At this moment there are at least 6 json libraries for scala, not counting the java json libraries. All these libraries have a very similar AST. This project aims to provide a single AST to be used by other scala json libraries.
At this moment the approach taken to working with the AST has been taken from lift-json and the native package is in fact lift-json but outside of the lift project.
This project also attempts to set lift-json free from the release schedule imposed by the lift framework. The Lift framework carries many dependencies and as such it's typically a blocker for many other scala projects when a new version of scala is released.
So the native package in this library is in fact verbatim lift-json in a different package name, this means that your import statements will change if you use this library.
import org.json4s._
import org.json4s.native.JsonMethods._
After that everything works exactly the same as it would with lift-json
In addition to the native parser there is also an implementation that uses jackson for parsing to the AST. The jackson module includes most of the jackson-module-scala functionality and the ability to use it with the lift-json AST.
To use jackson instead of the native parser:
import org.json4s._
import org.json4s.jackson.JsonMethods._
Parsing and formatting utilities for JSON.
A central concept in lift-json library is Json AST which models the structure of a JSON document as a syntax tree.
sealed abstract class JValue
case object JNothing extends JValue // 'zero' for JValue
case object JNull extends JValue
case class JString(s: String) extends JValue
case class JDouble(num: Double) extends JValue
case class JDecimal(num: BigDecimal) extends JValue
case class JInt(num: BigInt) extends JValue
case class JBool(value: Boolean) extends JValue
case class JObject(obj: List[JField]) extends JValue
case class JArray(arr: List[JValue]) extends JValue
type JField = (String, JValue)
All features are implemented in terms of above AST. Functions are used to transform the AST itself, or to transform the AST between different formats. Common transformations are summarized in a following picture.
Summary of the features:
- Fast JSON parser
- LINQ style queries
- Case classes can be used to extract values from parsed JSON
- Diff & merge
- DSL to produce valid JSON
- XPath like expressions and HOFs to manipulate JSON
- Pretty and compact printing
- XML conversions
- Serialization
- Low level pull parser API
You can add the json4s as a dependency in following ways. Note, replace XXX with correct Json4s version.
For the native support add the following dependency to your project description:
val json4sNative = "org.json4s" %% "json4s-native" % "3.2.6"
For the Jackson support add the following dependency to your project description:
val json4sJackson = "org.json4s" %% "json4s-jackson" % "3.2.6"
For the native support add the following dependency to your pom:
<dependency>
<groupId>org.json4s</groupId>
<artifactId>json4s-native_${scala.version}</artifactId>
<version>3.2.6</version>
</dependency>
For the jackson support add the following dependency to your pom:
<dependency>
<groupId>org.json4s</groupId>
<artifactId>json4s-jackson_${scala.version}</artifactId>
<version>3.2.6</version>
</dependency>
Download following jars:
- http://repo1.maven.org/maven2/org/json4s/json4s-core_2.9.2/3.2.6/json4s-core_2.9.3-3.2.6.jar
- http://repo1.maven.org/maven2/org/json4s/json4s-native_2.9.2/3.2.6/json4s-native_2.9.3-3.2.6.jar
- http://mirrors.ibiblio.org/pub/mirrors/maven2/com/thoughtworks/paranamer/paranamer/2.5.6/paranamer-2.5.6.jar
- scalap (Only for Scala-2.9 compatible versions)
Support for Enum, Joda-Time, ...
Applicative style parsing with Scalaz
Support for Box
JField is no longer a JValue. This means more type safety since it is no longer possible to create invalid JSON where JFields are added directly into JArrays for instance. Most noticeable consequence of this change is that map, transform, find and filter come in two versions:
def map(f: JValue => JValue): JValue
def mapField(f: JField => JField): JValue
def transform(f: PartialFunction[JValue, JValue]): JValue
def transformField(f: PartialFunction[JField, JField]): JValue
def find(p: JValue => Boolean): Option[JValue]
def findField(p: JField => Boolean): Option[JField]
//...
Use *Field functions to traverse fields in the JSON, and use the functions without 'Field' in the name to traverse values in the JSON.
