Dataclass takes no arguments in a subinterpreter in 3.12

Bug report

Bug description:

In 3.12, when I construct an instance of a dataclass in a subinterpreter, I get an "X takes no arguments" error, for cases that work fine in the main interpreter.

A fairly minimal code example:

#include <stdio.h>

#include <Python.h>

const char* code = "\n\
from dataclasses import dataclass\n\
\n\
@dataclass\n\
class MyClass:\n\
  id: str\n\
\n\
c = MyClass(id='abc')\n\
print(c)\n\
";

int main(int argc, char** argv) {
  Py_InitializeEx(0);
  printf("Python version: ");
  fflush(stdout);
  PyRun_SimpleString("import sys; print(sys.version)");
  PyThreadState* main_thread = PyThreadState_Get();

  printf("In main interpreter:\n");
  PyRun_SimpleString(code);

  PyThreadState* interpreter_thread = Py_NewInterpreter();
  printf("\nIn subinterpreter:\n");
  PyRun_SimpleString(code);
  Py_EndInterpreter(interpreter_thread);

  PyThreadState_Swap(main_thread);
  Py_Finalize();
  return 0;
}

Compiled with gcc -o py_dataclass py_dataclass.c -I /usr/include/python3.11 -lpython3.11 I get the expected:

Python version: 3.11.5 (main, Aug 25 2023, 13:19:50) [GCC 11.4.0]
In main interpreter:
MyClass(id='abc')

In subinterpreter:
MyClass(id='abc')

Compiled with gcc -o py_dataclass py_dataclass.c -I /usr/include/python3.12 -lpython3.12 I get:

Python version: 3.12.0 (main, Oct  2 2023, 15:04:50) [GCC 11.4.0]
In main interpreter:
MyClass(id='abc')

In subinterpreter:
Traceback (most recent call last):
  File "<string>", line 8, in <module>
TypeError: MyClass() takes no arguments

This is using the python3.11-dev and python3.12-dev packages from the deadsnakes PPA.

I'm a little lost about what the interaction between the subinterpreter and the dataclass is that would cause this. The non-mangled Python code is just

from dataclasses import dataclass

@dataclass
class MyClass:
  id: str

c = MyClass(id='abc')
print(c)

Let me know if there's any more info I can provide!

CPython versions tested on:

3.11, 3.12

Operating systems tested on:

Linux

An interesting datapoint; the following code does work as expected in the subinterpreter:

from dataclasses import dataclass

@dataclass
class MyClass:
  id: str

@dataclass 
class MySubClass(MyClass):
  pass

c = MySubClass(id='abc')
print(c)

@ericsnowcurrently

cc: @ericsnowcurrently

@ericvsmith

Any ideas, @ericvsmith?

FWIW, it seems to work fine on main. I was able to reproduce the failure on 3.12:

from test.support import interpreters

interp = interpreters.create()
interp.run("""if True:
    from dataclasses import dataclass

    @dataclass
    class MyClass:
      id: str

    c = MyClass(id='abc')
    print(c)
    """)

Ah nice, I didn't know that test module was available to try it out in pure Python.

Also working for me at head, same for our internal usage (which is a bit more complex than the reduced example here - there's something else going wrong there which I'm still digging through).

I have no idea what's going on there. I don't have handy access to my dev box right now, but maybe inspecting the signature for MyClass.__init__ will shed some light.

bisected to ba65a06 on 3.12
looks like it was fixed again on main in b72947a

Do you know if that made it to the 3.12 branch (so would be in 3.12.1)?

I had decided not to backport it. Clearly I need to revisit that decision. 😄

I just tried e658413 (the head of 3.12) and I still reproduce the failure

For what it's worth, this has broken using Flask in Apache with mod_wsgi, meaning I cannot upgrade to Python 3.12 until this is fixed. So please consider getting this into 3.12.1.

I've merged the backport fix. Could someone verify whether or not this is fixed now. Thanks!

@ericsnowcurrently

@ericsnowcurrently if you point me to the patch to try, I can do it. I have the same problems with mod_wsgi in a Django application.

I reported this in Fedora's bug tracker: https://bugzilla.redhat.com/show_bug.cgi?id=2252567

They made a build with the patch; I tested their build and confirmed that Flask apps no longer fail with the patch.

Does anyone know if this fix in included in 3.12.1?

The fix is indeed part of 3.12.1.

Thanks, also fixed for me!

The fix is in Fedora 39 as of a day or two ago.

Navigation Menu

Repository navigation

Dataclass takes no arguments in a subinterpreter in 3.12 #110279

Bug report

Bug description:

CPython versions tested on:

Operating systems tested on:

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Dataclass takes no arguments in a subinterpreter in 3.12 #110279

Description

Bug report

Bug description:

CPython versions tested on:

Operating systems tested on:

Activity

peterebden commented on Oct 3, 2023

AA-Turner commented on Oct 3, 2023

ericsnowcurrently commented on Oct 4, 2023

ericsnowcurrently commented on Oct 4, 2023

peterebden commented on Oct 4, 2023

ericvsmith commented on Oct 4, 2023

graingert commented on Oct 4, 2023

ericvsmith commented on Oct 4, 2023

ericsnowcurrently commented on Oct 4, 2023

graingert commented on Oct 4, 2023

anthonyfk commented on Nov 3, 2023

ericsnowcurrently commented on Nov 28, 2023

thierryba commented on Dec 6, 2023

uckelman commented on Dec 8, 2023

thierryba commented on Dec 8, 2023

ericsnowcurrently commented on Dec 8, 2023

peterebden commented on Dec 13, 2023

uckelman commented on Dec 19, 2023

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions