I am worried about performance since I have a fairly large XML file to create on a frequent basis.
Is it best to simply start with a skeleton structure containing dummy data and then just populate all of the nodes Values and attributes ? That structure will have a 1 instance declaration in the case of "occurs 0 or 1 to X" nodes.
The other alternative is to build it up from scratch for each data set to be transformed.
If structural changes are imposed, I don't think one approach has a distinct advantage over the other as far as programming effort goes.
Are there any other issues to consider that I have forgotten about ?