Try other branches in smart union in case of omit errors by mikeedjones · Pull Request #12758 · pydantic/pydantic

mikeedjones · 2026-01-30T16:56:31Z

Change Summary

fix #12750 by updating the UnionValidator.validate_smart in union.rs to handle ValError::Omit errors by continuing to try other degrees of exactness instead of propagating the omit error and causing the value to be omitted.

Added a new test, test_td_smart_union_omit, in test_union.py to verify that the union validator properly supports the omit behavior when multiple union branches are present, ensuring that omitted errors do not prevent successful validation. This test fails on primary:

__________ test_td_smart_union_omit __________

    def test_td_smart_union_omit() -> None:
        validator = SchemaValidator(
            core_schema.typed_dict_schema(
                fields={
                    'x': core_schema.typed_dict_field(
                        core_schema.union_schema(
                            [
                                core_schema.with_default_schema(core_schema.int_schema(), on_error='omit'),
                                core_schema.with_default_schema(core_schema.bool_schema(), on_error='omit'),
                            ]
                        )
                    )
                },
            )
        )
>       assert validator.validate_python({'x': '123'}) == {'x': 123}
E       AssertionError: assert {} == {'x': 123}
E         
E         Right contains 1 more item:
E         {'x': 123}
E         Use -v to get more diff

Related issue number

fix #12750

Checklist

The pull request title is a good summary of the changes - it will be used in the changelog
Unit tests for the changes exist
Tests pass on CI
Documentation reflects the changes where applicable
My PR is ready to review, please add a comment including the phrase "please review" to assign reviewers

Selected Reviewer: @Viicos

codspeed-hq · 2026-01-30T17:09:07Z

Merging this PR will not alter performance

✅ 212 untouched benchmarks

_{Comparing mikeedjones:bugfix/pydantic-core/omit-errors-in-unions (ff87040) with main (4f818e0)}

github-actions · 2026-01-30T17:10:52Z

Coverage report

This PR does not seem to contain any modification to coverable code.

mikeedjones · 2026-01-30T17:27:25Z

please review :)

davidhewitt

Thanks, I think it's reasonable to intercept Omit errors in this way, however we probably need to ensure that if no validator succeeds we will re-raise the Omit error.

mikeedjones · 2026-02-02T15:16:53Z

Thanks, I think it's reasonable to intercept Omit errors in this way, however we probably need to ensure that if no validator succeeds we will re-raise the Omit error.

Thanks for review! Yes, I think the behaviour is correct but its not following the ValError::Omit code path when that error has been raised (potentially with the legitimate user-set exactness). I'm not sure it's worth adding the overhead to propagate "legitimate" Omit Errors now - what do you think? Have added that test and expanded the comment in union.rs.

If we did want to re-raise the omit error - do you think its better to add a MaybeOmits and associated interface or extend MaybeErorrs? Leaning towards the former.

davidhewitt

Thanks, I think we're getting there. As this is adjusting behaviour (arguably for the better), I just want to be extra sure that we have enough tests to be comfortable we're not missing any corner cases.

Co-authored-by: David Hewitt <[email protected]>

mikeedjones · 2026-02-05T15:35:39Z

@davidhewitt looking at the list it seems most prudent to just raise the Omit error if there's an omit. An omit in the union is currently taking precedence over a raised error - I think this is the best option and I'm not sure there's a way to do both?

davidhewitt

Agreed, I think propagating an omit error seems most correct given that is the existing behaviour.

So the PR modification is that in case of one branch raising an omit error, we now allow the other branches to try to succeed. If none of the branches succeed, we still raise the omit error as we would have previously.

