Shouldn't glibc-devel r: kernel-headers?

[wrobell]

On Wed, Oct 16, 2002 at 12:21:59PM +0200, Jakub Bogusz wrote:
> On Tue, Oct 15, 2002 at 10:10:58PM +0200, Michal Moskal wrote:
> > On Tue, Oct 15, 2002 at 08:58:54PM +0200, Jakub Bogusz wrote:
> > > On Sat, Oct 12, 2002 at 09:21:54PM +0200, Andrzej Krzysztofowicz wrote:
> > > > > There are some situations when you should recompile
> > > > > your glibc, but they are rare.
> > > > 
> > > > Kernel 2.4+ headers are no longer intended to be used in userspace.
> > > > Eg., you must hack them in some places to use them with C++.
> > > 
> > > Where? I didn't change anything in 2.4 headers and don't see such
> > > problems (after recompiling glibc on those headers).
> > > But I did see strange type conflicts when I tried to use fresh kernel
> > > headers with glibc compiled on older headers (that was in 2.2.1x era).
> > > Those problems dissapeared after recompiling glibc.
> > > 
> > > > So providing own copy of hacked kernel-headers included into glibc-devel
> > > > seems to be the only choice when we switch into 2.4/2.6.
> > > 
> > > I agree - kernel headers in /usr/include/{asm,linux}, in glibc-devel
> >                                                 ^^^^^
> > If program does include <linux/...> it probably should
> > -I/usr/src/linux.
> 
> I don't think so... user space programs usually include <linux/*> only
> for ioctl numbers and similar things.
new structures, etc... :-)

> Also, many glibc headers and <asm/*> headers include <linux/*>.
> 
> Using -I%{_kernelsrcdir}/include to compile programs would require
> recompiling glibc after each kernel headers upgrade (required to build
> modules for new kernel).

Take iptables (I mean user space programs, not modules) for example. For
several months I had been using glibc from Ra, used non-distro kernel,
compiled and used iptables without any problems.

    wrobell <wrobell at ite.pl>