On Wed, May 11, 2022 at 06:33:32AM -0700, Rob Clark wrote: > > And ofc we want the expectations to be in the kernel tree because > there could be, for example, differences between -fixes and -next > branches. (Or even stable kernel branches if/when we get to the point > of running CI on those.) There are tradeoffs both ways, whether the patches are kept separate, opr in the kernel tree. In the file system world, when we discover a bug, very often a test case is found to test the fix, and to protect us against regressions. It has one other benefit; since the tests (xfstests) are kept separate from the kernel, it's a useful way to identify when some patch didn't get automatically backported to a LTS or distro kernel. (For example, because the patch didn't cherry-pick cleanly and the manual backport process fell through the cracks.) It does make things annoying when we have bugs that can not be safely backported (which results in tests that fail on the LTS kernel without kernel-version exclude files), and/or when the expectations change between versions. (Although to be honest, for us, the more common annoyance is when some userspace package --- e.g., bash or coreutils or util-linux --- changes their output, and we have to add filter functions to accomodate expected output differences.) - Ted
