use XML::Rules instead of XML::Simple if you want sensible and predictable data structures :)

use Data::Diver to drill down into complex data structures

You could use block eval to prevent death, and ref to test what type of reference you have , and wrap that up in a subroutines ... but you're inventing Data::Diver :) so just use Data::Diver

See Re: Parsing SOAP::Lite results (with Data::Diver), Re: $var{el} += 1 vs $var{el} = $var{el} + 1, Re: Problem with counting the nodes in a hash created by XML::Simple (use XML::Rules, Data::Diver )

See Re: XML::Simple Not getting the proper output. ( xml2XMLRules, Re^3: Parsing XML (XML::Rules xml2XMLRules )

See xml2XMLRules the better the rules it spits out for you to plug into XML::Rules, the more stable the resulting data structure

or use XML::LibXML + xpather ... see all the links here Re: Retrieve select information from HTML, they're examples(for tree-xpath and others)/walkthroughs/tutorials ... tools like xpather.pl/htmltreexpather.pl can give you paths to start with