Re: [RFC PATCH] fstests: add support for hfsplus

[Eryu Guan <guaneryu@gmail.com>]

On Sun, Sep 02, 2018 at 12:13:26AM -0500, Eric Sandeen wrote:
> 
> 
> On 9/1/18 11:03 PM, Ernesto A. Fernández wrote:
> > On Sun, Sep 02, 2018 at 01:09:57AM +0800, Eryu Guan wrote:
> >> On Fri, Aug 31, 2018 at 11:05:03PM -0300, Ernesto A. Fernández wrote:
> >>> It is not possible to set file system size when running mkfs.hfsplus,
> >>> so use the device mapper as a workaround.
> >>
> >> I'd prefer _notrun the test in _scratch_mkfs_sized() instead of this
> >> workaround, as the harness is expecting to operate $SCRATCH_DEV but this
> >> workaround may break this assumption in subtle ways, e.g. now
> >> _check_scratch_fs fails to create hfsplus-tmp device because
> >> $SCRATCH_DEV may still be mounted.
> > 
> > I didn't realize $SCRATCH_DEV could still be mounted, sorry. Maybe I can
> > unmount it before _dmsetup_create(), and remount after _dmsetup_remove()?
> > Like _check_generic_filesystem() does. Or would that bring other problems?

I haven't seen other problems yet (after just a few simple test runs),
but I suspect it may break tests in more subtle ways, the
_check_scratch_fs failure is just an example. And even it works for now,
and we have to worry about making it continue to work when adding new
features in the future and that's a maintaining burden we'd better to
avoid.

> > 
> > I know it's ugly, but one test that uses _scratch_mkfs_sized() has helped
> > me find a number of bugs already. It would be really useful to get it to
> > work.
> 
> Has anyone proposed a patch to mkfs.hfsplus to accept a filesystem size?
> I'd expect that might be less complicated than this sort of devicemapper
> setup ;)

Yeah, this would be the best way to let _scratch_mkfs_sized() support
hfsplus :)

Thanks,
Eryu