PyType_GenericAlloc doesn't initialize the content to NULL

[khatharsis42]

Bug report

Bug description:

I spent a few hours yesterday debugging this, and thus thought I had to point it out.
Here's my simplified C++ code:

struct Example {
   // Lots of stuff, pointers, stuff like that
}

typedef struct {
  PyObject ob_base;
  struct A Example;
} ExampleObject;

static PyTypeObject ExampleType = {
  .ob_base = PyVarObject_HEAD_INIT(NULL, 0).tp_name = "my_module.Example",
  .tp_basicsize = sizeof(ExampleObject),
  .tp_flags = Py_TPFLAGS_DEFAULT,
  .tp_new = PyType_GenericNew,
};

static PyObject * foo(PyObject* self, PyObject * args) {
     auto* temp = PyObject_New(ExampleObject, &ExampleType);
    // The issue is here
}

The issues lies in the fact that, according to the documentation, the content of temp should be initialized at NULL, but that is clearly not the case here: I printed each byte of temp, and some of them clearly were not NULL.
This silent error caused me many a troubles. I fixed this by using memset, but I really shouldn't have to, so please fix this.
Thanks !

CPython versions tested on:

3.13

Operating systems tested on:

Linux

[ericvsmith]

I think you're saying that you expect PyObject_New to zero out the allocated memory (since you're referring to temp). But its documentation at https://docs.python.org/3.13/c-api/allocation.html#c.PyObject_New explicitly says "Fields not defined by the Python object header are not initialized."

[ZeroIntensity]

PyObject_New is strictly an allocator, typically used by tp_new (PyType_GenericNew in this case). It's not something to create usable instances, you'll want something like PyObject_CallNoArgs to do that.

We're already working on improving the object lifecycle documentation, so hopefully there'll be less confusion around allocators sometime soon.

I fixed this by using memset, but I really shouldn't have to

And will be wrong! memset will also zero-out the object's type and reference count, which you really don't want to do.

[khatharsis42]

@ericvsmith Indeed, it seems I got lost in the documentation there. Since I was using C++, it simply didn't occur to me that a function called "new" wouldn't call a constructor / set the memory to 0. I should have read more.
@ZeroIntensity thanks for the tip ! I only did the memset on the non-python part of my struct, don't worry ;)

Good luck on improving the documentation !