In fixing some bugs with color bars (https://github.com/pydata/xarray/pull/3601), we had to do some clunky workarounds because of limitations of matplotlib's API for modifying colormaps - this prompted a matplotlib issue https://github.com/matplotlib/matplotlib/issues/16296#issuecomment-1629755861. That issue has now been closed, and apparently the limitations are now fixed, so it should be possible to tidy up some of the colorbar code (at least at some point, once the oldest xarray-supported matplotlib includes the new API).

I have no time to look into this myself, but opening this issue to flag the new matplotlib features in case someone is looking at refactoring colorbar or colormap code.

What happened: If a 0d dask array is passed as an argument to isel(), an error occurs because dask arrays do not have a .item() method. I came across this when trying to use the result of da.argmax() from a dask-backed array to select from the DataArray.

What you expected to happen: isel() returns the value at the index contained in the 0d dask array.

Minimal Complete Verifiable Example:

```python import dask.array as daskarray import numpy as np import xarray as xr

a = daskarray.from_array(np.linspace(0., 1.)) da = xr.DataArray(a, dims="x")

x_selector = da.argmax(dim=...)

da_max = da.isel(x_selector) ```

Anything else we need to know?:

I think the problem is here https://github.com/pydata/xarray/blob/a198218ddabe557adbb04311b3234ec8d20419e7/xarray/core/variable.py#L546-L548 and k.values.item() or int(k.data) would fix my issue, but I don't know the reason for using .item() in the first place, so I'm not sure if either of these would have some undesirable side-effect.

May be related to #2511, but from the code snippet above, I think this is a specific issue of 0d dask arrays rather than a generic dask-indexing issue like #2511.

I'd like to fix this because it breaks the nice new features of argmin() and argmax() if the DataArray is dask-backed.

Environment:

It seems like set_coords() doesn't create an index variable. Is there a reason for this? I was surprised that the following code snippets produce different Datasets (first one has empty indexes, second one has x in indexes), even though both Datasets have a 'dimension coordinate' x:

(1) ``` import numpy as np import xarray as xr

ds = xr.Dataset() ds['a'] = ('x', np.linspace(0,1)) ds['b'] = ('x', np.linspace(3,4)) ds = ds.rename(b='x') ds = ds.set_coords('x')

print(ds) print('indexes', ds.indexes) ```

(2) ``` import numpy as np import xarray as xr

ds = xr.Dataset() ds['a'] = ('x', np.linspace(0,1)) ds['x'] = ('x', np.linspace(3,4))

print(ds) print('indexes', ds.indexes) ```

Environment:

Adding a variable that's not a 1d "dimension coordinate" with the same name as a dimension is an error. This makes sense. However, if I have a .nc file that has such a variable, is there any workaround to get the badly-named variable into xarray short of altering the .nc file or loading it separately with netCDF4? I.e. to make the following work somehow ``` import xarray as xr import netCDF4

f = netCDF4.Dataset() f = netCDF4.Dataset("test.nc", "w") f.createDimension("x", 2) f.createDimension("y", 3) f["y"] = np.ones([2,3]) f["y"][...] = 1.0 f.close()

ds = xr.open_dataset('test.nc') `` rather than getting the current errorMissingDimensionsError: 'y' has more than 1-dimension and the same name as one of its dimensions ('x', 'y'). xarray disallows such variables because they conflict with the coordinates used to label dimensions.`

I think it might be nice to have something like a rename_vars argument to open_dataset(). Similar to how drop_vars ignores a list of variables, rename_vars could rename a dict of variables so the example above could do ds = xr.open_dataset("test.nc", rename_vars={"y": "y_not_dimension"}) and get a Dataset with a dimension "y" and a variable "y_not_dimension".

I think this is due to documented 'undefined behaviour' of scipy.interpolate.interp1d, so not really a bug, but I think it would be more user-friendly if xarray gave an error in this case rather than producing an 'incorrect' result.

What happened:

If a DataArray contains a NaN value and is interpolated, output values that do not depend on the entry that was NaN may still be NaN.

What you expected to happen:

The docs for scipy.interpolate.interp1d say

Calling interp1d with NaNs present in input values results in undefined behaviour.

which explain the output below, and presumably mean it is not fixable on the xarray side (short of some ugly work-around). I think it would be good though to check for NaNs in DataArray/Dataset.interp(), and if they are present raise an exception (or possibly a warning?) about 'undefined behaviour'.

scipy.interpolate.interp2d has a similar note, while scipy.interpolate.interpn does not mention it (but has very limited information).

What I'd initially expected was an output would be valid at locations in the array that shouldn't depend on the NaN input: interpolating a 2d DataArray (with dims x and y) in the x-dimension, if only one y-index in the input has a NaN value, that y-index in the output might contain NaNs, but the others should be OK.

Minimal Complete Verifiable Example:

```python import numpy as np import xarray as xr

da = xr.DataArray(np.ones([3, 4]), dims=("x", "y"))

da[0, 0] = float("nan")

newx = np.linspace(0., 3., 5)

interp_da = da.interp(x=newx)

print(interp_da) ```

On my system, this gives output: <xarray.DataArray (x: 5, y: 4)> array([[nan, 1., 1., 1.], [nan, 1., 1., 1.], [ 1., 1., 1., 1.], [nan, nan, nan, nan], [nan, nan, nan, nan]]) Coordinates: * x (x) float64 0.0 0.75 1.5 2.25 3.0 Dimensions without coordinates: y [Surprisingly, I get the same output even using method="nearest".]