I think this is a sensible improvement, and we explicitly allow ourselves to improve the smart union logic:

We reserve the right to tweak the internal smart matching algorithm in future versions of Pydantic. If you rely on very specific matching behavior, it's recommended to use union_mode='left_to_right' or discriminated unions.

davidhewitt · 2026-02-07T10:00:22Z

        let old_fields_set_count = state.fields_set_count;

        let mut errors = MaybeErrors::new(self.custom_error.as_ref());
+        let mut omits = MaybeOmits::new();


I think we can just keep a boolean marker, looks like the labels are never used:

Suggested change

let mut omits = MaybeOmits::new();

// propagate an omit error if all branches failed

let mut should_omit = true;

Viicos · 2026-02-07T10:28:45Z

Would it be possible to have pydantic tests instead? That is, have a proper TypedDict defined, and use TypeAdapter to make the assertions?

In this case, it would look like:

class TD(TypedDict): x: OnErrorOmit[int] | OnErrorOmit[str]

I'm also wondering if this should be valid at all. If you simply do:

class TD(TypedDict): x: OnErrorOmit[int | bool] # or even just OnErrorOmit[int]

We get a schema build error:

TypeAdapter(TD) """ pydantic_core._pydantic_core.SchemaError: Error building "typed-dict" validator: SchemaError: Field 'x': 'on_error = omit' cannot be set for required fields """

And doing this works fine:

class TD(TypedDict, total=False): x: OnErrorOmit[int | bool] ta = TypeAdapter(TD) ta.validate_python({'x': '123'}) #> {'x': 123}

So presumably we're not catching a schema build error when doing OnErrorOmit[int] | OnErrorOmit[str], which leads to inconsistent validation results?

(these tests would go in tests/types/test_union.py )

I'm also wondering if this should be valid at all.

To be fair, this is a great question. We could potentially just start rejecting this at build time (or at least a deprecation warning that we'll reject it in v3).

I think

list[OnErrorOmit[int] | OnErrorOmit[bool]]

should be valid even if

list[OnErrorOmit[int | bool]]

Produces the same result and is arguably clearer.

What about

list[OnErrorOmit[int] | bool]

... I guess that should almost certainly be invalid?

I included test_smart_union_omit_with_defaults to document how that's treated in this PR.

pydantic.TypeAdapter(list[pydantic.OnErrorOmit[int] | bool]).validate_python(["1","True"]) > [1]

Is current behavior in 2.12 which I think would be unexpected by users. I can add a schema build warning at least?

@davidhewitt I added a warning on constructing a union schema including OnErrorOmit :)

I'm -0 on having this warning, as it doesn't take into account all cases (e.g. what if you do: type A = OnErrorOmit[int]; U = A | OnErrorOmit[str]).

Ideally we would simplify such unions when building the validator (to end up with a single one), which would give us the ability to catch the error for typed dicts. But this is not trivial.

So I think let's remove the warnings logic, move the tests to pydantic (https://github.com/pydantic/pydantic/pull/12758/changes#r2777402600), and it should be good to merge.

Thank you @Viicos ! I added pydantic side tests to tests/types/test_union.py in ccd07fa but have now removed the pydantic-core side tests and warning logic. Is there anything else this needs?

…MaybeOmits struct

davidhewitt

Thanks!

