Port JSR166 LinkedTransferQueueTest flakiness fix
This CL cherry-picks the following upstream (*) test changes ("upstream"
here is JSR166's CVS repository
cvs -d ':pserver:anonymous@gee.cs.oswego.edu/home/jsr166/jsr166'
This is in contrast to the code under test, whose upstream is OpenJDK).
=====
src/test/tck/LinkedTransferQueueTest.java
revision 1.70
date: 2017/02/18 16:37:49; author: jsr166; state: Exp; lines: +11 -5
waitForThreadToEnterWaitState should fail on timeout; tests should tolerate transient blocking Thread.State at any time (e.g. due to classloading)
src/test/tck/JSR166TestCase.java
revision 1.219
date: 2017/02/18 16:37:49; author: jsr166; state: Exp; lines: +43 -3
waitForThreadToEnterWaitState should fail on timeout; tests should tolerate transient blocking Thread.State at any time (e.g. due to classloading)
src/test/tck/PhaserTest.java
revision 1.45
date: 2017/02/18 15:05:55; author: jsr166; state: Exp; lines: +10 -10
use default timeout of LONG_DELAY_MS with
waitForThreadToEnterWaitState
======
The effect of this CL is to fix flakiness in LinkedTransferQueueTest's
testTransfer2() and testWaitingConsumer().
These test methods need to wait until a background thread was blocked
waiting on a LinkedTransferQueue; before this CL, they did so by
calling a helper method that waited for the background thread's state
to become WAITING, TIMED_WAITING, or BLOCKED. After this CL, they also
check that the LinkedTransferQueue is in the expected state before
they stop waiting.
The additional check is necessary because LinkedTransferQueue uses
LockSupport.park(), which on Android involves a synchronized lock
and therefore BLOCKED state. This caused the test methods to
prematurely stop waiting, which caused the test to fail.
The concrete failure could also have been prevented by waiting for
the thread to enter WAITING or TIMED_WAITING state rather than BLOCKED.
This was considered but not chosen by upstream since it would still
have made assumptions about those states not being entered elsewhere.
This CL also changes the waiting logic to fail() the test upon
timeout. PhaserTest was changed to use longer timeouts (10sec
rather than 250msec) to avoid flaky timeout failures.
Bug: 34814528
Test: LinkedTransferQueueTest
Test: PhaserTest
Test: Checked that testTransfer2() and testWaitingConsumer() are
non-flaky by running their body 10,000 times each in a loop.
Change-Id: I112ee1fba8aea6ca97ca0f99bba0fc9f00e5c0c2
3 files changed