5. Introspection and customization

CtClass provides methods for introspection. The introspective ability of Javassist is compatible with that of the Java reflection API. CtClass provides getName(), getSuperclass(), getMethods(), and so on. CtClass also provides methods for modifying a class definition. It allows to add a new field, constructor, and method. Instrumenting a method body is also possible.

Javassist does not allow to remove a method or field, but it allows to change the name. So if a method is not necessary any more, it should be renamed and changed to be a private method by calling setName() and setModifiers() declared in CtMethod.

Javassist does not allow to add an extra parameter to an existing method, either. Instead of doing that, a new method receiving the extra parameter as well as the other parameters should be added to the same class. For example, if you want to add an extra int parameter newZ to a method:

void move(int newX, int newY) { x = newX; y = newY; }

in a Point class, then you should add the following method to the Point class:

void move(int newX, int newY, int newZ) {
    // do what you want with newZ.
    move(newX, newY);
}

Javassist also provides low-level API for directly editing a raw class file. For example, getClassFile() in CtClass returns a ClassFile object representing a raw class file. getMethodInfo() in CtMethod returns a MethodInfo object representing a method_info structure included in a class file. The low-level API uses the vocabulary from the Java Virtual machine specification. The users must have the knowledge about class files and bytecode. For more details, the users should see the javassist.bytecode package.

5.1 Inserting source text at the beginning/end of a method body

CtMethod and CtConstructor provide methods insertBefore(), insertAfter(), and addCatch(). They are used for inserting a code fragment into the body of an existing method. The users can specify those code fragments with source text written in Java. Javassist includes a simple Java compiler for processing source text. It receives source text written in Java and compiles it into Java bytecode, which will be inlined into a method body.

The methods insertBefore(), insertAfter(), and addCatch() receives a String object representing a statement or a block. A statement is a single control structure like if and while or an expression ending with a semi colon (;). A block is a set of statements surrounded with braces {}. Hence each of the following lines is an example of valid statement or block:

System.out.println("Hello");
{ System.out.println("Hello"); }
if (i < 0) { i = -i; }

The statement and the block can refer to fields and methods. However, they cannot refer to local variables declared in the method that they are inserted into. They can refer to the parameters to the method although they must use different names $0, $1, $2, ... described below. Declaring a local variable in the block is allowed.

The String object passed to the methods insertBefore(), insertAfter(), and addCatch() are compiled by the compiler included in Javassist. Since the compiler supports language extensions, several identifiers starting with $ have special meaning:

$0, $1, $2, ...	Actual parameters
$args	An array of parameters. The type of $args is Object[].
$$	All actual parameters. For example, m($$) is equivalent to m($1,$2,...)
$cflow(...)	cflow variable
$r	The result type. It is used in a cast expression.
$w	The wrapper type. It is used in a cast expression.
$_	The resulting value
$sig	An array of java.lang.Class objects representing the formal parameter types.
$type	A java.lang.Class object representing the formal result type.
$class	A java.lang.Class object representing the class currently edited.

$0, $1, $2, ...

The parameters passed to the methods insertBefore(), insertAfter(), and addCatch() are accessible with $0, $1, $2, ... instead of the original parameter names. $1 represents the first parameter, $2 represents the second parameter, and so on. The types of those variables are identical to the parameter types. $0 is equivalent to this. If the method is static, $0 is not available.

These variables are used as following. Suppose that a class Point:

class Point {
    int x, y;
    void move(int dx, int dy) { x += dx; y += dy; }
}

To print the values of dx and dy whenever the method move() is called, execute this program:

ClassPool pool = ClassPool.getDefault();
CtClass cc = pool.get("Point");
CtMethod m = cc.getDeclaredMethod("move");
m.insertBefore("{ System.out.println($1); System.out.println($2); }");
cc.writeFile();

Note that the source text passed to insertBefore() is surrounded with braces {}. insertBefore() accepts only a single statement or a block surrounded with braces.

The definition of the class Point after the modification is like this:

class Point {
    int x, y;
    void move(int dx, int dy) {
        { System.out.println(dx); System.out.println(dy); }
        x += dx; y += dy;
    }
}

$1 and $2 are replaced with dx and dy, respectively.

