Servant-util
This package contains the core primitives which directly participate in API and some common utilities.
Build Instructions
Run stack build servant-util
to build everything.
Usage
For the following examples, we consider a simple server with two endpoints.
import Universum
import Data.Aeson (FromJSON, ToJSON)
import Data.Aeson.TH (defaultOptions, deriveJSON)
import qualified Network.Wai.Handler.Warp as Warp
import Network.Wai.Middleware.RequestLogger (logStdoutDev)
import Servant ((:<|>) (..), (:>), FromHttpApiData, Get, JSON, PostCreated, QueryParam, ReqBody,
Server, serve)
import Servant.Util ()
newtype Isbn = Isbn Word64
deriving (Eq, Show, ToJSON, FromJSON)
data Book = Book
{ isbn :: Isbn
, bookName :: Text
, author :: Text
}
deriveJSON defaultOptions 'Book
newtype Password = Password Text
deriving (FromHttpApiData)
type GetBooks
= Get '[JSON] [Book]
type AddBook
= QueryParam "password" Password
:> ReqBody '[JSON] Book
:> PostCreated '[JSON] Isbn
type BooksAPI = "books" :> (
GetBooks :<|>
AddBook
)
booksHandlers :: Server BooksAPI
booksHandlers =
(return [])
:<|>
(\_ book -> return (isbn book))
warpSettings :: Warp.Settings
warpSettings = Warp.defaultSettings
& Warp.setHost "127.0.0.1"
& Warp.setPort 8090
serveBooksServer :: IO ()
serveBooksServer =
Warp.runSettings warpSettings $
serve (Proxy @BooksAPI) booksHandlers
Sorting
When an endpoint is extended with a SortingParams
combinator, it starts to accept a sorting
specification in sortBy
query parameter. This way, the user can supply a sequence of fields
from the set of allowed fields. They will be applied lexicographically in the specified order.
For example, GetBooks
from the example above can be extended as
type GetBooks
= SortingParams
["isbn" ?: Isbn, "name" ?: Text, "author" ?: Text]
'["isbn" ?: 'Asc Isbn]
:> Get '[JSON] [Book]
The first list required by SortingParams
combinator should consist of fields that
are allowed to participate in sorting.
The first argument of ?:
operator stands for a field name from front-end's point
of view; the second argument corresponds to the field type and used primarily to avoid
mistakes in the implementation.
(Soon, it will also be used to distinguish between nullable and mandatory fields.)
The second list required by SortingParams
stands for the base sorting that will
always be applied last disregard the user's input. It allows for more
deterministic results and is, in fact, essential when paired with pagination.
Examples of valid requests to this server (using httpie):
http :8090/books sortBy=='asc(name)'
— sort alphabetically by name
(for equal names — by isbn
);
http :8090/books sortBy=='asc(name),desc(author)'
— sort alphabetically by name
, for equal names — by author
in reversed order (and for entries with the same name and author - by isbn
).
http :8090/books sortBy==+name,-author'
— same as above.
The server handler will be supplied with SortingSpec
argument, use neighbor servant-util-*
packages for applying this specification to your backend.
You can also use methods from Servant.Util.Dummy
for a trivial in-Haskell
implementation, suitable for a server prototype:
import Servant.Util.Dummy (sortBySpec, fieldSort)
getBooks :: ToServer GetBooks
getBooks sortingSpec = do
let
-- Correlate user input with fields of our response type
sortingApp Book{..} =
fieldSort @"name" bookName .*.
fieldSort @"author" author .*.
HNil
sortBySpec sortingSpec sortingApp <$> allBooks
Since a list of fields that can participate in sorting is usually determined by the response
type, you can extract this fields list to an instance of the dedicated type family helper:
type instance SortingParamBaseOf Book =
["name" ?: Isbn, "author" ?: Text]
type instance SortingParamProvidedOf Book =
["isbn" ?: Asc Isbn]
type GetBooks
= SortingParamsOf Book -- same as the definition at the section's top
:> Get '[JSON] [Book]
In case you need to construct a SortingSpec
manually (for instance, to pass to a client handler),
take a look at Servant.Util.Combinators.Sorting.Construction
module.
Filtering
This package provides support for many types of filtering: exact matching, comparisons,
text search; these are called automatic filters. Complex filtering conditions which do
not (and cannot) fall into one of the mentioned categories are also allowed and here they
are called manual filters. When user specifies multiple filters, their conjunction is
applied.
Let's consider filters on the previous example. Your API should be extended with
FilteringParams
combinator which is pretty similar to SortingParams
.
type GetBooks
= FilteringParams ["isbn" ?: 'AutoFilter Isbn, "name" ?: 'AutoFilter Text]
:> Get '[JSON] [Book]
Your endpoint implementation will be provided with FilteringSpec
argument. Note that
this time parameters list contains not only parameter name and type, but also the type of
filter.
Just like for sorting, here you can use SortingParamTypesOf
type family to reduce the
boilerplate.
