Scales Xml, in addition to immutability, aims to focus on these separate concerns:

Structure and manipulation
XML Comprehensions - XPath
Pull API
  // think resumable folds without cps plugin
  type ResumableIter[E,A] = IterV[E, (A, IterV[E,_])]
Comprehensive Serialization Support
JAXP Support

As a disclaimer the only theoretical paper and research used was Huets Zipper paper, which forms the basis of much of Scales Xmls approach. This will change after 0.1 is released, there are many other ideas to build on.

How to use

The M2 style repo for snapshots is at

So for sbt 0.7 its:

val scalesSnapshots = "Scales Snapshots" at ""
val scalesRepo = "Scales Repo" at ""
val scalesXml = "scales" %% "scales-xml" % "0.2.1"

Maven repos should therefore use scales-xml_2.9.1 as the dependency.

The documentation site is here and zip of the site documentation is available at

_Warning_ local file based sites do not work in Chrome, use Firefox or IE in preference.

Simple Usage Examples

To use ScalesXml you must use the following imports (where the objects ScalesUtils and ScalesXml import implicits).

import scales.utils._
import ScalesUtils._
import scales.xml._
import ScalesXml._

Parse and XPaths

  val testXml = loadXml(new FileReader("./tests/data/BaseXmlTest.xml"))

  val path = top(testXml)

   Note that \+ is the only major difference to XPath style syntax
   and allows expansion of child nodes (E1\E2 in XPath requires E2 to
   be evaluated within the context of E1, which is tricky to mimic
   in static code)


  val ns = Namespace("test:uri")
  val nsa = Namespace("test:uri:attribs")
  val nsp = nsa.prefixed("pre")

  val builder = 
      <(ns{ "Elem" }) /@ (nsa("pre", "attr1") -> "val1", // prefixed attribute
      	    	   "attr2" -> "val2", // no namespace attribute
		   nsp { "attr3" } -> "val3") // prefixed attribute
      	    ns("Child"), // no prefix
	    Text("Mixed Content"),
	    <(ns("Child2")) ~> "All previous nodes below are replaced with this text"

  val removed = builder -/@("attr") -/(ns("Child"))

In-Place Transformations

    import Elements.Functions.localName

    val builder = <(ns("i0")) / (ns("i2"), ns("i3"), ns("i40"), ns("i20"), ns("i5"), ns("i7"), ns("i10"), ns("i50"), ns("i11"), ns("i14") )

    val folded = foldPositions(all)( implicit p =>
      localName match {
	// test inserting before the start
	case "i2" => AddBefore(Right( ns("i1") ))
	// replace in the middle
	case "i40" => Replace(ns("i4")) // for some reason a single param makes the inference freak
	// remove in the middle
	case "i20" | "i50" => Remove()
	// after followed by after
	case "i5" => AddAfter(Right( ns("i6") )) 
	// after followed by before
	case "i7" => AddAfter(Right( ns("i8") ))
	case "i10" => AddBefore(Right( ns("i9") ))
	// after followed by before without any previous after
	case "i11" => AddAfter(Right( ns("i12") ))
	case "i14" => AddBefore(Right( ns("i13") ))
	// just copy it over - noop
	case _ @ x => AsIs()

Nested Transformations


  def fooIdBits(i : Int) : Stream[XmlTree] = Stream[XmlTree]( <("foo"l)/@("id" -> i.toString) /( ("bar"l)/@("id" -> "0")) /(
      (("baz"l)/@("id" -> "0", "blah" -> "blah", "etc" -> "etc")) /( (("buz"l)/@("id" -> "0")) ),
      (("buz"l)/@("id" -> "0"))
    ) ).append( fooIdBits( i + 1 ) )
  val fooIdBuilder = <("root"l) /( fooIdBits(1).take(5) )
  // replace every child's id attribute with the id param
  // Note with \.\\ the leading \ is necessary as the first \\ also includes
  // the context node in XPath, and we don't want that here
  def toId( id : String )( op : XmlPath ) = 
    foldPositions( top(op.tree).\.\\.*@("id").\^ ){ p => Replace(Right{ elem(p) /@("id"-> id) toTree}) }
  import Elements.Functions.attributes

  val folded = foldPositions( top(fooIdBuilder).\* )( p => 
    ReplaceWith( toId(attributes(p)("id"l).get.value) ) )

The above attributes(p) and elem(p) can also use an implicitly scoped path.