Path expressions were changed after 2.2 version. Previous versions returned JField which unnecessarily complicated the use of the expressions. If you have used path expressions with pattern matching like:
val JField("bar", JInt(x)) = json \ "foo" \ "bar"
It is now required to change that to:
val JInt(x) = json \ "foo" \ "bar"
Any valid json can be parsed into internal AST format. For native support:
scala> import org.json4s._
scala> import org.json4s.native.JsonMethods._
scala> parse(""" { "numbers" : [1, 2, 3, 4] } """)
res0: org.json4s.JsonAST.JValue =
JObject(List((numbers,JArray(List(JInt(1), JInt(2), JInt(3), JInt(4))))))
scala> parse("""{"name":"Toy","price":35.35}""", useBigDecimalForDouble = true)
res1: org.json4s.package.JValue =
JObject(List((name,JString(Toy)), (price,JDecimal(35.35))))
For jackson support:
scala> import org.json4s._
scala> import org.json4s.jackson.JsonMethods._
scala> parse(""" { "numbers" : [1, 2, 3, 4] } """)
res0: org.json4s.JsonAST.JValue =
JObject(List((numbers,JArray(List(JInt(1), JInt(2), JInt(3), JInt(4))))))
scala> parse("""{"name":"Toy","price":35.35}""", useBigDecimalForDouble = true)
res1: org.json4s.package.JValue =
JObject(List((name,JString(Toy)), (price,JDecimal(35.35))))
You can generate json in 2 modes either in DoubleMode
or in BigDecimalMode
; the former will map all decimal values
into a JDouble the latter into a JDecimal.
For the double mode dsl use:
import org.json4s.JsonDSL._
// or
import org.json4s.JsonDSL.WithDouble._
For the big decimal mode dsl use:
import org.json4s.JsonDSL.WithBigDecimal._
-
Primitive types map to JSON primitives.
-
Any seq produces JSON array.
scala> val json = List(1, 2, 3) scala> compact(render(json)) res0: String = [1,2,3]
-
Tuple2[String, A] produces field.
scala> val json = ("name" -> "joe") scala> compact(render(json)) res1: String = {"name":"joe"}
-
~ operator produces object by combining fields.
scala> val json = ("name" -> "joe") ~ ("age" -> 35) scala> compact(render(json)) res2: String = {"name":"joe","age":35}
-
Any value can be optional. Field and value is completely removed when it doesn't have a value.
scala> val json = ("name" -> "joe") ~ ("age" -> Some(35)) scala> compact(render(json)) res3: String = {"name":"joe","age":35} scala> val json = ("name" -> "joe") ~ ("age" -> (None: Option[Int])) scala> compact(render(json)) res4: String = {"name":"joe"}
-
Extending the dsl To extend the dsl with your own classes you must have an implicit conversion in scope of signature:
type DslConversion = T => JValue
object JsonExample extends App {
import org.json4s._
import org.json4s.JsonDSL._
case class Winner(id: Long, numbers: List[Int])
case class Lotto(id: Long, winningNumbers: List[Int], winners: List[Winner], drawDate: Option[java.util.Date])
val winners = List(Winner(23, List(2, 45, 34, 23, 3, 5)), Winner(54, List(52, 3, 12, 11, 18, 22)))
val lotto = Lotto(5, List(2, 45, 34, 23, 7, 5, 3), winners, None)
val json =
("lotto" ->
("lotto-id" -> lotto.id) ~
("winning-numbers" -> lotto.winningNumbers) ~
("draw-date" -> lotto.drawDate.map(_.toString)) ~
("winners" ->
lotto.winners.map { w =>
(("winner-id" -> w.id) ~
("numbers" -> w.numbers))}))
println(compact(render(json)))
}
scala> JsonExample
{"lotto":{"lotto-id":5,"winning-numbers":[2,45,34,23,7,5,3],"winners":
[{"winner-id":23,"numbers":[2,45,34,23,3,5]},{"winner-id":54,"numbers":[52,3,12,11,18,22]}]}}
Example produces following pretty printed JSON. Notice that draw-date field is not rendered since its value is None:
scala> pretty(render(JsonExample.json))
{
"lotto":{
"lotto-id":5,
"winning-numbers":[2,45,34,23,7,5,3],
"winners":[{
"winner-id":23,
"numbers":[2,45,34,23,3,5]
},{
"winner-id":54,
"numbers":[52,3,12,11,18,22]
}]
}
}
Two JSONs can be merged and diffed with each other. Please see more examples in MergeExamples.scala and DiffExamples.scala.
