Updated the common test suite

[iainbeeston]

This PR just updates json-schema to use the latest version of the common test suite.

Disclaimer: I've just added some tests around how defaults are expected to work (i.e. they're for documentation purposes only) and this will probably fail with the json-schema gem.

[pd]

The invalid defaults will only cause validation to fail if we use insert_defaults, it seems. I took the first example from your commit to the common test suite (json-schema-org/JSON-Schema-Test-Suite@f57888c):

schema = { 'properties' => { 'foo' => { 'type' => 'integer', 'default' => [] } } }
obj = {}

JSON::Validator.validate(schema, obj) #=> true
obj #=> {}

JSON::Validator.validate(schema, obj, insert_defaults: true) #=> false
obj #=> {"foo"=>[]}

I suppose we should add a test to demonstrate this case. Or actually change the behavior so that default values are filled in after validation? But the interaction of required + default prolly gets weird there.

[pd]

Meanwhile, 👍, I see no reason not to merge this as is. It at least validates that the standard calls to validation will work as draft04 expects.

[iainbeeston]

Oh really? I thought that we always insert defaults during validation? (But copy them across to the original object at the end, if insert defaults is enabled) Maybe my understanding is wrong

Anyway, I'm glad to see the tests pass!

[RST-J]

I believe the current behavior is to insert defaults into a copy of the current object under validation and if this object does not pass validation the original object is restored and the validation error propagated.

But I like @pd's idea to first validate and insert defaults afterwards. I think this is the most intuitive approach (considering the fact that for validation they are irrelevant as per the spec) to handle defaults if we decide to do so.
However, we should only introduce/change this handling of defaults if it is not too messy to handle (anyOf, allOf pops to my mind again) and otherwise just get rid of insert defaults.

And for the original subject: 👍 ;)

[iainbeeston]

Just for completeness - I was wrong and you guys were right - defaults are only ever inserted (even into the working copy) of insert_defaults is enabled.=