Regarding uniqueItems error interpretation

[nuno-andre]

JSON Schema specs define an array as an ordered sequence of zero or more values and its instance equality as both are arrays and have an equal value item-for-item but, AFAIK, do not state which mathematical object corresponds to this type and therefore neither its concept of multiplicity.

So, for instance, in a uniqueItems validation against [1, 1, 2, 2, 2, 3], how many errors should we say have occurred?

• Two errors: 1 and 2 are not unique.
• Three errors: the second 1, the second 2, and the third 2 are repeated elements.
• Five errors: {1, 1, 2, 2, 2}.

[handrews]

@gregsdennis would you handle this as one entry in the error structure, and any information about how many / what kind of problems would go into the detail string? Or multiple error entries, one for each way in which it could fail? What do you do for additionalProperties: false? One error for all extra properties or one error per extra property?

[karenetheridge]

In my implementation I generate one error (all the validation vocabulary keywords generate just one error), and an indication of where in the data the error lies.

For example, the data [ 1, 1, 2, 2, 1 ] evaluated against the schema {"uniqueItems": true } generates this evaluation result:

{
  "errors" : [
    {
      "error" : "items at indices 0 and 1 are not unique",
      "instanceLocation" : "", 
      "keywordLocation" : "/uniqueItems"
    }
  ],
  "valid" : false 
}

(edited data instance to add more duplicate items, to make it clear that even with longer arrays only one error is generated)

[gregsdennis]

I think a single output unit would suffice here. It's the keyword that failed to validate, so it makes sense to only include a single unit for the keyword.

---

Bearing that in mind, #973 is considering adding an instance-based structure (instead of a schema-based structure) as a possible output format. If this is used, it may make sense to include each violation. This hasn't been hashed out yet though.

[nuno-andre]

Yes, I was not precise enough. I was assuming one validation error, the question was about the need to identify the individual errors in verbose or detailed outputs. Luckily, both questions were solved :)

I see that @karenetheridge's implementation adopts the third option, which is consistent with the definition of an array as an indexed set and also probably the most robust approach.

I am going to close this issue and subscribe to #973 to be aware of upcoming changes.

Thank you all for this answer and for all the others, this repo has become my JSON-Schema Bible ;)