Serialization

Copy

• copy.copy(d) returns a shallow AttributeDict (nested values shared).
• copy.deepcopy(d) returns a deep AttributeDict; nested AttributeDict instances stay AttributeDict. Self-references and cycles terminate and are preserved.

import copy

d = AttributeDict(a=1, nested=AttributeDict(x=1))
shallow = copy.copy(d) # shallow.nested is d.nested
deep = copy.deepcopy(d) # deep.nested is not d.nested; same value

Pickle

pickle.dumps/pickle.loads are supported across all protocols (0–5), preserving structure, the AttributeDict type, and reference cycles:

import pickle

d = AttributeDict(a=1, nested=AttributeDict(x=[1, {"y": 2}]))
loaded = pickle.loads(pickle.dumps(d))
# type(loaded) is AttributeDict; type(loaded.nested) is AttributeDict

How it works: AttributeDict.__reduce__ returns (reconstruct, (cls), None, None, iter(self.items)) — the 5-tuple form lets pickle apply items lazily through its memo, so cyclic structures pickle without recursion. attributedict._pickle_support.reconstruct creates the empty instance.

JSON / YAML / dataclasses

Interop with dataclasses / pydantic / SQLAlchemy / TypedDict is supported (see compatibility.md); the caveat is that those layers normalize an AttributeDict to a plain dict. For JSON, converting to a plain dict first remains the cleanest path:

import json
json.dumps(dict(d))