Re: [Swig-user] AttributeError: No constructor defined. Why?

[William S F. <ws...@fu...>]

FermiPhysics wrote:
> 
> 
> Josh Cherry wrote:
>>
>>
>> On Fri, 31 Jul 2009, FermiPhysics wrote:
>>
>>> Looking at the SWIG-generated .py module, the __init__ function simply
>>> says:
>>>
>>> def __init__(self, *args, *kwargs): raise AttributeError("No constructor
>>> defined")
>>>
>>> The constructor is defined in the source, and declared in the header as
>>> well
>>> as the SWIG input (.i) file.  Anyone had a similar problem?
>> Have a look at http://www.swig.org/Doc1.3/SWIGPlus.html#SWIGPlus_nn9.
>>
>> It might be nice if SWIG generated different exception text in the case of 
>> abstract classes, e.g., "Class is abstract".
>>
Good idea. If one of the reasons the constructor is missing because the 
class is abstract, the code generated is now:

     def __init__(self, *args, **kwargs): raise AttributeError("No 
constructor defined - class is abstract")


> That's what I initially thought.  I tried running swig with the -Wall option
> so that it would raise a warning if it thought my class were abstract, and
> there were no warnings.
> 
> The class that has the __init__ function which raises the error is derived
> from another C++ class.  If SWIG thought the BASE class were abstract, would
> it raise a warning with the -Wall option?  That may be my problem.
> 
No it does not raise a warning. There is no reason to do so, the 
designer of the C++ class intended it to be this way and the c++ 
compiler doesn't issue such a warning, it gives an error when attempting 
to invoke the method, so SWIG follows this approach.

William