Collapse `anyOf` with only one option?

[stevekuznetsov]

If I have a schema like:

        "rings": {
          "type": "array",
          "description": "Array of the deployment ring object.",
            "items": {
                "anyOf": [
                    {
                        "$ref": "#/definitions/deploymentRing"
                    }
                ]
            }
      }

In a sense, it's equivalent to:

        "rings": {
          "type": "array",
          "description": "Array of the deployment ring object.",
          "items": {
              "type": "object",
              "$ref": "#/definitions/deploymentRing"
          }
      }

For users that generate Go types using a JSONSchema they do not control, the use of anyOf in the above example will generate []interface{}, which is unfortunate and very much so not ergonomic, either on generation or parsing. Would it be acceptable to generate a Go struct in these cases with a definite type for the slice, even if that is not perfectly what the author of the JSONSchema intended?

I am not sure, but it seems like there may be some other tool that generates these "loose" JSONSchemas with anyOf array items with only one type, perhaps also a bug in that generator. Would be awesome to work around such cases on the consumer end with this tool.

[stevekuznetsov]

If this proposal is acceptable, I am happy to implement such a thing, also happy to hide it behind a flag.

[stevekuznetsov]

@omissis any thoughts on this? Happy to start work ASAP if you are OK with the idea.

[omissis]

Hey, I think that while this approach might be considered attractive for one-child AnyOfs, it exposes the client code that is using the generated code to breaking changes in those cases where the AnyOfs eventually evolve to contain more than one item: for this reason I am not very keen on going down this path. Unfortunately Go's type system is what it is and Go generics are what they are, so I don't see a good way to provide great ergonomy for now.

[stevekuznetsov]

@omissis wouldn't the transition from a one-length anyOf to a n-length anyOf just mean that the generated code would change from something like

type Thing struct {
    Member MemberType `json:"member"`
}

type MemberType struct {
    Field string
}

To something like the current generation, which is

type Thing struct {
    Member any `json:"member"`
}

type MemberType struct {
    Field string
}

---

In what cases would this be breaking for consuming code? Naively I was expecting no impact for some consumer like:

func main() {
    myThing := Thing{
        Member: MemberType{
            Field: "whoa",
       },
    }
}

For example, on the playground

---

Changes from a one-length anyOf to some other one-length anyOf would be breaking, yes, but they would be breaking regardless if that were the true intent of the schema (and in the current generation, this break would not be caught by the compiler!)

[stevekuznetsov]

@omissis thoughts? On coding this I realized we just ignore these fields today. The update would be trivial:

$ git diff -- pkg/generator/schema_generator.go
diff --git a/pkg/generator/schema_generator.go b/pkg/generator/schema_generator.go
index 25bea18..24d384c 100644
--- a/pkg/generator/schema_generator.go
+++ b/pkg/generator/schema_generator.go
@@ -704,6 +704,12 @@ func (g *schemaGenerator) generateTypeInline(
                        }
                }
 
+               for _, collection := range [][]*schemas.Type{t.AllOf, t.AnyOf, t.OneOf} {
+                       if len(collection) == 1 {
+                               return g.generateDeclaredType(collection[0], scope)
+                       }
+               }
+
                typeIndex := 0
 
                var typeShouldBePointer bool

Please let me know what backwards compatibility issues you forsee - I hope my comment above helps to show my thinking. Simple summary:

len(anyOf) Before	len(anyOf) After	Type Before	Type After	Impact
1	1	typeA	typeB	breaking change, but the user will appreciate this since the schema did in fact break
1	N>1	typeA	interface{}	not a breaking change
N>1	N>1	interface{}	interface{}	not a breaking change
N>1	1	interface{}	typeA	breaking change, but a user will appreciate this since the schema did in fact break

I believe some upstream tooling that generates JSON Schemas from e.g. .NET applications may be incorrectly using anyOf even where only one type is possible, so consuming these generated JSON Schemas is very tricky without this kind of functionality.

[stevekuznetsov]

@omissis any thoughts here?

[stevekuznetsov]

@omissis any chance you have feedback on this?

[stevekuznetsov]

@omissis are you still maintaining this project? Do you have any feedback on this proposal?

[stevekuznetsov]

@omissis sorry to be a nag - I really would like to not have a long-lived fork for this project, are you able to chime in on this proposal?

[mestafin]

@stevekuznetsov i support your proposal

[perher]

I was initially skeptical to this feature request as it introduces extra complexity to the already complex codebase, but then realized I'm also stuck with external schemas that has this anti-pattern.

In my mega branch (#473) I added basic support behind a new AliasSingleAllOfAnyOfRefs (not sure about naming this) config option that when enabled creates an alias type definition instead of duplicating the types when only a single subschema. So feel free to test that if this is still relevant.