Regression 2 in query join promotion logic

[Tim Graham]

(master only) I think there's a regression in the following commit - test case that passes before the commit attached. I imagine we may want to introduce a new model rather than modify ObjectC as done in the patch.
​
​https://github.com/django/django/commit/ecaba3602837d1e02fe1e961f7d3bf9086453259

(regression 1 was #21366)

[Anssi Kääriäinen]

Is this really a regression? For the query to produce any results objectb must be 1, and if objectb is 1, then LEFT JOIN will not produce any more results. This of course assumes that foreign key constraints are in effect, but that assumption has always been there for the ORM.

Rewriting the test to show that wrong results are produced would make it clear what the regression is. Not producing a left join in a case where it was done earlier isn't a regression.

[Tim Graham]

I'll rewrite the test.

[Tim Graham]

Test should now be sufficient, I think.

[Anssi Kääriäinen <akaariai@…>]

In 6fe2b001dba45134d7c10729c57959995e241a88:

Fixed #21376 -- New implementation for query join promotion logic

This commit introduced a new class JoinPromoter that can be used to
abstract away join promotion problems for complex filter conditions.
Query._add_q() and Query.combine() now use the new class.

Also, added a lot of comments about why join promotion is done the way
it is.

Thanks to Tim Graham for original report and testing the changes, and
for Loic Bistuer for review.