1

The Visitor Design Patternand Java Tree Builder

Cheng-Chia Chen

2

The Visitor Pattern

• A design pattern enabling the definitions of new operations on an object structure without the requirement to change the classes of the objects.

• Separate (the specification of) extrinsic operations on an object structure from the structure of objects.

3

UML Representation of Visitor Pattern

4

Visitor Structure

• Different elements have different concrete interfaces• Element abstract base class adds accept interface• Double hand-shake between concrete element and visitor allows

visitor to call the appropriate concrete element method

Element

accept ()

ElementA

accept (); A_method ();

Visitor

ConcreteVisitor

visitA () = 0;visitB () = 0;

visitA ();visitB ();

ElementB

accept (); B_method ();

5

Visitor Interactions

A_method ();

visitA ();

accept ();

ElementB ConcreteVisitorElementA

B_method ();

visitB ();

accept ();

6

Example : integer List

public abstract class List { }

public class NullList extends List {}

class ConsList extends List { int head; List tail;}

7

Operation : summing an integer list• First Approach: Instanceof and Type Casts List l; // The List-object int sum = 0; boolean proceed = true; while (proceed) { if (l instanceof NullList) proceed = false; else if (l instanceof ConsList) { sum = sum + ((ConsList) l).head; l = ((ConsList) l).tail; // Notice the two type casts! }}Advantage: The code is written without touching the classes NullList and ConsList.Drawback: The code constantly uses type casts and instanceof to determine what class of object it is considering.

8

2nd approach : attach a method to each class

• The first approach is not object-oriented!• To access parts of an object, the classical approach is to

use dedicated methods which both access and act on the subobjects.

abstract class List { int sum(); }

• We can now compute the sum of all components of a given List-object l by writing l.sum().

9

2nd approach (continued)

class NullList extends List { public int sum() { return 0; } }

class ConsList extends List { int head; List tail; public int sum() { return head + tail.sum(); } }

Advantage: The type casts and instanceof operations have disappeared, and the code can be written in a systematic way.

Disadvantage: For each new operation (say, multiply all integers, count #integers, print all integers,…) on List-objects, new dedicated methods have to be written, and all classes must be recompiled.

10

Third Approach: The Visitor PatternThe Idea:• Divide the code into an object structure and a Visitor • Insert an accept method in each class. Each accept

method takes a Visitor as argument. • A Visitor contains a method visit(C x) for each possible

object class C. (overloading!).

abstract List { void accept(Visitor v); }

interface Visitor { void visit(NullList x); void visit(ConsList x); }

11

Third Approach (continued)

• The purpose of the accept methods is to invoke the visit method in the Visitor which can handle the current object.

class NullList extends List { public void accept(Visitor v) { v.visit(this); } //1}class ConsList extends List { int head; List tail; public void accept(Visitor v) { v.visit(this); } //2}Note: 1 and 2 are the same and can be lifted to parent

List.class List {public void accept(Visitor v) // default accept implem. { v.visit(this); } }

12

Third Approach : (continued)• The control flow goes back and forth between the visit methods in

the Visitor and the accept methods in the object structure.class SumVisitor implements Visitor { int sum = 0; public void visit(NullList x) {} public void visit(ConsList x) { sum = sum + x.head; x.tail.accept(this);}}.....SumVisitor sv = new SumVisitor();l.accept(sv);System.out.println(sv.sum);• Notice: The visit methods describe both (1) actions, and (2) access

of subobjects. (2) can in fact be determined by object class instead of visitor.

13

Third Approach : (Continued)

• Extend your List applications to permit– multiplying all integers in a list or– counting the number of integers in a list– printing out all integers, ….

• Each can be implemented by defining a new visitor.– without the need to change code of List (and NullList and ConsList …).

main advantage of the visitor pattern.

class ProductVisitor implements Visitor { int prod = 1; public void visit(NullList x) {} public void visit(ConsList x) { prod = prod * x.head; x.tail.accept(this);}}

14

class CountVisitor implements Visitor { int count = 0; public void visit(NullList x) {} public void visit(ConsList x) { count++; x.tail.accept(this);}}class PrintVisitor implements Visitor { String result = “”; public void visit(NullList x) {} public void visit(ConsList x) { if( “”.equals(result) ){ result = result + x.head; } else { result = rersult + “ ,” + x.head ;} x.tail.accept(this);}}…CountVisitor cv = new CountVisitor(); PrintVisitor pv = new CountVisitor(); cv.accept(l); System.out.println(cv.count);pv.accept(l); System.out.println(“[“ + pv.result +”]”);

15

Comarision

• The Visitor pattern combines the advantages of the two other approaches.

