Right now we feed every supported part type into custom extractors, even if we know we can only handle specific types. This is mainly a problem for the text-based extractor, as all types can be converted to that.