Allow zero sized ticks #7959

scopatz · Jan 27, 2017

This addresses an issue with savefig() when there are no ticks present. In axis.py,

    def get_tick_space(self):
        ends = self.axes.transAxes.transform([[0, 0], [1, 0]])
        length = ((ends[1][0] - ends[0][0]) / self.axes.figure.dpi) * 72.0
        tick = self._get_tick(True)
        # There is a heuristic here that the aspect ratio of tick text
        # is no more than 3:1
        size = tick.label1.get_size() * 3
       return int(np.floor(length / size))

if size == 0, then the this will be effectively, int(np.floor(np.infty)) which fails on integer conversion with,

>>> int(np.floor(np.infty))
OverflowError: cannot convert float infinity to integer

Since this is directly used in ticker.py in _raw_ticks.py in a minimization, it is sufficient to just return a 'big' number. The full traceback we are seeing in the xonsh test suite is,

__________________________________ test_mpl_preserve_image_tight ___________________________________

    def test_mpl_preserve_image_tight():
        """Make sure that the figure preserves height settings"""
        f = create_figure()
        exp = mplhooks.figure_to_rgb_array(f)
        width, height = f.canvas.get_width_height()
>       s = mplhooks.figure_to_tight_array(f, 0.5*width, 0.5*height, True)

test_mpl.py:94: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
../xontrib/mplhooks.py:105: in figure_to_tight_array
    array = figure_to_rgb_array(fig, shape=(height, width, 4))
../xontrib/mplhooks.py:48: in figure_to_rgb_array
    array = np.frombuffer(_get_buffer(fig, dpi=fig.dpi, format='raw').read(), dtype='uint8')
../xontrib/mplhooks.py:24: in _get_buffer
    fig.savefig(b, **kwargs)
../../miniconda/lib/python3.5/site-packages/matplotlib/figure.py:1572: in savefig
    self.canvas.print_figure(*args, **kwargs)
../../miniconda/lib/python3.5/site-packages/matplotlib/backends/backend_qt5agg.py:222: in print_figure
    FigureCanvasAgg.print_figure(self, *args, **kwargs)
../../miniconda/lib/python3.5/site-packages/matplotlib/backend_bases.py:2244: in print_figure
    **kwargs)
../../miniconda/lib/python3.5/site-packages/matplotlib/backends/backend_agg.py:526: in print_raw
    FigureCanvasAgg.draw(self)
../../miniconda/lib/python3.5/site-packages/matplotlib/backends/backend_agg.py:464: in draw
    self.figure.draw(self.renderer)
../../miniconda/lib/python3.5/site-packages/matplotlib/artist.py:63: in draw_wrapper
    draw(artist, renderer, *args, **kwargs)
../../miniconda/lib/python3.5/site-packages/matplotlib/figure.py:1143: in draw
    renderer, self, dsu, self.suppressComposite)
../../miniconda/lib/python3.5/site-packages/matplotlib/image.py:139: in _draw_list_compositing_images
    a.draw(renderer)
../../miniconda/lib/python3.5/site-packages/matplotlib/artist.py:63: in draw_wrapper
    draw(artist, renderer, *args, **kwargs)
../../miniconda/lib/python3.5/site-packages/matplotlib/axes/_base.py:2409: in draw
    mimage._draw_list_compositing_images(renderer, self, dsu)
../../miniconda/lib/python3.5/site-packages/matplotlib/image.py:139: in _draw_list_compositing_images
    a.draw(renderer)
../../miniconda/lib/python3.5/site-packages/matplotlib/artist.py:63: in draw_wrapper
    draw(artist, renderer, *args, **kwargs)
../../miniconda/lib/python3.5/site-packages/matplotlib/axis.py:1136: in draw
    ticks_to_draw = self._update_ticks(renderer)
../../miniconda/lib/python3.5/site-packages/matplotlib/axis.py:969: in _update_ticks
    tick_tups = [t for t in self.iter_ticks()]
../../miniconda/lib/python3.5/site-packages/matplotlib/axis.py:969: in <listcomp>
    tick_tups = [t for t in self.iter_ticks()]
../../miniconda/lib/python3.5/site-packages/matplotlib/axis.py:912: in iter_ticks
    majorLocs = self.major.locator()
../../miniconda/lib/python3.5/site-packages/matplotlib/ticker.py:1794: in __call__
    return self.tick_values(vmin, vmax)
../../miniconda/lib/python3.5/site-packages/matplotlib/ticker.py:1802: in tick_values
    locs = self._raw_ticks(vmin, vmax)
../../miniconda/lib/python3.5/site-packages/matplotlib/ticker.py:1744: in _raw_ticks
    nbins = max(min(self.axis.get_tick_space(), 9),
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <matplotlib.axis.XAxis object at 0x7fe9effe7390>

