ffc team mailing list archive

[Johan Hake <johan.hake@xxxxxxxxx>]

On Tuesday April 26 2011 08:50:31 Martin Sandve Alnæs wrote:
> On 26 April 2011 17:47, Martin Sandve Alnæs <martinal@xxxxxxxxx> wrote:
> > On 26 April 2011 17:08, Johan Hake <johan.hake@xxxxxxxxx> wrote:
> >> On Tuesday April 26 2011 01:01:55 Martin Sandve Alnæs wrote:
> >> > I'm not sure if this is safe or even solves anything.
> >> > If there are circular references a.b = b; b.a = a,
> >> > a.__del__ won't be called if the reference from
> >> > b.a is still intact, since a.__del__ is called when
> >> > the reference count of a goes to 0. Adding the
> >> > __del__ function will also stop the cyclic reference
> >> > detection in Python from cleaning up. That is my
> >> > understanding after reading
> >> > http://docs.python.org/reference/datamodel.html
> >> > just now. Correct me if I'm wrong!
> >> 
> >> If the above situation was correct you would be right. As it is now a
> >> preprocessed form keeps a reference to form_data and form_data keeps a
> >> reference to the preprocessed form => Ciruclar dependency.
> > 
> > I.e. a.b = b; b.a = a.
> > 
> > I tried to break that by deleting the preprocessed form from its
> > form_data,
> > 
> >> when the original form is deleted (not the preprocessed one.) When this
> >> happens the preprocessed form will be deleted (I guess).

We now have:

  a.b = b
  b.a = a
  c.a = a

when c is deleted it deletes b from a. So I guess it would work.

> > A guess is not enough :)
> > 
> > I don't feel confident about the __del__ patch.
> > As you say, removing the circular references sounds like a better
> > solution.

Yes, that would be much better!

> A possible tool for breaking the cycle:
> http://docs.python.org/library/weakref.html

Yes, that would help.

Johan

> Martin