2.8 SQLAlias leaks

[Chris McDonough]

When you bind a Shared.DC.ZRDB.Results.Results class to an instance of a 2.8 extensionclass (e.g. self.results = <result instance>), it creates a cycle that eventually manifests itself as SQLAlias reference leaks. This appears to be a resurrection of an ancient bug (ala http://mail.zope.org/pipermail/zope-dev/2000-December/008280.html).

[Andreas Jung (ajung)]

hm...what is the usecase for binding a Results instance as instance attribute? I can not think of a case where this would make sense.

[Tres Seaver (tseaver)]

Andreas wrote:

> hm...what is the usecase for binding a Results instance as
> instance attribute? I can not think of a case where this would
> make sense.

1. I have a view for a content object, which execute an SQL
   query and stores the results as an instance of the view, which
   should be only for the life of the request. However, due to
   this bug, the view and all associated objects (the request,
   the SQL results, etc.) leak.

2. I can also imagine this being problematic for caches of the
   SQL results.

[Tres Seaver (tseaver)]

Uploaded: results_leak.py

Attaching a script which (I think) demonstrates the leak. Here
is output on my machine:

$ bin/zopectl run ../results_leak.py
================================
2.3.5 (#1, Apr 11 2005, 13:18:06)
[GCC 3.3.5 (Debian 1:3.3.5-8ubuntu2)]
================================
--------------------------------
Leak test: None
--------------------------------
leaked: 0

--------------------------------
Leak test: by attribute
--------------------------------
leaked: 16

--------------------------------
Leak test: by item
--------------------------------
leaked: 15