decide about `try_instance`

[pydantic]

decide about `try_instance`

[PrettyWood]

It took me some time to understand at first the from_orm behaviour. Personally I like to think about "dict representation" (which try_instance kind of explains better).
I still don't understand why we need to add this in pydantic-core. Wouldn't it work to convert in pydantic world a from_orm to a function (dict representation) -> dict validation in pydantic-core?

[samuelcolvin]

Well we could, but it would be slower.

The real problem would be with nested objects - hard to know the the python function how deep to go.

Since we have the functionality in pydantic-core, I think it's just about deciding when to enable it.

[PrettyWood]

Ah yes you're right. It's just that the current pydantic-core interface is neat and quite generic. Having a specific case of conversion for dict feels a bit odd. And I'm just thinking right now how we had issues in my company with this issue.
I don't know if it would make sense to generalize this behaviour for other types like lists where users want to use a custom representation my_object_as_dict, my_object_as_list, ...
It could be a simple "converter" key in the schema 🤔

Instead of

v = SchemaValidator({'type': 'dict', 'keys': {'type': 'str'}, 'try_instance_as_dict': True})

we would write

def cls_instance_to_dict(x):
    if hasattr(x, '__dict__'):
        return {k: serialize(v) for k, v in x.__dict__.items()}
    return x

v = SchemaValidator({'type': 'dict', 'keys': {'type': 'str'}, 'convertor': cls_instance_to_dict})

[samuelcolvin]

I think we just need a config flag.

[samuelcolvin]

"converters" are covered by the functions already implemented - e.g. you could use a function with mode "before" and convert the data before calling the main validator.

[samuelcolvin]

think I'm going to use the name from_attributes.

I think this is better than try_instance and from_orm.

[PrettyWood]

the name is way better 👍