Question about login_tty source

[Adhemerval Zanella]

On 19/08/2020 00:01, Soumendra Ganguly via Libc-help wrote:
> Hello. In /login/login_tty.c, while setting fd to be our controlling
> tty, we first check #ifdef TIOCSCTTY. This question is about the #else
> part. In
> 
> if (fdname)
> {
>     if (fd != 0)
>        (void) close (0);
>    if (fd != 1)
>        (void) close (1);
>    if (fd != 2)
>        (void) close (2);
>    newfd = open (fdname, O_RDWR);
>    (void) close (newfd);
> }
> 
> can anyone please explain why we need the following block?
> 
>    if (fd != 0)
>        (void) close (0);
>    if (fd != 1)
>        (void) close (1);
>    if (fd != 2)
>        (void) close (2);
> 
> If necessary, dup2 should close 0,1, and 2 later anyway.

Since this was added by Roland almost 25 years ago, it seems to be trying to
emulate the TIOCSCTTY for system derived from System V (as indicated by this
stackflow thread [1]). 

Back in the day glibc tries to enclose fallbacks to be portable to the various
Unix implementations.  Currently this is just a dead code with various
shortcomings (as for a lot of the glibc fallback implementation where proper
kernel support was the correct answer) since neither Linux nor Hurd uses it.

And it seems that Unixes derived from System V also has moved on from this
interface, neither Solaris nor Illumos seems to provide it.

https://stackoverflow.com/questions/51593530/code-explanation-for-glibc-login-tty-function-openttyname-immediately-f