Native-Image cross-compilation from amd64 to arm64 fails on `gcc -v` check

I want to create native images for both amd64 and arm64, and when I attempt that on my x86 machine (Mac), I start the image ghcr.io/graalvm/native-image:21.3 with --platform=linux/arm64.

Out of the box, this fails during the toolchain-check, as running gcc -v from within the Java process fails.

Adding this resolved the issue for me:

ENV JAVA_TOOL_OPTIONS="-Djdk.lang.Process.launchMechanism=vfork"

Should that be added to the graalvm-ce / native-image Docker images?

Btw, adding gcc variants with different target architectures and the related libs to the Docker image would be awesome - then I could run native-image in my home-architecture while creating native images in the target architecture without having to run the whole thing through qemu (which increases the build time by a factor of 20 for me).

Start by reproducing the gcc -v toolchain check in ghcr.io/graalvm/native-image:21.3 with --platform=linux/arm64, then compare behavior with JAVA_TOOL_OPTIONS set to the reported launch mechanism. Review the Docker image definitions for graalvm-ce and native-image; done means the intended cross-compilation workflow completes reliably, with the scope of additional gcc target variants decided.