netCDF4 should error when assigning multi-dimensional arrays as attributes

[shoyer]

Right now, they are silently flattened:

In [13]: import netCDF4

In [14]: with netCDF4.Dataset('testing.nc', 'w') as nc:
    ...:     nc.foo = ['multi', 'dimensional']
    ...:     nc.bar = [[1, 2], [3, 4]]
    ...:

In [15]: with netCDF4.Dataset('testing.nc') as nc:
    ...:     print(nc.foo)
    ...:     print(nc.bar)
    ...:
multidimensional
[1 2 3 4]

This was reported downstream by an xarray user: pydata/xarray#2251

[jswhit]

Agreed - this is a bug. Will submit a PR.

[jswhit]

I hope no one is depending on this behavior. Maybe we should issue a FutureWarning instead, and change it to ValueError in the next release.

[jswhit]

Note that nc.foo = ['multi', 'dimensional'] produces [u'multi', u'dimensional'] in master, so this PR will have no affect for this case. However nc.foo = [['multi'], ['dimensional']] will now raise a ValueError.

[shoyer]

A FutureWarning sounds like a good choice for now.

…

On Wed, Aug 29, 2018 at 12:13 PM Jeff Whitaker ***@***.***> wrote: Note that nc.foo = ['multi', 'dimensional'] produces [u'multi', u'dimensional'] in master, so this PR will have no affect for this case. However nc.foo = [['multi'], ['dimensional']] will now raise a ValueError. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#841 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABKS1iB3GOnB0PFrOZLedGPHXfBgNUxRks5uVufKgaJpZM4WR3bj> .

[jswhit]

changed pull request #842 to FutureWarning

[jswhit]

pull request #824 merged. FutureWarning will change to ValueError in 1.4.3