10 KiB
Introduction
Okapi is a microframework for web development in Haskell based on monadic parsing. The inspiration for Okapi originally came from looking at web frameworks in other language ecosystems such as Python’s Flask, Nim’s Jester, OCaml’s Dream, and F#’s Giraffe, which the name of this Haskell framework is related to. I noticed that many Haskell web frameworks tend to require a lot of boilerplate code, and/or make use of a lot of advanced type level language features that make it hard to understand the internals of the framework. The goal of Okapi is to create a Haskell web framework with an ergonomic developer experience that is idiomatic to the host language.
Parsers
In Haskell, a simple String
parser can be modeled as a function with the type String -> (Either ParserError a, String)
.
This function takes values of type String
and returns either a ParserError
(if it fails) or a value of some type a
(if it succeeds), along with a new String
that's missing the characters that were consumed by the parsing function. We could use the function like so:
This is great, but issues start to arise when we try to compose parsers with other parsers. For example, let's say we wanted to parse blah blah:
To avoid the explicit passing of state from one parser to the next, we can use monads. You may have already noticed that the type of our parser can be simplified with the State String
monad.
We can transform our function of type String -> (Either ParseError a, String)
to a value of type State String (Either ParserError a)
. A value of type State String (Either ParserError a)
represents a value of type Either ParserError a
that was computed in a stateful context, where the state is of type String
. Now that our parser is defined as a monad we can use do
notation
, and it becomes easier to compose our parsers with other parsers because we don't have to manually pass the state from a previous parser to the next one.
Let's try the parser composition we tried above with our new parser definition:
As you can see our parsers compose a lot better, but we still have to explicitly handle the result of the parsers because they may return a ParserError
.
Functions that return values of the type Either ParserError a
can be modelled using the Except ParserError
monad. A value of the type Except ParserError a
represents a value of type a
that is computed in a context that may not succeed, but instead throw an error value of type ParserError
. In our case we want
our parser's computations to happen in a context in which there is state of type String
, and the possibilty of throwing an error value of type ParserError
.
To get both of these useful abilities, let's combine the Except ParserError
monad with our State String
monad using monad transformers. Our simplified parser
now has the type ExceptT ParserError (State String a)
, where ExceptT
is a monad transformer that gives our base State String
monad the ability to throw error
values of type ParserError
upon failure. To make the code examples easier on our eyes, let's make a type synonym defined as type Parser a = ExceptT ParserError (State String a)
.
Now, any value anottated with the type Parser a
represents a value of some type a
that is computed in a context that has access to state of type String
AND may throw error
values of type ParserError
upon failing. Let's redefine the example we defined above:
Great.
HTTP Request Parsers
Now, let's redefine type Parser a = ExceptT ParserError (State String a)
as type Parser a = ExceptT HTTPError (State HTTPRequest a)
. This is an HTTP Request Parser. Instead of parsing happening in a context where the computation has access to state of type String
and can throw errors of type ParserError
, it happens in a context where the computation has access to state of type HTTPRequest
and can throw errors of type HTTPError
. Just like the string parser above had a concept of "consuming" parts of a String
, the HTTP request parser "consumes" values of the type HTTPRequest
. By consume we mean .... If you break values of type String
into its smallest consituents, you get values of type Char
. A String
value is a list of Char
values. What are the smallest constituents of a HTTPRequest
value? The data type HTTPRequest
is defined as follows:
data HTTPRequest = HTTPRequest
{ method :: Method
, path :: [Text]
, query :: Query
, body :: ByteString
, headers :: Headers
}
Our HTTP request parser consumes different parts of the HTTP request like the method
and query
. Once a piece of the HTTP request is parsed, it is removed from the request before it is implicitly passed to the next parser.
There are 2 types of parsers:
- Data parsers
- Checker parsers
There are 5 types of parsers for each of the 5 parts of a HTTP request.
