ANTLR duplicate a tree

问题:

I use ANTLR to build a tree (CommonTree) like follwing (language: JAVA):

  Parser.prog_return r = parser.prog();
  CommonTree t = (CommonTree) r.getTree();

Now, I need to pass "t" as a parameter, and make some changes without affecting the original tree. However, with Java's pointer, this could not been done, so I need to duplicate the tree.

I have been search on the internet, the cloested thing I could find is the dupTree() method of ASTFactory class.

Any suggestion or advises on how to achive this would be appreciated!

EDIT

@Bart Kiers, Thanks for your answer, it absolutely works!

I see you are doing a depth first walk over the tree, and create a CommonTree object for each node that was visited.

My question is now, what is the relation between CommonToken and CommonTree, and what are these attributes do:

cTok.setCharPositionInLine(oTok.getCharPositionInLine());
cTok.setChannel(oTok.getChannel());
cTok.setStartIndex(oTok.getStartIndex());
cTok.setStopIndex(oTok.getStopIndex());
cTok.setTokenIndex(oTok.getTokenIndex());

回答1:

Try something like this:

public static CommonTree copyTree(CommonTree original) {
  CommonTree copy = new CommonTree(original.getToken());
  copyTreeRecursive(copy, original);
  return copy;
}

private static void copyTreeRecursive(CommonTree copy, CommonTree original) {
  if(original.getChildren() != null) {
    for(Object o : original.getChildren()) {

      CommonTree originalChild = (CommonTree)o;

      // get the token from the original child node
      CommonToken oTok = (CommonToken)originalChild.getToken();

      // create a new token with the same type & text as 'oTok' 
      CommonToken cTok = new CommonToken(oTok.getType(), oTok.getText());

      // copy all attributes from 'oTok' to 'cTok'  
      cTok.setLine(oTok.getLine());
      cTok.setCharPositionInLine(oTok.getCharPositionInLine());
      cTok.setChannel(oTok.getChannel());
      cTok.setStartIndex(oTok.getStartIndex());
      cTok.setStopIndex(oTok.getStopIndex());
      cTok.setTokenIndex(oTok.getTokenIndex());

      // create a new tree node with the 'cTok' as token
      CommonTree copyChild = new CommonTree(cTok);

      // set the parent node of the child node
      copyChild.setParent(copy);

      // add the child to the parent node
      copy.addChild(copyChild);

      // make a recursive call to copy deeper
      copyTreeRecursive(copyChild, originalChild);
    }
  }
}

...

// get the original tree
CommonTree tree = (CommonTree)parser.parse().getTree();

// create a copy of the tree
CommonTree copy = copyTree(tree);

// change the contents of the right node of the right node of the root 
((CommonTree)tree.getChild(1).getChild(1)).getToken().setText("X");

System.out.println(tree.toStringTree());
System.out.println(copy.toStringTree());

which would produce:

(&& a (|| b X))
(&& a (|| b c))

for the input "a && (b || c)". I.e., the tree has X, but the copy would will have the original contents: c.

Note that I choose CommonTree and CommonToken objects because those are the default Token and Tree implementations. If you choose to create your own Token and/or Tree, chances are that you'll subclass the CommonTree and CommonToken classes, in which case my suggestion would not break.

A CommonTree is nothing more than a wrapper around a CommonToken, holding a bit of extra information: a parent node, and child nodes. That is why I also copy all the information from the CommonToken objects.

回答2:

I had the same issue stumbling over dupTree(), which seemed to be deprecated and then Bart's posting, which steered me into the right direction. I finally ended up with the following, which is using the constructor of CommonTree accepting a CommonTree - abstracting it from the need to copy the individual fields.

  private static CommonTree copyTreeRecursive(CommonTree original) {

    CommonTree copy = new CommonTree(original); // Leverage constructor

    if(original.getChildren() != null) {
      for(Object o : original.getChildren()) {
        CommonTree childCopy  = copyTreeRecursive((CommonTree)o);
        childCopy.setParent(copy);
        copy.addChild(childCopy);
      }
    };
    return copy;
  }

NB: I stuck with Bart's naming convention.