Add support for non-call @root_validator #114
Conversation
In Pydantic v1 having a plain decorator like `@root_validator`
(without `()`) was common and worked fine.
In v2, `@model_validator` needs a `mode` argument.
Add a test and adjust implementation, so that the program
replaces
```py
@root_validator
def foo(cls, values):
...
```
with:
```py
@model_validator(mode="after")
@classmethod
def foo(cls, values):
...
```
Thanks 馃檹 |
bump_pydantic/codemods/validator.py
Outdated
| else: | ||
| raise RuntimeError("decorator as attribute is not implemented yet.") # this should never happen ATM |
There was a problem hiding this comment.
We shouldn't have this conditional. Can you add a xfail test for which this would happen?
There was a problem hiding this comment.
OK. Removed RuntimeError and moved the Name case under else.
Can you add a xfail test for which this would happen?
I think the test is already there in test_import_pydantic, no?
| else: | ||
| # Assuming node.decorator matches Name() | ||
| # (once ROOT_VALIDATOR_DECORATOR supports Attribute() this needs a separate case). | ||
| # Changing `@root_validator` (Name, not Call) into `@model_validator(...)` (Call) | ||
| decorator = cst.Call(func=cst.Name(new_name), args=self._args) |
There was a problem hiding this comment.
But it shouldn't be an else, should it? Does the test you mention fails differently now?
There was a problem hiding this comment.
If I make it an elif (like before) without the else branch, mypy complains about decorator potentially being unbound.
Does the test you mention fails differently now?
Well, the test fails the same everytime ATM, because we never reach that case due to what ROOT_VALIDATOR_DECORATOR is matching (it's not matching decorators as attributes now, so those cases won't be processed even).
In Pydantic v1 having a plain decorator like
@root_validator(without()) was common and worked fine.In v2,
@model_validatorneeds amodeargument.Add a test and adjust implementation, so that the program replaces
with:
The implementation maybe is not beautiful, but seems to work 馃槄.
While working on this I noticed that having a
@model_validator()(without args) in v2 produces an error. One has to provide themodearg.I can open a separate PR for this,
but this change most likely will need a separate
visitmethodfor
ROOT_VALIDATOR_DECORATOR,because
modeinfield_validatorcan be omitted.