Saturday, November 22, 2008

Double-clicked Java

Nice article from Dmitry Leskov:

"How do I make an .EXE file from my Java application?",

"Need help converting jar to exe",

"Is it possible to create a Windows executable using Java?"

--- these and similar questions are among the most popular topics on Java developer forums. Should you start such a topic today, you are likely to encounter the following three types of replies:

1/ "You cannot"
2/ "You should not, because that would kill the very purpose of Java"
3/ "You can do that with third party software X and Y"

The truth is that there exist two completely different approaches to the creation of native executables from Java applications, addressing different sets of problems. Moreover, under certain conditions some of those problems may be solved without making an EXE. So the most correct way to reply to such a post would be a request for more information, namely what is the goal of conversion to EXE. And the most frequent answer would be

=== Simplify Java Application Deployment ===

Java compiles to platform-independent bytecode (.class files), which is not directly supported by PC hardware. So a Java program needs a Java Runtime Environment (JRE) to run, which would either interpret the bytecode instructions or compile them to native code on the fly. This in turn means that the author of that program has to ensure somehow that the proper version of the JRE is installed on an end user system.

In a general case you may not expect that your end users will know what a JRE is, how to check its version, and how to download and install it. This is especially true for consumer applications, such as games or multimedia. And those who already have a JRE installed may not like the idea about installing a different version, because it may break their existing Java applications and favorite applets.

Then, even if you can make sure the right version of the JRE is properly installed on enduser systems, which is quite possible in a classroom or enterprise environment, the command line required to launch your Java application can be quite long:

java -Xmx200m -cp whatever.jar MyApp

Yes, you may put that line into a batch file and call it runme.bat, but it looks so much easier to give your program to a friend, teacher or colleague as a single file that can be run by a double-click. Or, even better, enable it to be installed and uninstalled in a native manner without affecting other applications.

So it comes as no surprise that the primary motivation for seeking a way to convert a Java application into an EXE file is to make its deployment and use simpler and safer for an average user, that is, a Windows user. What surprises newbie Java developers is that the JDK does not offer such functionality. Before J2SE 1.4, all you could make with JDK tools were

=== Executable Jars ===

You can make your Java application runnable via a double-click by packaging it into a so called executable jar. You do that by specifying the main class of your application, any extra jar files it may require and so on in the jar's manifest file

Main-Class: MyAppMain
Class-Path: mylib.jar

Then you use the jar utility from the Java SDK to package your classes and resource files, specifying the m option and the name of your manifest file:

jar cvfm MyApp.jar *.class *.gif

This will result in the creation of MyApp.jar. Now, if you type

java -jar MyApp.jar

the Java launcher will read the manifest from MyApp.jar and invoke the main method from the class MyAppMain. Moreover, if you double-click that jar file on a system that has JRE installed, the java launcher will be invoked automatically.

Note: As of J2SE 5.0, jar files are associated with the javaw launcher on Windows, which does not open a console on startup. If your application needs a console, write a batch file which would start it using the java launcher.

If your application consists of more than one jar file, there is an open source tool called One-JAR that claims to correctly repackage multiple jars into one.

The major problem with executable jars is compatibility. The default JRE may be of an older version than is required by your application or may not have the necessary Java Optional Packages (previously known as Standard Extensions) installed. For instance, if your app uses the java.nio package introduced in Java 2 version 1.4, it will not work on JRE 1.3.x. Similarly, if it uses JavaMail 1.3, and the default JRE has JavaMail 1.2 or JavaMail is not present at all, the double-clicked jar will not run.

Fortunately, Sun has created a Java application deployment technology that eliminates this compatibility problem and adds some nice features. It is part of the Java 2 platform since version 1.4 and is called

=== Java Web Start ===

Java Web Start (JWS) and the underlying Java Network Launch Protocol (JNLP) enable Java application delivery from a standard Web server. The end user initiates application installation by clicking on an URL. If the Java Web Start engine is not present on the system, the user is prompted to download and install it. Once Java Web Start is in place, clicking on the same URL will initiate the application download and installation procedures. It may involve download and installation of the required version of the JRE and Optional Packages. Upon their successful completion, the application is launched. The application will be cached on the user's system so next time the user clicks on the same URL, the JWS engine will launch the local copy of the application from the cache, if it detects that the computer is offline or the application was not updated on the Web site.

