contourf on cylindrical projection with irregular latlon grid (tri=True) is calling shiftdata when it shouldn't

从 lib/mpl_toolkits/basemap/init.py 开始，沿着 contourf 包装器跟踪到 shiftdata 以及 traceback 中显示的 itemindex 检查。使用圆柱投影、不规则 1D 数组、tri=True 和 latlon=True 重现该示例，然后与正射投影进行比较；完成的标准是圆柱投影情况下不再引发 ambiguous-array ValueError，同时规则网格的行为保持不变。

I ran into this problem while trying to plot interpolated data NOT on a regular latlon grid. Orthographic projection works but cylindrical does not.

Relevant code:

fig = plt.figure()
ax = fig.add_subplot(111)
levels = 20
m = Basemap(projection="cyl")  # causes ValueError
# m = Basemap(projection="ortho", lat_0=0., lon_0=0., resolution='l')  # works, does not cause ValueError
MC = m.contourf(lons_deg, lats_deg, vals, levels, ax=ax, tri=True, latlon=True)  # latlon=True interprets first two args as lon and lat respectively

Note that lats_deg, lons_deg, and vals are 1d np arrays with irregular point spacing (no meshgrid). I used tri=True to allow interpolation. This works fine on orthographic projection. But with cylindrical I get the following error:

Traceback (most recent call last):
  File "IcosahedralGeodesicLattice.py", line 146, in <module>
    test_lattice.plot_data(data)
  File "/home/wesley/programming/Mapping/Lattice.py", line 86, in plot_data
    MC = m.contourf(lons_deg, lats_deg, vals, levels, ax=ax, tri=True, latlon=True)  # latlon=True interprets first two args as LON and LAT RESPECTIVELY
  File "/home/wesley/.local/lib/python3.5/site-packages/mpl_toolkits/basemap/__init__.py", line 544, in with_transform
    fix_wrap_around=plotfunc.__name__ not in ["scatter"])
  File "/home/wesley/.local/lib/python3.5/site-packages/mpl_toolkits/basemap/__init__.py", line 4797, in shiftdata
    if fix_wrap_around and itemindex:
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

Looking at the relevant file, the docstring for shiftdata says "Only valid for cylindrical/pseudo-cylindrical global projections and data on regular lat/lon grids. longitudes and data can be 1-d or 2-d, if 2-d it is assumed longitudes are 2nd (rightmost) dimension."

So should it not be being called with tri=True? Sorry if I have misunderstood something, let me know. Thanks!