Frequent Frequent type casts? recompilation?Instanceof and type casts Yes NoDedicated methods No YesThe Visitor pattern No No

• The advantage of Visitors: New methods without recompilation!• Requirement for using Visitors: All classes must have an accept

method.• Tools that use the Visitor pattern:

– JJTree (from Sun Microsystems) and the Java Tree Builder (from Purdue University), both frontends for The Java Compiler Compiler from Sun

Microsystems.-SableCC

16

Visitor: Summary

• Visitor makes adding new operations easy. – Simply write a new visitor.

• A visitor gathers related operations. – It also separates unrelated ones.

• Adding new classes to the object structure is hard. – Key consideration: are you most likely to change the algorithm applied

over an object structure, or are you most like to change the classes of objects that make up the structure.

– As to compiler design, it requires many operations on syntax tree, which is the main object structure of compiler and is relatively stable.

• Visitors can accumulate state/information.• Visitor can break encapsulation.

– Visitor’s approach assumes that the interface of the data structure classes is powerful enough to let visitors do their job. As a result, the pattern often forces you to provide public operations that access internal state, which may compromise its encapsulation.

17

The Java Tree Builder (JTB)

• Developed at Purdue University.– http://www.cs.purdue.edu/jtb/

• A frontend for The Java Compiler Compiler.• Supports the building of syntax trees which can be

traversed using visitors.• JTB transforms a bare JavaCC grammar into three

components:– a JavaCC grammar with embedded Java code for building a

syntax tree;– one class for every form of syntax tree node; and– a default visitor which can do a depth-first traversal of a syntax

tree.

18

JTB (continued)• The produced JavaCC grammar can then be processed

by the Java Compiler Compiler to give a parser which produces syntax trees.

• The produced syntax trees can now be traversed by a Java program by writing subclasses of the default visitor.

Bare JavaCC Grammar

Default Visitor

JTB

Parser

JavaCC

JavaCC Grammar with embeded code

Program

SyntaxTree Node Classes

SyntaxTree with accept methods

MyApplicationVisitor

19

Using JTB

• jtb myGrammar.jj // will generates many files:– jtb.out.jj -- the original grammar file with syntaxTree building

code inserted. – /syntaxtree/**.java -- a class for each production– /visitor/**– Visitor.java, DepthFirstVisitor -- the interface and

implemenation– ObjectVisitor.java, ObjectDepthFirstVisitor.java– -- the visitor interface (& implementation) with return

value and argument

• javacc jtb.out.jj // generates a parser with a specified name• javac Main.java // Main.java contains a call of the parser and

calls to visitors• java Main < prog.f // builds a syntax tree for prog.f, and

executes the visitors

20

Example:• A visitor which operates on syntax trees for Java programs. The

program prints the right-hand side of every assignment.

public class VprintAssignRHS extends DepthFirstVisitor {void visit(Assignment n) {PrettyPrinter v = new PrettyPrinter();n.f2.accept(v); v.out.println();n.f2.accept(this);} }• When this visitor is passed to the root of the syntax tree, the depth-

first traversal will begin, and when Assignment nodes are reached, the method visit in VprintAssignRHS is executed.

• Notice the use of PrettyPrinter. It is a visitor which pretty prints Java 1.1 programs.

• JTB is bootstrapped (JTB written using JTB)

21

Details

• jtb.out.jj, the original grammar file, now with syntax tree building actions inserted

• The subdirectory/package syntaxtree which contains a java class for each production in the grammar

• The subdirectory/package visitor which contains Visitor.java, the default visitor interface, also– DepthFirstVisitor.java, a default implementation which visits each

node of the tree in depth-first order. 

• ObjectVisitor.java, another default visitor interface that supports return value and argument. – ObjectDepthFirst.java is a defualt implemetation of

ObjectVisitor.java.

22

General Instructions

• To generate your parser, simply run JavaCC using jtb.out.jj as the grammar file. 

• Let's take a look at all the files and directories JTB generates.   

23

The grammar file

• Named jtb.out.jj • This file is the same as the input grammar file except that

it now contains code for building the syntax tree during parse. 

• Typically, this file can be left alone after generation.   • The only thing that needs to be done to it is to run it

through JavaCC to generate your parser

24

The syntax tree node classes

• This directory contains syntax tree node classes generated based on the productions in your JavaCC grammar. 