Now you need to tell which automatic filter types are allowed for your types:
type SupportedFilters Isbn = '[FilterMatching, FilterComparing]
type SupportedFilters Text = '[FilterMatching, FilterComparing]
-- the latter is already defined in this library
On the frontend side the following query parameters will be allowed:
isbn=12345
- plain matching filter.
isbn[eq]=12345
- same as above.
isbn[neq]=12345
- values not equal to the given one.
isbn[in]=[12345,23456]
- any value from the given list matches.
isbn[gt]=12345
- higher values are allowed.
isbn[lte]=12345
- the opposite to the previous predicate.
name[gte]=D&name[lt]=E
- everything starting with latter D
.
Now let's suppose you need your server backend to support a much more complex
predicate: for instance, the book's name is longer than 10
characters. You can either define
your own filter or use a manual one; let's demonstrate the latter case:
type GetBooks
= FilteringParams ["hasLongName" ?: ManualFilter Bool]
:> Get '[JSON] [Book]
This way user can supply only hasLongName=false
or hasLongName=true
query parameter,
but the filter implementation can be arbitrarily complex.
In case we put FilteringParams
with all the fields mentioned above into our GET
method,
a dummy implementation of its handler may look like:
getBooks :: ToServer GetBooks
getBooks filterSpec = do
let
filterApp Book{..} =
filterOn @"isbn" isbn .*. -- automatic fields require only the field
filterOn @"name" bookName .*.
manualFilter @"hasLongName" -- manual filter requires a predicate
(\needLongName -> (length bookName >= 10) == needLongName) .*.
HNil
filterBySpec filterSpec filterApp <$> getAllBooks
If for any reason you need to construct a FilteringSpec
manually, take a look at
Servant.Util.Combinators.Filtering.Construction
module.
Pagination is applied via PaginationParams
combinator. It accepts a settings
type argument,
which is currently just either 'DefPageSize n
or 'DefUnlimitedPageSize
that define
the default page size (defined statically in order to be reflected in documentation).
An endpoint supplied with this combinator starts accepting offset
and limit
query
parameters, both are optional.
Your endpoint implementation will be given a PaginationSpec
object which can be applied
with an appropriate function.
import Servant.Util (PaginationParams, PaginationSpec)
import Servant.Util.Dummy (paginate)
type GetBooks
= PaginationParams
:> Get '[JSON] [Book]
getBooks :: ToServer GetBooks
getBooks pagination = paginate pagination <$> getAllBooks
Logging
Problem
One can enable logging of incoming requests in a very simple way using logStdoutDev
function from wai-extra
package.
Let's try it:
http POST :8090/books isbn:=12312 bookName='Some book' author=unknown password==qwerty123
Produced log:
POST /books
Params: [("password","qwerty123")]
Request Body: {"isbn": 12312, "bookName": "Some book", "author": "unknown"}
Accept: application/json, */*
Status: 201 Created 0.000086s
Note the problem: logs contain the password which makes them unusable in production.
And if we use
logStdout
,
logs are missing the most part of the request data:
127.0.0.1 - - [29/Jan/2019:02:32:31 +0300] "POST /books?password=qwerty123 HTTP/1.1" 201 -
"" "HTTPie/0.9.8"
Proposed solution
A reasonable way to resolve this would be to display objects depending on their semantics.
This package provides an implementation of such logging, it can be used as follows:
serveBooksServer :: IO ()
serveBooksServer =
Warp.runSettings warpSettings $
serverWithLogging loggingConfig (Proxy @BooksAPI) $ \sp ->
serve sp booksHandlers
where
loggingConfig = ServantLogConfig putTextLn
This will wrap your API into an internal LoggingAPI
combinator, resulting API Proxy
should be passed to serve
method; meanwhile, handlers remain unchanged.
You will also need to provide Buildable
instances for all request parameters:
instance Buildable Isbn where
build (Isbn i) = "isbn:" <> build i
instance Buildable Password where
build _ = "<password>"
instance Buildable Book where
build Book{..} =
"{ isbn = " +| isbn |+
", title = " +| bookName |+
", author = " +| author |+
" }"
and instance Buildable (ForResponseLog *)
for all types appearing as a response.
Note that semantics of ForResponseLog
newtype wrapper is displaying a reasonable part of
a response: not too little to stay informative, not too large in order to keep logs small.
instance Buildable (ForResponseLog Isbn) where
build = buildForResponse
instance Buildable (ForResponseLog Book) where
build = buildForResponse
instance Buildable (ForResponseLog [Book]) where
build = buildListForResponse (take 5)
Now logs look like
POST Request #1
:> books
:> 'password' field: <password>
:> request body: { isbn = isbn:12312, title = Some book, author = unknown }
Response #1 OK 0.013415s > isbn:12312
For Contributors
Please see CONTRIBUTING.md for more information.
About Serokell
Servant-util is maintained and funded with ❤️ by Serokell. The names and logo for Serokell are trademark of Serokell OÜ.
We love open source software! See our other projects or hire us to design, develop and grow your idea!