Index dtype changes when assigning column from index

• Creating a new ddf column from the ddf’s existing index erroneously changes the dtype of the index
• Impact: causes downstream ddf operations to fail or behave in unintended ways, e.g. dd.concat

>>> ddf = dd.from_pandas(pd.DataFrame(dict(i=['a','b'], a=[1,2])), npartitions=1)
>>> ddf = ddf.set_index('i)

>>> ddf.index

Dask Index Structure:
npartitions=1
a    object
b       ...
Name: i, dtype: object
Dask Name: sort_index, 4 tasks

>>> ddf.assign(b=ddf.index).index

Dask Index Structure:
npartitions=1
a    int64
b      ...
dtype: int64
Dask Name: assign, 6 tasks

Versions

$ python --version
Python 3.6.0

$ cat requirements.txt | egrep 'dask/dask' | grep -v '^#'
git+https://github.com/dask/dask.git@00ed56e

[jcrist]

This is fixed by #4695:

In [1]: %load test.py

In [2]: # %load test.py
   ...: import pandas as pd
   ...: import dask.dataframe as dd
   ...:
   ...: df = pd.DataFrame(dict(i=['a','b'], a=[1,2]))
   ...: ddf = dd.from_pandas(df, npartitions=1).set_index('i')
   ...: ddf2 = ddf.assign(b=ddf.index)
   ...: assert ddf2.index.dtype == object
   ...:

In [3]: %load test2.py

In [4]: # %load test2.py
   ...: import pandas as pd
   ...: import dask.dataframe as dd
   ...:
   ...: df = pd.DataFrame([dict(x='c', y='a'), dict(x='d', y='b')])
   ...: ddf = dd.from_pandas(df, 2)
   ...: ddf2 = ddf.rename(columns=dict(x='index'))
   ...: ddf3 = ddf2.groupby('index')['y'].apply('-'.join, meta=str)
   ...: ddf4 = ddf3.reset_index()
   ...: assert ddf4['index'].dtype == 'int64'
   ...:

[jdanbrown]

@jcrist Thanks!