This is a simple utility for "scrambling" the contents of an XML document while preserving the structure of the XML. It only modifies the text nodes or the stuff between the tags. Anything inside the tags (including attribute values) is preserved as-is.
The motivation for writing this was to allow customers with sensitive content in their XSL-FO documents to be able to send us samples that were useful for testing and debugging.
This is actually the third iteration (that I know of) of an XML scrambler developed at Antenna House. It was written to solve two problems:
- Scrambling large documents. The other two scramblers both used a DOM parser so they needed to load the entire document into memory before they could operate on it. By contrast, my implementation uses Expat which is a stream parser. It only loads a small chunk of the document at a time.
- Security. Primarily I was trying to answer the question, "how can we be sure it's truly random"?
The security question is one that I had given some thought and discussed with my coworker, Alex, several times. How do we (or our customers) know it's really random? I was only able to find the source code for one of the two scramblers we were sending out to customers. And without being able to see the source, it would be difficult to determine if the output being produced was really random or merely encrypted with a reversible algorithm of some sort. Something as simple as rot-13 would appear random to the untrained eye, but is, in fact, trivial to reverse. We need a way for customers to verify beyond any doubt that the document was irrecoverably scrambled. This led me to the decision that instead of "scrambling" the data, what I should be doing is "masking" the data.
The way the current "scrambler" works is by replacing all uppercase latin characters with 'X', all lowercase latin characters with 'x', and all digits with '0'. All other characters are passed through unmolested. This allows anyone to, at a glance, verify that the document has been securely stripped of sensitive information while still producing a document that is (usually) suitable for our testing.