Why does weakproxy not always preserve equivalence in python ?

Question

MySQLDb uses weak proxy to prevent circular dependencies between cursors and connections.

But you would expect from the documentation on weakref that you could still tests for equivalence. Yet:

In [36]: interactive.cursor.connection.thread_id()
Out[36]: 4267758

In [37]: interactive.web_logic.conns.primary.thread_id()
Out[37]: 4267758

In [38]: interactive.cursor.connection == interactive.web_logic.conns.primary
Out[38]: False

In [39]: interactive.cursor.connection
Out[39]: <weakproxy at 0x3881c60 to Connection at 0x94c010>

In [40]: interactive.web_logic.conns.primary
Out[40]: <_mysql.connection open to 'xendb01' at 94c010>

How do I tell if the connections are the same ?

Answer 1

If the object is a standard weakref, you need to call it to get the object itself.

import weakref
class Test(object): pass
a = Test()
b = weakref.ref(a)
a is b() # True
a == b() # True

Using weakrefs here seems wrong, though: if I construct a connection, create a cursor from it, and discard the connection object, the cursor should remain valid. There shouldn't be a circular dependency unless the connection is keeping a list of all cursors, in which case that is what should be the weakref.

Answer 2

I've long found weakref.proxy's design and implementation to be somewhat shaky. Witness...:

>>> import weakref
>>> ob=set(range(23))
>>> rob=weakref.proxy(ob)
>>> rob==ob
False
>>> rob.__eq__(ob)
True

...DEFINITELY peculiar! In practice what I use from weakref are weak-key or sometimes weak-value dictionaries; but weakref.ref is sounder than the proxy wrapper on top of it:

>>> wr=weakref.ref(ob)
>>> wr()==ob
True

The need to "call" the ref to get the object (or None if the object has since disappeared) unfortunately makes it non-transparent (so a DB API module couldn't do it while staying compliant to the API). I don't understand why MySqlDb wants weak cursor->connection referencing at all, but if they do I see why they felt they had to use proxies rather than refs. However, one pays a very high price for that transparency!

Btw, the "explicit __eq__" trick (or an equivalent one with __cmp__, depending on the type of the underlying object) may help you, even though it's definitely inelegant!

Answer 3

Wrap the non proxy with weakref.proxy and use the identity operator:

>>> interactive.cursor.connection is weakref.proxy(interactive.web_logic.conns.primary)
True

Calling weakref.proxy() twice will return the same proxy object.