Metadata repository and polyglot programming demo

Preparation

This really only works on Linux systems as of 08 Dec 2022

Start with a recent Linux distribution; this is tested on Linux 8.9 and 9.x.

• Install Oracle GraalVM Enterprise Edition v 22.3 or later.
• Install gradle
• Clone this repository.

Metadata

GraalVM compilation depends on the idea that the compiler can trace the main entry point and identify all the classes required to execute the code. It works pretty well unless the design firewalls some part of the code, making it impossible for the compiler to determine what classes are air-gapped. There are 2 strategies for dealing with this:

1. run the code once using the JIT compiler and instrument the code. The instrumentation can tell what classes are loaded and mark them for inclusion.
2. use similar data, pre-built from standard distributions.

Oracle maintains a metadata repository, allowing users to post data needed to compile many standard open source packages. One such package is the H2 in-memory database. Since the JDBC driver is instantiated with Class.forName(...), the compiler can't trace down into H2's client to determine what needs to be included in the final executable. With the Metadata repository, that information can be looked up at compile time, and a complete executable generated.

Inspect the code: In src/main/java/com/jrandrews/jsc/polyglot/App.java, you can see that we instantiate an H2 JDBC driver using Class.forName(...), then use it to perform some basic database operations. Inspecting the build.gradle file, you can see how we have configured the GraalVM plugin to compile the code.

Run the app:

./gradlew run

Note that a database file is created in the logged in user's home directory. Now build and run a native image:

./gradlew nativeRun

The build succeeds, but the resulting application throws a null pointer exception, because the JDBC client code is not included in the executable.

Navigate to (GraalVM Reachability Matrix)[https://github.com/oracle/graalvm-reachability-metadata]; drill in to H2 and note the version supported.

Update build.gradle; set metadataRepository = true

Now build and run again:

./gradlew nativeRun

...to show that the JDBC driver works this time.

Polyglot

GraalVM can run multiple languages at once.

First, install runtimes:

gu available        # to see what's available
gu install python
gu install r

Now change the included project from “app” to “polyglot”:

cp settings.polyglot settings.gradle

Inspect the code in the polyglot project. App.java loads a Python script and an R script into an execution Context, then executes the Python script. That script finds the R script in the context, and executes that. The results are printed to the screen.

Now build and run using the JIT compiler:

./gradlew run

Isn't that cool?

Then build the native image and run it. The R native image build is experimental so for now, comment out the R code by editing the App.java source code, and commenting out the line that says "app.rFromPythonGraph();". Then:

./gradlew clean run --args="-agentlib:native-image-agent” # to get a set of configuration files. ‘find . –name ““.json”’ to see the files.
./gradlew nativeRun