Loss32: a proposal for a first‑class Win32 personality on Linux

Loss32 sketches a clear, opinionated goal: make Win32 a first-class runtime on a Linux kernel, not just a compatibility layer you coax into working. What’s notable here isn’t inventing from scratch-Wine/Proton, DXVK, and friends already do the heavy lifting-but packaging those pieces as a coherent ABI with strict boundaries and conformance. Under the hood, that means a PE loader wired via binfmt_misc, an ntdll/kernel32 surface with disciplined syscall thunking, and CI gated on Windows API tests rather than app-specific hacks. Wrap it in containers, standardize GPU/audio/input paths (Vulkan via DXVK/VKD3D, PipeWire, Wayland/evdev), and you get a predictable target that vendors and ops can treat like an OS personality, not a science project.

The bigger picture is operational: a Win32/Linux lets teams host legacy Windows apps on Linux fleets without VMs, with tighter sandboxing (namespaces, seccomp) and saner packaging (OCI images). Worth noting: semantics matter more than benchmarks-case-insensitive FS (FUSE overlay), file-locking, APCs/IOCP, timer granularity, console and service control need first-class treatment to avoid per-app shims. The ceiling is also clear: kernel drivers, anti-cheat, DRM, and deep OS hooks remain out of scope; UWP/WinRT isn’t the target. If Loss32 sticks to a well-defined ABI and continuous conformance, it could turn today’s “works on my Wine prefix” into a reproducible, supportable runtime-less hype, more contract.