What I meant to say was turning it off properly.
When we were first putting small computers in the office (long before the PC), we had problems with the mindset of the users, that, in my opinion, have only been partially solved today.
The first thing was the power switch. Users would have, say, a GL application open and enter some data and simply reach for the power switch after the last entry. After all, that's what you did when you were through using your desk calculator, typewriter or radio. Although we had our own OS, I know that this problem plagued CP/M applications also. Files open for writing present a problem--if you lose some data because of a power failure, how do you (a) make sure that the user knows that some data will be lost and (b) how does a transaction get backed out such that it's completely reversed and not just partially reversed?
The other problem was diskette changing. Users would simply pop diskettes in and out in advance of their being needed, often leaving files open for writing on the old set. Worst case was that the application would continue writing without realizing that the media had been changed and all hell would break loose. CP/M tried to solve this by periodically checksumming the first few sectors of the directory to make sure that the disk hadn't been changed. If it had, the status of the new disk was set to read-only. Their "cure" was almost worse than the problem--few programs knew how to gracefully back out of the situation. We simply sampled the write-protect sensor every quarter-second or so, and if we detected a disk being changed while it still had files open, beeped and flashed a message to put the disk back in--and didn't continue until it happened. Issues with corrupted files dropped amazingly.
MS-DOS was worse--because it keeps the directory and allocation information separate, you can easily get the two out of sync. Microsoft's answer was to give us CHKDISK. You can go over old problem reports with MS-DOS right through at least version 6.22 and read about DOS not seeing disk changes. Ultimately, the thing that saved DOS (and Windows) was the use of the "disk changed" status reporting on 1.2M/720K/1.44M floppies.
If you had a Mac, you simply couldn't eject a floppy manually (without using a straightened paperclip).
Perhaps with the dual-floating gate memory that I see being promised as the next big thing, we'll have systems that we can simply power off. I
hate soft power switches--I don't think they're safe.