flask-sqlalchemy

flask-sqlalchemy copied to clipboard

Description

By default, the pagination query methods enforce .unique() on the session/execute level. Since limit/offset is done beforehand at the query level, if the retrieved elements are not unique, the selected objects will not reflect the uniqueness/distinct filter and the query will return less elements than defined by per_page. See the code below.

class SelectPagination(Pagination):
    """Returned by :meth:`.SQLAlchemy.paginate`. Takes ``select`` and ``session``
    arguments in addition to the :class:`Pagination` arguments.

    .. versionadded:: 3.0
    """

    def _query_items(self) -> list[t.Any]:
        select = self._query_args["select"]
        select = select.limit(self.per_page).offset(self._query_offset)  # **The pagination limits are applied to the select statement**
        session = self._query_args["session"]
        return list(session.execute(select).unique().scalars())  # **But only here the `.unique()` is applied**

(...) 

On a side-note, I don't think .unique() should be enforced at this level; I believe this should be achieved manually by the user at the select statement level or, at least, be optional based on an attribute in the Pagination class. For me, at least, the current behaviour is highly undesireable because 1) I might want non-unique objects, and 2) it might obfuscate problems I might have in my query.

Please consider adding this flag for now for users to opt-out of this (to avoid breaking deployments) and possibly remove this filter altogether in a future major (breaking) change.

Fix

• Apply .distinct() in the select statement and remove . unique() from the session.execute
• (Optional) Make this unique filter optional

Environment

• Python version: 3.11
• Flask-SQLAlchemy version: main branch (3e3e92b), latest release 3.1.1
• SQLAlchemy version: 2.0.36