Welcome to the Monastery. If you want to be able to find out which files have non-empty "fail-message" elements, so that you can see how many different values there are for fail messages, you probably want to use XPath expressions (which you can look up elsewhere - it's not strictly a Perl thing).

For example, a while back I figured out how to use the XPath facilities in XML::LibXML, and it was so easy and so cool, I posted some sample code for a generic tool to take an XPath expression and an XML file as command-line args, and output the portions of a given file (if any) that matched a given XPath: Re: XPath command line utility....

Consider a sample XML file like this (let's call this file "test.xml"):

<foo>
<bar id="t1">
<fail-message></fail-message>
</bar>
<baz id="t2">
<fail-message>yar</fail-message>
</baz>
</foo>
[download]

If I want to extract non-empty "fail-message" elements from that sort of XML data, the XPath expression would be:

//*[fail-message!='']
[download]

To apply that expression using my "exp" tool, the (bash) command line would be:

exp -p "//*[fail-message!='']" test.xml
[download]

Since I might want to see all the markup (including attributes) for nodes that match the expression, I included a '-x' option on my "exp" script to do just that. In this case, seeing the markup would be helpful if I want to look at the empty fail-messages (by using "=" instead of "!=" in the XPath expression).