> want your end users to do the work for you

They don't: they skipped release preview step this time.

That's even worse, because they rolled this to non-RP users straight away. Though I must admit I never saw it in the updates list.

It's worse than that: my files were deleted as Windows Update downloaded the files before it even asked me if I wanted to install the update. I never even had a chance.

Current branch is the release preview.

They just want everyone to start at square one.

You can apply this statement to every software bug, probably ever.

No, I only apply it to companies that have dropped all QA and are having users test their software. (In this case, Microsoft and the Insider programme.)

Your statement is not true. Microsoft hasn't "dropped all QA".

Somehow I don't see an `apt-get update` deleting user files. User config files, maybe. But not documents.

Depends on how competent the package maintainers are/remain. You might just be one script bug away from wiping random files (https://www.slashgear.com/steam-on-linux-bug-can-delete-all-...)

Yike. I had forgotten about that one.

The need for safe, robust updates and dependency management is a pressing issue of our time, and it's a tough problem to solve. While the Debian systems I use have proved to be impressively reliable, in particular never causing any sort of failure with their unattended upgrades in many years, that is unfortunately the exception rather than the rule.

Windows updates in previous versions have caused more downtime for my businesses than any other sort of technical failure. Sometimes that has included data loss.

Package managers for various programming languages are another recurring source of danger. They often have debatable architecture with limited security and sandboxing functionality, sometimes allowing any rogue package's installation script to damage the system or cause data loss.

Apple's style with iOS updates seems to favour bricking old (or not so old) devices or crippling reductions in performance over data loss, but they still have a pretty bad track record, and they still rely on a ratchet-style update system where you can't easily go back if something does break.

I still find it baffling that in 2018 we retain so much emphasis on things like complicated filesystems, applications built on top of shared libraries and dynamic linking, and complicated plug-in based architectures. There are remarkably few good arguments for these fragile, risky, over-complicated strategies on modern systems. It's just momentum, but the collective cost to society from all the things that break as a result (and worse, break but can't then easily be fixed again) must be staggering.

Fedora Silverblue (previously Fedora Atomic Workstation) provides some interesting solutions[0] for the issue you mention in the first paragraph.

[0] https://docs.fedoraproject.org/en-US/fedora-silverblue/#_wha...