I have a tag extension which produces html containing a data-* kind of attribute. It looks like the mediawiki parser later parses and modifies the contents of this attribute (it adds some PRE, BR and P tags). So I base64-encode that attribute (and later in the code, where the attribute is needed by some Javascript I base64-decode it again). This workaround is working :-)
Now I found Manual:Tag extensions#How can I avoid modification of my extension's HTML output? which probably is a better solution (no base64 encode-decode). However, when I wrap my HTML with an array( "..html...", "markerType" => 'nowiki') and have my extension return the parser nevertheless tampers with the attribute (adding some BR and P tags, but no PRE tags). So I am back with my base64 method for the moment. Which is *not* elegant.
What am I doing wrong? What did I misunderstand from the manual entry?