2d plots may fail for some choices of `x` and `y`

[johnomotani]

What happened:
When making a 2d plot with a 1d x argument and a 2d y, if the two dimensions have the same size and are in the wrong order, no plot is produced - the third plot in the MCVE is blank.

What you expected to happen:
All three plots in the MCVE should be identical.

Minimal Complete Verifiable Example:

from matplotlib import pyplot as plt
import numpy as np
import xarray as xr

ds = xr.Dataset({"z": (["x", "y"], np.random.rand(4,4))})

x2d, y2d = np.meshgrid(ds["x"], ds["y"])

ds = ds.assign_coords(x2d=(["x", "y"], x2d.T), y2d=(["x", "y"], y2d.T))

fig, axes = plt.subplots(1,3)

h0 = ds["z"].plot.pcolormesh(x="y2d", y="x2d", ax=axes[0])
h1 = ds["z"].plot.pcolormesh(x="y", y="x", ax=axes[1])
h2 = ds["z"].plot.pcolormesh(x="y", y="x2d", ax=axes[2])

plt.show()

result:

Anything else we need to know?:

The bug is present in both the 0.17.0 release and current master.

I came across this while starting to work on #5084. I think the problem is here

xarray/xarray/plot/plot.py

Lines 678 to 684 in ddc352f

	# check if we need to broadcast one dimension
	if xval.ndim < yval.ndim:
	dims = darray[ylab].dims
	if xval.shape[0] == yval.shape[0]:
	xval = np.broadcast_to(xval[:, np.newaxis], yval.shape)
	else:
	xval = np.broadcast_to(xval[np.newaxis, :], yval.shape)

as the check xval.shape[0] == yval.shape[0] doesn't work if the single dimension of x is actually the second dimension of y, but happened to have the same size as the first dimension of y? I think it needs to check the actual dimensions of x and y.

Why don't we just do something like

xval = xval.broadcast_like(darray)
yval = yval.broadcast_like(darray)

if either coordinate is 2d before using .values to convert to numpy arrays?

Environment:

Output of xr.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.8.6 | packaged by conda-forge | (default, Oct 7 2020, 19:08:05)
[GCC 7.5.0]
python-bits: 64
OS: Linux
OS-release: 5.4.0-70-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_GB.UTF-8
LOCALE: en_GB.UTF-8
libhdf5: 1.10.6
libnetcdf: 4.7.4

xarray: 0.17.0
pandas: 1.1.5
numpy: 1.19.4
scipy: 1.5.3
netCDF4: 1.5.5.1
pydap: None
h5netcdf: None
h5py: 3.1.0
Nio: None
zarr: None
cftime: 1.3.0
nc_time_axis: None
PseudoNetCDF: None
rasterio: None
cfgrib: None
iris: None
bottleneck: None
dask: 2020.12.0
distributed: 2020.12.0
matplotlib: 3.3.3
cartopy: None
seaborn: None
numbagg: None
pint: 0.16.1
setuptools: 49.6.0.post20201009
pip: 20.3.3
conda: 4.9.2
pytest: 6.2.1
IPython: 7.19.0
sphinx: 3.4.0

[dcherian]

👍 for making this change.

IIRC the "resolving intervals" section later will break and is somewhat annoying to fix. This is why the current ugly code exists.