On Tue, Jan 12, 2016 at 11:48:59PM -0800, Christoph Hellwig wrote:
> On Wed, Jan 13, 2016 at 08:24:06AM +1100, Dave Chinner wrote:
> > I can't apply this straight off. Most of the libxfs code that is
> > changed is shared with the kernel code, and so the definitions of
> > the variables need to be the same as the kernel code. There are
> > reasons for the kernel code using __[u]int*_t type variants (e.g. I
> > think the endian conversion static checker requires the __ variants
> > for host order variables), and so before making sweeping changes
> > like this we need to ensure that we can make the equivalent changes
> > to the kernel code as well...
>
> There is no functional requirement for these types. I'm not sure if
> they are an IRIXism, or something that was added during the early
> Linux port, but they certainly aren't nessecary.
>
> I'd love to kill them in favour or either the kernel __u/s types or the
> C99 types which are like the urrent ones just without the __ prefix.
>
> Maybe we should get an agreement which ones we want and ask Felix for
> a patch to the kernel tree?
Either way is fine by me, but it's not a small patch - there's ~600
variables declared in fs/xfs with the __[u]int*_t types.
Cheers,
Dave.
--
Dave Chinner
david@xxxxxxxxxxxxx
|