Fix union validator to correctly identify inherited classes in discriminated unions #1613
Conversation
…minated unions Update the `validate_smart` function to identify the correct object class in a discriminated union by using the class with the greatest number of matching fields and the greatest percentage of matching fields. * Modify `validate_smart` function in `src/validators/union.rs` to calculate the number and percentage of matching fields for each class and select the best match. * Add test cases in `tests/validators/test_union.py` to verify the updated `validate_smart` function for discriminated unions and ensure correct class selection based on the number and percentage of matching fields. --- For more details, open the [Copilot Workspace session](https://copilot-workspace.githubnext.com/pydantic/pydantic-core/tree/main?shareId=XXXX-XXXX-XXXX-XXXX).
There was a problem hiding this comment.
This seems like a reasonable heuristic to me when picking what the best match should be, thanks.
cc @dmontagu we've talked about this kind of idea long ago, this seems like a reasonable small iteration towards a good result 👍
| let mut errors = MaybeErrors::new(self.custom_error.as_ref()); | ||
|
|
||
| let mut best_match: Option<(Py<PyAny>, Exactness, Option<usize>)> = None; | ||
| let mut best_match: Option<(Py<PyAny>, Exactness, Option<usize>, usize, f64)> = None; |
There was a problem hiding this comment.
I think now that we have so many fields here, we should create a struct for these so we can give them names.
src/validators/union.rs
Outdated
| let new_exactness = state.exactness.unwrap_or(Exactness::Lax); | ||
| let new_fields_set_count = state.fields_set_count; | ||
| let new_fields_set_percentage = new_fields_set_count.map(|count| { | ||
| let total_fields = input.as_dict().map_or(0, |dict| dict.len()); |
There was a problem hiding this comment.
This doesn't depend on the validator so can be done once at the start of the function (i.e. above the self.choices loop.)
src/validators/union.rs
Outdated
| let new_fields_set_count = state.fields_set_count; | ||
| let new_fields_set_percentage = new_fields_set_count.map(|count| { | ||
| let total_fields = input.as_dict().map_or(0, |dict| dict.len()); | ||
| count as f64 / total_fields as f64 |
There was a problem hiding this comment.
To avoid a divide by zero here I think that we should keep total_fields as Option<usize>. For example consider an int or str input, these have no fields.
… in a discriminated union * Modify `validate_smart` to iterate through choices and calculate the number and percentage of matching fields for each class * Adjust `validate_smart` to select the class with the highest number and percentage of matching fields * Add test cases in `test_union.py` to verify the updated `validate_smart` function for discriminated unions * Include test cases to check the correct class selection based on the number and percentage of matching fields
|
After a little research, I realized that the |
2 participants
Update the
validate_smartfunction to identify the correct object class in a discriminated union by using the class with the greatest number of matching fields and the greatest percentage of matching fields.validate_smartfunction insrc/validators/union.rsto calculate the number and percentage of matching fields for each class and select the best match.tests/validators/test_union.pyto verify the updatedvalidate_smartfunction for discriminated unions and ensure correct class selection based on the number and percentage of matching fields.For more details, open the Copilot Workspace session.