[Sheepdog] [PATCH] sheep: configure should check if libattr exists

[Christoph Hellwig]

On Fri, Dec 30, 2011 at 12:50:47AM +0800, Liu Yuan wrote:
> On 12/29/2011 11:01 PM, Christoph Hellwig wrote:
> 
> > The right fix is to replace the attr/xattr.h include with sys/xattr.h,
> > which is provided by glibc.
> > 
> 
> 
> Looks that Farm uses it wrong. I include attr/xattr.h as man 2 getxattr
> suggests.
> 
> When I tried sys/xattr.h, I have to additionally manually change XNOATTR
> into *XNODATA* or meet an compile error.
> 
> So, I just wonder which is right way to use getxattr/setxattr system
> call? If we are supposed to use sys/xattr.h, I think there is some
> manual accessible to tell that we might have to use XNODATA instead of
> XNOATTR for error check. Do I miss anything?

Given that Linux doesn't have the ENOATTR error code I'd suggest to
simply not use it.  It's left in libattr as a porting aide for code from
IRIX that had it.  But only the libattr highlevel interface from
<attr/attributes.h> is portable anyway, so using it for the lowlevel
Linux interface doesn't make a whole lot of sense.