libgdx/packr

About

Packages your JAR, assets and a JVM for distribution on Windows, Linux and macOS, adding a native executable file to make it appear like a native app. Packr is most suitable for GUI applications, such as games made with libGDX.

On the topic of games, Packr version 2.4.2+ supports Java 14 and the Z garbage collector has been verified to work. Because who doesn't want GC pause times guaranteed to not exceed 10ms with work in progress for sub 1ms GC pauses. When bundling Java 14+ make sure to use --useZgcIfSupportedOs instead of passing --vmargs XX:+UseZGC because versions of Windows before Windows 10 1803 are not supported by the Z garbage collector.

Starting with Java 14, there's a new tool that is included with the JDK called jpackage. There's a lot of overlap between jpackage and packr. Considering jpackage is supported by the broader OpenJDK community, it's worth looking into. It might be a better solution for your product. There's also Conveyor which can build self-updating packages, do signing and notarization, and cross-build (Windows/Mac packages from Linux, Linux/Mac from Windows etc). It's free for open source projects but requires payment for commercial projects.

Download

The latest build is available for download here.

Resource artifacts are available at Maven Central

• Until Maven central publishing is working, the following Maven repository is available:
  • http://artifactory.nimblygames.com/artifactory/ng-public/

Usage

You point packr at your JAR file(s) containing your code and assets, some configuration parameters, and a URL or local file location to a JDK build for your target platform.

Invoking packr from the command line may look like the following. For a more complete example look at the PackrAllTestApp/packrAllTestApp.gradle.kts:

java -jar packr-all.jar \
     --platform mac \
     --jdk OpenJDK11U-jre_x64_mac_hotspot_11.0.10_9.tar.gz \
     --useZgcIfSupportedOs \
     --executable myapp \
     --classpath myjar.jar \
     --mainclass com.my.app.MainClass \
     --vmargs -Xmx1G \
     --resources src/main/resources path/to/other/assets \
     --output out-mac

Alternatively, you can put all the command line arguments into a JSON file which might look like this:

{
    "platform": "mac",
    "jdk": "/Users/badlogic/Downloads/OpenJDK8U-jdk_x64_mac_hotspot_8u252b09.tar.gz",
    "executable": "myapp",
    "classpath": [
        "myjar.jar"
    ],
    "removelibs": [
        "myjar.jar"
    ],
    "mainclass": "com.my.app.MainClass",
    "vmargs": [
       "-Xmx1G"
    ],
    "resources": [
        "src/main/resources",
        "path/to/other/assets"
    ],
    "minimizejre": "soft",
    "output": "out-mac"
}

You can then invoke the tool like this:

java -jar packr-all.jar my-packr-config.json

It is possible to combine a JSON configuration, and the command line. For single options, the command line parameter overrides the equivalent JSON option. For multi-options (e.g. classpath or vmargs), the options are merged.

This is an example which overrides the output folder and adds another VM argument. Note that the config file name is delimited by -- because the option prior to it, --vmargs, allows multiple arguments:

java -jar packr-all.jar --output target/out-mac --vmargs -Xms256m -- my-packr-config.json

Finally, you can use packr from within your Java code. Just add the JAR file to your project, either manually, or via the following Gradle dependency:

repositories {
   mavenCentral() // Packr artifacts will be published to Maven Central in the future
   maven(uri("https://oss.sonatype.org/content/repositories/snapshots/")) // Packr snapshot artifacts will be published to Maven Central in the future

   // The following repositories are available until artifacts can be published to Maven Central
   maven(uri("http://artifactory.nimblygames.com/artifactory/ng-public-snapshot/"))
   maven(uri("http://artifactory.nimblygames.com/artifactory/ng-public-release/"))
}
dependencies {
   implementation("com.badlogicgames.packr:packr:3.0.3")
}

To invoke packr, you need to create an instance of PackrConfig and pass it to Packr.pack():

PackrConfig config = new PackrConfig();
config.platform = PackrConfig.Platform.Windows32;
config.jdk = "/User/badlogic/Downloads/openjdk-for-mac.zip";
config.executable = "myapp";
config.classpath = Arrays.asList("myjar.jar");
config.removePlatformLibs = config.classpath;
config.mainClass = "com.my.app.MainClass";
config.vmArgs = Arrays.asList("-Xmx1G");
config.minimizeJre = "soft";
config.outDir = new java.io.File("out-mac");
config.useZgcIfSupportedOs = true;

new Packr().pack(config);

macOS notarization and entitlements

The following entitlements when signing the PackrLauncher executable are known to work on macOS 10.15 (Catalina) and Java 14.

