fix: handle empty last chunk correctly in 'Query._chunkify'

[tseaver]

Closes #487.
Supersedes #488.

[noxxious]

In the #488 I also had a fix for the None case of the bulk_writter being not correctly initialized.

[tseaver]

@noxxious Your change (calling self.bulk_writer rather than instantiating BulkWriter() directly) didn't change the behavior: the client's bulk_writer method takes an optional options argument, but it isn't being passed.

[noxxious]

Sure, passing options would have changed the client API interface. With the current API of bulk_writer being optional and current coded behavior of constructing bulk_writer, using a default constructor results in non-functional BulkWriter not bound to a client, hence all the recursive delete actions are not performed.

My additional tests for the None BulkWriter just show that.

So this means, either the method API needs to change to indicate that the bulk_writer is not optional or the options need to be added.

[tseaver]

@noxxious Thanks for the follow-up!

It is indeed a problem that we have code constructing BulkWriter() without passing a client: that argument should not be optional (now fixed in db9599b).

a2e5469 refactors the bits which construct the bulk writer if none is passed to the publicrecursive_delete methods: the private _recursive_delete helpers now have it as a required argument, and the public methods construct it using self.bulk_writer(), as in your patch.

[noxxious]

Glad to help.

[chrisrossi]

LGTM!