On Tue, Dec 10, 2024 at 05:08:27PM +0100, Christoph Hellwig wrote: > I think the diffstat alone makes it pretty clear that moving away > form that is a benefit, and it's also a lot easier to understand than > that ext2 and ext3 magically run ext4 tests. We talked about this on the weekly ext4 video chat, and I think what we'd think is actually cleaner is to have a single directory for all ext2/ext3/ext4 tests, and then eventually, have feature-specific guards which skip a test if a particular feature isn't supported by a particular file system. It's always been my position that ext2, ext3, and ext4 are effectively the same file system from a conceptual perspective, with multiple implementations that support different subsets of file system features. This includes /usr/src/linux/fs/ext2, /usr/src/linux/fs/ext3 (before we removed it from more recent rernels), /usr/src/linux/fs/ext4, HURD's implementation of ext2, NetBSD/FreeBSD's implementation of ext2, etc. So effectively, what I'm proposing is that we use xfstests/tests/ext4 effectively as "extN", which would be used when testing with FSTYP=ext[234]. Yes, we'll need to do some cleanup to add feature guards (e.g., _require_metadata_journaling, and "_require_scratch_ext4_feature mmp") instead of _exclude_fs ext2, but in the end, I think this will be cleaner and easier to understand since we'll know exactly what the test is testing. Cheers, - Ted
