Removing dependencies?

[epoberezkin]

@handrews could you please point me to the discussion about adding the note about the possibility of removing dependencies?

If it wasn't discussed, I would propose to remove this note from the spec until such possibility is agreed.

Without "dependencies", a concise schema:

{
  "dependencies": {
    "foo": ["bar"],
    "baz": ["quux"]
  }
}

will have to be replaced with:

{
  "allOf": [
    {
      "if": {"required": ["foo"]},
      "then": {"required": ["bar"]},
    },
    {
      "if": {"required": ["baz"]},
      "then": {"required": ["quux"]},
    }
  ]
}

which seems both more verbose and more difficult to understand.

So I am not sure what could be the motivation to remove it.

[handrews]

You were the one who OK'd it:
#180 (comment)

The reason for putting the comment in is to provoke reactions from people who are actually using the keyword and want to keep it. The note is doing what it is intended to do. It is not in any way a commitment to removing the keyword.

I'm closing this not because your objections aren't useful (they are, and I've noted the point), but because I want people other than the usual suspects to file objections. That works better if they can't find one that's already open :-)

[handrews]

Thinking on this a bit more, it's really the schema form of dependencies that I would prefer to remove. I agree that the property name form is a nice shortcut. But I don't think that "if": {"required": ["foo"]} on its own, with an arbitrary then, is worth having a shortcut for.

Or if it is, I really don't like having the same keyword used for the string and schema forms. They're not really the same thing and it's hard to reason about the keyword as it can be either a subschema keyword or a value keyword which is weird.

[handrews]

type has two forms, but both are value forms (single string vs array of strings). items has two forms, but both are subschema forms (single subschema vs array of subschemas).

dependencies has one value form and one subschema form. I think the value form is worth keeping, but I'd like to encourage removal of the subschema form in favor of if. Perhaps I could clarify the note in this manner?

[epoberezkin]

I agree about schema form, this one was a bit confusing indeed.

I think I meant schema form in that comment too ;)

[pipobscure]

I find the schema form very useful. Please let's not remove it. If anything let's remove the values form.

{ "foo": { "required": [ "bar" ] } }

isn't that much worse than

{ "foo": [ "bar" ] }

but replacing the schema version with a combination of allOf/anyOf/oneOf/if/then/else is much worse.

It's much clearer than if/then/else.

Side-jibe at if/then/else: These are a bad idea to begin with, but since they are not obligatory to use, I'm fine keeping them. One reason why they are bad is that they mutate JSON-Schema from a schema to a programming language.

[handrews]

One reason why they are bad is that they mutate JSON-Schema from a schema to a programming language.

This was covered in excruciating detail in #180, and will not be revisited until the keywords have been available in a published draft to see how they are or are not used "in the wild."

[markchart]

Although I am a big fan of "if","then","else" I would not care to lose (either form of) "dependencies". Some constraints may be expressed much more clearly using "dependencies" (just as some things are better expressed using "if","then","else" than tangled "xxxOf" clauses).

It is counterproductive to try to squeeze out all redundancies at the cost of clarity, not just of exposition but of intent.