-
Notifications
You must be signed in to change notification settings - Fork 4.5k
Description
Currently transforms such as Select, DropFields, RenameFields, and AddFields can create schemas with unexpected order. The problem is that FieldAccessDescriptor stores top-level fields and nested fields separately, so there's no way to tell the relative order between them. To fix this we should refactor FieldAccessDescriptor: instead of storing these separately it should store a single list, where each item in the list might optionally have a nested FieldAccessDescriptor.
Expected behavior from the transforms:
DropFields: preserves order in original schema
RenameFields: preserves order in original schema
AddFields: adds fields in order specified. If multiple nested fields are selected, the first reference to the top field wins (e.g. adding "a.b", "c", "a.d" results in adding a before c.
Select: Select fields in the order specified.
Imported from Jira BEAM-7301. Original Jira may contain additional context.
Reported by: reuvenlax.
Subtask of issue #18898