[asukaminato0721]

Summary

Fixes #4131

Custom class union members no longer degrade to Any.

Invalid Pydantic boolean literals are rejected according to Pydantic’s accepted values (https://docs.pydantic.dev/latest/api/standard_library_types/#booleans), while valid alternate union branches remain accepted.

Test Plan

add test, updated the lax-conversion documentation.

[meta-codesync]

This pull request has been imported. If you are a Meta employee, you can view this in D114011818. (Because this pull request was imported automatically, there will not be any future comments.)

[codspeed]

Merging this PR will not alter performance

✅ 35 untouched benchmarks
⏩ 9 skipped benchmarks¹

---

_{Comparing asukaminato0721:4131 (2eb723a) with main (1af3067)}

Footnotes

1. 9 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

============================================================
SUMMARY
============================================================
Total: +21 new errors, -0 fixed errors
By preset: +21/-0 (default), +21/-0 (strict)

Projects with changes (3):
  openlibrary: +1 -0
  freqtrade: +1 -0
  pandera: +19 -0
============================================================

FULL DIFF DETAILS
------------------------------------------------------------

openlibrary (https://github.com/internetarchive/openlibrary)
+ ERROR openlibrary/bookworm/tests/test_opds.py:156:15-80: Argument `list[dict[str, str]]` is not assignable to parameter `links` with type `Iterable[Link]` in function `openlibrary.bookworm.opds.Publication.__init__` [bad-argument-type]

freqtrade (https://github.com/freqtrade/freqtrade)
+ ERROR freqtrade/rpc/api_server/api_ws.py:112:49-56: Argument `dict[str, Any]` is not assignable to parameter `data` with type `WSAnalyzedDFMessage.AnalyzedDFData` in function `freqtrade.rpc.api_server.ws_schemas.WSAnalyzedDFMessage.__init__` [bad-argument-type]

