subqueryload() generates redundant queries

[Former user]

When subqueryload() is used on a relation of a polymorphic table, it generates/executes the exact same query multiple times. This hurts badly when that query is a heavy one. 0.6 and 0.7 shows the same behavior, but e.g. 0.7.0 generates just 2 identical queries in the attached example while 0.7.5 generates 3.

[Mike Bayer]

great test. this is a side effect of something that might be improved within mapper loading overall at some point, but for now this is a patch that solves this issue for the case of subqueryloaders, as the "row processor" function created is much more expensive than any other.

--- a/lib/sqlalchemy/orm/strategies.py  Fri May 04 23:18:52 2012 -0400
+++ b/lib/sqlalchemy/orm/strategies.py  Tue May 08 10:41:17 2012 -0400
@@ -864,7 +864,13 @@

         q = context.attributes[reduced_path)](('subquery',)

-        collections = dict(
+        # cache the loaded collections in the context
+        # so that inheriting mappers don't re-load when they
+        # call upon create_row_processor again
+        if ('collections', reduced_path) in context.attributes:
+            collections = context.attributes[reduced_path)](('collections',)
+        else:
+            collections = context.attributes[reduced_path)](('collections',) = dict(
                     (k, [v[0](v[0) for v in v]) 
                     for k, v in itertools.groupby(
                         q,

[Mike Bayer]

d05319ec7e7868fb09792d07d975c029c178d2d0 7bb964e9e57849de921821b58260b86389b4b939

[Mike Bayer]

Removing milestone: 0.7.8 (automated comment)