On Wed, 9 Jun 2021 10:33:34 GMT, Maurizio Cimadamore <mcimadam...@openjdk.org> wrote:
> This tests fails intermittently but very rarely, on Windows it seems. The > subtest in question is `testLockSharedMultiThread`. This test is spawning a > number of threads, each of which: > > * increments a counter atomically > * acquire scope > * waits some time > * release scope > * decrement counter (in finally block) > > The main test thread tries (while the counter value is > 0) to repeatedly > close the scope, and asserts that if the scope closes successfully, then the > counter value should be exactly zero. > > Looking at the test closely, I realized there's an issue: the order in which > counter is incremented and scope is acquired is wrong. As such it is possible > for counter to be increased, but then for subsequent acquire to fail (e.g. if > segment is already closed). This would lead to the main test thread to fail, > as it would appear as if the segment has been closed successfully, but the > counter value is not zero. > > The fix is to only increment the counter after a successful acquire. I have > also tweaked the test so that it keeps trying closing the scope, even if the > counter value reaches zero permanently (e.g. all other threads have > completed). I agree with the analysis and test code changes. Trivially you should also update the copyright years on this file. ------------- Marked as reviewed by dfuchs (Reviewer). PR: https://git.openjdk.java.net/jdk/pull/4427
