Special Case Protocol Conformance: Long-term Tradeoffs for Near-term Conformance

[Joe_Groff]

@skaller

Cool! Thanks for info. Looking at parts of compiler now. Strangely:

public:
  FunctionTypeRepr(GenericParamList *genericParams, TupleTypeRepr *argsTy,
                   SourceLoc throwsLoc, SourceLoc arrowLoc, TypeRepr *retTy,
                   bool GenericParamsAreImplied = false,
                   ArrayRef<TypeRepr *> GenericSubs = {})

it looks like internally at least functions actually use tuples anyhow.

As others had noted, it used to be the case that Swift represented all functions as having a single argument, but that is no longer the case. Bits of the old representation still linger in the compiler.

@skaller

I have guessed as much. Separate compilation is a pain. Does this also explain why associated types returned as opaque types with a protocol don't work? The associate typed is only known in the instance used at the call site, which then has to support the protocol specified by the opaque return type. Problem is, its not known by the function until run time what the associated type is, because its passed from the caller. So the information is available but not at compile time, so static conformance verification can't work. I'm just guessing....

That used to be the case, but protocol witness tables now record all of the associated type information. The "Self or associated type" limitation was an artifact of that missing information that has lost its technical reasons for existing. You can see this thread for discussion about lifting the restriction:

Lifting the "Self or associated type" constraint on existentials

When the first seed of Swift came out many years ago, there were technical reasons for the "self or associated type" constraint on protocol existentials: At that time, protocol witness tables did not carry associated type information, so it was impossible to re-open the dynamic type in order to dispatch methods on an existential when a protocol had associated types. This was fixed a while ago in order to allow for recursive protocol constraints, but we held on to the restriction on existentials …