README

mx is a command line based tool for managing the development of (primarily) Java code. It includes a mechanism for specifying the dependencies as well as making it simple to build, test, run, update, etc the code and built artifacts. mx contains support for developing code spread across multiple source repositories. mx is written in Python (version 2.7) and is extensible.

The organizing principle of mx is a suite. A suite is both a directory and the container for the components of the suite. A suite component is either a project, library or distribution. There are various flavors of each of these. A suite may import and depend on other suites. For an execution of mx, exactly one suite is the primary suite. This is either the suite in whose directory mx is executed or the value of the global -p mx option. The set of suites reachable from the primary suite by transitive closure of the imports relation form the set that mx operates on.

Running mx

mx can be run directly (i.e., python2.7 mx/mx.py ...), but is more commonly invoked via the mx/mx bash script (which includes a Python version check). Adding the mx/ directory to your PATH simplifies executing mx. The mx/mx.cmd script should be used on Windows.

The general form of the mx command line is:

mx [global options] [command] [command-specific options]

If no options or command is specified, mx prints information on the available options and commands, which will include any suite-specfic options and commands. Help for a specific command is obtained via mx help <command>. Global options are expected to have wide applicability to many commands and as such precede the command to be executed.

For an example of mx usage, see the README for the Graal project.

Note: There is a Bash completion script for global options and commands, located in bash_completion directory. Install it for example by sourceing this script in your ~/.bashrc file. If used, a temporary file /tmp/mx-bash-completion-<project-path-hash> is created and used for better performance. This should be OK since the /tmp directory is usually cleaned on every system startup.
mx-honey provides richer completions for zsh users.

Suites

The definition of a suite and its components is in a file named suite.py in the mx metadata directory of the primary suite. This is the directory named mx.<suite name> in the suite's top level directory. For example, for the compiler suite, it is mx.compiler. The format of suite.py is JSON with the following extensions:

• Python multi-line and single-quoted strings are supported
• Python hash comments are supported

Java projects

Java source code is contained in a project. Here's an example of two Graal compiler projects:

"org.graalvm.compiler.serviceprovider" : {
  "subDir" : "src",
  "sourceDirs" : ["src"],
  "dependencies" : ["JVMCI_SERVICES"],
  "checkstyle" : "org.graalvm.compiler.graph",
  "javaCompliance" : "8",
  "workingSets" : "API,Graal",
},

"org.graalvm.compiler.serviceprovider.jdk9" : {
  "subDir" : "src",
  "sourceDirs" : ["src"],
  "dependencies" : ["org.graalvm.compiler.serviceprovider"],
  "uses" : ["org.graalvm.compiler.serviceprovider.GraalServices.JMXService"],
  "checkstyle" : "org.graalvm.compiler.graph",
  "javaCompliance" : "9+",
  "multiReleaseJarVersion" : "9",
  "workingSets" : "API,Graal",
},

The javaCompliance attribute can be a single number (e.g. 8), the lower bound of a range (e.g. 8+) or a fixed range (e.g. 9..11). This attribute specifies the following information:

• The maximum Java language level used by the project. This is the lower bound in a range. It is also used at the value for the -source and -target javac options when compiling the project.
• The JDKs providing any internal JDK API used by the project. A project that does not use any internal JDK API should specify an open range (e.g. 8+). Otherwise, a JDK matching the exact version or range is required to compile the project.

The multiReleaseJarVersion attribute is explained in the "Versioning sources for different JDK releases" section below.

Java distributions

A distribution encompasses one or more Java projects and enables the class files and related resources from projects to be packaged into a jar file. If a distribution declares itself as a module (see Java modules support), a JMOD file will also be produced when the distribution is built. The path to the jar file for a distribution is given by mx paths <distribution name>. For example:

> mx paths GRAAL
/Users/dnsimon/graal/graal/compiler/mxbuild/dists/jdk11/graal.jar

Java modules support

A distribution that has a moduleInfo attribute will result in a Java module being built from the distribution. The moduleInfo attribute must specify the name of the module and can include other parts of a module descriptor.

