dataclass, slots, __post_init__ and super

[voidspace]

Bug report

Bug description:

Using super() in __post_init__ on a dataclass with slots fails.

This code:

from dataclasses import dataclass

@dataclass(slots=True)
class Base:
    def __post_init__(self):
        pass

@dataclass(slots=True)
class Thing(Base):
    a: int
    b: int
    c: int = 0

    def __post_init__(self):
        self.c = self.a + self.b
        super().__post_init__()

t = Thing(1,3)

Produces this exception:

Traceback (most recent call last):
  File "C:\Users\michael.foord\Code\one-touch-switch-service\example.py", line 18, in <module>
    t = Thing(1,3)
        ^^^^^^^^^^
  File "<string>", line 6, in __init__
  File "C:\Users\michael.foord\Code\one-touch-switch-service\example.py", line 16, in __post_init__
    super().__post_init__()
    ^^^^^^^
TypeError: super(type, obj): obj must be an instance or subtype of type

The fix is to use Python 2 style super(Class, self).__post_init__().

CPython versions tested on:

3.11

Operating systems tested on:

Windows

Linked PRs

• gh-90562: Fix super() without args calls for dataclasses with slots #111538

[sobolevn]

I think that I found why this happens :)

First of all, I was not able to reproduce it without @dataclass, so I assumed it is related to the internal machinery.

This is my debug patch:

» git patch
diff --git Lib/dataclasses.py Lib/dataclasses.py
index 2fba32b5ffb..a7ce3c03e9b 100644
--- Lib/dataclasses.py
+++ Lib/dataclasses.py
@@ -1232,7 +1232,7 @@ def _add_slots(cls, is_frozen, weakref_slot):
 
     # And finally create the class.
     qualname = getattr(cls, '__qualname__', None)
-    cls = type(cls)(cls.__name__, cls.__bases__, cls_dict)
+    cls = type(cls)(cls.__name__ + '__slots', cls.__bases__, cls_dict)
     if qualname is not None:
         cls.__qualname__ = qualname
 
diff --git Objects/typeobject.c Objects/typeobject.c
index 25085693070..79db15579e5 100644
--- Objects/typeobject.c
+++ Objects/typeobject.c
@@ -10400,9 +10400,11 @@ supercheck(PyTypeObject *type, PyObject *obj)
         Py_XDECREF(class_attr);
     }
 
-    PyErr_SetString(PyExc_TypeError,
+    PyErr_Format(PyExc_TypeError,
                     "super(type, obj): "
-                    "obj must be an instance or subtype of type");
+                    "obj must be an instance or subtype of type %s %s",
+                    Py_TYPE(obj)->tp_name,
+                    type->tp_name);
     return NULL;
 }
 

Turns out, I was right. Given the code in the original bug report, it produces:

Traceback (most recent call last):
  File "/Users/sobolev/Desktop/cpython2/ex.py", line 32, in <module>
    Thing2(1, 2)
  File "/Users/sobolev/Desktop/cpython2/ex.py", line 19, in __init__
    self.__post_init__()
  File "/Users/sobolev/Desktop/cpython2/ex.py", line 24, in __post_init__
    super().__post_init__()
    ^^^^^^^^^^^^^^^^^^^^^
TypeError: super(type, obj): obj must be an instance or subtype of type Thing__slots Thing

Looks like we add slots and regenerate class in the very end. I think we can try doing it at the beggining. Maybe it will help.

I am working on this idea! :)

[sobolevn]

Here are some experiments that I had:

from dataclasses import dataclass
# from attr import dataclass

@dataclass(slots=True)
class Base:
    def m(self):
        return "DONE!"

class Thing(Base):
    def m(self):
        print(super().m())

Thing2 = dataclass(slots=True)(Thing)
Thing2().m()

It fails with the same error. However, if you pip install attrs and uncomment # from attr import dataclass it will work and print DONE!

So, it can be fixed (given that our _add_slots functions are similar): https://github.com/python-attrs/attrs/blob/52c1cd97d723cde6d75259a22d218f9305ecf8ee/src/attr/_make.py#L809-L914

[sobolevn]

Here's the fix: https://github.com/python-attrs/attrs/blob/52c1cd97d723cde6d75259a22d218f9305ecf8ee/src/attr/_make.py#L884-L912 :)

[voidspace]

Nice work @sobolevn

[JelleZijlstra]

Duplicate of #90562