On May 14, 2020 7:28 PM, Junio C Hamano wrote: > An early preview release Git v2.27.0-rc0 is now available for > testing at the usual places. It is comprised of 479 non-merge > commits since v2.26.0, contributed by 53 people, 16 of which are > new faces. All tests passed except, t9001 and t9020, which is normal. t0301 subtest 8 (1x), 11 (3x), 12 (1x), 26 (2x), still randomly fail when run 10 times, but I can't reproduce any of the failures 100% of the time. Number of failures of each is listed. For 26, the removal of a socket is not instantaneous on a MPP system, so this may be partly at fault. It can take a few microseconds, which is less than the time it takes to execute that subtest from end-to-end. If the system is heavily loaded, the results are different. Checking that a file that was created is no longer there, unless you are doing an explicit rm in the same script, is not a good plan in this platform. It looks like that 'git credential-cache exit' may be happening in a different logical CPU from the test script so it takes non-zero time for the file system on the script's logical CPU to see the removal. If I stick a sleep 5 (to be paranoid) into subtest 26, the subtest becomes a stable pass. Regards, Randall
