dylemma / xml-spac

Handle streaming XML data with declarative, composable parsers



Build Status Scaladoc

XML Streaming Parser Combinators is a Scala library for creating event-based StAX parsers that are:

  • Declarative - You write what you want to get, not how to get it.
  • Immutable - Parsers that you create have no internal state.
  • Composable - Combine and transform parsers to handle complex data structures.
  • Fast - With minimal abstraction to get in the way, speed rivals any hand-written handler.
  • Streaming - Parse huge XML documents from events, not a DOM.

You can jump into a full tutorial, or check out the examples, but here's a taste of how you'd write a parser for a relatively-complex blog post XML structure:

val PostParser = (
	Parser.forMandatoryAttribute("date").map(commentDateFormat.parseLocalDate) and
	Splitter(* \ "author").first[Author] and
	Splitter(* \ "stats").first[Stats] and
	Splitter(* \ "body").first.asText and
	Splitter(* \ "comments" \ "comment").asListOf[Comment]

Get it!

Add the following to your build.sbt file:

libraryDependencies += "io.dylemma" %% "xml-spac" % "0.5"

Main Concepts

The classes you'll interact with most in XML SPaC are Parser, Transformer, Splitter, and Consumer.

A Consumer[In, Out] knows how to consume a stream of In events, producing an Out result.

A Parser[Out] is like a Consumer of XMLEvents which wraps its results in a Try[Out].

A Transformer[A, B] turns a stream of A events into a stream of B events.

A Splitter[In, Context] combines with a Context => Parser[Out] to create a Transformer[In, Out]. A Splitter "splits" a stream into "substreams", assigning a context value to each substream. A parser can be run on each of the substreams, and the resulting values become the contents of the transformed stream.

Note that each of the four classes mentioned here are completely immutable, which means they can be shared and combined without worrying about state or thread safety. All of the mutable stuff is encapsulated in a Handler instance, which is created each time you want to process a stream. For the most part, you won't need to interact with Handlers; the library will do it for you.


Parser.forMandatoryAttribute("foo") is a parser which will find the "foo" attribute of the first element it sees.

val parser = Parser forMandatoryAttribute "foo"
val result = parser parse """<elem foo="bar"/>"""
assert(result == Success("bar"))

Suppose you have some XML with a bunch of <elem foo="..."/> and you want the "foo" attribute from each of them. This is a job for a Splitter. You write an XML Splitter sort of like an XPATH, to describe how to get to each element that you want to parse.

val xml = """<root><elem foo="bar"/><elem foo="baz"/></root>"""
val splitter = Splitter(* \ "elem")
val transformer = splitter through parser

val rootConsumer = transformer.consumeToList
assert(rootConsumer.consume(xml) == List("bar", "baz"))

val rootParser = transformer.parseToList
assert(rootParser.parse(xml) == Success(List("bar", "baz")))

Check out the docs for ContextMatcherSyntax, which defines helpers for creating the arguments to a Splitter, and Parser, which provides lots of convenience implementations of xml parsers.

Under the Hood

The core classes are able to be immutable because they act as factories for "handlers", which contain all of the mutable state and stream processing logic. The Handler class is fairly simple:

trait Handler[-In, +Out] {
	def isFinished: Boolean
	def handleInput(input: In): Option[Out]
	def handleError(err: Throwable): Option[Out]
	def handleEnd(): Out

When the stream wants to send a value, it calls handleInput. When the stream ends, it calls handleEnd. When something goes wrong, it calls handleError. If the handleInput or handleError methods returned a Some, the stream can stop and clean up. If the handler isFinished at any point, the stream can stop and clean up.

Now, I said "stream" a lot in the previous paragraph, but there is no "Stream" class in XML SPaC. Instead, there's the ConsumableLike[S, In] typeclass, which says that S is a stream of In. For example, List[A] could be a stream of A, so there would be an instance of ConsumableLike[List[A], A]. An InputStream could be a stream of XMLEvent, so there would be an instance of ConsumableLike[InputStream, XMLEvent].

trait ConsumableLike[-S, +In]{
	def apply[Out](source: S, handler: Handler[In, Out]): Out

The apply method asks the source (stream) to drive the handler until it produces a result Out.

There are many different ConsumableLike instances already, including generalized ones for Iterable collections and Iterators, and XML-specific ones for String, File, and InputStream. If you have a more specific "Stream" type, you can write your own ConsumableLike[StreamType, EventType].

Here's how the core classes act like handler factories:

trait Consumer[-In, +Out] {
	def makeHandler(): Handler[In, Out]

trait Parser[+Out] {
	def makeHandler(): Handler[XMLEvent, Try[Out]]

trait Transformer[In, B] {
	def makeHandler[Out](downstream: Handler[B, Out]): Handler[In, Out]