There are two things that diffxml understands that diff doesn’t; the syntax of XML documents (e.g. <br/> is equivalent to <br></br>) and the hierarchical structure they represent.

The advantages of understanding XML syntax are pretty easy to explain. Consider these two XML documents:

<a
>text<b/>
<c></c>
</a>

and

<a>text<b></b>
<d/>
</a>

If we compare these using diff, we get the following output:

1,3c1,2
< <a
<     >text<b/>
< <c></c>

Which is telling us that every line in the document has changed. However, if we use diffxml to difference the documents, we get:

<?xml version="1.0" encoding="UTF-8" standalone="no"?>
<delta>
<insert charpos="2" childno="4" name="d" nodetype="1" parent="/node()[1]"/>
<delete node="/node()[1]/node()[5]"/>
</delta>

Which is telling us that the difference between the documents is the insertion of an element “d” and the removal of another element¹.

The other major advantage of diffxml is that it understands the hierarchical, or “tree” structure of XML documents. It’s a little harder to explain what this means, but consider the following. The XML document:

<a><b><d/></b><c><e/></c></a>

Can be represented as:

And the XML document:

<a><b/><c><d/><e/></c></a>

Can be represented as:

It’s clear from the diagram that the only change is that the element “d” has moved from element “b” to element “c”. There is no way that a line-based differencing utility could tell us this, but diffxml gives us:

<?xml version="1.0" encoding="UTF-8" standalone="no"?>
<delta>
<move childno="1" new_charpos="1" node="/node()[1]/node()[1]/node()[1]" old_charpos="1" parent="/node()[1]/node()[2]"/>
</delta>

Which correctly identifies that the only difference is the move of a single element to a new parent.

I hope this makes it clear why tools such as diffxml which understand the heirarchical narture of XML documents are often a better choice than line-based equivalents for comparing XML documents.

1. Admittedly the output is a little hard for humans to read currently. There are a couple of things that can be done to improve this (use proper node names instead of using the node() axis and put a string in the nodetype attribute), but in the future I hope to provide some sort of graphical interface.