Restricting a dj.U() aggregation raises a bare AssertionError #1580

Restricting a universal aggregation, dj.U().aggr(...) & <condition>, raises an AssertionError with no message, on any backend. The aggregation itself returns the expected single row; only the restriction fails, including when the restricted aggregate is used to restrict another table.

Versions: v2.3.4 and master at 68fdea5, MySQL 8.0.

Reproduction

@schema
class File(dj.Manual):
    definition = """
    file_id : int32
    ---
    filename : varchar(255)
    """

File.insert([(1, "a_part1.dat"), (2, "a_part2.dat"), (3, "a_meta.json")])
total = dj.U().aggr(File, n="count(*)")

total.to_dicts()                      # [{'n': 3}]
(total & "n = 3").to_dicts()          # AssertionError
(total & {"n": 3}).to_dicts()         # AssertionError
len(File & (total & "n = 3"))         # AssertionError

Expected: (total & "n = 3") returns [{'n': 3}] (and [] when the count differs), and File & (total & "n = 3") keeps every row of File when the restricted aggregate is non-empty and none when it is empty, as restriction by an expression with no common attributes does elsewhere (condition.py, "without common attributes, any non-empty set matches everything").

Cause

Aggregation.make_sql (expression.py:1358):

assert self._grouping_attributes or not self.restriction

A restriction on an aggregation is emitted as HAVING (MySQL) or as an outer WHERE around a subquery (PostgreSQL), but both paths assume grouping attributes exist. The PostgreSQL branch is also gated on them (use_subquery_for_having = ... and self._grouping_attributes).

Proposed fix

• Remove the assertion.
• MySQL: emit HAVING without GROUP BY, which MySQL accepts for an ungrouped aggregate and which may reference the select alias (SELECT count(*) AS n FROM t HAVING n = 3).
• PostgreSQL: take the existing subquery path whenever the aggregation is restricted, with or without grouping attributes.
• Tests: restrict dj.U().aggr(T, n="count(*)") by a string and by a dict, on both backends, with matching and non-matching counts; and restrict another table by the restricted aggregate.

Workaround

Wrap the aggregate in a subquery before restricting it:

gate = dj.U().aggr(File, n="count(*)").make_subquery() & {"n": 3}
File & gate      # all rows when the count is 3, none otherwise

len(gate) still fails on this form; see the companion issue on len() of an expression with an empty primary key.