Recursively remove files with the same name as the ones that end in `.part`

It is incorrect for two reasons.

1. File names containing glob characters

This is an edge case scenario.

Consider this structure:

.
├── abc
├── abc.part
├── cde
└── c*e.part

The outermost Find will find

• abc.part, so base=abc and the innermost Find looks for files matching the glob abc*, which matches the abc file. Good.
• c*e.part, so base=c*e and the innermost Find looks for files matching the glob c*e*, which matches the cde file. Bad, because cde does not contain c*e.

2. File names with extra characters

If you have abcde and abc.part files, the former will be deleted because it matches abc* as should be clear from the previous case discussion.

This particular problem would be easily fixed by changing $base* -> $base.*.

Proposed solution

Point 1 is the real challenge: It is quite involved to feed the file names back again into another Find's -name argument and escape the meta-characters, which is always a mine field.

I propose instead to use a shell with support for **, the recursive glob, for example Bash or Ksh with globstar option set or Zsh.

#!/bin/bash
shopt -s globstar #Not needed in Zsh
for f in ./**/*.part; do
    rm ./**/"$(basename "$f" .part)".*
done

For a breakdown,

• In line 2, **/*.part matches ./a.part but also ./a/b/c.part (hence "recursive glob").
• In line 3, "$(basename "$f" .part)" removes all directory components of the file name and its .part extension. This would boil down to a and c in our example.
  So the full line rm ./**/"$(basename "$f" .part)".* recursively removes files matching the a.* and c.* patterns.

It is crucial not to quote the * characters in the example, because we want it to act as a glob (and not to be parsed literally).

posted about 2 years ago

CC BY-SA 4.0

2y ago

Quasímodo‭

906 reputation 15 24 114 40

I might be inclined to try...

find . -type f -name '*.part' -exec sh -c '
  [ -f "${1%.part}" ] && rm -i -- "${1%.part}"; 
  for f in "${1%.part}".*; do 
    [ -f "$f" ] && rm -i -- "$f"; 
  done
' -- {} \;

(newlines for readability; can be elided if one-liner means something to you...)

1. find . -type f -name '*.part' — find files ending with .part
2. -exec sh -c '...' -- {} \; — run a shell script ... for each found file; path to file is in $1 in child script
3. "${1%.part}" — strip .part from the end of the filename in $1 (same as basename but without the extra process)
4. [ -f "${1%.part}" ] && ...; — if a file exists with no extension, do the ... bit
5. rm -i -- "${1%.part}" — delete the file with no extension
6. for f in "${1%.part}".*; do ... done — loop each found path matching the filename with any extension; path is stored in $f (this includes the one with the .part extension)
7. [ -f "$f" ] && ...; — if the path in $f exists and is a file, do the ... bit
8. rm -i -- "$f" — remove the file in $f

Note that I'm using various checks that the thing I'm asking to delete is a file, not a directory, link, fifo, etc.

If limiting only to files is less of a concern, you might well be able to shorten this to...

find . -name '*.part' -exec sh -c 'rm -i -- "${1%.part}" "${1%.part}".*' -- {} \;

The shell may write errors if the args to rm don't expand to existing paths, hide that with judicious use of 2>/dev/null redirection, if you care.

For fewer subshells, you may be able to pass all found files to the same shell in one go, with...

find . -name '*.part' -exec sh -c 'while [ -n "$1" ]; do rm -i -- "${1%.part}" "${1%.part}".*; shift; done' -- {} \+

...but this might be painful for larger file lists.

In general, note there's is technically a race condition between the various tests and the eventual delete, but that's only a concern if multiple processes are acting on that directory tree. Not sure how to avoid that.

Finally, rm -i is used to prompt y/n for each file to delete, as a safety net. Remove the -i switch from the rm calls if you are confident.

posted over 1 year ago

CC BY-SA 4.0

1y ago

jimbobmcgee‭

31 reputation 0 3 3 5

For each file named foo.xyz, you want to delete foo.xyz.part. It doesn't matter if foo.xyz.part exists, you can just attempt it and skip errors.

You can get a list of all files with find etc. But you don't want the ones with .part, so you use grep to take them out: find | grep -v '\.part$'. $ means end of string and \. is because otherwise . means any character in regex.

You can then attempt to delete each one: find | grep -v '\.part$' | parallel rm {}

If the file doesn't exist, Parallel will show you the error message, but it will still delete the ones that do exist. You can do a bunch of extra filtering with comm to only attempt to delete those files which do exist, but there's no need in this case.

posted over 1 year ago

CC BY-SA 4.0

1y ago

matthewsnyder‭

1846 reputation 124 77 291 110