[PaulHoule]

Talk to a CS professor and they will tell you an OS is what is deep inside the machine but not see. Read a review in arstechnica and you would think an OS is about appearances and the exact shape of each UI component, etc.

My understanding is that the kernel is basically the same already, it is the services stacked on top that are different.

[srvmshr]

iOS always ran a stripped down XNU/Darwin based on the original CMU Mach blueprint. Right from the first version.

For a fact, Steve Jobs mentioned that in his iPhone introduction keynote in 2007

[PaulHoule]

It's completely practical to ship the (almost) same kernel with different a different userspace. That's the difference between desktop Linux and Android, XBOX and Windows, etc.

In fact if you care about performance and reliability (particularly add more cores and get more performance) there is no option rather than build on a mature kernel. Google's Fuchsia, for instance, is a high risk project which might never reach parity with Android.

(I think of Microsoft's "dual track" OS strategy that took 5 years to merge Windows 95 and Windows NT)

[neilalexander]

> Google's Fuchsia, for instance, is a high risk project which might never reach parity with Android.

If you write an app using the Android APIs and not native code then it wouldn’t matter if you were really running a Linux-based Android or a Fuchsia-based Android, as long as the API contracts were met. Google knows this — it’s not impossible to imagine that one day Android might not be Linux-based.

[PaulHoule]

It's possible, but Google has to get all the details right to make it "good enough" if not better.

For instance they have to get the story right w.r.t. to drivers and the needs of handset vendors, carriers, etc. Performance has to be good. Microkernel and capability-based systems are notorious for having bottlenecks.

Personally I liked the original WSL from Microsoft a lot but Microsoft didn't feel it was good enough because filesystem metadata operations are much slower in Windows NT than they are in Linux. If you do something that involves an excessive number of files (say build the Linux kernel) some people find the performance of WSL unacceptable. Yet, nobody complained about the slowness of filesystem metadata operations before on Windows -- people figured that was just the way it was, might not even have known some operating system was faster, and they stuffed data in SQLLite and otherwise reduced the number of files that they handled.

Microsoft felt they had to do something about it and came out with WSL2 which is inferior to WSL and to just running a normal Linux kernel under Hyper-V, VirtualBox, VMWare and the like -- because it is still closely coupled to Windows in ways that make "it just works" elusive. (e.g. just requiring that you install it from the Windows Store means you can't install it if your Windows Store is b0rked, which mine often is)

So Fuschia might be the future, but getting there involves confronting a lot of details that they might not want to confront. If there are ten critical "non-functional requirements" and they only get nine of them they are doomed.