$1, $2, $3 ... are updatable. If a new value is assigend to one of those variables, then the value of the parameter represented by that variable is also updated.

$args

The variable $args represents an array of all the parameters. The type of that variable is an array of class Object. If a parameter type is a primitive type such as int, then the parameter value is converted into a wrapper object such as java.lang.Integer to store in $args. Thus, $args[0] is equivalent to $1 unless the type of the first parameter is a primitive type. Note that $args[0] is not equivalent to $0; $0 represents this.

If an array of Object is assigned to $args, then each element of that array is assigned to each parameter. If a parameter type is a primitive type, the type of the corresponding element must be a wrapper type. The value is converted from the wrapper type to the primitive type before it is assigned to the parameter.

$$

The variable $$ is abbreviation of a list of all the parameters separated by commas. For example, if the number of the parameters to method move() is three, then

move($$)

is equivalent to this:

move($1, $2, $3)

If move() does not take any parameters, then move($$) is equivalent to move().

$$ can be used with another method. If you write an expression:

exMove($$, context)

then this expression is equivalent to:

exMove($1, $2, $3, context)

Note that $$ enables generic notation of method call with respect to the number of parameters. It is typically used with $proceed shown later.

$cflow

$cflow means "control flow". This read-only variable returns the depth of the recursive calls to a specific method.

Suppose that the method shown below is represented by a CtMethod object cm:

int fact(int n) {
    if (n <= 1)
        return n;
    else
        return n * fact(n - 1);
}

To use $cflow, first declare that $cflow is used for monitoring calls to the method fact():

CtMethod cm = ...;
cm.useCflow("fact");

The parameter to useCflow() is the identifier of the declared $cflow variable. Any valid Java name can be used as the identifier. Since the identifier can also include . (dot), for example, "my.Test.fact" is a valid identifier.

Then, $cflow(fact) represents the depth of the recursive calls to the method specified by cm. The value of $cflow(fact) is 0 (zero) when the method is first called whereas it is 1 when the method is recursively called within the method. For example,

cm.insertBefore("if ($cflow(fact) == 0)"
              + "    System.out.println(\"fact \" + $1);");

translates the method fact() so that it shows the parameter. Since the value of $cflow(fact) is checked, the method fact() does not show the parameter if it is recursively called within fact().

The value of $cflow is the number of stack frames associated with the specified method cm under the current topmost stack frame for the current thread. $cflow is also accessible within a method different from the specified method cm.

$r

$r represents the result type (return type) of the method. It must be used as the cast type in a cast expression. For example, this is a typical use:

Object result = ... ;
$_ = ($r)result;

If the result type is a primitive type, then ($r) follows special semantics. First, if the operand type of the cast expression is a primitive type, ($r) works as a normal cast operator to the result type. On the other hand, if the operand type is a wrapper type, ($r) converts from the wrapper type to the result type. For example, if the result type is int, then ($r) converts from java.lang.Integer to int.

If the result type is void, then ($r) does not convert a type; it does nothing. However, if the operand is a call to a void method, then ($r) results in null. For example, if the result type is void and foo() is a void method, then

$_ = ($r)foo();

is a valid statement.

The cast operator ($r) is also useful in a return statement. Even if the result type is void, the following return statement is valid:

return ($r)result;

Here, result is some local variable. Since ($r) is specified, the resulting value is discarded. This return statement is regarded as the equivalent of the return statement without a resulting value:

return;

$w

$w represents a wrapper type. It must be used as the cast type in a cast expression. ($w) converts from a primitive type to the corresponding wrapper type. The following code is an example:

Integer i = ($w)5;

The selected wrapper type depends on the type of the expression following ($w). If the type of the expression is double, then the wrapper type is java.lang.Double.

If the type of the expression following ($w) is not a primitive type, then ($w) does nothing.

$_

insertAfter() in CtMethod and CtConstructor inserts the compiled code at the end of the method. In the statement given to insertAfter(), not only the variables shown above such as $0, $1, ... but also $_ is available.

