Repository navigation
Travis CI: test_high_characters sometimes fails #2758
Description
I've seen it fail on some Docker build jobs. After rebuilding, it was fine.
Here's an example from DOCKER="debian-stretch-x86" from today at https://travis-ci.org/python-pillow/Pillow/jobs/279047420, which passed after rebuilding:
======================================================================
FAIL: TestFontPcf.test_high_characters
----------------------------------------------------------------------
Traceback (most recent call last):
File "/Pillow/Tests/test_font_pcf.py", line 59, in test_high_characters
self._test_high_characters(message)
File "/Pillow/Tests/test_font_pcf.py", line 55, in _test_high_characters
self.assert_image_equal(image, compare)
File "/Pillow/Tests/helper.py", line 85, in assert_image_equal
msg or "got size %r, expected %r" % (a.size, b.size))
AssertionError: got size (783, 22), expected (765, 22)
-------------------- >> begin captured logging << --------------------
PIL.PngImagePlugin: DEBUG: STREAM 'IHDR' 16 13
PIL.PngImagePlugin: DEBUG: STREAM 'IDAT' 41 2644
PIL.PngImagePlugin: DEBUG: STREAM 'IHDR' 16 13
PIL.PngImagePlugin: DEBUG: STREAM 'IDAT' 41 1400
--------------------- >> end captured logging << ---------------------
----------------------------------------------------------------------
Ran 1068 tests in 16.060s
FAILED (SKIP=25, failures=1)
Here's another from August (#2634 (comment)):
I'm not sure what's up with the Travis errors, as they're non-deterministic. This fails on debian-stretch in one commit or ubuntu-trusty-x86 in another, but the changes should not affect them.
======================================================================
FAIL: TestFontPcf.test_high_characters
----------------------------------------------------------------------
Traceback (most recent call last):
File "/Pillow/Tests/test_font_pcf.py", line 59, in test_high_characters
self._test_high_characters(message)
File "/Pillow/Tests/test_font_pcf.py", line 55, in _test_high_characters
self.assert_image_equal(image, compare)
File "/Pillow/Tests/helper.py", line 85, in assert_image_equal
msg or "got size %r, expected %r" % (a.size, b.size))
AssertionError: got size (770, 22), expected (765, 22)
(and when rerun, the ubuntu version passed, and the debian one still failed)
Activity
wiredfool commentedon Sep 24, 2017
I can't get them to do the same thing on my docker build machine. Leads one to believe that it's uninitialized memory or something similar.
hugovk commentedon Sep 28, 2017
Another fail on https://travis-ci.org/python-pillow/Pillow/jobs/280732802 passed when restarted. Here are full logs. Nothing much different in the setup.
ubuntu-trusty-x86-fail.log
ubuntu-trusty-x86-pass.log
wiredfool commentedon Sep 28, 2017
It's not deterministic, which sort of narrows down the possible roots of the problem. I've also never seen it happen on my docker build system, which makes me really suspicious.
(aside -- I thought that part of the point with docker was reproducibility -- If you run the same docker image, you should get the same results. Clearly it's not quite there)
hugovk commentedon Sep 28, 2017
Same numbers on debian-stretch-x86 from https://travis-ci.org/python-pillow/Pillow:
debian-stretch-x86-fail.log
debian-stretch-x86-pass.log
wiredfool commentedon Oct 2, 2017
2 of the runs to make linux wheels died on the 4.3.0 version bump, got a size of (769,22).
Include a null byte at the end of the string, length doesn't include it
Include a null byte at the end of the string, length doesn't include it
radarhere commentedon Nov 3, 2017
I'm finding now that the 'MB_PYTHON_VERSION=2.7 PLAT=i686' build on Pillow Wheels is failing continuously because of this issue - might be helpful in resolving this?
wiredfool commentedon Nov 3, 2017
I'm pretty sure that the patch in #2825 is the fix.
radarhere commentedon Nov 7, 2017
Yes, it fixes the problem. Thanks