commit fd62fcdf9626bf86c1b152a83e969152c3b47554
author Edwin Kempin <ekempin@google.com> Fri Jul 16 17:07:02 2021 +0200
committer Edwin Kempin <ekempin@google.com> Fri Jul 16 17:16:43 2021 +0200
tree 2a7b8c507035de43a6dca40c59e1dc60daf50e8d
parent 60d92c8f1dddbc23f33ed3bb4e1a578234dd28ed

GetCodeOwnersForPathInChange: Use a hash of the change number as a seed

If code owners have the same score they should be returned in a random
order, however for one change the same random order should be
consistently applied across requests. This is achieved by using the
change number as a seed for the randomisation. However it turned out
that with java.util.Random nearby seeds are prone to produce the same
random numbers. This is bad for change series, where the change IDs only
differ by 1, so all seeds are nearby and hence always the same code
owners will be suggested for all changes in the series. To avoid this we
are now using a hash of the change ID as a seed for the randomisation
(instead of the change ID itself). This way the seeds that are used for
a change series are ensured to be distant and hence the randomisation
should work better now.

While we are fixing this also add a test that ensures that using
different seeds produces a different order at least once. FWIW this test
was also succeeding with the old code.

Signed-off-by: Edwin Kempin <ekempin@google.com>
Change-Id: Idd038765f5e512280b4804d23e32a63b15b73573