Re: [kbuild-all] fs/xfs/xfs_ondisk.h:96:2: error: call to '__compiletime_assert_96' declared with attribute error: XFS: sizeof(xfs_dir2_sf_entry_t) is wrong, expected 3

[Fengguang Wu]

Hi Christoph,

On Sun, Oct 16, 2016 at 09:45:56AM -0700, Christoph Hellwig wrote:
> > head:   1001354ca34179f3db924eb66672442a173147dc
>
> That's Linux 4.9-rc1
>
> > 30cbc591 Darrick J. Wong 2016-03-09 @86  	XFS_CHECK_STRUCT_SIZE(xfs_dir2_data_unused_t,		6);
>
> But that's not how xfs_ondisk.h in 4.9-rc1 looks like

Yeah it's a bit confusing for old errors -- the git blame is on the
old first bad commit, which is

ab9d1e4f7b0217948a3b35a64178602ab30ff45d Merge branch 'xfs-misc-fixes-4.6-3' into for-next

> > 30cbc591 Darrick J. Wong 2016-03-09 @96  	XFS_CHECK_STRUCT_SIZE(xfs_dir2_sf_entry_t,		3);
>
> And that's line 119.
>
> Something is odd about this report.
>
> But both xfs_dir2_data_unused_t and xfs_dir2_sf_entry_t have one thing in
> common: they are strutures that aren't padded to a natural alignment at
> the end.  It seems like the openrisc gcc does implicit padding for them,
> which sounds like a nightmware waiting to happen for any sort of disk
> or network structure, so I'm really not sure if we should work around
> it in xfs.

FYI the error for 'xfs_dir2_sf_entry_t' only shows up in openrisc;
and the 'xfs_attr_shortform_t' error only shows up in cris builds.

There is no errors for 'xfs_dir2_data_unused_t' in recent kernels.

Thanks,
Fengguang