Custom traversal possibility via 'xml_tree_walker' interface

[DanN91]

I am referring to this method: PUGI__FN bool xml_node::traverse(xml_tree_walker& walker)

As I see now in the code, before calling for_each, we have this line xml_node arg_for_each(cur) which wraps the current node and thus a copy of the current node reference is passed to for_each.

The limitation with this is that if I want to skip the traversal of, let's say, all the siblings of a specific node inside my for_each implementation for the xml_tree_walker which I provided for traversal, I know not of a way to do this.

This is due to the fact that after the call to walker.for_each(arg_for_each), what is used in the traversal logic is not arg_for_each, but cur and so, what I do inside my implementation in order to change the node which would be traversed next ( node = node.parent().next_sibling() ), has no effect on the traversal logic.

My questions is if the way things are done now has a good purpose in making a copy of the reference to the current node. And if not, can you provide this custom traversal functionality? I think it's of benefit when you want to optimize the traversal based on some criteria...

[zeux]

This is somewhat related to #103; I was hoping to provide a functionality for skipping subtrees as part of the new interface.

As for the reference issue, the way it happened is mostly a historical accident - logically, the intention has always been to use a parameter passed by value. Unfortunately, pugxml (which pugixml is a fork of from more than 10 years ago) used a non-const reference argument and I failed to notice it until it was too late in that the backwards compatibility was a significant concern so changing this wasn't possible.

The internal implementation doesn't use xml_node to begin with - it uses a lower level xml_node_struct - so the code naturally has to wrap the object in a node handle that is then discarded.

[DanN91]

I will close this because it's more than covered in #103;

Thank you for considering this.