The variable $_ represents the resulting value of the method. The type of that variable is the type of the result type (the return type) of the method. If the result type is void, then the type of $_ is Object and the value of $_ is null.

Although the compiled code inserted by insertAfter() is executed just before the control normally returns from the method, it can be also executed when an exception is thrown from the method. To execute it when an exception is thrown, the second parameter asFinally to insertAfter() must be true.

If an exception is thrown, the compiled code inserted by insertAfter() is executed as a finally clause. The value of $_ is 0 or null in the compiled code. After the execution of the compiled code terminates, the exception originally thrown is re-thrown to the caller. Note that the value of $_ is never thrown to the caller; it is rather discarded.

$sig

The value of $sig is an array of java.lang.Class objects that represent the formal parameter types in declaration order.

$type

The value of $type is an java.lang.Class object representing the formal type of the result value. This variable is available only in insertAfter() in CtMethod and CtConstructor.

$class

The value of $class is an java.lang.Class object representing the class in which the edited method is declared.

addCatch()

addCatch() inserts a code fragment into a method body so that the code fragment is executed when the method body throws an exception and the control returns to the caller. In the source text representing the inserted code fragment, the exception value is referred to with the special variable $e.

For example, this program:

CtMethod m = ...;
CtClass etype = ClassPool.getDefault().get("java.io.IOException");
m.addCatch("{ System.out.println($e); throw $e; }", etype);

translates the method body represented by m into something like this:

try {
    the original method body
}
catch (java.io.IOException e) {
    System.out.println(e);
    throw e;
}

Note that the inserted code fragment must end with a throw or return statement.

5.2 Modifying a method body

CtMethod and CtConstructor provide setBody() for substituting a whole method body. They compile the given source text into Java bytecode and substitutes it for the original method body. If the given source text is null, the substituted body includes only a return statement, which returns zero or null unless the result type is void.

In the source text given to setBody(), the identifiers starting with $ have special meaning

$0, $1, $2, ...	Actual parameters
$args	An array of parameters. The type of $args is Object[].
$$	All actual parameters.
$cflow(...)	cflow variable
$r	The result type. It is used in a cast expression.
$w	The wrapper type. It is used in a cast expression.
$sig	An array of java.lang.Class objects representing the formal parameter types.
$type	A java.lang.Class object representing the formal result type.
$class	A java.lang.Class object representing the class currently edited.

Javassist allows modifying only an expression included in a method body. javassist.expr.ExprEditor is a class for replacing an expression in a method body. The users can define a subclass of ExprEditor to specify how an expression is modified.

To run an ExprEditor object, the users must call instrument() in CtMethod or CtClass. For example,

CtMethod cm = ... ;
cm.instrument(
    new ExprEditor() {
        public void edit(MethodCall m)
                      throws CannotCompileException
        {
            if (m.getClassName().equals("Point")
                          && m.getMethodName().equals("move"))
                m.replace("{ $1 = 0; $_ = $proceed($$); }");
        }
    });

searches the method body represented by cm and replaces all calls to move() in class Point with a block:

{ $1 = 0; $_ = $proceed($$); }

so that the first parameter to move() is always 0. Note that the substituted code is not an expression but a statement or a block.

The method instrument() searches a method body. If it finds an expression such as a method call, field access, and object creation, then it calls edit() on the given ExprEditor object. The parameter to edit() is an object representing the found expression. The edit() method can inspect and replace the expression through that object.

Calling replace() on the parameter to edit() substitutes the given statement or block for the expression. If the given block is an empty block, that is, if replace("{}") is executed, then the expression is removed from the method body. If you want to insert a statement (or a block) before/after the expression, a block like the following should be passed to replace():

{ before-statements;
  $_ = $proceed($$);
  after-statements; }

whichever the expression is either a method call, field access, object creation, or others. The second statement could be:

$_ = $proceed();

if the expression is read access, or

$proceed($$);

if the expression is write access.

javassist.expr.MethodCall

A MethodCall object represents a method call. The method replace() in MethodCall substitutes a statement or a block for the method call. It receives source text representing the substitued statement or block, in which the identifiers starting with $ have special meaning as in the source text passed to insertBefore().

