[GNU Crypto] Interesting behavior with crypt()

[Scott Sanders]

I was researching some odd behavior in Apache and mod_dbd and thanks
to the apr-devel mailing list I was informed that my problem was not a
bug in mod_dbd, but was actually the way crypt() behaved. I am posting
here because in my mind this is a bug, but I can also see how there
might be arguments to the contrary and would like the communities
opinion on the matter.

If the hash given to crypt() as a salt is an empty string, then the
crypt function returns an empty string and the resulting strcmp will
always match regardless of the password the user provides. This makes
an empty string invalid as a password, which is not the case on
Windows, BeOS, or Netware (according to Tom Donovan from the apr-dev
mailing list. http://thread.gmane.org/gmane.comp.apache.apr.devel/13795).

Modifying the demo source in the GNU Manual for crypt to the following
shows the bug rather plainly.

--begin--code--

#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <crypt.h>

int
main(void)
{
        /* Empty hash */
        const char *const pass = "";

        char *result;
        int ok;

        /* Read in the user's password and encrypt it,
                 passing the expected password in as the salt. */
        result = crypt(getpass("Password:"), pass);

        /* Test the result. (This will always succeed!) */
        ok = strcmp (result, pass) == 0;

        puts(ok ? "Access granted." : "Access denied.");
        return ok ? 0 : 1;
}

--end--code--

Please help me understand why this is behaving this way if it is
correct, otherwise I would be glad to file a bug report on it.

-Scott