issues

What happened: When creating a new variable based on an existing dimension, the attrs of the dimension are lost.

What you expected to happen: The attrs should be preserved.

Minimal Complete Verifiable Example:

```python import xarray as xr

ds = xr.Dataset() ds['x'] = xr.DataArray(range(10), dims='x') ds['y'] = xr.DataArray(range(len(ds['x'])), dims='x') ds['x'].attrs['foo'] = 'bar' print(ds['x']) # attrs of ds['x'] are preserved

print('\n****\n')

ds = xr.Dataset() ds['x'] = xr.DataArray(range(10), dims='x') ds['x'].attrs['foo'] = 'bar' ds['y'] = xr.DataArray(range(len(ds['x'])), dims='x') print(ds['x']) # attrs of ds['x'] are lost ```

Output of above code: ``` <xarray.DataArray 'x' (x: 10)> array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9]) Coordinates: * x (x) int64 0 1 2 3 4 5 6 7 8 9 Attributes: foo: bar

<xarray.DataArray 'x' (x: 10)> array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9]) Coordinates: * x (x) int64 0 1 2 3 4 5 6 7 8 9 ```

Environment:

This is a do-over of @gajomi's #2070 with minimal additional tests to preserve attrs on Datasets as well as DataArrays.

• [x] Closes #2049, closes #2070
• [x] Tests added
• [x] Passes isort . && black . && mypy . && flake8
• [x] User visible changes (including notable bug fixes) are documented in whats-new.rst

I've been working with some time-series data with occasional nans peppered throughout. I want to interpolate small gaps of nans (say, when there is a single isolated nan or perhaps a block of two) but leave larger blocks as nans. That is, it's not appropriate to fill large gaps, but it acceptable to do so for small gaps.

I was hoping interpolate_na() with the limit argument would do exactly this, but it turns out that if you specify, say, limit=2, it will fill the first two nans of nan-blocks of any length, no matter how long. There are definitely solutions for dealing with this, but it seems like a common issue, and has cropped up over on Pandas as well.

I'm not able to attempt tackling this right now, but I guess I wanted to put in a feature request for an additional argument to interpolate_na() that would do this.

• [x] Closes #3404
• [x] Tests added
• [x] Passes black . && mypy . && flake8

Testing could be improved, but as-is it would fix the same issue from happening again.

I recently upgraded to xarray 0.14.0. When running code that used to work in 0.13, I get a TypeError: ufunc 'isnan' not supported for the input types, and the inputs could not be safely coerced to any supported types according to the casting rule ''safe'' in 0.14 when running code similar to the MCVE below. The code should return the GroupBy bins instead of the error.

MCVE Code Sample

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

ts = pd.date_range(start='2010-08-01', end='2010-08-15', freq='24.8H')

ds = xr.Dataset() ds['time'] = xr.DataArray(pd.date_range('2010-08-01', '2010-08-15', freq='15min'), dims='time') ds['val'] = xr.DataArray(np.random.rand(*ds['time'].shape), dims='time')

ds.groupby_bins('time', ts) #error thrown here ```

Full error details below.

Output of xr.show_versions()

• [x] Closes #2887
• [x] Tests added
• [x] Fully documented, including whats-new.rst

Implement load_dataset() and load_dataarray()

BUG: Fixes #2887 by adding @shoyer solution for load_dataset and load_dataarray, wrappers around open_dataset and open_dataarray which open, load, and close the file and return the Dataset/DataArray TST: Add tests for sequentially opening and writing to files using new functions DOC: Add to whats-new.rst. Also a tiny change to the open_dataset docstring

Updates formatting to use .format() instead of % operator. Changed all instances of % to .format() and added test for using tuple as key, which errored using % operator.

BUG: Fix #2770 by changing ~ to np.logical_not() TST: Add test for scalar datetime64 value DOC: Add to whats-new.rst

I'm not sure I understand how the unlimited_dims option to to_netcdf() is supposed to work. Consider the following: python ds = xr.Dataset() ds['time'] = xr.DataArray(pd.date_range('2000-01-01', '2000-01-10'), dims='time') ds.to_netcdf('timedim.cdf', unlimited_dims='time') This results in a file that looks like this: ``` $ ncdump timedim.cdf netcdf timedim { dimensions: t = UNLIMITED ; // (0 currently) i = UNLIMITED ; // (0 currently) m = UNLIMITED ; // (0 currently) e = UNLIMITED ; // (0 currently) time = UNLIMITED ; // (10 currently) variables: int64 time(time) ; time:units = "days since 2000-01-01 00:00:00" ; time:calendar = "proleptic_gregorian" ; data:

time = 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 ; } `` Note the dimensions namedt,i,m,eall with zero length. Thetimedimension (which is the only one that should exist) is properly set toUNLIMITEDbut we shouldn't have the four extra dimensions. What's going on here? The same behavior occurs when setting viads.encoding['unlimited_dims'] = 'time'. Everything is as expected without theunlimited_dimsoption (but thetimedimension is notUNLIMITED`, of course).

I thought it could be related to the variable and dimension having the same name, but this also happens when they are different.

Expected Output

There shouldn't be extra 0-length dimensions

Output of xr.show_versions()

• [X] Closes #2044
• [X] Tests added (for all bug fixes or enhancements)
• [x] Tests passed (for all non-documentation changes)
• [x] Fully documented, including whats-new.rst for all changes and api.rst for new API (remove if this change should not be visible to users, e.g., if it is an internal clean-up, or if this is part of a larger project that will be documented later)

Migrated from Stack Overflow.

NetCDF supports the NC_STRING type, which can stores arrays of strings in attributes. Xarray already supports reading arrays of strings from attributes in netCDF files, and It would be great if it also supported writing the same.

Reading already works

```python import xarray as xr import netCDF4 as nc

rg = nc.Dataset('test_string.nc', 'w', format='NETCDF4') rg.setncattr_string('testing', ['a', 'b']) rg.close() ds = xr.open_dataset('test_string.nc') print(ds) gives <xarray.Dataset> Dimensions: () Data variables: empty Attributes: testing: ['a', 'b'] ```

This works because I used the setncattr_string method. Setting the attributes like rg.testing = ['a', 'b'] does not work and results in a concatenated list (just like the xarray example below).

Writing doesn't work

python import xarray as xr ds = xr.Dataset() ds.attrs['testing'] = ['a', 'b'] ds.to_netcdf('asdf.nc') ds = xr.open_dataset('asdf.nc', autoclose=True) print(ds) gives <xarray.Dataset> Dimensions: () Data variables: *empty* Attributes: testing: ab

Note the list elements have been concatenated. So this is a request for xarray to implement something like netCDF4's setncattr_string. I would be happy to help do this if someone pointed me in the right direction; I looked through the code but got lost pretty quickly. Thanks.