$0	The target object of the method call. This is not equivalent to this, which represents the caller-side this object. $0 is null if the method is static.
$1, $2, ...	The parameters of the method call.
$_	The resulting value of the method call.
$r	The result type of the method call.
$class	A java.lang.Class object representing the class declaring the method.
$sig	An array of java.lang.Class objects representing the formal parameter types.
$type	A java.lang.Class object representing the formal result type.
$proceed	The name of the method originally called in the expression.

Here the method call means the one represented by the MethodCall object.

The other identifiers such as $w, $args and $$ are also available.

Unless the result type of the method call is void, a value must be assigned to $_ in the source text and the type of $_ is the result type. If the result type is void, the type of $_ is Object and the value assigned to $_ is ignored.

$proceed is not a String value but special syntax. It must be followed by an argument list surrounded by parentheses ( ).

javassist.expr.FieldAccess

A FieldAccess object represents field access. The method edit() in ExprEditor receives this object if field access is found. The method replace() in FieldAccess receives source text representing the substitued statement or block for the field access. In the source text, the identifiers starting with $ have also special meaning:

$0	The object containing the field accessed by the expression. This is not equivalent to this. this represents the object that the method including the expression is invoked on. $0 is null if the field is static.
$1	The value that would be stored in the field if the expression is write access. Otherwise, $1 is not available.
$_	The resulting value of the field access if the expression is read access. Otherwise, the value stored in $_ is discarded.
$r	The type of the field if the expression is read access. Otherwise, $r is void.
$class	A java.lang.Class object representing the class declaring the field.
$type	A java.lang.Class object representing the field type.
$proceed	The name of a virtual method executing the original field access. .

The other identifiers such as $w, $args and $$ are also available.

If the expression is read access, a value must be assigned to $_ in the source text. The type of $_ is the type of the field.

javassist.expr.NewExpr

A NewExpr object represents object creation with the new operator. The method edit() in ExprEditor receives this object if object creation is found. The method replace() in NewExpr receives source text representing the substitued statement or block for the object creation. In the source text, the identifiers starting with $ have also special meaning:

$0	null.
$1, $2, ...	The parameters to the constructor.
$_	The resulting value of the object creation. A newly created object must be stored in this variable.
$r	The type of the created object.
$class	A java.lang.Class object representing the class of the created object.
$sig	An array of java.lang.Class objects representing the formal parameter types.
$proceed	The name of a virtual method executing the original object creation. .

The other identifiers such as $w, $args and $$ are also available.

javassist.expr.Instanceof

A Instanceof object represents an instanceof expression. The method edit() in ExprEditor receives this object if an instanceof expression is found. The method replace() in Instanceof receives source text representing the substitued statement or block for the expression. In the source text, the identifiers starting with $ have also special meaning:

$0	null.
$1	The value on the left hand side of the original instanceof operator.
$_	The resulting value of the expression. The type of $_ is boolean.
$r	The type on the right hand side of the instanceof operator.
$type	A java.lang.Class object representing the type on the right hand side of the instanceof operator.
$proceed	The name of a virtual method executing the original instanceof expression. It takes one parameter (the type is java.lang.Object) and returns true if the parameter value is an instance of the type on the right hand side of the original instanceof operator. Otherwise, it returns false.

The other identifiers such as $w, $args and $$ are also available.

javassist.expr.Cast

A Cast object represents an expression for explicit type casting. The method edit() in ExprEditor receives this object if explicit type casting is found. The method replace() in Cast receives source text representing the substitued statement or block for the expression. In the source text, the identifiers starting with $ have also special meaning:

$0	null.
$1	The value the type of which is explicitly cast.
$_	The resulting value of the expression. The type of $_ is the same as the type after the explicit casting, that is, the type surrounded by ( ).
$r	the type after the explicit casting, or the type surrounded by ( ).
$type	A java.lang.Class object representing the same type as $r.
$proceed	The name of a virtual method executing the original type casting. It takes one parameter of the type java.lang.Object and returns it after the explicit type casting specified by the original expression.

