How to delete everything except top-level numeric directories?

Question

I want to delete everything under some directory /path/to/foo, EXCEPT those sub-directories that match the meta-pattern

/path/to/foo/<DIGITS>/

For example, if the contents under /path/to/foo are initially like this:

/path/to/foo
├── 0/
│   ├── a
│   └── b
├── 232532/
├── 42
├── 73/
│   ├── d
│   └── e
├── 8xyz/
│   ├── i
│   └── j
├── _bar/
│   ├── x
│   ├── y
│   └── z
├── .baz/
│   ├── f
│   └── frobozz/
│       ├── g
│       └── h
└── quux/
    └── 123/

...I want to end up with

/path/to/foo
├── 0/
│   ├── a
│   └── b
├── 232532/
└── 73/
    ├── d
    └── e

I'm looking for a find ... -delete-based incantation, or a suitable zsh glob pattern (for rm -r), that will do this. I am using Linux.

Answer 1

With zsh:

set -o extendedglob # best in ~/.zshrc
rm -rf /path/to/foo/^<->(D) /path/to/foo/<->(^-/)

• ^something is not something (similar to ksh's !(something))
• <-> is <x-y> to match decimal integers from x to y, but with none of the bounds provided (so matches any sequence of decimal digits, similar to ksh's +([0-9])).
• (D) a glob qualifier to include hidden files (Dot files)
• (^-/) a glob qualifier to say only files that are not of type directory after symlink resolution (remove the - if you also want to remove symlinks to directories).

Answer 2

If you have access to GNU find, you can run:

find . -maxdepth 1 -not \( -type d -regex '.*/[0-9]*$' \) -exec rm -r {} +

The idea is to find those top level files that aren't i) directories and ii) whose name doesn't consist of nothing but numbers.

I strongly urge you to run it once with echo instead of rm -r to see what it will do.

Answer 3

cd /path/to/foo && \
find . -maxdepth 1 ! -type d -delete -o \
   ! -name . -name '*[!0-9]*' -exec rm -rf {} +

Which reads as:

• change over to the /path/to/foo directory and once successfully there
• invoke find to go just one level deep and delete any non-directories there.
• for all others, i.e., subdirectories in your /path/to/foo dir, reject those with the names of ".", and what remains out of these, select those that have at least one nondigit in it.
• The selected dirs are then led to the abbatoir, thereby ensuring the survival of digit-only dirs.

Answer 4

Apart from zsh, you can't select non-directories using wildcards (you can only make a wildcard pattern apply to all file types, or apply to directories and symlinks to directories by putting a / at the end).

In any shell, you can loop over the files and exclude the ones you want to keep. To enumerate all files including dot files and . and .., use the two wildcard patterns * .*. It's possible that * will not match any file and so will remain unexpanded; in this case, rm -rf is executed on *, which does nothing since rm -f ignores non-existent arguments.

for x in * .*; do
  case "$x" in
    .|..: continue;;                              # skip the . and .. entries
    *[!0-9]*) rm -rf -- "$x";;                    # not just digits -> delete
    *) if ! [ -d "$x" ]; then rm -f -- "$x"; fi   # just digits -> delete if not a directory
  esac
done

Answer 5

find ./* ! \( -type d -regex "./[0-9]+\(/.*\)?" \) -delete

or, if foo have too many files and the argument limit excess happens, use:

find . ! -path . ! \( -type d -regex "./[0-9]+\(/.*\)?" \) -delete