cc @xedin

@swift-ci create

Comment by Vuk (JIRA)

When type checking the example in the Description

 let x: Int = 1^^^

It does not attempt a disjunction choice for an implicitly unwrapped optional because the function ref kind is Unapplied so it returns false in this method in `ConstraintSystem.cpp`. which causes it not to introduce a disjunction and causes the failure in the example.

bool OverloadChoice::isImplicitlyUnwrappedValueOrReturnValue() const {

  ...

  switch (getFunctionRefKind()) {
  case FunctionRefKind::Unapplied:
  case FunctionRefKind::Compound:
    return false;
  case FunctionRefKind::SingleApply:
  case FunctionRefKind::DoubleApply:
    return true;
  }
  llvm_unreachable("unhandled kind");
}

Here is the expression dump

(postfix_unary_expr type='$T2' location=/Users/vuk/Desktop/test.swift:5:15 range=[/Users/vuk/Desktop/test.swift:5:14 - line:5:15]
  (declref_expr type='$T0' location=/Users/vuk/Desktop/test.swift:5:15 range=[/Users/vuk/Desktop/test.swift:5:15 - line:5:15] decl=test.(file).^^^@/Users/vuk/Desktop/test.swift:3:14 function_ref=unapplied)
  (paren_expr implicit type='($T1)' location=/Users/vuk/Desktop/test.swift:5:14 range=[/Users/vuk/Desktop/test.swift:5:14 - line:5:14]
    (integer_literal_expr type='$T1' location=/Users/vuk/Desktop/test.swift:5:14 range=[/Users/vuk/Desktop/test.swift:5:14 - line:5:14] value=1 builtin_initializer=**NULL** initializer=**NULL**)))

Should the FunctionRefKind be SingleApply for the postfix operator and if so would that be an issue prior to type checking or does that get resolved/changed during type checking? In an equivalent function call the function_ref is SingleApply. Any help is greatly appreciated!

vukrado (JIRA User) Yes, the FunctionRefKind should be SingleApply for an operator call. The function ref kind gets determined in PreCheckExpression by the call to markDirectCallee. Looks like we need to be calling it for all ApplyExprs instead of just CallExprs.

Comment by Vuk (JIRA)

It looks like that was it @hamishknight. Thank you! I made a PR here #32737.

Resolving as PR was merged

cc @hamishknight