This is best shown with examples from Truffle and Graal:

Here is an extract from the definition of the TRUFFLE_API distribution which produces the org.graavm.truffle module:

"TRUFFLE_API" : {
    "moduleInfo" : {
        "name" : "org.graalvm.truffle",
        "requires" : [
            "static java.desktop"
        ],
        "exports" : [
            "com.oracle.truffle.api.nodes to jdk.internal.vm.compiler",
            "com.oracle.truffle.api.impl to jdk.internal.vm.compiler, org.graalvm.locator",
            "com.oracle.truffle.api to jdk.internal.vm.compiler, org.graalvm.locator, com.oracle.graal.graal_enterprise",
            "com.oracle.truffle.api.object to jdk.internal.vm.compiler, com.oracle.graal.graal_enterprise",
            "com.oracle.truffle.object to jdk.internal.vm.compiler, com.oracle.graal.graal_enterprise",
        ],
        "uses" : [
          "com.oracle.truffle.api.TruffleRuntimeAccess",
          "java.nio.file.spi.FileTypeDetector",
          "com.oracle.truffle.api.impl.TruffleLocator",
        ],
    },
    ...
    "distDependencies" : [
        # These distributions must also have `moduleInfo` attributes and the corresponding
        # modules will be added to the set of `requires` for this module.
        "sdk:GRAAL_SDK"
    ],
}

The module-info.java created by mx from the above is:

module org.graalvm.truffle {
    requires java.base;
    requires static java.desktop;
    requires java.logging;
    requires jdk.unsupported;
    requires transitive org.graalvm.sdk;
    exports com.oracle.truffle.api to com.oracle.graal.graal_enterprise, jdk.internal.vm.compiler, org.graalvm.locator;
    exports com.oracle.truffle.api.impl to jdk.internal.vm.compiler, org.graalvm.locator;
    exports com.oracle.truffle.api.nodes to jdk.internal.vm.compiler;
    exports com.oracle.truffle.api.object to com.oracle.graal.graal_enterprise, jdk.internal.vm.compiler;
    exports com.oracle.truffle.object to com.oracle.graal.graal_enterprise, jdk.internal.vm.compiler;
    uses com.oracle.truffle.api.TruffleRuntimeAccess;
    uses com.oracle.truffle.api.impl.TruffleLocator;
    uses com.oracle.truffle.api.object.LayoutFactory;
    uses java.nio.file.spi.FileTypeDetector;
    provides com.oracle.truffle.api.object.LayoutFactory with com.oracle.truffle.object.basic.DefaultLayoutFactory;
    provides org.graalvm.polyglot.impl.AbstractPolyglotImpl with com.oracle.truffle.polyglot.PolyglotImpl;
    // conceals: com.oracle.truffle.api.debug
    // conceals: com.oracle.truffle.api.debug.impl
    // conceals: com.oracle.truffle.api.dsl
    // conceals: com.oracle.truffle.api.frame
    // conceals: com.oracle.truffle.api.instrumentation
    // conceals: com.oracle.truffle.api.interop
    // conceals: com.oracle.truffle.api.interop.impl
    // conceals: com.oracle.truffle.api.io
    // conceals: com.oracle.truffle.api.library
    // conceals: com.oracle.truffle.api.object.dsl
    // conceals: com.oracle.truffle.api.profilesLayoutFactory
    // conceals: com.oracle.truffle.api.source
    // conceals: com.oracle.truffle.api.utilities
    // conceals: com.oracle.truffle.object.basic
    // conceals: com.oracle.truffle.polyglot
    // jarpath: /Users/dnsimon/hs/graal/truffle/mxbuild/dists/jdk11/truffle-api.jar
    // dist: TRUFFLE_API
    // modulepath: org.graalvm.sdk
}

The provides clauses are automatically derived from the META-INF/services/ directory in the distribution's jar file.

