On Fri, Jul 05, 2019 at 11:40:53PM -0400, Theodore Ts'o wrote:
> On Fri, Jul 05, 2019 at 04:07:57PM +0800, Eryu Guan wrote:
> > > +_require_scratch_inode_limits()
> > > +{
> > > + _require_scratch
> > > + _scratch_mkfs > /dev/null 2>&1
> > > + _scratch_mount
> > > + if [ $(_get_free_inode $SCRATCH_MNT) -eq 0 ]; then
> > > + _notrun "$FSTYP does not have a fixed number of inodes available"
> > > + fi
> > > + _scratch_unmount
> > > +}
> >
> > I think testing against $TEST_DIR should be sufficient, so we could
> > avoid the mkfs & mount & umount SCRATCH_DEV time.
>
> I was following the pattern that I saw with other similar _require
> tests (for example: _require_scratch_shutdown). I *thought* the
> reason why this is was done is because if the test only uses the
> SCRATCH_DEV, there's no making it a requirement that TEST_DEV be
> available --- since IIRC, we do support SCRATCH_DEV being available,
> but not TEST_DEV.
>
> I personally don't use xfstests in that way --- when I run xfstests,
> TEST_DEV is always available and in some cases, SCRATCH_DEV won't be
> present. But I thought that's why _require_test exists --- so that
> tests can be skipped if TEST_DEV does not exist.
xfstests assumes TEST_DEV is always present and SCRATCH_DEV is optional
(but recommended). _require_test not only checks if TEST_DEV is avaiable
& properly mounted but also implies we should do fsck against TEST_DEV
after test, as _require_test does "touch ${RESULT_DIR}/require_test" as
well.
Regarding to _require_scratch_xxx, that's usually because the
requirement being checked might change when mkfs with different options.
But a filesystem will always have a fixed number of inodes no matter
what mkfs options it uses (the same is true for not having a fixed
number of inodes). So in this case, I think check against
TEST_DEV/TEST_DIR should be fine.
Thanks,
Eryu
