Not my story to tell, so I'm relaying it. Childhood friend works for a big company, you've heard their name, they make nuclear control systems for nuclear reactors; they have products out in the field they support and there are new reactors in parts of the world from time to time. We were scheduled to have lunch a couple years back and he bailed, we rescheduled, he bailed because that was the day you couldn't defer XP updates anymore, they came in and some XP systems became Windows 10. XP was "nuclear reactor approved" by someone, they had a tool chain that didn't work right on other versions of windows, it all gave me chills.
They ended up giving MS a substantial amount of money to extend support for their use case for some number of years. I can't remember the number he told me but it was extremely large.
It sounds like he said XP machines auto-updated to Windows 10, and they would have had to have been connected to the internet in order to download that update. (I'm assuming, optimistically, that these were more remote-control computers than actual nuclear devices.)
Eh. There are a great many problems that could befall a medical emergency systems that are unrelated to OS. Like power loss. I think the core problem here really is a lack of redundancy.
Just a few weeks ago I had an OpenBSD box render itself completely unbootable after nothing more than a routine clean shutdown. Turns out their paranoid-idiotic "we re-link the kernel on every boot" coupled with their house-of-cards file system corrupted the kernel, then overwrote the backup copy when I booted from emergency media - which doesn't create device nodes by default so can't even mount the internal disks without more cryptic commands.
Counter anecdote: I’ve been using Linux for 20 years, nearly half of that professionally. The only time I’ve broken a Linux box where it wasn’t functional was mixing Debian unstable with stable, and I was still able to fix it.
I’ve had hardware stop working because I updated the kernel without checking if it removed support, but a. that’s easily reversible b. Linux kept working fine, as expected.
I’ll also point out, as I’m sure you know, that the BSDs are not Linux.
Funny, i broke my Debian twice (on two separate laptops) by doing exactly that, mixing stable with testing. I was kinda obliged to use "testing" because Dell XPS would miss critical drivers.
In fairness, this is the number one way listed [0] on how to break Debian. That said, if you need testing (which isn’t that uncommon for personal use; Debian is slow to roll out changes, favoring stability), then running pure Sid is actually a viable option. It’s quite stable, despite its name.
I hope organisations start revisiting some of these insane decisions.