To be fair, this was always a problem, also with old computers and OSes. There was no link between pressing eject and magical things happening in the machine to commit data. The data was simply already committed so there was no need to have any additional layers of security.

The difference between then and now is that new computers have enormous amounts of memory and use it to provide write-back caching. This means the OS can tell applications that data is committed when it's not. This is an OS level decision which is done for performance and has nothing to do with the physical eject-buttons on your devices.

You may call this a small nit, but I think it's an important one. The eject-button on old machines would suffer the exact same problems if it used an OS which offered write-back cache.

You could say old Macs didn't have this problem, which is true. But it didn't have an eject-button. It relied on the OS doing that. And then instead you had the problem of people not being able to get their disks out of their machines if the OS for some reason refused to release it (or that users didn't realize they had to drag the disk-icon to trash).

My first time on a Mac, about 1990, I felt pretty confident having plenty of early PC experience. A graphic designer I shared a house with had left me alone with (and permission to use) her Mac.

I quickly discerned there was no software of interest actually installed, but found a floppy that looked something like a game. It wasn't, and I went to hit the ... wait, there's no eject button?

After a fruitless few minutes during which I failed to notice the eject key on the keyboard (of all places!) I eventually opted for tweezers, and gently teased the floppy out of the drive.

Imagine my surprise when I restarted the Mac to find it was unable to start, having apparently completely forgotten about it's own HD! The thing had to go back for servicing before it would even boot properly, iirc.

My reputation as the house's resident computer expert was severely impacted, needless to say.

Still, seeing how many people in my office will just pull USB flash drives without ejecting them I almost see the appeal now.

* Drag the disc icon to the trash can to eject the disc.

* Pull the power cord since there were no other way they knew of to turn it off.

Mac's reputation as a usable computer was severely impacted, needless to say :).

I didn't quite like the way Linux handled this though, not allowing the disk to be ejected as long as any file descriptors were still open, which meant you had to hunt down the processes keeping them open.

But it would be extremely helpful for writable media like flash to be able to do full write-back caching and only flush the cache when the eject button is pressed, then EBADF any further reads or writes to any open file descriptors.

Also useful when something is hogging the audio device in /dev.

For those completely oblivious to this: This was actually not about getting data written to disk, but to physically protect the drive from its own mechanical construction. The "park" command would cause the harddrive to move its magnetic read and write-heads away from the internal platters, so that when it was powered down and the disks stopped spinning, the platters would not be damaged or experience needless tear and wear.

Yeah, I'm glad we no longer have to do that exercise.

(This used to be how you'd tell the tape-spooler to rewind..)

Always, always sync -r after writing to a zip drive on Windows. Even minutes after you think it is finished writing, do not trust the OS to have finished flushing to disk.

Also, not syncing to disk immediately can improve the lifetime of flash media, which wasn't a concern in the old days.

However, I'm somewhat disappointed by current filesystems not using journaling to make it possible to pull out a USB drive without harm (other than losing the latest writes).

As for Macs not having the problem, this doesn't require qualification. Not only did Macs not have the problem, they treated disks as logical _volumes_ rather than physical devices. If you ejected a disk mid-save on a Mac it would complain, and if you inserted the wrong disk it would reject it and ask for the right one. If you did the same thing on a DOS/Windows machine it would correct the new disk (and leave the old disk corrupted too).

Similarly, you could copy files from one disk to another without having two disk drives in Finder (it was a pain in the neck, swapping the disks, but it worked). Other computers needed special software and relied on the user to insert the correct disk.

This advantage continues to this day. Windows still refers to devices by physical location.

Nope, not since NT. I learned that the hard way when moving a HDD containing Win2k around. Even though it was hooked to IDE master it would come up as D: as that was what it was assigned after it got formated with NTFS.

You can go into disk manager and assign a partition any letter than is not already in use.

You can even mount a partition into a existing FHS, much like on unix, but i can't say i have seen it used often.

Basically the letters are there for backwards compatibility.

There was on the Apple Lisa. Or rather ejecting a disk was something you did in the OS which made sure everything was written/closed before allowing the disk out.

It was definitely confusing to new users: Insert a disk was something which anyone could do, since it only involved simple movement of a physical object, but ejecting it required you to know and understand the OS.

This was quite an "impedance mismatch" between what should ideally be two symmetrical actions, and I remember I had to help out my mother quite a few times :)

Fortunately the paperclip trick was in there.

The media itself is fully passive, while a SD card or thumbdrive have logic sitting between the USB interface and the raw flash. Thus optical media can be "rescued" by simply getting a new drive.

And you really need to mistreat a optical disc to make it completely unreadable.

Or perhaps someone could buy the minidisc tech off Sony and resurrect magnet-optical storage?

https://en.wikipedia.org/wiki/Mount_Rainier_%28packet_writin...

Windows used to require you to 'eject' but doesn't any more, which I think is due to the above. Please correct me if I'm wrong.

No, it's always required, because applications may be writing to the drive.

I recall some distros tried to mount USB drives with the sync option back in the early days of USB media.

Problem was that this, in combo with FAT, would basically kill cheap drives lacking wear leveling.

This because every write action was accompanies with a allocation table update. And thus whatever flash cells housed that table would see a massive amount of writes.

Actually i think the mount man file still have a warning about this in the sync option entry.

I am so paranoid now about removing USB drives after I accidentally removed one while it was still writing to the drive (which I had no idea because there was no LED on the drive) and it bricked the entire device.

I generally wait until my computer has shutdown completely now before removing any important drives, just to be sure.

I don't even bother to unmount it; what would unmounting add, if the OS can recognize that the device is gone and clean up its mount tables? Is there some filesystem that waits until unmount to recalculate indexes or something?

(Okay, sure, the OS could tell me the disk is in use because I've got an overlay filesystem mounted on top of it, and would I kindly wait forever for that to stop being the case. But screw you, OS, I need this USB. Just send the FUSE server or SMB daemon or whatever a SIGPIPE.)

Also, unmounting helps you to be confident that nothing's started doing more IO since you ran sync.

#!/bin/bash for i in `ls /media/$USER` do cd /media/$USER umount $i done

ls /media/$USER |grep -v cdrom0 |grep -v cdrom> /tmp/disk_status file=/tmp/disk_status if [ ! -s $file ]; then kdialog --passivepopup "Nothing Mounted " 1 else kdialog --passivepopup "Mounted devices: `cat /tmp/disk_status`" 2 fi