continue checking smart union on omit error

62574b0

github-actions Bot added the relnotes-fix Used for bugfixes. label Jan 30, 2026

pydantic-hooky Bot added the ready for review label Jan 30, 2026

pydantic-hooky Bot assigned Viicos Jan 30, 2026

add assert that invalid value is omitted

d8d68aa

davidhewitt reviewed Feb 2, 2026

View reviewed changes

Comment thread pydantic-core/tests/validators/test_union.py Outdated

add list test

e8411a6

mikeedjones added 4 commits February 2, 2026 15:25

expand comment

26e202d

add

5a74d59

clarify comment on omit error handling in UnionValidator

20d0b7b

Merge branch 'main' into bugfix/pydantic-core/omit-errors-in-unions

1a1d6bf

mikeedjones requested a review from davidhewitt February 3, 2026 07:14

davidhewitt reviewed Feb 5, 2026

View reviewed changes

Comment thread pydantic-core/tests/validators/test_union.py Outdated

mikeedjones and others added 5 commits February 5, 2026 11:40

Update pydantic-core/tests/validators/test_union.py

acc16bc

Co-authored-by: David Hewitt <[email protected]>

added tests showing interaction with defaults and non-omitted values

93e4050

add a maybe omit

6150ac2

priortise omit over raise

7293499

Merge branch 'main' into bugfix/pydantic-core/omit-errors-in-unions

1f87705

mikeedjones requested a review from davidhewitt February 5, 2026 15:36

remove redundant type annotation

4160dc6

davidhewitt changed the title ~~fix: continue checking smart union on omit error~~ Try other branches in smart union in case of omit errors Feb 7, 2026

davidhewitt reviewed Feb 7, 2026

View reviewed changes

Viicos reviewed Feb 7, 2026

View reviewed changes

refactor: simplify omit handling in UnionValidator and remove unused …

ccd07fa

…MaybeOmits struct

update for 3.9

6324257

mikeedjones requested review from Viicos and davidhewitt February 9, 2026 13:19

mikeedjones added 4 commits February 12, 2026 08:59

emit warning when using omit within union schema

8c61e4f

Merge branch 'main' into bugfix/pydantic-core/omit-errors-in-unions

416dcec

Merge branch 'main' into bugfix/pydantic-core/omit-errors-in-unions

95b6ce5

Merge branch 'main' into bugfix/pydantic-core/omit-errors-in-unions

8867f99

Viicos added awaiting author revision awaiting changes from the PR author and removed ready for review labels Mar 5, 2026

mikeedjones added 5 commits March 12, 2026 07:56

Merge branch 'main' into bugfix/pydantic-core/omit-errors-in-unions

b68d863

rm warning

4fa8ee3

remove pydantic-core tests, only use pydantic tests

4b9b6b7

rm test for warning logic

de942f6

Merge branch 'main' into bugfix/pydantic-core/omit-errors-in-unions

ff87040

davidhewitt approved these changes Mar 18, 2026

View reviewed changes

davidhewitt merged commit 6178953 into pydantic:main Mar 18, 2026
81 checks passed

	let mut omits = MaybeOmits::new();
	// propagate an omit error if all branches failed
	let mut should_omit = true;

Uh oh!

Conversation

mikeedjones commented Jan 30, 2026 • edited by pydantic-hooky Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Change Summary

Related issue number

Checklist

Uh oh!

codspeed-hq Bot commented Jan 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will not alter performance

Uh oh!

github-actions Bot commented Jan 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Coverage report

Uh oh!

mikeedjones commented Jan 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

davidhewitt left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

mikeedjones commented Feb 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

davidhewitt left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

mikeedjones commented Feb 5, 2026

Uh oh!

davidhewitt left a comment

Choose a reason for hiding this comment

Uh oh!

davidhewitt Feb 7, 2026

Choose a reason for hiding this comment

Uh oh!

Viicos Feb 7, 2026

Choose a reason for hiding this comment

Uh oh!

davidhewitt Feb 7, 2026

Choose a reason for hiding this comment

Uh oh!

davidhewitt Feb 7, 2026

Choose a reason for hiding this comment

Uh oh!

mikeedjones Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

davidhewitt Feb 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mikeedjones Feb 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mikeedjones Feb 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Viicos Mar 5, 2026

Choose a reason for hiding this comment

Uh oh!

mikeedjones Mar 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

davidhewitt left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

mikeedjones commented Jan 30, 2026 •

edited by pydantic-hooky Bot

Loading

codspeed-hq Bot commented Jan 30, 2026 •

edited

Loading

github-actions Bot commented Jan 30, 2026 •

edited

Loading

mikeedjones commented Jan 30, 2026 •

edited

Loading

mikeedjones commented Feb 2, 2026 •

edited

Loading

davidhewitt Feb 11, 2026 •

edited

Loading

mikeedjones Feb 11, 2026 •

edited

Loading

mikeedjones Feb 12, 2026 •

edited

Loading

mikeedjones Mar 12, 2026 •

edited

Loading