• Each production will have its own class.  If your grammar contains 42 productions, this directory will contain 42 classes (plus the special automatically generated nodes--these will be discussed later), with names corresponding to the left-hand side names of the productions. 

• Like jtb.out.jj, after generation these files don't need to be edited.  Generate them once, compile them once, and forget about them

25

Example

• Let's examine one of the classes generated from a production.  Take, for example, the following production

void ImportDeclaration() :

{}

{   "import" Name() [ "." "*" ] ";“

}

26

What gets produced?Part 1

// Generated by JTB 1.1.2 //

package syntaxtree;

/**

* Grammar production:  

* f0 -> "import"  

* f1 -> Name()  

* f2 -> [ "." "*" ]  

* f3 -> ";"  

*/

public class ImportDeclaration implements Node {    public NodeToken f0;   

public Name f1;   

public NodeOptional f2;   

public NodeToken f3;

All parts of a production are represented in the tree, including tokens. 

27

The Syntax Tree Classes

• Notice the package "syntaxtree".  • The purpose of separating the generated tree node

classes into their own package is that it greatly simplifies file organization, particularly when the grammar contains a large number of productions. 

• It’s often not necessary to pay the syntax classes any more attention.  All of the work is to done to the visitor classes. 

• Note that this class implements an interface named Node.   

28

Automatically-Generated Tree Node Interface and Classes

Node All tree nodes implement this

Nodelistinterface List interface that NodeList, NodeListOptional, and NodeSeqeunce implement

Nodechoice Represents ( A | B )

Nodelist Represents ( A ) +

NodeListOptional Represents ( A ) *

NodeOptional Represents [ A ] or ( A )?

NodeSequence Represents nexted sequence like

[ "extends" Name() ]

NodeToken Represents a token string

29

Node

• The interface Node is implemented by all syntax tree nodes. Node looks like this: 

public

interface Node extends java.io.Serializable {   

public void accept(visitor.Visitor v);   

public Object accept(visitor.ObjectVisitor v,

Object argu);

}

30

Nodes and Accept

• All tree node classes implement the accept() method.  – In the case of all the automatically-generated classes,

the accept() method simply calls the corresponding visit(XXXX n) (where XXXX is the name of the production) method of the visitor passed to it.  

– Note that the visit() methods are overloaded, i.e. the distinguishing feature is the argument each takes, as opposed to its name. 

31

Two New Features

• Two features presented in JTB 1.2 may be helpful –   The first is that Node extends java.io.Serializable,

meaning that you can now serialize your trees (or subtrees) to an output stream and read them back in. 

– Secondly, there is one accept() method that can take an extra argument and return a value.   

32

What gets produced?Part 1// Generated by JTB 1.1.2 //

package syntaxtree;

/**

* Grammar production:  

* f0 -> "import"  

* f1 -> Name()  

* f2 -> [ "." "*" ]  

* f3 -> ";"  

*/

public class ImportDeclaration implements Node {    public NodeToken f0;   

public Name f1;   

public NodeOptional f2;   

public NodeToken f3;

All parts of a production are represented in the tree, including tokens. 

33

NodeListInterface

• The interface NodeListInterface is implemented by NodeList, NodeListOptional, and NodeSequence.   NodeListInterface looks like this: 

public interface NodeListInterface extends Node { public void addNode(Node n);    public Node elementAt(int i);    public java.util.Enumeration elements();    public int size(); }

34

Details

• Interface not generally needed but can be useful when writing code which only deals with the Vector-like functionality of any of the three classes listed above.  – addNode() is used by the tree-building code to add

nodes to the list.  – elements() is similar to the method of the same name

in Vector, returning an Enumeration of the elements in the list.

– elementAt() returns the node at the ith position in the list (starting at 0, naturally).

– size() returns the number of elements in the list

35

NodeChoice

• NodeChoice is the class which JTB uses to represent choice points in a grammar.  An example of this would be 

( "abstract" | "final" | "public" )• JTB would represent the production 

void ResultType() :

{}

{   "void" | Type() }– as a class ResultType with a single child of type NodeChoice. 

36

Details

• The type stored by this NodeChoice would not be determined until the file was actually parsed. 

• The node stored by a NodeChoice would then be accessible through the choice field. 

• Since the choice is of type Node, typecasts are sometimes necessary to access the fields of the node stored in a NodeChoice. 

37

Implementation

public class NodeChoice implements Node {    public NodeChoice(Node node, int whichChoice);    public void accept(visitor.Visitor v);    public Object accept(visitor.ObjectVisitor v, Object argu);    public Node choice;    public int which; }

38

Which One?

