Draft 4 - recursive references test requires internal referencing

[matt-allan]

Hello,

I updated the test suite which pulled in this test.

The schema's id is http://localhost:1234/tree and it contains {"$ref": "node"}. When my validator encounters the reference it resolves to http://localhost:1234/node. At this point it attempts to load http://localhost:1234/node and fails because that schema cannot be loaded via HTTP.

I see that http://localhost:1234/node is defined below within the same schema but I don't see which part of the Draft-04 spec says that you should attempt to load the schema from within the same document first.

It looks like Draft-05 covers that in section 8.2.1 but that section does not exist in Draft-04. The closest thing I see is section 7.2.3. It seems like this might be considered inline dereferencing but that isn't required by the spec.

Am I interpreting the spec incorrectly or should this behavior not be required for Draft-04?

[handrews]

I'll let @Julian and/or @awwright decide whether we want to do anything with the suite for this, but my view is that draft-04 is old and we want to encourage people to modernize, so unless something's really easy here let's not bother.

[Julian]

Hmmmm I wonder whether I've pulled in this test in my own validator -- in general on earlier drafts predating the current renaissance my recommendation has been "for very old tests leave em as is whenever the spec is unclear, because people have likely coded to them in the interim", so 2 criteria to be explicit:

• there's a feasible way to have understood the spec includes the behavior
• the test has been in the suite for awhile

[matt-allan]

Oh, I should have mentioned that the test in question was merged in commit 0178c74. I had the test suite locked to the last tag in my dependency manager so it wasn't pulled in until I switched to master. Since it's been in here for a year now I am fine with leaving it and skipping the test. I know 2 PHP validators have been written in PHP since it was merged that now pass that test, so it seems likely that quite a few implementations have implemented that behavior by now.

Thanks for the feedback 👍