fix(java): References within InvocationHandler, Issue #1364

[cn-at-osmit]

This is the fix for the issue #1364 including a test case ;)

[chaokunyang]

The try catch block is still needed, graalvm only capture this pattern and replace the offset. This invocation can be placed in catch block

[cn-at-osmit]

Should I leave the InvocationHandler stub initialization as is or move it to the static block as well? I'm new to grallvm but it's on my todo list 👍

[chaokunyang]

Leave it as it looks good to me

[cn-at-osmit]

done 👍

[cn-at-osmit]

Okay, I just stumbled over the try...catch mention. Is a try / catch really necessary or is the static block okay? 🤔

[cn-at-osmit]

Do you have something like this in mind?

static {
    try {
      // Make offset compatible with graalvm native image.
      PROXY_HANDLER_FIELD_OFFSET =
          Platform.objectFieldOffset(Proxy.class.getDeclaredField("h"));
    } catch (NoSuchFieldException e) {
      PROXY_HANDLER_FIELD_OFFSET = Platform.objectFieldOffset(Proxy.class, InvocationHandler.class);
    }
}

But I think i have to work with a temp assignment...

[cn-at-osmit]

This compiles but I'm not sure if it works for graalvm?

  static {
    long proxyHanlderFieldOffset;
    try {
      // Make offset compatible with graalvm native image.
      proxyHanlderFieldOffset = Platform.objectFieldOffset(Proxy.class.getDeclaredField("h"));
    } catch (NoSuchFieldException e) {
      proxyHanlderFieldOffset = Platform.objectFieldOffset(Proxy.class, InvocationHandler.class);
    }
    PROXY_HANDLER_FIELD_OFFSET = proxyHanlderFieldOffset;
  }

[cn-at-osmit]

Okay, I studied the redhat documentation in the link provided. So it seems for graalvm we assume that the name of the field is alway's 'h' and for other JVM's it could differ.

I updated the PR with this variant:

  static {
    if (GraalvmSupport.isGraalBuildtime()) {
      try {
        // Make offset compatible with graalvm native image.
        PROXY_HANDLER_FIELD_OFFSET = Platform.objectFieldOffset(Proxy.class.getDeclaredField("h"));
      } catch (NoSuchFieldException e) {
        throw new RuntimeException(e);
      }
    } else {
      // not all JVM implementations use 'h' as internal InvocationHandler name
      PROXY_HANDLER_FIELD_OFFSET =
          ReflectionUtils.getFieldOffset(Proxy.class, InvocationHandler.class);
    }
  }

My intention is to use the graalvm 'automatic fix' pattern when it's executed during graalvm build time and the other variant when executed during normal JVM execution.
Could this work?

[chaokunyang]

I'm not sure, may need a test. Could we add a test in integration_tests/graalvm_tests?

FYI, I tested following pattern, it works:

  private static final Field field;
  private static final long offset;

  static {
    field = ReflectionUtils.getField(Example.class, "f2");
    offset = Platform.objectFieldOffset(field);
  }

[chaokunyang]

Proxy in graalvm is a little complex, may need to add a json config file: https://www.graalvm.org/latest/reference-manual/native-image/guides/configure-dynamic-proxies/

If it's not easy to add, we can leave it in another PR

[chaokunyang]

Hi @cn-at-osmit , thanks for contribute this fix, I'll merge it first and add graalvm tests in a following-up PR

[chaokunyang]

LGTM