Comment 6 for bug 1902254

3) xfstests (aka fstests)

Compared the 'consistently fail set' and 'likelyhood to fail' in flaky tests (see below.)
No regressions.

Details:
---

Source:
- git://git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git
- commit 31f6949f ("ext4: verify unwritten extent conversion in buff-io")

The test set is the 'full' suite (default), not just the 'quick' group/suite.

The fail set is flaky (some tests fail/pass inconsistently), so the 'consistently fail set' (i.e., fail every time) across 10-20 runs has been compared between the original/patched kernels.

Also looked for increased likelyhood to fail (i.e., flaky tests that fail more often with the patches.)

More Details:
---

The xfstests output/log includes the fail set in the 'Failures:' line, and the kernel version in the 'PLATFORM' line.
For example:

 $ grep -e ^Fail -e ^PLATFORM xfstests.log.2020-10-27-13-26-44
 PLATFORM -- Linux/s390x mfo-s390x-focal 5.4.0-52-generic #57-Ubuntu SMP Thu Oct 15 10:52:40 UTC 2020
 Failures: btrfs/153 btrfs/200 btrfs/204 btrfs/205 btrfs/213 btrfs/220 btrfs/221 btrfs/222 btrfs/226 generic/166 generic/175 generic/260 generic/286 generic/301 generic/465 generic/484 generic/610

Those two lines from each log/run are used to identify, per kernel version (original/patched):
- how many runs have been done;
- which tests fail consistently (i.e., in all runs);
- which tests fail inconsistently (i.e., flaky) and how many times (to compare likelyhood to fail)