scala> import org.json4s._
scala> import org.json4s.jackson.JsonMethods._
scala> val lotto1 = parse("""{
"lotto":{
"lotto-id":5,
"winning-numbers":[2,45,34,23,7,5,3]
"winners":[{
"winner-id":23,
"numbers":[2,45,34,23,3,5]
}]
}
}""")
scala> val lotto2 = parse("""{
"lotto":{
"winners":[{
"winner-id":54,
"numbers":[52,3,12,11,18,22]
}]
}
}""")
scala> val mergedLotto = lotto1 merge lotto2
scala> pretty(render(mergedLotto))
res0: String =
{
"lotto":{
"lotto-id":5,
"winning-numbers":[2,45,34,23,7,5,3],
"winners":[{
"winner-id":23,
"numbers":[2,45,34,23,3,5]
},{
"winner-id":54,
"numbers":[52,3,12,11,18,22]
}]
}
}
scala> val Diff(changed, added, deleted) = mergedLotto diff lotto1
changed: org.json4s.JsonAST.JValue = JNothing
added: org.json4s.JsonAST.JValue = JNothing
deleted: org.json4s.JsonAST.JValue = JObject(List((lotto,JObject(List(JField(winners,
JArray(List(JObject(List((winner-id,JInt(54)), (numbers,JArray(
List(JInt(52), JInt(3), JInt(12), JInt(11), JInt(18), JInt(22))))))))))))))
JSON values can be extracted using for-comprehensions. Please see more examples in JsonQueryExamples.scala.
scala> import org.json4s._
scala> import org.json4s.native.JsonMethods._
scala> val json = parse("""
{ "name": "joe",
"children": [
{
"name": "Mary",
"age": 5
},
{
"name": "Mazy",
"age": 3
}
]
}
""")
scala> for {
JObject(child) <- json
JField("age", JInt(age)) <- child
} yield age
res0: List[BigInt] = List(5, 3)
scala> for {
JObject(child) <- json
JField("name", JString(name)) <- child
JField("age", JInt(age)) <- child
if age > 4
} yield (name, age)
res1: List[(String, BigInt)] = List((Mary,5))
Json AST can be queried using XPath like functions. Following REPL session shows the usage of '\', '\\', 'find', 'filter', 'transform', 'remove' and 'values' functions.
The example json is:
{
"person": {
"name": "Joe",
"age": 35,
"spouse": {
"person": {
"name": "Marilyn"
"age": 33
}
}
}
}
Translated to DSL syntax:
scala> import org.json4s._
scala> import org.json4s.native.JsonMethods._
or
scala> import org.json4s.jackson.JsonMethods._
scala> import org.json4s.JsonDSL._
scala> val json =
("person" ->
("name" -> "Joe") ~
("age" -> 35) ~
("spouse" ->
("person" ->
("name" -> "Marilyn") ~
("age" -> 33)
)
)
)
scala> json \\ "spouse"
res0: org.json4s.JsonAST.JValue = JObject(List(
(person,JObject(List((name,JString(Marilyn)), (age,JInt(33)))))))
scala> compact(render(res0))
res1: String = {"person":{"name":"Marilyn","age":33}}
scala> compact(render(json \\ "name"))
res2: String = {"name":"Joe","name":"Marilyn"}
scala> compact(render((json removeField { _ == JField("name", JString("Marilyn")) }) \\ "name"))
res3: String = {"name":"Joe"}
scala> compact(render(json \ "person" \ "name"))
res4: String = "Joe"
scala> compact(render(json \ "person" \ "spouse" \ "person" \ "name"))
res5: String = "Marilyn"
scala> json findField {
case JField("name", _) => true
case _ => false
}
res6: Option[org.json4s.JsonAST.JValue] = Some((name,JString(Joe)))
scala> json filterField {
case JField("name", _) => true
case _ => false
}
res7: List[org.json4s.JsonAST.JField] = List(JField(name,JString(Joe)), JField(name,JString(Marilyn)))
scala> json transformField {
case JField("name", JString(s)) => ("NAME", JString(s.toUpperCase))
}
res8: org.json4s.JsonAST.JValue = JObject(List((person,JObject(List(
(NAME,JString(JOE)), (age,JInt(35)), (spouse,JObject(List(
(person,JObject(List((NAME,JString(MARILYN)), (age,JInt(33)))))))))))))
scala> json.values
res8: scala.collection.immutable.Map[String,Any] = Map(person -> Map(name -> Joe, age -> 35, spouse -> Map(person -> Map(name -> Marilyn, age -> 33))))
Indexed path expressions work too and values can be unboxed using type expressions.
