Why does this compile?

[Mark Millard]

On 2017-Sep-25, at 9:38 PM, Russell Haley <russ.haley at gmail.com> wrote:

> I'm trying to compile the new dotnet core 2.0 and I've run into a C
> problem I don't understand. Since I ran the code to check it on my arm
> board, I'm going to ask here (the most knowledgeable fbsd C people I
> could ask).
> 
> The cmake file is trying to test for a linux struct in_pktinfo:
> 
> check_c_source_compiles(
>    "
>    #include <${SOCKET_INCLUDES}>
>    int main()
>    {
>        struct in_pktinfo;
>        return 0;
>    }
>    "
>    HAVE_IN_PKTINFO)
> 
> SOCKET_INCLUDES resolves to netinet/in.h so the final source is:
> 
> #include <netinet/in.h>
> 
> int main()
> {
>    struct in_pktinfo;
>    return 0;
> }
> 
> This compiles on FreeBSD current and apparently on 11 too. That's a
> bad thing because it's supposed to fail. I checked in.h and there is
> no struct for in_pktinfo. Not surprisingly, if I remove the include
> altogether, it still compiles.

struct in_pktinfo;

declares but does not define the struct type. Not even
the size is known --but nothing is done that needs
to use even the size.

By contrast the below would need the definition
of the struct type in question:

#include <netinet/in.h>

int main()
{
   struct in_pktinfo struct_instance;
   return 0;
}

> I assume then that the original author made a mistake? My C is too
> weak and most of my searches don't turn up anything close to what I'm
> looking for.

The program needs to have something that requires
seeing the definition of the type, such as needing
its size.

> Any suggestions would be awesome.  :)

check_c_source_compiles(
   "
   #include <${SOCKET_INCLUDES}>
   int main()
   {
       struct in_pktinfo struct_instance;
       return 0;
   }
   "
   HAVE_IN_PKTINFO)


===
Mark Millard
markmi at dsl-only.net