Filesystems, testing, and stable trees

> The key to having stable kernels with stable filesystems is being able to run fstests (formerly xfstests) on them.

This is important, but it would not spot any problems of the class linked higher in this article (which I have a personal interest in, being the person that bug bit). AIUI, fstests creates filesystems and then tests them, all in the same kernel version: situations in which filesystems touched by earlier kernels (or, alternatively, made by older mkfses) have properties that newly-mkfsed ones do not will not be spotted, which is how the bug in question sneaked in (the new kernel could handle stuff *it* had written fine, just not stuff older kernels predating the problematic change had written).

And, of course, almost all filesystems will have been written by "older kernels" by this definition: nobody re-mkfses all their writable filesystems on every kernel upgrade!

(Maybe a fstest run mode in which the mkfses and some of the writes are done by a stable baseline kernel and then the rest are done by the kernel under test might work, but where do you draw the boundary between which things should be done by an older kernel and which by a newer? Bugs of this nature might occur after any of those writes...)