While adding documentation for this feature, I came across a few areas that I think might be confusing for this feature. Specifically the difference between nil vs empty of a slice/map, along with the marshaler's behavior on when elements are set or not set on the user's side during marshal.

To help clarify this I think renaming NilAsEmpty to KeepEmpty to be more inline with the common, omitempty. Will review if a nil vs empty of a list/map needs additional separation, but a keepempty I think covers the nil and empty usecases. Need to see if or how keepempty applies to non map/slice members.

@jasdel do you use this code yourself ? we also have a need to save empty maps and slices correctly (not as nil) in dynamodb.

Thanks for asking @gobadiah. Could you share more information on your use case.

In making this change, it became evident that (un)marshaler's default behavior for handling of maps and slices insufficient. In building out this PR, I think the (un)marshaler's behavior should be that nil maps/slices are sent to DynamoDB as nil, and non-nil empty maps/slices should be stored as empty maps and lists in DynamoDB. With the same behavior for unmarshaling DynamoDB items to Go types. With the exception of members decorated as Set since they must not be empty.

I think the PR should be updated to include this behavior.

Our use case is pretty much exactly the same. Right now if we save empty map / slice we get back a nil value, which forces us to put all kind of custom logic to handle these cases.

I've read past issues but I don't really understand the reasoning behind the current behaviour, and why this change is not getting more support.

Do you have a workaround or are you using your own fork to get around this limitation ?

Fixes #2746

Replaced by #2834