↓Skip to main content

QEMU-iOS Rewrite Progress

·3 mins

This is going to be a quick update post. I’m in the process of attempting to revive QEMU-iOS, an iPod Touch 1G/2G emulator by Martijn de Vos; it is about 4 years old at this point and seems to be unmaintained1. I’ve wanted to work on a rewrite of QEMU-iOS for about 4 years at this point but it has been on the back burner. Last year I tried several hours to build the original project and run it locally, but I was getting segfaults (in retrospect, maybe I could have traced it better). A couple months ago, I threw Claude at it and it was able to write a small patch and get the project to boot again. I’ve been itching to work on a large emulation project so it was either this or contributing to Inferno. However, since QEMU-iOS is unmaintained, I thought it would be more worthwhile focusing on this project and as a bonus, early iPods are simpler than modern iPhones. QEMU is a mature open source emulation project and TCG is a fast JIT; however, to have more flexibility when writing this emulator and to better familiarize myself with the system overall, I decided to forgo QEMU23.

Why not release? #

I’m getting close to booting Springboot, but even when I reach this state, I don’t plan on releasing the source for now. While my opinion on this could evolve, for important projects I work on, I always want to have some contextual understanding of most of the core emulation code I write and be able to rearchitect it as I feel is appropriate. I’ve been using Claude to generate a test suite against the original QEMU-iOS project, which it can then use to assist itself when writing the current emulator; this is, in my opinion, a great way to use LLMs since it removes a lot of busywork. However, once the author is divorced from their code’s architecture, I suspect it will be challenging to vet the quality of code contributions or even evaluate their own code. This doesn’t matter as much for non-core emulation code (such as UI code) since that code is going to be more boilerplate anyway4.

Sidenote #

I got motivated back into working on my Game Boy emulator (it never ends)! I did a simple refactor recently, but I hope to build out its debugger and maybe even the APU if I get the time (but I’m fairly busy nowadays, so it’s hard to say).

  1. There was a blog post written here, a 37C3 talk here, and a FOSDEM talk here. ↩︎

  2. AFAIK the only prominent game emulator using QEMU is xemu ↩︎

  3. Using Claude, I initially attempted to gradually migrate off QEMU using Unicorn but I was getting memory access issues so I ended up just generating an ARMv6 interpreter instead. ↩︎

  4. Somewhat unrelated but here are AI contribution guidelines from RPCS3 and Dolphin ↩︎