Re: s390/s390-64 missing xattr exports

[Christoph Hellwig <hch at lst dot de>]

On Mon, Aug 19, 2002 at 09:22:29PM -0400, Roland McGrath wrote:
> Aside from the nits you already noted, that looks ok to me to go in as
> generic (though I think __BEGIN_DECLS belongs after the #includes).

I modelled it after <sys/socket.h>, which has:

#include <features.h>

__BEGIN_DECLS

#include <sys/uio.h>
#define __need_size_t
#include <stddef.h>

> don't have a strong opinion about the header file name.  I would say that
> if there is an extant system that this interface is otherwise compatible
> with, then use the same header file name.  AFAICT the SGI XFS code is the
> only previous thing to be compatible with, and it uses the (silly) name
> <attr/xattr.h>.

It's not XFS but the libattr package which is maintained in the SGI
repository.

> Despite the silliness of the name, if there is any prior
> system anyone ever used to be compatible with and all it takes to be
> compatible is a silly header file name, then compatibility seems a more
> worthwhile goal than nonsilliness. 

This is an addon library, so you had to add -lattr to your LDFLAGS to
use it.  So if you actually want the hard libattr names not to override
the weak glibc ones you have to change your package anyway.  Going to
a sensible name for the header during the process is a good idea (TM).

libattr assumes it owns ${prefix}/include/attr anyway, it has another
IRIX-compatiblity header there and people might add compatibliblity for
say AIX or FreeBSD.