How to validate JSON Schema based on combination of an array of enums?

Question

Given:

Say that I am defining a schema for Contacts. But, I can have "Primary Contact", "Student" or one who is both; and different properties that go with all three choices. The contact types are defined in an array of contact_type: [ "Primary Contact", "Student" ] which can be either one, or both.

Say that the fields are as such per contact type:

• If Primary Contact, then I want phone_number
• If Student, then I want first_name
• If Student and Primary Contact then I want phone_number and first_name

Usage

I use Ajv library to validate in Node.js using a code like such:

function validator(json_schema){
    const Ajv = require('ajv');
    const ajv = new Ajv({allErrors: true});
    return ajv.compile(json_schema)
}

const validate = validator(json_schema);

const valid = validate(input);

console.log(!!valid); //true or false
console.log(validate.errors)// object or null

Note: I've had trouble with allErrors: true while using anyOf for this, and I use the output of allErrors to return ALL the missing/invalid fields back to the user rather than returning problems one at a time. Reference: https://github.com/ajv-validator/ajv/issues/980

Schema

I have written the following schema and it works if I do either "Student" or "Primary Contact" but when I pass both, it still wants to validate against ["Student"] or ["Primary Contact"] rather than both.

 {
  "$schema": "http://json-schema.org/draft-07/schema",
  "type": "object",
  "required": [],
  "properties": {},
  "allOf": [
    {
      "if": {
        "properties": {
          "contact_type": {
            "contains": {
              "allOf": [
                {
                  "type": "string",
                  "const": "Primary Contact"
                },
                {
                  "type": "string",
                  "const": "Student"
                }
              ]
            }
          }
        }
      },
      "then": {
        "additionalProperties": false,
        "properties": {
          "contact_type": {
            "type": "array",
            "items": [
              {
                "type": "string",
                "enum": [
                  "Student",
                  "Primary Contact"
                ]
              }
            ]
          },
          "phone": {
            "type": "string"
          },
          "first_name": {
            "type": "string"
          }
        },
        "required": [
          "phone",
          "first_name"
        ]
      }
    },
    {
      "if": {
        "properties": {
          "contact_type": {
            "contains": {
              "type": "string",
              "const": "Student"
            }
          }
        }
      },
      "then": {
        "additionalProperties": false,
        "properties": {
          "contact_type": {
            "type": "array",
            "items": [
              {
                "type": "string",
                "enum": [
                  "Student",
                  "Primary Contact"
                ]
              }
            ]
          },
          "first_name": {
            "type": "string"
          }
        },
        "required": [
          "first_name"
        ]
      }
    },
    {
      "if": {
        "properties": {
          "contact_type": {
            "contains": {
              "type": "string",
              "const": "Primary Contact"
            }
          }
        }
      },
      "then": {
        "additionalProperties": false,
        "properties": {
          "contact_type": {
            "type": "array",
            "items": [
              {
                "type": "string",
                "enum": [
                  "Student",
                  "Primary Contact"
                ]
              }
            ]
          },
          "phone": {
            "type": "string"
          }
        },
        "required": [
          "phone"
        ]
      }
    }
  ]
}

Example Valid Inputs:

• For just ["Primary Contact"]:

    {
        "contact_type":["Primary Contact"],
        "phone":"something"
    }

• For just ["Student"]:

    {
        "contact_type":["Student"],
        "first_name":"something"
    }

• For ["Primary Contact", "Student"]

    {
        "contact_type":["Primary Contact", "Student"],
        "phone":"something",
        "first_name":"something"
    }

Question:

I would like this to validate even if allErrors: true , is this possible? If not, how should I change the schema?

Footnotes

I don't want to change the "contact_type" from being an array unless it is the last resort. (it is a requirement, but can be broken only if there's no other way)

I can't allow any additionalItems, therefore I'm fully defining each object in the if statements although contact_type is common. If I move contact_type out, then I get error messages about passing contact_type as an additionalItem (it looks at the if statement's properties and doesn't see contact_type when it's taken out to the common place). This is why my initial properties object is empty.

Answer 1

Here's how I might go about solving the validation issue: https://jsonschema.dev/s/XLSDB

Here's the Schema... (It's easier if you try to break up concerns)

{
  "$schema": "http://json-schema.org/draft-07/schema",
  "type": "object",

First, we want to define our conditional checking subschemas...

  "definitions": {
    "is_student": {
      "properties": {
        "contact_type": {
          "contains": {
            "const": "Student"
          }
        }
      }
    },
    "is_primay_contact": {
      "properties": {
        "contact_type": {
          "contains": {
            "const": "Primary Contact"
          }
        }
      }
    }
  },

Next, I'm assuming you always want contact_type

  "required": ["contact_type"],
  "properties": {
    "contact_type": {
      "type": "array",
      "items": {
        "enum": ["Primary Contact", "Student"]
      }
    },

And we need to define all the allowed properties in order to prevent additional properties. ( draft-07 cannot "see through" applicator keywords like allOf . You can with draft 2019-09 and beyond, but that's another story)

    "phone": true,
    "first_name": true
  },
  "additionalProperties": false,

Now, we need to define our structural constraints...

  "allOf": [
    {

If the contact is a student, first name is required.

      "if": { "$ref": "#/definitions/is_student" },
      "then": { "required": ["first_name"] }
    },
    {

If the contact is a primary contact, then phone is required.

      "if": { "$ref": "#/definitions/is_primay_contact" },
      "then": { "required": ["phone"] }
    },
    {

However, additionally, if the contact is both a student and a primary contact...

      "if": {
        "allOf": [
          { "$ref": "#/definitions/is_student" },
          { "$ref": "#/definitions/is_primay_contact" }
        ]
      },

Then we require both phone and first name...

      "then": {
        "required": ["phone", "first_name"]
      },

Otherwise, one of phone or first name is fine (which one is covered by the previous section)

      "else": {
        "oneOf": [
          {
            "required": ["phone"]
          },
          {
            "required": ["first_name"]
          }
        ]
      }
    }
  ]
 }

I'm not convinced this is the cleanest approach, but it does work for the requirements you've provided.

As for getting validation errors you can pass back to your END user... given the conditional requirements you lay out, it's not something you can expect with pure JSON Schema...

Having said that, ajv does provide an extension to add custom error messages, which given the way I've broken the validation down into concerns, might be useable to add custom errors as you're looking to do ( https://github.com/ajv-validator/ajv-errors ).