The other identifiers such as $w, $args and $$ are also available.

javassist.expr.Handler

A Handler object represents a catch clause of try-catch statement. The method edit() in ExprEditor receives this object if a catch is found. The method insertBefore() in Handler compiles the received source text and inserts it at the beginning of the catch clause. In the source text, the identifiers starting with $ have special meaning:

$1	The exception object caught by the catch clause.
$r	the type of the exception caught by the catch clause. It is used in a cast expression.
$w	The wrapper type. It is used in a cast expression.
$type	A java.lang.Class object representing the type of the exception caught by the catch clause.

If a new exception object is assigned to $1, it is passed to the original catch clause as the caught exception.

5.3 Adding a new method or field

Adding a method

Javassist allows the users to create a new method and constructor from scratch. CtNewMethod and CtNewConstructor provide several factory methods, which are static methods for creating CtMethod or CtConstructor objects. Especially, make() creates a CtMethod or CtConstructor object from the given source text.

For example, this program:

CtClass point = ClassPool.getDefault().get("Point");
CtMethod m = CtNewMethod.make(
                 "public int xmove(int dx) { x += dx; }",
                 point);
point.addMethod(m);

adds a public method xmove() to class Point. In this example, x is a int field in the class Point.

The source text passed to make() can include the identifiers starting with $ except $_ as in setBody(). It can also include $proceed if the target object and the target method name are also given to make(). For example,

CtClass point = ClassPool.getDefault().get("Point");
CtMethod m = CtNewMethod.make(
                 "public int ymove(int dy) { $proceed(0, dy); }",
                 point, "this", "move");

this program creates a method ymove() defined below:

public int ymove(int dy) { this.move(0, dy); }

Note that $proceed has been replaced with this.move.

Javassist provides another way to add a new method. You can first create an abstract method and later give it a method body:

CtClass cc = ... ;
CtMethod m = new CtMethod(CtClass.intType, "move",
                          new CtClass[] { CtClass.intType }, cc);
cc.addMethod(m);
m.setBody("{ x += $1; }");
cc.setModifiers(cc.getModifiers() & ~Modifier.ABSTRACT);

Since Javassist makes a class abstract if an abstract method is added to the class, you have to explicitly change the class back to a non-abstract one after calling setBody().

Mutual recursive methods

Javassist cannot compile a method if it calls another method that has not been added to a class. (Javassist can compile a method that calls itself recursively.) To add mutual recursive methods to a class, you need a trick shown below. Suppose that you want to add methods m() and n() to a class represented by cc:

CtClass cc = ... ;
CtMethod m = CtNewMethod.make("public abstract int m(int i);", cc);
CtMethod n = CtNewMethod.make("public abstract int n(int i);", cc);
cc.addMethod(m);
cc.addMethod(n);
m.setBody("{ return ($1 <= 0) ? 1 : (n($1 - 1) * $1); }");
n.setBody("{ return m($1); }");
cc.setModifiers(cc.getModifiers() & ~Modifier.ABSTRACT);

You must first make two abstract methods and add them to the class. Then you can give the method bodies to these methods even if the method bodies include method calls to each other. Finally you must change the class to a not-abstract class since addMethod() automatically changes a class into an abstract one if an abstract method is added.

Adding a field

Javassist also allows the users to create a new field.

CtClass point = ClassPool.getDefault().get("Point");
CtField f = new CtField(CtClass.intType, "z", point);
point.addField(f);

This program adds a field named z to class Point.

If the initial value of the added field must be specified, the program shown above must be modified into:

CtClass point = ClassPool.getDefault().get("Point");
CtField f = new CtField(CtClass.intType, "z", point);
point.addField(f, "0");    // initial value is 0.

Now, the method addField() receives the second parameter, which is the source text representing an expression computing the initial value. This source text can be any Java expression if the result type of the expression matches the type of the field. Note that an expression does not end with a semi colon (;).

5.4 Limitations

In the current implementation, the Java compiler included in Javassist has several limitations with respect to the language that the compiler can accept. Those limitations are: