published on 2025-11-05

In recent years, Apple has been producing some of the best hardware in the laptop market. The Apple Silicon chips are among the most efficient and performant processors available, often using as little power as a smartphone while delivering desktop-class performance. This combination allows Apple to build laptops that are both powerful and capable of lasting an entire day on a single charge. When these new MacBooks were released, I was excited to see what they could do, so I bought a MacBook Air. The Air is passively cooled, completely silent, and rarely overheats except under extreme workloads. It’s powerful enough for smooth video editing in full resolution using optimized software like Final Cut Pro. I’ve been using it to experiment with neural networks, taking advantage of Apple’s new MLX library. Since M-series chips use unified memory shared between the CPU and GPU, there’s no need to copy buffers back and forth, which significantly reduces overhead. This makes it easy to tweak parameters or fix errors without waiting tens of minutes for each new training run—all while carrying a 1.3 kg laptop in my backpack.

The issue: macOS

But as always, there are compromises. Now that Apple fully controls its hardware, they no longer publish driver documentation, making it very difficult to port other operating systems. The Asahi Linux project is working hard to reverse-engineer the platform, but the experience is still far from what macOS provides. GPU acceleration, for example, is not yet supported—an enormous limitation considering how capable the Apple Silicon GPUs are. I’ve been a Linux user since I was 12, and over time my workflow became completely dependent on free software. Since I mostly use C, the POSIX environment feels like home. macOS isn’t as bad as Windows for C development since its userland comes from FreeBSD and the kernel implements POSIX quite well, but there are still problems. First, I prefer lightweight software, and that’s nearly impossible on macOS. The windowing system can’t be replaced, the UI is heavy, and since macOS 26 it’s also quite buggy. Second, the system APIs are unpleasant to use. Apple forces developers to rely on technologies designed for NeXTSTEP in 1988, completely object-oriented and tightly coupled to one of the worst languages I’ve ever used (in my opinion): Objective-C. Even basic tasks, like creating a window without the system UI toolkit, require defining new classes, managing layers, and setting up autorelease pools (because apparently developers can’t handle memory even with modern tools like sanitizers). Doing the same in another language, say, plain C, means using the Objective-C runtime API, which is a mess. Simple things like working with strings require conversions between NSString and C strings through dynamic runtime calls, wasting both time and resources. Unfortunately, this issue extends beyond Cocoa. Metal has similar problems, and while it’s technically usable from C, it’s tedious. That’s disappointing because Metal 4 introduces features that could make graphics software much faster and more efficient. The same applies to Core ML and other frameworks: they’re powerful but so cumbersome that many developers just avoid them. Soon Objective-C will likely be replaced by Swift, which will make things even worse for low-level developers. At least Objective-C can be compiled by most C compilers like GCC and Clang. Swift cannot. Mixing Swift and C requires bridging headers and multiple compilers, adding unnecessary complexity.

What I’d like to see

If Apple cared a bit more about power users (still a small but relevant group) Macs could become the perfect development machines. That could happen by either providing proper Linux support or making macOS more customizable, with APIs accessible directly from C. Ideally, both. Realistically, that’s unlikely. Apple’s hardware is far ahead, and companies with such an advantage tend to vendor-lock users, much like NVIDIA did with CUDA. I hope other vendors eventually catch up and produce laptops that are just as efficient and powerful, maybe using RISC-V or ARM, so I can go back to running Linux like I do on all my other machines.