I filed https://issues.apache.org/jira/browse/SHINDIG-1107 Does anyone have any opinion about cleaning up those dependencies? We were pulling in json-lib which seems unnecessary since we have a native json serializer in place now.
Another simplification is deprecating nekohtml for htmlparser, which is used by caja. I asked the caja folks about using neko and this was their response: htmlparser was recommended by Ian Hickson, author of large chunks of the HTML5 spec as conforming closely to the spec. Nekohtml is indeed quite fast but htmlparser does a better job of more accurately producing the kind of DOM that you would get in an actual browser (which is what we're trying to codify) when parsing tag soup. Mike Samuel looked at nekohtml more recently (primarily to see if we could benefit from faster parsing by neko) and improved our own parsing speed to a point where it is comparable to neko. I am not sure I fully follow the benefit of removing dependency on icu4j.
