My Java/Clojure bytecode inspection workflow

Last summer I had to do a fair bit of Clojure compiler hacking. Since the primary objective of a compiler is to produce compiled code (in case of Clojure it is Java bytecode), this process involved reading classfiles every now and then. To make it easier I set up a nice environment using publicly available tools to make reading the bytecode as painless as possible.

Note that despite me using this workflow for Clojure, there is nothing here specific to it, so you can utilize the same setup when working with Java, Scala or any other JVM-compiled language out there.

javap

This helpful tool comes with a standard JDK distribution. It is likely to be already on your PATH, so you can call it like this:

javap -c target/classes/clojure/core\$bases.class

And get the following:

public final class clojure.core$bases extends clojure.lang.AFunction {
  public static final clojure.lang.Var const__0;

  public static final clojure.lang.Var const__1;

  public static {};
    Code:
       0: ldc           #12                 // String clojure.core
       2: ldc           #14                 // String seq
       4: invokestatic  #20                 // Method clojure/lang/RT.var:(Ljava/lang/String;Ljava/lang/String;)Lclojure/lang/Var;
       7: checkcast     #22                 // class clojure/lang/Var
  ...

Java decompiler

While having disassembled bytecode is nice and all, most of the time you would rather see the respective Java code. Jad comes to rescue. You can install it from your distro's repositories or just download and unpack the archive.

Jad tries its best to transform the bytecode back into Java. Of course it can't reconstruct everything, and specifically for Clojure the code looks quite weird. But in most cases it is still more lucid than digging through byte codes.

Say, you execute a command:

jad core\$empty_QMARK_.class

If everything goes well the decompiled class will be written to core\$empty_QMARK_.jad. View it and you will find the following:

package clojure;

import clojure.lang.AFunction;
import clojure.lang.IFn;
import clojure.lang.RT;
import clojure.lang.Var;

public final class  extends AFunction
{

    public Object invoke(Object coll)
    {
        return ((IFn)const__0.getRawRoot()).invoke(((IFn)const__1.getRawRoot()).invoke(coll = null));
    }

    public static final Var const__0 = (Var)RT.var("clojure.core", "not");
    public static final Var const__1 = (Var)RT.var("clojure.core", "seq");


    public ()
    {
    }
}

To those who haven't yet seen Clojure-produced bytecode before this may seem incoherent, but after some practice you'll get a handle on it. In this particular case, the code you see in the invoke method simply calls (not (seq coll)).

Emacs

Having these tools as shell commands is surely convenient, but because I spend almost all of my development time in Emacs I'd like to have them somehow integrated. Luckily, "Emacs" and "integrated" are exact synonyms. First, you will need to install javap package (available on MELPA and Marmalade) and also javad.el (put it into your ~/.emacs.d/ directory). Then insert the following code into init.el:

;; Load both modes
(load "~/.emacs.d/javad.el")
(require 'javap-mode)

(defun javad-find-class (&rest args)
  (interactive)
  (if (not (string= ".class" (substring (buffer-file-name) -6 nil)))
      nil
    (message "Show class as: [b]ytecode, [d]ecompiled or [i]dentity?")
    (let ((resp (read-char)))
      (cond
       ((= resp 98) (progn (javap-buffer) nil))
       ((= resp 100) (progn (javad-buffer) nil))
       (t nil))
      (let ((buff (current-buffer)))
        (switch-to-buffer buff)))))

(add-hook 'find-file-hook 'javad-find-class)

This is actually an abhorrent solution and I welcome anyone to fix it for me. What it does is creating a hook that will be executed when a file is opened in Emacs. If the file name ends with .class, the user will be interactively queried about what (s)he wants to do with the class: disassemble, decompile, or nothing. It is handy to be able to choose from both former options, as some tricky classes cannot be processed by Jad, so you can always fallback to reading the bytecode.

This setup is especially convenient for me since I use Emacs as my file manager too (via amazing Sunrise Commander). From there I can navigate to any classfile and open it selecting between javap and Jad. This feels much more empowering than typing the commands in the shell, plus you get syntax highlighting for both bytecode and decompiled Java.

Conclusion

I hope the provided information will be of help to those who must deal with bytecode extensively, or enthusiasts who occasionally want to see how their compiled code looks like. Enjoy!

Bytøpia