Discriminated unions raise TypeError for unions of a single type

[tommilligan]

Checks

• I added a descriptive title to this issue
• I have searched (google, github) for similar issues and couldn't find anything
• I have read and followed the docs and still think this is a bug

Bug

Output of python -c "import pydantic.utils; print(pydantic.utils.version_info())":

             pydantic version: 1.9.0
            pydantic compiled: False
                 install path: /home/tom/reinfer/env/lib/python3.8/site-packages/pydantic
               python version: 3.8.10 (default, Nov 26 2021, 20:14:08)  [GCC 9.3.0]
                     platform: Linux-5.11.0-44-generic-x86_64-with-glibc2.29
     optional deps. installed: ['dotenv', 'typing-extensions']

When a Union with only a single variant it tagged with a field as a discriminated union, a type error is thrown on defining the class. The below is a minimally reproducible example:

# repro.py
from typing import Literal, Union

from pydantic import BaseModel, Field


class DataAModel(BaseModel):
    kind: Literal["a"]


class RequestModel(BaseModel):
    data: Union[DataAModel] = Field(..., discriminator="kind")

produces

$ python repro.py
Traceback (most recent call last):
  File "repro.py", line 10, in <module>
    class RequestModel(BaseModel):
  File "/home/tom/r/env/lib/python3.8/site-packages/pydantic/main.py", line 204, in __new__
    fields[ann_name] = ModelField.infer(
  File "/home/tom/r/env/lib/python3.8/site-packages/pydantic/fields.py", line 488, in infer
    return cls(
  File "/home/tom/r/env/lib/python3.8/site-packages/pydantic/fields.py", line 419, in __init__
    self.prepare()
  File "/home/tom/r/env/lib/python3.8/site-packages/pydantic/fields.py", line 534, in prepare
    self._type_analysis()
  File "/home/tom/r/env/lib/python3.8/site-packages/pydantic/fields.py", line 605, in _type_analysis
    raise TypeError('`discriminator` can only be used with `Union` type')
TypeError: `discriminator` can only be used with `Union` type

The following code snippet works as intended. The only difference is I have introduced a second type in the discriminated union field. Note that simply repeating the existing type (such as Union[DataAModel, DataAModel]) still fails:

from typing import Literal, Union

from pydantic import BaseModel, Field


class DataAModel(BaseModel):
    kind: Literal["a"]


class DataBModel(BaseModel):
    kind: Literal["b"]


class RequestModel(BaseModel):
    data: Union[DataAModel, DataBModel] = Field(..., discriminator="kind")

---

After some poking, it looks like within _type_analysis, self.discriminator_key is set on the model DataAModel, rather than the parent field, Union[DataAModel]. I'm guessing there is a simplification step somewhere where singleton unions are simplified to their child type, which is no longer accurate?

Happy to submit a PR if you can point me further in the right approach.

[PrettyWood]

Hi @tommilligan
The thing is we can't do anything about it because python changes Union[A] into A at interpretation time.

from typing import Union
class A: ...
assert Union[A] is A

So at runtime, pydantic has no way to know A was supposed to be a Union

[tommilligan]

Ah, that makes sense. I was about to start poking to see if the simplification was internal to pydantic or not, but if it's at a higher layer I'll give it up as a lost cause.

Would you accept a PR noting this as an edge case in the documentation/error message?

I suppose the workaround is to add a dummy second type into the union, or just to remove the discriminator until it is required.

For documentation forward compatibility, we were using openapi-schema-pydantic, which now collides with the discriminator Field property nicely. But that's not your problem! I'll file a bug over there now.

[PrettyWood]

Documentation PRs are always welcome :)

[tommilligan]

Opened #3639