As I have investigated, data of joined models is not cached at all. The reason is that Doctrine_Record::serialize() removes relations before serialization:

unset($vars['_references']);

When I commented out this line, my test case passes successfully, but I'm not sure would it be OK to just delete this line?

Chapter "7.1.12 Serializing" of User Manual states: Records can be serialized, but remember: Doctrine cleans all relations, before doing this.

So, removing references is documented behavior, but it contradict examples of result cache usage.

Test Case.

May I stress the magnitude of this problem; join queries are more often than not more resource consuming that regular ones. Not being able to cache them (and moreso, creating overhead IF YOU DO) simply invalidates the caching mechanism itself and makes Doctrine unsuitable for large scale projects -- which is, exactly where using Doctrine makes a lot of sense.

I apologize for stating the obvious, obviously I've just bumped into the problem. I'll look through the code and see if I can come up with a solution, although my experience with Doctrine internals is pretty much non-existent.

Just to be clear, this is a known limitation. The result cache works fine with a) single objects, b) a flat list of objects, c) an array graph (HYDRATE_ARRAY), d) Other scalar hydration modes.

The only thing that does *not* work is caching object graphs because the associations are not reconstructed. We're open to proposed solutions but they need to be tested properly.

Automatic lazy-loading just makes this limitation worse when it is encountered, this is a negative effect of lazy-loading and the reason why lazy-loading will not be enabled by default in the future.

fixes #1636

Patch created.

from what I can tell,removing the line unset($vars_references?); from the seralization method in the Record class removes the issue and doesn't break anything. However I am a little concerned, about memory consumption. Will all the recursions be removed? I have been using it in dev environment too for two days and its looking ok so far.

(In [5301]) [1.0, 1.1] Fixing issue with resultset cache and joined records (closes #1636)

(In [5316]) [1.0, 1.1] Fixes regression in resultset cache (closes #1768) and (addresses #1636 again)