Use for loop in find exec

Question

 -> find  .. -name bin  -exec  for file in {}/* ; do echo $file ; done  \;

   -bash: syntax error near unexpected token `do'

What is the proper syntax for this command ?

Answer 1

To use multiple statements, such as a for-loop, as the argument to -exec, one needs to invoke a shell, such as bash, explicitly:

find .. -name bin -exec bash -c 'for file in "$1"/* ; do echo "$file" ; done' none {}  \;

This is safe even for filenames that contain spaces or other hostile characters.

How bash -c works

One can invoke bash with a command of the form:

bash -c some_complex_commands arg0 arg1 arg2 ...

In this case, bash will execute whatever is in the string some_complex_commands. Those commands can make use of the usual shell positional parameters. The first argument after command, arg0 above, is assigned to $0, the second to $1, the third to $2, etc.

When one executes a normal shell script, $0 is the name of the script and $1 is the first argument that appears on the command line. In keeping with that tradition, the bash -c command was written to assign the file name, {} in find's notation, to $1. Since this script does not have a sensible name, none is assigned as a placeholder to $0.

Answer 2

It looks like you've got things reversed from what you want. Try this:

for f in `find .. -name bin`
do
echo $f
done

Answer 3

You can approximate the output of your pseudo-code there with find primitives as is:

find .. -path \*/bin/\* ! -name .\* 

...which should print only files/dirs with names that do not begin with a . and which are rooted at some level both in the parent directory, and, at some greater degree, in a dir named /bin.

In general I can think of all kinds of practical purposes for looping in a find -exec child process, but not a one in which I could consider it practical to do shelld glob on an argument passed by find. To do the same thing your pseudo-code does with printf - because its use can guarantee literal translation of arguments to output - you might do...

find .. -type d -name bin -exec sh -c '
    printf %s\\n "$0/"*' {} \;

...which does the printing and the globbing without the for loop. One difference between this and your example command, though, is that without the -type d specification and type of result matching the name bin will be echod - and so you're highly likely to see a lot of bin/* being written to stdout. Of course, even w/ -type d, there's no guarantee that the * will resolve - an empty directory or one containing only .files will render no matches and so you might see it anyway.

Note also that the example pseudo code, because it uses the {} \; primitives might be a lot slower than some other ways. We'll try the printf thing again like:

find .. -type d -name bin -exec sh -c '
    for d do printf %s\\n "$d/"*; done' -- {} +

...which still risks the empty glob case but instead of -execing a shell per match, it rather gathers as many arguments as it might reasonably pass off to another process at exec time and passes the lot to sh in "$@" its positional parameter array - which we can loop over with for d do printf....

Now if you wanted to do something other than just print results - which is what I would typically consider useful about looping in an -exec statement - you can fallback to the earlier -path example and combine it with -exec like...

find .. -path \*/bin/\* -exec sh -c '
    for arg do : something with "$arg"
done' -- {} +