Re: [PATCH] generic: add missing $FSX_AVOID to fsx invocations

["Theodore Ts'o" <tytso@mit.edu>]

On Wed, Feb 08, 2023 at 02:26:56AM +0800, Zorro Lang wrote:
> fstests has merged below change 2 month ago:
>   [PATCH] fstests: update group name according to xfs_io command requirement
>   https://lore.kernel.org/fstests/20221108183242.3362013-1-zlang@kernel.org/
> 
> So I'd like to check if it helps for the problem you described above?
> If not, I think we can think about the patch you metioned above.

Oops, sorry, this got lost in my inbox.  :-(

It definitely helped, thanks.  My one observation about this patch is
that it's a one-time fix-up.  I tried rerunning the script referenced
in the patch, and there were 11 tests that it "fixed up".  Now, they
were all adding tests to the "prealloc" group, which I think you had
deliberately excluded, because they weren't actually testing prealloc,
but it's the worry that future fstests developers might forget to set
the group name correctly, which is why I still have "common: introduce
XFS_IO_AVOID env var"[1] as an out of tree patch.

[1] https://lore.kernel.org/all/1445107518-32022-1-git-send-email-tytso@mit.edu/

It's a small change, it almost never conflicts with upstream changes
(generally the only time I have to deal with a conflcit when rebasing
is when a new environment variable is added to the documentation in
README), and it means that when I run "kvm-xfstests --no-collapse", my
wrapper scripts do this:

    no_collapse)
        ALL_FSSTRESS_AVOID="$ALL_FSSTRESS_AVOID -f collapse=0"
        ALL_FSX_AVOID="$ALL_FSX_AVOID -C"
        ALL_XFS_IO_AVOID="$ALL_XFS_IO_AVOID fcollapse"
        FSTESTSET="$FSTESTSET -x collapse"
        ;;

and I'm *guaranteed* to make sure that any tests involving
collapse_range will be skipped.  Do I strictly speaking need the
out-of-tree patch in [1], probably not, assuming the group list is
always kept up to date, and to be honest it's been a *long* time since
I've never needed to use gce-xfstests --no-collapse or --no-insert.

However, the cost of keeping the out-of-tree patch in my local
xfstests git repo is quite low, so I've just kept it.  But do I *need*
it?  Arguably, no, which is why I haven't been bugging you about it.
:-)

						- Ted