript :: Query.Script Filter
myQueryScript = do
  color      <- Query.param "color"
  categoryID <- Query.param "category"
  isOnSale   <- Query.optional $ Query.flag "sale"
  pure Filter {..}

Each Script type has its own operations suited to parsing its respective part of the Request. These operations are covered in more detail below.

  1. Method

    The method field represents the HTTP Method that the Endpoint accepts.

    Its type is StdMethod from the http-types library.

    HTTP standard method (as defined by RFC 2616, and PATCH which is defined by RFC 5789).

  2. Path Script

    The path field defines the Request Path that the Endpoint accepts, including Path parameters.

    Path Scripts have two operations: static and param.

    myPathScript = Path.static "person" *> Path.param @Int "personID"
  3. Query Script

    The query field defines the Query that the Endpoint accepts.

    There are two operations for Query Scripts: param and flag.

    There are two modifiers for Query Scripts: optional and option.

    optional and option are specialized versions of the optional and option parser combinators found in the parser-combinators library.

    myQueryScript :: Query.Script (Text, Maybe Float, Int)
    myQueryScript = do
      lastName  <- Query.param "last_name"
      optSalary <- Query.optional $ Query.param "salary"
      minAge    <- Query.option 21 $ Query.param "min_age"
      pure (lastName, optSalary, minAge)
  4. Body Script

    The body field defines the Request Body that the Endpoint accepts.

    There are four operations for Body Scripts: json, form, param, file.

    There are two modifiers for Body Scripts: optional and option.

    myBodyScript :: Body.Script (Maybe Value)
    myBodyScript = Body.optional $ Body.json @Value
  5. Headers Script

    The headers field defines the Request Headers that the Endpoint accepts.

    There are two operations for Headers Scripts: param and cookie.

    There are two modifiers for Headers Scripts: optional and option.

    myHeadersScript :: Headers.Script _
    myHeadersScript = undefined
  6. Responder Script

    The responder field defines the Responses that the Endpoint's handler MUST return.

    Responder Scripts have to be more complex than the other Script types in order for the Endpoint to have a contract with its Handler. The contract ensures that the Handler will respond with the Responses defined in the Responder Script.

    This is done using a combination of higher order functions, linear types, and smart constructors.

    Responder Script operations have to take an Add Header Script as an argument to define what Headers will be attached to the Response.

    For now, there is only one operation for Responder Scripts: json.

    Add Header Scripts only have one operation as well: using.

    {-# LANGUAGE ApplicativeDo #-}
    {-# LANGUAGE RecordWildCards #-}
    {-# LANGUAGE BlockArguments #-}
    {-# LANGUAGE LinearTypes #-}
    
    data SecretHeaders = SecretHeaders
      { firstSecret :: Int -> Response -> Response
      , secondSecret :: Int -> Response -> Response
      }
    
    data MyResponders = MyResponders
      { allGood :: (SecretHeaders %1 -> Response -> Response) -> Text -> Response
      , notGood :: (() %1 -> Response -> Response) -> Text -> Response
      }
    
    myResponderScript = do
      allGood <- Responder.json @Text status200 do
        addSecret <- AddHeader.using @Int "IntSecret"
        addAnotherSecret <- AddHeader.using @Int "X-Another-Secret"
        pure SecretHeaders {..}
      notGood <- Responder.json @Text status501 $ pure ()
      pure MyResponders {..}
    
    myHandler someNumber _ _ _ _ (MyResponders allGood notGood) = do
      if someNumber < 100
        then return $ allGood
          (\(SecretHeaders firstSecret secondSecret) response -> secondSecret 0 $ firstSecret 7 response)
          "All Good!"
        else return $ notGood
          (\() response -> response)
          "Not Good!"

    More information about Responder and AddHeader are available in the Handler section.

Handler

Handlers are simple: they are contextful functions from the arguments provided by an Endpoint, to a Response.

The type synonym Handler represents the type of these contextful functions.

type Handler m p q b h r = p -> q -> b -> h -> r -> m Response

The type parameter m represents the context in which the Handler creates the Response.

The type parameters p, q, b, h and r represent the types of the values returned by the Endpoint's Path, Query, Body, Headers and Responder Scripts respectively.

Plan

A Plan is how your Endpoint and its designated Handler come together.

data Plan m p q h b r = Plan
  { transformer :: m ~> IO
  , endpoint :: Endpoint p q h b r
  , handler :: Monad m => p -> q -> b -> h -> r -> m Response
  }

The transformer field represents a natural transformation from your Handler's Monad m, to IO. This is where you decide how your custom effects are interpreted in an IO context.

The endpoint field represents your Endpoint.

