issue_comments

After seeing the discussion in #680, I'm wondering if showing the firsts values of the flattened array wouldn't be enough here, e.g., something like this:

```

d <xarray.DataArray (a: 2, b: 5, c: 2, d: 10)> array int64 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 ... Coordinates: * a (a) <U1 'A' 'B' * b (b) <U5 'Cat 1' 'Cat 2' 'Cat 3' 'Cat 4' 'Cat 5' * c (c) <U1 'J' 'K' * d (d) int64 0 1 2 3 4 5 6 7 8 9 ```

This example is more consistent with the repr of Dataset data variables, and similarly we could customize the repr of dask arrays and lazy arrays (loaded from netcdf files) like this:

```

d.chunk((10, 5, 5, 10)) <xarray.DataArray (a: 2, b: 5, c: 2, d: 10)> dask.array int64 chunksize=(10, 5, 5, 10) Coordinates: * a (a) <U1 'A' 'B' * b (b) <U5 'Cat 1' 'Cat 2' 'Cat 3' 'Cat 4' 'Cat 5' * c (c) <U1 'J' 'K' * d (d) int64 0 1 2 3 4 5 6 7 8 9 ```

```

d.name = 'myvar' d.to_netcdf('data.nc') xr.open_dataset('data.nc').myvar <xarray.DataArray 'myvar' (a: 2, b: 5, c: 2, d: 10)> lazy-array int64 Coordinates: * a (a) <U1 'A' 'B' * b (b) <U5 'Cat 1' 'Cat 2' 'Cat 3' 'Cat 4' 'Cat 5' * c (c) <U1 'J' 'K' * d (d) int64 0 1 2 3 4 5 6 7 8 9 ```

In most cases I found the DataArray repr useful for quickly checking the dimensions (both names and sizes), the attributes and the types/values of both data and labels (I mean just checking here if the values are consistent regarding their units, acceptable ranges, etc.), but rarely for in-depth checking of the data values along each dimension, hence my suggestion of a flat (subset) array.

To inspect the data of high dimensional datarrays, I've mainly used the indexing logic of xarray to extract slices of <3 dimensions. However, I admit that for quick inspection purposes I actually like your suggestion of having a specific repr method that would allow showing small data slices as labeled tables, especially if we choose to always use a flat array for the repr of Dataarray (i.e., even when the number of dimensions <3). Why not something like:

``` python

d.slice_repr(a=0, b=0) d 0 1 2 3 4 5 6 7 8 9 c
J 0 1 2 3 4 5 6 7 8 9 K 10 11 12 13 14 15 16 17 18 19 ```

This is equivalent to

``` python

dslice = d.isel(a=0, b=0) pd.DataFrame(data=dslice.data, index=dslice.c, columns=dslice.d) ```

Except that slice_repr() would return a string instead of a data object (or an array or a dataframe). Not sure about the name and/or signature of slice_repr(), though.