ENH: fill_value argument for shift

[adbull]

Feature Request

At the moment, the Series/DataFrame shift methods always fill with nans, which causes an upcast for boolean/integer types. In cases where there is an appropriate non-nan fill value, it would be useful to provide this value in shift, rather than having to correctly downcast the result. This would presumably also be more efficient than .shift().fillna() in those cases where no upcast is required.

Example Code

>>> pd.Series([True]).shift(fill_value=False)
0    False
dtype: bool

[jreback]

sure this is reasonable to add. Also wouldn't be very difficult, mostly just need to pass this thru to the lower level functions (where a fill_values is found).

care to do a PR?

[linebp]

This is a copy of #15527 where work seems to be in progress?

[jreback]

@linebp this is the original reported issue, #15527 is a possible patch for this.

[samuelsinayoko]

Looks like this still needs doing. I'm going to try and give this ago if nobody is working on it.

[samuelsinayoko]

First stab in this PR https://github.com/pandas-dev/pandas/pull/20460/files