+1. I don't know if I'm for #313 but, if we don't do #313 or this, we'll likely have to break backwards-compatibility (at least in a limited sense) to add linking from complex attributes down the line.

Adding for 1.0 final review.

I'm 👍 to finding a way to explicitly allow nested links for future consideration, but I think it is premature to have examples of this usage in the spec. What about simply reserving the links key at any level of nesting and leaving it at that?

Actually, reserving any reserved keywords at any level of nesting.

Actually, reserving any reserved keywords at any level of nesting.

Yes, this is sufficient I believe.

"Any reserved keywords" is not well-defined. Would this mean id, type, links, and meta?

Yes - any keywords reserved in a resource object.

@dgeb Gotcha. What about the example? Remove altogether, remove linking, or leave as is?

Example usage was requested in #289.

@bintoro I definitely don't want to include an example here. We are simply reserving the keys, without making a statement about their usage.

Updated!

👍

@bintoro please rebase

Rebased.

By the way, if there's a short answer to why linking from complex attributes is so contentious, I'd love to hear it!

For my part, it's lack of experience with implementations doing this in the wild. It could well be that the resulting impact is minor, but I feel it bears more consideration before we explicitly support this. I'd love to continue the discussion post 1.0, though!

I am out of town for most of March but when I return I am going to create a test suite using a reference database that any implementor can use to validate compliance with the spec. Once that is in place I think it will be much easier for us to reason about additions like this--we can temporarily add them to the test suite and reference database to see the real world implications.

@bintoro thanks for rebasing!

By the way, if there's a short answer to why linking from complex attributes is so contentious, I'd love to hear it!

The possibility of nested, n-level deep links that must be maintained and accessed independent of the containing object (ie. resource) adds a large burden of complexity to implementations. I would prefer that this be implemented as an extension first (and perhaps exclusively) as a proof of concept.

I am going to create a test suite using a reference database that any implementor can use to validate compliance with the spec.

Very excited for this @tkellen!

Has linking from complex attributes made any headway since this discussion?