Re: [Pyobjc-dev] Deprecation warnings turned into BadPrototypeError

[Ronald O. <ron...@ma...>]

 
On Monday, April 14, 2008, at 10:31AM, "Luc Heinrich" <lu...@ho...> wrote:
>Greetings,
>
>Revision 1993 of PyObjC trunk seems to change the "Not all Objective-C  
>arguments are present in the Python argument-list of foo" deprecation  
>warnings to hard BadPrototypeError exceptions, which makes my  
>application to fail on launch because of some decorated methods which  
>use variable arguments and/or keyword arguments.
>
>The problem was already discussed here and the diff in pyobjc-core/ 
>Modules/objc/libffi_support.m seems to show that work has been done to  
>better detect false positives.
>
>However, the following test in PyObjCFFI_MakeFunctionClosure...
>
>else if (stubUserdata->argCount == 0 && haveVarArgs && haveVarKwds)
>
>...should probably be...
>
>else if (stubUserdata->argCount == 0 && (haveVarArgs || haveVarKwds))
>
>...right ?

Maybe. The code is designed to trigger on definitions like this: def wrapper(*args, **kwds), which seems to be the most common way to define wrapping decorators.  The intention is to avoid varargs declarations other than wrapping decorators. 

Do you use "def wrapper(*args):" instead? 

BTW. Lib/objc/test/test_method_prototypes.py contains the unittests that are most relevant for this discussion.

Ronald

P.S. I just noticed that vararg definitions shouldn't be completely forbidden, but should be allowed when other metadata says that the ObjC method is actually a varargs method with an interface that PyObjC can deal with. The trunk doesn't support this at the moment :-(