[PATCH] vfs_gpfs: move btime stuff from stat to get_dos_attrs

[Christof Schmitt]

On Fri, Dec 16, 2016 at 08:07:19AM +0100, Ralph Böhme wrote:
> On Thu, Dec 15, 2016 at 10:29:23PM +0100, Ralph Böhme wrote:
> > On Thu, Dec 15, 2016 at 12:40:22PM -0700, Christof Schmitt wrote:
> > > $ git am ~/vfs_gpfs_btime.patch
> > > Applying: vfs_gpfs: update btime in vfs_gpfs_(f)get_dos_attributes
> > > Applying: vfs_gpfs: remove updating btime from stat VFS calls
> > > Applying: vfs_gpfs: simplify stat_with_capability() ifdef
> > > 
> > > For some reason, git am decides to switch the two parts for me. As long
> > > as the correct version ends up in master, i do not see a problem.
> > 
> > This sounds really strage, I'll dig into this tomorrow, afaict soemthing like
> > this should not happen. *scratches head really hard*
> 
> I could reproduce it: the line offsets in the hunks where off a few lines
> because in the branch where I created the patchset with git format-patch there
> was another patch for vfs_gpfs before the three patches I posted.
> 
> This caused the line offset, but I'm still puzzled that this can cause git am to
> behave like this and make it silently reorder commit hunks.
> 
> Pushed the patchset after rebasing my original patchset on a clean master
> branch.

Thanks for finding this. My guess would be that in this case the line
numbers were the only way to identify the place where the hunk should be
applied to. If the numbers were off, then git (or patch) did not have
any information.

I was concerned with the potential build breakage, and sometime we
should try to address this by e.g. including the gpfs.h header in
third_party/ and always build the module.

Christof