You might expect at least the following, with NaN only at y=0: <xarray.DataArray (x: 5, y: 4)> array([[nan, 1., 1., 1.], [nan, 1., 1., 1.], [ 1., 1., 1., 1.], [nan, 1., 1., 1.], [nan, 1., 1., 1.]]) Coordinates: * x (x) float64 0.0 0.75 1.5 2.25 3.0 Dimensions without coordinates: y

Environment:

I noticed what I think might be a performance bug: should .compute() be called on the input data in all the plotting methods (e.g. plot.pcolormesh()) like it is in .plot() here https://github.com/pydata/xarray/blob/bf27e2c1fd81f5e0afe1ef91c13f651db54b44bb/xarray/plot/plot.py#L166 See also discussion in https://github.com/pydata/xarray/issues/2390.

I was making plots from a large dataset of a quantity that is the output of quite a bit of computation. A script which made an animation of the full time-series (a couple of thousand time points) actually ran significantly faster than a script that made pcolormesh plots of just 3 time points (~2hrs compared to ~5hrs). The difference I can think of is that the animation script called .values before the animation function, but the plotting script called xarray.plot.pcolormesh() without calling .values/.load()/.compute() first. A modified version of the script that calls .load() before any plot calls reduced the run time to 30 mins even though I plotted 18 time points, not just 3.

2d plots might all be covered by adding a darray = darray.compute() call in newplotfunc()?https://github.com/pydata/xarray/blob/bf27e2c1fd81f5e0afe1ef91c13f651db54b44bb/xarray/plot/plot.py#L609 I guess the 1d plot functions would all need to be modified individually.

Is there an xarray way to make a surface plot, like matplotlib's plot_surface()? I didn't see one on a quick skim, but expect it should be fairly easy to add, following the style for contour(), pcolormesh(), etc.? For the matplotlib version, see https://matplotlib.org/stable/gallery/mplot3d/surface3d.html.

In the docs for Dataset.plot.quiver() http://xarray.pydata.org/en/latest/generated/xarray.Dataset.plot.quiver.html the u and v arguments are labelled as 'optional'. They are required for quiver plots though, so this is slightly confusing. I guess it is like this because the docs are created from a generic _dsplot() docstring, so don't know if it's fixable in a sensible way...

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:

```python 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 https://github.com/pydata/xarray/blob/ddc352faa6de91f266a1749773d08ae8d6f09683/xarray/plot/plot.py#L678-L684 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:

Minimal Complete Verifiable Example: ``` import numpy as np import xarray as xr

ds = xr.Dataset() ds["a"] = xr.DataArray(np.linspace(0., 1.), dims="x")

ds["x"] = xr.DataArray(np.linspace(0., 2., len(ds["x"])), dims="x") ds["x"].attrs["foo"] = "bar"

print(ds["x"])

ds["b"] = xr.DataArray(np.linspace(0., 1.), dims="x")

print(ds["x"]) ```

What happened: Attribute "foo" is present at the first print, but missing in the second, after adding b to ds.

What you expected to happen: Coordinate x should be unchanged.

Anything else we need to know?:

Environment:

What happened: The browser title bar displays an incorrect version when viewing the docs online. See below - title bar says 0.15.1 but actual version in URL is 0.16.0.

http://xarray.pydata.org/en/stable/ also displays 0.15.1 in the title bar, but I guess is actually showing 0.16.0 docs (?), which is confusing!

Once xarray (or dask) has loaded some array into memory, is there any way to force the memory to be released again? Or should this never be necessary?

For example, what would be the best workflow for this case: I have several large arrays on disk. Each will fit into memory individually. I want to do some analysis on each array (which produces small results), and keep the results in memory, but I do not need the large arrays any more after the analysis.

I'm wondering if some sort of release() method would be useful, so that I could say explicitly "can drop this DataArray from memory, even though the user might have modified it". My proposed workflow for the case above would then be something like: ``` da1 = ds["variable1"] result1 = do_some_work(da1) # may load large parts of da1 into memory da1.release() # any changes to da1 not already saved to disk are lost, but do not want da1 any more

da2 = ds["variable2"] result2 = do_some_work(da2) # may load large parts of da2 into memory da2.release() # any changes to da2 not already saved to disk are lost, but do not want da1 any more

... etc. ```

Sometimes it would be nice for isel() to be able to ignore a dimension if it is missing in the Dataset/DataArray. E.g. ``` ds = Dataset()

ds.isel(t=0) # currently raises an exception

ds.isel(t=0, ignore_missing=True) # would be nice if this was allowed, just returning ds ``` For example, when writing a function can be called on variables with different combinations of dimensions.

I think it should be fairly easy to implement, just add the argument to the condition here https://github.com/pydata/xarray/blob/65a5bff79479c4b56d6f733236fe544b7f4120a8/xarray/core/variable.py#L1059-L1062 the only downside would be increased complexity of adding another argument to the API for an issue where a workaround is not hard (at least in the case I have at the moment), just a bit clumsy.

xarray.merge() drops the attrs of Datasets being merged. They should be kept, at least if they are compatible

MCVE Code Sample

