Overriding Dictionary Object

[Christopher T King]

On Mon, 26 Jul 2004, Adonis wrote:

> I am subclassing the dict object, and having it serialize itself to disk for
> later access, but I am having some trouble with accessing the self._filename
> attribute, when the object is first created it functions fine, but when I
> access it on a later time, self._filename is not present, how can I go about
> this knowing that when objects are serialized they are not initialized on
> subsequent access? I was under the impression that everything in an objects
> scope was serialized by pickle, or is my assumption wrong?

Your assumption is correct -- the problem is elsewhere in your code:  the
line 'self = cPickle.load(data)' doesn't do what you think it does -- this
only sets the local variable 'self' to the object returned by load(), not 
change the object itself.  Hence, self._filename is setting the attribute 
of a soon-to-be-out-of-scope object, not the current object.

To get the effect you want, you would have to implement the __new__ method
of the object.  The __new__ method is supposed to return a new,
uninitialized instance of the class, but you can have it return an old
instance instead (though you must make sure it doesn't get initialized).

However, I suggest you try other ways of accomplishing your goal.  Writing 
the entire dictionary to disk every time one of its values is changed is 
very inefficient, and will also cause obscure bugs to appear, assuming 
it's used in a threaded context (e.g. a web server).  The dbhash module 
provides an efficient, thread-safe mechanism that does the same thing as 
your class:

 import dbhash

 data = dbhash.open('appweb.data','cl')	# 'cl' means create if necessary,
					# and use automatic locking

 data['foo']='bar'	# set a new value
 print data['apple']	# retrieve an old value
 data.close()		# all done here