Tjyylli

I'm trying to figure out what words does the -interactive shell of cp accepts as input.

For your convenience, here's code that sets up files for experimentation.

touch example_file{1..3}

mkdir example_dir

cp example_file? example_dir

cp -i example_file? example_dir

The shell then asks interactively for each file whether it should be overwritten. It seems to accept all sorts of random input.

cp: overwrite 'example_dir/example_file1'? q

cp: overwrite 'example_dir/example_file2'? w

cp: overwrite 'example_dir/example_file3'? e

I tried looking into the source code of cp, but I don't know C and searching for overwrite is of no help.

As far as I can tell it accepts some words as confirmation for overwriting, and everything else is taken as a no. The problem is even words like ys seem to be accepted as yes, so I don't know what works and what doesn't.

I'd like to know how exactly does this work and to have some proof of it by means of documentation or intelligible snippets of source code.

The POSIX standard only specifies that the response need to be "affirmative" for the copying to be carried out when -i is in effect.

For GNU cp,
the actual input at that point is handled by a function called yesno(). This function is defined in the lib/yesno.c file in the gnulib source distribution, and looks like this:

bool

yesno (void)

{

  bool yes;



#if ENABLE_NLS

  char *response = NULL;

  size_t response_size = 0;

  ssize_t response_len = getline (&response, &response_size, stdin);



  if (response_len <= 0)

    yes = false;

  else

    {

      /* Remove EOL if present as that's not part of the matched response,

         and not matched by $ for example.  */

      if (response[response_len - 1] == 'n')

        response[response_len - 1] = '';

      yes = (0 < rpmatch (response));

    }



  free (response);

#else

  /* Test against "^[yY]", hardcoded to avoid requiring getline,

     regex, and rpmatch.  */

  int c = getchar ();

  yes = (c == 'y' || c == 'Y');

  while (c != 'n' && c != EOF)

    c = getchar ();

#endif



  return yes;

}

If NLS ("National Language Support") is not used, you can see that the only reply that the function returns true for is a response that starts with an upper or lower case Y character. Any additional or other input is discarded.

If NLS is used, the rpmatch() function is called to determine whether the response was affirmative or not. The purpose of the rpmatch() NLS library function is to determine whether a given string is affirmative or not (with support for internationalisation).

On BSD systems, the corresponding function is found in src/bin/cp/utils.c:

/*

 * If the file exists and we're interactive, verify with the user.

 */

int

copy_overwrite(void)

{

        int ch, checkch;



        if (iflag) {

                (void)fprintf(stderr, "overwrite %s? ", to.p_path);

                checkch = ch = getchar();

                while (ch != 'n' && ch != EOF)

                        ch = getchar();

                if (checkch != 'y' && checkch != 'Y')

                        return (0);

        }

        return 1;

}

This is essentially the same as the non-NLS code path in the GNU code.

If you look at the source code and search for interactive you can see that it sets x.interactive = I_ASK_USER;, so you can search in this and other files where I_ASK_USER is checked.

In copy.c you will find this symbol near a function call overwrite_ok, and this function contains a call to a function yesno.

The rest is left as an exercise for the reader. ;-)