The handler field represents your Handler. The types must match the types of your endpoint and transformer.

Here's an example of a Plan:

myPlan = Plan
  { transformer = id
  , endpoint = myEndpoint
  , handler = myHandler
  }

myEndpoint = Endpoint
  { method = GET
  , path = do
      Path.static "index"
      magicNumber <- Path.param @Int
      pure magicNumber
  , query = do
      x <- Query.param @Int "x"
      y <- Query.option 10 $ Query.param @Int "y"
      pure (x, y)
  , body = do
      foo <- Body.json @Value
      pure foo
  , headers = pure ()
  , responder = do
      itsOk <- Responder.json @Int HTTP.status200 do
        addSecretNumber <- AddHeader.using @Int "X-SECRET"
        pure addSecretNumber
      pure itsOk
  }

myHandler magicNumber (x, y) foo () responder = do
  let newNumber = magicNumber + x * y
  print newNumber
  print foo
  return $ responder (\addHeader response -> addHeader (newNumber * 100) response) newNumber

The Server you build will be a combination of many Plans.

Server

A Server is the final type of value that you need to generate an Application or Specification.

data Server = Server
  { info :: Maybe Info
  , defaultResponse :: WAI.Response
  , artifacts :: [Artifact]
  }

The info field represents your Server's metadata. It's used in the generation of the Server's Specification. It's optional.

The artifacts field is a list of Artifact. A single Artifact is generated from a single Plan. An Artifact contains two values:

  1. An IO action that returns a Response. It's only executed if the Endpoint used to generate the IO action matches the Request.
  2. An OpenAPI PathItem value based on the structure of the Endpoint used to build the Artifact.

These two values, when combined with the Server's other Artifacts, are used to generate the final Application and OpenAPI Specification.

build ::
  forall m p q h b r.
  Monad m =>
  Plan m p q h b r ->
  Artifact
build = ...

plan1 :: Plan A B C D E F
plan1 = ...

plan2 :: Plan G X Z U I P
plan2 = ...

plan3 :: Plan Y T L G N Q
plan3 = ...

myServer = Server
  { info = Nothing
  , defaultResponse = default404
  , artifacts =
      [ build plan1
      , build plan2
      , build plan3
      , ...
      ]
  }

The types of the Plans used to build your Server don't have to be the same. The build function erases the types and gives us the end products we need. This allows us to mix and match various combinations of Endpoints, Handlers, and Monad transformations in the same Server definition. For example, you can have two Handlers that operate in two different Monads in the same Server.

Now that you have you your Server, you can use it to:

  1. Generate a WAI Application
  2. Generate an OpenAPI Specification
myServer :: Server
myServer = ...

api :: Application
api = genWaiApplication myServer

apiSpec :: OpenApi
apiSpec = genOpenAPISpec myServer

In the future, you should be able to automatically generate API clients as well.

Tips & Ideas

Not Using A Plan

You can also create Servers with out first creating Plans. If you want to do this, you can just use the buildWith function directly.

buildWith ::
  forall m p q h b r.
  Monad m =>
  (m ~> IO) ->
  Endpoint p q h b r ->
  Handler m p q h b r
  Artifact
buildWith transformer endpoint handler = ...

Assuming all of your handlers for the Server will run in the same context, you can just partially apply buildWith to a transformation function and use the partially applied function on your Endpoints and Handlers to produce Artifacts.

buildWithIO = buildWith id

myServer = Server
  { info = Nothing
  , defaultResponse = default404
  , artifacts =
      [ buildWithIO endpoint1 handler1
      , buildWithIO endpoint2 handler2
      , buildWithIO endpoint3 \x y z a b -> do
          doSomethingWith x
          log a y b
          ...
      ]
  }

DRY Endpoints

When implementing an API you will usually need the same path to have multiple methods, each with different parameters in the query, body and headers. Since Endpoints are records, this is easy to deal with. Let's say we have a typical /users/{userID : UserID} route that accepts GET and PUT requests for fetching and updating a specific user respectively. The GET variant doesn't need a Body, but the PUT variant will.

getUser = Endpoint
  { method = GET
  , path = do
      Path.static "users"
      userID <- Path.param @UserID
      pure userID
  , query = pure ()
  , body = pure () -- No Body Needed
  , headers = pure ()
  , responder = do
      ... -- The appropriate responses for a GET request
  }

putUser = getUser
  { method = PUT
  , body = Body.json @UpdatedUser -- Body Needed
  , responder = do
      ... -- The appropriate responses for a PUT request
  }

This way, we can define the putUser Endpoint by simply modifying getUser and avoid repeating our self.


