1.3rc1: TestDisplayWorldCoordinate.test_cube_coords fails on MIPS

[olebole]

On MIPS platforms, the TestDisplayWorldCoordinate.test_cube_coords test fails during the Debian build:

Full Python Version: 
2.7.13rc1 (default, Dec  4 2016, 14:12:39) 
[GCC 6.2.1 20161124]

encodings: sys: ascii, locale: ANSI_X3.4-1968, filesystem: ANSI_X3.4-1968, unicode bits: 20
byteorder: big
float info: dig: 15, mant_dig: 15

Numpy: 1.11.2
Scipy: 0.18.1
Matplotlib: 1.5.3
h5py: not available
Pandas: not available
Cython: 0.25.2b0
Using Astropy options: remote_data.

[...]
self = <astropy.visualization.wcsaxes.tests.test_display_world_coordinates.TestDisplayWorldCoordinate object at 0x6c260930>
tmpdir = local('/tmp/pytest-of-buildd/pytest-0/test_cube_coords0')

    def test_cube_coords(self, tmpdir):
        wcs = WCS(self.cube_header)
    
        fig = plt.figure(figsize=(4, 4))
        canvas = fig.canvas
    
        ax = WCSAxes(fig, [0.1, 0.1, 0.8, 0.8], wcs=wcs, slices=('y', 50, 'x'))
        fig.add_axes(ax)
    
        # On some systems, fig.canvas.draw is not enough to force a draw, so we
        # save to a temporary file.
>       fig.savefig(tmpdir.join('test.png').strpath)

astropy/visualization/wcsaxes/tests/test_display_world_coordinates.py:107: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/usr/lib/python2.7/dist-packages/matplotlib/figure.py:1563: in savefig
    self.canvas.print_figure(*args, **kwargs)
/usr/lib/python2.7/dist-packages/matplotlib/backend_bases.py:2232: in print_figure
    **kwargs)
/usr/lib/python2.7/dist-packages/matplotlib/backends/backend_agg.py:527: in print_png
    FigureCanvasAgg.draw(self)
/usr/lib/python2.7/dist-packages/matplotlib/backends/backend_agg.py:474: in draw
    self.figure.draw(self.renderer)
/usr/lib/python2.7/dist-packages/matplotlib/artist.py:62: in draw_wrapper
    draw(artist, renderer, *args, **kwargs)
/usr/lib/python2.7/dist-packages/matplotlib/figure.py:1159: in draw
    func(*args)
astropy/visualization/wcsaxes/core.py:335: in draw
    ticklabels_bbox=self._ticklabels_bbox)
astropy/visualization/wcsaxes/coordinate_helpers.py:434: in _draw
    self._update_ticks()
astropy/visualization/wcsaxes/coordinate_helpers.py:486: in _update_ticks
    tick_world_coordinates, self._fl_spacing = self.locator(*coord_range[self.coord_index])
astropy/visualization/wcsaxes/formatter_locator.py:268: in locator
    values = self._locate_values(value_min, value_max, spacing_deg)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

value_min = 30.621389143656998, value_max = 30.621389143656998, spacing = 0.0

    @staticmethod
    def _locate_values(value_min, value_max, spacing):
        imin = np.ceil(value_min / spacing)
        imax = np.floor(value_max / spacing)
>       values = np.arange(imin, imax + 1, dtype=int)
E       ValueError: array is too big; `arr.size * arr.dtype.itemsize` is larger than the maximum possible size.

astropy/visualization/wcsaxes/formatter_locator.py:99: ValueError

Full log here. The same happens on mip64el and sparc64. Other platforms (arm64, powerpc etc.) are fine.
This almost looks like a matplotlib problem; so I will disable this test on the Debian build.

[astrofrog]

Thanks for the report - we should probably still fix _locate_values to give a proper error if spacing is 0

[olebole]

I'd revert my assumption that it is a matplotlib problem.

Quickly checking though the code, I did't see a place where spacing could be set to 0. -- in this specific case (value_max == value_min; is this correct??) it should be the base_spacing.

(BTW, the marked range doesn't have fallback if both self.spacing and self.number are None, which is not generally excluded by the code).

[bsipocz]

@olebole - would you please check whether it's still an issue for 2.0rc1?

[olebole]

I am just preparing to test this and the others. I'll report back on success.

[olebole]

I can confirm that this works now (2.0rc1) for MIPS. Closing.