[PATCH] add attribute nonstring

[Paul Eggert]

[Dropping tz@iana.org as this email isn't relevant to tzdb.]

Martin Sebor wrote:
> I tried to make only minimal changes to suppress
> the new warning to avoid excessive churn.

As the main goal here is to find bugs not to pacify GCC, it's helpful to take 
these diagnostics as an opportunity to look more carefully at the strncpy calls. 
I just now looked at the other two data structures affected by the patch, and 
found some issues.

First, the patch to sysdeps/gnu/bits/utmp.h seems reasonable, as the members 
ut_user etc. have used strncpy format since the 1970s and people who deal with 
this format are used to it. However, this format does not appear to be 
documented in the manual. I suggest accompanying it with a patch to the part of 
manual/users.texi that talks about ut_user etc., so that the __NONSTRING issue 
is documented there.

Second, the patch to sysdeps/gnu/net/if.h is dubious. Its goal appears to be to 
pacify calls like the first call to strncpy in 
sysdeps/unix/sysv/linux/if_index.c. But that first call appears to be a bug in 
glibc, since the underlying Linux system call expects a null-terminated string. 
My guess is that
this particular data structure should indeed be a null-terminated string, and 
should not be marked with __NONSTRING, and that we should instead change the 
glibc code to not use strncpy here. I suppose the code should report an error 
instead of creating a non-null-terminated array, and that it need not initialize 
the rest of the array to zeros.