• Another feature of NodeChoice is the field which for determining which of the choices was selected

• The which field is used to see which choice was used–   If the first choice is selected, which equals 0

(following the old programming custom to start counting at 0). 

– If the second choice is taken, which equals 1.  The third choice would be 2, etc. 

– Note that your code could potentially break if the order of the choices is changed in the grammar.   

39

NodeList

• NodeList is the class used by JTB to represent lists.  An example of a list would be 

( "[" Expression() "]" )+ 

• JTB would represent the javacc production : void ArrayDimensions() :

{}

{ ( "[" Expression() "]" )+ ( "[" "]" )* }

– as a class ArrayDimensions() with children NodeList and NodeListOptional respectively. 

40

Details

• NodeLists use java.lang.Vectors to store the lists of nodes. 

• Like NodeChoice, typecasts may occasionally be necessary to access fields of nodes contained in the list. 

41

Implementation

public class NodeList implements NodeListInterface {    public NodeList();     public void addNode(Node n);    public Enumeration elements();    public Node elementAt(int i);    public int size();    public void accept(visitor.Visitor v);    public Object accept(visitor.ObjectVisitor v,

Object argu);    public Vector nodes; }

42

NodeToken• This class is used by JTB to store all tokens into

the tree, including JavaCC "special tokens" (if the -tk command-line option is used). 

• In addition, each NodeToken contains information about each token, including its starting and ending column and line numbers. 

43

Implementation

public class NodeToken implements Node {    public NodeToken(String s);    public NodeToken(String s, int kind, int beginLine,

 int beginColumn, int endLine, int endColumn); public String toString();    public void accept(visitor.Visitor v);    public Object accept(visitor.ObjectVisitor v, Object argu); // -1 for these ints means no position info is available.

// …}

44

Continued

public class NodeToken implements Node {// …. public String tokenImage;      public int beginLine, beginColumn, endLine, endColumn; // -1 if not available.    // Equal to the JavaCC token "kind" integer.       public int kind;   // Special Token methods below    public NodeToken getSpecialAt(int i);    public int numSpecials();    public void addSpecial(NodeToken s);    public void trimSpecials();    public String withSpecials();   public Vector specialTokens; }

45

Token Details

• The tokens are simply stored as strings.  • The field tokenImage can be accessed directly, and the

toString() method returns the same string.  • Also available is the kind integer. • JavaCC assigns each type of token a unique integer to

identify it.  • This integer is now available in each JTB NodeToken. 

For more information on using the kind integer, see the JavaCC documentation. 

46

Member Variables of Generated Classes

• Next comes the member variables of the ImportDeclaration class. 

• These are generated based on the RHS of the production.  Their type depends on the various items in the RHS and their names begin with f0 and work their way up. 

• Why are they public?  – Visitors which must access these fields reside in a different

package than the syntax tree nodes– Package visibility cannot be used. – Breaking encapsulation was a necessary evil in this case. 

47

What gets produced?Part 2

   public ImportDeclaration(NodeToken n0, Name n1, NodeOptional n2, NodeToken n3) {      

f0 = n0;f1 = n1;f2 = n2;f3 = n3;   

}   public ImportDeclaration(Name n0, NodeOptional n1) {      

f0 = new NodeToken("import");f1 = n0; f2 = n1; f3 = new NodeToken(";");   

}   

48

Constructors

• The next portion of the generated class is the standard constructor.  It is called from the tree-building actions in the annotated grammar so you will probably not need to use it. 

• Following the first constructor is a convenience constructor with the constant tokens of the production already filled-in by the appropriate NodeToken.  This constructor's purpose is to help in manual construction of syntax trees. 

49

What gets produced?Part 3

public void accept(visitor.Visitor v)

{       v.visit(this);   

}   

public Object

accept(visitor.ObjectVisitor v, Object argu) {      

return v.visit(this,argu);   

}

}

50

The Accept Methods

• After the constructor are the accept() methods.  • These methods are the way in which visitors

interact with the class. 

void accept(visitor.Visitor v) – works with Visitor

Object accept(visitor.ObjectVisitor v,

Object argu) – works with ObjectVisitor

51

References

• Java Tree Builder Documentation• Why Visitors?• Erich Gamma, Richard Helm, Ralph Johnson, and John

Vlissides. Design Patterns: Elements of Reusable Object-Oriented Software. Addison-Wesley, 1995