Saxon

Casual experiments with upgrading to 1.4.0 were unsuccessful. It will still build if we reduce the dependencies to only axiom-dom and axiom-impl, but 1.4.0 appears to introduce a new node type for CDATA. Attempting to treat CDATA as text is only partially successful as the resulting text nodes don't get merged.

Looking at the JDOM2 interface, I see that managing adjacent text nodes is spread across a few different methods. It's not immediately clear if the Axiom model can be approached in the same way.

Curiously, 1.2.15 has a OMNode.CDATA_SECTION_NODE but doesn't use it? Or maybe the Axiom API has methods for merging adjacent nodes and those have changed in 1.4.0?

According to the Axiom docs,

Preserving CDATA sections during parsing

By default, StAXUtils creates StAX parsers in coaelescing mode. In this mode, the parser will never return two character data events in sequence, while in non coaelescing mode, the parser is allowed to break up character data into smaller chunks and to return multiple consecutive character events, which may improve throughput for documents containing large text nodes. It should be noted that StAXUtils overrides the default settings mandated by the StAX specification, which specifies that by default, a StAX parser must be in non coalescing mode. The primary reason is compatibility: older versions of Woodstox had coalescing switched on by default.

A side effect of the default settings chosen by Axiom is that by default, CDATA sections are not reported by parser created by StAXUtils. The reason is that in coalescing mode, the parser will not only coaelsce adjacent text nodes, but also CDATA sections. Applications that require correct reporting of CDATA sections should therefore disable coalescing. This can be achieved by creating a XMLInputFactory.properties file with the following content:

javax.xml.stream.isCoalescing=false

But using System.setProperty to change values of javax.xml.stream.isCoalescing doesn't seem to have any effect in either version of the API.

Changed AxiomLeafNodeWrapper to treat a CDATA node on the child axis in the same way as a text node (ie, wrapping it in a NodeInfo of type text).

Changed AxiomTreeTests to set mergesAdjacentTextNodes to false, so the test no longer expects adjacent text nodes to be merged (the Axiom tree wrapper, like the DOM4J tree wrapper, makes no attempt to do this)