And we hit again the issue of #Mozilla #Firefox #XSLT engine not supporting disable-output-escaping='yes'
«But we process directly to DOM because it's faster, we can't support disable-output-escaping with that, it requires parsing the output of the template application.»
«Oh yeah because going the slow path only when output escaping is disabled is SO HARD. It's not like we have innerHTML in JavaScript or something.»
I have actually found a solution and it's as horrible as you'd think:
<xsl:variable name='inner'
select='document(concat("data:text/html;charset=UTF-8,<body>",atom:content,"</body>"))'/>
<xsl:apply-templates select='$inner/body/*'/>
So basically what I do is turn the atom feed content into data: uri and apply my templates to it. This is a bit fragile, since some of the special templates may apply still; the solution, if this is ever an issue, will be to specify different modes based on whether we're processing inner or outer data.
This is also a very good detector of broken XML pages, since they fail to parse and don't appear 8-)