You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am getting only one test failure for both Python 2.7.11 and 3.5.1 on Arch. Hope this helps :)
======================================================================
FAIL: matplotlib.tests.test_image.test_rasterize_dpi.test
----------------------------------------------------------------------
Traceback (most recent call last):
File "/usr/lib/python2.7/site-packages/nose/case.py", line 197, in runTest
self.test(*self.arg)
File "/build/python-matplotlib/src/matplotlib-1.5.1-py2/build/lib.linux-x86_64-2.7/matplotlib/testing/decorators.py", line 53, in failer
result = f(*args, **kwargs)
File "/build/python-matplotlib/src/matplotlib-1.5.1-py2/build/lib.linux-x86_64-2.7/matplotlib/testing/decorators.py", line 220, in do_test
'(RMS %(rms).3f)'%err)
ImageComparisonFailure: images not close: /build/python-matplotlib/src/tmp_test_dir/result_images/test_image/rasterize_10dpi_svg.png vs. /build/python-matplotlib/src/tmp_test_dir/result_images/test_image/rasterize_10dpi-expected_svg.png (RMS 5.661)
----------------------------------------------------------------------
Ran 5165 tests in 136.270s
FAILED (KNOWNFAIL=18, SKIP=1, failures=1)
Shape of x does not match that of z: found (9, 9) instead of (9, 10).
Just as another data point -- I can't reproduce in Python 2.7.11 or 3.5.1 (Conda on Fedora). Can someone who's hitting this post the SVG file so I can see how it's failing?
I think what we're running into is how Inkscape is handling image edge rendering -- which appears to be different in the version of Inkscape on Travis (which I think is 0.48.1), and that on our local machines (which is probably the latest 0.91). The old Inkscape will resample the image such that it "bleeds" outside of the image bounding box. This creates the same rendering before and after the matplotlib change. The new Inkscape seems to strictly crop the image to the image bounding box, creating a different rendering between the two images -- the old has bleed (since it wasn't cropped correctly) and the new one doesn't.
The bottom line is that I think it's safe to update the image to the latest generated in this case. There isn't actually a matplotlib bug here as far as I can tell -- the entire image content (and nothing more) is making it into the output SVG, which is exactly right.
Activity
tacaswell commentedon Jan 11, 2016
hmm, I saw the same thing locally in #5817 (I use arch 😄 ) but everything is passing on travis so I wrote it off as a local issue....
jenshnielsen commentedon Jan 11, 2016
I'm seeing it too. Not sure why Travis doesn't catch it
mdboom commentedon Jan 11, 2016
Just as another data point -- I can't reproduce in Python 2.7.11 or 3.5.1 (Conda on Fedora).
Can someone who's hitting this post the SVG file so I can see how it's failing?mdboom commentedon Jan 11, 2016
I'm able to reproduce it now -- I think I wasn't getting a full rebuild earlier.
tacaswell commentedon Jan 11, 2016
Sorry everyone 😞 .
mdboom commentedon Jan 11, 2016
I think what we're running into is how Inkscape is handling image edge rendering -- which appears to be different in the version of Inkscape on Travis (which I think is 0.48.1), and that on our local machines (which is probably the latest 0.91). The old Inkscape will resample the image such that it "bleeds" outside of the image bounding box. This creates the same rendering before and after the matplotlib change. The new Inkscape seems to strictly crop the image to the image bounding box, creating a different rendering between the two images -- the old has bleed (since it wasn't cropped correctly) and the new one doesn't.
The bottom line is that I think it's safe to update the image to the latest generated in this case. There isn't actually a matplotlib bug here as far as I can tell -- the entire image content (and nothing more) is making it into the output SVG, which is exactly right.
tacaswell commentedon Jan 11, 2016
Probably need to add a test on inkscape version too so we can skip on travis?
mdboom commentedon Jan 11, 2016
Hopefully by updating the baseline image, we won't be Inkscape-version-dependent anymore.
Fix matplotlib#5829. Update the baseline image.
Merge pull request #5834 from mdboom/fix-interpolation-10dpi
Fix matplotlib#5829. Update the baseline image.