On any modern system it's actually "sudo rm -rf / --no-preserve-root" and then entering your password while staring at the command.
"rm -rf ~/blue /" will not come close to deleting / unless you are in the habit of running every command as sudo, even ignoring the presence of --no-preserve-root
Much, much, much the worse is "rm -rf ~ /blue". I don't give a crap about 99% of the stuff outside of $HOME, but of course, the stuff in $HOME is the stuff that's trivial to destroy.
I think this would be a bit better for interactive cases. Note: written just now, I haven't actually felt the need for this safeguard... yet.
rmrf()
{
(echo "The following files are going to be deleted!!!"
for FILE in "$@"; do
echo "<<<" "$FILE" ">>>"
done) | less
read -p "Are you sure? " -n 1 -r
echo
if [[ $REPLY =~ ^[Yy]$ ]]
then
rm -rf "$@"
fi
}
What would be the "good" alternative ? I often tries "rmdir" or "rm -r" if the directory is not empty and very often there are some "protected files" so I add -f.
Thus it happens that I directly lauch "rm -rf".
Watch it fail first, verify that the failure makes sense, check to see if there's a way to delete one file with -f before deleting the rest with -r. Use -rf only as a last resort, and only by appending the f to an already-failed command whose syntax you've validated.
Hehe, I did something like that once, except that I typed "rm -rf ~ /blue/". There was no /blue/, but I managed to wipe out my home directory, and I did not have a backup. :-|
It was on my personal machine, so at least I did not delete anybody else's files, but I still got burned hard enough to learn a valuable lesson.
The / key was right next to Enter on a lot of old keyboards. It was quite easy to type 'rm -rf /tmp/garbage*' and have a simple fumble turn it into 'rm -rf /'. I mean, there's this guy I know, he did that once.
Ten years prior to that, Apple had a similar bug in one of its installer scripts on OS X. I have a hard time finding much about it online now, because it happened at a time when OS X and the Internet were a lot less popular than today, but what I recall is that an unexpectedly customized installation directory (say with spaces or one level closer to "/" than the default) would cause the installer to delete a whole lot of things.
The one that I did only a few months ago was something like
$ cp -r path/to/some/directory path/to/very/important/directory
$ (run some commands to verify copy did what I wanted)
$ rm -r path/to/some/directory path/to/very/important/directory
Of course, all I had meant to do was delete `path/to/some/directory`, but I just pressed 'up' in my history and switched `cp` to `rm`. Of course I hit Ctrl-C in an instant, but my FS was already hosed...
It's my habit to never use the -f flag until I get those annoying confirmation messages. I <CTL>-c to cancel that command, then scroll up and add the flag to run again. I think this is a good habit? Anyway, the worst thing I've done along these lines was resetting a dev DB that had seen considerable un-backed-up configuration work. I couldn't blame rm for that.
Eh, "rm -rf $TEMPDIR/$TEMPFILE" in a shell script is just a couple typos away from deleting everything on the network. Yet I've seen people put crap like that in build scripts even after they've previously inadvertently deleted half the network drives.
Fortunately, despite rm's poor choice of options and bash's poor default handling of variable name typos and the obvious PEBKAC, backups saved the day here.
