fix(openapiv2): exclude oneof fields from required with proto3 field semantics

[sessa]

Summary

• Fixes oneof fields being incorrectly marked as required when use_proto3_field_semantics=true is set
• Adds a field.OneofIndex == nil check in updateSwaggerObjectFromFieldBehavior so that oneof members are excluded from the required array
• Explicit REQUIRED field behavior annotations still override this correctly

Problem

When use_proto3_field_semantics=true, all non-optional proto3 fields are added to the OpenAPI required array. This includes oneof members, which is semantically wrong -- a oneof means at most one field can be set, so they should never all be required.

For example, this proto:

message Filter {
  oneof filter_group_or_value {
    FilterList list = 1;
    FilterRule rule = 2;
  }
}

Generates both list and rule as required, when neither should be.

Fix

One-line change in updateSwaggerObjectFromFieldBehavior:

// Before
required = !field.GetProto3Optional()

// After
required = !field.GetProto3Optional() && field.OneofIndex == nil

Proto3 optional fields use synthetic oneofs (OneofIndex != nil), but GetProto3Optional() already handles that case. The OneofIndex == nil check catches real oneof groups.

Test plan

• Added 3 new test cases to Test_updateSwaggerObjectFromFieldBehavior:
  • Oneof field not required with proto3 field semantics
  • Oneof field not required without proto3 field semantics
  • Oneof field still required when explicitly annotated with REQUIRED
• All existing tests pass with no regressions

Fixes #6334

Made with Cursor

[johanbrandhorst]

Thanks for your contribution, could you please submit an example proto file that reproduces this issue to internal/example/proto? Either add it to an existing one with the flag set, or add a new protobuf file and buf.gen.yaml file with the option set. It's a secondary type of testing we use in addition to template_test.go. Thanks!