Support subclass mocks on Graal VM.

[raphw]

Adds support for running Mockito on Graal native image using the (still experimental) experimental-class-define-support. Byte Buddy now picks up the Graal image code and creates more predicatable class files such that hashes during the JIT runtime and the image runtime match.

This leads to some minor restrictions, but generally this should not be a big issue. With future improvements in Graal, these restrictions might fade in the future.

As an example, with the mockto.jar that is produced in this PR:

#!/bin/bash
set -x
set -e
if [[ -z $1 ]]; then
  echo "Specify location of Graal VM as first argument"
  exit 1
fi
$1/bin/gu install native-image
rm -rf META-INF sample.build_artifacts.txt Sample.class Sample.java
echo "public class Sample {
  public static void main(String[] args) {
    Sample sample = org.mockito.Mockito.mock(Sample.class);
    org.mockito.Mockito.when(sample.m()).thenReturn(\"Hello Graal!\");
    System.out.println(sample.getClass());
    System.out.println(sample.m());
  }
  public String m() { return null; }
}" > Sample.java
wget -O bytebuddy.jar https://search.maven.org/remotecontent?filepath=net/bytebuddy/byte-buddy/1.12.9/byte-buddy-1.12.9.jar
#wget -O mockito.jar https://search.maven.org/remotecontent?filepath=org/mockito/mockito-core/XXX/mockito-core-XXXX.jar
wget -O objenesis.jar https://search.maven.org/remotecontent?filepath=org/objenesis/objenesis/3.2/objenesis-3.2.jar
export dependencies=bytebuddy.jar:mockito.jar:objenesis.jar:.
$1/bin/javac -cp $dependencies Sample.java
$1/bin/java -agentlib:native-image-agent=config-output-dir=META-INF/native-image,experimental-class-define-support -Dorg.graalvm.nativeimage.imagecode=agent -cp $dependencies Sample
$1/bin/native-image --no-fallback -cp $dependencies \
--rerun-class-initialization-at-runtime=\
net.bytebuddy.ClassFileVersion,\
net.bytebuddy.utility.dispatcher.JavaDispatcher,\
net.bytebuddy.utility.Invoker\$Dispatcher \
--initialize-at-build-time=\
net.bytebuddy.description.method.MethodDescription\$InDefinedShape\$AbstractBase\$ForLoadedExecutable,\
net.bytebuddy.description.type.TypeDescription\$AbstractBase,\
net.bytebuddy.description.type.TypeDescription\$ForLoadedType,\
net.bytebuddy.description.method.MethodDescription\$ForLoadedMethod,\
net.bytebuddy.implementation.bind.annotation.Argument\$BindingMechanic,\
net.bytebuddy.implementation.bind.annotation.Argument\$BindingMechanic\$1,\
net.bytebuddy.implementation.bind.annotation.Argument\$BindingMechanic\$2,\
net.bytebuddy.implementation.bind.annotation.Super\$Instantiation\$2 \
Sample
./sample

The setup is still a bit cluncky but Oracle is working on improving the process. There will likely be plugins for the tools in question to improve this.

[codecov-commenter]

Codecov Report

Merging #2613 (c0d7443) into main (8314824) will decrease coverage by 0.34%.
The diff coverage is 34.78%.

❗ Current head c0d7443 differs from pull request most recent head d23dc0e. Consider uploading reports for the commit d23dc0e to get more accurate results

@@             Coverage Diff              @@
##               main    #2613      +/-   ##
============================================
- Coverage     86.62%   86.27%   -0.35%     
+ Complexity     2762     2761       -1     
============================================
  Files           314      314              
  Lines          8237     8268      +31     
  Branches       1017     1026       +9     
============================================
- Hits           7135     7133       -2     
- Misses          842      869      +27     
- Partials        260      266       +6     

Impacted Files	Coverage Δ
.../creation/bytebuddy/SubclassBytecodeGenerator.java	74.25% <28.57%> (-10.26%)	⬇️
...al/creation/bytebuddy/SubclassInjectionLoader.java	54.71% <28.57%> (-2.73%)	⬇️
...ito/internal/creation/bytebuddy/ModuleHandler.java	21.42% <33.33%> (-1.40%)	⬇️
.../main/java/org/mockito/internal/util/Platform.java	84.61% <33.33%> (-4.58%)	⬇️
.../internal/creation/bytebuddy/MockMethodAdvice.java	80.40% <100.00%> (+0.11%)	⬆️
...al/creation/bytebuddy/InlineBytecodeGenerator.java	84.48% <0.00%> (-0.14%)	⬇️
...internal/creation/bytebuddy/BytecodeGenerator.java	100.00% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8314824...d23dc0e. Read the comment docs.

[TimvdLippe]

Haven't had the chance to look at the implementation yet, but would it be possible to add a mockito-graal test project to our regression suite? That would help me with reviewing the implementation as well. Thanks!

[raphw]

I looked at it, but this would require running on Graal and installing the native extension. I am not sure how realistic this is and I would want to avoid downloading a GB for each build. It also requires two execution circles.

For those reasons, I have decided against it in Byte Buddy and rather rely on manual tests so long as the general setup is still autotested. I was hoping that the Graal ecosystem spawns some more mature testing utilities in the future.

[TimvdLippe]

Nice! Only a couple of nits and clarifying questions. Other than that LGTM!

[TimvdLippe]

Nit: can we maybe use a Set here, so that we can skip the contains check for Serializable? E.g. on Graal we always add Serializable to the Set and then we create the CompoundList of that.

[raphw]

I need to replace it with a sorted set, actually as a different order would cause a different class file and hash and fail the Graal run.

[TimvdLippe]

Is there a specific reason we only want this on Graal? I can imagine we would want this for others as well, to keep everything consistent?

[raphw]

The very theoretical situation would be a OSGi setup where multiple modules of Mockito wanted to create a mock for the same class loader. It's unlikely but might trigger a regression. The question is if the stable name justifies this theoretical issue.

[TimvdLippe]

Nit: I think we no longer need this, as this particular class isn't included in this PR?

[raphw]

Indeed, this stems from an intermediate version to work around something that is now fixed in Graal.