    def get_tick_space(self):
        ends = self.axes.transAxes.transform([[0, 0], [1, 0]])
        length = ((ends[1][0] - ends[0][0]) / self.axes.figure.dpi) * 72.0
        tick = self._get_tick(True)
        # There is a heuristic here that the aspect ratio of tick text
        # is no more than 3:1
        size = tick.label1.get_size() * 3
>       return int(np.floor(length / size))
E       OverflowError: cannot convert float infinity to integer

../../miniconda/lib/python3.5/site-packages/matplotlib/axis.py:2024: OverflowError

story645 · Jan 27, 2017

Can you please add (py.test) tests?

scopatz · Jan 27, 2017

can you please point me where there are similar tests? I know nothing about testing mpl these days. thanks!

story645 · Jan 27, 2017

lib/matplotlib/axis.py

-        return int(np.floor(length / size))
+        if size > 0:
+            return int(np.floor(length / size))
+        else:


doesn't need the else. Also, I'd invert this and flag the special case

if size <= 0: return return 2**31 - 1

For what it is worth, keeping the else clause is about 10% faster. The order shouldn't matter and doesn't affect timings that I have noticed.

scopatz@artemis ~ $ def f(x): `·.,¸,.·*¯`·.,¸,.·* if x: `·.,¸,.·*¯`·.,¸,.·* return 42 `·.,¸,.·*¯`·.,¸,.·* else: `·.,¸,.·*¯`·.,¸,.·* return 0 `·.,¸,.·*¯`·.,¸,.·* `·.,¸,.·*¯`·.,¸,.·* scopatz@artemis ~ $ def g(x): `·.,¸,.·*¯`·.,¸,.·* if x: `·.,¸,.·*¯`·.,¸,.·* return 42 `·.,¸,.·*¯`·.,¸,.·* return 0 `·.,¸,.·*¯`·.,¸,.·* scopatz@artemis ~ $ timeit! f(True) 10000000 loops, best of 3: 57.6 ns per loop scopatz@artemis ~ $ timeit! g(True) 10000000 loops, best of 3: 63.6 ns per loop scopatz@artemis ~ $ timeit! f(False) 10000000 loops, best of 3: 58 ns per loop scopatz@artemis ~ $ timeit! g(False) 10000000 loops, best of 3: 64 ns per loop scopatz@artemis ~ $ 1 - (58.0/64) 0.09375

I know order doesn't matter, I just think from a clarity/maintance point of view the special case should be in the if.

Though now I'm wondering if there's a way to modify int(np.floor(length/size)) such that it'll catch the size<=0 case

It is probably a bit of apples and oranges since I find the way I wrote it clearest. I looked around for something in numpy that would cast infinities to integers better, but I didn't come up with anything.

I also find the if else statement clearer.

shrugs wasn't gonna put up a fight

story645 · Jan 27, 2017

lib/matplotlib/axis.py

-        return int(np.floor(length / size))
+        if size > 0:
+            return int(np.floor(length / size))
+        else:


same, don't need else

story645 · Jan 27, 2017

:/ ugh, unfortunately there are no tests for axis, so at least as a first pass maybe test that savefig now behaves as expected. Savefig tests are in test_backend_ps.py but the fix should maybe be in base if it applies everywhere.

