AnyOfFilter shouldn't impose distinct by default unless joining on another table

[rodrigoalmeidaee]

Context

Currently, AnyOfFilter, since it inherits from MultipleChoiceFilter, sets distinct=True by default on queries. This is for a good reason: when filtering on a field that may traverse relationships, rows may get duplicated as result.

Issues

More often that not, AnyOfFilter is used to filter on the table own fields, in which case distinct is not necessary as rows won't duplicate by definition (they all have different primary keys 😅). It is unclear whether PostgreSQL can figure the same thing out on its own (I've done a simple test locally against a table and it seems it doesn't, at least for the case I tested). In case it can't, there are possible performance gains of not using distinct when filtering own table fields.

Solution

distinct becomes a three-valued attribute for AnyOfFilter:

• True: uses distinct()
• False: doesn't use distinct()
• None (the default): infer based on the field_name: if it contains a navigation (identified by the presence of a __), use distinct.

Note: lookups are also double underscored and will become "false positives" (triggering distinct when not needed), but this matches the current behavior (which triggers distinct all the time) and is not a regression.