maven-native-image exec agent not generating target/ META-INF/native-image directory

[warrenc5]

Describe the problem
Guys - the plugin product is fantastic but the documentation is terribly difficult for me to follow/comprehend.

I've been working with maven and graal native images for a long time.
And have contributed to graalvm bug reporting previously. So please accept my constructive criticism kindly.

I love the way you use maven profiles but I think you should focus on the acutal plugin features with skip options clearly used and let the dev refactor their pom to use profiles and execution ids. Certainly mention them for larger project build optimizations as they are increadibly powerful.

There are exisiting 'defacto standards' for maven plugin documentation and your presenting seems mixed between a man page and a learn by example, but because you don't seem to have a complete cohesive example you have to keep annotating and referencing.. i.e. but it can be enabled in the pom.xml file - customize the plugin within the block.

I know nothing is perfect - I wrote https://github.com/warrenc5/jarwalker https://github.com/warrenc5/jain-slee-js https://github.com/warrenc5/graaljs-autobox and now I'm writing https://github.com/warrenc5/graalson https://github.com/warrenc5/graalson-trax and https://github.com/warrenc5/json-transformer - certainly not perfect :)

I am referring to https://github.com/warrenc5/json-transformer/blob/master/pom.xml

The exec is not picking up the graal agent from the native-maven-plugin and I'm unsure why.

I know/think the agent should generate a native-image directory in the target directory and then I can copy that to src/main/resources/META-INF for native-image compilation.

I understand how agents are loaded by java (--javaagent:) but I don't see that happening on the exec using mvn -X test. Should it happen automagically or do I need to add the jvmArgs to the exec plugin? - I can see how it could with surefire but exec plugin is isolated by the fork.

I looked at to see if there was a configure-agent mojo..

mvn help:describe -Dplugin=org.graalvm.buildtools:native-maven-plugin -Dfull

I am missing something conceptually but I don't know what and can't glean it from the documentation I briefly looked the examples - but couldn't find any thing relevant. Sorry.

There seems to be a duplication error in the documentation with the xml fragment - those two plugin configurations appear exactly the same. Then you refer to exec:exec@native but I don't see that defined anywhere

Is that my conflation? Should there be an additional configuration in the plugins?

https://graalvm.github.io/native-build-tools/latest/maven-plugin.html#agent-support-running-application

[DEBUG] Executing command line: [java, -classpath, /home/wozza/code/json-transformer/target/classes:/home/wozza/.m2/repository/mobi/mofokom/graalson-trax/1.0.3-SNAPSHOT/graalson-trax-1.0.3-SNAPSHOT.jar:/home/wozza/.m2/repository/org/slf4j/slf4j-api/2.0.7/slf4j-api-2.0.7.jar:/home/wozza/.m2/repository/mobi/mofokom/graalson/1.0.5-SNAPSHOT/graalson-1.0.5-SNAPSHOT.jar:/home/wozza/.m2/repository/org/graalvm/polyglot/polyglot/25.0.1/polyglot-25.0.1.jar:/home/wozza/.m2/repository/org/graalvm/sdk/collections/25.0.1/collections-25.0.1.jar:/home/wozza/.m2/repository/org/graalvm/sdk/nativeimage/25.0.1/nativeimage-25.0.1.jar:/home/wozza/.m2/repository/org/graalvm/sdk/word/25.0.1/word-25.0.1.jar:/home/wozza/.m2/repository/org/graalvm/js/js-language/25.0.1/js-language-25.0.1.jar:/home/wozza/.m2/repository/org/graalvm/regex/regex/25.0.1/regex-25.0.1.jar:/home/wozza/.m2/repository/org/graalvm/truffle/truffle-api/25.0.1/truffle-api-25.0.1.jar:/home/wozza/.m2/repository/org/graalvm/shadowed/icu4j/25.0.1/icu4j-25.0.1.jar:/home/wozza/.m2/repository/org/graalvm/shadowed/xz/25.0.1/xz-25.0.1.jar:/home/wozza/.m2/repository/org/graalvm/truffle/truffle-runtime/25.0.1/truffle-runtime-25.0.1.jar:/home/wozza/.m2/repository/org/graalvm/sdk/jniutils/25.0.1/jniutils-25.0.1.jar:/home/wozza/.m2/repository/org/graalvm/truffle/truffle-compiler/25.0.1/truffle-compiler-25.0.1.jar:/home/wozza/.m2/repository/jakarta/json/jakarta.json-api/2.1.3/jakarta.json-api-2.1.3.jar, JsonT, test-classes/template.js, test-classes/ALL_FINES.json, all_fines_nsw.json, UTF8]

I think it should be something including the agentlib -

-agentlib:native-image-agent

This blocked worked - but I would like to use the confuration to configure the agent - and I don't know how that works.

<plugin>
    <groupId>org.codehaus.mojo</groupId>
    <artifactId>exec-maven-plugin</artifactId>
    <version>3.0.0</version>
    <executions>
        <execution>
            <id>java-agent</id>
            <goals>
                <goal>exec</goal>
            </goals>
            <phase>test</phase>
            <configuration>
                <executable>java</executable>
                <workingDirectory>${project.build.directory}</workingDirectory>
                <arguments>
                    <argument>-agentlib:native-image-agent=config-output-dir=${basedir}/target/native-image/</argument>
                    <argument>-classpath</argument>
                    <classpath/>
                    <argument>${mainClass}</argument>
                </arguments>
            </configuration>
        </execution>
    </executions>
</plugin>

I'm confused - Any clarity your can provide around this is greatly appreciated - thanks for your time.

[lenborje]

Not familiar with maven, but at least when using Gradle, the agent writes the collected data to a temporary working area, similar to where compiled classes etc are written before packaging to a jar/application/whatever.

This collected data then needs to be moved to the src/main/resources/META-INF/native-image/ directory, or, if you already have metadata there from a previous run, merged with that data.

But this is where things have recently started to get tricky. In previous GraalVM versions, the metadata was split up into several different files, but now there seems to be only one, reachability-metadata.json. Problem is, the tool we previously have used to merge all the metadata files, native-image-configure, can't handle the new file! Nor does the Gradle task metadataCopy work either.

I'm still struggling with how to handle the new metadata. I have a hack in Gradle (basically using JsonSlurper and JsonOutput and some Groovy map-merging), but this is very primitive.

I'm hoping to be saved by a tool update, but the whole design of native image metadata seems to be in flux for the moment.

[warrenc5]

Thanks @lenborje for explaining that the "schmozzle" is the reason I was so confused.

Ironically - I'm upgrading Graalson a Graal/Truffle/Polyglot based implementation of Json-Api jsr 374 to support the following..

• JSON Patch - Object & Array data-type - explicit primitive add/replace/delete operations.
• JSON Merge Patch - Object-only key replacement or key addition.
• Custom JAXP javax.transform support for custom Javascript transformer template using Polyglot bindings. - Json-Transformer <-- see the ReachabilityTestIT
• Java Language 21
• jakarta.json - JEE 10

My question was really about how the agent gets it configuration when run from exec maven plugin.

I need to get up to speed on these Graal native image changes, before I can be any more useful :)
What version did the change happen in Btw?