pandera (https://github.com/pandera-dev/pandera)
+ ERROR tests/pandas/test_pydantic.py:263:42-54: Argument `type[SimpleSchema]` is not assignable to parameter `pa_schema` with type `SimpleSchema` in function `DataFrameModelPydantic.__init__` [bad-argument-type]
+ ERROR tests/pandas/test_pydantic.py:271:42-43: Argument `Literal[1]` is not assignable to parameter `pa_schema` with type `SimpleSchema` in function `DataFrameModelPydantic.__init__` [bad-argument-type]
+ ERROR tests/pandas/test_pydantic.py:274:42-66: Argument `DataFrameSchema` is not assignable to parameter `pa_schema` with type `SimpleSchema` in function `DataFrameModelPydantic.__init__` [bad-argument-type]
+ ERROR tests/pandas/test_pydantic.py:285:33-46: Argument `type[test_invalid_dataframemodel.InvalidSchema]` is not assignable to parameter `pa_schema` with type `test_invalid_dataframemodel.InvalidSchema` in function `PydanticModel.__init__` [bad-argument-type]
+ ERROR tests/pandas/test_pydantic.py:300:47-53: Argument `type[test_dataframemodel_inheritance.Parent]` is not assignable to parameter `pa_schema` with type `test_dataframemodel_inheritance.Parent` in function `PydanticModel.__init__` [bad-argument-type]
+ ERROR tests/pandas/test_pydantic.py:301:47-52: Argument `type[test_dataframemodel_inheritance.Child]` is not assignable to parameter `pa_schema` with type `test_dataframemodel_inheritance.Parent` in function `PydanticModel.__init__` [bad-argument-type]
+ ERROR tests/pandas/test_pydantic.py:307:51-59: Argument `type[test_dataframemodel_inheritance.NotChild]` is not assignable to parameter `pa_schema` with type `test_dataframemodel_inheritance.Parent` in function `PydanticModel.__init__` [bad-argument-type]
+ ERROR tests/pandas/test_pydantic.py:324:43-44: Argument `Literal[1]` is not assignable to parameter `pa_schema` with type `DataFrameSchema | None` in function `DataFrameSchemaPydantic.__init__` [bad-argument-type]
+ ERROR tests/pandas/test_pydantic.py:327:39-42: Argument `Literal['1']` is not assignable to parameter `pa_mi` with type `MultiIndex | None` in function `DataFrameSchemaPydantic.__init__` [bad-argument-type]
+ ERROR tests/pandas/test_pydantic.py:345:47-48: Argument `Literal[1]` is not assignable to parameter `pa_series_schema` with type `SeriesSchema | None` in function `SeriesSchemaPydantic.__init__` [bad-argument-type]
+ ERROR tests/pandas/test_pydantic.py:348:40-43: Argument `Literal['1']` is not assignable to parameter `pa_column` with type `Column | None` in function `SeriesSchemaPydantic.__init__` [bad-argument-type]
+ ERROR tests/pandas/test_pydantic.py:351:39-42: Argument `Literal['1']` is not assignable to parameter `pa_index` with type `Index | None` in function `SeriesSchemaPydantic.__init__` [bad-argument-type]
+ ERROR tests/polars/test_polars_pydantic.py:86:48-66: Argument `type[SimplePolarsSchema]` is not assignable to parameter `pa_schema` with type `SimplePolarsSchema` in function `PolarsDataFrameModelPydantic.__init__` [bad-argument-type]
+ ERROR tests/polars/test_polars_pydantic.py:94:48-49: Argument `Literal[1]` is not assignable to parameter `pa_schema` with type `SimplePolarsSchema` in function `PolarsDataFrameModelPydantic.__init__` [bad-argument-type]
+ ERROR tests/polars/test_polars_pydantic.py:97:48-78: Argument `DataFrameSchema` is not assignable to parameter `pa_schema` with type `SimplePolarsSchema` in function `PolarsDataFrameModelPydantic.__init__` [bad-argument-type]
+ ERROR tests/polars/test_polars_pydantic.py:112:47-53: Argument `type[test_polars_dataframemodel_inheritance.Parent]` is not assignable to parameter `pa_schema` with type `test_polars_dataframemodel_inheritance.Parent` in function `PydanticModel.__init__` [bad-argument-type]
+ ERROR tests/polars/test_polars_pydantic.py:113:47-52: Argument `type[test_polars_dataframemodel_inheritance.Child]` is not assignable to parameter `pa_schema` with type `test_polars_dataframemodel_inheritance.Parent` in function `PydanticModel.__init__` [bad-argument-type]
+ ERROR tests/polars/test_polars_pydantic.py:119:51-59: Argument `type[test_polars_dataframemodel_inheritance.NotChild]` is not assignable to parameter `pa_schema` with type `test_polars_dataframemodel_inheritance.Parent` in function `PydanticModel.__init__` [bad-argument-type]
+ ERROR tests/polars/test_polars_pydantic.py:174:49-50: Argument `Literal[1]` is not assignable to parameter `pa_schema` with type `DataFrameSchema | None` in function `PolarsDataFrameSchemaPydantic.__init__` [bad-argument-type]

[rchen152]

Thanks! Left a suggestion to avoid some false positives and another to make the LaxBool handling a little nicer-looking. The false positives definitely need to be addressed before this can go in; feel free to push back on the LaxBool refactor if it doesn't make sense.

[rchen152]

This is now too strict. The mypy primer delta shows some false positives that all seem to be caused by this change:

• openlibrary test_opds.py:156 and freqtrade api_ws.py:112: each passes a dict, or a list of dicts, to a field typed as a nested BaseModel (Link, AnalyzedDFData). Pydantic validates the dict into the model. The same applies to stdlib dataclass fields.
• pandera, 6 errors (pandas lines 263, 300, 301; polars lines 86, 112, 113): a field typed as pa.DataFrameModel accepts the model class itself, via pandera's __get_pydantic_core_schema__. These tests assert that the call succeeds.

Claude suggested this narrower rule:

• BaseModel and dataclass types convert to T | Mapping[str, Any].
• Classes that define __get_pydantic_core_schema__ stay Any, because they validate input their own way.
• Other classes stay as themselves.

[rchen152]

Adding a separate field for LaxBool looks sort of clunky. Here's a suggestion to avoid that (again from Claude, so feel free to push back if this doesn't make sense):

keep the separate literal check, but run it per union member and build it on the existing lax conversion.

For most members, "accepts" means is_subset_eq(literal, lax(member)), using the same conversion that builds the signature. The one refinement is bool: it accepts int literals 0/1 and the listed strings. Put that predicate next to the bool entry in the atomic lax conversion table in pydantic_lax.rs, so the definition of what bool accepts lives in one place.