# Bulk File Rename

A function that funnels all filenames into a file and opens that file in vim. The user then changes the names, saves, and quits. Finally the function renames the files in the folder with the new names provided in the file.

bulkrename() {
mkdir -p ~/.temp
ls > ~/.temp/newfilenames && nvim ~/.temp/newfilenames
local files=(*)
for filenum in $(seq 1$(ls | wc -l)); do
mv ${files[${filenum}]} $(sed "${filenum}q;d" ~/.temp/newfilenames)
done
rm ~/.temp/newfilenames
}


### Always double-quote variables used in command arguments

The arguments of mv should have been double-quoted, like this:

mv "${files[${filenum}]}" "$(sed "${filenum}q;d" ~/.temp/newfilenames)"


### The script is fragile

As it is, the script looks very fragile:

• The list of files for editing is generated by ls, and then the edited list items are paired up with the files array generated with *. I'm not sure the ordering is guaranteed to be consistent, and I think it would be painful to track down in the documentation if this is indeed the case. It would be easier to generate both lists in a way to ensure consistent ordering.

• Using ls to generate the list is problematic. The output of ls > ... will depend on active aliases. command ls > ... would be safer.

• Files whose names don't change will raise errors when executing mv same same

• If there are duplicate lines after editing, one of the original files may silently disappear.

• The script may behave unexpectedly in certain corner cases:

• the user deleted a line from the file
• the user inserted a line in the file
• the list of files in the directory changed while editing the file
• The user may not have a way to abort the operation. With default shell settings, even if nvim exits with failure, the script goes ahead with the renames, which is probably not what a user would want.

• Even with the double-quoting fixed, the script will not work for files whose names contain newlines. I think that's acceptable and not worth the pain to make it work, but it would be good to document (in a comment).

To mitigate these issues, I suggest:

• Create an array from *, let's call it oldnames
• Save oldnames to the work file: printf '%s\n' "${oldnames[@]}" > "$work" 
• Let the user edit the work file
• Check the exit code, and abort on failure (user can cause failure by exiting nvim with :cq)
• Load the content of the work file into another array: mapfile -t newnames < "$work" • Add a sanity check to verify that the number of files match before and after. • Run mv only for the files whose names changed, and use -i to avoid overwriting existing files, and -v to show what was actually renamed. ### Use mktemp to create temporary files The script is not safe to use concurrently. It's easy enough to create a unique temporary file using mktemp. ### Use trap to clean up temporary files To make sure that temporary files get cleaned up when the script exits, use trap, for example: trap 'rm -f "$tmpfile"' EXIT


Declare this trap right before creating tmpfile.

### Declare all local variables as local

It's good you declared local files. There is filenum too.

### Don't use seq

The seq utility is not installed by default in all systems, and Bash has a native way to use counting loops:

for ((i = 0; i < size; i++)); do ...; done


### If you use Bash arrays, reap all the benefits

Instead of $(ls | wc -l) to find the count of files, you already have that in the files array: ${#files[@]}.

### Improve performance

Calling sed in a loop to get the n-th line of a file is inefficient. It would be better to read the lines into an array, and then use a counting loop to with the two arrays, for example:

for ((i = 0; i < ${#oldnames[@]}; i++)); do old=${oldnames[i]}
new=${newnames[i]} if [[ "$old" != "$new" ]]; then mv -vi "$old" "\$new"
fi
done

• I implemented all of your suggestions. Learned a lot from them. Thank you. New question. – Tony J Jul 27 '19 at 19:42