Type substitution: When substituting SILFunctionTypes we substitute u…

[aschwaighofer]

…nbound Objective-C generic for bound ones

IRGen does so. So don't assert on this case.

rdar://63509292

[aschwaighofer]

@swift-ci Please test

[slavapestov]

IRGen's behavior has caused problems in the past. I think it's a bit of a hack. This fix is totally fine but it would be nice if we could wean IRGen off of UnboundGenericTypes entirely. They should not exist after type checking.

[aschwaighofer]

IRGen creates it here:

// Get the type that exists at runtime to represent a compile-time type.        
CanType IRGenModule::getRuntimeReifiedType(CanType type) {                      
  // Leave type-erased ObjC generics with their generic arguments unbound, since
  // the arguments do not exist at runtime.                                     
  return CanType(type.transform([&](Type t) -> Type {                           
    if (CanType(t).isTypeErasedGenericClassType()) {                            
      return t->getAnyNominal()->getDeclaredType()->getCanonicalType();         
    }                                                                           
    return t;                                                                   
  }));                                                                          
}  

[slavapestov]

Yeah, it seems like whoever consumes the result of that is unable to handle a BoundGenericType whose decl is an Objective-C generic class. It seems like it would be easier to fix that code to instead ignore the generic arguments there, rather than performing this transformation in the right place.

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 532f0cb

[aschwaighofer]

@swift-ci Please test linux

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 532f0cb

[aschwaighofer]

@swift-ci Please test linux

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 532f0cb

[aschwaighofer]

@swift-ci Please test linux

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 532f0cb

[aschwaighofer]

@swift-ci Please test linux

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 532f0cb

[jckarter]

@slavapestov Part of the issue is that we (at least historically) rely on the type identity being the same for types that are equivalent at runtime in parts of IRGen. There's likely to be better ways to do that, though.

[aschwaighofer]

@swift-ci Please test linux

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 532f0cb

[aschwaighofer]

@swift-ci Please test linux

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 532f0cb

[aschwaighofer]

@swift-ci Please test linux

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 532f0cb

[aschwaighofer]

@swift-ci please test linux

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 532f0cb

[aschwaighofer]

@swift-ci Please test linux

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 532f0cb

[aschwaighofer]

@swift-ci Please smoke test linux