It depends mostly on the reason why target code is used in the grammar. Is it only action code to do something with the found tokens (e.g. building a symbol table or alternative tree representation) then is indeed no problem do remove such native code and do the processing afterwards (using a parse tree walker or visitor).
However, predicates are a different. They are used to guide the parser and also require native code. What you can do is to move all the native code into a base class from which your generated parser derives. You then only need to re-write this base class in your target language and keep the grammar mostly free of native code (except for a single function call, which invokes the native code).
This approach has the advantage that no additional library reference is necessary (#include in C/C++, import in other languages), which also is native code preventing use for multiple targets.