- Method Parsers
method :: MonadOkapi m => m Method
matchMethod :: MonadOkapi m => Method -> m ()
get :: MonadOkapi m => m ()
get = matchMethod "GET"
- Path Parsers
path :: MonadOkapi m => m Path -- Parses entire remaining path
path = many seg
seg :: MonadOkapi m => m Text
matchPath :: MonadOkapi m => Path -> m ()
matchPath desiredPath = mapM_ matchSeg desiredPath
matchSeg :: MonadOkapi m => Text -> m ()
pathParam :: MonadOkapi m => FromHttpApiData a => m a
pathEnd :: MonadOkapi m => m ()
- Query Parsers
query :: MonadOkapi m => m Query -- parses entire query
queryParam :: MonadOkapi m => FromHttpApiData a => Text -> m a
queryFlag :: MonadOkapi m => Text -> m ()
queryParamRaw :: Text -> m Text
- Body Parsers
body :: MonadOkapi m => m Body
bodyJSON :: MonadOkapi m, FromJSON a => m a
bodyFormURLEncoded :: FromForm a, MonadOkapi m => m a
bodyFormMultipart :: FromForm a, MonadOkapi m => m (a, [File])
- Headers Parsers
headers :: MonadOkapi m => m Headers
header :: MonadOkapi m => HeaderName -> m Header
cookie :: MonadOkapi m => m Cookie
crumb :: MonadOkapi m => Text -> m Crumb
We can use these to create increasingly complex parsers. For example, let's say we wanted to implement a HTTP parser that matches the request GET /blog
. That would look like this:
blogRoute :: Parser ()
blogRoute = do
get -- Make sure that the request is a GET request
matchSeg "blog" -- Match against the path segment /blog
pathEnd -- Make sure that there are no more path segments remaining in the request
Just like earlier, with our monadic string parser, we can sequence HTTP request parsers using do
notation. This request parser isn't really useful though because it doesn't return anything. Let's make it return a response:
blogRoute :: Parser HTTPResponse
blogRoute = do
get
matchSeg "blog"
pathEnd
return ok
Now if we run our parser, it will return a 200 OK
response if we send a GET
request to the /blog
endpoint. On top of being able to sequence parsers with do
notation thanks to Parser
being an instance of the Monad
typeclass, we can also build parsers that "choose" between multiple subparsers. This is possible because the Parser
type is also an instance of the Alternative
typeclass, which provides the <|>
operator.
Explain <|>
then explain we can also parser combinators like many
, some
, optional
, option
, takeWhile
, etc.
Then explain the two types of errors and how to throw and catch them.
Then explain returning responses and executing a parser.
Explaining type safe URLs with patterns:
Patterns
Okapi uses bi-directional patterns to have typesafe urls. So you would have something like:
-- Matches routes of the form /blog/99
pattern BlogRoute :: Int -> Path
pattern BlogRoute uuid <- ["blog", PathParam uuid]
where
BlogRoute uuid = ["blog", PathParam uuid]
or just
-- Bidriectional Implicit
pattern BlogRoute :: Int -> Path
pattern BlogRoute uuid = ["blog", PathParam uuid]
pattern BlogCategoryRoute :: Text -> Path
pattern BlogCategoryRoute category = ["blog", PathParam category]
uses these bidrectional patterns with the route
parser, like so:
route :: MonadOkapi m => (Path -> m Response) -> m Response
route matcher = do
path <- parsePath
matcher path
myAPI :: MonadOkapi m => m Response
myAPI = route $ \case
BlogRoute uuid -> do
get
return ok
BlogRouteCategory category -> do
get
mbOrderBy <- optional $ queryParam @Order "order"
case mbOrderBy of
Nothing -> do
...
return ok
Just orderBy -> do
...
return ok
_ -> next
Since both routes are GET
requests, let's factor out the get
parser:
myAPI :: MonadOkapi m => m Response
myAPI = do
get
route $ \case
BlogRoute uuid -> return ok
BlogRouteCategory category -> do
mbOrderBy <- optional $ queryParam @Order "order"
case mbOrderBy of
Nothing -> do
...
return ok
Just orderBy -> do
...
return ok
_ -> next
URLs
There are two types of URLs that you can generate with Okapi:
- Relative URLs
- Absolute URLs
data URL = URL { unURL :: Text }
data RelURL = RelURL Path Query
data AbsURL = AbsURL Scheme Host (Maybe Port) RelURL
class ToURL a where
render :: a -> URL
instance ToURL Path where
instance ToURL Query where
instance ToURL RelURL where
render (RelURL p q) = render p <> render q
instance ToURL AbsURL where
instance ToURL Request where
blogRouteCategoryURL = render $ BlogRouteCategory "fiction"
class ToURL a => Matchable a where
match :: (a -> m Response) -> m Response
route :: (Path -> m Response) -> m Response
routeWithQuery :: (RelURL -> m Response) -> m Response
routeVirtual :: (AbsURL -> m Response) -> m Response