issues

Is there a way of over-specifying chunks upon opening a dataset without throwing an error?

Currently, giving chunk sizes along dimensions that are not present in the dataset fails with a ValueError: python da = xr.DataArray(np.arange(10), dims=("x", ), name="example_data") da.to_dataset().to_netcdf("example.nc") ds = xr.open_dataset("example.nc", chunks={"x": 5, "y": 1})

Currently, to_netcdf() will write files containing a variable named __xarray_dataarray_variable__ if self.name is None.

Should there be a warning to at least notifies the user that it would be a good idea to pick a decent variable name?

Problem

I am running into inconsistent time coordinates with a long climate model experiment that exceeds the limits of pandas.tslib.Timestamp (covers roughly 17th to 23rd century).

Currently, xarray.open_mfdataset delegates decoding of the time axis to xarray.open_dataset which decodes either to pandas time stamps or, of this fails, to netcdftime.datetime objects. xarray.open_mfdataset later combines the single-file datasets and just concatenates all the time axes.

Solution

1. Let auto_combine check for consistency and repair the time axis if necessary.
2. Let xarray.open_mfdataset prevent xarray.open_dataset from decoding the times for each file and only decode times after everything is combined.

The latter is equivalent to a workaround I use for the moment: Pass decode_times=False to xarray.open_mfdataset and then explicitly call xarray.decode_cf on the dataset.

Minimal example

The following shows that creating a time-coordinate with two dates before and after the latest date compatible to np.datetime64[ns] results in coordinates that have been decoded to np.datetime64[ns] for the first (and compatible) date and left at datetime.datetime for the second (and incompatible) date:

```python In [1]: from datetime import datetime ...: import numpy as np ...: import xarray as xr

In [2]: dates = [datetime(year, 1, 1) for year in [2262, 2263]] ...: ds = xr.Dataset(coords={"dates": dates})

In [3]: print(ds.coords["dates"][0]) <xarray.DataArray 'dates' ()> array('2262-01-01T00:00:00.000000000', dtype='datetime64[ns]') Coordinates: dates datetime64[ns] 2262-01-01

In [4]: print(ds.coords["dates"][1]) <xarray.DataArray 'dates' ()> array(datetime.datetime(2263, 1, 1, 0, 0), dtype=object) Coordinates: dates object 2263-01-01

In [5]: ds2 = xr.Dataset({})

In [6]: ds2["dates"] = (["dates", ], dates)

In [7]: ds2.coords["dates"][0] Out[7]: <xarray.DataArray 'dates' ()> array('2262-01-01T00:00:00.000000000', dtype='datetime64[ns]') Coordinates: dates datetime64[ns] 2262-01-01

In [8]: ds2.coords["dates"][1] Out[8]: <xarray.DataArray 'dates' ()> array(datetime.datetime(2263, 1, 1, 0, 0), dtype=object) Coordinates: dates object 2263-01-01

```

Problem description

I don't seem to find a way of passing time-coordinates to an xarray Dataset without having them decoded. This is problematic, because it makes it very hard (or impossible?) for a user to make sure a time axis does completely consist of, e.g., netcdftime objects.

Is there a simple way of inserting, say, a time-step in an xarray.DataArray?

Background

I have a year of gridded daily data with a few missing time steps. Each existing time step is represented by a file on disk. (To be specific: For 2016, there should be 366 files, but there are only 362.) In many cases, it would be nice to be able to just add masked data whereever a day is missing from the original data.

• [x] Closes #799
• [x] Tests added / passed
• [x] Passes git diff upstream/master | flake8 --diff
• [x] Fully documented, including whats-new.rst for all changes and api.rst for new API

This is meant to eventually make xarray.open_dataset and xarray.open_mfdataset work with pathlib.Path objects. I think this can be achieved as follows:

1. In xarray.open_dataset, cast any pathlib.Path object to string

2. In xarray.open_mfdataset, make sure to handle generators. This is necessary, because pathlib.Path('some-path').glob() returns generators.

Curently, tests with Python 2 are failing, because there is no explicit pathlib dependency yet.

With Python 3, everything seems to work. I am not happy with the tests I've added so far, though.

This has #799 in mind.

Currently, xarray/plot.py raises a TypeError if the data to be plotted are not among [np.floating, np.integer, np.timedelta64, np.datetime64].

This PR adds netCDF4.datetime objects to the list of allowed data types. These occur, because xarray/conventions.py passes undecoded netCDF4.datetime objects if decoding to numpy.datetime64 fails.