W3C home > Mailing lists > Public > www-dom@w3.org > April to June 1998

RE: Alternative to the Live NodeIterator

From: David Brownell <David.Brownell@Eng.Sun.COM>
Date: Fri, 29 May 1998 15:13:01 -0700
Message-Id: <199805292213.PAA18325@argon.eng.sun.com>
To: amarshal@usc.edu
Cc: www-dom@w3.org
[ Still finishing digging out from a few weeks worth of e-mail ... ]

> From: "Andrew n marshall" <amarshal@usc.edu>
> Date: Sun, 3 May 1998 10:45:02 -0700
>	[
>	  Re problems with iterator liveness; David was saying that
>	  liveness is less of a problem than "between nodes" semantics
>	]
> > > 2) The current node is undefined if it is deleted. The spec could
> > > define it in some way or another, but as I recall the discussion
> > > none of the options were without problems.  Plus, this STILL
> > > requires that iterators update
> > > their state whenever a delete occurs ...
> >
> > Hmm, my reading says that there are no nodes in a "deleted" state;
> > they're just detached from some other node.  So there's no need
> > to have iterators interacting with Node.remove!  (I just persuaded
> > myself of that ...)
> Unfortunately it isn't that easy.  If the iterator was acting as
> a filtered snapshot, you would be correct.

Actually, in all cases -- think about the null filter!  And I was
not assuming snapshot semantics at all; see later.

>	However, the iterator is currently defined to be live, with
> a filter relative to the to either an element or the Document
> (obviously this should be managed by your proposed revision to
> the document fragment).  Therefore removed nodes are also removed
> form the context of the filter and should not be returned by
> the iterator.

Right, but since it's live they _won't_ be returned!

> > That is, one would define iterators as encompasing a "current node" and
> > a "traversal algorithm" (such as "siblings", "children", "preorder
> > depth first"; perhaps filtering out some kinds of nodes).  Then there
> > would be two cases in node removal, ones that are familiar to anyone
> > who's ever edited a linked list:
> I'm not sure I understand you.

Consider a tree of nodes navigable with primitives like nextSibling,
firstChild, parent.  (Like DOM.Node!)  Use any node as iterator state.
The iterator itself has a "next" primitive (and maybe others).

If it's doing filtering ("only nodes with tag 'P'") the "next" primitive
just uses those node navigation primitives to find the next node,
skipping all that don't meet the filter criteria.  For example, if this
iterator was just for the parent's children, it only uses the node's
nextSibling primitive ... perhaps skipping those without a "P" tag.

    class Iterator {
	private Node current;

	public Iterator (Node start) { current = start; }

	public Node getCurrent () { return current; }

	public Node getNext () {
	    Node temp = current;

	    do {
		temp = internalGetNext (temp);
	    } while (temp != null && !isInteresting (temp));
	    current = temp;	// or, could save previous
	    return current;

	// default traversal: preorder depth first ... "TreeIterator"
	// subclasses could override to show just siblings, etc
	protected Node internalGetNext (Node start) {
	    Node	temp;
	    if (start == null)
		return start;
	    if ((temp = start.getFirstChild ()) != null)
		return temp;
	    do {
		if ((temp = start.getNextSibling ()) != null)
		    return temp;
		start = start.Parent ();
	    } while (start != null);	// stop at document root
	    return null;

	// default does no filtering
	// subclasses could insist on particular tags, types, etc
	protected boolean isInteresting (Node n) { return true; }

And of course, filtering out tags of some sort could look like:

    class TagIterator extends NodeIterator {
	private String tag;
	public TagIterator (Node n, String tag) {
	    super (n);
	    this.tag = tag;
	protected boolean isInteresting (Node n) {
	    return (n instanceof Element)
		&& (((Element)n).getTag ().equals (tag));

If one deletes any Node except the current one, it doesn't show up
from the iterator, regardless of filtering.  And if it's the current
node that's deleted, the two cases are:

> >     - You removed the node since it's the one you're interested in.
> >       Don't touch the iterator; remove the current node from its
> >       parent, and that iterator just shows the good parts.
> What are 'the good parts'?  And if you you don't touch the iterator,
> isn't it still pointing to the removed node?

For example:  After an iterative search, you concluded that THIS
paragraph is the one you wanted; it's the "good parts".  Remove
that node, you have the result of your search.  Put it someplace
else if you like, same document or another.

Yes, the iterator is still pointing inside of it -- that's what
you wanted, by definition of this scenario.  Continue to use it
to look at what's inside this paragraph, or perhaps what follows
it after you've relocated the paragraph elsewhere.

> >     - You removed the node since it's the one you're NOT interested in.
> >       Before you remove it, move the iterator to some other node.
> Why would the iterator ever point to a node it is not interested in?

For example, you might want to delete all sections that don't talk
about some particular topic.  The iterator scans sections, and you
found that the "current" one doesn't talk about that topic.  So it
gets nuked ... after the iterator's relocated, since you haven't
finished looking at all the sections.

> > Completely natural for single threaded code, and it's got relatively
> > nice behaviour for concurrent mutating/traversing threads (assuming
> > they coordinate somehow).
> Could you please re-explain how you intend to implement iterators/node
> removal without interaction between the two.

See above.  "Without interaction" is meant in the sense of "no special
lines of code or specification need to be written".  That's because the
iterator has a very clear semantic:  It's (a) a current node, coupled
with (b) some traversal algorithm (depth-first preorder, siblings,
children, etc), and optionally (c) a predicate used to filter out any
nodes that aren't desired (e.g. by tag or type).  Given that, it's
got clear and natural semantics when any node is removed.

- Dave
Received on Friday, 29 May 1998 18:12:45 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 20 October 2015 10:46:04 UTC