Different ways to read XML

One reason I originally went with XML is not having to write a parser. Well that's true... up to a point. You still have to recognize the different element names and attributes, and with a fairly complex DTD like XMLTV's that's quite a lot of code.

AFAICT there are three styles of interface to the XML parser:

• DOM, or tree-based. This parses the whole XML document into memory and then you call methods with names like getAttributeNode() to traverse it. Flexible, but slow and unscalable because the whole tree must be loaded into memory. Of course this is what we use :-). Apparently a C DOM implementation like Xerces would be faster than the pure-Perl XML::DOM, but I couldn't get it working.
• Token-based, for example SAX. Here you get a token at a time and have to do most of the work yourself. But you can read files gradually. Instead of SAX you can get really close to the metal and talk to expat directly.
• Generate a parser from the DTD. This is ideal but there's no tool which does exactly what I want. FleXML might come close.

Current answer is a layer on top of XML::DOM. XMLTV.pm table-based DOM handler routine thingies. You can call parsefile() to read a whole file into a Perl data structure and write_data() to write it out again. Should be fairly straightforward to replace the DOM code by something else. Also possible to cheat by serializing Perl data structures to disk.