On Wed, Sep 26, 2001 at 10:59:12PM +1000, Hamish Moffatt wrote:
> This is a pretty funky bug. I've been experimenting with it, as I have
> an hppa machine of my own. I have all the latest sid updates installed.
>
> The difference seems to be g++ 3.0 on hppa. You can replicate it on
> i386 (etc) with g++-3.0 too.
Interesting...I thought that I had tried it with g++-3.0 on i386 as well,
but I can confirm that it doesn't work now. I've reassigned the bug to
g++-3.0.
> If I do
>
> g++ -E -c -D_BSD_SOURCE /usr/include/netinet/udp.h
>
> then I can see the BSD definitions (uh_sport etc), both on i386 and hppa.
> But if I do
>
> g++-3.0 -E -c udphdrtest.c
>
> then I see the non-BSD definition on both. Works ok with g++ 2.95.4 on
> i386, but hppa doesn't have g* 2.x.
>
> Looking at the output of g++{,-3.0} -c -E -dM, version 3 fails
> to define __FAVOR_BSD, which the header file uses to choose the
> definition. <features.h> only #defines __FAVOR_BSD if _GNU_SOURCE
> is undefined; on g++ 3.0, _GNU_SOURCE is defined.
>
> touch blah.c
> g++ -c -E -dM blah.c shows no _GNU_SOURCE;
> g++-3.0 -c -E -dM blash.c shows _GNU_SOURCE, hence no __FAVOR_BSD
>
> I'm not sure if there is a bug in the library or not (I can't work out
> where _GNU_SOURCE is being defined; it's only with g++-3.0, not with
> gcc-3.0, too). You could #undef _GNU_SOURCE in your source though;
> that might be a hack.
Thanks for the analysis. I can't find anyplace where _GNU_SOURCE is
defined, even conditionally, in /usr/include or /usr/lib/gcc-lib. It isn't
defined by cpp (cpp -dM /dev/null doesn't show it). Hopefully the GCC
maintainers will have some ideas.
--
- mdz
