pickle.dump vs. yaml.dump vs. pydrake yaml_dump, accidental serialization

accidental_serialization.py

from enum import Enum
import io
import pickle

import numpy as np
import torch
import yaml

from pydrake.common.yaml import yaml_dump

def pickle_dump(obj):
  f = io.BytesIO()
  pickle.dump(obj, f)
  return f.getvalue()

class MyEnum(Enum):
  Value = 1

np_value = np.array([1.0, 2.0])
torch_value = torch.tensor([3.0, 4.0])

messy = {
  "enum": MyEnum.Value,
  "np": np_value,
  "torch": torch_value,
}
cleaned = {
  "enum": messy["enum"].value,
  "np": messy["np"].tolist(),
  "torch": messy["torch"].numpy().tolist(),
}

print("pickle.dump")
size = len(pickle_dump(messy))
print(f"  size: {size}")
size = len(pickle_dump(cleaned))
print(f"  size (cleaned): {size}")
print()

print("yaml.dump")
size = len(yaml.dump(messy))
print(f"  size: {size}")
size = len(yaml.dump(cleaned))
print(f"  size (cleaned): {size}")
print()

print("pydrake yaml_dump")
try:
  size = len(yaml_dump(messy))
  print(f"  size: {size}")
except Exception as e:
  print(e)
size = len(yaml_dump(cleaned))
print(f"  size (cleaned): {size}")

"""
Output:

pickle.dump
  size: 609
  size (cleaned): 82

yaml.dump
  size: 1113
  size (cleaned): 43

pydrake yaml_dump
('cannot represent an object', <MyEnum.Value: 1>)
  size (cleaned): 41
"""