Pull Parsing Iterator

Shows the basic iterating over a file, upon the last event the file source is closed.

    val pull = pullXml(new FileReader("./tests/data/BaseXmlTest.xml"))

    def out(it : String) : Unit = 
      () // write it to a file, processs the data etc...
    for{event <- pull}{
      event match {
	case Left(x) => x match {
	  case Elem(qname, attrs, ns) =>
	    out("<" + qname + x => " " +"='"+x.value+"'" ).mkString(" ") + ">")

	  case item : XmlItem =>
	case Right(EndElem(qname, ns)) =>
	  out("</"+ qname +">")

    assertTrue("Should have been closed", pull.isClosed)

Pull Parsing Iteratees

Drop all events until the end element of Fred is reached

    val pull = pullXml(reader)

    val isEndFred = (x : PullType) => {
      x match {
        case Right(EndElem(qname, _) ) if qname.local == "Fred" =>
      	case _ => 
    val iteratee2 = dropWhile[PullType]( isEndFred )
    val endOfFred = iteratee2( run

Pull Parsing ResumableIteratees

Note the cont item in the match, this is the continuation Iteratee to process the rest of the xml.

    val iter : Iterator[PullType] = .....

    val QNames = List("root"l, "child"l, "subChild"l)

    val eachSubChild = onDone(List(onQNames(QNames)))

    def processSubChild( res : ResumableIterList[PullType,QNamesMatch]) = 
      res match {
	case Done(((QNames, Some(x)) :: Nil,cont), y)  => 
	  // use the resulting Path, each child below subChild is captured
	case _ => // any other combination is likely Eol in this example

    var res = eachSubChild(iter).eval // to use eachSubChild(iter) eval a new line must follow
    // extract cont from the match, and process it again.
    res = extractCont(res)(iter).eval

As Iteratees are composable onDone uses this property and ResumableIter to allow nesting of many different folds. When an Iteratee returns Done its resumable state is included with the state of each Iteratee in the input list, allowing restart of all of the lists Iteratees as if they operated alone.

In the above example the QNames are returned as well from onQNames, allowing the caller to identify which of the QName lists actually matched onDone.

This allows constant space progress through XML with the following helpful and automatic collection patterns:

Alternating and Repeating Elements

      <ofInterest> <!-- Collect all of these -->
      <alsoOfInterest> <!-- Collect all of these -->
	just some text

It should be noted that monadic serial composition of onQNames would also work here, onDone is not absolutely necessary, although as we will see it is more general..

Grouped Repeating

      <ofInterest> <!-- Collect all of these -->
      <alsoOfInterest> <!-- Collect all of these -->
	just some text

Repeating nested

      <ofInterest> <!-- Collect all of these -->
          <smallKeyValues> <!-- Collect all of these -->

Sectioned Grouped Repeating, onDone calling onDone

      <!-- Necessary for processing the below events -->
      <sectionHeader>header 1</sectionHeader>

      <ofInterest> <!-- Collect all of these -->
      <ofInterest> <!-- Collect all of these -->
      <ofInterest> <!-- Collect all of these -->
      <!-- Necessary for processing the below events -->
      <sectionHeader>header 2</sectionHeader>

Its possible using onDone with onQNames to process the above document with a single call to:


and the events will be fired in the correct order. The only unpleasant issue is that a stack of current sectionHeader must be kept, which again looks like a fold.

