Bad configure test for getpwent() ?

["Bart Schaefer" <schaefer@candle.brasslantern.com>]

No patch because I'm not sure it's wrong, but there's this test:

--- 8< --- cut ---
dnl -----------
dnl test for faked getpwnam() entry, ie a single entry returned for any username
dnl for instance, BeOS R4.51 is not multiuser yet, and fakes getpwnam()
dnl test by looking up two usernames that shouldn't succeed, and compare entry
dnl -----------
if test $ac_cv_func_getpwnam=yes; then
    AC_CACHE_CHECK(if getpwnam() is faked,
    zsh_cv_sys_getpwnam_faked,
    [AC_TRY_RUN([
#include <pwd.h>
main() {
    struct passwd *pw1, *pw2;
    char buf[1024];
    sprintf(buf, "%d:%d", getpid(), rand());
    pw1=getpwnam(buf);
    sprintf(buf, "%d:%d", rand(), getpid());
    pw2=getpwnam(buf);
    exit(pw1!=0 && pw2!=0 && !strcmp(pw1->pw_name, pw2->pw_name));
}
],
      zsh_cv_sys_getpwnam_faked=no,
      zsh_cv_sys_getpwnam_faked=yes,
      zsh_cv_sys_getpwnam_faked=no)])
    if test $zsh_cv_sys_getpwnam_faked = yes; then
      AC_DEFINE(GETPWNAM_FAKED)
    fi
fi
--- 8< --- cut ---

I don't think getpwent() is required to return a unique pointer each time;
that is, I think it's allowed to re-use an internal static buffer for each
entry that it returns.  That would mean that this test always succeeds ...

I think the only right way is to test the first pointer for nonzero, copy
the entry into a local struct passwd, then call getpwent() again and do
the comparison.

-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com