How does this relate (if at all) with #5911?

On Mon, Feb 1, 2016 at 4:28 PM, Michael Droettboom <notifications@github.com

wrote:

The issue is that the clipping algorithm may create a moveto command if
the polygon is clipped outside of the image. A subsequent "close poly"
command will then close the polygon to that last moveto command, rather
than the original moveto location.

I'm not sure there's a good solution to this: One needs the closepoly
command to actually close the poly (if you just to a lineto to the original
point, the line isn't "joined" at the corner properly... not getting that
detail right has been the source of other bugs in the past.) The only
correct solution I can think of is to "rotate" the order of the vertices so
that the closepoly always ends up outside of the clipping box. But that
just sounds hard to do efficiently.

In the meantime, I think the best fix is to just turn off clipping for all
paths that have a closepoly on the grounds that large polygons are less
likely to require clipping anyway vs. the many-vertex line plots that are

quite common.

You can view, comment on, or merge this pull request online at:

#5955
Commit Summary

• Fix tilted line visible in generated pdf file #5948: Don't clip closed paths

File Changes

• M src/_backend_agg.h
  https://github.com/matplotlib/matplotlib/pull/5955/files#diff-0 (4)
• M src/_path.h
  https://github.com/matplotlib/matplotlib/pull/5955/files#diff-1 (6)
• M src/path_cleanup.cpp
  https://github.com/matplotlib/matplotlib/pull/5955/files#diff-2 (2)

Patch Links:

• https://github.com/matplotlib/matplotlib/pull/5955.patch
• https://github.com/matplotlib/matplotlib/pull/5955.diff

—
Reply to this email directly or view it on GitHub
#5955.

Failures look like

======================================================================
ERROR: matplotlib.tests.test_transforms.test_clipping_of_log
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/travis/build/matplotlib/matplotlib/venv/lib/python3.5/site-packages/nose/case.py", line 198, in runTest
    self.test(*self.arg)
  File "/home/travis/build/matplotlib/matplotlib/venv/lib/python3.5/site-packages/matplotlib-1.5.1+40.ga952891-py3.5-linux-x86_64.egg/matplotlib/tests/test_transforms.py", line 214, in test_clipping_of_log
    assert np.allclose(tcodes, [M, M, L, L, L, C])
  File "/home/travis/build/matplotlib/matplotlib/venv/lib/python3.5/site-packages/numpy/core/numeric.py", line 2278, in allclose
    res = all(isclose(a, b, rtol=rtol, atol=atol, equal_nan=equal_nan))
  File "/home/travis/build/matplotlib/matplotlib/venv/lib/python3.5/site-packages/numpy/core/numeric.py", line 2359, in isclose
    return within_tol(x, y, atol, rtol)
  File "/home/travis/build/matplotlib/matplotlib/venv/lib/python3.5/site-packages/numpy/core/numeric.py", line 2342, in within_tol
    result = less_equal(abs(x-y), atol + rtol * abs(y))
ValueError: operands could not be broadcast together with shapes (5,) (6,) 
----------------------------------------------------------------------

In this case, I think we update the test...

Isn't that test also modified in #5911?

On Mon, Feb 1, 2016 at 5:44 PM, Michael Droettboom <notifications@github.com

wrote:

In this case, I think we update the test...

—
Reply to this email directly or view it on GitHub
#5955 (comment)
.

How does this relate (if at all) with #5911?

#5911 fixes the problem that there was always a MOVETO command emitted at the beginning, even if that MOVETO command was out of bounds. This deals with not inserting MOVETO commands in the middle of a closed shape -- by bluntly just turning off clipping for closed paths.

The test changed to be more correct (coincidentally) in both cases. One by not having two MOVETO commands at the beginning, and the other by not running clipping at all on the path. So, in a sense, the broken test is now testing something really easy (copying values from one place to another), but it's probably worth keeping anyway.

@tacaswell is definitely related I'll pull your latest code and check later on. Thanks for the reference.