The Big Apple Silicon Switch

Wed Feb 15 12:01:19 EST 2023

Tags: docker java

Almost exactly two years ago now, I wrote a post describing using Apple Silicon for work while my iMac was in the shop. That was an interesting experience, but too thorny for me to want to really stick with. Some of that was due to the limitations of the hardware - it was a pretty low-end MacBook Air - but most of that was due to the very-much-in-progress world of Java tooling for ARM Macs.

Since then, I'd been itching to find a machine to replace the iMac Pro and the recently-released Mac mini finally fit the bill. It, now dubbed Tethys, arrived on Friday, and I've been going through the process of re-carving-out my working environment, this time starting clean.

Eclipse

Compared to my first dive into ARM Macs, things have gotten a lot smoother. The biggest one for me is the stability of Java generally and Eclipse specifically. Last time, I was on the bleeding edge of that, and I was fortunate enough to at least help diagnose and report some of the things barring the port from being complete. The experience was very janky but the sheer snappiness of Eclipse on ARM really stuck with me, and it's still there in spades.

And, really, there's not even a lot of special things to know. If you download the aarch64 build of Eclipse, it just works like you'd want it to. I had to make sure to download some x64 builds of Java for when I'm running legacy native code, but that wasn't really different from my current collection of different Java versions anyway.

This is all a huge distinction from before, where I was going to weird lengths to try to run Eclipse via X11 remotely just to get something responsive.

Notes and libnotes

For my needs, Notes works fine. It's a shame that it's not native, but, as long as I run my test suites with an x64 JVM, things work about as well as they do on x64.

One immediate problem I ran into, though, was that some test cases that push multithreading would hard crash with a native error to do with the x64-to-ARM translation and JIT. Running with an OpenJ9-based JVM, I found that specifying -Djava.compiler=NONE in the launch args avoided this trouble. I'm sure that the tests run a bit slower now, but they were already emulated anyway, so it's not noticeable.

Domino and Docker

One of the sticking points with my half-transition years ago was the need to run my Domino Docker containers in emulation. Though HCL will have to port Domino to ARM eventually, they haven't yet released such a thing, and so that situation remains the same.

And, unfortunately, the segfault I encounter during my use remains. I had hoped that changed in qemu would fix it, but it appears to be unchanged. Recent versions of Docker can also take advantage of Apple's support for Rosetta in Linux, but for my needs that just creates a different segfault earlier in the process.

So, on this front, I'm doing the same thing I started doing before: running Docker on a VM on actual x64 hardware and using the DOCKER_HOST environment variable to point to it. Fortunately, I've taken steps in the intervening years to make this more practical. The main thing to work around with such a setup is the inability to use filesystem binds, and so I've changed a lot of my Dockerfiles and Testcontainers setups to instead copy more into the image at build. I still haven't solved every problem there, in particular a huge in-container build I'd like to do that points at a giant pool of dependencies that would be onerous to put into the image, but I'm working on that.

Conclusion

Overall, I'm pleased as punch with this thing. It's zippy all around, I haven't once heard the fan even when really hitting the CPU and GPU, and the state of the software ecosystem is such that I only use a very few x64-compiled processes during my daily work. I'd expected to have more of a checklist to work down to clean them all up, but it's really just the for-now-required bits. Not bad at all.

New Comment