Use oneOf to represent nullable schemas #12089
Conversation
Instead of falling back to the general union case with `anyOf`, we represent the special case of a nullable schema using `oneOf` for compatibility with external tooling.
There are a few remaining tests which may reflect complexities highlighted in the discussion, so leaving assertions untouched for now.
CodSpeed Performance ReportMerging #12089 will not alter performanceComparing Summary
|
glopesdev
force-pushed
the
nullable-schema-oneof
branch
from
2f06031 to
4454a9e
Compare
|
Thanks for the contribution @glopesdev. I don't think using |
|
@Viicos Thanks for pushing the development on this. The option of having type arrays to represent nullable fields of primitive types would work for us, as it works with NJsonSchema and all our current models only use primitive value types anyway. I think there is definitely room for further discussion, and it will likely make sense to extend this feature to support other union formats. At least we now have an explicit place to further develop the functionality, since I am assuming for example if later we want to add the option for Feel free to close this PR if not appropriate anymore. It would be great to have #12147 merged into pydantic already for the next release (v1.18). |
Instead of falling back to the general union case with
anyOf, we represent the special case of a nullable schema usingoneOffor compatibility with external tooling.Change Summary
Following the suggestion in #7161 (comment), we propose here to replace
anyOfwithoneOfspecifically for representing nullable schemas. This is a special case which has validation in the code to extend compatibility ofOptional[T]to a large swath of tooling.Related issue number
fix #7161
Checklist