Re: [Pyobjc-dev] pickled Objective-C objects

[Just v. R. <ju...@le...>]

Ronald Oussoren wrote:

> I noticed the other day that our unicode subclass doesn't support
> pickling. Should we add support for pickling, and if so how? Should
> instances of our subclass be pickled just like 'normal' unicode
> objects (although I have no idea how to do that), or as NSString
> values?

I wouldn't care much either way, although I don't see much use in
pickling the NSString portion; might as well convert to a unicode
object. But ditto: I have no idea how to do that...

> Normal Objective-C classes also don't support pickling, [ ... ]

Synchronicity! I was thinking about that today. I have written a little
app that constructs hundreds of NSBezierPaths at app startup time; I
gather it would save a lot of time if I could just load them from a
pickle. Would be way cool. For now I think I'll play with
NS[Un]Archiver. I don't have a use case, but the reverse would also be
very cool: the ObjC wrappers/proxies for Python objects could map the
NSCoding protocol to the pickle protocol.

> should we add
> a reduce method to the proxy that calculates the object state using
> NSCoder (where possible)?

That sounds fine.

> It might also be possible to provide a
> default implementation of initWithCoder: and encodeWithEncoder: if
> the parent class implements those.

> The default encodeWithEncoder: implementation would call the
> superclass implementatation, then a NSDictionary containing the
> Objective-C instance variables and python __slots__ and then an
> NSDictionary containing the python __dict__.
> 
> The __reduce__ method would use the result of [self
> encodeWithEncoder:] as the object state, and the restore function
> would use that state to restore the object using initWithCoder:.
> 
> Does this look like a valid approach? If it is I'll probably file a
> bug report containing this scheme (I don't think I'll get around to
> implementing this anytime soon).

The __reduce__ part of the pickle protocol is completely new to me, so I
can't offer any advice at the moment. I'll try to find the time to look
into it. Opening a feature-request-bug is definitely a good idea.

Just