Add a blueprint for pydantic 2

[caarmen]

Add a blueprint to support pydantic models, using the pydantic 2.x version.

Here's the pydantic 2 migration guide: https://docs.pydantic.dev/latest/migration/

[caarmen]

ℹ️ Here's the difference between the two blueprints for pydantic.

The change is to use model_json_schema() instead of model_schema(), which takes a ref_template argument instead of ref_prefix.

Note, I didn't see in the pydantic migration notes anything about migrating from the old model_schema().

$ diff -U5 docs/blueprints/pydantic.py  docs/blueprints/pydantic2.py 
--- docs/blueprints/pydantic.py	2023-07-17 13:38:19.059201949 +0200
+++ docs/blueprints/pydantic2.py	2023-07-17 13:41:14.883652117 +0200
@@ -1,9 +1,9 @@
 from drf_spectacular.extensions import OpenApiSerializerExtension
 from drf_spectacular.plumbing import ResolvedComponent
 
-from pydantic.schema import model_schema
+from pydantic.json_schema import model_json_schema
 
 
 class PydanticExtension(OpenApiSerializerExtension):
     target_class = "pydantic.BaseModel"
     match_subclasses = True
@@ -12,11 +12,11 @@
         return self.target.__name__
 
     def map_serializer(self, auto_schema, direction):
 
         # let pydantic generate a JSON schema
-        schema = model_schema(self.target, ref_prefix="#/components/schemas/")
+        schema = model_json_schema(self.target, ref_template="#/components/schemas")
 
         # pull out potential sub-schemas and put them into component section
         for sub_name, sub_schema in schema.pop("definitions", {}).items():
             component = ResolvedComponent(
                 name=sub_name,

[codecov]

Codecov Report

Patch and project coverage have no change.

Comparison is base (fa568ce) 98.55% compared to head (5a02f01) 98.55%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1028   +/-   ##
=======================================
  Coverage   98.55%   98.55%           
=======================================
  Files          68       68           
  Lines        8367     8367           
=======================================
  Hits         8246     8246           
  Misses        121      121           

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[tfranzel]

awesome! thx @caarmen

[caarmen]

Actually, it seems there's an issue, sorry I didn't catch it earlier :/

For a model with a field which is a list of another pydantic model, the field isn't being generated correctly.

Looks the pydantic 1 model_schema and pydantic 2 model_json_schema aren't exactly equivalent, the subtype information isn't present in model_json_schema.

I'll update here if/when I find more information.

[tfranzel]

yes, I also noticed the 2 versions are subtly different. Stuff wasn't just renamed. Feel free to do another PR.

[caarmen]

I hit an issue trying to work around this. I opened an issue on Pydantic: pydantic/pydantic#6741

I haven't managed to generate an openapi json with either #/$defs/ as the location of definitions (which pydantic seems to want), or #/components/schemas (which drf-spectacular seems to want).

Looks like pydantic has an api to be able to put definitions in #/components/schemas, but there's an issue retrieving sub objects when I try to specify this location.

Maybe (hopefully) I just didn't understand the pydantic api enough. 🤞🏻

[caarmen]

I've figured out a workaround, but it's not very clean :/ It basically takes the schema and does a search/replace of #/$defs/ => #/components/schemas/ 😅

Tested in my project:

+import json
+
 from drf_spectacular.extensions import OpenApiSerializerExtension
 from drf_spectacular.plumbing import ResolvedComponent
 from pydantic.json_schema import model_json_schema
 
 
 class PydanticExtension(OpenApiSerializerExtension):
     target_class = "pydantic.BaseModel"
     match_subclasses = True
 
     def get_name(self, auto_schema, direction):
         return self.target.__name__
 
     def map_serializer(self, auto_schema, direction):
 
         # let pydantic generate a JSON schema
-        schema = model_json_schema(self.target, ref_template="#/components/schemas")
+        schema = model_json_schema(self.target)
 
         # pull out potential sub-schemas and put them into component section
-        for sub_name, sub_schema in schema.pop("definitions", {}).items():
+        for sub_name, sub_schema in schema.pop("$defs", {}).items():
             component = ResolvedComponent(
                 name=sub_name,
                 type=ResolvedComponent.SCHEMA,
                 object=sub_name,
                 schema=sub_schema,
             )
             auto_schema.registry.register_on_missing(component)
 
-        return schema
+        # Workaround to force placing definitions in components/schemas.
+        # Pydantic places definitions in $defs, and we want them in components/schemas,
+        # but using ref_template="#/components/schemas in model_json_schema() makes us lose sub objects.
+        # See https://github.com/pydantic/pydantic/issues/6741
+        schema_with_relocated_defs = json.loads(json.dumps(schema).replace("#/$defs/", "#/components/schemas/"))
+        return schema_with_relocated_defs

If you want, I can open a PR for this.

But I understand it's not very clean! 😄

We can wait for some news on the pydantic issue instead.

[caarmen]

scratch that, I can do a proper fix I think 😁
Coming soon (sorry for the noise).

[caarmen]

Here's a PR:

• Fix the blueprint for pydantic version 2 #1030

My confidence in this isn't at 100% for now, we could say 😅 . So do feel free to try it out a bit before merging.

I didn't find unit tests for blueprints in the project.