What if Linux ran Windows… and meant it? Meet Loss32
What if, rather than make a Linux distro that can run Windows apps, you built the whole distro around Windows binaries instead?
Loss32 is the most gleefully deranged idea for how to put together a Linux OS that we think we have ever read about in three and a half decades… but it's not impossible. Not only could it be done, there could be real advantages to doing it this way.
The idea comes from a blogger and developer known as Hikari no Yume ("Dream of Light" in Japanese) who made it public at the 39th Chaos Communication Congress in Germany at the end of December.
The gist of the idea is to run the whole user environment, desktop and all, inside WINE. So it's something like a bare-metal WINE sitting on top of the Linux kernel, with just enough plumbing to connect them up. This is significantly different from the current way, which is to run a completely Linux-based stack – the kernel, an init, a userland, a Linux display system, and a Linux desktop, and then run Windows programs inside that.
Nor is it just "a Linux that can run Windows apps." That's an old idea – it was the concept behind the Lindows distribution some 25 years ago, although the name got the company sued by Microsoft. Lindows became Linspire became Freespire, which is, unexpectedly, still around, and the included Click'n'Run Warehouse was pretty much the first app store on the web. There was also an effort to add direct support for Windows binaries to the Linux kernel, called Longene, over a decade ago.
It's also a profoundly different approach to emulating the entire Windows OS, as the ReactOS project is inching toward. It's been working on that for quite some time now: The Register first mentioned it in 2012, as far as we can see. It also reminds us of the Neptune OS project we covered in 2022, which is still in development.
It could be made to work. Even ReactOS itself considered a comparable approach. It's even possible to run WINE on Windows itself to restore compatibility with 16-bit Windows binaries, and there are efforts to make that easier such as BoxedWine.
Long before WINE itself was useful, Sun offered WABI – Oracle still has the manual [PDF]. Sun proposed making the Win16 API a formal standard. Later, Caldera offered a Linux version of WABI, and this vulture tried it. It worked remarkably well, and let us successfully install and run MS Office 4.3 under Linux with no VMs – or Windows licenses – involved. You can see some modern screenshots on VirtuallyFun.
Linux in 2026 is better at running Windows apps than it's ever been before, to the extent that there is mass-market consumer hardware sold for this, with an Arch-based distro whose selling point is its ability to run Windows games smoothly and well, and more such hardware is coming soon.
- The last supported version of HP-UX is no more
- You don't need Linux to run free and open source software
- Keeping Windows and macOS alive past their sell-by date
- pearOS is a Linux that falls rather close to the Apple tree
Much of this is down to the growing maturity of WINE, but it's not just WINE. Valve is sponsoring a lot of the work, including its Proton layer for running Windows games on Linux. You can check what will work and how well on ProtonDB. The forthcoming Steam Frame headset is Arm64-powered, but is designed to run x86-64 Windows games, thanks to FEX. A notable website aimed at this market, PCGamer.com, went on the record:
There are quite a lot of layers to unpack behind the ideas in the Loss32 proposal. The name is, of course, a pun on the name of the original Windows native API, Win32. The name "Loss," and the project's logo, also refer to the famed episode of that name from the long-running Ctrl+Alt+Del web comic. This installment itself became a meme.
One of perhaps the less obvious inspirations is a widely discussed blog post from 2022, titled "Win32 is the only stable ABI on Linux." This is a long-running joke in the Linux world, in the spirit of an ironic commentary on Linux compatibility over time. The kernel ABI itself is highly stable, and Linus Torvalds is notorious for defending this, but when you layer other components on top, it gets a lot more complicated. Even the lowest-level components: in the late 1990s, the transition from libc version 5 to libc 6, known as glibc, was a thorny issue and led to compatibility problems between distros that took about a decade and a half to subside.
Will Loss32 happen? It's too soon to say. Some people love the idea, some hate it, and some feel both, which we totally understand. But the bits are there. You could even boot such a Frankensteinian OS direct from NTFS – that's been possible for half a decade. Should it happen? That's a different question, but now that the challenge has been made, it may just be a matter of time. ®