I pointed out to a client that they're seeing failures parsing XML because some of the element content that they're producing contains characters illegal in XML content, like "&" (unencoded). They acknowledged that should be fixed, but they also said they could instead enclose all content with CDATA blocks. That seems bizarre to me, but I'm not sure I can immediately come up with all the cogent arguments against that. Can someone summarize specifically why you should NOT do that?