limit/offset criteria that forces a subquery b. the relationship
uses "secondary" c. the primaryjoin of the relationship refers to
a column that is either not part of the primary key, or is a PK
col in a joined-inheritance subclass table that is under a different
attribute name than the parent table's primary key column d. the
query defers the columns that are present in the primaryjoin, typically
via not being included in load_only(); the necessary column(s) would
not be present in the subquery and produce invalid SQL.
fixes #3592

scope of a :meth:`.Session.flush` operation that's raising an
exception, as has been observed in some MySQL SAVEPOINT cases, prevents
the original  database exception from being observed when it was
emitted during  flush, but only on Py2K because Py2K does not support
exception  chaining; on Py3K the originating exception is chained.  As
a workaround, a warning is emitted in this specific case showing at
least the string message of the original database error before we
proceed to raise  the rollback-originating exception.
fixes #2696

the many caveats and confusing effects of, the popular approach
of using contains_eager() to alter the natural result of a
related collection.  I'm not a fan of this technique as it changes
the semantics of a relationship in such a way that the rest of the
ORM isn't aware of and it also can be undone very easily; hence the
section needs as much text for warnings as for describing the
technique itself.   fixes #3563

to return ``datetime.timedelta`` in the same way as that of
:obj:`.types.Interval.python_type`, rather than raising
``NotImplementedError``.
fixes #3571

(cherry picked from commit 29d6f6e19b014bb5ce79032bd8803e32b4da0e5e)

relationship flag; this flag *does* have an effect when the baked
lazy loader plugin has been invoked.  clarify the intent of this
flag as an "opt out" but only has an effect when the baked system
is loaded anyway.  fixes #3572