Allow empty "required" array

[gibson042]

(copied from json-schema/json-schema#232 )

Draft v4 specifies that a required array "MUST have at least one element", which makes it impossible to explicitly indicate that all properties are optional (and therefore also impossible to distinguish intentional optionality from accidental omission of required). Empty values ("required": []) should be permitted.

[erosb]

which makes it impossible to explicitly indicate that all properties are optional

I don't think it is a problem. Being able to express the same thing in multiple ways (an explicit and an implicit notation) is not a goal of the specification.

[gibson042]

Then why is there an integer type, to point out just one from a multitude of counterexamples? And why not make required required, so author intention is always clear?

Put another way, the default value of minItems is 0. Why deviate from that default for required?

[erosb]

And why not make required required, so author intention is always clear?

That should be a different proposal.

[gibson042]

I was pointing out the inconsistency in claiming that JSON Schema is good about preventing multiple ways of expressing the same thing, not seriously suggesting requiring required. This (allowing required: []) is the serious proposal, because an empty value conveys information that is not possible if the property is absent (and in fact I've begun defining schemas with _required: [] as a workaround).

[awwright]

I would personally try to leave out a keyword if it's not going to have any effect on validation (since keywords generally add constraints to an instance), but not being overly pedantic about these sorts of things seems pretty reasonable.

There might be a good reason to want to leave in a keyword even if it's not going to do anything.

[handrews]

I agree with @awwright about trying to leave it out, but as long as required doesn't become required (no keyword should be required, as {} should always be valid and validate everything) I like the idea of allowing "required": []

I think mostly we allow putting in the default explicitly, so required is currently an exception to that pattern.

[Relequestual]

If it says in the spec, that if required is not included, then all are optional, then I don't see that there's a problem.

[epoberezkin]

No objection here to allowing required: [], it's backward compatible :)...

[awwright]

Er, as I mentioned in #112, it turns out "dependencies" has this behavior too.

I'd like to just modify "stringArray" so both "dependencies" properties and "required" allows multiple items. They're both used for the same purpose, we should probably modify this behavior together.

[handrews]

@gibson042 with #112 now merged (thanks, @awwright !) I think this is now fixed. If you agree could you please close?

[epoberezkin]

In parallel with other keywords in I-D (minLength, etc.), it can be said that {"required":[]} is equivalent to {}, so [] is default for required.

[awwright]

Good point, I'll reopen this briefly so I don't forget

[awwright]

Addressed in #171

[handrews]

@awwright I already included this fix in #168 with all the other meta-schema stuff.