Fields and projection API are very inconvenient [DATAMONGO-2637]

Zsombor opened DATAMONGO-2637 and commented

We use a couple of projections in our app to limit the amount of data to transfer, with something like this:

aggregationSteps += Aggregation.project(
   "aaaa",
   "bbbb",
   "cccc",
   "dddd"
 ....  (20 more similar lines)

However, in a refactor, we moved a couple of fields into a sub-entity, and blindly adjusted the projection to this:

aggregationSteps += Aggregation.project(
     "aaaa",
     "bbbb",
     "xxx.cccc",
     "dddd"
  ... (20 more similar lines)

Which surprisingly doesn't work as expected.

The smallest solution, which I could find is:

val fields = listOf(
     "aaaa",
     "bbbb",
     "xxx.cccc",
     "dddd"
  ... (20 more similar lines)
).map { Fields.field(it, it) }

aggregationSteps += Aggregation.project(Fields.from(*fields.toTypedArray()))

So the issue is:

• Aggregation.project(
  String... names)and Fields.fields(
  String... names) has a surprising default behavior
• There is no API which takes an array of strings, or a list of strings as field names for projections
• There is no API which takes a list of Fields for projections

Affects: 3.0.4 (Neumann SR4)

Start by tracing the Aggregation.project(String... names) and Fields.fields(String... names) APIs and their projection-related types. Check how nested names such as "xxx.cccc" are interpreted, then determine the API changes needed for string collections and lists of Fields. Done means projections support the requested field forms without the current surprising behavior, with regression coverage for nested fields.