ConstructingParser adds space between two predefined entities when no other text is present

[cayhorstmann]

scala.xml.parsing.ConstructingParser.fromSource(scala.io.Source.fromString("<div>&lt;&lt;</div>"), false).document
res0: scala.xml.Document = Document(<div>&lt; &lt;</div>)

Note the space between the two <.

[cayhorstmann]

The problem is that predefined entities are turned into Atom(c), where c is the character represented by the entity. For example, < turns into an Atom('<'). This happens in MarkupParser:

case '&' => // EntityRef or CharRef
      nextch(); ch match {
        case '#' => // CharacterRef
          nextch()
          val theChar = handle.text(tmppos, xCharRef(() => ch, () => nextch()))
          xToken(';')
         ts &+ theChar

The &+ operator is defined in NodeBuffer as follows:

def &+(o: Any): NodeBuffer = {
  o match {
    case null | _: Unit | Text("") => // ignore
    case it: Iterator[_]           => it foreach &+
    case n: Node                   => super.+=(n)
    case ns: Iterable[_]           => this &+ ns.iterator
    case ns: Array[_]              => this &+ ns.iterator
    case d                         => super.+=(new Atom(d))
  }
  ...
}

That sounds innocuous, but there is no good way to save such a thing. Calling XML.write or PrettyPrinter.formatNodes ultimately goes through Utilities.sequenceToXML and this branch:

else if (children forall isAtomAndNotText) { // add space
    val it = children.iterator
    val f = it.next()
    serialize(f, pscope, sb, stripComments, decodeEntities, preserveWhitespace, minimizeTags)
    while (it.hasNext) {
      val x = it.next()
      sb.append(' ')
      serialize(x, pscope, sb, stripComments, decodeEntities, preserveWhitespace, minimizeTags)
    }
}

When saving an element that contains only non-text atoms, they are separated by spaces.
The simplest remedy would be to change

ts &+ theChar 

to

ts &+ Text(theChar.toString)

in MarkupParser.

[ashawley]

Good analysis of the problem. Not sure if that change to MarkupParser will fix it though.

The parser will need some work to be smart enough to preserve white space between entities. Obviously, white space shouldn't be preserved, when it's not desired, between other types of XML "nodes". However, that feature seems to be the source of the problem. An Atom is considered a Node, but entities are indeed a special case. They should be treated like text, so any adjacent whitespace should be treated as text.

The MarkupParser doesn't have a method to consume white space characters in to a Text object. The appendText method would be useful here. Unfortunately, the preserveWS affects the behavior of appendText.

[ashawley]

I've taken a first pass at writing unit tests should anyone want to take this one on.

[piyush-jaiswal]

Hi @ashawley, I have tried to fix this issue and also added the tests you provided. It would be great if you could take a look at it. Thanks!