sandmark.watermark.ct.trace
Class Preprocessor

java.lang.Object
  |
  +--sandmark.watermark.ct.trace.Preprocessor

public class Preprocessor

extends java.lang.Object

Field Summary

static int	ADDEDCODESIZE The amount of code we're adding to each method.
(package private) ClassFileCollection	cfc
static java.lang.String	FRAMECOUNTER Name of static variable in sandmark.watermark.ct.trace.Annotator that holds the current stack ID count.
(package private) java.util.Properties	props
static java.lang.String	STACKID Name of the local variable stored in each method to record the unique ID of that call frame.

Constructor Summary

Preprocessor(java.util.Properties props)

Method Summary

void	preprocess() Go through every method in every class of the program and add the code long sm$stackID = sandmark.watermark.ct.trace.Annotator.stackFrameNumber++; to the very beginning of the method.
(package private) void	preprocessClass(EditedClass ec) Make the modifications to every method of the class.
(package private) void	preprocessMethod(EditedClass ec, de.fub.bytecode.generic.MethodGen mg) Create a local variable to hold the stack ID.
void	save() Save the edited classfiles under a new name.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

cfc

ClassFileCollection cfc

props

java.util.Properties props

STACKID

public static final java.lang.String STACKID

Name of the local variable stored in each method to record the unique ID of that call frame.

See Also:

Constant Field Values

FRAMECOUNTER

public static final java.lang.String FRAMECOUNTER

Name of static variable in sandmark.watermark.ct.trace.Annotator that holds the current stack ID count.

See Also:

Constant Field Values

ADDEDCODESIZE

public static final int ADDEDCODESIZE

The amount of code we're adding to each method. Use this to compute the actual offset of a mark() call within the unprocessed (original) source. We're adding the following code, whose size adds up to 11:

   GETSTATIC   X                        3
   DUP2                                 1
   LCONST_1                             1
   LADD                                 1
   GETSTATIC   X                        3
   LSTORE_1    X   (or LSTORE)          1 (or 2)
   (NOP)                                0 (or 1)
 

See Also:

Constant Field Values

Constructor Detail

Preprocessor

public Preprocessor(java.util.Properties props)
             throws java.io.IOException

Method Detail

preprocess

public void preprocess()
                throws java.io.IOException

Go through every method in every class of the program and add the code

    long sm$stackID = sandmark.watermark.ct.trace.Annotator.stackFrameNumber++;
 

to the very beginning of the method. This will give every stackframe a unique ID.

java.io.IOException

save

public void save()
          throws java.io.IOException

Save the edited classfiles under a new name.

java.io.IOException

preprocessClass

void preprocessClass(EditedClass ec)

Make the modifications to every method of the class.

Parameters:

ec - the class to be edited

preprocessMethod

void preprocessMethod(EditedClass ec,
                      de.fub.bytecode.generic.MethodGen mg)

Create a local variable to hold the stack ID.

Parameters:

ec - the class to be edited

mg - the method to be edited Example:

    void P() {
       ...
    }
 

The resulting method would look something like this:

    void P() {
       long sm$stackID = sandmark.watermark.ct.trace.Annotator.stackFrameNumber++;
       ...
    }
 

Or, in bytecode:

   GETSTATIC   sandmark.watermark.ct.trace.Annotator.stackFrameNumber
   DUP2
   LCONST_1
   LADD
   GETSTATIC   sandmark.watermark.ct.trace.Annotator.stackFrameNumber
   LSTORE      sm$stackID-index
 

One complication:

   LSTORE
 

could either be

      LSTORE_1
 

or

      LSTORE 5
 

In the former case the instruction is 1 byte, in the latter, 2. We want all code sequences we insert to be the same length, so that when mark() is called in the preprocessed bytecode, we can figure out what its "real" offset is (the offset in the unprocessed code). Therefore we add an extra NOP in the case of LSTORE_1.