BaseModel.copy produces copy that modifies the original

[physicsrob]

Bug

When copying a model, changing the value of an attribute on the copy updates the value of the attribute on the original. This only happens if deep != True. If this is expected behavior, I would suggest the documentation needs to be updated, because it's different than the intuition one would have for shallow copies of standard python objects.

My expectation of a shallow copy: Modifying a mutable attribute on the child SHOULD modify the mutable attribute on the parent (the attribute should be the same object). Setting the value of an attribute on the child, however, SHOULD NOT set the value on the original.

Output of python -c "import pydantic.utils; print(pydantic.utils.version_info())":

             pydantic version: 1.4
            pydantic compiled: True
                 install path: /Users/rob/opt/anaconda3/envs/words_ml/lib/python3.8/site-packages/pydantic
               python version: 3.8.2 (default, Mar 25 2020, 11:22:43)  [Clang 4.0.1 (tags/RELEASE_401/final)]
                     platform: macOS-10.15.4-x86_64-i386-64bit
     optional deps. installed: []

Reproduction:

from pydantic import BaseModel
from typing import Optional, List

class X(BaseModel):
    val: int

x = X(val=1)

y = x.copy()
y.val= 2

print(f"{x.val =}") # x.val =2
print(f"{y.val =}") # y.val =2 -- setting the value on the copy also set the value on the original?


# My expectation would be that the behavior would be similar to standard python objects:
x = {'val': 1}
y = x.copy()
y['val'] = 2
print(f"{x['val'] =}") # x.val =1
print(f"{y['val'] =}") # y.val =2

[StephenBrown2]

copy() has a deep kwarg that defaults to False, detailed in the docs: https://pydantic-docs.helpmanual.io/usage/exporting_models/#modelcopy
The behavior is similar to standard python objects, where a normal copy merely duplicates the reference (a shallow copy), and a deepcopy duplicates the data with a new reference: https://docs.python.org/3/library/copy.html
Here is the code that pydantic uses: https://github.com/samuelcolvin/pydantic/blob/master/pydantic/main.py#L528-L529

[physicsrob]

I disagree that this is similar to normal python behavior. To be clear, I'm talking specifically about shallow copying behavior. I agree that a standard python shallow copy copies the "reference". The difference is that for a normal object when you update the "reference" of a copy you do not update the "reference" of the original. If you modify what the "reference points" to, then of course that will modify what the reference points to in the original.

This is standard python behavior:

class myobj:
    def __init__(self, val):
       self.val = val
x = myobj(1)
import copy
y = copy.copy(x)
y.val = 2

print(f"{y.val =}") # y.val =2
print(f"{x.val =}") # x.val =1

This is pydantic behavior:

from pydantic import BaseModel

class X(BaseModel):
    val: int

x = X(val=1)
y = x.copy()
y.val = 2

print(f"{y.val =}") # y.val =2
print(f"{x.val =}") # x.val =2

[samuelcolvin]

I 100% agree this is weird, it's not what I would expect to happen, and it's a bug.

But it seems to have been inadvertently fixed in v1.5 (as yet unreleased).

To resolve this, I'm going to:

• submit a PR with tests to make sure this fixed and doesn't get broken again, and add a change description so it's in the release notes
• release v1.5 asap Version 1.5 release - what needs doing? #1341

[physicsrob]

Awesome! Thanks :)