Allow model_validate_json to return valid JSON string upon raising error

[timothylimyl]

Initial Checks

• I have searched Google & GitHub for similar requests and couldn't find anything
• I have read and followed the docs and still think this feature is missing

Description

Currently, I am using model_validate_json to validate the JSON string output.

By default, I am hoping to get the JSON output without the fields that Pydantic has raised error on.

For example, if my class is as such:

class Human(BaseModel):
    name: str
    age: float
    hobby: list[str]

but the JSON output is:
{ "name" : "Tim", "age": 20.0, "hobby" : "coding"}

It will raised the validation error, indicating "hobby" is type str but it is suppose to be type list[str]. Given that the validate function knows which field is errorneous, can it remove that field away from the JSON and return me the valid JSON instead?

so the code typically look something like this:

try:
  human_json_string = model_output_json(....)
  Human.model_validate_json(json_data = human_json_string)
  # now we know the JSON is all good, we can load it 
  human_data = json.loads(human_json_string)
except ValidationError as e:
  print(f"Error message from Pydantic: {e}\n")

within this code, I want to get the valid json that I can load to still display valid fields instead of failing the entire JSON output from the model

Docs: https://docs.pydantic.dev/latest/api/base_model/#pydantic.BaseModel.model_validate_json

Affected Components

• Compatibility between releases
• Data validation/parsing
• Data serialization - .model_dump() and .model_dump_json()
• JSON Schema
• Dataclasses
• Model Config
• Field Types - adding or changing a particular data type
• Function validation decorator
• Generic Models
• Other Model behaviour - model_construct(), pickling, private attributes, ORM mode
• Plugins and integration with other tools - mypy, FastAPI, python-devtools, Hypothesis, VS Code, PyCharm, etc.

[timothylimyl]

I reckon this is a common feature that people will want for use cases where JSON data output is non-deterministic and partial information is still valid for the application to "move on".

[timothylimyl]

As an example, I believe Jason Liu himself also brought up this use case @ 15.22: https://www.youtube.com/watch?v=yj-wSRJwrrc

If we set up a validator to check whether the JSON output of the field has answer that is containing in the text that we have provided it, I believe in an application flow, we will want to continue with the answers that we have and deal with the fields that has been validated as false (no reference in text).

[sydney-runkle]

Hi @timothylimyl,

This is a solid feature request. I think our partial JSON parsing offers some support for this, but doesn't entirely cover the cases you've outlined above.

Leaving this open, PRs + API design ideas are welcome!