[lldb] Implement TypeSystemSwiftTypeRef::IsMeaninglessWithoutDynamicResolution

[kastiglione]

Implement TypeSystemSwiftTypeRef::IsMeaninglessWithoutDynamicResolution.

rdar://68171389

[kastiglione]

Running the tests results in one SwiftASTContext cross validation assertion, for the symbol $sxIgr_D:

% swift demangle -expand sxIgr_D
Demangling for $sxIgr_D
kind=Global
  kind=TypeMangling
    kind=Type
      kind=ImplFunctionType
        kind=ImplConvention, text="@callee_guaranteed"
        kind=ImplResult
          kind=ImplConvention, text="@out"
          kind=Type
            kind=DependentGenericParamType
              kind=Index, index=0
              kind=Index, index=0
$sxIgr_D ---> @callee_guaranteed () -> (@out A)

It seems something about this tree pattern will need to be special cased.

[kastiglione]

I had a new thought, maybe this is a bug and the existing implementation is incorrect? Should IsMeaninglessWithoutDynamicResolution return true, or false, for the type @callee_guaranteed () -> (@out A)?

[adrian-prantl]

This is a generic type, so naively one would think it is dynamic. However, since it is a function type, we don't actually have any means to bind it like we do for other types, since it the binding would only happen at one specific call site.

I think it would be most intuitive if both implementations returned false. Probably by special-casing function types.

[kastiglione]

Sounds reasonable.

Probably by special-casing function types.

Do you have any thoughts on how to narrow which function types? There are a number of node kinds named with "FunctionType":

% rg FunctionType include/swift/Demangling/DemangleNodes.def
72:NODE(DifferentiableFunctionType)
73:NODE(EscapingDifferentiableFunctionType)
74:NODE(LinearFunctionType)
75:NODE(EscapingLinearFunctionType)
87:NODE(NoEscapeFunctionType)
99:NODE(FunctionType)
126:NODE(ImplFunctionType)
216:NODE(ThinFunctionType)
233:NODE(UncurriedFunctionType)

[adrian-prantl]

I would just defer this question to TypeSystem::IsFunctionType() and worry about whether it is implemented correctly when/if that becomes a problem.

[kastiglione]

I see a IsFunctionType predicate already exists. It checks for only FunctionType and ImplFunctionType. I'll use that.

[kastiglione]

note that this will demangle the type twice.

[adrian-prantl]

That's ok, we will do an NFC performance pass once the implementation is complete. For now let's focus on clarity!

[kastiglione]

@swift-ci test

[kastiglione]

@swift-ci test

[kastiglione]

@swift-ci test