[Zope-dev] im_self of methods accessed via non trivial acquisition

[Phillip J. Eby]

At 01:12 PM 1/18/02 +0100, Stefan Bund wrote:
>"Phillip J. Eby" <pje@telecommunity.com> writes:
> >
> > Method rebinding is done only when an item is retrieved from the aq_self
> > side, and only if im_self points to aq_self.  If these conditions are met,
> > a new binding is created which points to the acquisition wrapper.
> >
> > This is a feature, not a bug.  If you rebind im_self on the aq_parent side,
> > and your method assigns a value to an attribute of self, it will be
> > assigned to the wrong object!  It is safe in your example to bind m.im_self
> > to B, because B.aq_base is f, the true self of m.  But it is not safe to
> > bind m.im_self to A, because A.aq_base is g -- another object altogether.
> >
> > I guess another way to look at it is that a method retrieved from an
> > acquisition-wrapped object will always meet the condition that
> > method.im_self.aq_base is the original object the method was retrieved
> > from.  This ensures that the method simply works with an
> > acquisition-wrapped version of its original self.
>
>Of course, im_self cannot be A, that would be wrong, but as far as I
>understand, it could be a *new* acquisition wrapper, something like /B
>o A/ or even /[f] o A/:
>
>C--A--O--[a]
>|  |  |
>|  |  [b]
>|  |
>|  O--B--O--[c]
>|  |  |  |
>|  |  |  [d]
>|  |  |
>|  |  O--[e]
>|  |  |
>|  |  [f]
>|  |
>|  [g]
>|
>B--O--[c]
>|  |
>|  [d]
>|
>O--[e]
>|
>[f]
>
>(that's what I refer to in the definition of /C/). So if /C/ is one of
>the aforementioned new wrappers, /C.aq_base/ is indeed [f] and
>therefore, /C/ would be safe as im_self?
>
>Hmm. Is there another reasun, why method rebinding works the way it
>does (returning a subtree and not a new acquisitionwrapper like
>described above) or am I completely amiss?

Beats me.  I guess nobody's asked for the feature, and in any event it 
requires global knowledge of the tree structure that would be very 
difficult to code.  The current behavior is a rule that is applied locally 
at each acquisition wrapper, and the wrappers don't know what wrapper is 
pointing to them, so it's really not practical for them to try and do 
something that "smart".

It may also be that what you're asking for is a bad idea, in which case Jim 
Fulton can probably explain why.  :)