I have a dict subclass like this:

class MyDict(dict):def __getitem__(self, name):return globals()[name]

This class can be used with eval and exec without issues:

>>> eval('bytearray', MyDict())
<class 'bytearray'>
>>> exec('print(bytearray)', MyDict())
<class 'bytearray'>

But if I instantiate a function object with the types.FunctionType constructor, the function can't access any builtins:

import typesfunc = lambda: bytearray
func_copy = types.FunctionType(func.__code__,MyDict(),func.__name__,func.__defaults__,func.__closure__)print(func_copy())
# Traceback (most recent call last):
#   File "untitled.py", line 16, in <module>
#     print(func_copy())
#   File "untitled.py", line 8, in <lambda>
#     func = lambda: bytearray
# NameError: name 'bytearray' is not defined

Replacing MyDict() with globals() or dict(globals()) or event {'__builtins__': __builtins__} makes the code print <class 'bytearray'> as expected.

I don't understand where this exception is coming from. Can anyone explain this behavior? Why does it work with eval but not with a function object?

Not a complete answer, but what seems to be happening is that CPython ignores the custom __getitem__ when it accesses the builtins. It seems to treat MyDict like a normal (not subclassed) dict. If the '__builtins__' key is actually present in the dict, then everything works correctly:

class MyDict(dict):def __getitem__(self, name):return globals()[name]import typesglobs = MyDict()
globs['__builtins__'] = __builtins__func = lambda: bytearray
func_copy = types.FunctionType(func.__code__,globs,func.__name__,func.__defaults__,func.__closure__)print(func_copy())
# output: <class 'bytearray'>

The question remains why this only happens with FunctionType, and not with eval and exec.