can't pickle dict_keys objects

[brl0]

Using Dask.distributed, which uses cloudpickle for serialization, I encounter an error when passing a dict_keys object as a parameter to a function. I reported the issue there, where it was suggested that I also report an issue here for consideration.

As a test, I attempted to use cloudpickle directly on this type of object and received the following error:
TypeError: can't pickle dict_keys objects

Would it be possible to add the ability to pickle dict_keys object?

[ogrisel]

I wonder why this is not the picklable by default in Python:

>>> import pickle                                                                                  
>>> pickle.dumps({}.keys())                                                                        
Traceback (most recent call last):
  File "<ipython-input-6-6d75fd452bb1>", line 1, in <module>
    pickle.dumps({}.keys())
TypeError: cannot pickle 'dict_keys' object

Maybe I miss something. Maybe because unsuspecting users would risk sending the dict values as part of the pickle without realizing it?

I suppose we could write a reducer in cloudpickle that would ship the whole dict (which should be picklable) and reconstruct the dict_keys view on the other side. Same for dict values and items.

Would you be interested in contributing a PR?

[brl0]

Thanks for your response.

I would be interested in attempting to contribute, but after taking a look at the codebase, I am honestly not sure where/how to start. I am open to guidance if you are willing to point me in the right direction.

My initial thought is that it would be best not to ship the whole dict, because it could be large, complex, or contain sensitive information that is not needed.