scala> val json = parse("""
{ "name": "joe",
"children": [
{
"name": "Mary",
"age": 5
},
{
"name": "Mazy",
"age": 3
}
]
}
""")
scala> (json \ "children")(0)
res0: org.json4s.JsonAST.JValue = JObject(List((name,JString(Mary)), (age,JInt(5))))
scala> (json \ "children")(1) \ "name"
res1: org.json4s.JsonAST.JValue = JString(Mazy)
scala> json \\ classOf[JInt]
res2: List[org.json4s.JsonAST.JInt#Values] = List(5, 3)
scala> json \ "children" \\ classOf[JString]
res3: List[org.json4s.JsonAST.JString#Values] = List(Mary, Mazy)
Case classes can be used to extract values from parsed JSON. Non-existing values can be extracted into scala.Option and strings can be automatically converted into java.util.Dates. Please see more examples in ExtractionExampleSpec.scala.
scala> import org.json4s._
scala> import org.json4s.jackson.JsonMethods._
scala> implicit val formats = DefaultFormats // Brings in default date formats etc.
scala> case class Child(name: String, age: Int, birthdate: Option[java.util.Date])
scala> case class Address(street: String, city: String)
scala> case class Person(name: String, address: Address, children: List[Child])
scala> val json = parse("""
{ "name": "joe",
"address": {
"street": "Bulevard",
"city": "Helsinki"
},
"children": [
{
"name": "Mary",
"age": 5
"birthdate": "2004-09-04T18:06:22Z"
},
{
"name": "Mazy",
"age": 3
}
]
}
""")
scala> json.extract[Person]
res0: Person = Person(joe,Address(Bulevard,Helsinki),List(Child(Mary,5,Some(Sat Sep 04 18:06:22 EEST 2004)), Child(Mazy,3,None)))
By default the constructor parameter names must match json field names. However, sometimes json field names contain characters which are not allowed characters in Scala identifiers. There's two solutions for this (see LottoExample.scala for bigger example).
Use back ticks.
scala> case class Person(`first-name`: String)
Use transform function to postprocess AST.
scala> case class Person(firstname: String)
scala> json transformField {
case ("first-name", x) => ("firstname", x)
}
Extraction function tries to find the best matching constructor when case class has auxiliary constructors. For instance extracting from JSON {"price":350} into the following case class will use the auxiliary constructor instead of the primary constructor.
scala> case class Bike(make: String, price: Int) {
def this(price: Int) = this("Trek", price)
}
scala> parse(""" {"price":350} """).extract[Bike]
res0: Bike = Bike(Trek,350)
Primitive values can be extracted from JSON primitives or fields.
scala> (json \ "name").extract[String]
res0: String = "joe"
scala> ((json \ "children")(0) \ "birthdate").extract[Date]
res1: java.util.Date = Sat Sep 04 21:06:22 EEST 2004
DateFormat can be changed by overriding 'DefaultFormats' (or by implmenting trait 'Formats').
scala> implicit val formats = new DefaultFormats {
override def dateFormatter = new SimpleDateFormat("yyyy-MM-dd'T'HH:mm:ss.SSS'Z'")
}
JSON object can be extracted to Map[String, _] too. Each field becomes a key value pair in result Map.
scala> val json = parse("""
{
"name": "joe",
"addresses": {
"address1": {
"street": "Bulevard",
"city": "Helsinki"
},
"address2": {
"street": "Soho",
"city": "London"
}
}
}""")
scala> case class PersonWithAddresses(name: String, addresses: Map[String, Address])
scala> json.extract[PersonWithAddresses]
res0: PersonWithAddresses("joe", Map("address1" -> Address("Bulevard", "Helsinki"),
"address2" -> Address("Soho", "London")))
Case classes can be serialized and deserialized. Please see other examples in SerializationExamples.scala.
