xr.concat loses coordinate dtype information with recarrays in 0.9

Simple example script:

import numpy as np
import xarray as xr

p1 = np.array([('A', 180), ('B', 150), ('C', 200)], dtype=[('name', '|S256'), ('height', int)])
p2 = np.array([('D', 170), ('E', 250), ('F', 150)], dtype=[('name', '|S256'), ('height', int)])

data1 = np.arange(50, 80, 1, dtype=np.float)
data2 = data1 * 2

dims = ['measurement', 'participant']

da1 = xr.DataArray(
    data1.reshape(10, 3),
    coords={
        'measurement': np.arange(10),
        'participant': p1,
        'samplerate': 1
    },
    dims=dims
)

da2 = xr.DataArray(
    data2.reshape(10, 3),
    coords={
        'measurement': np.arange(10),
        'participant': p2,
        'samplerate': 1
    },
    dims=dims
)

combined = xr.concat([da1, da2], dim='participant')
print(da1.participant.dtype)
print(da2.participant.dtype)
print(combined.participant.dtype)

With xarray version 0.9.5 (and 0.8.2), this results in:

[('name', 'S256'), ('height', '<i8')]
[('name', 'S256'), ('height', '<i8')]
object

We have to go all the way back to xarray version 0.7 to get the expected behavior:

[('name', 'S256'), ('height', '<i8')]
[('name', 'S256'), ('height', '<i8')]
[('name', 'S256'), ('height', '<i8')]

[shoyer]

Is there an advantage to using pandas for concatenation over numpy.concatenate?

Yes, otherwise we loose pandas (d)type information. For example, a MultiIndex gets turned into a dtype=object array of tuples.

[aldanor]

Looks like this is still a problem, just tested on 0.11.3 and it still results in object…