Choosing a Tree Model

Saxon provides two implementations of the internal tree data structure (or tree model). The tree model can be chosen by an option on the command line (-dt for the tiny tree, -ds for the standard tree) or from the Java API. The default is to use the tiny tree model. The choice should make no difference to the results of a transformation (except the order of attributes and namespace declarations) but only affects performance.

There is an exception to this: the so-called "standard" tree model (-ds) does not support type annotations. If you use Saxon-SA to annotate element and attribute nodes, use the "tiny tree" (-dt).

Generally speaking, the tiny tree model is faster to build but slower to navigate. It therefore performs better when you visit each node on the tree once or less. The only case where the tiny tree is known to be significantly slower than the standard tree is when there is a large "fan-out" (a parent node with thousands of immediate children) combined with very frequent use of the parent axis.

The tiny tree model gives most benefit when you are processing a large document. It uses a lot less memory, so it can prevent thrashing when the size of document is such that the standard tree doesn't fit in real memory.

If in doubt, stick with the default.

Expand

Next