Nullable properties with a default of null are not possible according to the 3.0 spec

[johncmunson]

From the spec...

default - The default value represents what would be assumed by the consumer of the input as the value of the schema if one is not provided. Unlike JSON Schema, the value MUST conform to the defined type for the Schema Object defined at the same level. For example, if type is string, then default can be "foo" but cannot be 1.

This, combined with the fact that properties can be nullable, means that it's not possible to have a property with a default value of null.

You can see below that the folks over at Swagger have interpreted the spec literally, as they should. It's the spec that seems to be wrong.

Here's a sample response body on my /todos/{id} route in the Swagger UI...

{
  "id": 0,
  "title": "string",
  "complete": false,
  "note": "null",
  "dueDate": "null",
  "categoryId": 0
}

And here's the OpenAPI schema for my todo objects...

Todo:
  description: A todo item helps you track what you need to get done
  type: object
  required:
    - title
    - complete
  properties:
    id:
      type: integer
    title:
      type: string
    complete:
      type: boolean
      default: false
    note:
      type: string
      nullable: true
      default: null
    dueDate:
      type: string
      nullable: true
      default: null
    categoryId:
      type: integer
      nullable: true
      default: null

Looking at the note property, we can see that the default value of null gets interpreted as the string "null".

Looking at categoryId, it seems that 0 is the closest integer to null they could think of.

[tedepstein]

@johncmunson , as I read the spec, it is only saying that the default value must conform to the type specified in the schema.

Unlike JSON Schema, the value MUST conform to the defined type for the Schema Object defined at the same level.

This is just to contrast the requirement with JSON Schema's less stringent guidance:

It is RECOMMENDED that a default value be valid against the associated schema.

As we're attempting to clarify in this proposal, nullable should function as a type modifier. Under that interpretation, your note Schema Object would translate to a JSON Schema with "type" : ["string", "null"]. In that case, "default" : null should be interpreted as a literal null value, not the string "null", and it should be allowed.

I'm not sure which Swagger component you were referring to. I would say that the behavior of this Swagger component is wrong; it should allow null as a default value. But honestly, the current spec is very unclear as to how nullable should be interpreted, and how it should interact with other keywords. This is yet another case of that, so I'll add it to the proposal for completeness.

Until the proposal is accepted and the v3.0.3 patch of the specification is released, we're kind of in a gray area here. But you might raise the issue in the appropriate Swagger project, and see if they agree that they should change the default validation logic. They might want to make this change even before the proposed clarification is official.

[johncmunson]

@tedepstein thanks for some of the background details surrounding this. I'm glad that it's on the team's radar to clear up some of the ambiguity surrounding null values. After your comments, I now see it from your perspective that the spec is a little ambiguous in this regard, rather than contradicting or wrong. I will raise this issue in the appropriate Swagger repo and see what their perspective is on this. Thanks!

Update:
Here is the related issue I created for Swagger.

[handrews]

If we need to further clarify the language in OAS 3.1, we should track this in #2099. Otherwise it sounds like maybe this can be closed?

[tedepstein]

I would like to clarify the language in 3.1. In fact, #1977 doesn't update the nullable definition to the final language that we agreed to, so we need a PR to cover that anyway. I can propose some other minor clarifications in the same PR.

[darrelmiller]

@tedepstein Could that be because we updated the nullable language in 3.0.3 and it hasn't been ported over to 3.1 yet?

[handrews]

@tedepstein I did the language for 3.1 did it not get merged into the PR? I don't understand how GitHub's "suggestions" work. https://github.com/OAI/OpenAPI-Specification/pull/1977/files#diff-b83206d953b4be5d312b068f82ba0217R2295

[webron]

@handrews suggestions need to be accepted by the original author of the PR, which possibly did not happen in this case. Would be easier to just submit another PR for it.

[philsturgeon]

@tedepstein I added that language from @handrews to #2124 so if there is more improvements to come lets do it over there.

[tedepstein]

Thanks, @philsturgeon , @darrelmiller & @handrews .

Aside from the nullable definition, which is now complete, I would still like to review the other parts of the spec that mention nullable or relate to it, just to see if any part of it needs to be clarified or realigned with the updated definition. It's on my to-do list, and once I get to it I'll open a separate PR.

@philsturgeon , you might want to add this to your punch list in #2099 as @handrews suggests.