Implement raiseload for deferred columns

Added "raiseload" feature for ORM mapped columns.

As part of this change, the behavior of "deferred" is now more strict;
an attribute that is set up as "deferred" at the mapper level no longer
participates in an "unexpire" operation; that is, when an unexpire loads
all the expired columns of an object which are not themselves in a deferred
group, those which are mapper-level deferred will never be loaded.

Deferral options set at query time should always be reset by an expiration
operation.

Renames deferred_scalar_loader to expired_attribute_loader

Unfortunately we can't have raiseload() do this because it would break
existing wildcard behavior.

Fixes: #4826
Change-Id: I30d9a30236e0b69134e4094fb7c1ad2267f089d1

1 files changed, 3 insertions, 0 deletions

diff --git a/lib/sqlalchemy/orm/loading.py b/lib/sqlalchemy/orm/loading.py
index 106ea7985..8de7d5a8b 100644
--- a/lib/sqlalchemy/orm/loading.py
+++ b/lib/sqlalchemy/orm/loading.py
@@ -21,6 +21,7 @@ from . import exc as orm_exc
 from . import path_registry
 from . import strategy_options
 from .base import _DEFER_FOR_STATE
+from .base import _RAISE_FOR_STATE
 from .base import _SET_DEFERRED_EXPIRED
 from .util import _none_set
 from .util import aliased
@@ -384,6 +385,8 @@ def _instance_processor(
                 # searching in the result to see if the column might
                 # be present in some unexpected way.
                 populators["expire"].append((prop.key, False))
+            elif col is _RAISE_FOR_STATE:
+                populators["new"].append((prop.key, prop._raise_column_loader))
             else:
                 getter = None
                 # the "adapter" can be here via different paths,