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
	<key>com.apple.security.cs.allow-jit</key>
	<true/>
	<key>com.apple.security.cs.allow-unsigned-executable-memory</key>
	<true/>
	<key>com.apple.security.cs.disable-executable-page-protection</key>
	<true/>
	<key>com.apple.security.cs.disable-library-validation</key>
	<true/>
	<key>com.apple.security.cs.allow-dyld-environment-variables</key>
	<true/>
</dict>
</plist> 

If all the bundled dylibs are signed, fewer entitlements might be possible. When using Java 8, com.apple.security.cs.allow-unsigned-executable-memory, and com.apple.security.cs.disable-executable-page-protection were not needed.

Example macOS code signing and notarization command line steps

These steps assume you have an Apple developer account, have saved your Apple code signing certificate into Keychain and have generated an app-specific password for your Apple developer account, allowing you to pass your username and token as command line arguments. The example commands also assume you saved the app-specific password in your Keychain allowing these commands to run in an automated way, e.g., your CI pipeline can execute all these commands.

1. codesign --sign <keychain id for certiticate> --verbose=10 --timestamp --force --options runtime --entitlements <path-to-entitlements-file> <path to exe or shared lib>
   • You have to codesign every executable and shared library, --deep is for "emergency repairs".
2. /usr/bin/ditto -c -k --keepParent <app path> <app path>.zip
   • ditto is a commandline zip tool, any tool that creates a zip file from a directory can be used.
3. xcrun altool --notarize-app --verbose --primary-bundle-id com.mydomain.myproduct --username '<username>' --password "@keychain:<app-specific password>" --file <app path>.zip
   • If this step fails, it will exit with a non-zero return code and provide good output as to why it failed. E.g., "You must first sign the relevant contracts online."

Optional steps, you can choose to wait for an email notification

1. xcrun altool --notarization-history 0 -u <username> -p "@keychain:<app-specific password>" --output-format xml
   • This command grabs the history for the last call to xcrun altool --notarize-app, this will obviously fail if you're running multiple xcrun altool --notarize-app processes in parallel. You'll have to come up with a better way to parse the history.
2. Parse the XML output for the last request UUID, regex: <string>(.*?)</string>
3. In a loop, every minute check the notarization status.
   • xcrun altool --notarization-info <parsed uuid> -u <username> -p "@keychain:<app-specific password>"
4. Parse the output for the status, regex: .*?Status:\s+(.*?)$
5. When the status no longer matches in progress exit the loop.
6. If the Status did not end up as success the output will provide a description of what went wrong.
7. xcrun stapler staple --verbose <app path>

Minimization

Unless you're stuck with using Java 8, it's best to create a minimized JRE using jlink. See TestAppJreDist/testAppJreDist.gradle.kts for an example Gradle build script which generates JREs from downloaded JDKs.

JRE

A standard OpenJDK 8 JRE is about 91 MiB unpacked. Packr helps you cut down on that size, thus also reducing the download size of your app.

To minimize the JRE that is bundled with your app, you have to specify a minimization configuration file via the minimizejre flag you supply to Packr. A minimization configuration is a JSON file containing paths to files and directories within the JRE to be removed.

As an example, have a look at the soft profile configuration:

{
  "reduce": [
    {
      "archive": "jre/lib/rt.jar",
      "paths": [
        "com/sun/corba",
        "com/sun/jndi",
        "com/sun/media",
        "com/sun/naming",
        "com/sun/rowset",
        "sun/applet",
        "sun/corba",
        "sun/management"
      ]
    }
  ],
  "remove": [
    {
      "platform": "*",
      "paths": [
        "jre/lib/rhino.jar"
      ]
    },
    {
      "platform": "windows",
      "paths": [
        "jre/bin/*.exe",
        "jre/bin/client"
      ]
    }
  ]
}

This configuration will unpack rt.jar, remove all the listed packages and classes in com.sun.* and sun.*, then repack rt.jar again. By default, the JRE uses zero-compression on its JAR files to make application startup a little faster, so this step will reduce the size of rt.jar substantially.

Then, rhino.jar (about 1.1 MiB) and, in the JRE for Windows case, all executable files in jre/bin/ and the folder jre/bin/client/ will be removed.