Another important feature of JWS is its ability to run your application in a sandbox - a restricted container based on Java security architecture. But, unlike an applet, your application can gain access to local system resources like the filesystem, printer and system clipboard using the JNLP API even if it comes from an untrusted environment, after prompting the user for confirmation.

Java Web Start is available for Windows, Linux, and Solaris, and is part of MacOS X since v10.1. There are also third-party implementations of the JNLP protocol, some of them also include tools that assist you in the creation and maintenance of JNLP packages.

That was the bright side. Now, what is not so good about JNLP? First off, for seamless operation both the browser and the Web server that hosts the JNLP-enabled application must support application/x-java-jnlp-file MIME type. Some hosting providers do not support it. Moreover, versioning and incremental updates require additional support from the Web server, which has to be implemented using servlets, cgi-bin scripts, etc.

On the client side, a major browser would be configured to recognize the above MIME type during installation of the JWS engine, but users of less popular browsers, such as Opera, may have to do that manually.

JNLP-enabling an application may involve minor changes in its code and (re)packaging it into a set of jar files.

Before J2SE 5.0, JWS had very little to offer in terms of desktop integration - all it could do was create a desktop icon and/or a Start Menu entry for the application. On Windows, the application will not show up in Add/Remove Programs, so end users would have to run the Java Web Start application manager in order to remove your application.

Finally, JWS user interface needs much polishing. As of J2SE 5.0, users still complain about ugly windows with incomprehensible messages.

To sum it up, JWS can be a viable option in a controlled environment, such as corporate intranet, but it is not ready for the consumer market, where you may be better off using

=== Custom Java Launchers And Wrappers ===

When a Java program is invoked using one of the methods discussed above (batch file, executable jar, or Java Web Start/JNLP), the operating system runs a Java launcher from the JRE. The Windows version of the JRE has separate launchers for command-line and GUI apps, called java.exe and javaw.exe respectively.

As a result, all running Java applications have the same Taskbar/Alt-Tab icons and appear in the Windows Task Manager as either java.exe or javaw.exe. If you have two or more Java apps running, you have no means to distinguish between multiple instances of the standard Java launcher in the Task Manager.

In fact, those launchers are just small native programs that load the Java Virtual Machine from a DLL/shared library and then feed your program to that JVM using the Invocation API. That API is part of the Java Native Interface (JNI), so it is standardized, and it is also very simple. This makes it relatively easy to write your own launcher with a unique name and icon. What it has to do is find a suitable JRE on the end user's system (unless you bundle the JRE with your application), load and initialize the JVM, and run your application on it.

If you do not have the right tools, skills, or time to develop a custom launcher for your Java application, there are quite a few third-party Java launcher generators listed in the Tools section of the frame. Some of them provide additional features such as instant splash screen, stdout and stderr redirection, and so on, the most notable being wrapping.

A Java wrapper is essentially a custom Java launcher that is also a self-extracting archive containing all the application's classes, jars and auxiliary files. The wrapper unpacks those files on startup and removes on termination. This way, your application is distributed as a single executable.

A wrapper normally looks up the JRE upon startup. If the JRE is not present or its version does not match the application's compatibility requirements, some wrappers may install the JRE (if you have included it when wrapping your application) and/or download and install the required version of the JRE.

The most sophisticated wrappers may also setup file associations and create shortcuts on first run. But if you need something more complex, such as support for automatic updates or uniform cross-platform deployment, have a look at

=== Java-Aware Setup Authoring Tools ===

If all you need is install a private copy of the JRE alongside your application and create shortcuts that run your application on that JRE, you may use any setup generator. However, using a Java-aware tool may give you the following benefits:

* Install-time JRE detection and download
* Generation of native launchers
* User-editable JVM parameter files
* Redirection of stderr and stdout for saving logs and exception stack traces.
* Registration of Java applications as Windows services and Unix daemons

This category is the most diversified in terms of tool pricing and functionality. The differences are explained below by example.

Windows-centric tools, such as Advanced Installer for Java enable you to build MSI (Windows Installer) packages.

Multi-platform tools can generate native installers for multiple platforms - Windows, Linux, Mac OS X, as well as RPMs and tarballs. install4j is one such tool.

There exist also Java-based setup authoring tools enabling you to create cross-platform installations, Those installations are essentially executable jars with platform-specific logic selected at run time. InstallAnywhere is perhaps the most well known tool of this type, but if its pricing is beyond your budget, consider the cheaper JExpress or the open source IzPack.