Matchpoint

A Matchpoint is a pattern that matches on Request values.

pattern Matchpoint :: Request -> Matchpoint

You can use the Matchpoint pattern synonym to create your own pattern synonyms that match specific Requests.

newtype UserID = UserID Int
  deriving ({- various typeclasses -})

pattern GetUserByID :: UserID -> Request
pattern GetUserByID userID <- Matchpoint
  GET
  ["users", PathParam @UserID userID]
  _
  _
  _

The GetUserByID pattern defined above would match against any Request of the form GET /users/{userID : UserID}. The Handler on the RHS of this pattern in a case statement will then be able to use the userID parameter in its function body if the Request matches sucessfully. If not, the next Matchpoint in your case statement is checked, just like regular patterns that we use all the time.

PathParam is a pattern synonym that you can use in your Matchpoints to match against path parameters of any type that are instances of both ToHttpApiData and FromHttpApiData. This is required since PathParam is a bidirectional pattern synonym. This property of PathParam makes it useful for generating URLs.

If your matching logic is more complicated, pattern synonyms alone may not be enough. For more complicated routes, we can use Okapi's DSL inside our Matchpoints by using -XViewPatterns. As an example, let's reimplement the first Endpoint on this page as a Matchpoint. Here's the Endpoint version first.

-- | Define Endpoints using an Applicative eDSL
myEndpoint = Endpoint
  { method = GET
  , path = do
      Path.static "index"
      magicNumber <- Path.param @Int
      pure magicNumber
  , query = do
      x <- Query.param @Int "x"
      y <- Query.option 10 $ Query.param @Int "y"
      pure (x, y)
  , body = do
      foo <- Body.json @Value
      pure foo
  , headers = pure ()
  , responder = do
      itsOk <- Responder.json @Int status200 do
        addSecretNumber <- AddHeader.using @Int "X-SECRET"
        pure addSecretNumber
      pure itsOk
  }

Here's the equivalent Matchpoint version.

-- | Define Matchpoints using the same DSL
pattern MyMatchpoint magicNumber pair foo = Matchpoint
  GET
  (Path.eval $ Path.static "index" *> Path.param @Int -> Ok magicNumber)
  (Query.eval xyQuery -> Ok pair)
  (Body.eval $ Body.json @Value -> Ok foo)
  _

xyQuery = do
  x <- Query.param @Int "x"
  y <- Query.option 10 $ Query.param @Int "y"
  pure (x, y)

We can simplify MyMatchpoint further by using more pattern synonyms.

pattern MyMatchpoint n pair bar <- Matchpoint
  GET
  (MagicNumber n)
  (XYQuery pair)
  (Foo bar)
  _

pattern MagicNumber n <- (Path.eval $ Path.static "index" *> Path.param @Int -> Ok n)

pattern XYQuery pair <- (Query.eval xyQuery -> Ok pair)

pattern Foo baz <- (Body.eval $ Body.json @Value -> Ok baz)

xyQuery = do
  x <- Query.param @Int "x"
  y <- Query.option 10 $ Query.param @Int "y"
  pure (x, y)

Pattern synonyms like MagicNumber or XYQuery in the example above come in handy when we need to use the same pattern inside multiple Matchpoints.

You can use the Matchpoint we defined above in a case statement with other Matchpoints to define a Server.

type Server m = Request -> m Response

myServer :: Server IO
myServer = \case
  MyMatchpoint n (x, y) foo -> do
    ...
  _ -> do
    ...

instantiate :: Monad m => (m ~> IO) -> Server m -> Application
instantiate transformer server = ...

api :: Application
api = instantiate id myServer

The Server type for Matchpoints is much simpler than the Server type for Endpoints.

Matchpoints vs. Endpoints

We recommend using Endpoints. Matchpoints are great if you're not worried about safety and just want to get something up and running quickly. Here are some downsides to using Matchpoints to implement your Server:

  1. We can't perform any static analysis on them. This means you can't generate OpenAPI specifications for Matchpoint Servers or perform any optimizations on them. You can perform static analysis on the Scripts that you use in your Matchpoints though, if there are any.

  2. All Handlers in a Matchpoint Server must operate within the same context. For Endpoints, this is not the case.

  3. Endpoints are more modular. You can achieve some level of modularity with your Matchpoints by using nested -XPatternSynonyms though.

  4. Matchpoint Servers have no knowledge of what Responses you will return to the Client. Endpoint Servers know every possible Response you may return from your Handlers, besides the ones returned by IO errors (the goal is for Endpoints to know about these as well).

  5. Requires knowledge of the -XPatternSynonyms and -XViewPatterns language extensions.

In short, if you don't care about safety, use Matchpoints.


Servant <> Okapi

Coming Soon