scala> import org.json4s._
scala> import org.json4s.native.Serialization
scala> import org.json4s.native.Serialization.{read, write}
scala> implicit val formats = Serialization.formats(NoTypeHints)
scala> val ser = write(Child("Mary", 5, None))
scala> read[Child](ser)
res1: Child = Child(Mary,5,None)
If you're using jackson instead of the native one:
scala> import org.json4s._
scala> import org.json4s.jackson.Serialization
scala> import org.json4s.jackson.Serialization.{read, write}
scala> implicit val formats = Serialization.formats(NoTypeHints)
scala> val ser = write(Child("Mary", 5, None))
scala> read[Child](ser)
res1: Child = Child(Mary,5,None)
Serialization supports:
- Arbitrarily deep case class graphs
- All primitive types, including BigInt and Symbol
- List, Seq, Array, Set and Map (note, keys of the Map must be strings: Map[String, _])
- scala.Option
- java.util.Date
- Polymorphic Lists (see below)
- Recursive types
- Serialization of fields of a class (see below)
- Custom serializer functions for types which are not supported (see below)
Type hints are required when serializing polymorphic (or heterogeneous) Lists. Serialized JSON objects will get an extra field named 'jsonClass' (the name can be changed by overriding 'typeHintFieldName' from Formats).
scala> trait Animal
scala> case class Dog(name: String) extends Animal
scala> case class Fish(weight: Double) extends Animal
scala> case class Animals(animals: List[Animal])
scala> implicit val formats = Serialization.formats(ShortTypeHints(List(classOf[Dog], classOf[Fish])))
scala> val ser = write(Animals(Dog("pluto") :: Fish(1.2) :: Nil))
ser: String = {"animals":[{"jsonClass":"Dog","name":"pluto"},{"jsonClass":"Fish","weight":1.2}]}
scala> read[Animals](ser)
res0: Animals = Animals(List(Dog(pluto), Fish(1.2)))
ShortTypeHints outputs short classname for all instances of configured objects. FullTypeHints outputs full classname. Other strategies can be implemented by extending TypeHints trait.
To enable serialization of fields, a FieldSerializer can be added for some type:
implicit val formats = DefaultFormats + FieldSerializer[WildDog]()
Now the type WildDog (and all subtypes) gets serialized with all its fields (+ constructor parameters). FieldSerializer takes two optional parameters which can be used to intercept the field serialization:
case class FieldSerializer[A: Manifest](
serializer: PartialFunction[(String, Any), Option[(String, Any)]] = Map(),
deserializer: PartialFunction[JField, JField] = Map()
)
Those PartialFunctions are called just before a field is serialized or deserialized. Some useful PFs to rename and ignore fields are provided:
val dogSerializer = FieldSerializer[WildDog](
renameTo("name", "animalname") orElse ignore("owner"),
renameFrom("animalname", "name"))
implicit val formats = DefaultFormats + dogSerializer
We've added support for case classes defined in a trait. But they do need custom formats. I'll explain why and then how.
For classes defined in a trait it's a bit difficult to get to their companion object, which is needed to provide default values. We could punt on those but that brings us to the next problem, the compiler generates an extra field in the constructor of such case classes. The first field in the constructor of those case classes is called $outer
and is of type of the defining trait. So somehow we need to get an instance of that object, naively we could scan all classes and collect the ones that are implementing the trait, but when there are more than one: which one to take?
I've chosen to extend the formats to include a list of companion mappings for those case classes. So you can have formats that belong to your modules and keep the mappings in there. That will then make default values work and provide the much needed $outer
field.
trait SharedModule {
case class SharedObj(name: String, visible: Boolean = false)
}
object PingPongGame extends SharedModule
implicit val formats: Formats =
DefaultFormats.withCompanions(classOf[PingPongGame.SharedObj] -> PingPongGame)
val inst = PingPongGame.SharedObj("jeff", visible = true)
val extr = Extraction.decompose(inst)
extr must_== JObject("name" -> JString("jeff"), "visible" -> JBool(true))
extr.extract[PingPongGame.SharedObj] must_== inst
It is possible to plug in custom serializer + deserializer functions for any type. Now, if we have a non case class Interval (thus, not supported by default), we can still serialize it by providing following serializer.
scala> class Interval(start: Long, end: Long) {
val startTime = start
val endTime = end
}
scala> class IntervalSerializer extends CustomSerializer[Interval](format => (
{
case JObject(JField("start", JInt(s)) :: JField("end", JInt(e)) :: Nil) =>
new Interval(s.longValue, e.longValue)
},
{
case x: Interval =>
JObject(JField("start", JInt(BigInt(x.startTime))) ::
JField("end", JInt(BigInt(x.endTime))) :: Nil)
}
))
scala> implicit val formats = Serialization.formats(NoTypeHints) + new IntervalSerializer
Custom serializer is created by providing two partial functions. The first evaluates to a value if it can unpack the data from JSON. The second creates the desired JSON if the type matches.
