coerce_numbers_to_str needs a per-field variant

[jkugler]

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

We are migrating our code from Pydantic 1 to 2, and came across a situation where we do want some fields to coerce numbers to strings, but other fields in the model should not be coerced. As far as I can tell from the docs, coerce_numbers_to_str is an all-or-nothing thing, and will affect the entire model. Being able to specify fields would be nice. Maybe something like:

model_config = ConfigDict(coerce_numbers_to_str="field1,field2,field3")

I don't know if that's the best syntax, but that would serve our purposes.

If there is a way to work around this feature not being there, could it be added to https://docs.pydantic.dev/latest/api/config/#pydantic.config.ConfigDict.coerce_numbers_to_str ?

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.

[sydney-runkle]

@jkugler,

Seems like a reasonable feature request 👍. I think we should still stick with True and False as the allowed settings, but we could support Field(..., coerce_numbers_to_str=True) or something along those lines.

[jkugler]

Seems perfectly reasonable. I'm a Pydantic newbie thrown in the deep end by migrating a project from v1 to v2, so will quicly admit not being fully up to speed with the API.

[sydney-runkle]

@jkugler,

No worries. Let us know if there's anything else you need help with in terms of your migration, or if you notice anything missing from the migration guide.

[NeevCohen]

Hey @sydney-runkle, I was looking into this issue and I think I can work on this. I looked at the StrValidator in core, and saw that it only reads the coerce_numbers_to_str configuration from the config. My thoughts were to add a coerce_numbers field to the str_schema and make StrValidator read that value from the schema as well as from the model config.

[sydney-runkle]

@NeevCohen,

Sounds great! Go for it - I think you have the right approach :). Maybe we stick with coerce_numbers_to_str as the name of the flag just to be consistent. Prioritize the field level setting, of course :).

[NeevCohen]

@sydney-runkle I opened a PR in core to support this - pydantic/pydantic-core#1249

When that will get merged I'll start working on a change in pydantic, which should be pretty straight forward

[NeevCohen]

@sydney-runkle
Thanks for your help with the PR in core! Now that support was merged into core, I started looking into implementing the change in pydantic. I added a coerce_numbers_to_str option to FieldInfo, and I was wondering where would be the correct place to pass that value on to str_schema(). I thought maybe in GenerateSchema._common_field_schema, I could inject it into the schema variable? Not 100% sure on that, perhaps you have a better idea?

[sydney-runkle]

@NeevCohen,

Hmm, I think here would probably be the most appropriate location:

pydantic/pydantic/_internal/_known_annotated_metadata.py

Line 159 in 477e0a2

def apply_known_metadata(annotation: Any, schema: CoreSchema) -> CoreSchema | None: # noqa: C901

That function is used during the schema generation process :). You can also ensure that said setting is only specified for types with which it's compatible.

[NeevCohen]

@sydney-runkle Thanks!

I've opened a PR #9137, but it can't be merged until a new version of core (with support for this) is released and the version is updated in pydantic.

[jkugler]

Brilliant, thanks!