// transformer can return null or input on no change
// will not call transformer on individual words
sS lambdaMapLike gazelle_deepTransform(IF1<S> transformer, S input, O... _) {
  if (transformer == null) ret input;
  
  // call transformer on whole input, return if change
  S transformed = transformer.get(input);
  if (transformed != null && transformed != input) ret transformed;

  // transformer didn't do anything, tokenize & go through parts
  O tokenizer = optPar tokenizer(_, f javaTokWithBrackets);
  LS tok = callTokenizer(tokenizer, input);
  
  bool changes;
  for ping (int i = 1; i < l(tok); i += 2) {
    S t = tok.get(i);
    if (tok_isRoundOrCurlyBracketed(t)) {
      S unBracketed = tok_deRoundOrCurlyBracket(t);
      transformed = gazelle_deepTransform(transformer, unBracketed, _);
      if (transformed != unBracketed) {
        if (!changes) tok = cloneList(tok); // in case a caching tokenizer was used
        
        // enclose in brackets & replace
        tok.set(i, startsWith(t, "(") ? roundOpt(transformed) : curlyOpt(transformed));
        set changes;
      }
    }
  }
  
  ret changes ? join(tok) : input;
}