I've added a supports_weakref attribute to ClassIR and a check to add a weakref slot in has has_dict is false but supports_weakref is True

I have not yet added any logic for setting supports_weakref, it is currently always set == cls.has_dict. I'm thinking to implement one or more of the options listed here: mypyc/mypyc#1102

Something in this PR has broken your serialization tests, it would be helpful if somebody could share some pointers wrt where I should look.

Well... I tried fixing the serialization issue and now I just get segfaults instead. Not sure where to go from here.

Great, the segfaults are fixed and the serialization tests are green

Next steps:

• add tests for mypyc_attr(supports_weakref=True)
• add tests for has_dict=False but supports_weakref=True

Distant todo (maybe?):

• Detect weakref attr defined in init and set ir.supports_weakref=True

Hell yeah, it works!
@JukkaL let me know what you think, and if the additional test cases are even necessary given that these newly added tests all pass and the to_dict classes didn't break

TODO: I found this in cpython's docs:

The only further addition is that tp_dealloc needs to clear any weak references (by calling PyObject_ClearWeakRefs()) if the field is non-NULL:

static void
Trivial_dealloc(TrivialObject *self)
{
    /* Clear weakrefs first before calling any destructors */
    if (self->weakreflist != NULL)
        PyObject_ClearWeakRefs((PyObject *) self);
    /* ... remainder of destruction code omitted for brevity ... */
    Py_TYPE(self)->tp_free((PyObject *) self);
}

https://docs.python.org/3.8/extending/newtypes.html#weak-reference-support

I added some logic to the dealloc function as per the python docs above, it fails on all python versions

for python 3.8-3.11 it fails due to the if check, and the fact that weakreflist is not a member of the struct. I believe this can either be fixed by adding the member to the struct (difficult, I noticed you have this commented in the code as a #TODO) or looking up the weakref field instead (potentially easier). I am blocked by not knowing how to access the weakref field here, or whether we can even access a python attribute in the dealloc function.

for python 3.12 and 3.13 I think it segfaults because we need to #include weakref.h so we can call PyObject_ClearWeakRefs. I'm not sure of this however, and don't know how to emit the #include so I can test my theory.

Leaving this PR as-is for now, will come back in a few days and work on it some more.