Message3182

And here is a change to the PyDerivedDict#__cmp___ that make us pass the
TestSuite. 

    public int __cmp__(PyObject other) {
        PyType self_type=getType();
        PyType[] where_type = new PyType[1];
        PyObject impl = self_type.lookup_where("__cmp__", where_type);
        // Full Compatibility with CPython __cmp__:
        // If the derived type don't override __cmp__, the 
        // *internal* PyDictionary#__cmp__ should be called, not the
exposed 
        //  one. The difference is that the exposed dict.__cmp__ throws a 
        // TypeError if the argument is an instance of dict.   
        //  
        // BTW, impl should never be null, because, if __cmp__ has not been 
        // overidden,  it should be the *exposed*
PyDictionary#dict___cmp__. 
        // But checking for that won't hurt.
        if (impl == null || where_type[0] == PyDictionary.TYPE) {
        	return super.__cmp__(other);
        }
        PyObject res=impl.__get__(this,self_type).__call__(other);
        if (res instanceof PyInteger) {
            int v=((PyInteger)res).getValue();
            return v<0?-1:v>0?1:0;
        }
        throw Py.TypeError("__cmp__ should return a int");
    }

That was before I was told that the method was generated by gderived.py.
Now I don't know if I should adapt the object.derived template to
generate similar code, or not. Thoughts?