[BUG][RUST] `#[serde(tag)]` conflicts with discriminator property in nested objects

[ThomasVille]

Bug Report Checklist

• Have you provided a full/minimal spec to reproduce the issue?
• Have you validated the input using an OpenAPI validator?
• Have you tested with the latest master to confirm the issue still exists?
• Have you searched for related issues/PRs?
• What's the actual output vs expected output?
• [Optional] Sponsorship to speed up the bug fix or feature request (example)

Description

When generating Rust code from an OpenAPI spec that uses oneOf with a discriminator, the generator produces #[serde(tag = "type")] for the enum. However, when the enum variants are newtype wrappers around objects that also contain the discriminator field, serde fails to deserialize because it cannot find the type field in the expected location.

The rust models generated look like this:

#[derive(Clone, Default, Debug, PartialEq, Serialize, Deserialize)]
pub struct PersonResource {
    #[serde(rename = "type")]
    pub r#type: Type,
    #[serde(rename = "id")]
    pub id: String,
    #[serde(rename = "name", skip_serializing_if = "Option::is_none")]
    pub name: Option<String>,
}

#[derive(Clone, Default, Debug, PartialEq, Serialize, Deserialize)]
pub struct OrganizationResource {
    #[serde(rename = "type")]
    pub r#type: Type,
    #[serde(rename = "id")]
    pub id: String,
    #[serde(rename = "companyName", skip_serializing_if = "Option::is_none")]
    pub company_name: Option<String>,
}

#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
#[serde(tag = "type")]
pub enum ResourceUnion {
    #[serde(rename="person")]
    Person(models::PersonResource),
    #[serde(rename="organization")]
    Organization(models::OrganizationResource),
}

When deserializing JSON like this:

{
  "type": "person",
  "id": "123",
  "name": "John Doe"
}

Serde fails with: Error("missing field 'type'", ...)

This happens because:

1. #[serde(tag = "type")] tells serde that the enum uses an internally-tagged representation
2. For newtype variants with tag, serde expects the tag field to be separate from the variant's data
3. But the inner PersonResource struct also expects a type field
4. Serde removes the discriminator field when deserializing the enum, so PersonResource can't find it

openapi-generator version

7.14.0 I believe it's not a regression.

OpenAPI declaration file content or url

https://gist.github.com/ThomasVille/06ebf97ed17966c6d0cb2565a89c5aa2

Generation Details

java -DloggerPath=conf/log4j.properties generate -i "spec.json" \
        -g rust \
        -o ${TMP_PATH} \
        --additional-properties=packageName=my_client \
        --additional-properties=publishRustRegistry=false \
        --additional-properties=library=reqwest \
        --additional-properties=supportMiddleware=true \
        --additional-properties=avoidBoxedModels=true \
        --additional-properties=preferUnsignedInt=true

Steps to reproduce

1. Generate the SDK using the spec and the command above.
2. In the src/lib.rs file add this test:

#[test]
fn deserialize_resource_union() {
    let json = r#"{"type": "person", "id": "123", "name": "John Doe"}"#;
    let result: models::ResourceUnion = serde_json::from_str(json).unwrap();
}

3. Run the test with cargo test.
4. The test fails with:

running 1 test
test deserialize_resource_union ... FAILED

failures:

---- deserialize_resource_union stdout ----

thread 'deserialize_resource_union' panicked at src/lib.rs:16:68:
called `Result::unwrap()` on an `Err` value: Error("missing field `type`", line: 0, column: 0)
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


failures:
    deserialize_resource_union

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

Related issues/PRs

• These issues seem similar:
  • [BUG] [Rust] Conflicting type fields for discriminator unions, causing deserialization failures. #22018
  • [BUG] [RUST] Enum Varients Merging Incorrectly, breaking change #17869
• This PR seems to fix the issue for Axum but I don't understand the fix: [Rust-Axum][Breaking Change] Improve the oneOf model generator #20336.

This pattern is very common in JSON:API specifications where all resource objects have a type field, so I believe this is a widespread issue that may already have a solution.

Suggest a fix

I don't see any clean solution for now.

Here are a few leads:

• Using #[serde(untagged)] doesn't work when all variants have the same structure, which is typical when conforming to the JSON:API spec and is possible even in other situations as described in the example above.
• Automatically detecting this situation (there's a discriminator and all the newtype variants have the discriminator field) and removing the field from the newtypes when generating the model: likely a breaking change.
• Making the newtype variants not have the discriminator field in the OpenAPI spec, but this doesn't match with the JSON:API spec when a response has an included field where the items of the field should be valid resource objects (thus have the type field):

{
  "data": [
    {
      "type": "node",
      "id": "019a5ee2-862f-77d1-801c-1a08f9172167",
      "attributes": {},
      "relationships": {}
    }
  ],
  "included": [
    { // This is expected to be a valid resource object with `type`, `id`, ...
      "type": "mesh",
      "id": "019a5ee2-862a-7011-88f1-4b1c590a00d2",
      "attributes": {},
      "relationships": {}
    }
  ]
}

[ThomasVille]

Implementing #[serde(default...)] on the discrimintator field like in #20336 solves the issue for me.

However, in #20336, we inject #[serde(default = "{{{classname}}}::_name_for_{{{name}}}")] depending on if the field is the discriminator. Another solution is to give a default value to the field in the OpenAPI definition, however doing this doesn't inject serde(default on the field, so there may be a feature missing there.