[Proposal] Retroactive protocol refinement

[johnno1962]

Hi Apple,

A proposal for this feature that is mentioned in the generics manifesto given that attempts at an implementation are now showing fruit.

[Jumhyn]

IIUC, this section is meant to be a discussion of whether/how library authors may use the proposed feature in a resilient library. Specific to this proposal, I think the Effect on API resilience section should communicate whether library authors may add or remove retroactive refinements in an API compatible (and ABI compatible, for libraries built in evolution mode) way.

[johnno1962]

Thanks for the comment, the proof of concept and resilient protocols is not a good mix indeed you can’t really use the feature to extend the conformances of a protocol in the standard library which is resilient as it needs to know the witness table layout to extend. Ultimately this precludes a lot of use cases even though they may work for now. Can you think of a wording that would capture this?

[Jumhyn]

Is this limitation just a by-product of the current prototype implementation, or is it an inherent limitation of the feature as proposed? If it's the latter, I would say something like,

Retroactive refinement is not available for protocols declared in libraries with evolution mode enabled. If such a retroactive conformance is attempted, the compiler will emit an error of the form "".

I think you should also address the compatibility implications of adding a new retroactive refinement within a resilient library. For instance, the current Library Evolution page has this to say about protocols (among other items):

• New protocol conformances can be added to classes, structs, and enums. (Even if they’re @Frozen.)
• Changes to a protocol’s list of refined protocols are not allowed.

So this section of your proposal should answer the question, "can a resilient library add a retroactive conformance to an existing protocol?". This is subtly different than the question that your comment addresses, which is, "can a client of a resilient library retroactively refine a protocol from that library?". Both questions should be addressed!

[johnno1962]

I’ve updated the proposal to outline this limitation in more detail.

[hartbit]

Name needs updating :)

[johnno1962]

Thanks, David, I've updated it. Its not like the proposal is going anywhere anyway 🙂

[rjmccall]

The Language Workgroup is trying to be better about not allowing proposals to sit for long periods without making progress. Unfortunately, this PR has fallen into that state, and it's not clear to us that anyone is still actively engaged with it. I am closing it as stale, but that's not a judgment on the underlying proposal; please feel free to reopen it if you're interested in bringing it through evolution.

This proposal will also need extensive implementation feasibility review.