cc @xedin @hborla looks like a regression since 5.2. It seems like we have a failed constraint (argument conversion from Int to String) in the system and so salvage() bails out when it sees it and we don't end up actually solving the system in salvage mode.

This problem seems to be related to this early return on CSSimplify

This also leads to ambiguous diagnostics in situations like

infix operator ---
 
func --- (_ lhs: String, _ rhs: String) -> Bool {
  return true
}
let x = 1
 
x --- x // type of expression is ambiguous without more context

It is basically the same case

IIRC you get a diagnostic if you have two different variables (x and y) so it seems like this only happens when you have >1 failure associated with the same argument?

Do you mean like

func twoargs(_ x: String, _ y: String) {}

func test() {
  let x = 1
  let y = 1

  twoargs(x, y) // type of expression is ambiguous without more context
}

infix operator ---

func --- (_ lhs: String, _ rhs: String) -> Bool {
  return true
}
let x = 1
let y = 1

x --- y // type of expression is ambiguous without more context

In this case, we still get ambiguous for the same reason, the problem is when having more than one argument failure (for any argument) e.g. in those cases args #0 and #1 fail on the same call anchor. I really don't know what would be an approach to solve this, as the comment in line 3644 states ("we can consider overload unrelated"), but this leads to problems like those examples here if all overloads are ignored ... and change this would impact a lot of other diagnostics.
Maybe we could change this to return this as diagnosed, and implement a custom logic on `ArgumentFix::diagnoseAmbiguity`... not really sure

Oh sorry, I think I got confused when I was testing out a few other cases. I thought I remembered seeing a diagnostic when using different variables instead of the same one.

I think it might work to start recording all of the argument failures and increase a score significantly after each one to indicate that the more argument are mismatching the least desirable this overload choice is...

@xedin I've seen that there is already a score increasing for argument fix here

@LucianoPAlmeida I'm taking about this bit

Right @xedin let's try that!

Fixed on master @marcrasi can you please test on the next available snapshot? thanks

I verified it. Thanks!!