Adding instruction to "helper.literal_eval_extended" for comparing custom object

[atah1991]

Hi! Thanks for this great lib! I happened to get a bug on a custom case when I compare dictionaries where key is the dataclass(simplified example below). The output of the DeepDiff shows the path to fields as None.

@dataclass(frozen=True)
class ClassA(object):
    a: int = 1
    

@dataclass(frozen=True)  
class ClassB(object):
    b: int = 2


left = {ClassA(): "ClassA"}
right = {ClassB(): "ClassB"}
DeepDiff(left, right, verbose_level=2)

The script above outputs to

stringify_param was not able to get a proper repr for "ClassB(b=2)". This object will be reported as None. Add instructions for this object to DeepDiff's helper.literal_eval_extended to make it work properly: malformed node or string: <_ast.Call object at 0x7f5f3abfa3a0>
stringify_param was not able to get a proper repr for "ClassA(a=1)". This object will be reported as None. Add instructions for this object to DeepDiff's helper.literal_eval_extended to make it work properly: malformed node or string: <_ast.Call object at 0x7f5f7845fc70>
{'dictionary_item_added': {None: 'ClassB'},
'dictionary_item_removed': {None: 'ClassA'}}

After surfacing the code, I understand that so far support is for the following cases in helper:

LITERAL_EVAL_PRE_PROCESS = [
    ('Decimal(', ')', _eval_decimal),
    ('datetime.datetime(', ')', _eval_datetime),
    ('datetime.date(', ')', _eval_date),
]

Is there a way for the user to add the custom instruction to this list and pass it in DeepDiff interface so the resulted diff shows the path (not None) to fields that are different? Or maybe there is a better way to have comparison for such objects? Note, I would like to compare them as dictionaries, instead of iterating over keys, values and comparing them

Thanks!

[atah1991]

Any comments or suggestions on the above? That would extremely help for my current project!

[seperman]

Hi @atah1991
The problem here is that deepdiff wants to store some string presentation of your paths but also wants to make sure the string representation can be converted to your actual path. Due to security reasons, we don't allow it to "eval" arbitrary code.

Hence right now the eval is limited to very few things in the list you have posted (LITERAL_EVAL_PRE_PROCESS) other than what Python's restricted eval (literal_eval) allows.

The problem is not specific to the data classes. If you even remove the data class decorators, you will still run into this issue when your keys are custom objects.

Maybe in the future we can add a flag to allow eval instead of restricted eval. Then it will work better with cases like this when users are sure the data they are diffing doesn't contain Python code lines.

[emonsler]

Hi,

@atah1991 , did you find any workaround for this?

My use case has modifying helper.py as problematic. I was getting the same warning for classes deriving from enum.Enum.

I seem to be able to work around the issue by setting a repr for the class, returning

'"' + FULLY_QUALIFIED_PATH + "ModelB." + self.name + '"'

So it displays as

"some.org.atah1991.ModelB.a"

[seperman]

We do allow a "safe" list for Delta objects. We can use the same logic to allow eval to be run on paths too.
https://zepworks.com/deepdiff/current/delta.html#delta-safe-to-import-parameter
I don't currently have the bandwidth to look into it. PRs are welcome!