pandas (original) (raw)

Currently, DataFrame.to_feather() tries to import feather, which only works if feather-format is installed. The same capabilities could also be imported as follows: import pyarrow.feather as feather.

The issue is that pyarrow is more easily available (e.g. conda install feather-format doesn't work without specifying a specific channel; but pyarrow works), and is under more active development.

I asked if there is a preference between the two imports in wesm/feather#341, and Wes answered [my bold]:

You can use either method; there is no explicit recommendation right now. import feather is provided for backwards compatibility

I think the import should at least try import pyarrow.feather as feather as well, if import feather fails.