The GRAAL distribution shows how a single exports attribute can be used to specify multiple exports clauses:

"GRAAL" : {
    "moduleInfo" : {
        "name" : "jdk.internal.vm.compiler",
        "exports" : [
            # Qualified exports of all packages in GRAAL to modules built from
            # ENTERPRISE_GRAAL and GRAAL_MANAGEMENT distributions
            "* to com.oracle.graal.graal_enterprise,jdk.internal.vm.compiler.management",
        ],
        ...
    },
    ...
},

This results info a module-info.java as that contains qualified exports, a small subset of which are shown below:

module jdk.internal.vm.compiler {
    ...
    exports org.graalvm.compiler.api.directives to com.oracle.graal.graal_enterprise, jdk.internal.vm.compiler.management;
    exports org.graalvm.compiler.api.replacements to com.oracle.graal.graal_enterprise, jdk.internal.vm.compiler.management;
    exports org.graalvm.compiler.api.runtime to com.oracle.graal.graal_enterprise, jdk.internal.vm.compiler.management;
    exports org.graalvm.compiler.asm to com.oracle.graal.graal_enterprise, jdk.internal.vm.compiler.management;
    exports org.graalvm.compiler.asm.aarch64 to com.oracle.graal.graal_enterprise, jdk.internal.vm.compiler.management;
    exports org.graalvm.compiler.asm.amd64 to com.oracle.graal.graal_enterprise, jdk.internal.vm.compiler.management;
    ...

The jars build for a distribution are in <suite>/mxbuild/dists/jdk*/. The modular jars are in the jdk<N> directories where N >= 9. There is a modular jar built for each JDK version denoted by the javaCompliance values of the distribution's constituent projects.

Specifying required modules

If a project uses a package from a module other than java.base or a module implied by a dependency (e.g., the JVMCI_API library defined by Graal), it must specify these additional modules with the requires attribute. For example:

"org.graalvm.compiler.hotspot.management.jdk11" : {
    ...
    "requires" : [
        "jdk.management"
    ],
    "javaCompliance" : "11+",
    ...
},

The requires attribute is used for two purposes:

• As input to the requires attribute of the descriptor for the module encapsulating the project.
• To derive a value for the --limit-modules javac option which restricts the modules observable during compilation. This is required to support separate compilation of projects that are part of a JDK module. For example, org.graalvm.compiler.hotspot.amd64 depends on org.graalvm.compiler.hotspot and the classes of both these projects are contained in the jdk.internal.vm.compiler module. When compiling org.graalvm.compiler.hotspot.amd64, we must compile against classes in org.graalvm.compiler.hotspot as they might be different (i.e., newer) than the classes in jdk.internal.vm.compiler. The value of --limit-modules will omit jdk.internal.vm.compiler in this case to achieve this hiding. In the absence of a requires attribute, only the java.base module is observable when compiling on JDK 9+.

Use of concealed packages

Concealed packages are those defined by a module but not exported by the module. If a project uses concealed packages, it must specify a requiresConcealed attribute denoting the concealed packages it accesses. For example:

"org.graalvm.compiler.lir.aarch64.jdk11" : {
    "requiresConcealed" : {
        "jdk.internal.vm.ci" : [
            "jdk.vm.ci.aarch64",
            "jdk.vm.ci.code",
        ],
    },
    "javaCompliance" : "11+",
},

This will result in --add-exports=jdk.internal.vm.ci/jdk.vm.ci.aarch64=ALL-UNNAMED and --add-exports=jdk.internal.vm.ci/jdk.vm.ci.code=ALL-UNNAMED being added to the javac command line when the org.graalvm.compiler.lir.aarch64.jdk11 project is compiled by a JDK 9+ javac.

Note that the requires and requiresConcealed attributes only apply to projects with a minimum javaCompliance value of 9 or greater. When javac from jdk 9+ is used in conjunction with -source 8 (as will be the case for projects with a minimum javaCompliance of 8 or less), all classes in the JDK are observable. However, if an 8 project would need a requires or requiresConcealed attribute were it a 9+ project, then these attributes must be applied to any module containing the project. For example, org.graalvm.compiler.serviceprovider has "javaCompliance" : "8+" and contains code that imports sun.misc.Unsafe. Since org.graalvm.compiler.serviceprovider is part of the jdk.internal.vm.compiler module defined by the GRAAL distribution, GRAAL must include a requires attribute in its moduleInfo attribute:

"GRAAL" : {
    "moduleInfo" : {
        "name" : "jdk.internal.vm.compiler",
        "requires" : ["jdk.unsupported"],
        ...
    }
}

Selecting JDKs

Specifying JDKs to mx is done via the --java-home and --extra-java-homes options or via the JAVA_HOME and EXTRA_JAVA_HOMES environment variables. An option has precedence over the corresponding environment variable. Mx comes with a select_jdk.py helper that simplifies switching between different values for JAVA_HOME and EXTRA_JAVA_HOMES.

Unit testing with Junit

The unittest command supports running Junit tests in mx suites.

The unit test harness will use any org.junit.runner.notification.RunListener objects available via java.util.ServiceLoader.load().

Executing tests on JDK 9 or later can be complicated if the tests access packages that are publicly available in JDK 8 or earlier but are not public as of JDK 9. That is, the packages are concealed by their declaring module. Such tests can be compiled simply enough by specifying their Java compliance as "1.8=". Running the tests on JDK 9 however requires that the concealed packages are exported to the test classes. To achieve this, an AddExports annotation should be applied to the test class requiring the export or to any of its super classes or super interfaces. To avoid the need for a dependency on mx, unittest harness simply looks for an annotation named AddExports that matches the following definition:

import java.lang.annotation.ElementType;
import java.lang.annotation.Retention;
import java.lang.annotation.RetentionPolicy;
import java.lang.annotation.Target;

/**
 * Specifies packages concealed in JDK modules used by a test. The mx unit test runner will ensure
 * the packages are exported to the module containing annotated test class.
 */
@Retention(RetentionPolicy.RUNTIME)
@Target(ElementType.TYPE)
public @interface AddExports {
    /**
     * The qualified name of the concealed package(s) in {@code <module>/<package>} format (e.g.,
     * "jdk.vm.ci/jdk.vm.ci.code").
     */
    String[] value() default "";
}

Coverage testing with JaCoCo

To enable code coverage testing with JaCoCo, the JaCoCo agent needs to be injected through VM command line arguments. For this, mx provides the convenience method mx_gate.get_jacoco_agent_args() which returns a list of those arguments if coverage is requested (e.g. by using mx gate --jacocout ...), otherwise None. Here is an example how it is used to enable coverage testing of the sources of the Graal compiler. Running code with the JaCoCo agent enabled outputs a jacoco.exec which can be converted into an HTML or CSV report with the mx jacocoreport command.

Versioning sources for different JDK releases

Mx includes support for multiple versions of a Java class. The mechanism is inspired by and similar to multi-release jars. A versioned Java class has a base version and one or more versioned copies. The public signature of each copy (i.e., methods and fields accessed from outside the source file) must be identical. Note that the only API that is visible from the JAR is the one from the base version.

Versioned classes for JDK 9 or later need to be in a project with a javaCompliance greater than or equal to 9 and a multiReleaseJarVersion attribute whose value is also greater or equal to 9. The versioned project must have the base project as a dependency.

Versioned classes for JDK 8 or earlier need to be in a project with a javaCompliance less than or equal to 8 and an overlayTarget attribute denoting the base project.

URL rewriting

Mx includes support for the primary suite to be able to override the source URLs of imported suites. The suite level urlrewrites attribute allows regular expression URL rewriting. For example:

  "urlrewrites" : [
    {
      "https://git.acme.com/(.*).git" : {
        "replacement" : r”https://my.company.com/foo-git-cache/\1.git",
      }
    },
    {
      "https://hg.acme.com/(.*)" : {
        "replacement" : r”https://my.company.com/foo-hg-cache/\1",
      }
    }
  ],

The rules are applied in definition order. Only rewrite rules from the primary suite are used meaning a suite may have to replicate the rewrite rules of its suite dependencies. This allows the primary suite to retain full control over where its dependencies are sourced from.

