I left a few comments/questions that I hope are helpful.

Posting a response to all the comments here since they all relate to the same concept: the constraint associated with this implementation that we only apply transformations to subschemas that we can resolve a JsonTypeInfo for. The subschemas that we don't recurse into are:

• schema.not
• schema.allOf
• schema.oneOf
• schema.additionalProperties

If a user sets one of these subschemas explicitly in their own transformer, we have no way of constructing a complete context object for it because we don't know the type associated with the schema.

options.UseSchemaTransformer((schema, context, cancellationToken) =>
{
	// What is the JsonTypInfo assocaited with this schema?
	schema.Not = new OpenApiSchema { Type = "string" };
});

There's two approaches to solving this:

We could relax the constraints on the OpenApiSchemaTransformerContext so that the JsonTypeInfo is optional and recurse into the subschemas above. At the point, there's no way for you to know what the context of the schema you're modifying is. Is it a not schema, is it an allOf? There's no way to tell without adding additional state to the context object.

Alternatively, we could set the OpenApiSchemaTransformerContext.JsonTypeInfo property to the parent type associated with the schema (for example, a not inside a Todo schema would have JsonTypeInfo = Todo. This has the opposite problem though: you'd recurse into a schema with the wrong type and potentially make unexpected modifications to the target schema.

So, stuck between a rock and a hard place, I opted not to recurse into these as part of the default implementation. If a user explicitly needs it though, it should be possible to construct their own schema transformer implementation that supports the recursion into these properties and they can initialize the OpenApiSchemaTransformerContext with whatever state makes sense for their scenario. I'll add a test to model how I think this approach would look.

Let me know if you have any other questions or if there is some other way to solve the ambigous context problem.

Bump for reviews! Hoping to get this in before Friday...

FWIW I had a look at it last week and didn't spot anything.