On Apr 22 07:13, Klaus Birkelund Jensen wrote:
> On Apr 22 01:12, Shinichiro Kawasaki wrote:
> > On Apr 21, 2020 / 09:33, Klaus Jensen wrote:
> > >
> > > -device_requires() {
> > > - ! _test_dev_is_zoned || _have_fio_zbd_zonemode
> > > -}
> > > -
> > > test_device() {
> > > echo "Running ${TEST_NAME}"
> > >
> > > @@ -25,6 +21,10 @@ test_device() {
> > > local zbdmode=""
> > >
> > > if _test_dev_is_zoned; then
> > > + if ! _have_fio_zbd_zonemode; then
> > > + return
> > > + fi
> > > +
> >
> > This check is equivalent to device_requires() you removed, isn't it?
> > If the skip check in test_device() is the last resort, it would be the
> > better to check in device_requires(), I think.
> >
>
> I did fix something... just not the real problem I think.
>
> Negations doesnt really work well in device_requires. If changed to
>
> ! _require_test_dev_is_zoned || _have_fio_zbd_zonemode
>
> then, for non-zoned devices, even though device_requires returns 0,
> SKIP_REASON ends up being set to "is not a zoned block device" and skips
> the test in _call_test due to this.
>
> There are two fixes; either we add a _require_test_dev_is_not_zoned
> again or put the negated check in an arithmetic context, that is
>
> (( !_require_test_dev_is_zoned )) || _have_fio_zbd_zonemode
>
> I think the second option is a hack, so we'd better go with the first
> choice.
Doh.
The _is_not_zoned version would of course just cause the test to be
skipped for zoned devices instead.
So I actually think my original fix is the best option here.