Module json4s-ext contains extensions to extraction and serialization. Following types are supported.
// Lift's box
implicit val formats = org.json4s.DefaultFormats + new org.json4s.native.ext.JsonBoxSerializer
// Scala enums
implicit val formats = org.json4s.DefaultFormats + new org.json4s.ext.EnumSerializer(MyEnum)
// or
implicit val formats = org.json4s.DefaultFormats + new org.json4s.ext.EnumNameSerializer(MyEnum)
// Joda Time
implicit val formats = org.json4s.DefaultFormats ++ org.json4s.ext.JodaTimeSerializers.all
JSON structure can be converted to XML node and vice versa. Please see more examples in XmlExamples.scala.
scala> import org.json4s.Xml.{toJson, toXml}
scala> val xml =
<users>
<user>
<id>1</id>
<name>Harry</name>
</user>
<user>
<id>2</id>
<name>David</name>
</user>
</users>
scala> val json = toJson(xml)
scala> pretty(render(json))
res3: {
"users":{
"user":[{
"id":"1",
"name":"Harry"
},{
"id":"2",
"name":"David"
}]
}
}
Now, the above example has two problems. First, the id is converted to String while we might want it as an Int. This is easy to fix by mapping JString(s) to JInt(s.toInt). The second problem is more subtle. The conversion function decides to use JSON array because there's more than one user-element in XML. Therefore a structurally equivalent XML document which happens to have just one user-element will generate a JSON document without JSON array. This is rarely a desired outcome. These both problems can be fixed by following transformation function.
scala> json transformField {
case ("id", JString(s)) => ("id", JInt(s.toInt))
case ("user", x: JObject) => ("user", JArray(x :: Nil))
}
Other direction is supported too. Converting JSON to XML:
scala> toXml(json)
res5: scala.xml.NodeSeq = <users><user><id>1</id><name>Harry</name></user><user><id>2</id><name>David</name></user></users>
Pull parser API is provided for cases requiring extreme performance. It improves parsing performance by two ways. First, no intermediate AST is generated. Second, you can stop parsing at any time, skipping rest of the stream. Note, this parsing style is recommended only as an optimization. Above mentioned functional APIs are easier to use.
Consider following example which shows how to parse one field value from a big JSON.
scala> val json = """
{
...
"firstName": "John",
"lastName": "Smith",
"address": {
"streetAddress": "21 2nd Street",
"city": "New York",
"state": "NY",
"postalCode": 10021
},
"phoneNumbers": [
{ "type": "home", "number": "212 555-1234" },
{ "type": "fax", "number": "646 555-4567" }
],
...
}"""
scala> val parser = (p: Parser) => {
def parse: BigInt = p.nextToken match {
case FieldStart("postalCode") => p.nextToken match {
case IntVal(code) => code
case _ => p.fail("expected int")
}
case End => p.fail("no field named 'postalCode'")
case _ => parse
}
parse
}
scala> val postalCode = parse(json, parser)
postalCode: BigInt = 10021
Pull parser is a function Parser => A
, in this example it is concretely Parser => BigInt
.
Constructed parser recursively reads tokens until it finds FieldStart("postalCode")
token. After that the next token must be IntVal
, otherwise parsing fails. It returns parsed
integer and stops parsing immediately.
Q1: I have a JSON object and I want to extract it to a case class:
scala> case class Person(name: String, age: Int)
scala> val json = """{"name":"joe","age":15}"""
But extraction fails:
scala> parse(json).extract[Person]
org.json4s.MappingException: Parsed JSON values do not match with class constructor
A1:
Extraction does not work for classes defined in REPL. Compile the case class definitions with scalac and import those to REPL.
-
The original idea for DSL syntax was taken from Lift mailing list (by Marius).
-
The idea for AST and rendering was taken from Real World Haskell book.