Re: [Pyobjc-dev] Access to 'hidden' class-methods

[Ronald O. <ous...@ci...>]

On Sunday, Jan 12, 2003, at 05:17 Europe/Amsterdam, bb...@ma... wrote:

> On Saturday, Jan 11, 2003, at 17:31 US/Eastern, Ronald Oussoren wrote:
>> Some time ago there was a discussion about access to class-methods 
>> with the same name as an instance-method. Currently this is not 
>> possible.
>>
>> One of suggestions was to add pyobjc-specific attributes to proxy 
>> objects. I've implemented, but not yet fully tested this. After I 
>> check this in you can do:
>>
>> 	NSObject.pyobjc_classMethods.description()	# Call +description
>> 	a = NSObject.new()
>> 	a.pyobjc_classMethods.description()  # Also call +description
>> 	a.pyobjc_instanceMethods.description() # Call -description
>>
> Could this be simplified by requiring the developer to go through 
> class() to get to the class method?
No, that's why one of the tests in Lib/objc/tes/test_objc.py is 
failing. If a class has a class and instance method with the same name, 
the former will be inaccessible from Python using the normal 
mechanisme. In python, attributes are first looked up in the __dict__ 
of the instance and then in the __dict__ of its class (and 
superclasses). All methods are only present in the class dict.

Note that my patch doesn't change the default behaviour, 
NSObject.alloc() will still work.

>
> NSObject.pyobjc_classMethods.description() # invoke class method
> a = NSObject.new()
> a.class().description() # invoke class method
> a.description() # invoke instance method
>
> This doesn't solve the overall problem that Ronald's solution 
> addresses, but it would work most of the time [I think].
>
>> I've not yet added an pyobjc_instanceMethods attribute to 
>> class-proxies. Another restriction is that reflection doesn't work 
>> yet for the contents of these attributes, 
>> dir(NSObject.pyobjc_classMethods) will not return a list of class 
>> methods.
>
> Since we have to go through this extra hoop anyway, maybe the answer 
> is to hang the classMethods and instanceMethods attribute off of the 
> class object?
>
> a.class().pyobjc_classMethod('description')
> a.class().pyobjc_instanceMethod('description')
>
> Actually -- maybe the right thing to do is to bridge the following two 
> methods on NSObject...
>
>     - (IMP)methodForSelector:(SEL)aSelector;
>     + (IMP)instanceMethodForSelector:(SEL)aSelector;
>
> ... such that they return unbound class/instance methods, as 
> appropriate.  This would make it possible to do:
>
>     a.class().methodForSelector_("description")(a.class())
>     a.class().instanceMethodForSelector("description")(a)
>
> May look odd, but it is certainly 'pythonic' in that returning a 
> callable from a method/function is fairly common practice.
>
> This also obviates the need for doing introspection at the level of 
> generating an inventory of methods for the class with some 
> categorization.

Sounds good, but methodForSelector: is simular to description in that 
it is both a class and instance methods, and in your example your using 
the class method. Because of the way the bridge and python are working 
you cannot access the class-method at the moment.

Ronald