Bug#1092867: pyglet: Autopkgtests uses timing tests that fails on s390x (flaky?)

[Niels Thykier]

Package: pyglet
Version: 2.0.17+ds-2
Severity: serious
X-Debbugs-Cc: ni...@thykier.net
Control: affects -1 src:pillow

Hi,

The pyglet autopkgtests fails on s390x due to an assumption on the precision about wake ups from sleep vs. time elapsed clock.time() between such sleep calls. This intend turn prevents unrelated packages from migrating. Concretely, pillow is blocked by such a failure.

112s =================================== FAILURES 
===================================
112s ________________________ test_elapsed_time_between_tick 
________________________

112s     def test_elapsed_time_between_tick():
112s         """
112s         Test that the tick function returns the correct elapsed
112s         time between frames, in seconds.

112s         expect a small error (1%) from the expected value.
112s         """
112s         sleep_time = 0.2

112s         clock.tick()

112s         sleep(sleep_time)
112s         delta_time_1 = clock.tick()

112s         sleep(sleep_time)
112s         delta_time_2 = clock.tick()

112s E       assert 0.2045531269395724 == 0.2 ± 4.0e-04

112s E         Obtained: 0.2045531269395724
112s E         Expected: 0.2 ± 4.0e-04

112s =========================== short test summary info 
============================
112s FAILED tests/unit/test_clock_freq.py::test_elapsed_time_between_tick - 
assert...
112s ==== 1 failed, 610 passed, 106 skipped, 93 deselected, 2 xfailed in 28.04s 
=====

112s 112s @skip_if_continuous_integration() 112s 112s Because we are measuring time differences, we 112s 112s # initialize internal counter 112s 112s # test between initialization and first tick 112s 112s # test between non-initialization tick and next tick 112s 112s > assert delta_time_1 == pytest.approx(sleep_time, rel=0.01*sleep_time) 112s E 112s E comparison failed 112s 112s tests/unit/test_clock_freq.py:64: AssertionError

Some options include:
 * Skipping the concrete test
 * Changing the limits to make it less likely (though at some point,
   the extra leeway will invalidate the test)
 * Mark the autopkgtests as flaky, which will make the test less likely
   in catching regression.

Best regards,
Niels

OpenPGP_signature.asc
Description: OpenPGP digital signature