The locmem cache currently returns the objects from the cache directly. But it is based on a in-memory dictionary and so the returned response could be changed outside and the change would be made to the cached object itself. Think for example about a middleware that changes the response headers - if that is fed a response from the locmem:/// cache, it will change the cached response itself and the next cache hit will deliver a changed object.
This patch should solve that problem:
Index: django/core/cache.py
===================================================================
--- django/core/cache.py (revision 804)
+++ django/core/cache.py (working copy)
@@ -230,6 +230,7 @@
import cPickle as pickle
except ImportError:
import pickle
+import copy
from django.utils.synch import RWLock
class _LocMemCache(_SimpleCache):
@@ -250,7 +251,7 @@
elif exp < now:
should_delete = True
else:
- return self._cache[key]
+ return copy.deepcopy(self._cache[key])
finally:
self._lock.reader_leaves()
if should_delete:
The CacheMiddleware? itself currently does a copy.copy() on the cached response, but that only is a shallow copy - and shallow copies don't help with the above header-changing scenario. I think with this patch, the copy.copy() in the CacheMiddleware? can be dropped, as the other caches all use pickling and unpickling to store and retrieve the objects (and the memcache interface does it's own pickling/unpickling) and so already do something similar to deep copying.