Packr comes with two such configurations out of the box: soft and hard. The hard profile removes a few more files, and repacks some additional JAR files.

The "removelibs" option

Minimization aside, packr can remove all dynamic libraries which do not match the target platform from your project JAR file(s):

This step is optional. If you don't need it, just remove the configuration parameter to speed up packr. This step doesn't preserve the META-INF directory or files in the jar.

Caching

Extracting and minimizing a JRE can take quite some time. When using the cachejre option, the result of these operations are cached in the given folder, and can be reused in subsequent runs of packr.

As of now, packr doesn't do any elaborate checks to validate the content of this cache folder. So if you update the JDK, or change the minimize profile, you need to empty or remove this folder manually to force a change.

Output

Windows

When packing for Windows, the following folder structure will be generated

outdir/
   myapp.exe
   myjar.jar
   myapp.json
   jre/

Linux

outdir/
   myapp
   myjar.jar
   myapp.json
   jre/

Mac OS X

outdir/
   Contents/
      Info.plist
      MacOS/
         myapp
      Resources/
         myjar.jar
         myapp.json
         jre/
         icons.icns [if config.icon is set]

You can further modify the Info.plist to your liking, e.g. add icons, a bundle identifier etc. If your output folder has the .app extension it will be treated as an application bundle by Mac OS X.

Executable command line interface

By default, the native executables forward any command line parameters to your Java application's main() function. So, with the configurations above, ./myapp -x y.z is passed as com.my.app.MainClass.main(new String[] {"-x", "y.z" }).

The executables themselves expose an own interface, which has to be enabled explicitly by passing -c or --cli as the very first parameter. In this case, the special delimiter parameter -- is used to separate the native CLI from parameters to be passed to Java. In this case, the example above would be equal to ./myapp -c [arguments] -- -x y.z.

Try ./myapp -c --help for a list of available options.

Note: On Windows, the executable does not show any output by default. Here you can use myapp.exe -c --console [arguments] to spawn a console window, making terminal output visible.

Building from source code

If you want to modify the code invoke Gradle.

$ ./gradlew clean assemble

This will create a packr-VERSION-all.jar file in Packr/build/libs directory, you may invoke as described in the Usage section above.

Gradle project structure

The Gradle build is set up as a multi-project build. In order to fully build the multi-project you must have a compatible JRE (Java 8+) and C/C++ build tools that the Gradle build can find.

DrOpt Gradle sub-project

This is a downloaded and unzipped https://github.com/jamesderlin/dropt/releases version 1.1.1 source code with a Gradle script used to build it for consumption by the PackrLauncher Gradle project. The DrOpt source required a few modifications to get it compiling, namely some explicit casting in the C code.

Packr Gradle sub-project

This is the Java code for creating application bundles that can use the native launcher executables. This project also builds the packr-all uber/shadow jar that works as an executable jar.

PackrLauncher Gradle sub-project

This contains the platform native code for loading the JVM and starting the packr bundled application.

PackrAllTestApp Gradle sub-project

This is an example Hello world style application that bundles itself using packr and is used as a high level test suite to help reduce breaking changes.

TestAppJreDist Gradle sub-project

This project downloads JDKS 8, 11, and 14 and runs jlink on the 11 and 14 versions to create minimal JREs for use by PackrAllTestApp.

Limitations

• Only Adopt OpenJDKs 8, 11, and 15 are tested (other JDKs probably work)
• Icons aren't set yet on Windows and Linux, you need to do that manually.
• Minimum platform requirement on MacOS is OS X 10.10 (Only 10.15 macOS Catalina is actively tested, there are users that report 10.14 works).
• JRE minimization is very conservative. Depending on your app, you can carve out stuff from a JRE yourself, disable minimization and pass your custom JRE to packr. If you're using Java 11+ you should create a JRE using jlink.
• On MacOS, the JVM is spawned in its own thread by default, which is a requirement of AWT. This does not work with code based on LWJGL3/GLFW, which needs the JVM be spawned on the main thread. You can enforce the latter with adding the -XstartOnFirstThread VM argument to your MacOS packr config.

License & Contributions

The code is licensed under the Apache 2 license. By contributing to this repository, you automatically agree that your contribution can be distributed under the Apache 2 license by the author of this project. You will not be able to revoke this right once your contribution has been merged into this repository.

Security

Distributing a bundled JVM has security implications, just like bundling any other runtimes like Mono, Air, etc. Make sure you understand the implications before deciding to use this tool. Here's a discussion on the topic.