On Tue, Apr 09, 2013 at 09:07:34AM +1000, Dave Chinner wrote:
> On Wed, Mar 27, 2013 at 05:32:32PM -0300, Carlos Maiolino wrote:
> > Hi, working on xfstests after its new directory structures I noticed we
> > should
> > use now something like:
> >
> > ./check xfs/<test>
> >
> > to run some tests, IMHO this is not intuitive and I was working on a patch
> > to
> > make us able to use something just like the old way:
> >
> > ./check <test>
>
> What do you do with duplicate test names?
>
> The main reason that the interface changed was to keep the changes
> to the test harness down to a minimum as it was just moving tests
> around. All that code needs to be revisited to support arbitrary
> test names, so there wasn't much point in doing a massive rework
> only to have to rework it again...
>
> But the question is: is the old way a sane way to specify tests in
> the brave new world? I'd much prefer that test specification is
> explict, and doesn't implicitly select tests. Indeed, if it
> implicitly selects tests (e.g. when there are duplicates it runs all
> duplicates) then we still need a method for running specific
> tests.....
>
> Note that what you are seeing is how the $have_test_arg code
> processes the test name. It requires that you tell it the directory
> so it knows where to look for the specific test. You could make it
> look in each test directory like get_group_list()/get_all_tests() do
> so we don't need to specify a directory.
>
> > But, since xfstests is becoming more generalist than xfs specific, I wonder
> > if
> > we should still keep xfs as default.
>
> The default is whatever filesystem is on the $TEST_DEV, and I don't
> see that changing. i.e. what we set FSTYP to is the default. Note
> that get_group_list() and get_all_tests() specifically include the
> FSTYP directory,
>
Hi, this makes sense to me Dave, thanks to the explanation.
We have some documentation which says xfstests will run tests for a xfs
filesystem by default if no other fstype is specified, I'll change this into
documentation then.
Cheers,
--
Carlos
|