Re: Q: wrappers push/pull and models

From: Joseph A. Knapka <[email protected]>
> Samuele Pedroni wrote:
> >
> > > Samuele Pedroni wrote:
> > > >
> > >
> > > [scissors of brevity]
> > >
> > > > indeed but not messing with vs. keeping them alive through storing are
> > > > different things. It seems (but maybe I'm wrong) that the current code
> > needs
> > > > the latter, which would be complexity "pushed" down in the wrappers
again.
> > >
> > > Are you concerned with models being GC'd? I would think that if
> > > a model is used, the application would keep a reference to it,
> > > so Anygui doesn't need to worry about that.
> > >
> >
> > Honestly if I set a property to a model in a component a would aspect the
> > component to keep it alive. But anyway the point was another and I was
> > imprecise. The problem:   if the proxy and the wrapper don't keep a
reference
> > to the model and just keep the value or forget about the model how can they
> > behave properly, i.e. e.g. update the model.
> >
> > Maybe the code is correct and it is just me missing something.
>
> OK, I understand. I think you are right and the code is wrong,
> since AFAIK you are the first person to bother about getting
> model behavior correct.

To be precise <wink>, I was trying to look at how optimize the call chains
starting from Attrib.__g/setattr__, and found that Attrib.__setattr__ generates
a pull because of models and that behavior is probably wrong, and thought that
things first should be made work because optimizing not-working code is
pointless <wink>.

> > But if I set a
> > sofar-not-set property to a model it seems that following happens:
> >
> > model goes in proxy state, model is pushed down to wrapper, wrapper sets
model
> > value (if it can deal with model) into widget and forget about model.
> >
> > Now if I try to set the property again: first property value is pulled from
> > wrapper, wrapper get the value from widget (this is the aformen. plain
value),
> > plain value is put in state,  thus any notion of prop being bound to a
model is
> > lost...
> >
> > The relevant code is:
> > Attrib __setattr__ calls set that calls rawSet which does:
> >
> >             old_val = getattr(self, key, None)
> >
> > that triggers the discussed behavior through Attrib __getattr__ that
invokes:
> >
> >                     self.pull(*(name,))
> >
> > triggering in Proxy.pull:
> >
> >         self.wrapper.pull(state)
> >         self.state.update(state)
> >
> > now  wrapper.pull knows nothing about models and simply does with state:
> >
> >       for attr in [k for k in state.keys() if k not in unhandled]:
> >             state[attr] = lstate[attr]
>

The code in Models.py seems just sketched or at least not up-to-date (?), we
would need something along the line of:
  if state[attr] is a model:
    set state[attr] value to lstate[attr] without triggering a push

(( it was something like state[attr]._set_value() and maybe should become
.setValue() ))

I think for the moment we can just keep in mind that that line as it is cannot
deal with models.

IMO
            old_val = getattr(self, key, None)
            try: old_val.removed(self, key)
            except: pass
in rawSet can be safely rewritten as:

  if self.state.has_key[key]:
     try: self.state[key].removed(self, key)
     except: pass

then modattr (in Attribs.py) needs also to be revised.

Anyway all this seems to me too prone to throw exceptions on common paths for
non-models. Maybe we need to keep a ismodel := { 'name': 0/1 ... } dictionary
parallel to state...

regards.



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf