On Wed, Feb 26, 2025 at 03:41:07PM +0100, David Sterba wrote: > On Fri, Feb 21, 2025 at 11:45:31PM +0800, Zorro Lang wrote: > > > This may bring the question what is the policy for filesystem support. > > > I've seen "case $FSTYP" with widely inconsistent cases of filesystems, > > > probably the longest list is in common/rc:_require_scratch_nocheck with > > > 9p, virtiofs, afs, pvfs2 or ubifs. It would be good to document that in > > > README, I haven't found anything in that regard. > > > > Good idea, maybe we should have a list about that. > > > > There's not a limitation about what filesystems are supported or not. > > Especially the generic test cases, some "not in list" filesystems might > > run generic tests too. So we don't exclude any filesystem which wants > > to try fstests. > > > > Generally when someone wants to help fstests to support a new filesystem, > > there're 3 phases: > > 1) Check if fstests can mkfs, mount, unmount and fsck on this fs natively. > > If it can't, add the fs to the "case...esac" list, to do special treatment. > > 2) Run generic test case on the new fs, fix some test failures which are > > not suitable for this fs. > > 3) Add more test cases for this fs, might need a new tests/$FSTYP directory > > and more common helpers. > > > > Actually some of filesystems which are "supported" by xfstests, just satisfy > > the phase#1, or a few phase#2. That depends on how deeply that fs list > > would like to use xfstests. > > > > But everything is still changing. For example, we don't support nfs much, > > just in phase#1. But as more and more nfs folks start to use xfstests, > > it's good to run on nfs now, even reach to part of phase#3. In contrast, > > although cifs is in the list, and xfstests can be run on it, but too > > many test failures, no one from cifs would like to care about that. > > > > So about the "support list", what's the standard to count in a fs :) > > Thanks, this is useful, can you please add this text to the README? OK, I'll send a patch to add the list to get more reviewing. Thanks, Zorro >
