The change looks good, but function type seems not allowing to do it.
Could you please also add this error handling?

======================================================================
FAIL: test_delete___dict__ (test.test_funcattrs.FunctionDictsTest.test_delete___dict__)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/runner/work/RustPython/RustPython/pylib/Lib/test/test_funcattrs.py", line 360, in test_delete___dict__
    self.fail("deleting function dictionary should raise TypeError")
AssertionError: deleting function dictionary should raise TypeError

======================================================================
FAIL: test_setting_dict_to_invalid (test.test_funcattrs.FunctionDictsTest.test_setting_dict_to_invalid)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/runner/work/RustPython/RustPython/pylib/Lib/test/test_funcattrs.py", line 331, in test_setting_dict_to_invalid
    self.cannot_set_attr(self.b, '__dict__', None, TypeError)
  File "/home/runner/work/RustPython/RustPython/pylib/Lib/test/test_funcattrs.py", line 42, in cannot_set_attr
    self.fail("shouldn't be able to del %s" % name)
AssertionError: shouldn't be able to del __dict__

The related type is PyFunction.

Adding a getset is possible like:

// This impl must be one of `#[pyclass]` annotated
impl PyFunction {
    #[pygetset(setter)]
    fn __dict__(zelf: &Py<Self>, value: dict: PySetterValue<PyDictRef>,, vm: &VirtualMachine) -> ...
}

I found out, how to delegate the __dict__ access to the underlying object. This brings a new problem:

foo.__dict__ contains the keys __doc__, __module__ and __annotations__ in RustPython, but not in CPython (there it's just an empty dict).

I guess this happens, because execute_make_function in frame.rs assigns these three, but they are not explicitly added as members in PyFunction.

Should i add these three to the PyFunction struct? And if yes, in another pull request?

Thank you for analysis. That's a good point. Yes, making them as its own getset totally makes sense. And yes, in separated PR would be better.

I was able to remove __module__ and __annotations__ from the function __dict__, but the __doc__ does not go away, even if i implement a "pygetset" for it. The functions are not even getting called.

I declared them like this

#[pygetset(magic)]
fn doc(&self) -> PyObjectRef {...}

#[pygetset(magic, setter)]
fn set_doc(&self, value: PyObjectRef) {...}

So, something is intercepting the calls to my pygetset. Any idea what that might be? I looked at other pygetset for __doc__ but none of them is getting called either.

It also does not matter whether the function actually has a doc-string. If i comment out the part in compile.rs where it generates op codes to store the doc-string in the function then the doc-string of the function-type is used when writing foo.__doc__.

@nielsbuwen Making a PR of that changes will be helpful for me to understand what's happening. My first guess is execute_make_function in frame.rs

@youknowone here you go: #5392