NelleV · Jan 27, 2017

Hi @scopatz
The patch looks good.
As for testing, I think a simple test that shows it doesn't crash would be ok, so just add the code you used to illustrate the bug to test_axes.py in a test function, and it should be fine. No need to worry about pytest in this particular cas.

scopatz · Jan 27, 2017

Thanks @NelleV! I have added a minimal test that reproduces the issue. Let me know if this needs anything else.

story645 · Jan 27, 2017

lib/matplotlib/tests/test_axes.py

@@ -2341,6 +2341,16 @@ def test_manage_xticks():
    assert_array_equal(old_xlim, new_xlim)


+@cleanup
+def test_size0_ticks():


can you name this test_tick_space_size_0 so it refers to the function being tested?

:) LGTM, just waiting for the tests to clear

NelleV

Thanks @scopatz !

dstansby · Jan 29, 2017

👍 Looks good, thanks @scopatz

scopatz · Jan 29, 2017

Thanks All!

Allow zero sized ticks

NelleV · Jan 29, 2017

backported to v2.0.x as 729d219

dstansby · Jan 29, 2017

Woops, sorry for missing the backport and thanks for picking it up.

Allow zero sized ticks

874ccdd

scopatz mentioned this pull request Jan 27, 2017

pin mpl to 1.5.3 for tests xonsh/xonsh#2114

Merged

story645 reviewed Jan 27, 2017

View reviewed changes

tacaswell added this to the 2.0.1 (next bug fix release) milestone Jan 27, 2017

tacaswell added the Release critical For bugs that make the library unusable (segfaults, incorrect plots, etc) and major regressions. label Jan 27, 2017

zero-sized tick test

1006c08

story645 reviewed Jan 27, 2017

View reviewed changes

test rename

7df6ef2

NelleV approved these changes Jan 27, 2017

View reviewed changes

NelleV changed the title ~~Allow zero sized ticks~~ [MRG+1] Allow zero sized ticks Jan 27, 2017

story645 closed this Jan 27, 2017

story645 reopened this Jan 27, 2017

dstansby approved these changes Jan 29, 2017

View reviewed changes

dstansby merged commit 0799250 into matplotlib:master Jan 29, 2017

dstansby changed the title ~~[MRG+1] Allow zero sized ticks~~ Allow zero sized ticks Jan 29, 2017

scopatz deleted the size0tick branch January 29, 2017 20:16

NelleV pushed a commit that referenced this pull request Jan 29, 2017

Merge pull request #7959 from scopatz/size0tick

729d219

Allow zero sized ticks

tacaswell mentioned this pull request Feb 23, 2017

If division by zero, python cannot convert float infinity to integer in get_tick_space() #8136

Closed

Search code, repositories, users, issues, pull requests...

Uh oh!

Allow zero sized ticks #7959

Allow zero sized ticks #7959

Uh oh!

Conversation

scopatz commented Jan 27, 2017

Uh oh!

story645 commented Jan 27, 2017

Uh oh!

scopatz commented Jan 27, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

story645 Jan 27, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

story645 Jan 27, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

story645 commented Jan 27, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

NelleV commented Jan 27, 2017

Uh oh!

scopatz commented Jan 27, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

NelleV left a comment

Choose a reason for hiding this comment

Uh oh!

dstansby commented Jan 29, 2017

Uh oh!

scopatz commented Jan 29, 2017

Uh oh!

NelleV commented Jan 29, 2017

Uh oh!

dstansby commented Jan 29, 2017

Uh oh!

Uh oh!

story645 Jan 27, 2017 •

edited

Loading

story645 Jan 27, 2017 •

edited

Loading

story645 commented Jan 27, 2017 •

edited

Loading