Georg,

Thanks for reposting my proposal. Since BDFL is +1 on the idea, I feel motivated to polish the patch.  In your early comment you noted that kwdict == NULL check may be superficial.  I agree.  It does not look like it is possible that ext_do_call is called with CALL_FLAG_KW set and a null pointer at the top of the stack.  It looks like this may only happen if the stack is corrupted by a misbehaving c extension.  In this case, however, throwing a type error is too gentle. PyErr_BadInternalCall() seems to be more appropriate if anything at all.

I propose to eliminate the null check for kwdict alltogether. This would be consistent with the way stararg is handled later in this function.  I am going to run the testsuit without the null check and will post a revised patch shortly.