[PATCH] xfstests: enable many tests to run on ext2/3/4

[Eric Sandeen]

Christoph Hellwig wrote:
> Wow, that's a nice start.  The only important thing missing is checking
> the filesystems after each test run for the non-xfs case.

Yep, that was in the back of my mind.  Probably needs more abstraction
to make that cleaner.

> Maybe we should put this in in stages?  The _supported_fs generic
> thing is a nice cleanup already for the existing xfs/nfs/udf setup
> and should go in ASAP.

sure, I can break it up.

> The _scratch_mkfs output fix in 069 could also be a separate patch.
> 
> The _setup_generic_testdir should be generalized to match XFS for the
> default case and just set testdir in _setup_testdir instead of
> another function.  Also the comment there should be updated.
> 
> Same for _cleanup_testdir.
> 
> Btw, the way udf and nfs are currently handled look not very nice to me.
> We should not set up the test device by default for any filesystem but
> rather have a -setup or similar option to set it up if needed.

many of them likely fail, too, some of the acl & attr tests have some
assumptions about xfs limits.

> In common I would indeed prefer a new fstype option, but we might aswell
> put the current version in as-is.  Especially if we could tie up a really
> generic fstype= that wouldn't require listing the filesystems if they
> don't require special mount options or similar.

Ok.  I'd even thought that maybe by default, w/o options, it should just
run as whatever $TEST_DEV is formatted to (though that's trickier for
nfs I guess)

> The only thing preventing that is as far as I can see the current difference
> in _require_scratch for xfs and udf vs the rest.  Which looks really weird
> to me, need to investigate what's going on.

I think this is because even for udf etc, it still expects $TEST_DIR to
be xfs, so swizzles around test & scratch. yeah, I agree that's messy.

> As for the generic group I must say I don't like it very much, the
> filtering of notrun (maybe only notrun because of the filesystem type
> mismatch) sounds much better to me.

yeah, after I ran it a bit more I think I tend to agree....

I'll work on breaking this up a bit and tidying up some of the loose
ends, since the basic approach seems sane to more than one person now  :)

Thanks,
-Eric