cross-compilers and include-fixed... how does this work?

[Paul Smith <paul at mad-scientist dot net>]

Hi all.

To try to restore my sanity after attempting to get a multiarch (i686 &
x86_64) version of GCC 4.2.4 to build, I've decided to go with 4.5.2
which compiles just fine for multiarch.  But I have a problem that seems
like it would be common and I wonder how I'm intended to deal with it.

I'm using --sysroot to compile code for a different target, with a very
different set of headers and libraries than my build system.

However, the contents of the include-fixed directory in my GCC 4.5.2
installation are from the (very old) system that I built GCC on, while
I'm trying to compile for a target with a much newer set of headers and
libraries.  This is causing my builds to fail because the fixed includes
are not matching up with the includes in the sysroot.

For example, I compiled GCC 4.5.2 on a system running Red Hat EL 4 (very
old!).  The contents of the 4.5.2 include-fixed directory, for example
the features.h file, are thus ancient and don't have support for, for
example, the 2008 POSIX spec.

Now when I try to compile with a --sysroot for a newer target that does
have this support, I can't enable it because features.h is being used
from the old system, via include-fixed, not the sysroot.


Can anyone explain how the interaction between include-fixed and
cross-compiling is intended to work?

Thanks!