issue_comments

Thanks to @topper-123's PR being merged upstream, this is now fixed. The next time our CI runs it should be green (I tested things out offline). If we want we can consider leveraging _id in CFTimeIndex later, but this particular issue can be closed.

Thanks @topper-123 -- I see you also made https://github.com/pandas-dev/pandas/pull/37321, making Index._id optional again, which I appreciate!

We do rely on private attributes in some places, but we're typically cautious about doing that, because their behavior can change without warning. For instance, in this case pandas-dev/pandas#37087 would have been a breaking change in a different way, since _reset_identity went from returning an Index, to operating on the Index in place.

Looking at the history of pandas.Index, it generally seems like outside the recent change, _reset_identity has been pretty stable over the years, though we'd still need to mull over the trade-off between the performance benefit and the robustness of the subclass to changes in pandas.

I’ve looked at the CFTimeIndex.__new__ class method, and recommend you add a line result._id = result._reset_id() there.

This will fix this bug, but should be done even if this particular error didn`t appear, because it will give a performance boost when comparing indexes.

I posted a question in the pandas repo related to this: https://github.com/pandas-dev/pandas/issues/37213.

Indeed a bisect reveals pandas-dev/pandas#37087 was the cause: ``` abd3acf05516611e9e90d57ae363f1567e30f49a is the first bad commit commit abd3acf05516611e9e90d57ae363f1567e30f49a Author: Terji Petersen contribute@tensortable.com Date: Wed Oct 14 13:26:45 2020 +0100

CLN: clean Index._id (#37087)

pandas/core/indexes/base.py | 21 +++++++++++++-------- pandas/core/indexes/multi.py | 4 +++- pandas/tests/arithmetic/test_object.py | 3 ++- 3 files changed, 18 insertions(+), 10 deletions(-) ```

I think it's that _id used to automatically be initialized to None and now it needs to be explicitly initialized within the constructor. They added some try-except logic in the is_ method to handle the case where _id has not been initialized, but there's an error happening now within the view method. Here's a more minimal example:

``` In [1]: import xarray as xr

In [2]: xr.cftime_range("2000", periods=2).view()

AttributeError Traceback (most recent call last) <ipython-input-2-5226cd5e3bc1> in <module> ----> 1 xr.cftime_range("2000", periods=2).view()

~/Software/pandas/pandas/core/indexes/base.py in view(self, cls) 630 result = self._shallow_copy() 631 if isinstance(result, Index): --> 632 result._id = self._id 633 return result 634

AttributeError: 'CFTimeIndex' object has no attribute '_id' ```

I'll have to think about things a little more to see if we should pursue an upstream fix or whether we can address it in xarray.

pandas-dev/pandas#37087 recently touched Index._id but doesn't seem to change its functionality so might not be the cause.