Rewrite rules can also be specified by the MX_URLREWRITES environment variable. The value of this variable must either be a JSON object describing a single rewrite rule, a JSON array describing a list of rewrite rules or a file containing one of these JSON values. Rewrites rules specified by MX_URLREWRITES are applied after rules specified by the primary suite.

Environment variable processing

Suites might require various environment variables to be defined for the suite to work and mx provides env files to cache this information for a suite. Each suite can have an env file in suite/mx.suite/env and a default env file can be provided for the user in ~/.mx/env. Env files are loaded in the following order and override any value provided by the shell environment.

1. ~/.mx/env is loaded first.

2. The primary suite's env file is loaded before loading of the suites begins.

3. The env files of any subsuites are loaded in a depth first fashion such that subsuite env files are loaded before their dependents.

4. The primary suite's env file is reloaded so that it overrides any definitions provided by subsuites.

The -v option to mx will show the loading of env files during suite parsing.

Multiple suites per repository

Sometimes it might be convenient to group multiple suites inside a single repository. In particular, this helps ensure that all these suites are synchronized and tested together.

• A suite inside a 'big repo' must be in a directory that has the same name as the suite
• If you depend on a suite that is inside a 'big repo', you have to set subdir to True in the suite import.
• If you depend on a suite that is in the same 'big repo' as the current suite, you should not specify urls in the suite import.
• In order to sclone something that is inside a 'big repo' you have to use the --subdir argument for sclone which tells in which directory the suite that you want to clone is
• In order to dynamically import a suite that is inside a 'big repo' you have to use --dynamicimport bigrepo/suite (e.g., --dynamicimport graal-enterprise/substratrevm)

Note that a suite in a "big repo" should not have a dependency to a suite in a different repository that in turn has a transitive dependency to the same "big repo". In other words, there should be no back-and-forth to the same repo.

mx versioning

mx uses a major.minor.patch versioning scheme. To figure out if the version is sufficient for a given mx suite, first compare the major version number of your mx version against the major number of the required version specified in the suite. If these versions are not equal, you cannot expect mx to be compatible with this suite. The minor number has to be greater or equal to the specified minor version number. Compatibility is ensured regardless of the patch level. However, if your patch level is lower than the required patch level you might trigger bugs in mx.

From an mx developer point of view this versioning scheme enforces the following update policy:

• If you make a change that prevents the new version of mx from loading older files, increase the major number and reset both the minor and the patch level to 0.
• If you add new functionality without breaking backward compatibility, leave the major as it is, increase the minor number and reset the patch level.
• If you only fixed a bug without changing the public API (i.e., all files for the current version can still be loaded with the new version and vice versa), leave the major and minor versions as they are but increase the patch level.

The version update strategy is designed to help users to detect if their mx version is compatible with a suite. Thus, changes to the code that do not affect users do not require a change in the version number. See the following examples. In these examples, by user we mean command line clients or mx extensions (for example mx_graal-core.py).

• "I found a for-loop in the code that could be expressed using a map function. I changed it accordingly." This change has no influence on users. Thus, no version change is required!

• "I added a new mx command." Since this function was not available to users before, old scripts will continue to work with the new version. New scripts, however, might not work with old versions. This is a minor update and requires a new minor number and a reset of the patch level.

• "I fixed a bug that caused a wrong result of a publicly available function." This is a bugfix that is user visible. The patch level should be increased since users of old versions can expect at least the bug that was just fixed.

• "I fixed some documentation." This fix has no impact on the usage of mx and should thus not change the version of mx.

• "I fixed a function. The result now differs from the results before. A user cannot call this function." Since this function is invisible to the user, no version update is required.

• "I fixed a function. The result now differs from the results before. A user could call this function." Since the semantics of the function changed and the function is part of the API, old scripts might not work properly anymore. Since this change is not backward compatible, this is a major update.

• "I added some internal functions." Since the functions are internal, they have no impact on users. No version changed is required.

• "I added some new commands." Since the commands did not change the old commands, old scripts will continue to work as expected. New scripts that depend on the new commands will not work with older versions of mx. Thus, we need a new minor release.

• "I removed some commands from mx. There are alternative commands now." This change essentially changed the API. Thus, we require a new major release.