col_wrap arg to da.plot() can be a string?

[dschneiderch]

What happened:
if you try to plot a data array with 3 dimensions but the third dim has length 1, then col_wrap accepts a dim name and plots facets with pcolormesh. according to the documentation col_wrap argument should be an int so I assume this is unintentional http://xarray.pydata.org/en/stable/generated/xarray.DataArray.plot.html

if dim 3 has length >1, then col_wrap still takes a str but uses hist()

What you expected to happen:
I expected an error if col_wrap is not an int. I also expected da.plot(col='3rd_dim') to give me one facet if the dim has length 1 - this (mis)behavior is described in #620

all a bit confusing :). I thought I understood plotting but then ran into bug #620 and started fumbling around. I am trying to plot a dataarray in a function but I don't know ahead of time how many values 3rd dim has. I also wanted to avoid limiting my function to pcolormesh() only. da.plot.pcolormesh(col='3rd_dim') works to plot the images regardless of dim length.

Minimal Complete Verifiable Example:

da_list = []
for i in np.arange(1,3):
    data = np.ones((64, 48), dtype = 'uint8') * (200-i*15)
    oneframe = xr.DataArray(
        data=data[..., None],
        dims=('x', 'y','measurement'),
        coords={'measurement':[f't{i*40}']}
    )

    da_list.append(oneframe)

twoframes = xr.concat(da_list, 'measurement')

oneframe.plot(col_wrap='measurement') # Oddly, facets with pcolormesh()
twoframes.plot(col_wrap='measurement') # Oddly, a single facet with hist()
oneframe.plot(col='measurement') # ValueError: IndexVariable objects must be 1-dimensional ala issue #620
twoframes.plot(col='measurement') # facets with pcolormesh()
oneframe.plot.pcolormesh(col='measurement') # facets with pcolormesh()
twoframes.plot.pcolormesh(col='measurement') #facets with pcolormesh()

Anything else we need to know?:
@nfahlgren noted a workaround using FacetGrid directly

from xarray.plot.facetgrid import FacetGrid
gd = FacetGrid(data=oneframe, col="measurement")
gd.map_dataarray(xr.plot.imshow, "x", "y")

is consistent with:

gd = FacetGrid(data=twoframes, col="measurement")
gd.map_dataarray(xr.plot.imshow, "x", "y")

Environment:
Windows 10, python 3.7

Output of xr.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.7.10 | packaged by conda-forge | (default, Feb 19 2021, 15:37:01) [MSC v.1916 64 bit (AMD64)]
python-bits: 64
OS: Windows
OS-release: 10
machine: AMD64
processor: Intel64 Family 6 Model 158 Stepping 10, GenuineIntel
byteorder: little
LC_ALL: None
LANG: None
LOCALE: None.None
libhdf5: None
libnetcdf: None

xarray: 0.17.0
pandas: 1.2.3
numpy: 1.20.1
scipy: 1.6.0
netCDF4: None
pydap: None
h5netcdf: None
h5py: None
Nio: None
zarr: 2.7.0
cftime: None
nc_time_axis: None
PseudoNetCDF: None
rasterio: None
cfgrib: None
iris: None
bottleneck: None
dask: 2021.02.0
distributed: 2021.02.0
matplotlib: 3.3.4
cartopy: None
seaborn: None
numbagg: None
pint: None
setuptools: 49.6.0.post20210108
pip: 21.1.2
conda: 4.10.1
pytest: 6.2.2
IPython: 7.24.1
sphinx: None

[dcherian]

col_wrap should be an int. So we should raise an error there if col_wrap is not None and (col is None or row is None) as you say. And another if not isinstance(col_wrap, int). Pull requests are very welcome!

We will have to go through a deprecation cycle to fix #620

[dschneiderch]

small correction above, this first 2 cases are not being facetted and I think col_wrap is just being ignored.
oneframe.plot(col_wrap='measurement') outputs:
<matplotlib.collections.QuadMesh at 0x16ea840f7c8>
compared with
oneframe.plot.pcolormesh(col='measurement') outputs:
<xarray.plot.facetgrid.FacetGrid at 0x1b26a1f5ec8>

i'll see if i can make sense of the code base and submit a PR.