Finally, there is One Tool to Rule Them All - InstallShield, which can create both Windows desktop (MSI) and cross-platform installations, plus server and mobile ones, for any type of application and for a multitude of platforms. And yes, it does support JRE lookup and bundling, native launchers, and so on.

For straightforward installations, however, InstallShield is an overkill. Also note that InstallAnywhere and InstallShield are aimed at the enterprise developer and are priced accordingly.

All the above solutions do not change the fundamental principle mentioned in the first section of this article. Whether you make an executable jar or create a sophisticated installer, your Java program is still deployed as platform-independent bytecode. In the early days of Java, the only way to execute a Java program on a common PC hardware was to interpret that bytecode. Today, any decent J2SE implementation contains a Just-In-Time (JIT) compiler that compiles frequently executed methods to native code. So it sounds quite natural to take one step further and compile the entire application down to native code before it is deployed. Such tools exist and they are called

=== Ahead-Of-Time Compilers ===

AOT compilers are known also as "static compilers" and "native code compilers". The latter term is the most used and, as it often happens, the least correct from the technical standpoint, because JIT compilers also produce native code.

An Ahead-Of-Time (AOT) compiler takes as input your jars and class files and produces a conventional native executable for the target platform, such as Windows EXE or Linux ELF binary. Just like any other technical solution, this has its advantages and drawbacks.


* Performance. A JIT compiler works at application's runtime and shares CPU and memory resources with the application it compiles and possibly other applications. An AOT compiler runs on the developer's system with no resource or compilation time constraints. Therefore it can potentially use more powerful resource-intensive optimizations, yielding better code.

This advantage is amplified if you application will be deployed to embedded systems or low-end desktop PCs, where the JIT compilers simply may not have enough resources to work.

* Intellectual Property Protection. Java bytecode is very easy to decompile - just google for "download java decompiler" and you will get your source code back in 5 minutes. Yes, you may obfuscate names of public classes and methods not accessed via reflection, but control flow obfuscation can render your bytecode unverifiable on future JVMs and hinders optimizations implemented in JIT compilers. Finally, encrypting your Java bytecode does not protect it at all regardless of the encryption algorithm you use.

In contrast, native code produced by an optimizing AOT Java compiler is about as hard to reverse engineer as if you have coded the original program in C++. Needless to say, there is no performance loss. If you are concerned about protecting your intellectual property, have a closer look at native compilation.

* User Perception. Java client applications often suffer from the so called warm-up cycle syndrome. Starting up a Java application involves bytecode interpretation, profiling and JIT-compilation. So Java programs tend to start much longer than their native counterparts and the initial response time of a Java app GUI element is much worse than after it has been used several times, which are the two major reasons for Java still being perceived as slow by many users.

A native executable runs directly on hardware, without the interpret-profile-compile overhead, so it may start faster and immediately demonstrates the best response times.

* Native deployment. Even the most sophisticated Java-aware setup tools have to generate native launchers for better desktop integration, and may need to take care of the JRE download and installation.

Executables produced by an AOT Java compiler do not depend on the JRE and can be deployed using any setup authoring tool available for the target platform. Moreover, AOT compilers may come with specifically tailored setup generators that create compact, professional installers.


* Dynamic applications. Classes that the application loads dynamically at runtime may be unavailable to the application developer. These can be third-party plug-ins, dynamic proxies and other classes generated at runtime and so on. So the runtime system has to include a Java bytecode interpreter and/or a JIT compiler.

Moreover, in the general case only classes that are loaded by either system or application classloader may be precompiled to native code. So applications that use custom classloaders extensively may only be partially precompiled.

* Hardware-specific optimizations. A JIT compiler has a potential advantage over AOT compilers in that it can select code generation patterns according to the actual hardware on which the application is executing. For instance, it may use Intel MMX/SSE/SSE2 extensions to speedup floating point calculations. An AOT compiler must either produce code for the lowest common denominator or apply versioning to the most CPU-intensive methods, which may result in code size increase.

FYI, there used to be half a dozen AOT Java compilers on the market in the year 2000, but the only two that have survived are Excelsior JET and GCJ (GNU Compiler for Java).


To discuss about this article, please go to JavaLobby forum :-)


dleskov said...

Hey, you have stolen my article!

Mediocre Ninja said...

Hello D. Leskov ,
Sorry, your article impresses me so much that I post it again in this blog.
I have included your name and link to original article already .

It would be very kind of you to give me the permission to post the article here. Otherwise I will delete the article from my blog (soon after you reply).

- ninja,