The javadocs for both WDF and WDGF include a pretty detailed discussion about the proper use of the "combinations" parameter, but no such parameter exists. I don't know the history here, but it sounds as if the docs might be referring to some previous incarnation of this filter, perhaps in the context of some (now-defunct) Solr configuration.
I think it sounds as if there is some sound wisdom underlying the advice in the docs that is worth preserving, but it needs to be updated to match the current state of the code. I can take a stab at rewriting, but I want to make sure I understand the intent of the comment there. Essentially what it is saying is that a typical usage of WD(G)F is an asymmetric setup where splitting and subsequent token generation is done when indexing, but something less aggreesive (at least no generation, maybe also no splitting) is done when querying. I would probably recommend simply omitting this filter from query-side analysis. Is there a consensus on the best way to use this filter today? -Mike