Sectioned with Folds

  val Headers = List("root"l,"section"l,"sectionHeader"l)
  val OfInterest = List("root"l,"section"l,"ofInterest"l)

  val ofInterestOnDone = onDone(List(onQNames(Headers), onQNames(OfInterest)))

  val total = foldOnDone(xml)( (0, 0), ofInterestOnDone ){ 
    (t, qnamesMatch) =>
    if (qnamesMatch.size == 0) {
      t // no matches
    } else {
      // only one at a time
      assertEquals(1, qnamesMatch.size)
      val head = qnamesMatch.head
      assertTrue("Should have been defined",head._2.isDefined)
      // we should never have more than one child in the parent
      // and thats us
      assertEquals(1, head._2.get.zipUp.children.size)

      val i =  text(head._2.get).toInt
      if (head._1 eq Headers) {
	assertEquals(t._1, t._2)
	// get new section
	(i, 1)
      } else (t._1, i)
  assertEquals(total._1, total._2)

However often its easier to structure the code as a for comprehension upon xml.

For/Monad Comprehension - parsing SVN logs

Sometimes a foreach or flatMap the most appropriate choice for a developer to use.

    val LogEntries = List("log"l,"logentry"l)

    val bits = for{ entry <- iterate(LogEntries, xml).view
	revision <- entry.\.*@("revision"l).one // ensure its only got one revision
	author <- entry.\*("author"l).one
	path <- entry.\*("paths"l).\*("path"l) // more than one path is allowed
	kind <- path.\.*@("kind"l)
	action <- path.\.*@("action"l)
    } yield (text(revision), value(author), text(kind), text(action), value(path))

Bits is lazy in this case, remove the .view and its eager, but will not retain memory used for xml parsing (outside of any unpleasant substring reuse leaks).

NB: instead of .one which implicitly forces that only one matches, oneOr can be used allowing throwing of exceptions, logging etc, calling one is optional when the developer knows the data has only one.

Combination Fold and For Comprehension - parsing svn logs

A benefit of Scales XML is that the types for xml are the same for both XML Pull and Push. In particular the developer need not care what produced the XPath.

The path processing logic can therefore be separated from what generated it.

    val LogEntries = List("log"l,"logentry"l)
    val ionDone = onDone(List(onQNames(LogEntries)))

    val entries = foldOnDone( List[(String,String,String,String,String)](), ionDone ){ 
      (t, qnamesMatch) =>
	if (qnamesMatch.size == 0) {
	  t // no matches
	} else {
	  val entry = qnamesMatch.head._2.get
	  val bits = for{ 
			 revision <- entry.\.*@("revision"l).one // ensure its only got one revision
			 author <- entry.\*("author"l).one
			 path <- entry.\*("paths"l).\*("path"l) // more than one path is allowed
			 kind <- path.\.*@("kind"l).one
			 action <- path.\.*@("action"l).one
		       } yield (text(revision), value(author), text(kind), text(action), value(path))
	  t ++ bits

Position based pulls

When the xml contents itself is unknown and the processing is dependent on the type it can be useful to identify based on position information, for example what the root element is or a doc-literal first element in a soap message.

Another possible scenario is that you know you are only interested in a given message element but you don't want to parse a 50mb xml file to find out if it was that message type.

Because of these two use cases it is possible to perform a search based on position information. doc-literal SOAP identifying would be List(2,1) representing Envelope, (Header, Body), and finally the request node. The first root position is assumed.

  var res = skip(List(2, 1))(iter) run
  val path = res.get // can be None (see below)
  println("Request nodes qname "+ Elements.Functions.qname(path))    

The result from skip is not a ResumableIterable and simply returns Option[XmlPath]. If the stream runs out or its no longer possible to get that position it is None. Only as much of the stream is read as needed, it will stop on the Left(Elem) event.

skip also has a variable arg version, so skipv(2,1) is also usable.

Capturing and restarting / Buffering Pull Events

If the developer wishes to "peek" deep into an event steam then the events must be captured to allow replaying. This allows, for example, using qname or index (then presumably qname) based matching to identify a message type and the correct processing option.

The correctly identified processing can then restart from the beginning with the expectation of the message structure.

A simple example is processing soap messages based on the first body element, you may want to choose different code paths based on this, but require elements in the header to do so. The usage is simple:

  val xmlpull = // stream capture

  val captured = capture(xmlpull)
  // either the path or None if its EOF or no longer possible
  val identified = skip(List(2, 1))(captured) run

  val processor =

  // restart the stream from scratch