Inconsistent handling of empty lists in Document partial updates.

[albertisfu]

Hi! At work we are currently developing an API that relies on Elasticsearch responses. Within our document mapping, we have defined multi-value fields, which we aim to represent as empty arrays ([]) in the API response when these fields are empty. When initially indexing the document, it successfully store empty lists into ES as [] by passing these fields with skip_empty=False.

However, we are encountering an issue with partial updates. After updating a field to become empty, it is stored as None instead of [], even though we are passing the field values as [].

The root cause appears to be within the current implementation of the Document.update method for partial document updates. This method prepares data for Elasticsearch by invoking the to_dict method, which, by default, ignores empty fields. Consequently, empty lists are treated as None and updated accordingly.

Proposed Fix:
To address this issue, I propose modifying the update method to accept skip_empty=False and passing this parameter to the to_dict method. This adjustment will ensure that empty lists are included in the update payload as empty lists instead of being interpreted as None.

Steps to Reproduce:

1. Save a document with multi-value fields.
2. Perform a partial update using the Document.update method, passing multi-value fields as empty lists [].
3. Verify that the empty lists are updated as None instead of remaining as empty lists.

I am willing to contribute to fixing this issue if you believe it should be fixed in this way.

[miguelgrinberg]

I was able to reproduce this easily. I think update() should always use to_dict(skip_empty=False), I don't see why you would send an empty value to update but would like it to be skipped. Working on a fix.

[albertisfu]

You're right; there is no reason to pass skip_empty=True on an update. The default option, False, is correct.

Thank you!