On Wed, May 18, 2016 at 03:44:46PM -0500, Eric Sandeen wrote:
> On 5/18/16 11:37 AM, Jeffrey Bastian wrote:
> > There was a discussion a few months ago about adding a guard for the
> > fsxattr struct [0] because it's defined in two places, the Linux kernel
> > header linux/fs.h [1] and xfsprogs header xfs/linux.h [2].
>
> > xfs/linux.h has a FS_IOC_FSGETXATTR guard around the struct fsxattr
> > definition, but this only works if linux/fs.h is included *before*
> > xfs/linux.h (or xfs/xfs.h). If you include linux/fs.h after, then you
> > get a struct redefinition error.
> >
> > Is it a requirement that linux/fs.h is included first? If so, then
> > there is a bug in xfstests because it includes them in the wrong order
> > [3] and fails to build. If there is not an order requirement, then both
> > header files should probably have a HAVE_FSXATTR guard around the struct
> > definition.
>
> It seems best to me to include fs.h first. That may not be written in
> stone, but it's at least common practice.
>
> Having the same definition in both places, and guards going both ways,
> seems a little odd though.
>
> Maybe xfsprogs' include/linux.h should just directly include
> the kernel's linux/fs.h at the top - would that make sense?
That's the easiest solution - stops people wasting even more time on
this.
Cheers,
Dave.
--
Dave Chinner
david@xxxxxxxxxxxxx
|