Bug#108663: Help with bug #108663

["Carlos O'Donell Jr." <carlos@megatonmonkey.net>]

> I know this bug doesn't belong to my package, but I don't yet know where to
> reassign it.  It is definitely related to g++, and may be specific to hppa.
> Could someone take a look at it and see if they can make any sense of it?
> 
> The problem involves the operation of this small test program:
> 
> #define _BSD_SOURCE
> #include <netinet/udp.h>
> 
> void f(void) {
>   struct udphdr header;
>   header.uh_sport = 0;
> }
> 
> Which works with both gcc and g++ on i386, and works with gcc on hppa, but
> fails on hppa with g++.  See the BTS for details:
> 
> http://bugs.debian.org/108663


Hrm... whom to assign blame?

The udphdr struct is encased in #ifdef __FAVOR_BSD which is set in:

/usr/include/features.h (part of glibc)
---
/* If _BSD_SOURCE was defined by the user, favor BSD over POSIX.  */
#if defined _BSD_SOURCE && \
    !(defined _POSIX_SOURCE || defined _POSIX_C_SOURCE || \
      defined _XOPEN_SOURCE || defined _XOPEN_SOURCE_EXTENDED || \
      defined _GNU_SOURCE || defined _SVID_SOURCE)
# define __FAVOR_BSD    1
#endif
---

Though it seems the comment is backwards?
Since setting any of those will cause __FAVOR_BSD to stay unset...

udp.h -> cdefs.h -> features.h -> (no more includes)

Logical assumption: 
	Some define somewhere is setting one of those?
	or
	Something is undef'ing _BSD_SOURCE

So I begin ripping out the defines in the boolean expression of features.h

<Carlos plunks away>

Hrm... I've removed all of them. It just seems like _BSD_SOURCE is
_not_ defined even though it _is_ defined in the source (gcc issue?).

#ifdef _BSD_SOURCE
# define __FALSE_DEFINE    1
#else
# define __FAVOR_BSD    1
#endif

Works. Because _BSD_SOURCE is _not_ defined (even though it is :).

---

So I try:

#undef _BSD_SOURCE
#define _BSD_SOURCE
#include <netinet/udp.h>

void f(void) {
  struct udphdr header;
  header.uh_sport = 0;
}

int main(void) {

        f();
        return 0;

}

It works. And oddly enough, it continues to work thereafter
even if I remove the #undef line @_@

I haven't tried rebooting and seeing what happens (the 712/60 I have
has a tendency to be tempermental, I'll try it tommorow when I'm there
physically)

I'm not sure who to blame!
gcc isn't as stateful as glibc, and this problem seems like a bug
in glibc... hung jury.


On another note...

My current understanding:

2.2.3-9 - Doesn't compile anymore on HPPA with gcc 3.0.2
2.2.4-1 - Merging changes to get it to compile without dying

I'm helping along with the work for 2.2.4-1 on HPPA.
Some people have decided to wait until the next compiler comes out 
in order to fix other issues.

Currently 2.2.4-1 dies randomly with a seg'ing Zic while 
walking timezones... odd :}

Hope this helps with understanding the problem.
Get someone to try the 'undef' beforehand case.

Cheers,
Carlos O'Donell Jr.