Find Command – Understanding find(1)’s -exec Option with Curly Braces and Plus Sign

The syntax is:

find ... -exec cmd {} +

find will find a number of files based on the criteria in ... and run cmd with that list of file paths as arguments, as many as possible without surpassing the limit on the size of the arguments to a command.

If needed it may split the list of files and call cmd several times. For instance, it may end up calling:

cmd ./file1 ./file2 ... ./file3000
cmd ./file3001 ./file3002 ... ./file4321

A limitation with that is that {} has to be last. You can't for instance write:

find ... -exec cmd {} other args +

like you could with ';' instead of '+'.

You can write:

find ... -exec echo foo {} +

but not:

find ... -exec echo {} foo +

So, if you do need to add some extra arguments to cmd after the list of files, you have to resort to calling a shell. (Other reasons why you'd need to call a shell would be any time you need to use a shell feature like redirections, pipes, some string expansions....)

In sh -c 'inline-script' x a b c, for the inline-script, $0 is x, $1 is a, $2 is b... so "$@" is the list of those 3 arguments: a, b and c. So in:

find ... -exec sh -c 'cmd "$@" other arg' find-sh {} +

For the inline script, $0 (which is used for instance when displaying error messages) is set to find-sh and "$@" is the list of files (what find expands {} to).

By using the exec special builtin of the shell:

find ... -exec sh -c 'exec cmd "$@" other arg' find-sh {} +

We tell the shell not to fork an extra process to run cmd, but instead to run it in the same process (replacing the running shell process with that command). Some shells like bash, zsh and some implementations of ksh do that implicitly for the last command in a script.

In this case, you can work around find's exec grammar by capturing a brace expression and using a back reference in the replacement text:

$ cat f1 f2
f1: hello{}a
f2: hello{}a
$ find . -type f -exec sed -i 's/hello\([{][}]\)a/hello\1b/g' '{}' +
$ cat f1 f2
f1: hello{}b
f2: hello{}b

Or, more simply (as noted in the comments):

find "$dir" -type f -exec sed -i 's/\(hello[{]}\)a/\1b/g' {} +

Note that the -i option for Sed is not portable and will not work everywhere. The given command will work on GNU Sed only.

For details, see:

• How to achieve portability with sed -i (in-place editing)?