Translation RQL to SQL with ORDERBY COUNT

With this schema:

class Person(EntityType):
    __unique_together__ = [("nom", "prenom")]
    nom = String(maxsize=64, fulltextindexed=True, required=True)
    prenom = String(maxsize=64, fulltextindexed=True)
    travaille = SubjectRelation(
        "Societe",
        __permissions__={
            "read": (),
            "add": (),
            "delete": ("managers",),
        },
    )

    evaluee = SubjectRelation("Note")

class Societe(EntityType):
    nom = String(maxsize=64, fulltextindexed=True)
    evaluee = SubjectRelation("Note")


class Note(EntityType):
    date = String(maxsize=10)
    type = String(maxsize=1)
    para = String(maxsize=512)

The following RQL: Any X, COUNT(Y) GROUPBY X ORDERBY COUNT(Y) WHERE Y is IN (Societe, Person), Y evaluee X is translated into:

SELECT T1.C0, COUNT(T1.C1) FROM (
    SELECT rel_evaluee0.eid_to AS C0, _Y.cw_eid AS C1, COUNT(_Y.cw_eid) AS C2
    FROM cw_Personne AS _Y, evaluee_relation AS rel_evaluee0
    WHERE rel_evaluee0.eid_from=_Y.cw_eid
    UNION ALL
    SELECT rel_evaluee0.eid_to AS C0, _Y.cw_eid AS C1, COUNT(_Y.cw_eid) AS C2
    FROM cw_Societe AS _Y, evaluee_relation AS rel_evaluee0
    WHERE rel_evaluee0.eid_from=_Y.cw_eid
) AS T1
GROUP BY T1.C0
ORDER BY 2

Which is syntactically wrong (no GROUP BY in the 2 "inside" SELECTS). Here is what we might want:

SELECT T1.C0, COUNT(T1.C1) FROM (
    SELECT rel_evaluee0.eid_to AS C0, _Y.cw_eid AS C1
    FROM cw_Personne AS _Y, evaluee_relation AS rel_evaluee0
    WHERE rel_evaluee0.eid_from=_Y.cw_eid
    UNION ALL
    SELECT rel_evaluee0.eid_to AS C0, _Y.cw_eid AS C1
    FROM cw_Societe AS _Y, evaluee_relation AS rel_evaluee0
    WHERE rel_evaluee0.eid_from=_Y.cw_eid
) AS T1
GROUP BY T1.C0
ORDER BY 2

Fun facts:

• If we remove "Y is IN (Societe, Person)", RQL is correctly translated ;
• This happens only when we have a relation with several objects and/or several subjects, it seems the error occurs when we need a SQL JOIN.