We are, in fact, compiling on native hardware (AWS Graviton) for linux-aarch64 (where we just completed ros-galactic-desktop-full, by the way!).
For the Apple M1 it’s more challenging as there is no Azure Agent ([Question] Azure Pipeline Agent support for Apple M1 · Issue #3184 · microsoft/azure-pipelines-agent · GitHub) and we don’t have any sponsoring for RoboStack (if we find someone
to pay for a M1 at Hetzner, which would run at around 50 Euro a month, that would be awesome).
I could also compile everything locally, but in general it’s nice to build these things on a “public” infrastructure. And cross-compilation has worked incredibly well for the entire conda-forge stack on Apple M1, which includes things like BLAS, numpy, scipy, cartographer, ompl, … and many other “hard-to-build” libraries.
In fact, I believe that making cross-compilation easy is an awesome thing.