len() raises QuerySyntaxError for any query expression whose primary key is empty and that is not an Aggregation, including dj.U() & table, which the query-algebra spec documents as a single-row result.

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"), (2, "b")])

len(dj.U().aggr(File, n="count(*)"))                        # 1
len(dj.U() & File)                                          # QuerySyntaxError
len(dj.U().aggr(File, n="count(*)").make_subquery())        # QuerySyntaxError
len(dj.U().aggr(File, n="count(*)").make_subquery() & "n = 2")   # QuerySyntaxError

Expected: 1 for each, or 0 when the restriction leaves no row. to_dicts() and bool() on the same expressions work.

Cause

QueryExpression.__len__ (expression.py:1163-1174) picks the single-column branch when len(primary_key) <= 1, so an empty primary key produces:

SELECT count(DISTINCT ) FROM ...

Aggregation.__len__ is not affected, because it counts SELECT count(1) FROM (<subquery>).

Proposed fix

When the primary key is empty, the expression holds at most one row, so return int(bool(result)), or count over SELECT 1 FROM ... LIMIT 1. Add tests for len(dj.U() & T) on empty and non-empty T, and for len() of a restricted dj.U() aggregate subquery.

Related: #1580 (restricting a dj.U() aggregation).