[swift-evolution] Proposal: Introduce User-defined "Dynamic Member Lookup" Types

[Paul Cantrell]

I’m not at all a fan of _forcing_ IUOs. Depending on the application, return types of Self!, Self?, always-succeeds Self, Self + throws, or Self + fatalError could all be the best approach.

In Python and JS, where objects have a dictionary-like shape, IUOs make sense: code can test for member presence when appropriate, while keeping general member access ergonomic.

In Ruby, where member lookup is fundamentally shaped more like a method call than a dictionary lookup, and where accessing a nonexistent member is a runtime error in the native language, then either throws or fatalError would be more appropriate.

I can imagine data-centric languages where traversing nonexistent members is typical and expected, either because the language has behavior like old-school Obj-C nil handling or because has create-on-traversal semantics. That behavior would be baked into to the shape of the APIs designed for the language. In that case, never-failing Self would be appropriate.

• • •

I like the Chris’s approach where LookupType is an associated type (or, if we go with the empty protocol, behaves like one). It handles all these possibilities nicely, and then some.

It gives implementors latitude to choose the appropriate approach, and thus places the burden the authors of language-specific bridges to make language-appropriate design choices — which is _exactly_ as it should be.

Per my earlier message, I am interested in making well-designed libraries work as beautifully as possible, not in beating the authors of ill-designed ones over the head! The latter is futile, and this is a quintessential example of where pursuing the latter undermines the former.

Cheers,

Paul

> On Dec 9, 2017, at 11:20 AM, Steven Brunwasser via swift-evolution <swift-evolution at swift.org> wrote:
> 
> Just wanted to give my 2¢
> 
> ¢
> I don’t like empty protocols—they feel like an abuse of the feature. I think attributes are the right way to go, since this proposal is about enabling syntactic sugar for types which can’t yet be described in the language as-is. This prevents retroactive conformance on preexisting types, which some have raised as a concern.
> 
> ¢
> I think the discussion about whether or not implementations should throw, return optional, or be implicitly unwrapped is a larger discussion on its own, and should probably be a separate proposal to steer the language towards a more well defined convention. That being said, I’m of the opinion that they should always return an implicitly unwrapped value. The precedent is already in the language, it allows for cleaner syntax while also explicitly stating “hey, just so you know, I might not work, so be careful, ok?”, and callers can choose to be more cautious by explicitly using the ? operator.
> 
> That is all,
> - Steve
> 
> On Dec 8, 2017, at 16:34, Xiaodi Wu via swift-evolution <swift-evolution at swift.org <mailto:swift-evolution at swift.org>> wrote:
> 
>> On Fri, Dec 8, 2017 at 18:08 Jon Gilbert via swift-evolution <swift-evolution at swift.org <mailto:swift-evolution at swift.org>> wrote:
>> See below.
>> 
>> On Dec 6, 2017, at 02:45, Nick Keets via swift-evolution <swift-evolution at swift.org <mailto:swift-evolution at swift.org>> wrote:
>> 
>>> Apologies, I may have misunderstood you. What I wanted to say is that I see no problem allowing "dangerous" stuff that may be abused.
>> 
>> You see no problem with danger and abuse?
>> 
>> I guess we have differing philosophies...
>> 
>> https://developer.apple.com/swift/ <https://developer.apple.com/swift/> states:
>> 
>> “Swift eliminates entire classes of unsafe code.”
>> 
>> Lets keep it that way.
>> 
>> I’m all for this proposal if it can be tweaked to where any of the dangerous invocations contain the word, “Unsafe”, or equivalent.
>> 
>> Again, in Swift, “safety” means something very specific. Trapping at runtime is safe; in fact, trapping at runtime is *precisely the means by which safety is achieved* in the case of integer overflow and array indexing. This proposal introduces nothing that is unsafe.
>> 
>> ~Jon
>> 
>> 
>> _______________________________________________
>> swift-evolution mailing list
>> swift-evolution at swift.org <mailto:swift-evolution at swift.org>
>> https://lists.swift.org/mailman/listinfo/swift-evolution <https://lists.swift.org/mailman/listinfo/swift-evolution>
>> _______________________________________________
>> swift-evolution mailing list
>> swift-evolution at swift.org <mailto:swift-evolution at swift.org>
>> https://lists.swift.org/mailman/listinfo/swift-evolution <https://lists.swift.org/mailman/listinfo/swift-evolution>
> _______________________________________________
> swift-evolution mailing list
> swift-evolution at swift.org
> https://lists.swift.org/mailman/listinfo/swift-evolution

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.swift.org/pipermail/swift-evolution/attachments/20171209/ab005566/attachment.html>