Difference between command `ls 2 >tmp >tmp` and command `ls > tmp`

Question

I don't understand what is the difference b/w ls 2>tmp >tmp and ls > tmp. They both seem to do essentially the same thing, creating a file tmp and storing the result of ls command.

Answer 1

A short answer: ls 2>tmp >tmp redirect both stdout and stderr to file tmp. while ls > tmp only redirect stdout to file tmp.

Try this to see the difference:

$ ls asdsadasd 2>tmp >tmp
$ cat tmp
ls: cannot access asdsadasd: No such file or directory

$ ls asdsadasd > tmp
ls: cannot access asdsadasd: No such file or directory
$ cat tmp
<Nothing happen here>

Answer 2

Note that you should not do:

ls 2> tmp > tmp

Above the shell opens tmp for writing on file descriptor 2, then it opens tmp for writing again on file descriptor 1 and then executes ls. You'll have two file descriptors pointing to two separate open file descriptions to the same file.

If ls writes on both its file descriptor 1 (stdout for the file list) and file descriptor 2 (stderr for the error messages), those outputs will overwrite each other (assuming tmp is a regular file and not a named pipe).

Actually, the stdout output will be buffered, so is more likely to be written at the end before ls exits, so it will overwrite the stderr output.

Example:

$ ls /x /etc/passwd 2> tmp > tmp
$ cat tmp
/etc/passwd
ccess /x: No such file or directory

You should use:

ls > tmp 2>&1

or

ls 2> tmp >&2

In that case, the shell opens tmp on fd 2 and then duplicates that fd onto fd 1, so both fd 1 and 2 will share the same open file description to tmp and outputs won't overwrite each other.

$ ls /x /etc/passwd 2> tmp >&2
$ cat tmp
ls: cannot access /x: No such file or directory
/etc/passwd