compareDocumentPosition issues

I am the original implementor of (the since renamed) compareTreePosition
in Mozilla.  I'm interested in updating our implementation to the latest
draft, but have a few issues with the way the 20021022 draft is defined.

First, the following constants (among others) are defined on the Node

     const unsigned short DOCUMENT_POSITION_DISCONNECTED            = 0x00;


And, Definition group DocumentPosition says:

     If the two nodes being compared are the same node, then no flags
     are set on the return.

If I were using the JavaScript language (as all my examples will use),
it would then follow that:

   var flags = foo.compareDocumentPosition(foo);
   flags == 0; // true
   flags == Node.DOCUMENT_POSITION_DISCONNECTED; // true!?!?

Clearly, this seems to be a problem.  If the SAME_NODE bit has been
removed, then that is fine, but two identical nodes should not be

Further on in the Definition group DocumentPosition:

     If there is no common container node, then the order is based upon
     order between the root container of each node that is in no container.
     In this case, the result is disconnected and implementation-dependent.
     This result is stable as long as these outer-most containing nodes remain
     in memory and are not inserted into some other containing node. This
     would be the case when the nodes belong to different documents or
     fragments, and cloning the document or inserting a fragment might change
     the order.

So, with another example, if I have two nodes foo and bar which are in
different documents:

   var flags = foo.compareDocumentPosition(bar);
   flags & 0; // false
   flags & Node.DOCUMENT_POSITION_DISCONNECTED; // false!?!?

In general, I do like the way the WD is going with the updated version of
compareDocumentPosition.  However, unless I misunderstood the latest WD,
by removing the SAME_NODE constant and by allowing disconnected nodes to
have other bits set, it appears there is a need to give disconnected its
own bit, instead of none.  That would solve both of the issues I raised.



Received on Thursday, 31 October 2002 17:57:01 UTC