Handle nullable schemas with serialization schema available during JSON Schema generation

[Viicos]

As per https://github.com/pydantic/pydantic/pull/10121/files#r1715043570:

It might be that the serialization schema has a specific when_used mode set to unless-none or json-unless-none. In that case, if the base core schema is of type nullable, we need to account for this as the serialization schema might not be used.

This also fixes an issue where unless-none was not taken into account when trying to figure out if the default value should be passed into the plain serializer function to get the JSON Schema default. This was missed in #10121.

Change Summary

Related issue number

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

[cloudflare-workers-and-pages]

Deploying pydantic-docs with    Cloudflare Pages

Latest commit:	423345b
Status:	✅  Deploy successful!
Preview URL:	https://a758d129.pydantic-docs.pages.dev
Branch Preview URL:	https://nullable-ser.pydantic-docs.pages.dev

View logs

[Viicos]

I'm not too happy with how this is done. Ideally, self.nullable_schema should be used but it will use schema['schema'] to generate the inner schema:

pydantic/pydantic/json_schema.py

Lines 1084 to 1094 in a42fb67

	def nullable_schema(self, schema: core_schema.NullableSchema) -> JsonSchemaValue:
	"""Generates a JSON schema that matches a schema that allows null values.
	Args:
	schema: The core schema.
	Returns:
	The generated JSON schema.
	"""
	null_schema = {'type': 'null'}
	inner_json_schema = self.generate_inner(schema['schema'])

While we need schema['serialization'] to be used instead.

[sydney-runkle]

Should we model the below more closely?

if (
    self.mode == 'serialization' 
    and 'serialization' in schema_or_field
    and (ser_schema := schema_or_field['serialization']
    and (json_schema := self.ser_schema(ser_schema))
    and ser_schema.get('when_used') in ('unless-none', 'json-unless-none')
    and schema_or_field['type'] == 'nullable'
):
    json_schema = self.get_flattened_anyof([{'type': 'null'}, json_schema])

Or something like that

[sydney-runkle]

I understand your concern, but I agree, I think this is a fine patch for now. Maybe add a comment explaining the pattern here - we have so many conditionals, it's a bit hard to understand all together 👍

[Viicos]

Should we model the below more closely?

I think I prefer the two conditionals, as it shows the logic more clearly imo:

• if we are in serialization mode and a serialization schema is available:
  • shortcut JSON Schema generation and use ser_schema
  • special case nullable schema depending on the value of when_used

[codspeed-hq]

CodSpeed Performance Report

Merging #10132 will not alter performance

_{Comparing nullable-ser (423345b) with main (7abafa4)}

Summary

✅ 17 untouched benchmarks

[sydney-runkle]

Should we model the below more closely?

if (
    self.mode == 'serialization' 
    and 'serialization' in schema_or_field
    and (ser_schema := schema_or_field['serialization']
    and (json_schema := self.ser_schema(ser_schema))
    and ser_schema.get('when_used') in ('unless-none', 'json-unless-none')
    and schema_or_field['type'] == 'nullable'
):
    json_schema = self.get_flattened_anyof([{'type': 'null'}, json_schema])

Or something like that

[sydney-runkle]

I understand your concern, but I agree, I think this is a fine patch for now. Maybe add a comment explaining the pattern here - we have so many conditionals, it's a bit hard to understand all together 👍

[github-actions]

Coverage report

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