[PEAK] Circular class adapters..

[Bob Ippolito]

On Feb 19, 2004, at 6:39 PM, Phillip J. Eby wrote:

> At 06:34 PM 2/19/04 -0500, Bob Ippolito wrote:
>> I have a use case where I'd like to have a class adapter chain that 
>> can
>> be circular in some cases, such as:
--- code that breaks ---
>> How do I make this work?  Is it not possible, or is this a bug in
>> PyProtocols?  With 0.9.2, I get this nasty traceback:
>
> Ugh.  It looks like advising an object (the object itself, as opposed 
> to its class) with a circular adapter path doesn't work right.  I have 
> no idea why, I'll have to investigate further (and add your example to 
> the tests.
>
> As a workaround in the meantime, you could give each class a 
> metaclass, and tell the metaclasses that their instances provide the 
> interfaces, instead of telling the class that the class provides the 
> interface.

That's what I figured.. I actually named all of my class interfaces 
"IMeta..." in anticipation of this kind of problem, so it will be easy 
to track it all down ;)

PyProtocols is great so far though.  I decided to port one of my 
libraries that depended on twisted.python.components (which I had to 
hack in order to make it do what I wanted), and the (theoretically 
still working, once this is resolved) code got a whole lot cleaner with 
PyProtocols due to neat tricks such as sequenceOf and a much more 
flexible design.

-bob