[LTP] [PATCH v6 5/8] tst_test.sh: Add $TST_ALL_FILESYSTEMS

[Petr Vorel]

> Hi!
> > > Whatever the API ends up I guess that we should check the support and
> > > mkfs here as well in order to TCONF early before we create a device and
> > > attempt to format/mount it.
> > I was worried to add mkfs.foo dependency unless there is a requirement for it
> > (i.e. TST_NEEDS_DEVICE=1). Do you really want to add this dependency for *all*
> > tests? (I'd prefer no, i.e. to keep here check only for test not being skipped,
> > because check for mkfs.foo is in tst_mkfs which does the job, thus not only we'd
> > add an extra unneeded dependency but also duplicity in the check).

> Isn't this mostly theoretical problem? I mean how common it would be to
> have test that tests different filesystem but does not need
> corresponding mkfs?

Have you noticed !=, i.e.:
if [ "$TST_ALL_FILESYSTEMS" != 1 ]; then
The check here is for single filesystem testing, just with disabled on some
filesystems (i.e. skip on tmpfs). Equivalent of fcntl33.c (skips on tmpfs,
ramfs - BTW not in fs_type_whitelist, thus not being checked, nfs; but not using
.all_filesystems).

That's why I'm not going to add check for kernel support and mkfs.foo (hope I
was correct you overlooked !=, or did I miss something?)

For testing on all filesystems (TST_ALL_FILESYSTEMS=1) there is obviously check
for kernel and mkfs support in testcases/lib/tst_supported_fs.c.

> > But if you want it, i.e. if we require mkfs.$TST_FS_TYPE (unless tmpfs of
> > course, which is handled properly) here, then we don't need to separate checks
> > for mkfs.foo and skip list in testcases/lib/tst_supported_fs.c.

> I would go that way for now.
I dare to merge this approach:

* all filesystems
tst_supported_fs [-s skip_list]
   print the list of supported filesystems
   if fs_type is supported and not in skip_list (optional),
   print list of supported filesystems and return 0
   if fs_type isn't supported or in skip_list, return 1

* single filesystem
tst_supported_fs fs_type
   if fs_type is supported, return 0 otherwise return 1

tst_supported_fs -s skip_list fs_type
   if fs_type is in skip_list, return 1 otherwise return 0


> I guess that the cleanest solution would be keeping the funcionality
> really orthogonal, i.e. separating the kernel and mkfs checks so that we
> would have -m to enable mkfs check -k to enable kernel check and -s to
> enable skiplist. But that is something that can and should be done after
> the relase.

If really needed, sure.

Kind regards,
Petr