[ntp:bugs] [Bug 1862] Compilation error, ntp_io.c:1678: undefined reference to `in6addr_any'

[bugzilla-daemon at ntp.org]

http://bugs.ntp.org/show_bug.cgi?id=1862

--- Comment #13 from Stefan olsson <misca at hotmail.com> 2011-03-29 07:03:20 UTC ---
Hi


> Next step (and I'm working on another bug right now...) is to find
> the minimum changes to the test needed to make it not optimize away the reference.

One idea could be to force ./configure to use avoid optimization flags for it's
internal tests, such as the ac_fn_c_try_link and ac_fn_c_try_run, like this

my_ac_link=$(echo $ac_link | sed 's:-O[0-9]*::g')

and then use my_ac_link instead of ac_link when compiling/linking the test
code.
In this way the user can use CFLAGS, LDFLAGS as usual but there will be a
slight difference in how ./configure "works" and how make "works".





Also, I tried a number of different use cases to find a small one and here is
the result

------- Use cases that the compiler will "optimize" away
The following constructions will be "optimized away"

#1
        struct in6_addr in6;
        in6 = in6addr_any;


#2
        struct in6_addr in6;
        struct in6_addr in61;
        in6 = in6addr_any;
        in61 = in6addr_any;

#3
        struct in6_addr in6;
        struct in6_addr in61;
        in6 = in6addr_any;
        in61 = in6;


#4
This one have information about the internal structure of struct in6_addr

        struct in6_addr in6;
        in6 = in6addr_any;
        if (in6.s6_addr[0] == 0);


#5
This one have information about the internal structure of struct in6_addr

        struct in6_addr in6;
        in6 = in6addr_any;
        if (in6.s6_addr[0] == 0)
            return 0;

#6
This one have information about the internal structure of struct in6_addr

        struct in6_addr in6;
        in6 = in6addr_any;
        if (in6.s6_addr[0] == 0)
            in6.s6_addr[0] = 1;


------- Use cases that the compiler will NOT "optimize" away
The following constructions will NOT be "optimized away"

#7
This one is trapped
This one have information about the internal structure of struct in6_addr
But it will return '1', which ./configure will identify as an error and this is 
not correct in the case that the test program actually works

        struct in6_addr in6;
        in6 = in6addr_any;
        if (in6.s6_addr[0] == 0)
            return 1;

#8
This one is trapped
        #include <stdio.h>
        #include <string.h>
        #include <sys/types.h>
        #include <sys/socket.h>
        #include <netinet/in.h>



int
main ()
{

        struct in6_addr in6;
        in6 = in6addr_any;
        memcpy(&in6, &in6addr_any, sizeof(in6addr_any));


  ;
  return 0;
}

#9
This one is trapped
        #include <stdio.h>
        #include <string.h>
        #include <sys/types.h>
        #include <sys/socket.h>
        #include <netinet/in.h>



int
main ()
{

        struct in6_addr in6;
        in6 = in6addr_any;
                if (memcmp(&in6, &in6addr_any, sizeof(in6addr_any)) == 0)
                   return(1);


  ;
  return 0;
}


> I'm a little frightened, though, 
> because this sort of simple test is extremely common for configure, 
> including for tests I can't so easily modify as they're part of autoconf.  

I feel uncomfortable too, I have just installed some 50 different software's or
so on the system.

> If I were you'd I'd be tempted to stick with -O1 or whatever works with the tests as written.
I will for now on have a modified in6addr_any test in the ./configure 

> Thanks for tracking this down.

No problem, glad to help and it is fun to help things to become more generic 
in order to spread it more.



Live well


/Stefan

-- 
Configure bugmail: http://bugs.ntp.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.