```python

Your code here

import xarray as xr

ds1 = xr.Dataset() ds1.attrs['a'] = 42 ds2 = xr.Dataset() ds2.attrs['a'] = 42

merged = xr.merge([ds1, ds2])

print(merged) the result is <xarray.Dataset> Dimensions: () Data variables: empty ```

Expected Output

<xarray.Dataset> Dimensions: () Data variables: *empty* Attributes: a: 42

Problem Description

The current behaviour means I have to check and copy attrs to the result of merge by hand, even if the attrs of the inputs were identical or not conflicting.

I'm happy to attempt a PR to fix this. Proposal (following pattern of compat arguments): * add a combine_attrs argument to xarray.merge * combine_attrs = 'drop' do not copy attrs (current behaviour) * combine_attrs = 'identical' if attrs of all inputs are identical (using dict_equiv) then copy the attrs to the result, otherwise raise an exception * combine_attrs = 'no_conflicts' merge the attrs of all inputs, as long as any keys shared by more than one input have the same value (if not raise an exception) [I propose this is the default behaviour] * override copy the attrs from the first input, to the result

This proposal should also allow combine_by_coords, etc. to preserve attributes. These should probably also take a combine_attrs argument, which would be passed through to merge.

Versions

Current master of pydata/xarray on 17/3/2020

When using matplotlib, the set_under() method can be used to set values below the range of a colormap to a certain color, for example ``` import matplotlib from matplotlib import pyplot import numpy

dat = numpy.linspace(0, 1)[numpy.newaxis, :]*numpy.linspace(0, 1)[:, numpy.newaxis]

cmap = matplotlib.cm.viridis

cmap.set_under('w')

pyplot.contourf(dat, vmin=.3, cmap=cmap) pyplot.colorbar() pyplot.show() ``` produces

while uncommenting the cmap.set_under() call produces

However, using xarray to do the same thing, ``` import matplotlib from matplotlib import pyplot from xarray import DataArray import numpy

da = DataArray(numpy.linspace(0, 1)[numpy.newaxis, :]*numpy.linspace(0, 1)[:, numpy.newaxis])

cmap = matplotlib.cm.viridis cmap.set_under('w')

da.plot.contourf(vmin=.3, cmap=cmap) pyplot.show() `` produces ![xarray](https://user-images.githubusercontent.com/3958036/70076465-bc174080-15f6-11ea-9f59-37f20569cf6e.png) where it seems the call tocmap.set_under('w')` had no effect. Expected behaviour would be output like the second plot.

Output of xr.show_versions()

These lines were recently changed in #3358 https://github.com/pydata/xarray/blob/3f9069ba376afa35c0ca83b09a6126dd24cb8127/xarray/backends/locks.py#L87-L92

If the 'cloudpickle' package is not installed, then dask.multiprocessing is not available. The try/except that used to be wrapped around if actual_get is dask.multiprocessing.get meant that _get_scheduler() worked in that case, returning "threaded" (I assume this was the expected behaviour). After #3358, _get_scheduler() raised an AttributeError: module 'dask' has no attribute 'multiprocessing' until I installed 'cloudpickle'.

Suggest either reverting the changes that removed the try/except or making 'cloudpickle' a dependency.

To reproduce: 1. check 'cloudpickle' is not installed, but 'dask' is 2. execute the following commands ```

import xarray xarray.backends.api._get_scheduler() Expected result: `"threaded"` Actual result:

AttributeError Traceback (most recent call last) <ipython-input-2-20da238796b7> in <module> ----> 1 xarray.backends.api._get_scheduler()

~/.local/lib/python3.6/site-packages/xarray/backends/locks.py in _get_scheduler(get, collection) 87 pass 88 ---> 89 if actual_get is dask.multiprocessing.get: 90 return "multiprocessing" 91 else:

AttributeError: module 'dask' has no attribute 'multiprocessing' ```