[cmaloney]

The tp_new slot is inherited. Because the base type name was embedded in the errors this meant argument errors showed as the base type rather than the subclass. Change them to be the subclass name.

While tp_name would be slightly faster to use it includes the module name like _thread.RLock() which adds confusion. Use _PyType_Name(type) which returns RLock(). Include a test to check.

_PyArg_NoPositional() only short-circuits when args is NULL, which tp_new never gets; use _PyArg_NoPositionalStack() to keep the name lookup off the success path.

The vectorcall cases keep using constant strings since that slot does not inherit.

---

This is needed to keep parity moving frozenset to Argument Clinic I think in general it makes these error messages a little better for new.

[cmaloney]

Changed my mind on this, pure python is the other way and I'd argue AC should match that:

>>> class Base:
...     def __new__(cls, a): ...
... class Sub(Base): pass
... 
... Sub(1, 2)  # Base.__new__() takes 2 positional arguments but 3 were given
... Sub(x=1)   # Base.__new__() got an unexpected keyword argument 'x'
... 
Traceback (most recent call last):
  File "<python-input-0>", line 5, in <module>
    Sub(1, 2)  # Base.__new__() takes 2 positional arguments but 3 were given
    ~~~^^^^^^
TypeError: Base.__new__() takes 2 positional arguments but 3 were given